United States                      Atmospheric Sciences
                  Environmental Protection              Research Laboratory
                  Agency       '                   Research Triangle Park NC 27711

                  Research and Development             EPA/600/9-87/024 Oct.  1987

                   OF CHEMICAL MECHANISMS

                    Dr. Roger Atkinson, Workshop Chairman
                          Statewide Air Pollution Center
                        University of California-Riverside
                              Riverside, CA  92521
                               Steering Committee

                               Dr. Harvey Jeffries
                    University of  North Carolina-Chapel Hill

                                Dr. Gary Whitten
                            Systems Applications, Inc.

                               Dr. Fred Lurmann
                    Environmental Research and Technology
                                 Project Officer

                                Basil Dimitriades
                   Atmospheric Chemistry and Physics Division
                    Atmospheric Sciences Research Laboratory
                       Research Triangle Park, NC  27711
                    Atmospheric Sciences Research Laboratory
                      Office of Research and Development
                      U.S. Environmental Protection Agency
                       Research Triangle Park, NC  27711

Although the workshop described in this report has been funded wholly by the
U.S. Environmental Protection Agency under contract 68-02-4183 to Systems
Research and Development Corporation, the report has not been subjected to the
Agency's required peer and policy review and therefore does not necessarily
reflect the views of the Agency and no official endorsement should be inferred.
Mention of trade names or commercial products does not constitute endorsement
or recommendation for use.


       Atmospheric photocheraists  and  developers  and users  of
air quality models need  to  discuss  the  evaluation and documen-
tation of  chemical  mechanisms  used  in  air  quality  simulation
models.  A Workshop, tnerefore, was organized  and conducted on
December 1-3,  1987  by  EPA  to  discuss  the  latest  evidence and
viewpoints on the subject and to solicit trom experts recommen-
dations on optimum approaches  to  mechanistic model evaluation,
documentation, and further development.   Previous practices and
underlying issues in the subject areas  were  reviewed  and dis-
cussed in  background  documents  prepared   and  distributed  in
advance of the Workshop.  Participants agreed that smog chamber
data provide  the most  unambiguous  test  of urban  atmospheric
photochemistry mechanisms. They also agreed,  however, that there
are uncertainties associated with the representation of chamber
radical sources  and  of  photolytic  rates  in outdoor  chambers,
with smog  chamber measurement  errors, and  with the representa-
tion of  as  yet  unknown  reaction pathways.   The  participants
recommended that task forces and/or review groups be established
to discuss and resolve  existing  smog chamber methodology issues,
to assemble  a required  smog chamber data  base  for  mechanism
testing, and  to  review/evaluate  relevant  kinetic  and  smog
chamber data  and mechanism testing  results.  Recommendations
were also  developed  on future  research needs  and  on mechanism
documentation procedures.

The purpose of this workshop is to solicit from experts input that would help
EPA meet its commitment to provide guideline models for use in the development
of effective air pollution control strategies.  Specific objectives are (a) to
develop procedures that could be agreed upon for documenting chemical mechanism
modules so as to ensure error-free transfer of the modules from the developer
to the user, (b) to critically examine and rank approaches to evaluating the
accuracy of gas phase chemical mechanisms  for the ozone, aerosol, and acid rain
aspects of ambient air  quality, (c) to identify all important factors that
determine the reliability, comprehensiveness and precision of each mechanism
evaluation approach, (d) to define a "standard" data base needed for evaluating
a mechanism by a given approach, and finally, (e) to develop, if possible,
standard mechanism evaluation procedure(s).

The underlying problem and the need for standard procedures for documenting and
evaluating chemical mechanisms were first recognized and discussed during a
1983 workshop on the Empirical Kinetic Modeling Approach (EKMA). At that time,
EPA was disturbed by  the fact that models of insufficiently documented validity
were offered as official Agency guidelines for development of costly control
strategies. Furthermore, the plethora of ozone mechanisms in existence, their
differences in terms  of ozone prediction, and the difficulty in documenting
even the relative  validities of the various models caused another problem: they
encouraged the inappropriate practice of State government officials responsible
for development of State Implementation Plans (SIP), to select for application
those mechanistic models that happen to support pre-conceived control targets.
Therefore, for EPA to  be able to defend its control policies and regulations,
and for the States to be able  to develop objective and effective SIP's, it is
imperative that the model-type guidelines issued by EPA be of well documented
utility and validity.

The workshop objective to develop standard  procedures for documenting
mechanistic  models is an important and difficult one, but its  achievement does
not necessarily hinge upon resolution of complex scientific issues.  Therefore,
while detailed discussions will be needed to derive the collective judgment of
the Workshop group, no dichotomy in viewpoint or intensive debate among the
workshop participants is to be expected.  It should be possible to arrive at a
consensus, based on the personal experiences of the participants in dealing
with the problem of using someone else's model. In contrast,  the objectives
related to the evaluation of chemical mechanisms are almost certain to reveal
considerable diversity in viewpoint and to incite intensive debate.  This is
the normal result of the fact that evaluation  of chemical mechanisms is fraught
with unresolved  issues  most of which, in the  lack of convincing experimental
evidence, have acquired a subjective, even philosophical character.  For
example, questions on the evaluation approach for which there are no consensus
answers are: Can  field  data be used to evaluate chemical mechanisms?  Is
comparison against smog chamber data the best approach to evaluating
mechanisms? Given the conceptual and practical problems of the experimental

approaches, is the theoretical approach (formulation of multi-thousand step
"master" mechanism) the preferable approach to developing valid mechanisms?  In
addressing these questions, there is unavoidable clash between those believing
in the unchallengeable validity of real world data and those believing that
real world is hopelessly complex; and also between those advocating that theory
shall be used as "gospel" and those for whom experimental evaluation is the
only option available.

It is indeed unrealistic to expect that the workshop participants will resolve
all technical issues and agree, for example, on a single mechanism evaluation
procedure or on a "standard" data base of reasonable proportions.  It is
expected,  however, that it will be possible for specific alternatives or
options to be identified, clearly explained, thoroughly scrutinized, and,
finally, rated based on their relative merits and limitations. It is with
these latter expectations in mind that  the workshop organizers decided on the
format and size of the workshop, and, indeed, on the workshop itself.
                                                  Workshop Chairman,
                                                  Basil Dimitriades


Abstract	 iii
Introduction	  iy
Summary and  Recommendations	   1


1  The Science of Mechanism Development and Testing
       Harvey Jeffries
       University of North Carolina-Chapel Hill

2  How Do You  Know  a  Good  Model  When You See One?
       Joseph Tikvart
       U.S. Environmental Protection Agency

3  Chemical Mechanism Development for Applications  in Atmospheric Simulation
       Kenneth Demerjian
       State University of New York-Albany

4  Development  and Testing  of  Gas-Phase Chemical  Mechanisms: Present Status
   and Future Research Needs
       Roger  Atkinson
       University of California-Riverside

5  Review  of  the "Strawman" Document  for the EPA Workshop on  Evaluation/
   Documentation of  Chemical Mechanisms
       Michael Gery
       Systems Applications,  Inc.

6  Uncertainty Analysis and  Captive-Air Studies as Aids in Evaluating
   Chemical Mechanisms for  Air Quality Models
       Alan Dunker
       General Motors Research Laboratories

7  Development  and Testing  of  a Surrogate  Species Reaction Mechanism for
   Use in Atmospheric Simulation Models
       William Carter
       University of California-Riverside

8  Evaluation  and Intercomparison of Chemical Mechanisms
       Fred Lurmann
       Environmental Research  and  Technology, Inc.


A.  Need for Chemical  Mechanism Documentation
       Kenneth Sexton  and  Harvey Jeffries
       University of North Carolina

B.  The Science  of Photochemical Reaction Mechanism Development and Evaluation
    [Strawman  Document]
       Harvey Jeffries  and Jeffrey  Arnold
       University of North Carolina

C.  Workshop Agenda
D.  Workshop  Attendees


      EPA Workshop on
Evaluation/Documentation of
     Chemical Mechanisms
  Summary and Recommendations
            prepared by
       Workshop Steering Committee:
        Roger Atkinson, Chairman,
       Harvey Jeffries, Fred Lurmann,
           and Gary Whitten


The EPA conducted this workshop to discuss the evaluation and documentation
of chemical mechanisms used in air quality simulation models.  A chemical mecha-
nism is the set of chemical reactions and associated rate constants which describes
the transformation of emitted chemicals into intermediate and final products.  In
the context of this workshop, the initially emitted chemicals are hydrocarbons and
oxides of nitrogen, and ozone is the product species of major interest.  From this
list of chemical reactions, a set of coupled, non-linear, differential equations  which
describe  the rates of change of chemical species  concentrations with time are de-
rived.  These equations can be integrated using a computer to model the changing
concentrations of these chemical species. The major use of chemical mechanisms
is to serve as a component of comprehensive air quality models that predict  atmo-
spheric smog formation by simulating the emissions, transport, dispersion, chemical
reactions, and scavenging process that affect air pollution.

    A chemical mechanism must have a certain level of detail to describe adequately
the complex and detailed photochemistry  of smog formation.  Limitations on the
size and speed of computers and in the availability of data for inputs to air quality
models place restrictions on the detail allowed in  a chemical mechanism.  Thus the
development of practical chemical mechanisms for use in urban air quality models
requires a careful balance between detail and simplicity.

    As discussed by Atkinson and Jeffries  in their presentations, the goals of this
workshop were:
     •  to assess present practice in photochemical reaction mechanism development
       and testing for those mechanisms intended for use in urban air quality control
     •  to determine if there might be a commonly agreed upon mechanism evalua-
       tion procedure; and
     •  to determine if there might be a standard data base that would be useful in
       distinguishing among different mechanisms.

    A review of previous practice and a discussion of the underlying issues was pre-
sented in background documents "The Science of Photochemical Reaction Mecha-
nism Development and  Evaluation" by Jeffries and Arnold and "Need for Chemical
Mechanism Documentation" by Sexton and Jeffries, which were distributed to  work-
shop participants prior to the meeting.

    Four scientists and an EPA user of models were asked to respond to the back-
ground documents and to offer their viewpoints on evaluation procedures and testing
databases. These were Kenneth Demerjian, Roger Atkinson, Micheal  Gery, Allan
Bunker,  and Joseph Tikvart.  These four scientists were in agreement that there
is, and has been, a generally accepted procedure for testing the extent of "reason-
able agreement" between model predictions and experimental measurements. This
procedure involves the use of  laboratory kinetic, mechanistic, and product data,
the use of smog chamber data, and the use of other test data, such as captive
air irradiation measurements and ambient air measurements. Many participants
believed, however, that the latter type of comparisons (e.g., model predictions com-
pared to ambient measurements and, for some, even captive air studies) require so
many approximations and suffer from such instrumental limitations that the ex-
tent of agreement expected would  be quite  limited and thus such efforts would not
be clear  tests of our understanding of the  chemical transformation processes. All
four scientists agreed that, although  there certainly were problems with their data,
environmental or smog chambers still provided the most unambiguous data for the
testing of urban chemical transformation mechanisms.  Subsequent discussion con-
firmed that this approach or method was generally the accepted approach used by
the workshop attendees.

    The  EPA model user (J. Tikvart) strongly supported the draft proposal for
mechanism documentation. Other workshop attendees, including William Carter,
Fred Lurmann, Gary Whitten, and  Gregory McRae described their current prac-
tice and  recent model  testing  strategies and results.   While various mechanisms
have been developed and tested against limited numbers of smog chamber experi-
ments, only two mechanisms, the SAPRC/ERT mechanism and the latest Carbon
Bond Mechanism, have been tested against  a large number of chamber experiments
(ca. 500) from different chambers.

    Both of these well-tested mechanisms agree with the large body of experimental
environmental chamber data to within about ±30% for ozone maxima and show
varying levels of "reasonable agreement"  for other measurements (for example, see
Figure 1, 2a, and 2b). Discussion revealed that there is still some disagreement
among the modeling community over how best  to represent the chamber radical
source(s), how  best to represent the photolytic rates  in outdoor chambers, and
whether  some of the species  measurement techniques at the different chambers
perform satisfactorily and how these measurements intercompare among different
chamber groups. In addition, the unknown reaction pathways after hydroxyl attack
on the  aromatic species are represented differently in the two mechanisms.

                Q   SAPRC EC

                -J-   SAPflC ITC
                    SAPKC ore
                    UNC Ch»mt}«r
                          Observed Maximum Oiont (ppmf
Figure 1.  Predicted  and observed maximum ozone concentrations
    for irradiations of complex organic mixtures and NOX, carried out
    in the SAPRC evacuable (EC), indoor Teflon (ITC) and outdoor
    Teflon  (OTC), and in the  UNC outdoor dual chamber (UNC).
    From Carter et a/., EPA/600/3-86/031,  1986.

                                                e. is
                                  I    T

                                K (N02)
i    r
                                                                             i    i
                             "0   80  160  ?40  320  400  «60  560 6-50
                                        TIME  (MINUTES!
                  Figure  2a. Example of SAI model performance.  The
                      lines  are  Carbon Bond Four predictions  and the
                      symbols are experimental measurements from one
                      side  of the UNC outdoor chamber for the Sept. 20,
                      1981 experiment using a 14-component synthetic ex-
                      haust mixture.

              l   I    I     1    I   r
         50  100 150 200 250 300 350 400
                TIME (MINUTES)
                                             * 0.10
                                                                            I    T
                                                   0  100 200  300  400  500 600 700 800
                                                              TIME  (MINUTES)
          I    I
         RONE   '
t 0-06
                      I    I    1    I
                  J	I
     "0   10  80  120  160 200 2
    While many of the species predictions (e.g., nitric oxide (NO), nitrogen dioxide
(NOj), ethylene, toluene, nitric acid (HMOs)) in urban air quality applications agree
remarkably well between the the two mechanisms, ozone predictions can still dis-
agree by 20 to 30% (see Figure 3), leading to differences  in hydrocarbon control
predictions.  The Steering Committee concluded that the different methods of deal-
ing with the uncertainties described above are the most likely cause of the difference
in the mechanism predictions.

    The Steering Committee concluded that, without further work, it was not pos-
sible to choose one of these mechanisms over the other on the basis of scientific
evidence,  and that the EPA should be encouraged to use both mechanisms as a
method to estimate the present uncertainties in control requirement predictions.
Furthermore, similar situations are likely to occur whenever two or more mecha-
nisms are developed over similar time periods.

    The Steering Committee also concluded that several review groups should  be
assembled to review the kinetic and chamber data bases and to review the extent
of agreement among the models and these data bases. These recommendations will
be described below.

    It is clear from the data presented at this workshop, together with work pub-.
lished over the past five years, that a vast amount of progress has been made during
the past decade, both  with respect to urban-area chemical mechanism development
and the data base upon which these mechanisms are based and tested.

    Because  of these advances, chemical mechanisms are now being used for appli-
cations other than urban oxidant control, for example, rural  oxidant, long range
transport and deposition, and the impact of toxic chemical emissions on the ecosys-
tem. These  present and future societal needs make it necessary that the develop-
ment  and testing of even more complex chemical  mechanisms continue, and that
present  chemical mechanisms be updated and/or superseded by mechanisms able
to deal accurately with the formation of gas phase precursors to acid deposition
and to include or be interfaced to aqueous phase and cloud water chemical mecha-
nisms. It is,  however, clear that, despite the great advances made over the past few
years, new challenges remain which require a wider fundamental chemistry data
base coupled with more  detailed and accurate environmental chamber measure-
ments of a wide range of difficult-to-measure species than  is presently customary.
An integrated experimental and theoretical approach, which is closely coordinated
with other proposed or upcoming programs such as the NSF/NASA "Global Tro-
pospheric Chemistry"  project  is needed to address these needs.

                                                       5s S 8 3  g
                                                       C— tj v C

                                                       IS) "3 O m «
                                                           O '-^ f
                                                       u „ •« T "5

                                                       £ 11"? .=
                                                       o « >• a x
                                                       •2 "5 §

                                                       75 £ S

                                                       -   a
                                                       C v. 2
                                                       V 3 J
                                                       u O v


                                                       "2 o so
                                                       0) -O
                                                       >• e
                                                           _: S ae _3

                                                           «j «, .£ 0

                                                           ^ §-S 1
                                                           E •§ E IS
3 3 3 S  3  3  a 3 3 3  !  I 3 S

                                                 « o '= -
                                                 o- o £ '5
                                                 < «   •-
                                                 (fl 'T
    | e 8
    ^ § a
                                                 .H .§ c >
                                                 •a « o tc

                                                 2 3 -E 2 X
                                                 ? a ^ s g
                                                 8 E{" 8-3
                                                     ! •= c
                                                *w gj 00  £•

                                                S2° s =
                                                c b * -3 s
                                                S o J W °
                                                t/3 2 -* JJ '5 £-<
                                                ja *!! I ti ^
                                                " TJ ^ 8 H §
                                                3 5 E "S E —

    Recommendations for future chemical mechanism development, documentation,
and testing intended to extend the present urban-oxidant-only mechanisms to the
new areas that will confront EPA in the coming years are also given below.

General Approach
Guidelines for Mechanism Development and Testing

General Approach
Demerjian  described what most workshop participants accepted as a general ap-
proach to  mechanism development and testing.  The major components of this
approach are shown in Figure 4. Within this generally accepted approach, however,
there are differences in the practice among different modeling groups.

    The mechanism development process begins with the assembly of reactions and
rate parameters which are obtained from a consideration of chemical and ther-
modynamic theory, the chemical literature, recommendations and evaluations for
the simpler species, and from consideration of the available literature data for the
more complex portions of the  overall mechanism. This step is subject to several
     o not  all rate data or mechanistic  pathways of interest  are available  in  the
      literature, and
     o laboratory kinetic measurements must often be performed at conditions sig-
      nificantly different from those in the atmosphere, thus  frequently requiring
      extrapolation and introducing uncertainty, and
     o important competing processes may not have been present in the laboratory
      kinetics measurement system leading to either unintentionally ignoring im-
      portant reactions or, where this problem  is recognized, requiring the use of
      estimation methods to account for their absence.
Therefore, throughout the mechanism development process some form of testing is
used to determine the importance of inclusion or exclusion of a given reaction and to
assess the sensitivity of the mechanism's predictions to the choice of rate parameters
which have some uncertainty  either due to  the use of estimation  techniques  or
to experimental measurement  uncertainty. In addition, some form of testing for
completeness of description must be performed.

    For most model developers, these tests have included the simulation of smog
chamber experiments.  Dunker stated that,  "No mechanism developed entirely from
theory and  the literature is satisfactory without testing with chamber or captive air
bag data."  Such data, however, also have problems with regard to model testing.
Some of these are:

                                                General Approach
 Laboratory  Data
  Smog Chamber
      Use  in
 Testing against

Ambient Air Data
 for Consistency
Figure 4.  Schematic method for development and evaluation of

   chemical mechanisms. After Kenneth Demerjian.

Mechanism Evaluation Procedures
     o  identifying and quantifying the chamber conditions that influence the chem-
       ical transformation processes, and
     o  coping with the sparsity and quality of measurements of reactant species and
       reaction products.
In the first category are the magnitude and spectral distribution of the photolytic
rates in the chamber, the temperature, humidity, and mixing characteristics of the
chamber, the magnitude and production  mechanisms of chamber radical sources,
and chamber walls as sinks for intermediate and final products (e.g., HNO3, h^Oj).
In the second category are quality control problems, lack of analytical techniques
that are both highly sensitive and specific for important chemical species. Some of
the most important species, such as HO and HO2, for example, can not be measured
in chambers. Also there are differences among the measurements for intermediate
and very important species such as formaldehyde.

    Both sets of problems, the incomplete kinetics  data base problems and the
chamber effects and measurements problems, limit the extent to which each source
of knowledge is expected  to agree with the the other.  This expected extent of
agreement ultimately causes problems. As Demerjian said, "Some modelers like to
have the lines for the model predictions fit every point  from the chamber, while
others are pleased if they are just close."

    Chemical mechanisms that have enough detail to describe the features of sev-
eral sets of smog chamber data are too detailed to be used in air quality models.
Therefore generalization, distortion,  and deletion processes  are applied to these
more complex models to produce a more practically sized mechanism for use in the
air  quality models without sacrificing too much accuracy of prediction.  This  step
is generally described as producing a  "condensed chemical mechanism."

    Finally, Demerjian expressed a need to "test against ambient air data for consis-
tency."  This step is fraught with problems and was seen by many at the workshop
as being ambiguous in terms of decision making about the chemical mechanisms.
That is, because so  many other factors could be responsible for the  disagreements
between ambient air measurements and model predictions, that assignment of any
part of the error to the chemical mechanism would be very difficult, if not impos-
sible. But it was stated by Whitten that, "If the models and data disagree in the
ambient air, then the modeler ought to be required to consider explanations  that
at least include problems with the chemical mechanism."

Mechanism Evaluation Procedures
This section reviews some specific procedures for mechanism development and evalu-


                                                     Mechanism Evaluation Procedures
ation that were presented by Lurmann, Gery, Dunker, Carter, Whitten, and McRae.
There was general agreement that
     o mechanism testing should be performed according to a "hierarchy of species"
      as shown in Figure 5,
     o data from at least two, and preferably more, chambers should be used in the
     o testing should include at least two phases:
          >  testing and refining the representations of chamber dependent  phe-
             nomena, and
          >  testing and refining the complex organic species and mixtures of species,

     o as much data as is available should be used in the testing, and at at minimum,
      15 experiments should be used in the first phase and 50 experiments in the
      second phase for each chamber.

    In the first testing phase,  the so-called "core" mechanism or the "inorganic
and carbonyl reaction set,"  which includes the reactions for  the lower hierarchy
species (i.e.,  the inorganic species, carbon monoxide (CO), formaldehyde (HCHO),
acetaldehyde  (CHsCHO) and PAN) should be tested with the following types of ex-
     o NOx-air,
     o NOx-CO-air,
     o HCHO-air,
     o NOx-HCHO-air,
     o NOx-HCHO-CO-air,
     o CH3CHO-air,
     o NOx-CH3CHO-air,
     o NOx-CH3CHO-CO-air.
Because the core  mechanism is generally based on widely accepted recommenda-
tions, such as the NASA and IUPAC evaluations and reviews, the primary purpose
of this phase is not to test the adequacy of the chemistry. Rather,  the purpose
of this phase  of testing is to test and refine the representations of chamber depen-
dent phenomena such as solar radiation intensity and spectral distribution, and
wall processes including ozone destruction, nitrous acid  (MONO) production from
NO2, nitric acid (HNO3) production from NjOs, and possible NO, NO2, HCHO, HONO,


Mgchanitm Evaluation Procedures
             JOLEFINS |      I AROMATICS
         - ECHO (R^H)   - Dicarbonyls
              HCHO  HCO
                   • ICQ
         NO N02 N03 N2O6
         HONO HNO, HO' HOj H2O2
Higher Caxbonyls
 PAN Compounds
                                                  Lowest Level
     Figure 5. Hierarchy of Species

                                                     Mechanism Evaluation Procedures
and  HO  production from chamber-related processes.  There is agreement that a
uniform treatment of photolytic processes should be used. That is, the photolytic
rates should be derived from species cross-sections and quantum yields and from
the spectral distribution of the radiation sources in the different chambers. To the
extent  that the processes are understood, a uniform treatment should also be used
for chamber sources and sinks of reactants. Data from each chamber should be used
to determine the parameters for the descriptions of these processes. In some cases,
knowledge of prior chamber use may be important in  determining the stability of
values  of parameters.

    In  the second phase, the mechanism's performance against more complex hy-
drocarbon (HC)-NOx-air experiments is tested.  The mechanism's reactions for other
carbonyls, alkanes, alkenes, and aromatic species are first evaluated independently
by testing against experiments with single organic compounds, and parameteriza-
tion  of uncertain reaction rate constants and product yields is generally carried out
using these single HC-NOX experiments. The mechanism's performance for complex
mixtures of organics with NOX is then evaluated using known mixtures with 2 to
20 compounds. Additional evaluation using experiments with automobile exhaust
and with urban air (i.e., captive air experiments) are recommended because these
mixtures are most representative of ambient hydrocarbon mixtures. For these real
very complex mixtures, however,  analysis frequently results in a  5  to 20%C un-
known hydrocarbon fraction, and therefore tests using  such experiments may yield
ambiguous results depending upon the speciation of the unknown HC.

   As  indicated in the general discussion above, assessing goodness-of-fit between
the experimental data and the model predictions is a topic that requires further
work and one which the  Steering Committee recommends the creation of a Task
Group. Problems that arise in measuring model performance are:
     o  the large number of experiments and individual  measurements, (i.e., having
       to manipulate large quantities of information),
     o  both the time-to-events and the magnitude of results are important,
     o  the actual path taken to a point is  important, (i.e.,  the  "shape of the
     o  some species concentrations depend upon the ratio of two other events rather
       than the absolute levels, thus seemingly correct predictions can arise for the
       wrong reasons,
     o  in some cases small errors in representing chamber characteristics or temporal
       changes in parameters can cause large errors in  predictions at the very end
       of the simulation, but not for most of the experiment (e.g., see Figure 2a in


Mechanism Evaluation Procedures
      which the final 03 is over-predicted in the last 60 minutes, yet the model's
      predictions agreed well with all data up to that point),
     o uncertainties in experimental measurements, especially the initial conditions,
      can lead to apparently poor predictions,
     o lack of correspondence between the surrogate species in the mechanism and
      the experimentally measured species makes comparison difficult (i.e., PAR in
      CBM or chemiluminescent NOj measurements as in Figures 2 a and b),
     o inaccuracies at very low concentrations and sample line losses or conversions
      make some comparisons difficult,
     o for outdoor experiments, the sky condition or cloud cover condition, which
      may not be well represented in the model's photolytic rates, may be a major
      source of errors in the prediction.

    As an example of one approach to the goodness-of-fit problem, the developers
of the SAPRC/ERT mechanism used simple statistical measures to describe the
goodness-of-fit for their model. These included:
     o for the NOx-air experiments, plots of observed and predicted total change in
      NO and N02,
     o for experiments that produced O3, scatter diagrams of observed and predicted
      maximum Os, and an average rate of production, 
                                                 Development of Condensed Mechanisms
for example, the typical reader simply can not comprehend the detail presented in
so many graphs. No generalization of the information on the graphs is attempted
in the SAI approach.  Thus, the SAPRC/ERT approach tends to flatten all the
dimensions of the problem to a  few, while the SAI approach maintains all of the
nearly overwhelming dimensionality and does little to provide an overview. Hence,
some members of the Steering Committee assert that  further work is required for
demonstrating goodness-of-fit.

Development of Condensed Mechanisms
In general, the detailed chemical mechanism, developed and evaluated as discussed
above, must be condensed for use in air quality simulation models for control strat-
egy assessment purposes. It is strongly recommended that the condensed chemical
mechanism be derived from the final tested version of the detailed mechanism with
testing and comparison against the detailed mechanism at each step of the conden-
sation process.  Procedures for condensing mechanisms and testing the condensed
mechanism against simulations of the expanded  or explicit mechanisms were dis-
cussed in three reports: Whitten, Johnson, and Killus ("Development of a Chemical
Kinetic Mechanism for the U.S. EPA Regional Oxidant Model," EPA-600/3-85-026,
1985); Whitten and Gery  ("Development of CBM-X  Mechanisms for Urban and
Regional AQSMs," EPA-600/3-86-012, 1986); and Lurmann, Carter, and Coyner
("A  Surrogate Species Chemical Mechanism for  Urban-scale Air Quality Simula-
tion  Models, Adaptation of the Mechanism," EPA-600/3-86-031, 1986).

   The condensing and testing procedures are intended to minimize compensating
errors, minimize the number of species simulated, and yet retain the most important
features of the original mechanism. A progressive approach is recommended that
actually provides a series of condensed mechanisms, the final mechanism being the
most condensed.  Each of the mechanisms in the series would relate to  the less
condensed versions above it and thus could not be easily modified independently.
Thus, changes or updates should be first introduced into the detailed or explicit  (i.e.,
the  first) mechanism, and any adjustments or modifications should be propagated
down through the various condensed mechanisms.

   Specific methods  for mechanism condensation include eliminations of species
that are unreactive or do  not change  in concentration during the reactions and
thus should not change the numerical results of simulations using the intermediate
condensed mechanism when compared to the detailed mechanism. Species  that
satisfy the latter are M for  pressure-dependent reactions and O2 when the pressure
or altitude are constant. These species can then be incorporated into the reaction
parameters. For some applications, water may also be treated in this manner, but


Mechanism Documentation
frequently water varies in the experimental situation and must be treated explicitly.
An example of unreactive species are product species which are of no interest and
for which the mechanism contains no further reactions, for example H2, CO2, and
formic acid.

    Other steps include the use of fractional stoichiometric coefficients and interme-
diate species which have unimolecular branched pathways of reaction. As long as the
unimolecular decay lifetimes are reasonably short, the numerical simulation results
will not be significantly affected. Tests are required for lifetimes longer than a few
minutes. Other techniques include the use of a  "mass balance" or "counter species"
which are given arbitrarily high decay rates to form "intentional" steady-state-like
species.  For example, several versions of the Carbon Bond chemistry employ a
mass balance species usually called X, which rapidly removes the species  PAR.  Such
species can be eliminated by  the use of negative stoichiometric coefficients  (e.g.,
— 1PAR). Both mechanisms use a generalized RO2 species to act as the sum of all
alkylperoxy radicals in radical-radical termination steps, thus eliminating a  large
number of reactions while still approximately accounting for their effects with only
two or three reactions per radical.

    Condensation steps beyond those above involve assumptions about typical at-
mospheric situations, and the condensed versions of the mechanism must be tested
to determine the bounds of such assumptions.  Techniques have been developed to
identify unimportant reactions and species.  Elimination of unimportant reactions
provides little benefit to simulation costs, because the costs are most sensitive to
the number of species. Of course, fewer reactions do make a  mechanism easier to

Mechanism Documentation
It was an accepted fact among the Steering Committee that clear and comprehensive
documentation of chemical mechanisms by  their developers  is needed  to ensure
their proper use. At this Workshop because of other  pressing issues, little time
was spent discussing what "documentation  of chemical mechanisms" means.  In
his presentation, Joseph Tikvart reviewed  and strongly supported the approach
proposed in the workshop background document  "Need for Chemical Mechanism
Documentation," by Sexton and Jeffries. In this document an example  outline of
a guidance document for the application of a chemical mechanisms was  proposed.
In his writing for the Steering Committee after the workshop, Fred Lurmann also
described several items that need to be documented in a mechanism.  These are
given in Table 1.


                                                         Mechanism Documentation
            Table 1.  Items Needed to Document a Mechanism.

 1. A name, version number, and date.
 2. A listing of the reactions and species.
 3. A listing of all species  assumed to be in steady-state, including the  diag-
    nostic equations, solution method for computing steady-state species con-
    centrations, and discussion of the range of applicability (i.e., temperature,
    pressure, concentration, etc.).
 4. A listing of the rate constants and procedures to calculate their dependence
    on temperature, pressure, and actinic radiation.
 5. A listing of absorption  cross-sections and quantum yields as a function of
    wavelength including specifying averaging intervals and method  of integra-
    tion between points for  all photolytic species.
 6. The values of all photolytic reaction rates should be given at 10 degree zenith
    angle increments for a standard set of actinic fluxes such as those reported by
    Peterson in 1976 for the earth's surface with a standard estimate of albedo.
 7. A complete listing of references to the sources of the kinetic and mechanistic
    data for each reaction and the actual form and value of the literature rate ex-
    pression should be given (i.e., 6.1 x 10~13(T/Tref)~*s  cc-molecules^-sec"1).
 8. For use in  computer codes that only accept Arrhenius expressions for rate
    constants, the non-standard form of rate constants (including fall off expres-
    sions) should be restated in Arrhenius rate form in the units typically used
    in the mechanism (e.g.,  ppm~1-min~1) and the value at 760 mmHg and 298
    K should be given.
 9. The actual values of, or formulas for, all assumed concentrations of reactants,
    such as M, O2, and H2O that have been incorporated into rate constants should
    be given.
10. Clear  identification of any differences in reactions and rate  constants for
    modeling environmental chambers and modeling the ambient air.

 continued on next page...

Mechanism Documentation
           Table 1, cont. Items Needed to Document a Mechanism.

   11. Complete documentation on the hydrocarbon representation system, includ-
         a) describing the basic approach,
         b) describing the number of carbons in each lumped organic species,
         c) rules for converting organic mixtures into mechanism input,
         d) tables listing the assignment of commonly occurring compounds to
            lumped classes,
         e) complete examples of the conversion of detailed organic mixtures to
            lumped mixtures.
   12. An address where "official" versions of the mechanism can be obtained.

                                                          Mechanism Documentation
    Implementation of a chemical mechanism by non-developers is often difficult
not only because of deficiencies in documentation, but also because of software and
computer differences.  Computed solutions to mechanism test problems are essential
to proper implementation of chemical mechanisms by non-developer users.  These
test problems should be designed to test all reactions in the mechanism. The Steer-
ing Committee believes that a minimum of four test problems are needed.  These
initial test  problems should not involve dilution, entrainment, emissions injection,
or deposition, but rather be examples of the pure chemical  kinetics. Ideally, the
solutions should be computed using a high quality algorithm such as the Gear algo-
rithm with tight error control. In these test cases, the algorithm used, the method
of computing the  Jocobian, the use of absolute or relative  error tolerances, and
the minimum, initial, and maximum integrator step sizes should be documented.
All species should be  integrated  rather than determined from the steady-state as-
sumptions. The first two test problems should employ constant light intensity and
temperature and one should be at a low HC-to-NOx ratio while the other should be
at a high ratio. The  second two problems should employ diurnally varying light
intensity, spectral distribution, and temperature, again with a low and high  HC-to-
NOx ratio.  The low HC-to-NOx ratio test should be designed  to test the nighttime
chemistry also.

   All inputs and methods used in the test problems should be fully documented.
It is especially important to document the method (s) of updating the time-varying
parameters. Frequent output intervals should be used and the output should also
include the values of the time varying parameters.

   A final documentation item discussed by the Steering  Committee was the need
for a standard set of conditions for comparing mechanism prediction of VOC control
requirements when using the OZIPM program.  It was suggested that the regulatory
groups in EPA should be involved in developing these test cases to insure that they
cover typical cases of interest for regulatory purposes.

Differences Among Well-Tested Mechanisms
Differences Among Well-Tested Mechanisms

When the guidelines  for chemical mechanism development described in previous
section are followed by multiple research groups, it is expected that large sections
of the mechanisms developed will be very similar, if not identical. This will arise
because of a) ongoing NASA and IUPAC evaluations for the inorganic and small
(< Ca) organic species reactions (these have changed little in recent years),  and
b) because certain reactions and reaction sequences for the more complex organic
compounds in the detailed  mechanism are sufficiently supported by kinetic  and
mechanistic data that little uncertainty is associated with their magnitudes  and
structure.  The portions of the detailed chemical mechanism, however, which are
either unknown or only poorly understood will have to be assembled using estima-
tions or arguments by analogy, or perhaps simply be parameterized. Thus, different
methods of representing these unknown or poorly known sections of the chemistry
will arise in different mechanism developments.

    The latter process will most likely lead to detailed (and subsequently condensed)
overall mechanisms which, based on past experience, may differ  in  their control
strategy predictions, despite the fact that each chemical mechanism may be con-
sistent within the bounds of reasonable agreement with the elementary reaction
kinetics,  mechanisms, and products databases and with environmental  chamber
data. Other than further efforts to refine the measure of uncertainty in the both
the kinetics and chamber data and new formulations and testing of the mechanisms,
there may well be no scientific reason to accept or reject one of these chemical mech-
anisms over the other.

    Given that EPA must proceed with applications of the mechanisms and given
that the existing data might not allow a further reduction of the uncertainties among
the mechanisms offered as an explanations of the transformation processes, it  is
recommended that the EPA not select only a  single mechanism for making control
strategy predictions. Without further experimental data, such a choice would have
to be made on other than a scientific basis. The use of even two mechanisms  will
suggest how the present uncertainties in the experimental databases translate  into
uncertainties in predicted control strategies.  EPA must make the  final decision of
which strategy to use, and the uncertainty in the predicted control requirements
along with other  relevant societal values and issues should be considered in  this

    While the need for regulatory consistency,  and therefore a desire to have as long
a time between changes in the tools used to make control decisions, is recognized,


                                                Differences Among Well-Tested Mechanisms
our past experience has shown that sufficient changes occur in theoretical ideas and
in experimental data in even a two year period as to result in potentially significant
reductions in  the  uncertainties of control predictions.   A failure to account for
these changes in a timely manner could  result in the view that the EPA is not
using the best science.  This  problem may be exacerbated by the fact that the
EPA may be sponsoring some of the  "best" science being performed in this  area.
Therefore, the workshop participants  recommend that EPA create a set of review
processes,  coordinated to the regulatory  time line, that would use a base  time
period of two years for  assessing the quality of control strategy predictions.  A
major scientific finding, such as a breakthrough in understanding aromatic reaction
product mechanisms, may require a more rapid review and response. The exchange
of detailed information in these review meetings is itself expected to accelerate the
rate of improvement in the  measure of agreement between  models and data and
therefore to improve the  certainty for  control strategy predictions.

Talk or Review Groups Needed
Task or Review Groups Needed

The workshop participants recommended that  several interacting sets of review
processes  or review groups be established, each with a goal of assessing and docu-
menting the state of knowledge and degree of reasonable agreement to be expected
in a given domain. At least four different domains were identified for such activi-
    1) kinetic and mechanistic data needed in constructing photochemical transfor-
      mation mechanisms;
    2) environmental chamber data needed for comparison with mechanism predic-
    3) mechanism intercomparisons tests; and
    4) user or application tests.
The first  domain not only  includes the traditional review process as exemplified
by  the Atkinson and Lloyd 1984 review, it also includes statements about what
components of models are expected to look alike and where models might differ due
to lack of convincing evidence.

    The second domain includes all sources of information that influences judgments
about the extent to which experimental chamber data contribute to the uncertainty
in the agreements with models.   This includes  the type of information found in
QA documents, such  as calibration sources, histories, intercomparisons, and special
tests.  It  includes agreement on  what constitutes chamber  characterization and
would involve an assessment on  the state of characterization of the chambers in
current  use and would include historical trends  or adjustments needed  to recover
or use older data.  It also includes assessment on  the extent  to which the various
environmental chambers in use agree or disagree on various measures of consistency,
e.g., formaldehyde yields from ethylene experiments. An expected product of this
group would be  the  identification and documentation in a uniform format of at
least 100  high quality chamber experiments that  could be used by the modeling
community in general, or by EPA, to intercompare chemical mechanisms offered as
explanations of the smog transformation process.  It should be clearly recognized
that such a data set  would be  seen as merely a necessary test, not as  a sufficient
test. That is, if mechanism predictions do not reasonably agree with these data,
then the mechanism is  not satisfactory for use in urban applications, but such
agreement does not  imply  that a given mechanism would result  in a particular
set  of control requirements in atmospheric applications.  Furthermore, the data
set  presently available is not suitable for testing the predictions of the final acid


                                                        Talk or Review Groups Needed
generation products such as hydrogen peroxide and nitric acid, although it can
most likely test the production of precursors to these species, e.g., formaldehyde.

    The third domain encompasses those activities associated with merging infor-
mation from the first two domains. It includes reaching agreements on the processes
for describing how the authors actually constructed the particular mechanism, and
how the choices in the areas exhibiting uncertainties were made.  This domain in-
cludes clearly identifying the experiments used to develop the model (i.e., those
that may  have resulted in changes in model content) and experiments used to test
the model (i.e., ones used to assess accuracy of the predictions). This domain also
includes defining what constitutes fit between predictions and data, e.g., how may
species must  be fitted, the extent  of expected agreement expected given the uncer-
tainties caused  by uncertainties in rate constants and mechanism components and
the uncertainties  in the data as described by the assessment group in the second
domain.  Activities in this domain also include the identification and comparison
of the parts of both explicit  and compressed mechanisms which  are different as the
result of differing  interpretations in the mechanism development process. Activities
to design  experiments, either laboratory kinetic or environmental chamber, which
have some potential to discriminate among the representations are included in this
area.  Other efforts to identify additional constraints on the comparison of model
and data  are included also, for example, the use of dual chamber data to test two
different regimes  at the same time. Mechanism authors may be asked to provide
documents that describe in  detail why they  believe that their  own mechanism is
superior to other  mechanisms.

    The fourth domain includes  understanding prediction differences among  the
models, that  is, understanding causes of the differences when there are no compar-
ison empirical data.  This domain includes tests of predictions using atmospheric
data,  the  identification of critical points in  each  mechanism which result  in pre-
dictions being different, the establishment of the sensitivity of the predictions to
various input choices, and the sensitivity of the predictions to "correction" or scal-
ing techniques as  examples.  This activity should also establish a series of test cases
which all  mechanisms for use by EPA must perform.  Significant  or meaningful
measures  of performance other than control requirements should be identified.

    The Steering  Committee suggests  that  EPA's role in these review activities
should be one of support: providing funds to hold meetings, produce and dissemi-
nate documents. It should be further stressed, however, that each group is likely to
make strong recommendations for needed work to  resolve uncertainties further and
to improve the  science.  In this case, EPA's role, especially in the area of environ-


Ta«k or Review Groups Needed
mental chamber data production is seen as vital. No other governmental agency is
supporting work in this area and EPA is seen as the major client for such data.

                                                        Short-Term Recommendations
Short-Term Recommendations

Task Force on Chamber Light and Wall Process
The EPA has presently in hand two state-of-the-art chemical mechanisms for use in
EKMA in the upcoming SIPs revisions. Each of these mechanisms has been tested
against a large set of environmental chamber data, and the resulting mechanisms
are very similar in many respects.  There are several areas, however, in  which,
because  of uncertainty, or lack of data,  the model developers  diverged in their
mechanism construction. There are new experimental data that can help resolve
some of the uncertainties and other differences are perhaps resolvable by direct
.intercomparison of the modelers techniques in a single setting. Thus, the Steering
Committee believes that it may be possible to further refine these mechanisms on
a short-time scale and without the need for any new experimental data.  While
this advancement needs no new experimental data, recommendations  for further
well-defined research may result from this effort.

    The first priority is to use the presently available environmental chamber data
base to refine the testing of present chemical mechanisms.

    The  Steering Committee therefore recommends the establishment of  a task
force of non-EPA scientists to evaluate and resolve chamber effects and light in-
tensity/spectral distribution issues.

    The first step will be to use new light intensity and spectral distribution data
recently obtained at the UNC chamber to retrospectively better  define this highly
important chamber characteristic for the UNC chamber. In addition, similar spec-
tral measurements should be made in the UCR Outdoor Teflon  Chamber so that
a consistent treatment can be applied to all outdoor chamber data. Likewise, an
intercomparison of spectral measurements using the UNC spectroradiometer at the
UCR chamber facilities would be useful in ensuring a uniform treatment of pho-
tolytic rates in all of these chambers.

    After agreement has been reached on how to treat the photolytic rates in all of
the chambers in use, the next step would be for the group of modelers and chamber
operators to review and resolve the remaining divergent opinions concerning the
best methods to represent chamber effects.

    The Steering Committee believes that the best way to resolve these differences
is to have a modeling session involving both modeling groups and  both chamber
groups at a single location. The group, for example, could meet at the University of


Task Force on Chamber Light and Wall Process
California, Riverside facility, where both of the recent mechanisms would be running
on computer facilities that accessed the basic chamber characterization experimen-
tal data for all chambers.  This would allow the various proposals to be immediately
tested and discussed. In  addition to the modelers and experimental scientists in-
volved, a knowledgeable and impartial scientist in the atmospheric chemistry field
must be appointed to act  as moderator and ultimate decision maker.

    This group must first concur on the appropriate photolytic rate and spectral
distributions to use  in all chambers, especially the UNC outdoor chamber. A re-
modeling of the chamber characterization data bases must then be performed using
the different methods for  treating all chamber's wall processes.  These tests must
    a) chamber control  and characterization experiments, e.g., NOx-CO-air irradia-
   b) NOx-formaldehyde-air irradiations,
    c) NOx-acetaldehyde-air irradiations,
   d) NOx-simple organic compound-air irradiations,
    e) NOx-multi-organic mixture-air irradiations in order of increasing complexity.

    Where  side-to-side test data exists, both sides  must be modeled with consis-
tent assumptions  and  simultaneously  compared.  In addition,  recommendations
will be made concerning  a limited amount of experimental environmental cham-
ber data which would lead to further significant decreases in uncertainties concern-
ing chamber effects.  These experimental data will probably involve monitoring of
formaldehyde, using analytical techniques capable of monitoring formaldehyde in
sub-parts-per-billion levels, during NOx-air, NOx-CO-air, and NOx-propene-n-butane-
air irradiations in the UNC and various UCR chambers.

    If it is possible, all parties will agree on a single best representation of photolytic
rates  and chamber wall  processes for all chambers, or they will agree and describe
the evidence that  supports different representations and will suggest experiments
or measurements needed to support or refute the evidence offered.  As a result of
this re-analysis, the  respective  chemical mechanisms will almost certainly require
some refinements and adjustments, possibly leading to a narrowing of the differences
between the mechanisms with regard to their control strategy predictions.

   The expected benefits from this task force include:  a decrease  in the uncertainty
associated with the chamber data  bases;  an improvement in the representation of


                                             Task Force on Chamber Light and Wall Process
chamber wall processes in both EPA-sponsored models; and a significant increase
in the confidence of the predictions of both models.

    The time required for this work is about one month of initial preparation and
about one-week of meeting, followed by about two-weeks of analysis and documen-

Task Force To Select Environmental Chamber Data for Testing
Task Force To Select Environmental Chamber Data for Testing
One of the goals of this workshop was to identify a standard data base that would
be useful in distinguishing among different  mechanisms. The Steering Committee
concluded that it was  possible to assemble a "necessary" environmental chamber
data set, that is, one containing chamber experiments which all mechanisms for use
in urban areas would have to simulate.  The data base is described as "necessary"
because if a mechanism could not successfully simulate these experiments it most
likely would be considered unsatisfactory for EPA applications, but the data base
would not be "sufficient" to resolve all questions associated with oxidant prediction.
Mechanisms that could successfully simulate all experiments in the "necessary" data
base might still give different predictions in  urban atmospheric applications. Other
recommendations given below  address this situation.

    The Steering Committee recognized that to assemble such a database would
require significant input from chamber operators at both UNC and UCR and from
model developers at  SAI and UCR, as well as other interested parties who might
wish to use the data. Furthermore, such a data base could not be assembled until
the task force on chamber light and wall processes had finished its work.

    It is envisioned that the data base would contain about 100 experiments from
several chambers.  The data would be maintained in a uniform format. Each ex-
periment would contain detailed  recommendations and supporting information on
photolytic rates, chamber wall processes, and initial and temporal conditions. These
would be the result of review, discussion, and consensus among task force members.

    The benefits expected from this effort are a set of consensus data needed to as-
sure a minimum level of performance among  all mechanisms that may be potentially
used by EPA or by state agencies in meeting EPA requirements.

    The time required to accomplish this task is at least one year and would require
funding for individuals at both chamber facilities and at the modeling groups, as
well as for several working meetings of members of the task force group.

                                                       Longer Term Recommendations
Longer Term Recommendations

Establishment of Review Groups
 Review Group for Evaluation of Fundamental Kinetic and
     Mechanistic Data For Use in Model Development
It is evident that a critical necessity for present and future chemical mechanism
development is that a continual effort be initiated to critically review and evaluate
the kinetics, mechanisms, and products under atmospheric conditions for organic
compounds. Because of the existence of the NASA  and IUPAC evaluations panels
dealing, respectively, with the stratospheric and tropospheric reactions of inorganic
species and of the up to organics (methane, ethane, propane, ethene, propene, and
acetylene and  their degradation products  in the presently ongoing IUPAC evalu-
ation), this evaluation effort will not need to duplicate these present evaluations,
but rather focus entirely on the atmospheric chemistry of the greater  than Ca or-
ganics. This ongoing evaluation, together with  the NASA and IUPAC evaluations,
will provide the "recommended" reaction list for use in detailed chemical mecha-
nism development. This evaluation will identify the chemical reactions and reaction
schemes which are unknown or only poorly understood, as carried out  in the 1984
Atkinson and  Lloyd review article, and will provide the impetus for needed labo-
ratory experiments. It must be pointed out that this evaluation is extremely cost
and time effective, avoiding the necessity for each chemical mechanism developer to
independently review and evaluate the huge amount of literature available.

   It is recommended that this evaluation effort be carried out in loose cooperation
with the ongoing IUPAC evaluation.  A 5-6 member review team is recommended,
including at least one international expert in this area. To facilitate literature re-
trieval, close cooperation with the IUPAC panel with respect to publication must be
carried out, and the entire evaluation must be readily available through a computer
database system. Dr. Roger Atkinson, who is a  present member of the IUPAC eval-
uation panel and senior author of the 1984 Atkinson and Lloyd review  article, is a
potential chairman of the EPA-funded panel.  Publication and up-dates should be
carried out at  two-year intervals with the computer data base being updated more
often for use by the tropospheric modeling community.

   This effort could begin within one year and operate at an estimated yearly fund-
ing level of approximately $100,000 to $150,000, including publication and travel

Establishment of Review Groups
Review Group For Evaluation of Environmental Chamber Data
This standing review group would be an out-growth of the Task Force to select the
standard or "necessary" data base that was described under "short-term recommen-
dations" above. Its role would be to review and make recommendations with regard
to the extent of "reasonable agreement" among various environmental chamber re-
sults and to make recommendations for additions or deletions from the standard
data base.  This effort could begin  within one year and operate at an estimated
yearly funding level of approximately $50,000 - $100,000 including publication and
travel costs.

Review Group for Mechanism Intercomparison
This group would be the out-growth of the Task Force on chamber photolytic rates
and wall effects that was  described under "short-term recommendations" above.
The full domain of this group was described above; its role would be to formally
compare and resolve differences among different  mechanisms in comparison  with
chamber data.

   This effort could begin within one year and operate at an estimated yearly
funding level similar to that of the Evaluation of Environmental Chamber review
group above.

Review Group for Mechanism Predictions in Applications.
This group would provide scientific support for activities presently conducted by
OAQPS staff. The full domain of this group  was described above; its  role would
be to understand and explain differences in predictions from mechanisms in urban
control strategy calculations.

                                                Needs for Future Mechanism Development
Needs for Future Mechanism Development

D.ata Needs Concerning the Atmospheric Chemistry of Organics

Presently the further scientific refinement of chemical mechanisms is hindered by our
lack of knowledge concerning several key areas of the chemical reactions occurring
in the atmospheric degradation of certain classes of organics. While a detailed set
of data needs will emerge from the data evaluation panel, it is abundantly clear that
key areas of uncertainty involve the aromatic hydrocarbon chemistry, the chemistry
of the  larger (>Ce) alkanes, and aspects of the ozone-alkene reactions.  Further
laboratory  product (and kinetic) data need to be  obtained for these compounds
before more accurate chemical mechanisms can be developed.

    Specifically, the most important priorities are:
    a) to determine the fates of the HO -aromatic adducts under atmospheric condi-
      tions. This includes the determination of the fraction of these HO -aromatic
      adducts which result in the formation of phenolic compounds, and the elu-
      cidation of the ring-opened degradation products.
    b) to determine the reactions of the >Q alkoxy and alkylperoxy radicals under
      atmospheric conditions and the subsequent reactions of their products.
    c) to determine the radicals formed, and  their yields, from ozone-alkene reac-
      tions under atmospheric conditions.

    In addition, there is an urgent need for absorption cross-section and photodis-
sociation quantum yields and product data for the carbonyl compounds formed as
intermediate products in the degradation schemes of organics. Immediate candidate
carbonyls are:
    > methyl ethyl ketone
    o propionaldehyde
    !> acetone
    > glyoxal
    > methylglyoxal

    These laboratory studies must be coordinated with other ongoing or proposed
efforts, for example the NSF/NASA "Global Tropospheric Program" and NSF At-
mospheric Sciences Division projects, to avoid and unnecessary duplication of effort.


Needs for Future Mechanism Development
Needed Environmental Chamber Data
It is evident that for any future advances in the accuracy and breadth of a data
base upon which detailed chemical mechanisms will be  tested, new high quality
environmental chamber data will be required. This need will arise from a number
of reasons:
    a)  The need to discriminate between present divergence approaches to repre-
       senting chamber effects-i.e., HCHO and HOMO ofFgassing verses heterogeneous
       formation of HO" radicals (probably via the intermediate formation of MONO).
       Because of the relatively slow rate of HCHO photodissociation, measurement
       fo HCHO with sub-ppb sensitivity in the environmental chambers used will
       resolve this divergent opinion.
    b)  The need for high quality environmental chamber data, together with corre-
       spondingly high quality chamber effects (ofFgassing rates, deposition rates,
       hydrolysis rates, photolysis intensity and spectral distribution, etc.] to pro-
       vide more stringent tests of detailed chemical mechanisms.
    c)  The need for data concerning the production of acidic species (HNO3, H2SO4)
       and/or of their precursors (HjOj, HOj, organic acids, organic hydroperoxides)
       for testing of acid deposition models. In addition, experimental data obtained
       under low  NOX conditions will  also be needed,  especially for more regional
       and tropospheric model testing.

    Concurrent with these data collection efforts is the need for continued develop-
ment and use of analytical techniques to detect and routinely monitor key labile
species such as HONO, HCHO, H2O2, NO2  (not as  NOX-  NO),  HMOs, etc. Analytical
measurement development must be  viewed as an integral part of this effort.

Studies of Chamber-Dependent Effects
It is clear that at present a major uncertainty in the  use of environmental cham-
ber data for testing of chemical mechanisms concerns chamber-dependent effects.
In particular, the role of chamber walls in the production of radicals  during irra-
diations, either from contaminant offgassing or from the heterogeneous formation
of such species as HONO from NO2, with subsequent photolysis to yield radicals, is
an area of much uncertainty.  While  for the short and median terms  a more ac-
curate parameterization of these effects and their magnitudes is required,  on the
longer time scale fundamental studies of these heterogeneous chamber  wall effects
are needed to elucidate the chemistry/physics of these processes and to investigate
the optimum surface materials for use in future environmental chambers.  These
studies will  require the involvement of scientists of other  disciplines, for example,


                                                 Needs for Future Mechanism Development
surface chemists and physicists and experts in catalysts systems, as well as material

    This area of research will be crucial to the development of future environmental
chamber facilities for the production of high quality data for the testing and devel-
opment of chemical mechanisms for, for example, acid deposition and long range
transport  applications where experimental data concerning hydrogen peroxide, or-
ganic hydroperoxides, organic acids, and nitric acid are required.

Investigation of the Applicability of the present EKMA Method
When the EKMA method is used with different chemical mechanisms, it has been
common that each mechanism predicts a different control strategy.   While it  is
possible that all of the  difference in predicted control requirement are caused by
differences in the chemical mechanisms themselves, it often appears that the control
requirements are more sensitive than the  predictions of absolute levels of 0$, for
example.  It is possible that the "scaling" process the EKMA technique  requires
when the whole OZIPM model and its inputs  do not correctly predict the observed
Os might be responsible, at least in part,  for some of the differences  in predicted
control requirements in this situation.

    This might be so because in the present EKMA method all of the uncertainty
in the model's fit to the world is placed in the absolute emissions  level: it is the
emissions  that are adjusted to get a  "fit" between the  observed 0$ and the pre-
dicted O3. (Actually, in OZIPM, the emissions are expressed relative to the initial
concentrations in the starting  box, and the initial concentrations are  increased or
reduced to obtain the observed Os level. This is equivalent to adjusting the total HC
mass in the simulation, and therefore the absolute emissions.)  The applicability of
this technique has not been demonstrated with modern  mechanisms. The original
support for this approach was based upon the now obsolete Dodge Mechanism, and
essentially every rate  constant or species mechanistic pathway has changed since
that study was performed.

    The initial and final mixing height and the rate of rise (the  magnitude of the
dilution rate)  are probably the real factors that vary from day to day, and these
are more likely to be the controlling factors in the real situation. It would not be
any more difficult to program a computer search for the set of these conditions that
give the observed 0$ than it is now to look for  the correct combination of NMOC and
NOX.  Then the mixing height  parameters would  be fixed and the NMOC and NOX
reductions could  be explored in an absolute  manner. Different mechanisms may


Needs for Future Mechanism Development
give much closer predictions of control requirements under this situation than they
do under the standard EKMA approach.

    In any event, the Steering Committee believes that the scientific foundations
and assumptions in the present EKMA technique require further examination in
conjunction  with the adoption of the new mechanisms. This would be a modeling
and sensitivity study and may require re-writing of the OZIPM code to  implement
new scaling methods.

                      Paper 1
 The Science of Mechanism Development and Testing
                  Harvey Jeffries
Department of Environmental Science and Engineering
      University of North Carolina-Chapel Hill

                             The Science
                   of Mechanism  Development
                             and Testing

                             Harvey  Jeffries
                    Department of Environmental Science
                             and Engineering
                        University of North Carolina
                           Chapel Hill, NC 27514

This talk is about the practice of science related to photochemical kinetics mecha-
nism development for urban atmospheres.
    o One of its goals is to introduce a meta-model of modeling:
          o to describe the process of what modelers do to develop  and test a
            mechanism as opposed to the content of the models themselves;
          o to examine the components of the process and to understand how
            these contribute to the advancement of knowledge about atmospheric
    o Another goal is to place  before you questions that  I believe highlight the
      issues we are gathered to  discuss.
    o A final goal is to suggest actions that  I believe will maximize the benefits
      from all of our efforts.

Why Does EPA Need Science?

As described in the Government Accounting Office (GAO) report on EPA models,
air quality (AQ) models are used as tools in administering the Clean Air Act.

   EPA, therefore, wants models that predict, accurately and reliably, the effects
of changes in emissions of chemical precursors on the magnitudes of secondary
pollutant concentrations formed  hours to days after the emissions and 10s to 1000s
of kilometers away from sources.

   Quantitative  observations of such  atmospheric transformation processes are
nearly beyond our abilities, both from an experimental viewpoint and from a social
viewpoint as well.

                                                        Why Does EPA Need Science?
    Because science, as a way to provide growth in knowledge, has been preemi-
nently successful especially in the 20th century, government often turns to science
for help in achieving its goals. For many involved in this enterprise science is seen
as especially useful because they believe that "science makes possible knowledge of
the world beyond its accessible, empirical manifestations."

    Thomas Kuhn has  described the aim of science as "to  invent theories  that
explain observed phenomena  and that do so in terms of real objects, whatever the
latter phrase many mean."

    If this aim of science were  fully accomplished, then near-perfect  prediction
should be possible.

    The EPA has supported  scientific work with an aim to provide,  if not near-
perfect, then certainly acceptable prediction. Yet, according to some, the 'scientific
community' apparently has not accomplished this goal in 15 years, for example:
    > the GAO says EPA's  models are based on "assumptions, approximations,
      and judgments";
    > the GAO says that EPA's most reliable models estimate—they  do not  even
      say predict—"actual pollution concentration within a range of minus 50% to
      plus 200%";
    > the GAO quotes  un-named EPA officials as saying "more precise results are
      unlikely because  of the limitations in the science";
    o the House Committee  on Oversight wants a timetable for improvements.

    Are public and  Congressional expectations for science unreasonable? One of
the things the public expects from science is the  capacity to explain things and
concurrently to enable  at least scientists to predict what will happen  in various
combinations of circumstances.  One normally explains a phenomenon by citing
the mechanism that brought  it about or sustains it, where the mechanism itself is
understood by subsumption under causal, or at least time-dependent  laws. When
done 'right' in the minds of the public, the explainer not only derives the explanation
correctly,  but he also gets others to understand and accept his derivation. These
actions, explain, predict, accept, are at the crux of the issues we are here to discuss.

Problems With Our Science.
Problems With  Our Science.

I believe that our problems arise, in part, because of a lack of agreement.  For
    > there is no single, generally accepted chemical mechanism to describe the
      atmospheric transformation of precursors into the regulated and potentially
      regulated air pollutants.
    > there is such dis-agreement and rapid change in the modeling community that
      many, both inside and outside, wonder about whether any of the solutions
      proposed  are legitimate.

   To many people this is a frustrating situation.  After more than a decade and
a half of air pollution modeling research, we still cannot predict ozone to  many
people's satisfaction.

   The results of this situation are:
    > The sense of frustration  with  conflicting and contradictory results has led
      some administrators at EPA to question the policy of reliance on atmospheric
      models for predictive use.
    > Some have even doubted the necessity of further attempts to resolve the di-
      visive modeling questions within the scientific community that created them.
      Throwing good money after bad, they seem to be saying, makes even less
      sense in science than it does in economics, and so EPA perhaps should not
      seek a solution to the problems arising from models by funding  additional
      model research.

   Users of models want to understand:
   1. Why do we have conflicting and contradictory results in the applications of
   2. Are some models wrong and other models right?
   3. Are no models right?
   4. Are smog chambers data wrong and theory right?
   5. Are atmospheric data the ultimate judge of correctness?
   6. What tools or strategies are available to help EPA, and the 'scientific com-
      munity' of modelers themselves, make decisions about chemical mechanisms?
   7. Given our situation, how should  the (urban) atmospheric chemistry com-
      munity now achieve progress?  That is, how should we further actualize the

                                                          Problems With Our Science.
      promise of science to explain atmospheric chemical transformations, and do
      so in a manner sufficient to meet EPA's needs?

    A little reading of scientific history and philosophy of science will show that, in
the broader sense, these questions are  not specific to the problems of atmospheric
chemical models or to EPA's needs for such models, but they address something of
core importance in all sciences.

    As explained in the background document, "The Science of Photochemical Re-
action Mechanism Development and Evaluation,"  I have relied on the writing of
an acclaimed philosopher of science, Thomas Kuhn, for many of the  concepts I
have used to investigate our situation. Kuhn's careful analysis of many scientific
situations provides significant insight into the process of scientific advancement.

    Further, he suggests that science requires a decision process that which permits
rational people to disagree. He is also clear on how the ultimate judgments are
made in science; we will discuss this process later.

Who Agrees With Whom About What?
EPA's Needs Verses Scientists' Needs

We must clearly realize,  however,  that the needs of EPA and the  needs of the
scientists do not necessarily go hand-in-hand.

    From EPA's short-term perspective, the best outcome of this workshop might
be the development of a  "standard mechanism evaluation procedure(s)" and the
definition of a "standard data base" needed for performing the evaluation.

    Such an procedure may not be useful from the scientist's perspective.
    >  The scientist is concerned to understand the world and to extend the preci-
       sion and scope with which it has been ordered.
    >  If intense scrutiny of this world reveals pockets of apparent disorder, these
       challenge him to a new refinement of his observational techniques or to a
       further articulation of his theories so as to extend the precision and scope of
       his understanding.
    >  Often the theories (and so the  mechanisms) that arise in this process must
       be in complete contradiction  to the ones that have gone before them. This is
       so because the new theory must be more successful precisely where the old
       theory failed. A new theory may necessitate new experimental evidence and
       a radically (so to speak) different mechanism.
    >  This has  certain implications for the concept that there may be a single
       mechanism evaluation procedure and its concomitant database suitable for
       all present and future theories and mechanisms.  Such a concept ignores the
       evolutionary character that has brought about the  very success it seeks to

    But the difference in the need for rules goes beyond the evolutionary character
of mechanism development.

    Kuhn says that a lack  of a standard interpretation or of an agreed reduction
to rules will not prevent a 'scientific community' from conducting research that is
acceptable to the community. This is true, however, only as long as the relevant sci-
entific community accepts without question the particular problem-solutions already

Who Agrees  With Whom  About  What?

I think it is not it not much of an exaggeration to suggest  that every  modeler dis-
believes major aspects of every other  modeler's model.  And that some modelers

                                                  Who Agrees With Whom About What?
have serious doubts about all smog chamber data. And that many smog chamber
experimenters think that the models fit  the data poorly because the models are
incomplete or just plain wrong.

    Thus rules can become important wherever the underlying guiding principles of
a science are successfully challenged  or even where the community only perceives
that the principles are weakening.

    On the other hand, if we have arrived at the point where rules are now required
to sort theories and mechanisms,  then  something fundamental  in the science of
urban air chemistry modeling is under challenge.

    This challenge arises, I believe, from two situations:
    1) as a result  of the inability to resolve the  differences among  competing expla-
      nations of the chemistry of paraffins and olefins; and
    2) as a result of the failure to offer acceptable explanations  of the aromatics

    These are failures not only of the experimental  arts, but also of theoretical

How is Scientific Knowledge Obtained?
How is Scientific Knowledge Obtained?

Kuhn says that scientists often work from patterns of behavior (mental constructs)
"acquired through education and through subsequent exposure to the literature
often without quite knowing or even needing to know what characteristics have
given their methodologies and approaches acceptable status in the community that
they practice in. If they have learned such abstractions at all, they  show it mainly
through their ability to do successful research."

    Kuhn further states,  "Traditional discussion of scientific method have sought a
set of rules that would permit any individual who followed them to produce sound
knowledge.  I have tried to insist, instead, that, though  science is practiced by
individuals, scientific knowledge is intrinsically a group product and  that neither its
peculiar efficacy nor the  manner in which it develops will be understood without
reference to the special nature of the groups that produce it."
    A.  Normal science and revolution. Once a specific science has been individuated
       at  all, it characteristically passes through a sequence of normal science—
       crisis—revolution—new normal science.  Normal science is  chiefly  puzzle-
       solving activity, in which research workers  try both to extend successful
       techniques, and to remove problems that exist in some established body of
       knowledge.  Normal science is conservative, and its researchers are  praised
       for doing more the same, better. But from time to time anomalies in some
       branch of knowledge get out of hand, and there seems no way to cope with
       them.  This is a crisis. Only a complete rethinking of the material will suffice,
       and this produces  revolution.
    B.  Paradigms.  A normal science is characterized by a 'paradigm' [(PAIR-uh-
       dime), a pattern]. There is the paradigm-as-achievement.  This is  the ac-
       cepted way of solving a  problem which then serves as a model for future
       workers. Then there is the paradigm-as-set-of-shared-values. This means the
       methods, standards, and generalizations shared by those trained to carry
       on  the work that models itself on the paradigm-as-achievement.  The social
       unit that transmits both kinds of paradigm may be a small group of perhaps
       one hundred or so  scientists who write or telephone each other, compose the
       textbooks, referee  papers, and above all discriminate among problems that
       are posed for solution.

    The term "paradigm" enters in close proximity, both physical  and logical, to
the phrase "scientific community." A paradigm is what the members of a scientific
community, and they alone,  share. Conversely, it is their possession of a common

                                                  How is Scientific Knowledge Obtained?
paradigm that constitutes a scientific community of a group of otherwise disparate

    One thing that binds the members of any scientific community together and
simultaneously differentiates them from the members of other apparently similar
groups is their possession of a common language or special dialect.  Kuhn has
suggested that in learning such a language, as they must to participate in their
community's work, new members  acquire a set of cognitive commitments that are
not, in principle, fully  analyzable within that language itself. Such commitments
are a  consequence  of the ways in which the terms,  phrases, and sentences of the
language are applied to nature, and it is its relevance to the language-nature link
that makes the original narrower sense of "paradigm" so important.

    The paradigm-as-shared-values is a  source of commitments for the scientists
that share it. These include:
    1. Explicit statements of scientific law and about scientific concepts and theo-
      ries. While these continue to be honored, such statements help to set puzzles
      and to limit acceptable solutions.
    2. A multitude of commitments to preferred types of instrumentation and to
      the ways in  which accepted instruments may legitimately be employed.
    3. Quasi-metaphysical commitments  that tell scientists what sorts of entities
      the universe does and does not contain, what ultimate laws and fundamental
      explanations must be like, and what research  problems should be.
    4. A commitment to be concerned to understand the world and to extend the
      precision and scope with which it has been ordered, therefore, says Kuhn, "to
      scrutinize, either for himself or through colleagues, some aspect  of nature in
      great empirical detail. And if that scrutiny displays pockets of apparent dis-
      order, then these must challenge him to new refinements of his observational
      techniques or to a further articulation of his theories."  Though his concern
      with nature may be global  in extent, the problems on which he works must
      be problems of detail.
    5. More important, the solutions that satisfy him may not be merely personal
      but must instead be accepted as  solutions  by  a group. This group may not,
      however, be drawn at random from society as a whole,  but is rather the
      well-defined community of the scientist's professional compeers.
    6. One of the strongest, if still unwritten, rules of scientific life is the prohibition
      of appeals to heads of state or to the populace at large  in matters scientific.

How is Scientific Knowledge Obtained?
       Recognition of the existence of a uniquely competent professional group and
       acceptance of its  role as the exclusive arbiter of professional achievements
       has further implications.
       The group's members, as individuals and by virtue of their shared train-
       ing and experience, must be seen as the sole possessors of the rules of the
       game or of some equivalent basis for unequivocal judgments. To doubt that
       they shared some such basis for evaluations would be to admit the existence
       of  incompatible standards of scientific achievement.  This admission would
       inevitably raise the question whether truth in the sciences can be one.
Kuhn  says,  "Normal science is a highly determined activity, but  it need not be
entirely determined by rules."  There are rules for  moving the chess  pieces;  there
are few rules for playing a game of chess.

    For Kuhn, "Normal science consists in the actualization of that promise [of suc-
cess offered by the paradigm], an actualization achieved by extending the knowledge
of those facts that the paradigm displays as particularly revealing, by increasing the
extent of match between those facts and the paradigm's predictions, and by further
articulation of the paradigm itself." These are the three areas into which Kuhn says
all journal literature of normal science can be classified.

    The business of science, Kuhn says, is puzzle-solving. Though this may sound
demeaning, he certainly  intends no offense; he says that this process has  required
the highest scientific talents. In fact, he choose the  metaphor because  of the strik-
ing similarity between solving jigsaw  puzzles and  the  search for explanations  of
phenomena and the ways to harness nature.

    Kuhn contends that this puzzle-solving activity is a vital part  of science so long
as the  puzzles are interesting and the solutions are accepted by the community. In
their day-to-day work, scientists do not try to disprove the theories that direct their
research, nor do they seek unexpected  and unpredicted  results.

    To understand what  is meant by puzzle solving  we must examine a schema for
scientific practice:

               ,p,     , Auxiliary     D  j- A.      Facts to
               Theory + _       J   = Prediction fa ,      . .   ,
                        Statements                be explained

In the daily practice of  normal science, a scientists  seeks statements of his best
guesses (auxiliary statements)  about the proper way to connect  his own  research
problem with the corpus of accepted scientific knowledge (Theory).

                                                   How ia Scientific Knowledge Obtained?
    For example, he may conjecture that a newly discovered product in a chamber
experiment is to be understood as an effect of HO'-attack on a particular hydro-
carbon. The next steps  in his research are  intended to try out the conjecture or
hypothesis.  If it passes enough or stringent enough tests, the scientist  has made
a discovery or has at least resolved the  puzzle he had been set.  If not, he must
either abandon the puzzle entirely or attempt to solve it with the aid of some other

    The practitioner must often test the conjectural puzzle solution that his inge-
nuity suggests. But only his personal conjecture is tested.  If it fails the test, only
his own ability, not the corpus of current science is impugned. As Kuhn says,  "in
the final analysis it is the individual scientist rather than current theory which is

    Tests of this sort are a standard component of what Kuhn called normal science.
In no usual sense, however, are such tests directed to current theory. The scientist
does not immediately look for  a new oxidizing theory, on the contrary, the scientist
must premise current theory as the rules of his game.  His object is to solve a puzzle,
preferably one at which others have failed, and current theory is required to define
that puzzle and to guarantee that, given sufficient brilliance,  it can be  solved.

    As normal science advances, the promise of the paradigm is actualized and the
paradigm  itself is refined, reformulated, and further articulated.  This enterprise
seems to the  outsider (one who does not share the commitments described above)
to be an attempt to force nature into the preformed and relatively inflexible box
that the paradigm supplies.

    Theories are accepted because they have  real  explanatory successes. Although a
theory may legitimately be preserved by changes in the auxiliary statements which
are, in a sense,  ad hoc although not unreasonable,  its successes must not be ad
hoe. That we have no new paradigm 15 years after the proposition of HO -attack
does not mean that air pollution science has failed to progress in a Kuhnian sense.
Rather, the puzzle-solving done in those 15  years to expand  and ratify, or modify
the theories and auxiliary statements of HO -attack has been  the success promised
by that paradigm.  And  it was that promise, Kuhn would say and  I would agree
after looking  at the actual documentation of those early years, that led us to accept
the HO -attack theory in the first place.

    The areas investigated by normal science are minuscule; the enterprise has dras-
tically  restricted vision.  Kuhn says, "But those  restrictions,  born from confidence


How is Scientific Knowledge Obtained7
in a paradigm, turn out to be essential to the development of science. By focusing
attention upon a small range of relatively esoteric problems, the paradigm forces sci-
entists to investigate some part of nature in a detail and depth that would otherwise
be unimaginable."

    Thus the juxtaposition of our theoretical constructs and the observations asso-
ciated with them leads to an extension of the precision and scope of the ordering
of the world.

    This progress of the paradigm, the practice of normal science—by its very na-
ture (the  finer and finer attention to detail)—encounters anomalies that require
distortion to include  in the paradigm, or deletion to ignore as  important to the

    The hypotheses of individuals are tested, the commitments shared by his group
being presupposed. Group commitments, on the  other hand, are not tested, and
the process by which they are displaced differs drastically from that involved in the
evaluation of hypotheses. This is the crisis situation, in which the anomalies created
by the progress of normal science so strain, the old theory that many seek revolution,
a turning away from the original paradigm to a new theory, a new way of looking
at the world, that again offers real explanatory success and is full of promise for the
future.  We have not had such a  revolution in the still young field of atmospheric
chemistry:  the HO -chain theory still  holds much promise and, in our daily scientific
activities,  we still seek solutions to puzzles  assuming that  an HO -chain process is
central to  atmospheric chemistry.

                                                        Our Situation: Actions Needed.
Our Situation:  Actions  Needed.

We currently find many anomalies hidden in the region of "reasonable agreement"
between mechanism predictions and chamber data.  And because modelers have a
real need to succeed in solving the puzzle set by themselves, solutions to puzzles
have sometimes been claimed with little vindication. Further, the community has
apparently accepted these solutions at the time, only to find themselves in a position
of denying the solution a year later.

    This makes us appear somewhat less than legitimate or not very serious about
how much we really do understand about urban atmospheric chemistry.

    Therefore, I believe we must work as a community to re-establish explicit recog-
nition of a set of rules.  These rules, in the words of Kuhn "limit both the nature of
acceptable solution and the steps by which they are to be obtained."

    If we take 'limit' not to mean restrict, but rather to define, then these rules
are not unnecessary burdens on science. Also the interpretation of limit as "define"
explains why rules can lie unexpressed below the surface of regular scientific activity:
definitions, as we have seen, must often be taken for granted if they are to function
at all.

    These definitionary rules do not define what the state of the science, or even
what the paradigm, is. But rather, they define whether or not solutions fall within
the paradigm, require modification or distortion of the paradigm, or fall completely
outside the  paradigm.   Such rules can change in direct  response to the input of
puzzle- solutions, but clearly they are not infinitely flexible.

    Frequently in times of crisis, more solutions are ruled out of the paradigm than
are included within it.  When the rules function in this way to define the paradigm
more and to prevent its unnecessary distortion, the rules can save a still-functional
paradigm and work goes on.

    When the rules  have strengthened and protected the paradigm in these ways,
they—now accepted by consensus in  the community—recede again from explicit
recognition and indirectly guide the community toward interesting puzzles in need of
solution. Thus the definitionary rules can be changed in response to the increasingly
refined solutions proposed within the  paradigm, but they direct research  best  only
when they are sufficiently accepted by  the community to  be implicit.

Our Situation. Actions Needed
    For science to advance at all there must be reasonable agreement between three
sets of evidence:
    1.  between theory and observation;
    2.  between different sets of observations; and
    3.  between explanation and prediction.
Without agreement, no detailed comparison could be made between the three sets of
evidence: no puzzle-solutions could be evaluated. And the puzzles most meaningful
in maintaining or challenging a paradigm are exactly those that lie on the interfaces
of the three sets listed above.

    Thus the first sense in which reasonable agreement determines scientific progress
is in defining what sorts of evidence can be compared. That is to say, that by defin-
ing what qualifies as evidence of theories, observations, explanation, and predictions,
the community establishes  reasonable agreement for quantitative precision.

    Once evidential standards of  reasonable agreement have been recognized, the
puzzles we have described  as at the interfaces of theory and observation, between
observations, and between  explanation and prediction can be defined.

    When solutions to the  problems are offered, the second  function of  reasonable
agreement is activated: agreement for better quantitative and qualitative precision
to discriminate among the solutions.   Again,  this  function  is determined by con-
sensus within  the  community and again as  a standard it varies directly with the
progression of the  paradigm from potential to actualization.

    Thus, while I see the need for  a "mechanism evaluation procedure" I do not see
it as static or as something EPA regulates.

    I see  the process as consisting of several groups of practicing scientists struggling
with what constitutes "reasonable agreement" among the sets of evidence. I see this
as requiring face-to-face meetings in which evidence is presented, exchanged, argued
and agreed upon or questioned.  Part  of the current problem is that some of the
most important  evidence (e.g., plots of model predictions and observational data
and even mechanism listings) are sometimes not presented for judgment.  Significant
operational procedures are omitted for the sake of brevity.  Some evidence is treated
as 'proprietary.'  Clearly, these meetings must be open to the 'scientific community'
and must be supportive of those with differing opinions and with different sources
of information.


                                                       Our Situation: Actions Needed.
    Because the development of knowledge is not static, these assessments must
be periodic. The products of these assessments should allow modelers to develop
vindication statements for their models.

    Finally, it should be our goal not to seek multiple representations of reality—
for  example, it  is not the  Carbon Bond Mechanism, nor should it be the Carter,
Atkinson, Lloyd, Lurmann mechanism—instead, it should be
    > 'What is  the best representation of what we know with a high degree of
      certainty?', and
    > 'What is an adequate representation for what we probably know?' and
    > 'How creative can we  be about what we are guessing at?'

                  Discussion After Jeffries' Presentation
    Jeffries:  I have passed around some questions that are keyed to certain pages
in the background document I wrote. I believe that these questions highlight some
of the particular details and issues that we want to discuss. [These questions are
included here  as  Table 1.]  Dr.  Atkinson's basic proposal made  in his  opening
remarks was that  if a model fits smog chamber data, then the mechanism should be
satisfactory. Question 10 asks, "Is that really true?" I discussed several situations
were mechanisms  'fit' smog chamber data very well, only to later be shown to have
a have a major kinetic parameter wrong.  Questions 11 and 12, I think define what
this conference is  really about. "How do we prevent mechanisms from being mere

    Dimitriades: Do you intend to provide answers to these questions?

    Jeffries:  These are questions to stimulate discussion. I do not know if there
are "answers" to the questions. I certainly have provided discussion of them in the
background document. For example,  on page 42 of the document you will see that
in the Hecht-Sienfeld-Dodge paper, it was proposed that validation consisted of two
steps: 1) validate  against chamber data, and 2) validate against atmospheric data.
But they never explained why they wanted to validate against atmospheric data.
So, what is the real  need to validate against atmospheric data? What do you gain
by doing that? How would you actually use this process?

    Dimitriades:  That's fine.  I think the discussion can, perhaps, address these
questions of yours.

    Jeffries: I think it's premature,  Basil, I think we should wait for some others to
have a chance to reflect on some of the things that have been said.

    Atkinson:  Maybe we can go through the rest of the day and bring  that up
tomorrow morning.

    Riordan: As I understand it, in the normal evolution of sciences, as stated by
Kuhn,  there is an adoption of a  paradigm, then people apply the paradigm, and
over time anomalies develop as people look at the detail in the world. To deal with
the anomalies  they jerryrigged this paradigm.  Then all of a sudden these people
realize  that you cannot jerryrigged it anymore; then you get someone questioning
the basic paradigm and you get crisis,  and then revolution. Do you picture what


           Table 1. Discussion Questions Submitted by Jeffries

 1. Are "more precise results" in applying models unlikely "because of the limi-
    tations in the science" as was reported to have been by some EPA official in
    the GAO report? (page l)

 2. How should a scientist deal with the "delemma of the underdetermination of
    theory?"  (page 32)

 3. What are the conditions necessary to believe in a true correspondence be-
    tween entities in a mechanism and the real world? (page 38-39)

 4. How else, other than by the  use of  smog chambers, can we produce the
    detailed facts that must be compared unambiguously and directly with the
    model while preserving as much of the whole chemical transformation process
    as  possible to allow for the evaluation of significant over-generalization and
    deletion? (page 40)

 5. How, or even do, comparisons of model predictions with chamber data deal
    with the question of under-determination of theory? (page 40)

 6. What is the real need to validate in the atmosphere? (page 42)

 7. What defines "reasonable agreement?" (page 48)

 8. Given the comment made by Whitten on page 76, how can we believe the pre-
    diction of the mechanism outside of the conditions on which its explanatory
    function was demonstrated?

 9. Given the ability to predict  a chamber radical source strength, but  not the
    ability to explain  the source, can mechanisms be adequately tested with
    chamber data? (page 81)

10. What does testing mean? What does failure to fit chamber data mean for a
    mechanism?  What does success in fitting chamber data mean? (page 95)

11. How do we prevent mechanisms from being mere opinion? What constitutes
    "probative force to justify" a modeler's claims?

12. What is required to  make  the vindication statement on page 97?  Is this
    statement satisfactory for EPA's needs?


has been proposed here as a process that is beneficial, as trying to accelerate that,
or as being sort of complimentary, or as being contradictory to that.  I guess what I
am trying to figure out is, if you have this normal evolution of science and it occurs
somewhat naturally, what are we doing here trying to talk about rules of evidence
that would be appropriate in trying to discover anomalies?

    Jeffries: Within the Kuhn structure of things, I think our present field is still
in the middle of practicing normal science. We have not come across any of the
kinds of crisis that Kuhn has described. In other words, none of us sitting around
this room disbelieves that HO exists or that the HO -chain theory can account for
observed atmospheric chemistry. We still, I think, fundamentally believe that  there
are solutions out there operating under our current conceptual framework. And so
that's not what is at issue here at all. That is, we are not facing a Kuhnian crisis in
the sense of a failure in our paradigm. What is unique about our particular situation,
is that we're wrapped up in your regulatory issues. It  is you who comes along and
imposes a time table on the  science.  Given long enough time, our  problems, the
ones that I described as problems of the  science, would go away, by themselves.
We would just shake things  out.  So,  if someone discovered something that was
accepted by the community for a while, and then later we find out that it was not
true, we would just change it.  The problem is that you take this first result and
embed it in law and use it. That's where I think the interface between the science
and the regulatory issues cause us to have a problem. You are forced to use tools
prematurely. Or, to put it another way, as I said in the write- up, we have 'forced
actualization over promise.' We've been forced to achieve an actualization of the
promise of the paradigm before its time, in a sense, without adequate foundation.
That's where I think the real crux of  the problem  lies.  And, we're forced to do
that because 1) we have to succeed in solving the puzzles or we are seen as failures
as scientists, and 2)  we are  forced to conclude that  we  always  have reasonable
agreement because we have to meet the terms of your  contract or the grant.

    Whitten: I have two questions to bring up. First of all I think that the discussion
is not talking about engineering.  I think that  there are  science kinds of problems
and there is the need for a regulatory agency to have  a model.  What is needed is
an engineer in the middle that uses the  information in the scientific community and
constructs something based on it.  We need some sort  of an engineering approach.
There has always been a problem historically between  engineers and scientists and
this is a lot of what we have been talking about here. The second point is that I have
problems with your  use of the word paradigm in the context of the HO mechanism.
The paradigm that  I  see that is more  involved with the production of ozone and
smog, is the conversion of NO to NOj. I see that as a central issue that is currently


been challenged to a certain extent. There are two cycles of smog chemistry.  One
is the HO -organic cycle and the other one is the inorganic and NOX, and they are
connected with this NO to NOj conversion.  I would say the paradigm is the NO to
N©2 conversion in connection with these cycles. So that all the organic chemistry
is run through the HO  mechanism and production of hydrogen containing radicals.
I offer that as an alternative to the paradigm structure. It's not inconsistent with
what you've said but it's a little different.

    Jeffries:  I believe that EPA wants a credible, and therefore, 'scientifically sound'
approach. And for the second point, I do not see what you have described as being
any different at all. The paradigm-as- shared-values is a general acceptance of a set
of beliefs as  to how the world functions with respect to smog chemistry. I certainly
believe the general description that you gave.

    Atkinson:  Why must we accept a  single mechanism for use in an individual
air shed  or regional model?  Provided that mechanisms are  all based on the  best
available data and are tested equivalently, you can still end  up with different for-
mulations of those mechanisms. Indeed, they well may differ in their predictions.
But that is because the testing against environmental chamber data, or whatever
data, really is not testing under conditions which are applicable to the atmosphere.
We are quite a ways away from true atmospheric conditions in the testing. So you
can easily end up with different predictions and different estimations. But, based
upon our present data, they still could be wrong. They  do give you, if you have
multiple  mechanisms, based upon those same data base, some idea of the poten-
tial uncertainties  in the data.  I would not say accuracies, but certainly potential

    Jeffries:  I was not  very clear in making my  point, Roger. I do not advocate a
single mechanism, by any means.  But what has come to be true outside of our own
small  group, is that users tend to think of the mechanisms as an entity: it is the
'Carbon  Bond Mechanism,' it  is the Carter-Atkinson-Lloyd-Lurmann Mechanism.
It's an entity.  It  is like one  is right and one is wrong. What I am suggesting, is
that there are elements of knowledge—as you say—there are kinetic basis for which
all of us probably believe that they are  true representations of what  is really going

    Atkinson: Well, they are our best representations....

    Jeffries:  ... Ok,  but they are adequate enough  for a lot of people to agree
that they are probably the best explanations we  have at the current time.   And


further we would expect, if a mechanism is going to include that information, then
it should be the same in the different mechanisms.  In other words, there is a set of
information which we believe has to be the same in all mechanisms. ...

    Atkinson: Right, which means that they're based on the same data base.

    Jeffries:  Yes, but then there is the next group  of information for which we
are less certain about what is going on,  and for which  we do not have adequate
data, either kinetically or mechanistically to make distinctions among them.  To
build a model we have  to  make estimates of what we believe is true or right  or
wrong. Different people do make different estimates, and thus come up with different
answers that then led to different predictions. That is  what I called the second class
of knowledge I wanted to identify.  That is, 'What is an adequate representation
for what we probably know?'  And  what are the consequences of different  people
having chosen that  information differently. This approach is different in concept
you see, than simply saying, 'the Carbon Bond Mechanism is wrong, or the Carter-
Atkinson-Lloyd-Lurmann mechanism is right.'  Instead  of thinking about it that
way, which tends to give the impression that there are different representations, we
need to move to a three-tiered representation. In simple terms, there is a part  of
the chemistry which we all believe in, there is an area in which we are uncertain  so
we  make different choices, and there is another whole area which is wide open  to

    Atkinson: Yes, but those two mechanisms ultimately, how ever many there may
be, will ultimately converge as our knowledge gets better.

    Jeffries:  Granted.

    Atkinson: So, I do not think you should differentiate too much between different
mechanisms. I mean, clearly there're all aiming  towards the same end purpose.

    Jeffries:  Yes. But that's not how they're perceived outside.

    Atkinson: They  serve to define the uncertainties and how best to attack them, if
done correctly. They serve to define the uncertainties in our knowledge—the areas
where further input data are needed. By having those two or three representations
of the same data base.

    Jeffries:  There is another side of that coin:  they  serve to confuse the hell out
of people who do not know the details of the data  base.  And I think that is what



leads to part of the problem here. That people see that one mechanism predicts
one way and one predicts another way.  Then they begin to say, 'Which one is right
and which one  is  wrong?' In truth, aspects of both are right, aspects of both are
probably ok, and  other aspects of both are probably wrong.

    Atkinson: Of  course.

    McRae:  Harvey, since you set your report up as a straw man, I would like to
challenge several points you made. First of all, I'd like to challenge a basic premise
that you say that there are lots of assertions about model  performance and  our
inability  as a 'community to predict outcomes'. I think one of the things  that
would be very useful in  this workshop  is simply to define what we think are these
uncertainties. Are they scientific uncertainties? Are they educational uncertainties?
For example, somebody  in the White House or on the Hill may not be able to
distinguish between nitric oxide and nitrous oxide. That is an  educational issue
as distinct from a science issue. I think it would be frequently referred to as what
constitutes a good prediction. To whom and why? And what exactly do we mean by
that? The third area when it comes to  the basic premises that I think  is extremely
important to distinguish between uncertainties and the effects of those uncertainties.
It is very easy to build a laundry list of what we do not know. The critical question
is, 'What is the effect of those uncertainties in terms of predicted outcome?'

    The second  thing I would like to challenge is in use of Kuhn's work as a paradigm
for two reasons. One is that it doesn't  really reflect what's been  going on over the
last 25 years in artificial intelligence and common good psychology. There's lots of
insights in terms of problem solving, that  are helpful in that area. But, my major
criticism with the use of Kuhn's description is that in fact it is descriptive. It is not
constructive. It does not provide direct guidance in terms of what we should do to
improve our behavior in the future. The second point is that I'd like  to challenge
the view that you stated a number of times, is that the revolutions in science are a
group project and I can think of lots of examples in which that's not true. The view
that change— most revolutions in science do not easily fit into Kuhn's paradigm.
For example, Newton's Law of Gravity or  Einstein's General Theory, are not good
products.  They are step changes.  There was no pre-existing science,  pre-existing
theories, they were just something that was created.

    Kuhn  doesn't really discuss  the role  of data in discriminating between alter-
native theories. One of the  difficulties in  reading this  report is  that,  it would be
interesting to write a historical analysis of the  evolution of the quality of smog
chamber data.  And, to what extent that  has influenced the  choices that people in


modeling community might make. The final thing about the use of Kuhn's theory
that I'm disturbed about is the essentially a post theory view. It is looking back.
It doesn't tell you anything about how to identify whether you're in a paradigm or
about to undergo the gestalt switch or whatever. You frequently have eluded to the
fact that the modelers have prematurely declared victory.  I would offer the  view
that we do not know which race we're running in. In many cases the nature of the
problem is evolving.  So, the fact that we prematurely claim victory really, I've used
the case where a lot more about  how to define what victory is at the time you're
conducting it.

    Another area, which  Gary Whitten also raised, which I think is extremely im-
portant in terms of this group and the regulatory agencies is the view that between
science and engineering, that the science is unraveling the details of the process, the
engineering is actually the use of those what you know in a regulatory environment.
For example, we do not understand all of the details of materials on the mechanical
level but we still build bridges. We need to cross the river. We have an  analogous
form in air pollution. Thirty percent of the population is exposed to  levels of above
the Federal Ambient Air Quality Standard. Twenty-five cities are not going to meet
the containment deadlines.  That's a real problem: an engineering  problem.  One
that we need to suggest approaches to.

    The final area, which is a structure  to follow what Roger suggested, is  that
one thing that  as a community we could do, is to simply define what we think to
be the current  uncertainties and  nature of the  problem.  There are lots and lots
of assertions from GAO reports that we can't predict anything. What exactly are
the problems?  I think that we can  define them in terms  of the science issue or
engineering issue. And then I think there's lots of other  details which we can call
up based on largely  what Roger put together as a series of questions. So, I think
that Kuhn's idea is an interesting way to write  historical perspective,  but I do not
think that it is  particular constructive in this environment.

    Jeffries: [These comments were added after the workshop]. I believe that I stated
the source of the workshop issues quite clearly  in both the background  document
and  in my workshop presentation. I believe  that large parts of the paradigm-as-
shared-values which  I hold, along with others  here,  are under challenge, in  part
because of the  community's inability to  resolve the differences among competing
explanations of the chemistry of paraffins and olefins, and as a result of the failure
to offer acceptable explanations of the aromatics chemistry.  These are scientific
uncertainties  that lead to problems in application areas, which might be described
as educational problems.  Papers  presented at this workshop will show the effects


of these scientific uncertainties: differences in predicted control requirements that
translate into million dollar differences in costs.  Thus, the issues discussed here are
not just scientific 'laundry lists,' their resolutions are important to EPA's progress
in meeting goals established by Congress.  Furthermore, those not in the field have
questioned whether modelers are honest, or more politely, whether there is any truth
in what they present.  Why is this so?  What  is required to prevent mechanisms
from being mere opinion?

    With regard to the use of Kuhn's work to organize the background discussion,
as I said in the document, "There is no one truth about what happened."  Kuhn's
description in relevant for me—it fits  my experience and provides detail in my
map of reality.  I very much disagree with your  statement that Kuhn's view  is not
constructive, that is, 'does not provide direct guidance in terms of what we should
do to improve our behavior in the future.' Kuhn's work provided me the concept
of "reasonable agreement" and explained  how data and theory function together.
His description of normal science is analogous to your concept of the moving target
or that 'the nature of the problem is evolving.' As for behavioral guidance, Kuhn's
work suggested to me that if this workshop attempted to develop a set of standards
and method of testing mechanisms that did not conform with current practice, then
the workshop's product would  be doomed  to failure, it would simply be ignored by
the practicing scientists  because it would not  conform to his needs, in spite of
how well it might meet EPA's needs for regulatory consistency. Therefore Kuhn's
philosophy certainly influenced my outlook and approach to this workshop and thus
has been very constructive for  me and also for others here.

                     Paper 2
How Do You Know a Good Model When You See One?
                  Joseph Tikvart
    Office of Air Quality Planning and Standards
        U.S. Environmental Protection Agency

  How Do You  Know a Good Model When You See  One?

                             Joesph Tikvart
             Office of Air Quality Planning and Standards
                U.S. Environmental Protection Agency
                      Research Triangle Park, NC


The background document written by Jeffries and Arnold is an interesting historical
and philosophical work, but I do not think it is really a "straw man." I am not
sure that a group of people should tell those developing or doing the science how
to go about the task. They should go at it as they would, based on their training
as scientists, and then let others judge how they did.  I do not think there should
be a test to tell people how to go about that process or to set up a procedure for
doing this work. I view the issue that we are dealing with as basically, "How I know
a good model when  I see one."  The background document did not  deal with that
issue directly.

    There was a second background document written by Sexton and Jeffries enti-
tled "Need  for Chemical Mechanism Documentation"  that  I think did attempt to
address the issue. This second document considered several points.

    First, it expressed the need for guidance from the developer. What should be
done about questions where there are other versions of the models or the chemical
mechanism developed and where the user modifies that mechanism?  How should
this problem be handled?

    Second, there was discussion of confusion about the support programs for the
chemical mechanism; that is, how do we get the input needed to drive the mecha-
nism. I think an example here would be the extent of pre-processors needed to run
the SAI airshed model, and how one should deal with these.

    Third, the document discussed machine-oriented problems. That is, most codes
are written for one type of machine; to apply it  to another type of machine might
require extensive changes. For example, the simple Gaussian models in UNAMAP
are written  for the UNIVAC computer and you can not merely take that UNAMAP
code and put it on the IBM. There are some minor changes that need to be made.
Another example is the air shed model that was written for the PRIME. To run it
on the  UNIVAC is not  necessarily a  simple undertaking. So, how should we deal
with machine oriented problems?

                                                                 Model Guidance
   Next, the document discussed how to deal with errors that are found. How are
the changes formally made?

   Finally the document proposed what the documentation should consider.  For
example, what are documentation standards or what are the criteria for a user's
guide.  The document included a suggestion that there should be EPA guidelines
on model use, and there should be EPA distribution of the models.  There was a
suggestion  that there should be EPA review of all new  techniques.  When dealing
primarily with Gaussian dispersion models, EPA has already addressed many of
these problems in terms of the documentation and in terms of attempting to answer
the question of how you know a good model when you see it.

   There are two parts to my presentation. First, I will describe our regulatory
modeling guidance.  Second, I will discuss some of the qualities of a good model.

Model Guidance

While my personal experience is primarily with Gaussian models, my organization
does have some experience with  more chemically oriented models.  I believe  that
the present Agency  regulatory procedure could be applied specifically to chemical
mechanisms or models that include chemical mechanisms. First of all, let us discuss
the availability of these techniques.  All the Gaussian models that are to be used in
regulatory programs are a part of UNAMAP; these  cost approximately $1,000 to
obtain.  In  all cases, there are user's guides for these models. There are  codes for
the models and those that are most appropriate for regulatory application have a
default option. Throw a single switch in the model and it then operates in a specific
mode that has been subject to public comment and criticism. In addition, there is
a test data set to go with these models to make sure that new users can operate
them as the developer intended.

   Perhaps I could  introduce here the concept of documentation standards for
models. Bruce Turner, jointly with a contractor, prepared a handbook on how to
prepare user's guides for  air quality models. So, in terms of how to document a
model, i.e., preparing a user's guide for it, there is guidance already available.

   For the models  presently applicable to  regulatory programs, there are codes,
user's guides, and test data sets. I see no reason why this could not be done for
chemical mechanisms that this group wishes to entertain.

   With regard to guidance for regulatory programs, we do have a guideline on air
quality models that specifies data bases and models to be used, and within  that

Model Quality
guideline, EKMA and the airshed model are specifically included. Since we can
not write down everything in one document that indicates how to use the models
for regulatory programs, we have a model clearinghouse, whereby the EPA regional
offices and states can ask for clarification or for guidance on how to treat a unique
problem.  We are in the process of setting up or expanding such  a clearinghouse
activity to include EKMA.

    You should also be aware that in each regional office, and for that matter in
most states, there is somebody who is specifically responsible for modeling. In the
present situation, when we change the modeling guidance, we have  to go through a
process of public comment and review as part of the regulatory process.

Model  Quality

What was the criteria for  the models and data that were included in the modeling
guideline? We tried to include the models that were considered to be the best. The
first criteria was an evaluation and scientific peer review of the model.

    This evaluation or peer review for many of the models did not turn  out to be as
definitive  as we would have liked, since the peer reviewers felt that the models were
dated and did not represent the state of the art.  The state-of-the-art  models had
not yet been documented and systematized well enough for a wide variety of people
to use them; so, in many cases, we had to include models that were less rigorous,
scientifically.  They were,  however, the best-documented, off-the-shelf items that
could be used now.  Where the evaluation and peer reviews of these models were
ambivalent, recommendations were based on  consistency, public familiarity with
techniques, and use in the past.

    A question was posed  before, "Why can't  there be more than one  model used
in regulatory situations?"  The answer is that  those being regulated have a natural
tendency to use techniques that are credible and give them the least control require-
ments. If that is different from another source  downwind, then we find  ourselves in
a regulatory consistency issue on a national level. Congress has told EPA that  we
need to be consistent. So, in answer to the question, "Why can't  two models be
used?," the answer is to be consistent we must use only the best one. Other factors
are availability and resources.

    There was a question  in the background  document relating to  changes  made
to the model, whether they are associated with running on different computer, or
making a modification to a code, to satisfy particular purposes. In several cases, our

                                                                  Model Quality
approach has been to set up an equivalency test, where we provide a set of data and
say, "If you can run your model with this set of data and come up with essentially
the same answers within a certain percentage, then the codes are considered to be
equivalent."  This applies whether it's a whole different model that you constructed
on your own or just changing from one machine to another machine. Thus, we have
set a precedent for what we term a equivalency test.

    Where the user believes that the model suggested by EPA is not appropriate,
there is a mechanism by which they can do an on-site evaluation and thereby show,
for their data base and their source, that another model is more appropriate. That
approach has been used in a number of cases.

    Finally, for situations where there is no  model recommended, there  are some
general criteria that we use. The model used must be scientifically sound, be one
for which something  about its accuracy is known, and be one  that we know  does
not underestimate the design concentrations for an area.

    The steps described above are, in general terms, the  program that we  have
underway; they fit with the needs expressed in  the background documents for this
workshop. In other words, these are things suggested for consideration that we are
successfully implementing.

    Where did the models included  in our program come from? In 1980, we  pub-
lished a Federal Registry notice that asked for models to be considered for use in
regulation.  We placed six requirements on those models, however.

    First, a  model had to be  a computerized  code unless  it  was so simple  that
calculation  could be done on  the back of an envelope.  A model should  not be
limited to a paper published in a journal with a lot  of equations. That might be the
basis for a model, but it does not constitute a completely implementable model. It
has to  be a computerized code. Second, there has  to be documentation in a user's
guide.  Third, there  has to be a test  data set, so that other people can run the
model and test it. Fourth, it has to have an air pollution control application. Fifth,
there has to be some comparison with observed data so that we know how good it
is. And, sixth, there has to be public assess at reasonable cost.

    Most of the models that we received were Gaussian models, for single  source or
urban applications. We also received reactive models.

    A formal model evaluation has  been  performed for  five categories of models:
rural, urban, complex  terrain, long-range transport,  and mobile. All the models

Model Quality
were applied to one or more data bases and a comparative evaluation of the models
was conducted for those data bases. The performance was statistically evaluated for
each model for the same data. For the first three categories, we had the American
Meteological Society (AMS),  under a cooperative agreement with EPA, do a peer
scientific review of those models  based on the performance evaluation and their
scientific assessment of the  models.

    The peer review for the first  three categories is complete.  The AMS is now
considering peer reviews for the long range transport and for mobile models.

    What was the outcome of this peer review? I have already said that, for the rural
models, the available models were not state of the art—we could not distinguish
among them—they are all 20 years old. I will comment that one of the peer reviewers
was a very bright guy, and subsequently submitted a model that fulfilled all the
recommendations of the peer reviewers.  Unfortunately there was a fatal  flaw in
one parameter that he chose, thus his model did worst than all the other models.
I think he has since corrected that parameter.  The point I wish to make is  that
just because a model looks better scientifically does not mean that it necessarily
performs better.  This is related to one of the questions which was posed  to  you,
"Is a model a good model because it is scientifically satisfying?" I have had several
experiences with scientifically satisfying models that performed worse than  models
already in hand.

    In conducting the performance evaluation, we exercised care.  For example, we
wanted to avoid any situation where the developer would say, 'You did not run my
model right.' So,  we had the developers first review the data bases, and make any
coding changes in the models necessary to accommodate those data bases. Then
they verified that we were running the models as the developers would have us run
the models with the test data set.  Next the developer selected the specific elements
and parameters to be used in the model and the options. We wanted to run the
model just like the  developer thought it should be run. The developers were not
always happy in the end, sometimes they said, "Oh, I made the wrong choice."  But
for most  cases, this procedure worked well. Finally, we allowed the developers to
review the findings, comment on the draft final report, and make any final changes,
short of changing the way the model was run.

    In the urban situation,  the reviewers said that the four annual average  models
had performances that were fairly narrowly grouped,  and they again  could not
distinguish among them. That is, the annual models, which smear out a lot of
information, did not perform differently in a statistical sense and had much of the

                                                                   Model Quality
same theory. For the two short- term models, the peer reviewers essentially said
that they have similar abilities as the rural models, but they thought that one was
slightly better.

    For the complex terrain models, the peer reviewers did feel that one model ap-
peared to perform better than the other models, and, as a result, EPA has requested
public comment to  include this model in our modeling guidance. And that public
comment period is about to close. So we have reacted to getting better science into
our guidance.

    I have one final comment. Jeffries began his background document with a quote
by the GAO of an EPA official that said something to the effect that with models
"more precise results are unlikely because of the limitations of the science." I think
I am the official quoted there, but the GAO auditors did not interpret me correctly.
Essentially, what I said was that there is beginning to evolve concern that even if we
had a perfectly accurate model, because of the atmosphere's inability to replicate
its turbulent state, one may not know he has a perfectly accurate model. Because of
turbulence in the atmosphere, we may  have different realizations of the same event.
Models calculate a mean for a specific event, while the atmosphere does not. For
each replication of the same event, the turbulent nature of the atmosphere may give
us a different answer. Therefore, in dealing with the evaluation of models, one has
to deal with the  accuracy of the data bases, the physical realism of the models, and
the natural variability in the atmosphere. Due to natural variability, we may not be
able to  establish that we have a better model even though, scientifically we might
believe we should.  The AMS, under the EPA cooperative agreement, is struggling
with this problem and I presume that atmospheric chemistry and its problems just
compound the problems.

               Discussion Following Tikvart's Presentation
    Demerjian: Joe, given the fact that EPA has accepted as a guidance criteria
modeling technology that is twenty years old, has limited data sets to evaluate the
models, and has presumably identified certain kinds of performance statistics based
on these, why would EPA want from the chemical models more than, let us say, a
factor of two and a half? Actually, I am sure we do better than two and a half.

    Tikvart:  Tesche made a presentation on the airshed model at the  conference
here last week. He reviewed ten applications and his conclusion was that the airshed
model was accurate to within plus or minus 30 percent. Based on experience we
have had, I'm inclined to  agree with that.  The airshed model does appear to be
accurate within plus or minus 30 percent. It does tend to be on the low side, and you
can argue about why that  is, but that certainly is as good as the Gaussian models.
Maybe there is a better chemical mechanism since we have been using Carbon Bond
II in the version of the airshed model that has been evaluated. Perhaps there could
be a better chemical mechanism, a better way to handle the numerical calculations
for dispersion, and so forth; but then there is a need to test that version  to find out
that the better science does, in fact, give you more accurate answers. I'm not trying
to say that we do not need better science but the science we are using  appears to
be not that bad.

    Demerjian:  When  you looked at the point  source models, I remember that
basically there was a lot of discussion about the fact that they were pretty much all
the same, that they were all mostly based on Gaussian approximations. So people
that were dealing in the science  itself, whatever science was there, were basically
saying,  "They are all the same except for a few  thousand whistles on how to treat
this or that." Then there was another issue about whether the data that was being
used for the observations could be even defined within those bells and whistles of
those models which were better.  But, isn't it  true that, if they were used in a
regulatory sense, some of those models might require more conservative reductions
than others, or do they also do that the same too?

   Tikvart: No. Three models, all of which perform the same within your ability
to make a statistical distinction, could give three different requirements for control.

   Demerjian:  Given that, how did you make a decision in  terms of which would
be the best?

    Tikvart: We have reverted to past use and familiarity with codes. That was
our position. Scientifically, you cannot distinguish between the models.  The peer
reviewers essentially refused to distinguish between them.

    Demerjian: So, you use precedence then?

    Tikvart: It was precedence. In other words, if you have to have a consistent,
stable regulatory program, and one of the models—the model that has been the
basis for your regulatory program up to now—does not perform differently, and at
least as good as  other candidate  models, why should you switch?  Now there are
cases where somebody might get a more desirable answer with another model. You
do not want to foster inconsistencies by  saying go ahead  and use whatever model
gives the answer you  want, unless for that specific plant or facility it can be shown
that the model  performs better.  There  have been several cases where somebody
has taken on-site data and performed an evaluation for a regulatory model and one
or two other models  and used the model that performed best. The chosen model
was not the regulatory model in these site-specific cases.

    Demerjian: Ok, hypothetically we have the EKMA mechanism, we have some
version of Carbon Bond, we  have a  version of the SAPRC/ERT model and there
might some others. Presumably we  go through a similar exercise as you have de-
scribed and we come to the  conclusion that within the uncertainties of the data,
where ever it may come from, whether it is the Riverside chamber or Harvey's
chamber, that they all fit within the limits of uncertainty  of the information. And,
from what we can tell, they're all based on pretty much the same standard sets of
scientific information and they  all perform equally well. Now we discover when we
exercise them that one requires 20%  more reduction in hydrocarbons than another,
or 30% more. Now, what is your position at that point?

    Tikvart: You've  gone through a performance evaluation  and  a peer  review.
Right?  In  a sense we have already established for the regulatory  program what
to do because the EKMA model is the guidance for a screening type of discussion,
although it is good enough to base a control strategy on and the sophisticated model
of choice in the modeling guideline is the SAI airshed model.

    Demerjian: Let's  just  talk about the mechanisms and not the  airshed model.
Let's say we're talking about mechanisms that can plug into EKMA. Even if right
now, if there was a consensus opinion that whether it is the Dodge chemistry or any
of these three others that they all were indistinguishable in terms of a performance

    Tikvart: Then why not use EKMA and OZIP the way it is now?

    Demerjian: Even if it turned out that EKMA was, let's say 20% or 30% more
conservative. Would that ever be factored into your 'decision?

    Meyer: Let me try to get Joe off the hook a little bit here. Our thinking right
now is that, we are most probably going to change the recommended mechanism.
I think probably  the major basis for the recommendation is the more up-to-date
experimental work and theory that the newer mechanisms reflect. We're going to
try to select one mechanism however, for use in regulatory applications. And, our
thinking is basically that this will be used for perhaps a 3 or 4 year period at which
time there will once again be a reassessment by the scientific community and the
same kind of question will be asked, should we change this mechanism?

    Demerjian: Ok, let us move up to this next tier then. We have eliminated the
old version mechanism in EKMA. Now we can move to some new mechanism. Now
we have the same scenario: three mechanisms, presumably all state  of the science,
presumably all show reasonable performance credentials in terms of data. But now
still showing this difference in terms of policy requirements, in terms of control.

    Tikvart: Have you already made the change or are you in the process of selecting
the new mechanism?

    Demerjian: We are in the selection process and  we discovered this unfortunate
problem: three mechanisms, all performing within the standards of the data.

    Tikvart: All three of them are better than what you have now; you know that?

    Demerjian: We as a group are reaching  a consensus. We have established that
they are all  definite advancements over where we were.  Now we have discovered
that even  though it is still within the experimental data, there is enough leverage
within this to give different control requirements.

    Tikvart: The regulatory blackboard in that sense is sort of blank and prepared
for the scientific community to write on. We would ask the scientists to give us
your best  assessment of what this most scientifically credible technique is.

    Meyer: One other dimension for this kind of problem is what kind of information
is required by the mechanisms that regulators would have to try to come up with
in terms of inputs to  the mechanisms.  I think perhaps one factor that might be

considered is that, although two mechanisms might be equally valid as far as we
can make a judgment on, if one requires a lot of information that we have to wave
our hands about a lot, presumably the other one would be preferred.

    Tikvart: And here you get into terms like familiarity, resource requirements, ease
of use, etc.If you honestly cannot distinguish between the techniques scientifically,
then I think you degenerate to those sort of criteria.

    Demerjian: Again, we have a curious situation. We have one mechanism that
requires 20% more control. Now, we use this criteria that has just been given and
we find out that the one that requires 20% more control  actually is  a lot easier to
use. It turns out that you can pass it out in a form that runs on a PC computer and
cities  only have to collect a few data points and they can run it. When the people
use this tool they  find out that that this  extra 20% control is costing  them $100
million. Now, what's your justification for making the decision that  presumably it
is the application part that helps you to decide which model to use?

    Tikvart: Then there's another possible answer and that is the models go out for
public comment. Since people are going to regulated by that model, present the facts
in the Federal Register notice and say that you are seeking input on what direction
you should take.  That is one alternative. Since it seems that the blackboard is
blank, you can say, "It is your choice, World, go with whatever approach you want
to use."  However, very quickly I believe you will get into a  situation of inconsistency
and irregularities being jerked back and forth by what is most convenient or by what
is least costly to specific interests. This could result in one model in one part of the
country and another model in another part of the country.

    Meyer: I think another point too, to raise is, that it is probably unlikely that,
across the board, one model would predict 20% less control being needed than the
other one.  And, indeed if there is a difference, the difference may hinge on some
of the inputs to one  of the other models, for which there might be a great deal of
uncertainty. I think  what  we would want  to do would be to select the mechanism
that then has fewer of these very uncertain inputs.

    Lloyd:  Part of this discussion leaves me very confused.  We are meeting, as  I
understood it, to try to resolve some critical questions in the chemistry area and to
help out  EPA. Now,  I am hearing from Mr. Tikvart that EPA does not  necessarily
need the best mechanism, but probably the most readily available, best documented


    Tikvart: No, first of all we want the best mechanism.  I do not think there's any
equivocation about that.

    Lloyd:  But your time frame, as Mr. Meyer was saying, is not going to allow
you to use the best mechanism. You are talking a 3 to 4 year lag time in terms of
identifying mechanisms and having any changes.  Our earlier discussion led us  to
believe that evolvement of a model is a constantly changing thing as you get more
input data. If in fact there's no way in which you can allow changes for 3 to 4 years,
then by definition you're coming up with something which is outdated.

    Tikvart: I have a couple of comments on that. First of all, does the scientific
community know when to let loose?  This is a complaint  I have; I do not  think
that the scientific community does know when to let loose of a new technique.
Frequently, they have to be forced to let loose.  I should not say forced, I  mean
preempted.  And then somebody may take it and do something dumb with it. So,
the scientific community is oriented to continue improvement; it is difficult to decide
when this is the best shot. Another comment is that just because somebody thinks
he has an improvement, does not necessarily mean that he actually does; it might
be just one interim step to something better. So, I believe there is merit in having
some lag between when a new idea pops  out and when it is actually implemented.  It
dogs need some testing before you go jerking around multi-million dollar programs
and just saying, 'Here is a new technique.' A perfect example is work on plume rise
equations.  At one point, about 10 years ago, you  would have been changing the
plume rise equation used in Gaussian models every six months, as new information
evolved.  I do not  think that is in anybody's interest. The third point  is, that just
because somebody think's they  have a  better technique,  doesn't necessarily  mean
that they have a better technique until it's evaluated and compared to other existing
and new techniques. So, there are some tests I believe that are appropriate before
you impose new technology on the regulated community.

    Lloyd: One of the points, I think we must make clear here from the chemistry
point, is we are not just dealing with the ozone as we have in the past with some  of
those. We have to look ahead and we are already addressing many of the problems
in the acid deposition area. Secondly, one simply question. Supposing with the end
of this meeting we came up with a consensus,  best mechanism. How long would
that be before EPA would use that?

    Tikvart: Ok, I believe you would have to address some of the questions I just

    Lloyd: Supposing we did.


    Tikvart:  Addressing them  includes a user's guide,  codes, documentation in
general, evaluation and peer review,  all of which take time.  Perhaps some of you
have data you brought with you. But, let's say all of this has been taken care of
and you now feel you have the  best technique.  First of  all, as a regulator what I
would want to do is, see what that new technique does to control programs. Is there
a major relaxation or tightening of the emissions required? That question is asked
first because when you put this out for the public, they need the answer. You would
have to go out for public comment and announce that there is a new technique that
EPA wants to use.  Here is the information on how accurate it is.  Here is what
the scientific community thinks  of it and here is what it will do to emission control
requirements.  World, comment  on this as a regulatory tool.

    Lloyd: So, what are we talking about in a time frame?

    Tikvart: I think we are talking about 18 months.

    Jacob: I am quite uneasy with the distinction between urban and rural models,
especially as we move towards more regional problems. What is a rural air quality
model? Some rural areas have NOX conditions which are typical of cities and other
times have NOX conditions  which are typical of a  clean troposphere, so in such a
situation, what would  constitute a rural model?  I  think a  much more objective
criterion  that  we have got to  have  in the model documentation is what is the
range of  NOX that it can be applied to.  What  is  the range of hydrocarbons can
be applied?  What  kind of hydrocarbons can be  present in the system without
making the model go haywire?  I think all these questions must certainly be a part
of the documentation.

    Tikvart: I agree with what you are saying. If the model developer would lay out
specific limitations on the applicability of the model that will help the regulator who
wants to use this technique in a credible scientific framework. What happens too
often is there is a regulatory need for a model, the regulator interrupts a research
program and says give  me what you have now because I need it. Then we take it
and apply it to a problem that  it may not have  really been intended for but there
may be no other choice. So  the burden of responsibility for the model is transferred
in that case from the developer to the person who wants to use it. I think we would
much rather see the burden of applicability be on the person who developed it. So,
I agree with what you are saying.

    Whitten: I just have a couple of comments. First  of all, in the beginning of your
talk you wanted to ask the  question how do you recognize a good model when you


see one? I think that that is a very difficult question to answer, and we can make
more of an answer to a related question anyway, which is, 'How do you know a bad
model when you see one?'  I think in the archives of data that we have, we have
certain types of experimental conditions we know of that present grounds to reject
certain types of mechanisms that have been used in the past because they just can
not simulate a certain type of condition. Of course, you have to ask the question,
'Are those conditions of any general applicability?'

    Secondly, at the end of your talk, you mentioned about the randomness of the
atmosphere and it is true that meteorologically speaking there is a turbulence and a
randomness and that makes it difficult to determine where a certain pollutant might
be going, where a certain parcel of air is going. The nature of a chemistry model
does not have that random aspect to it.  It simulates what happens in a well-mixed
parcel of air.  So the randomness then applies as to  we do not know where that
parcel of air might go on a given day because of the turbulence or whether or not
it might  be perfectly well mixed, so that the chemistry is much more deterministic.
It does not have that randomness. What randomness it does have is more in the
eighth or ninth decimal place type of thing.

    Finally, I am a little troubled by Meyer's comment on discounting models which
need more data. I think the general approach has been, as you have outlined, to have
sort of a  default for various things. And in the  case of atmospheric mechanisms, the
more recent ones have many more species to define the various aspects of reactivity
that we see in the atmosphere.  The original EKMA for instance, more or less, fixed
the reactivity so there was no way to change it.  I think the newer mechanisms can
be used for a default type of reactivity and so therefore that is always there.  So you
still do not need any reactivity data. If you have some  reactivity data which shows
a given urban area or something like that, is far  out from the norm, then more
recent mechanisms have the ability to accommodate that, but it is not necessarily
a data requirement that use them all in the first place.

    Carter: You were talking about wanting to keep on using the existing model if
it performs adequately, but what do you do in a case when you get new information
that shows that the theoretical underpinnings of the existing model are wrong?
Like, for  example, the original EKMA model. One  of the most important reactions
in it and was subsequently shown to have a rate constant that was off by an several
orders of magnitude, yet that model is still being used. I was just wondering what
the attitudes are.  Do you continue using it because it seems to perform adequately
on the validation data sets or what?


    Tikvart: I guess there are a couple of concerns that I would have. First, how
serious is the  deficiency?  Is there some other better choice? And given the de-
ficiency and the other choice, how accurate are both when you compare them to
measurements ?

    Carter: Let us say that we have the other newer models which have not been
used as long but that they have that particular deficiency corrected.

    Tikvart: Are they more accurate when compared to observed data?

    Atkinson:  We know that they are scientifically more correct.

    Tikvart: Are they more accurate though?

    Carter: They are as equally accurate.

    Atkinson:  They fit the same environmental chamber data base to the same
degree, and one is scientifically more accurate.

    Tikvart: I would have to say that the criteria that I have used relative to the
Gaussian models, is that if you have something that is scientifically  better and it
performs as well as what you have now you should take steps to make a change.

    Meyer: I agree with that. I think that maybe our concern was that these would
not be continual changes that they would be done in discrete periods, three or four
year periods so that we are not constantly changing the tools that we are trying to
use for regulatory purposes.

    Tikvart: But, just to make sure that we are clear, let's assume that if you came
to me and said, "There is something wrong with what you are using now and I have
something that I know is theoretically better."  However,  unless you  can show me
that statistically it performs better, I would say come back later  after you know

    Carter: But if this discovery showed that the existing model was sort of invalid
at the core, would it at least accelerate the process of finding a new one?

    Tikvart: Perhaps, but this  has happened more than once. The scientific com-
munity has said, "What you use is twenty years  old! We  do not think the answers
you are getting you are getting for the right reasons! And, Oh by the way, here is


something that we think fulfills everything that you want and you should be using
it instead."  However, I find out that, because of some misjudgement as to what
value to assign a parameter in this new model, it is grossly wrong, then you can
understand some caution and some need for proof of the better aspects of the new

    Jeffries: There is a hidden assumption here that we have all been making. Ken
started out making the assumption that all the models fit the data equally well.
A major point of my presentation is that we hide many of  the anomalies  that
are present in the system in the asserted "reasonable agreement" with the data. In
reality, if you examine each mechanism in detail it will make unique predictions that
can be validated, verified, or tested, as a way to choose one from the other.  But that
requires that we have an agreement among us as to what the data really are.  And
so, for example, in a case where there is a disagreement about the formaldehyde
prediction of one mechanism verses another mechanism, we presently can not resolve
the issue because sometimes Carter's measurements of formaldehyde are different
from my measurement of formaldehyde and we don't know which one is right. If we
concentrate on improving the data, we can separate one mechanism from the other.

    Tikvart: Are you limited just to smog chamber data?

    Jeffries: Mostly.

    Tikvart: OK, that  is a tougher problem I would think.  If you could take a
dispersion model and test a variety of chemical codes in one framework, that might
provide you with  a test. But if you don't have the ambient data, then ...

    Jeffries: Most of the time the problem  is difficult enough trying to unambigu-
ously compare the data with  the theoretical predictions.  In the  atmospheric model,
there is ambiguity about what is causing the change in concentration because it is
a non-linear interactive process and so even though you may have the measurement
out there you do not know that it is the chemistry that is causing you not to pre-
dict that number correctly. It could be a transport, or a chemistry, or an emissions
problem and this is too complicated.  So you must resort, you  must go back to a
simpler system to juxtapose unambiguously the data, the observations, with the
predictions.  It does not do any good to compare observations with predictions if
the situation is so ambiguous as to cause and effect.

    Bradow: I think  that the use of chemical models in the acid rain program to
predict products  of other than ozone may be  the key  element in creating  some


of the differences in discussion here. In other words, it may be possible that all
the chemical modules do reasonably well in predicting ozone in those cases that
are most important for controlling ozone.  This  may not be the  case if we use
these same chemical modules, in  the case that we wish to evaluate the importance
of the oxidation processes in creating hydrogen peroxide and nitric acid and thus
influencing acid rain. A key element that Jeffries speaks of, and others here too, has
been how well do these models predict products other than ozone? It is possible that
engineering designers figuratively have built a birch bark canoe in order to control
ozone and then find themselves faced with transporting Noah's Ark animals.

    Atkinson: I would like to make one comment on that and actually it bares on
Jacob's comment about the developer putting on the areas of applicability of his
model.  If you use the  data base we presently have, unfortunately I  am not sure
that you could ever say that you test a model for acid disposition against certain
environmental chamber data and  I can see that it fits with your comment. We really
are extrapolating, I believe and I may be wrong, but I believe we are extrapolating
quite a way beyond  our data base.  I mean we understand hopefully, pretty well,
the chemical reactions which occur but we don't really have  a data base, a global
data base to test them against. Do you have a comment on that Daniel?

    Jacob:  I think I quite understand the problem with smog chamber data, but I
just want to  point out that it  is  extremely dangerous to extend such models that
were obtained at high concentrations.

    Atkinson: I fully agree. I mean anybody who has to have  NO in their chemistry,
clearly it has to be applicable to nighttime conditions or multiday conditions. There
are two sides of the coin. There are always problems in extrapolating data for the
long-range  consequences.

    Jacob:  Yes, I think aside from  the fact that the smog chamber data does not
cover a wide range of  compulsory conditions, there is also  problems within the
development of the chemical mechanism itself. Based on smog chamber data which
is done satisfactory at pretty high concentrations you can say, 'I understand pretty
well how the system works and even if the concentrations are lower that is what
I should get.' But we all know that there is a big difference  between for example,
the low NOX and the high NOX  region and the chemistry imbedded in the chemical
mechanism may just not work.  Aside from not  having  the observation data to
support it is the fact that based on chemical principles it will not go below a certain
range of NOX.


    Atkinson: Or two mechanisms may give very different results under those con-

    Dimitriades: I am not sure that we have been reacting to Joe Tikvart's presenta-
tion in the right context. I think what Joe has been saying is that what we are doing
in the dispersion model part, which would perhaps be applicable in the chemical
mechanism area also, is that if we have any questions on the model's performance,
or if we want to choose a model among two or three based on performance, we do
two things. In essence, we are  using existing field data processed according to the
American Meteorological Society procedures to derive some statistical measures of
the performance of the models. Then we throw all this information into the hands
of a group of reviewers and we ask them to look at them and give us their judg-
ments. Out of five reviewers you may  have three reviewers that say yeah and two
saying nay and the yeahs have  it and that is the judgment. That is the outcome of
this process.  What we are dealing with  here in this workshop is a different thing

    I think we are concerned,  number one, with the  data base that was used  to
derive statistical measures of the performance. And this data base may be good  or
bad, of appropriate comprehensiveness or not, etc.The other question that we are
dealing with is 'How do the reviewers  go about making their judgments?' If it is
subjective judgments, we do not want  them.  Therefore, we want to come up with
some guidance in essence telling the reviewers how to go about judging the data  so
that they come up with a valid conclusion. Thus another group of reviewers should
come  up with the same judgment.  Or if the same reviewers are asked the same
questions one, two, or three weeks later they will come up with the same judgment,
provided they have this procedure by which they go about making their judgments.
So, Tikvart's procedure is something which is perhaps acceptable in some respects,
but we are dealing with something entirely different. It is not a question of whether
his procedure is applicable to the mechanism area or not. We are dealing with an
entirely different subject and entirely different questions. So in a way what we are
aiming in doing here presumably will be used in the process that  Joe Tikvart has
described and the two things then should be considered complementary. Would you
agree  with that assessment?

    Tikvart:  Fine,  I think the  key that  I did not understand earlier was the im-
portance of the data base question as  you just  described. There is some question
as to  how to  interpret  the basic data  used to develop and evaluate the chemical
mechanisms themselves.  Whereas in the straightforward evaluation we took the
data as absolutes, even though we knew there was some error.


    Atkinson:  I have a further comment which is that on  the one hand we  are
being asked to look into how we get more accurate chemical mechanisms, gas-phase
chemical mechanisms, and  how we go about that; this in a way is the purpose,
or part of the purpose, of this Workshop.  On  the other hand, what I heard this
morning would lead me to believe that even if we  do develop them  they may well
never be used, just because of time scale of the EPA. How is that going to impact

    Tikvart:  I do not think  that should be a concern. If there is a better technique
to be had, I think the people involved in Research  and Development should find it
within available funds. And those of us on  the regulatory side get enough pressure
that we will want to have the best science out there. I  would hope that eventually
best science would prevail even though there are those with  different stances, etc.,
that may find other approaches better. But I would hope that ultimately the best
scientific technique, given that we can demonstrate that it is, would prevail.

    Jeffries:  Let me see if I understand what you are telling us.   A lot  of  the
discussion has been premised on the assumption that the mechanisms all fit the data
equally well. If it turns out that we could demonstrate that a new mechanism based
on better theory or more generally accepted theory could also be demonstrated to
fit new data better than some other mechanism, would it be the choice?

    Tikvart:  I think the answer is yes, if I understood everything you said.

    Jeffries: If we can narrow the range of  reasonable agreement between observa-
tions and  predictions from  mechanisms to  a point where we believe this one is a
better representation than that one, we accept this one over that one and you will
essentially be forced to do that because that now becomes the definition of better
science.  You do not want to go forward with a mechanism that is not good science.

    Tikvart:  Ned, have you heard anything here that you would find fault with?

    Meyer: Well,  the only thing  I guess  I would like to say again  is that barring
some major breakthrough, our inclination would be to make these changes  in dis-
creet increments.  That is,  to say we would look to see all the developments that
have occurred in the last three to four years, try  to get some judgment from  the
scientific community about  what the best state-of-the-art is  at that  time and then
try to incorporate those recommendations into our regulatory procedures and then
essentially freeze our regulatory procedures for three or four years while science goes
on. We would then have another one of these reassessments  in three or four years.


The only exception to that I think would be if some major breakthrough came in
that showed that what we were doing was just entirely wrong.

    Atkinson: Seems like an awfully long change over this time scale, three to four

    Jeffries: Ned, I think it might be  useful for you to say how many SIPS have
been done with which mechanisms.

    Meyers: They have almost all been done with the Dodge  mechanism, which is
ten years old. Or more.

    Tikvart: But can we say that in the interim there has been a clear agreement
that there is something better?

    Dodge: Oh, yes!

    Jeffries: Yes.

    Whitten:  I think that perhaps we should be a little more clear in our definition
of what data we are talking about here. The Dodge mechanism ran into trouble
not so much from smog  chamber data  but from laboratory data  which had shown
a particular rate constant or two rate constants were in considerable error and that
was based on laboratory data, not smog chamber data. And then, when those new
rate constants based on laboratory data were then put into the mechanism and you
went back, the procedure used say to adjust the propylene to butane ratio could have
been readjusted and that particular type of mechanism could have been updated at
that time. But it was felt that the, at least I  believe, it was felt that the mechanism
wasn't that far off.  With its compensating shall we say errors between a rate
constant which  didn't agree with  laboratory data and  a percentage of propylene/
butane there was an empirically adjusted to fit smog chamber data. So it was felt
that the fit of the smog chamber  data wouldn't be improved significantly enough
if you stayed with the old rate constant and the old propylene/butane split. I just
wanted  to use this as an example. I think it shows that there is laboratory data
that is really different from the smog chamber data and they are used in a different
way and I think that we just need to be conscious of it.

   Lurmann: I am still curious as to  how the EPA decides  that one mechanism
that is the chosen mechanism to use in their planning and so forth.  Is there a formal
review process,  what evaluation criteria have been used in the past and so forth.


Some of the same issues that we have been want to deal with in this conference
today have been  dealt with in previous situations and I guess that is really my
question for now.

    Meyer: Well,  at the time we have to make the decision, which again is almost
ten years now, we obviously did not know as much about the sensitivity of control
estimates and chemical assumptions that go into these models, and from an OAQPS
standpoint, we asked for the research people's best judgment, you know, what is the
.best available mechanism at the time and the recommendation was this propylene
butane mechanism which since became known  as the Dodge mechanism.  I think
one of the major  purposes of this Workshop as we pointed out this morning is to
come up with some more formalized rules that might be used periodically by the
scientific community in making/helping EPA make judgments about,  all right, as
of this point in time what is the most appropriate mechanism for use in subsequent
regulatory analyses and I guess the point I wanted to make was that this would
probably have to be done  on a periodic basis.  I suggested every three  or four
years. In the meantime, of course,  there will be intervening research and people's
perspectives maybe will change between iterations of this group. I don't really know
what all the criteria should be.

    Lloyd: Just for clarification here are you saying that the next  round of SIPS
that the Dodge mechanism will be used and that you are looking  for input from
this group to see what criteria should be used for replacement?

    Meyer: No, I  think that for the next round probably we don't know yet but I
think it is more than likely  that we are not going to be using Dodge.

    Lloyd: So then what set of criteria are you using to replace that mechanism?

    Meyer: Well, again, I will have  to pass the buck and in the sense that we have
asked our people  in the Office of Research and  Development what their judgment
is about the best mechanism at this time.

    Dimitriades: Actually, what Ned is going to do is he is going to  come to us
(ORD) will ask us, 'What is it that  you recommend at this time?' This is what the
practice has been in the past.  We  have to look at the existing mechanisms with
some help from consultants. We may organize workshops, (had an EKMA workshop
three or four years ago  if you remember)  and we use all this information internally
for our own purposes. Then we derive a judgment and we pass it back  to QAQPS.
This has been the process in the past. Now, in the last two or three years there has


been a lot of emphasis on having everything that we do peer reviewed in a more or
less systematic fashion. So I expect that any recommendations and any suggestions
that we will pass onto the regulatory program will have to be supported through
some kind of peer review process.  But we still have all the faults of the previous
process in the sense that we don't have a standard way that everybody would agree
is the best way of judging a mechanism. That we do not have this we recognize as
a lack and this is what we are trying to do at this workshop. But our process so far
has been relying mostly on our own judgment which was helped  through the use
of consultants, workshops, but in these workshops the judgment was not left up to
the participants to make. We simply used the advice and took it  as best we could
and made our own judgments.

    McRae:  I want to ask Basil and Joe a more general  question.  Part of what we
are doing in this Workshop is to think about the future and one of the things I am
concerned about and I am curious about your reaction is as mechanisms becoming
more complicated, and data requirements becoming more complicated, the process
of peer reviewing models and mechanisms is becoming much more difficult. Have
we thought about the question of whether there is enough people out there to carry
out these tests?

    Tikvart: Are there enough different people from the developers who have the
knowledge to do an unbiased review?

    McRae:  Well, let me say it in a more direct sense. There are twenty-five urban
areas that currently violate the Federal Ambient  Air Standard for ozone.  In each
of those regions how many people do you think there are that understand how to
pick the difference between two competing chemical mechanisms and translate the
evolving knowledge of chemistry into data collection procedures within their local

    Meyer:  I think that is exactly the reason why we find ourselves in the position
of having to recommend a mechanism. To take that prerogative basically out of the
hands of the State and local agencies.

    Tikvart: The answer is, "No there  aren't very many people out  there." What
typically happens is  the area  hires a consultant and the consultant imposes  his
knowledgeable assessment of the situation.  That consultant  tends to drive it the
way  he feels  it should go and that way generally, because  of the interest of his
client,  will  result in lower controls. So I  think you're right, there aren't enough
people out there to have a lot of freewheeling discussion and assessment and there


are only a few that can do the job, but that is a fact of life. So I am not sure that
there is anything that can be done because, as you pointed out, it is getting very
complicated.  The number of people who have the knowledge to assess it is small.

    McRae: Doesn't it argue for things like training programs?

    Tikvart: Is this so complicated though that training programs to understand
the mechanisms might be of limited use and training programs to apply a basic tech-
nique like EKMA might  be more appropriate. I would question a training program
to teach people how to use the Urban Airshed model because it is so complicated.  It
takes a staff who have a wide variety of knowledge and abilities to run the model.  A
prime example is the New York State Department of Environmental Conservation,
which does have that ability.  But, I would say that  New York State is probably one
of the best states to take this job on. I won't expect that every state  could take
on the Urban Airshed model, nor would I want them to, nor would I try and train
them to. So yes, perhaps there is a desirability of training, but I am not sure that
understanding the chemical mechanisms in the right way at a level you are talking

    McRae: There is a lot more than just the chemistry. For example, if you look
at what's happened to  the structure  of the hydrocarbon chemistry in most [tape
change]  ... people who have to give  their emission inventories.  And I am a little
concerned that if we spend a lot of time worrying about getting the mechanism right
and not thinking about the  future implications of these mechanisms in terms of
emission inventories that we will lose  the whole ballgame. We have a very accurate,
highly characterized, precise mechanism based on smog chamber data but lousy
emissions data to drive that.

    Tikvart:   I am  more concerned about that point, about  the accuracy  of the
emissions inventory, than I am in understanding the chemical mechanisms. Ned,
to what extent has our guidance addressed the specifics of the emissions inventory
requirements? We  have attempted to address it to some  extent but, whether we
addressed it fully enough relative to this comment, I  am not sure.

    Meyer:  Well, I  think we have something  called the VOC speciation manual
which is basically set up I think to subdivide the organic emissions categories suit-
able for use with the Carbon Bond type of approach. You know, if another kind
of approach were made,  then obviously that kind of a document would have to be
changed so that  it would be consistent with the categories needed.


    Tikvart: I would be more concerned with the emissions inventory. Perhaps that
comment ties into something that Ned said this morning about what mechanism
to use. It has to be a mechanism that you could have data sufficiently detailed to
drive and the number of species, etc., that you can include in your inventory might
be one of those limitations.

    Lurmann: Yeah, I think I agree with Greg. The mechanisms are getting more
complicated.  You have eight, ten, twelve classes of hydrocarbons instead of maybe
four or five or six that you had in mechanisms of five or ten years ago. But quite
frankly, I don't see the relevance of choosing what mechanism to use just because
one is only has five classes of hydrocarbons and your data base happens to  be set
up, or your program set  up for only five.  I think  the approach that should be used
is that is to try to figure out which mechanism appears to be more accurate and
then build a program based on that.  The data species manual, it gives profiles for
speciation based on individual compounds. Those can be run into any chemical
class that you want. For example, the State of California keeps computerized files
for  three different mechanisms so that they are  not forced into  the situation of
having the speciation profile set up for one mechanism because they use more than
one and that  is the approach that I would recommend here.

                      Paper 3
Chemical Mechanism Development for Applications in
               Atmospheric Simulation
               Kenneth L. Demerjian
        State University of New York-Albany

                      Chemical Mechanism Development
                for Applications in Atmospheric Simulation


                           Kenneth L. Demerjian
                       State University of New York
             Presented at the Environmental Protection Agency

           Workshop on Evaluation and Documentation of Chemical
                   Mechanisms Used in Air Quality Models
The Mechanism Development Process

    The development of chemical mechanisms for the purpose of representing
the transformation of pollutants and background trace constituents in the
atmosphere has evolved over approximately a twenty year period and has
considered several principal components in the development process.  The
components that form the basis for the methodology of this development
process are illustrated in Figure 1.

    Historically, each of these components has assumed a dominant role in
the mechanism development process, somewhat reflecting the state of the
science during the various development stages.  For example, in the early
1970's at the onset of research and development activities in chemical
mechanisms for the simulation of atmospheric transformations in polluted
environments (Demerjian et al.,  1974; Niki et al., 1972;), a majority of
the elementary reaction steps were generally theorized from thermochemical
kinetic estimates based on the methods introduced by Benson in 1968.
Mechanisms were developed using data from smog chamber experiments as a
basis set for truth, that is, mechanisms were judged on their success in
fitting the concentration - time profiles of the experimental data.  The
assumption being that the smog chamber experiments provided a
representative analogue of the chemical systems operative in real world
polluted atmospheres, therefore allowing the extrapolation of the developed
mechanisms to simulating the chemical transformations in polluted

    Many critical elementary reaction steps were identified in the process
described above for which no laboratory chemical kinetic data existed.  The
importance of these reactions in understanding the mechanistic
transformations of pollutant species in the atmosphere created a forcing
function which stimulated laboratory chemical kinetic studies.   Rate
parameters and elementary modes  of reaction for a large variety of species
and reactions were provided by the chemical kinetic community which
introduced significant advances  and refinements in mechanisms in the late
70's and 80's.

    As the feedback process above was occurring, so also was a process
between the model development and smog chamber communities.   The
mechanistic modelers attempted from the start to develop and test their
mechanisms for as many chamber systems and data sets as possible.  In doing

so the modeling community recognized certain limitations in the data bases
and chamber systems they were utilizing and initiated some guidelines for
the smog chamber experimentalists.  This resulted in additional
enhancements in the mechanism development and represents an important
methodological component in the evolutionary process.

    The scientific community has,  on a continuing basis, been developing
and evaluating mechanisms against smog chamber data sets.  As the science
of chemical mechanism development has become more sophisticated, the
community's requirements for quality and performance have also become more
refined.  The feedbacks and intercomparisons that were components of the
methodological development process began to identify limitations in the
chamber experiments.  Scientist began to question the degree of effort one
should make to fit individual runs,  series of runs,  and runs between
different experimental smog chamber systems.  The debate arising from the
fact that many smog chambers whose data sets had been used for mechanism
development had not been adequately characterized with respect to wall
effects.  The chamber walls which can act as both sources and sinks for
important chemical constituents introduce noise in the chamber data.  This
limits the precision and accuracy claims which might be inferred by precise
fits of modeled and observed concentration - time profiles.

    In addition these limitations are not always explicitly characterizable
and are thought to ultimately contaminate mechanisms which have been
developed from them.  This contamination can take the form of inherent
noise or a systematic bias.  This phenomenon represents a significant
scientific challenge to the community and is the subject of this discussion
and the focal point of this convened workshop.

    The utilization of atmospheric observations as a vehicle for mechanism
development and evaluation is intuitively the most satisfying
scientifically.  But until recently, the instrumentation technology
necessary to characterize the detailed chemical components of the
atmospheric system were beyond reach. Also the complexities introduced by
the dynamics of the atmosphere introduced considerable uncertainties which
make diagnostic interpretations of mechanisms quite difficult.  But even
with these caveats it would seem that progress in instrumentation
technology and the importance of studying real world chemical systems
suggests that atmospheric observations of increasing sophistication become
a major component in the development of new generation chemical mechanisms.

Application of Chemical Mechanisms

    The Environmental Protection Agency's interest in the research and
development of chemical mechanisms of polluted atmospheres stems from their
responsibility to manage air quality and its associated environmental
effects.  The chemical mechanisms are a critical component in modeling
techniques which provide a quantitative relationship between the emission
of chemical precursors which react in the atmosphere both in the gas- and
liquid- phases and in sunlight as well as in the dark, to produce chemical
species of environmental concern.   Currently these compounds include:
ozone, nitrogen dioxide, fine particulate matter (primarily sulfates) and
acid bearing substances, and most likely the list will expand with time.

    A case in point and relevant to the subject of this workshop is the

modeling of the formation of ozone in urban atmospheres.  The air quality
simulation model incorporates a chemical mechanism in conjunction with
emissions information and some treatment of the transport and diffusion of
the chemical species under study.  The model is exercised to provide
quantitative guidance as to the amount of precursor emission control (non
methane hydrocarbons and oxides of nitrogen) that is required to meet a
specified concentration of ozone.  At issue is the precision and accuracy
of this quantitative relationship, methods for its evaluation, and
standards of acceptability (or success).

    Chamber noise discussed in the previous section has important
implications on the limits of precision and accuracy of the quantitative
relationship desired in regulatory applications.   If the chamber effects
are unknown or incorrectly specified, the result is a chemical mechanism
that has most likely over- or under- compensated the radical production
processes in order to achieve an acceptable fit of the chamber data.  The
application of such a mechanism in a regulatory model will result in a
systematic bias in the quantitative relationship between precursor
emissions and ozone production.

Scientific versus Regulatory Success

    It is quite apparent that the precision and accuracy requirements which
would meet the scientific community's standards for success may not be
acceptable to the regulatory community.  By this I mean that the regulatory
application may require better precision and accuracy performance from the
mechanism in order for it to be an effective tool for developing
quantitative control strategies.

    For example, the scientific community in reviewing the various sources
of error associated with the chemical mechanism development process
anticipates uncertainties of the order of plus or minus 30% in the
mechanisms predictive capability when compared with chamber data.
Establishing that this is a reasonable error limit when the mechanism is
applied under real atmospheric conditions remains a critical issue which
must be demonstrated if these approaches are to have any scientific
credibility.  But more importantly if the 30% uncertainty is reflected in
the control requirement of precursors (e.g.,. non methane hydrocarbons) to
meet the ozone standard in a given city, the associated cost differentials
can be both economically and socially prohibitive.

    The recognition of this uncertainty and factoring it into the
quantitative procedures required in state implementation plans to
demonstrate a course of action for attaining the national ambient air
quality standard for ozone would seem a logical first step to be taken by
the regulatory community.  The next step should be to establish realistic
precision and accuracy performance standards for mechanisms to be used in
quantitative models for control strategy development.

Recommendations: Research Agenda to Address
"Control" Precision and Accuracy

    Working under the assumption that there is a desire to improve the
precision and accuracy of chemical mechanisms and their ability to provide
quantitative relationships between precursor emissions and ozone production

the following recommendations are presented.   It should be noted that
though these may not be all inclusive,  they are thought to be the most
critical. Finally, the recommendations  represent an integrated approach
which requires each to be carried out in order to successfully achieve the
overall objective.

    1. Mechanism Mapping - develop procedures for and perform flow diagrams
of all mechanisms under research and development or currently used in
applications to identify critical nodes for radical initiation,
propagation, and termination.

    2. Laboratory Studies - perform laboratory studies to reconcile
differences in critical nodes between mechanisms which result from lack of
knowledge of the chemical details of a reaction process (e.g.,. elementary
reaction steps, fragmentation channels  and yields).

    3. Smog Chamber Characterization -  develop guidelines and standards for
smog chamber operations and basic requirements for chamber
characterization.  The chamber performance characteristics must be
documented and submitted to peer review, preferable through publication in
the open literature.

    4. Smog Chamber Intercomparison Studies - develop guidelines and
perform smog chamber intercomparison studies to determine internal bias
among systems, variations in artifact effects and overall reproducibility
of results between systems.

    5. Atmospheric Observations - initiate a program to systematically
build an atmospheric chemical observational data base for the purpose of
diagnostic interpretation and evaluation of mechanisms.

Benson, S.W., Thermochemical Kinetics, John Wiley, New York,

Demerjian, Kenneth L.,  J.A. Kerr, and Jack G. Calvert, "The
Mechanisms of Photochemical Smog Formation", Adv. Environ.  Sci.
Technol., Vol. 4, 1-262, (1974).

Niki, H., E.E. Daby, and B. Weinstock, Adv. Chem. Ser. 113, 16

   Figure 1.   Components of Chemical Mechanism Development
                     Chemical Mechanism
Smog Chamber


                          Paper 4
Development and Testing of Gas-Phase Chemical Mechanisms:
             Present and Future Research Needs
                     Roger Atkinson
           Statewide Air Pollution Research Center
             University of California-Riverside



            Roger Atkinson

            Statewide Air Pollution Research Center
            University of California
            Riverside, California  92521
     In a real sense, atmospheric chemistry can be said to date
from approximately 1969-1970, when the hypothesis that the OH
radical is a key reactive intermediate in atmospheric reactions
was proposed (Heicklen et al., 1969; Stedman et al., 1970).  It is
now recognized that the OH radical plays a major role in
initiating organic consumption and, in combination with H02 and
RO^ radicals, NO to NOo conversion and the formation of ozone.

     Since 1970, chemical mechanism development has proceeded in
parallel with progress in elucidating the kinetics, mechanisms and
products under atmospheric conditions of elementary reactions and
in conducting environmental chamber experiments.  Table 1 lists a
number of chemical mechanism development and review/evaluation
efforts which have led to a steady and continual advance in our
detailed understanding of the chemical reactions of organic
compounds leading to ozone formation.  The dates given in Table 1
are those of publication; in most cases these mechanism
development or review efforts were initiated several years
earlier, as, for example, the Atkinson and Lloyd (1984) review
which in its initial form was completed in 1980 and which was used
as the basis for formulating the ALW mechanism (Atkinson et al.,

     To a large extent only two environmental chamber facilities
have been operational over the past few years, at the Statewide
Air Pollution Research Center at the University of California,
Riverside, and at the University of North Carolina, and only a
small number of groups have been active in chemical mechanism
development.  Despite this and the fact that much of the funding
for these programs came from the U.S. Environmental Protection

Table 1.  Selected Chemical Mechanism Development Programs
Niki, Daby and Weinstock (1972)
Demerjian, Kerr and Calvert (1974)

Hecht, Seinfeld and Dodge (1974)1
Durbin, Hecht and Whitten (1975)(
Falls and Seinfeld (1978)
Hendry et al. (1978)

Carter et al. (1979)

Whitten et al. (1979, 1980a)
Whitten et al. (I980b)
Atkinson et al. (1980)
McRae et al. (1982)
Atkinson, Lloyd and Winges (1982)

Killus and Whitten (1982)
Atkinson and Lloyd (1984)
Leone and Seinfeld (1984)
Leone and Seinfeld (1985)

Lurmann, Lloyd and Atkinson (1986)

Stockwell (1986)

Carter et al. (1986)
Detailed propene mechanism
Review of atmospheric
chemical reactions
Airshed mechanisms
Airshed mechanism
Detailed alkane, alkene and
aromatic mechanisms
Detailed propene, n-butane
Detailed organic mechanisms
"Carbon-bond" mechanism
Detailed toluene mechanism
Airshed mechanism
"Detailed mechanism" for
floating-box applications
Detailed toluene mechanism
Review/evaluation of the
detailed atmospheric chemical
mechanisms of eight
Detailed toluene mechanism
Detailed "master" mechanism
and comparison of then
current mechanisms
Detailed and condensed
mechanisms, for long-range
transport/acid deposition
Airshed mechanism for acid
deposition application
Detailed chemical mechanisms

Agency, in retrospect the development of chemical mechanisms does
not appear to have been a totally coordinated effort.

     The fact that the two major reviews concerning  the  chemical
reactions occurring in the atmosphere have led to continued
chemical mechanism development points to the crucial need  for an
ongoing effort to critically review and evaluate the available
kinetic, mechanism and product data for the elementary inorganic
and organic reactions occurring in the troposphere.  This  aspect
of chemical mechanism development, and of tropospheric chemistry
in general, has received short shrift to date.  The  stratospheric
chemistry community have had the NASA and, until recently, CODATA
evaluation panels on a continuing basis.  CODATA has terminated
its evaluation effort, and this has been picked up by IUPAC with a
slant towards tropospheric chemistry — with organics up to C-j
(propane, propene and their atmospheric degradation  products)
being included.

     A major problem in providing a review and evaluation  program
for organics of the complexity needed for regulatory purposes is
the lack of basic data — I estimate that of the hundreds  of
elementary organic reactions involved in the latest  Carter et al.
(1986) mechanism, actual kinetic or product data exist for no more
than 10-20% of these reactions.  Clearly, speculation and  argument
by analogy plays a major role, and review evaluation efforts
should include "directions" for such data gaps, utilizing
thermocheraical arguments and any other theoretical techniques

     In my opinion, the development of a chemical mechanism for
use in ambient atmospheric calculations must follow the following

     (a)  Assembly of a detailed chemical mechanism, whether it be
of the "carbon bond" approach or the "representative species"
approach, which is totally consistent with our current understand-
ing of the elementary chemical reactions which occur under atmos-
pheric conditions.  This chemical mechanism, or list of reactions,
hence utilizes the best kinetic, mechanistic and product data base
available from experimental laboratory and theoretical studies.

     After assembling a detailed chemical mechanism (or modify-
ing/updating an existing mechanism) which is consistent with the
available kinetic, mechanistic and product data base, this
mechanism must be tested against a relevant data base.  These

data are not those for elementary reactions, but rather they
concern the overall effects of complex reaction sequences.

     These observed effects involve, for example, organic consump-
tion rates, NO to N02 conversion rates, ozone formation and
product formation rates and yields, and may be termed "global"
data.  In theory, these can be obtained either from environmental
chamber irradiations of N0x-organic-air mixtures or from ambient
air measurements.  Neither of these data bases are free from
problems, but at the present time I think it is clear that the
environmental chamber data base involves the least number of
variables to be taken into account.

     To test (but not validate) a chemical mechanism against
environmental chamber data, a number of extra reactions to take
into account chamber effects need to be added to the list of
reactions in the chemical mechanism.  These include:

     •  Wall decay/offgassing of 03, N02, H202, HN03, N205 and
other species.

     •  Heterogeneous formation of MONO from NC^.

     •  Formation of "chamber-dependent" radicals, probably of OH
radicals from the photolysis of heterogeneously formed HONO.

     •  The spectral distribution and light intensity of each
chamber need to be accurately represented.

These "chamber-dependent" effects all introduce uncertainties
which affect mechanism testing.

     Clearly, there is a need for more work to be carried out
concerning these chamber effects, at two levels:

     •  Characterization or parameterization of chamber effects.
For example, the chamber radical source is presently parameterized
as a flux of OH radicals, with the magnitude of this flux being
determined from associated chamber control/characterization

     •  Fundamental understanding of the problem:  for example,
understanding the detailed chemical formation routes which give
rise to the chamber OH radical source.

     It should be recognized that chemical mechanism testing is
actually carried out under limited NO  and hydrocarbon
concentration regimes/ratios, which are not often representative
of ambient atmospheres, as shown schematically in Fig. 1.  If at

          1000 r~

                      10        100        1000

                              HC (ppbC)
Fig. 1.  NO  and hydrocarbon concentration regimes which are
         representative of ambient atmospheres and of environ-
         mental chamber data.
all possible, the chemical mechanism should be tested against
single organic data, for example, against N0x-air irradiations of
HCHO, CHoCHO, other carbonyls, alkanes, alkenes and aromatic
hydrocarbons in a hierarchical manner.  Testing should then
continue with N0x~air irradiations of mixtures.  This stepwise
testing procedure should be carried out since compensating errors
can allow complex mixtures to be fit well, but not simple single

     Finally, we obtain a detailed chemical mechanism which has
been compared/tested against a "global" data base, for example,
environmental chamber data.  For use in an airshed computer model
we must then:

     •  Condense the mechanism to fit within the limits of the
computer resources.
     •  Remove the chamber-effects reactions.

     Condensation of the chemical mechanism should be carried out
by a gradual procedure, testing the condensed versions at each
step of the process against the fully detailed mechanism over a
wide range of organic/NO  concentrations and ratios.  Retesting
against environmental chamber data is not sufficient for any major
condensation of the chemical mechanism.

     Chemical mechanisms must be well documented, both in terms of
the chemistry upon which the detailed mechanism is based as well
as the testing against environmental chamber data and subsequent
condensation.  This includes a complete documentation about how
chamber effects and light intensity and spectral distribution were

     However, it is not totally obvious that the removal of
certain of the chamber-effects is correct.  Thus, the major
chamber effect in terms of the reactivity of environmental chamber
irradiations concerns the "chamber-dependent" radical source.
This probably originates from a photo-enhanced heterogeneous wall
formation of HONO from NC>2»  The rate coefficient for the dark
formation of HONO at 50% relative humidity and 298 K in ~6000
liter volume environmental chambers is surprisingly similar to the
HONO formation rate coefficients derived from ambient atmospheric
measurements of nighttime HONO.  This suggests that there may well
be an atmospheric OH radical (or HONO) source which, for a given
N02 concentration, is similar in magnitude to those needed for
chamber simulations.  If indeed there is an atmospheric
heterogeneous source of OH radicals, this presumably could have
significant implications on control strategy decisions.

     To summarize, gas-phase chemical mechanism development
requires close collaboration and parallel research efforts in

     •  Basic gas-phase kinetic, mechanistic and product studies
carried out under laboratory conditions.

     •  An ongoing data evaluation and review process of the
kinetics, mechanisms and products of elementary reactions of
importance to atmospheric chemistry.

     •  Research studies to elucidate several significant environ-
mental chamber effects.

     •  Environmental chamber studies.

     •  Computer modeling and chemical mechanism development, with
this research area providing the critical input to identify the
basic kinetic, mechanistic and product data and the environmental
chamber/chamber effects data required.

     Only a cohesive approach will provide advances in the
"accuracy" of chemical mechanisms in the next few years.

     At the present time, we are data limited with respect to both
basic kinetic, mechanistic and product data and. high quality
environmental chamber data (including high quality
characterization data).  Indeed, further computer modeling
programs utilizing our present data base are a waste of both time
and effort.

Basic Data

     Vast advances in our understanding of the chemistry occurring
in the atmosphere have been made over the past 15 years.  However,
because of funding constraints the rate of advance has markedly
slowed since the EPA-sponsored workshop "Chemical Kinetic Data
Needs for Modeling the Lower Troposphere" was held in Reston, VA,
in 1978.  Indeed, for example, the limited amount of experimental
work carried out since then concerning the atmospheric chemistry
of aromatic hydrocarbons has led to an advance in our actual
knowledge of their chemistry, but has shown us how little we still
know about the atmospheric chemistry of this class of compounds.
The same is true for the long-chain alkanes.

     We need (a) basic mechanistic and product data (under atmos-
pheric conditions) for most classes of organics emitted from
anthropogenic and biogenic sources.  This includes, but is not
limited to, aromatic compounds and the larger alkanes, and (b) an
integral and ongoing review and evaluation effort, either as a
stand-alone effort for chemical mechanism development or as a
"piggy-back" onto other presently ongoing evaluation efforts.

Environmental Chamber Data Needs
     The environmental chamber data necessary for future chemical
mechanism development include the following:

     •  A review/evaluation of chamber-effects is needed, to
determine the nature of chamber effects and how they should be
taken into account.

     •  Further basic research is needed to elucidate the
chemistry/physics of certain of these chamber effects.

     •  High quality chamber data, combined with high quality
characterization/chamber effects data, are needed.  For the

upcoming generation of chemical mechanisms, such species as HONO,
HCHO, N02 (not as NOX~NO), HNO-j and H202 must be measured in real-
time.  Light intensity and spectral distributions must also be
measured accurately.

     •  Chamber data under low N0x~organic concentration
conditions, and in the absence of NO, are necessary for mechanism
development for long-range transport and multi-day applications.

     In addition, ambient air data are needed to aid in chemical
mechanism development.  A prime example concerns whether or not OH
is formed in the atmosphere from N02» as it is in environmental

     At the present time, we are at a critical juncture.  Thus,
the present generation of chemical mechanisms are close to being
as up-to-date with respect to the available data base as is
possible, and any further advances in the predictive accuracy of
urban and regional airshed computer models will require a renewed,
closely coordinated, effort by research groups with diverse
interests as well as by funding and regulatory agencies.

Atkinson, R., Carter, W. P. L., Darnall, K. R., Winer, A. M., and
     Pitts, J. N. Jr., 1980, Int. J. Chem. Kinet., 12:779.
Atkinson, R., Lloyd, A. C., and Winges, L., 1982, Atmos. Environ.,
Atkinson, R. and Lloyd, A. C., 1984, J. Phys. Chem. Ref. Data,
Carter, W. P. L., Lloyd, A. C., Sprung, J. L., and Pitts, J. N.,
     Jr., 1979, Int. J. Chem. Kinet., 11:45.
Carter, W. P. L., Lurmann, F. W., Atkinson, R., and Lloyd, A. C.,
     1986, "Development and testing of a surrogate species
     chemical reaction mechanism," Final Report to EPA Contract
     No. 68-02-4104, May.
Demerjian, K. L., Kerr, J. A., and Calvert, J. G., 1974, Adv.
     Environ. Sci. Technol., 4:1.
Durbin, R. A., Hecht, T. A., and Whitten, G. Z.,  1975, "Mathe-
     matical modeling of simulated photochemical  smog" EPA-650/4-
     75-026, June.
Falls, A. H. and Seinfeld, J. H., 1978, Environ.  Sci. Technol.,
Hecht, T. A., Seinfeld, J. H., and Dodge, M. C.,  1974, Environ.
     Sci Technol., 8:327.
Heicklen, J., Westberg, K., and Cohen, N., 1969,  Report No.  115-
     69, Center for Air Environment Studies, Pennsylvania State
     University, University Park, PA.

Hendry, D. G., Baldwin, A. C., Barker, J. R., and Golden D. M.,
     1978, "Computer modeling of simulated photochemical smog."
     EPA-600/3-78-059, June.
Killus, J. P. and Whitten, G. Z., 1982, Atmos. Environ., 16:1973.
Leone, J. A. and Seinfeld, J. H., 1984, Int. J. Chem. Kinet.,
Leone, J. A. and Seinfeld, J. H., 1985, Atmos. Environ., 19:437.
Lurmann, F. W., Lloyd, A. C., and Atkinson, R., 1986, J. Geophys.
     Res., 91:10905.
McRae, G. J., Goodin, W. R., and Seinfeld, J. H., 1982, "Mathe-
     matical modeling of photochemical air pollution."  Final
     Report to California Air Resources Board Contract Nos. A5-
     046-87 and A7-187-30, April.
Niki, H., Daby, E. E., and Weinstock, B., 1972, Adv. Chem. Series,
Stedman, D. H., Morris, E. R., Jr., Daby, E. E., Niki, H., and
     Weinstock, B., 1970, 160th National Meeting of the Am. Chem.
     Soc., Chicago, IL, Sept.
Stockwell, W. R., 1986, Atmos. Environ., 20:1615.
Whitten, G. Z., Hogo, H., Meldgin, M. J., Killus, J. P., and
     Bekowies, P. J., 1979, "Modeling of simulated photochemical
     smog with kinetic mechanisms."  Vol. I, Interim Report EPA-
     600/3-79-OOla, January.
Whitten, G. Z., Killus, J. P., and Hogo, H., 1980a, "Modeling of
     simulated photochemical smog with kinetic mechanism."  Vol.
     I, Final Report EPA-600/3-80-028a, February.
Whitten, G. Z., Hogo, H., and Killus, J. P., 1980b, Environ. Sci.
     Technpl., 14:690.
































                 NOZ  (not  NOX-NO)










                         Paper 5
Review of the "Strawman" Document for the EPA Workshop on
     Evaluation/Documentation of Chemical Mechanisms
                    Michael W. Gery
                Systems Applications, Inc.

         Review of the  "Strawman" Document for the  EPA Workshop on
              Evaluation/Documentation of Chemical  Mechanisms
                             Michael W. Gery

                        Systems Applications, Inc.
                           101 Lucas Valley Rd.
                           San  Rafael,  CA   94903

     In the following review of the Jeffries and Arnold document, I shall
take a somewhat different approach than that taken by the authors in
describing what I feel to be the essential workshop goals concerning areas
of uncertainty and the need for documentation in the model development,
evaluation and application practices.  Initially, I shall present a review
of the important points in the strawman document with respect to
historical development of our science.  Following this I shall describe
the current status of what I perceive to be our present modeling
methodology, developing out of this a discussion pertaining to
uncertainties in approach and data, future needs and options, and possible
courses of action.

                   The Current Tension and the Approach

     There appear to be at least three distinct views  concerning the
degrees of validity in photochemical  kinetics modeling, each dependent on
the needs of the respective groups who utilize these models in different
ways.  At the policy level, as exemplified in the authors'  quotes from the
GAO, legislators want to know how reliable different modeling tools are in
an effort to justify the complex decisions regarding issues of the Clean
Air Act.  The authors point out that since the quantification of model
validity based on certainty in well-grounded principles is  difficult,
vindication maybe a better description of the of the  actual requirement
for model performance at this level.  Within the Agency, however, costly
control strategy decisions must be made based on predictive estimates.
The need for determining the relative validity of different chemical
mechanisms based on scientific evaluation is, therefore, a  necessity.
Unfortunately, because the scientists who develop these models cannot
describe all real situations with absolute precision,  inherent uncertainty
often masks the differences between models.  This conflict, between
scientists with goals of absolute validity and regulators who need to use
the "best" tool currently available, creates a tension resulting from the
what the authors describe as a "forced realization of  the paradigm."  That
is, the photochemical kinetics models are needed to make predictions
beyond the bounds of certainty within which the scientists  presently feel
secure.  Hopefully, however, because this is a continually improving
process, it may be possible to describe the types and  measures of
uncertainties which exits at present, in an effort to both minimize these
error bands and prescribe tests which could detect mechanisms which do lie
beyond the realm of what is presently acceptable.  I feel that the best
way to do this is to examine each process in our current modeling
methodology, defining the structure and associated uncertainties in
detail.  This paper is based on such an approach.

     As noted above, Jeffries' and Arnold's description of the historical
and technical progression  in photochemical kinetics mechanism development
has provided us with a review of  those works that set the scientific

                                   - 2 -

foundation for the current practice of our science.   This description of
the development and status of our normal science allows us to look at the
structure and rules for various distinct processes which make up the
paradigm-as-set-of-shared-values.  That is, our methods, standards and
ways of solving problems.  This is critical to our present needs for two
reasons; first, because it is necessary to understand.the chemical model
development and evaluation processes that we currently employ to modify or
enhance them, and secondly, because a review of past work provides us with
descriptions of practical aspects of methodologies in our field which are
often overlooked or taken for granted because the purely analytical
results from the earlier works have been superseded.  Therefore, I will
briefly review the historical development of the basic paradigm-as-set-of-
shared-values within the philosophical descriptors set up by Jeffries and
Arnold.  I will later describe individual processes or methodologies to
Isolate the current status in the form of basic rules and current
uncertainties of each.  It is these processes that Involve the subjective
decisions by the model developer with respect to process and data
uncertainty and are therefore encumbered with uncertainty requiring
validation.  The approach I take 1s aided by a diagram (Figure 1) which
Illustrates the historical development of these processes and can later be
used as a basis for discussion.  Obviously, such a diagram represents my
idealized conception of the current model development methodology.
            The Historical  Development of Mechanism Development
           Methodologies Through the Eyes of Jeffries and Arnold

     The preparadigm period is described by Jeffries and Arnold as that
time when the need to describe the atmospheric chemistry of ozone
formulation was apparent but the science of air pollution chemistry was
lacking in its ability to explain the reasons for the chemical
transformations resulting in the formation of secondary pollutants.
Collections of chemical reactions along with some kinetic and mechanistic
information existed, but the relative significance of individual reactions
or cycles was not yet elucidated.  This pool of information, which has

                                   - 3 -

grown significantly since the preparadigm period,  is represented by the
top box of Figure 1.  This data is drawn upon in the initial  compilation
(and later revaluations) of a photochemical  kinetics mechanism.

     The first mechanistic descriptions were  developed prior  to 1970, but
many of the steps were empirical, lacking any real  physical or chemical
meanings.  About 15 years ago, an essential  "revolution" occurred with the
initial mechanistic descriptions of hydroxyl  radical chemistry and its
role in organic oxidation through the OH-H02  cycle.  Jeffries and Arnold
coin this the beginning of the "OH-paradigm," and suggest that because
this description and its attendant auxiliary  hypotheses (coupled with the
possibilities of computer simulation) were more successful at solving the
problems then recognized, rapid Gestalt switches made quickly by many
scientists practicing in different fields.  Because of these  developments,
the first successful mechanisms that were based on "well-founded
scientific principles" were put forth.  The paradigm had taken a big step
in that a paradigm-as-set-of-shared-values was clearly developing.  In
fact, sets of intuitive rules, many of which  still  govern aspects of
mechanism development today, were postulated  at that time.  These rules
were often put forth as a method of limiting  uncertainty through the
normalization of scientific process.

     This process of compiling a chemical kinetics mechanism through the
evaluation of the existing pool of data, combined with needed assumptions
is depicted in Figure 1 as the data evaluation process.   I will discuss
this process more below, however, it is important to note that the
formulation of a mechanism from the pool of data is a "mapping process."
Hence, some measure of ambiguity is inherited from the generalization,
deletion or distortion of reality which accompanies the creation of a
map.  Of these, deletion, especially deletions which occur because missing
reactions were simply unanticipated and therefore unstudied, is the most
difficult to  identify and deal with.

     The authors note that one of the early applications  of a mechanism
developed through this process was the  Niki, Dalby and Weinstock  (NOW)

                                   - 4 -

mechanism.  In their work, NOW simulated older data from the irradiation
of mixtures of propylene and NOX.  From these simulations, which utilized
the newly developing tool of computer modeling, it became apparent that
additional tools and data were needed to further the science.  This was,
however, an example of an initial application of such a mechanism.  It
supported the OH-paradigm as an explanation of the chemical dynamics in
the smog chamber experiments.  Jeffries and Arnold refer to this
occurrence as the first class of Kuhn's "normal science, where the model
based on the OH-paradigm is used to reveal the nature of things as
described by a paradigm that is becoming more worthy of testing and

     The first well known example of normal science following the adoption
of the OH-paradigm was the Demerjian, Kerr and Calvert (DKC) mechanism
development and application.  DKC were careful to document their method of
mechanism development, and they presented many of the formulation and
testing rules that are the basis of our current paradigm-as-set-of-shared-
values.  They noted the need for a good pool of kinetic and mechanistic
information, their evaluation of the existing data was a solid foundation
for the mechanistic development effort.  One particular extension of the
normal science of mechanistic development was the clear utilization of
smog chamber data to refine assumptions inherent in the mechanism.  This
process is shown in Figure 1 as a feedback loop of evaluation and
verification of the current mechanism using "real" data; in this case,
smog chamber data.  It is, in effect, the altering of the mechanistic map
through comparisons with a new set of "truth"  (the smog chamber data) to
minimize distortions and identify unintended deletions in the original
map.  Thus, while most earlier studies that attempted to simulate chamber
data were mainly concerned with verifying the paradigm, DKC also used
simulations to articulate and strengthen the paradigm by resolving
ambiguities.  DKC saw the need to build on the foundation of basic data
with new  information derived from smog chamber simulations.  The authors
say that DKC implied the need for smog chamber data to evaluate
"alternative mechanisms and reaction rate constants ... even though they
[DKCj did not explain how this occurred."  In their discussion of the DKC
                                   - 5 -

"asymmetry for the treatment of experimental  fact,"   Jeffries and Arnold
state, however, that while the chamber data was used to refine the
mechanistic map, it was not considered a requirement that the mechanism
"fit" all data.  It should be noted that the  amount  and quality of smog
chamber data was limited at this time.  While DKC could not yet see the
utility of smog chamber validation with the data they.had, they did use
their model in a way that manifested the need for the better data that now
exists, and the validity tests that can now be employed.

     The 1974 work of Hecht, Seinfeld and Dodge (HSD) 1s discussed next.
HSD reaffirmed and added to the description of the model development and
verification processes described above, and also clearly identified the
need for both fundamental kinetic data and smog chamber data.  That is,
evaluation of basic kinetic data and simulations versus smog chamber data
are two very compatible processes which, when treated equally, minimize a
great deal of uncertainty associated with the mapping process of mechanism
development.  The basic kinetic and mechanistic data contains the
descriptions of conditional chemical variability upon which atmospheric
simulations are based and legitimized.  The smog chamber data provides a
real, controlled, although somewhat limited,  test bed for detecting
deletions or distortions 1n the basic data.  Hence,  the iterative
adjustment of mechanistic assumptions through smog chamber simulation is a
way to link the strengths of both types of data (fundamental kinetic and
smog chamber) so that uncertainties in each may become obvious through
comparison with the other.

     A second key feature of the HSD work developed out of their different
objective.  The authors state that, "While a prime objective for DKC was
to have  'a complete reaction scheme1 and thus to explain, the prime
objective for HSD was "that the mechanism predict the chemical behavior of
a complex mixture of many hydrocarbons, yet that it include only a limited
degree of detail1 which would allow the mechanism to be applied to the
'prediction of both smog chamber reaction phenomena and atmospheric
reaction phenomena....This work, therefore, in  its conception and
definition, set out toward a different end than the DKC work:  DKC sought

                                   - 6 -

scientific explanation; HSD wanted prediction."  The need for prediction,
specifically the eventual need for prediction of atmospheric chemistry,
resulted in the formulation of two additional processes:  (1) condensation
of the mechanism (computers were and still are limited by the large number
of calculations necessary to run complete chemical kinetics simulation
code), and (2) verification that the predictions of the mechanism are
valid 1n the physical and chemical ranges of the predictive application.

     These processes are graphically presented in Figure 1.  Note that
there are two apparently parallel pathways leading to a verified,
condensed mechanism.  One describes the condensation of an explicit
mechanism that has first been tested versus smog chamber data, while the
other implies the production of an explicit mechanism from "first
principles," followed by evaluation versus experimental data.  In both
cases, the evaluation against smog chamber data 1s still required as an
essential test against real data to establish the completeness of the
explicit compilation.  It is my personal preference to proceed along the
path which provides a validated explicit mechanism prior to condensation,
since:  (1) compensating errors in the mechanism cannot be masked by
condensation prior to testing versus real data, and (2) a complete
explicit mechanism that has been verified against all available data can
be used to produce a number of condensed mechanisms at different levels of
condensation and for different applications.  The alternate pathway
requires .a new condensation for every iterative step in the smog chamber
evaluation loop, resulting in a more complicated task to produce the same

     As shown in Figure 1, the need for a simplified mechanism to predict
data collected at contaminant monitoring stations in an urban area is a
drastic variation on a complete reaction scheme used only for explanatory
purposes.  Recognizing this, HSD said, "a kinetic mechanism, once
developed, must be validated."  Hence, the nature of this workshop.
However, it is important to note that the validation requirement
originally derived from the HSD need for atmospheric simulation and
prediction, and this should be the focus of the workshop as well as the
initial makeup of and verification of the initial explicit mechanisms.
                                   - 7 -

                  The Current State of  "Normal Science"

     In the above section I  have developed an idealized structure of the
paradigm-as-set-of-shared-values that  I  believe we tend to follow in our
current version of normal science.   I  would now like to discuss the
present status of these  processes by isolating each and considering the
current "rules" and the  remaining sources of uncertainty.   If we use
Figure 1 as an idealized representation  of the mechanism development
process, we see that the boxes in the  schematic represent real (although
sometimes temporary) data, information,  compilations, mechanistic
versions; in short, they are starting, nodal and stopping points.
Listings of data, results and mechanisms can be produced at any time to
represent the current status at a point.  The ovals are processes, often
requiring sufficient ingenuity,  test data and validation to proceed
through.  These processes are the sources of uncertainty which must be
clearly described in the formulation of a predictive chemical kinetics
mechanism from basic data.  Because of this they are also the areas which
must be clearly documented in each application.  The amount of uncertainty
attained at each process, the past methods derived to limit uncertainty,
and the future prospects for better data and clear tests are the topics
for discussion at this workshop.

     The evaluation of existing chemical kinetics and mechanistic data is
clearly a necessity in the initial development and periodic updating of
chemical kinetics mechanisms.  As noted above, this process was
established in the comprehensive reviews in the NOW and DKC papers.  The
authors observed that one of the primary goals of DKC was, "to evaluate
the various alternative mechanisms and reaction rate constants proposed in
view of the best kinetic data in hand today."..Two of the rules DKC used
to normalize this process and minimize uncertainty are reported by the
authors as:  (1) mechanism reactions must be elementary reactions, and (2)
rate constants must be based on experimental determination of rate
constant or must be estimated using thermodynamic techniques.  In other
words, given well founded data and reasonable estimation techniques, use
them; rate constants cannot be considered adjustable parameters.

                                   - 8 -

     The compilation of existing data is far more objective that another
of OKC's goals, "to identify some of those potentially important reactions
in the mechanism of photochemical smog formation for which there was
insufficient basic kinetic data to allow for a realistic judgment of their
importance."  This enters the more subjective realm of mechanistic
mapmaking  as described above).  Because the methods of making data
evaluations and assumptions are based on Individual understanding and
intuition, different mechanism developers will make different choices
based on similar information or lack of information.  Thus, the errors of
the mapping process, distortion and overgeneralization of basic
information, combined with unanticipated deletion, will occur.  As noted
earlier, however, the second process (mechanism evaluation versus real
data)  allows some of these errors to be detected.

     Although an excellent compilation of the current pool of information
is almost always produced or referred to by mechanistic developers (for
instance, the compilation by DKC, the reviews by Atkinson and coworkers,
Kerr and Calvert, and the NASA and CODATA evaluations), the information of
how or why certain decisions were made in the development of a mechanism
from these data 1s often poorly conveyed.  Instead, an explicit
mechanistic description 1s simply asserted as a listing which reflects the
results, but not the reasons, for choices made by the authors.  I hasten
to add that this is not the case in all work, but only a trend apparent
from this paper.  This occurrence necessitates a third rule; the mapping
decisions, as well as the sources of data, must be well documented to
provide a more viable base of formulation.  Such a description will aid in
the future location of anomalies and help to identify the range of
applications for which the mechanism was originally based.

     If there is such a measure as goodness of fit for this process it is
through the peer review process.  I see no problem with this method,
certainly the reviews published to date have followed a trend of rigorous
evaluation, and more importantly, clear delineation between actual
measurements and intuitive estimates.  What does contribute to the overall
uncertainty in the model development is the relative errors associated

                                   - 9 -

with the kinetic and mechanistic data available and the unavailability of
certain data.  The solution to this problem is primarily related to the
amount and direction of research.  Periodic reevaluation and compilation
of extensive reviews directed at the chemistry of air pollution must be
enacted to make the wide range of data generally available and
deciphered.  These reviews require funding and, to some degree, guidance
toward desired goals.  They are not necessary, however, if no new data is
available to review.  Hence, a major method of reducing overall
uncertainty in chemical mechanisms, especially in areas of new interest
and little information, is to support new research into elementary
chemistry in those areas.  As more data comes in and the pool if
information is enhanced, many of the individual choices which contribute
to uncertainty in mechanism development are unnecessary and the map is a
more formidable description of reality.

     The evaluation/adjustment of the current explicit mechanism using
smog chamber data is the second process described in Figure 1.  The need
for such an approach has already been stated.  Basically, smog chamber
experiments provide realistic data, which Inherently include tests to
identify unintentionally deleted chemistry or overgeneralized or distorted
mechanisms.  As with most evaluation processes, this is iterative in form,
resulting at times in what the authors describe as "knob twittering."
Because of the large number of variables, data and subjective decisions,
this process probably provides the greatest accumulation of uncertainty  in
the entire mechanistic development and application practice.  On the other
hand, it must be noted that elimination of such a verification against
real data would most certainly yield far greater uncertainty in any
resulting mechanism.

     Although the process is shown as an iterative loop (Figure 1) with
one decision block, the process  is really a multi-step operation  (shown  in
Figure 2)  involving:   (1) the mechanical testing of model predictions
using the current mechanism versus data, (2) the more subjective
evaluation of comparison results, and  (3), the intuitive process of
adjustment of the mechanism to provide "better" predictions.  This process

                                   -  10 -

has evolved over the last two decades as more and higher quality data has
become available.  However, we are still confounded by the basic dilemma
of when to exit the loop.  This dilemma results from the uncertainty in
two areas:  (1) how good is the data used in the evaluation, and (2), how
does one measure "goodness of fit" between mechanism prediction and data
(and thus, decide he has reached "reasonable agreement" and terminate the
adjustment process)?  Clarification of this dilemma has stimulated the
formulation of many of the rules and methods that I will discuss next;
however, though these methods have improved our ability to measure
reasonable agreement (or in some cases, locate unreasonable agreement),
the decision as to when the fit is good enough is still ill-defined.
Because of this, it is not unlikely that the actual number of iterations
may sometimes be based on the level of remaining funding rather than any
reasonable method of testing.  This is unfortunate and must be a primary
focus of the workshop.  It is probably not possible to define the
necessary requirements for a mechanism to be considered "valid", but it is
possible to establish some minimum requirements needed for the claim of
reasonable agreement.

     It is Important to minimize the error associated with data
variability in the test data and to clearly define an evaluation program
that will enhance data evaluation as well as prove the strength of a
mechanism to the scientific community.  One way this is done is through
the formulation of basic rules of evaluation and adjustment.  One such
rule, noted early on by DKC and HSD, is that it is essential to establish
a strong data set prior to evaluation of and adjustments to the
mechanism.  This should include:  (1) data from different sources, (2) a
variety of hydrocarbons studied, (3) single hydrocarbon and mixture
experiments, and (4), different initial conditions (including a broad
HC/NOX ratio).  When adjustments are made to the mechanism or when test
data are missing, these values must be supplemented in a rational way.   It
is also critical that the conditions and characteristics of each smog
chamber be defined or estimated, especially chamber artifacts that differ
from ideal.  Although I will not discuss specific chamber artifact
representations, it should be noted that the uncertainty developed from

                                   -  11  -

different hypotheses concerning these effects must be overcome prior to
any meaningful comparison of mechanisms.   This should be a primary concern
of the workshop.  The reactions that are  chamber dependent should be
consistently applied within their definitions for all simulations
attempted in each study, and hopefully, with the future agreement of the
modeling community, this consistency will result from.a clearer
understanding of these processes.  Of course, these assumptions, plus all
other definitions and estimates, must be  documented.

     Since I was asked to specifically comment on the hierarchy of
chemical species concept developed by Whitten, I will extend this section
concerning design of the evaluation process.  It is logically intuitive
and operationally clear that the process  of mechanism evaluation against
smog chamber data be based on a stepwise  hierarchy of simulations.  Such
an approach was first suggested by DKC in their study methodology when
they noted that the understanding of smog chamber data was best revealed
by starting with simple systems and increasing the complexity of the
hydrocarbon components stepwise.  Whitten later put forth the principles
of the hierarchical approach, a method of model development and testing
intended to clarify (or identify) the sources of uncertainty in
simulations.  Figure 3 shows a hierarchy of species for the Carbon Bond X
mechanism.  The concept is that one should validate mechanisms in a
stepwise order starting from the lowest level first.  Once acceptable
agreement between simulation results and measurements is obtained, changes
in rate constants and reaction stoichiometry for the already tested part
of the mechanism is prohibited and the simulations increase to higher
hierarchical  levels.  If disagreement between simulation results and
measurements occurs at a higher hierarchical level and all lower species
data have already been simulated successfully, it is most likely that the
new chemistry for the higher level species is in.error and not the whole
mechanism.  Such a methodology minimizes the possibility of fortuitous
agreement between simulations and measurements from the effect of
compensating  errors which can occur in a mechanism created directly from
"first principles."  Jeffries and Arnold point out that the hierarchical
approach narrows the macroscopic facts of the smog chamber so that they

                                   -  12 -

more clearly confront the theory.  The opposite method, direct creation of
a mechanism to simulate a mixture of hydrocarbons and NOX in the
atmosphere, without the benefit of simulation of the products of the
hydrocarbon mixture, is a less certain process because of the possibility
of mechanistic compensating errors.

     Because this process appears to possess such a large degree of
uncertainty, it should be a direct concern of the workshop.  Hopefully, we
can take initial steps to improve both the standard test data and the
measure of "reasonable agreement".  We can reduce the uncertainty in the
standards of mechanistic comparison (smog chamber measurements) by
agreement on a good set of basic test data, consisting of many smog
chamber experiments from different facilities for at least some of the
lower hierarchical level species.  This will provide a minimum requirement
for photochemical kinetics models to fulfill.  That is, the initial goals
of a mechanism should be to handle time varying photolysis rates and the
chemistry of basic inorganic and simple organic species.  Because of its
uniform nature, such a dataset may also allow initial attempts at the
clarification of a measure of goodness of fit.  More complex test sets
could be agreed upon.  However, these sets.must be carefully conceived to
account for the mechanistic tests that the original experimental design
intended to develop.  Finally, we must also recognize that we are
specifically discussing ozone and oxidant chemistry here.  As new areas
entered, there is usually little or no experimental data to develop or
evaluate a new mechanism with.  We must consider new areas and decide
whether there are substitutes to smog chamber data which could be utilized
in the reality evaluation loop.

     The need for prediction applications of photochemical kinetics
mechanisms coincided with the requirement that a condensed chemical
mechanism be used in place of an explicit version because of the high
computational costs.  The process by which a condensation technique is
verified and a condensed mechanism is formulated is again an iterative
process similar to the smog chamber evaluation of explicit mechanisms.   It
is shown schematically in Figure 4.  There are different conditions in

                                   -  13 -

this process, however, which have led to more clearly defined
methodologies than in the smog chamber evaluation loop.   This is because
the object of this process is not to simulate smog chamber experiments
correctly, nor, in fact to improve the theory at all, but to duplicate the
ability of the explicit mechanism with a more applicable configuration.
In fact, the condensed mechanism cannot simulate the chamber results more
accurately if all It is based on is simplification of the explicit
mechanism since, in terms of the mapping process, it is  a more generalized
or distorted map with additional deletions.

     I will not'discuss specific condensation methodologies, however, some
basic rules have, again, been laid down to help unify or normalize this
process in HSD and later works.  These are mainly related to the
condensation of oxidant mechanisms.  Inorganic reactions are usually
treated explicitly, although some deletions occur.  Hydrocarbon
representation must occur through a logical methodology, cannot be overly
simple, and must include a specified method for representing complex
atmospheric mixtures in the mechanism.  The condensation/evaluation
process should be performed in a stepwise or incremental fashion, with
stoichiometry and mechanism structure derivable from underlying chemistry,
and rate constants related to elementary reactions or within realistic
range of measured values.  In such a process, which differs with every
researcher and application, it is clearly beneficial, both to the
developer and the user, to document nearly every step.

     In terms of the actual conditions used in the comparison of explicit
and condensed mechanism simulations results for the purpose of verifying
accurate condensation, it is not always necessary to use real measurements
(since there is no attempt in this process to verify that the mechanisms
simulate real measurements) although such data sets often provide a good
foundation.  It is a better idea, rather, to not only verify that the
condensed mechanism is an accurate map of the explicit mechanism within
the range of physical and chemical conditions available in smog chamber
data, but also  to test the condensation results  in more extreme conditions
that will be applied  in the intended application.  Hence, it may be

                                   -  14  -

necessary to fabricate test conditions to perform such a comparison.
Recall that the basic tenet of this process is that "after validation" of
explicit chemistry, the results of any simulation using the explicit
chemistry can serve as data to be used for the derivation and evaluation
of a condensed mechanism.  Therefore, the explicit chemistry, although
encumbered with uncertainty from the previous development, is viewed  here
as "truth."  Because there is no experimental variability in the test data
(the test data are the results of the same explicit mechanism every time),
a much tighter measure of goodness of fit must apply in the verification
decision of Figure 4.

     As noted above, the condensation process inherently requires
generalization, distortion and deletion choices.  For this reason, the
decisions as to the specific condensation approach must be oriented
towards the application.  Of course, these choices have most often been
based on the simulation of urban ozone formation.  However, other
applications may have different chemical requirements or involve new data
sets.  In addition, new validation in the previous process loop
(reverification of the explicit mechanism), may occur.  All of these
occurrences necessitate a new condensation, and therefore, new

     The production of a photochemical kinetics mechanism that has been
developed and tested with smog chamber data and has been condensed to the
size requirements for a specific model, does not necessarily make the
mechanism valid for the intended applications.  Often it takes a
significant leap of faith to accept the predictions of such a model.   For
instance, one might say that even given ideal fit to all smog chamber
tests and a very good condensation technique, the chemical mechanism has
only been validated for the chemical and physical conditions exhibited in
those data.  In addition, any attempt at validation against atmospheric
data carries the inconsistencies of the entire model into the simulation
results.  To take this a step further, while smog chamber validation
simulations were probably performed with the Gear algorithm, many large
air quality simulation models utilize lower order solution methods that
                                  -  15 -

could Induce error unassociated  with the chemical  kinetics  mechanism.
This is a formidable set of "ifs",  but careful  mechanism development
combined with an understanding of the sources of data involved in that
process can minimize some error  and aid in the  quantification of the

     The extended atmospheric range of physical and concentration
differences must be approached from two directions.  First, to the best of
our ability, we must ensure that any condensation-type processes are valid
over the entire range in the intended application.  Second, we must
recognize that the description of conditional variation in the chemical
mechanism is not based solely on smog chamber data.  Rather, while the
smog chamber simulations are used to test the completeness of the
mechanistic representation, the  inherent conditional variations are
usually based on kinetic studies with much wider physical ranges than the
smog chamber experiences.  Therefore, although  the explicit mechanism is
only verified against real measurements within the range of conditions
available in a smog chamber, there is no reason to expect rapid
deterioration of chemical description in conditions somewhat beyond those
of a smog chamber.

     The differences between smog chamber and ambient air concentrations
are more dramatic  Reactions or products which were felt to be
unimportant, or went unnoticed in higher concentrations smog chamber
conditions, may become significant in atmospheric applications.  Whereas
it is trivial to remove chamber artifact reactions from a chemical
mechanism to simulate the atmosphere, it is far more difficult to include
reactions which go unnoticed in the chamber but are significant in the
atmosphere.  For example, because of the orders of magnitude in
differences between urban reactive hydrocarbon concentrations and smog
chamber initial conditions, background species such as urban C0£ and
methane are of little importance in clean chamber simulations but must be
included  in atmospheric models.   Of course, documentation of the changes
needed in a mechanism for proper application to a given situation is
critical.  However, if we are to clearly understand and separate the

                                   -  16  -

chamber artifacts from the mechanism, better smog chamber experiments
which provide test data at concentrations nearer to ambient values must be
obtained.  This will require new or different facilities that limit
artificial processes, coupled with better descriptions of incident light
and background chemistry.

     The question of the error induced by the solution algorithm used to
calculate the changes in concentrations with time for a given modeling
application is a peripheral issue, but should be addressed briefly.  As
noted above, the chemical kinetics mechanisms are often developed with a
higher order solution technique which can be utilized because of the
physical simplicity of the system.  Atmospheric simulations requiring
simultaneous, multi-cell solutions of chemistry and meteorology usually
use a lower order technique which can develop mathematical errors that
translate into Inaccurate predictions.  It should be the responsibility of
the model developer to provide and document comparisons between whatever
solution techniques were used to develop the mechanism and represent the
chemistry in the application model.  The performance of the same
photochemical kinetics model over a simulated application range should be

     Finally, as noted earlier, not all atmospheric applications have smog
chamber datasets which can be used to verify the mechanism against real
data (locate deletions or distortions in the mechanistic map).  This is
becoming clear with the new generation of acidic precipitation gas-phase
models.  At present, comparison with the sparse set of atmospheric
measurements is the only evaluation possible.  These models will benefit,
to a degree, from advances made for all atmospheric models in the
meteorological and solar radiation algorithms, but we may have to face the
fact that unless some controlled tests similar to smog chamber experiments
are forthcoming, the predictions of these models will be viewed with far
more skepticism than the urban ozone models.
                                   -  17  -


     When I first considered the  present state of the mechanism
development and evaluation practice,  it appeared to be encumbered by
uncertainties to the point where  we could not distinguish one mechanism
from another.  That, of course,  is a primary goal of the workshop.  After
reviewing the present state of our science,  however, I believe we must
consider this apparent shortcoming in conjunction with the progress we
have made in this area.  Most of  the facts and processes reported in this
paper were developed at or after  the time when the Dodge mechanism was
first implemented in the EKMA.  Since then,  both the amount of information
available to model developers and the methodology of mechanistic
development have grown significantly.  This  new knowledge, the gathering
of which was largely supported by EPA, has significantly enhanced the
science of photochemical kinetics mechanism development.  Whereas it was
not yet possible to develop a comprehensive mechanism for the description
of urban ozone formation from reactive hydrocarbon mixtures at the time of
the original EKMA, we now agree on a good deal of this chemistry and can
clearly identify specific area of uncertainty that must be addressed
experimentally.  Thus, the objective of this workshop, the identification
and minimization of uncertainties in the photochemical kinetics mechanism
development process, must be viewed with the significant advancement of
our science in mind.  The chemical representations in most of the
mechanisms being compared today have converged to the point where, within
the bounds of uncertainty, it may be difficult to distinguish the benefits
of one over another.  Thus, there seem to be two goals for this
workshop:  (1) to define testing procedures which allow one to verify that
a mechanism is within those bounds of uncertainty which with we presently
feel can be attained (that it is the state-of-the-science), and  (2), to
find new ways to limit the uncertainty in tests which we use to
distinguish between mechanistic representations.

     The uncertainty resides  in the test data  (mainly smog chamber data)
and the measurement technique (our estimate of reasonable agreement
between simulation results and data).  The foundation of a photochemical

                                   -  18  -

kinetics mechanism is the elementary kinetic and mechanistic data upon
which it is based.  Support of new data gathering and periodic evaluation
of the data base will diminish uncertainty in the basic chemistry.  The
Input of mechanism developers cannot be overlooked in this process since
the areas of largest uncertainty sometimes only become apparent after
application of a model.  We must also strengthen our existing smog chamber
data sets.  Besides the continuing need for new experiments (possibly from
new or different facilities), we must study the artificial processes in
existing chambers so that uncertainty in existing data sets can be
minimized.  Study of the wall radical issue is of immediate importance.
In addition, the careful characterization of light sources in every
chamber should be carried out and made available with the data.  Chamber
intercomparisons, intercomparisons of instruments and calibration schemes,
and clear descriptions of uncertainty bands in measured concentrations
must be made available.

     With improvements in experimental data sets such as those noted, it
should be possible to define some basic tests with which the validity of a
chemical mechanism can be compared to another.  As stated earlier,
however, such test may not be powerful enough to point out a truly
excellent mechanism compared to a good mechanism within the current bounds
of our uncertainties.  On the other hand, the continuing limitation of
uncertainty in the test data will allow such tests to demonstrate some
basic weaknesses in poorer mechanistic descriptions.  At a minimum, the
participants of the workshop should try to agree on a method for producing
a set of good smog chamber data regardless of the intended use.  Such a
set would be a great benefit to model developers.  For some types of
mechanisms, however, we must recognize that we are in the early phase of
normal science.  Finally, mechanistic representations for hydrogen and
higher molecular weight peroxides, formic and higher molecular weight
organic acids, aqueous chemistry, and secondary aerosol formation are
under development, but lack the test data base available for urban oxidant
mechanisms.  This will eventually impact on the credibility of these
mechanisms and steps to develop this data must now be considered.
                                  -  19 -

Figure 1.  Schematic Diagram of the Process of Mechanism Devlopment and

                                     Current Explicit
                                     vs. Data
                                     Evaluate Fit.
                                     Reasonable Agreement?
                                     Validated Explicit
Figure 2.  Schematic Description of the Mechanism Evaluation Process.

        IfMethane I
             I Ketones H
                          HCHO HCO

NO NO, NO, N20,

Higher Carbonyla

 PAN Compounds

                                                           Lowest Lerel
Figure 3.   Schematic Description  of  the Hierarchy of Species for the
           Carbon Bond  Mechanism-IV.

                             More Explicit
                             Verification vs.
                             Explicit Mechanism
Figure 4.  Schematic Description of the Mechanism Condensation Process.

                      Paper 6
Uncertainty Analysis and Captive-Air Studies as Aids in
Evaluating Chemical Mechanisms for Air Quality Models
                  Alan M Dunker
         Environmental Science Department
        General Motors Research Laboratories


     Chemical mechanisms employed in air quality models to describe

the formation of pollutants, such as ozone, are simplifications of

•what actually occurs in the atmosphere, and therefore the mechanisms

must be evaluated to determine their accuracy.  Concern has arisen

regarding how mechanisms are evaluated because different mechanisms

can predict different emission control requirements.  The procedure

for evaluating mechanisms can be improved through the use of

uncertainty analysis and the results of captive-air experiments.

     In uncertainty analysis, probability distributions are

estimated for the parameters in a mechanism, chiefly rate constants

and initial concentrations.  Using these probability distributions,

the variances in the concentrations predicted by the mechanism are

calculated.  The variances can be related to uncertainty limits, and

comparison of the uncertainty limits on the predicted concentrations

with the uncertainty limits on the experimental data shows whether

differences between the mechanism and the data are significant.

     In captive-air experiments, ambient air is loaded into Teflon

bags in the early morning and irradiated by sunlight throughout the

day.  The advantages of such experiments are that an ambient  mix of

hydrocarbons at ambient concentrations is used and the photolysis

rates correspond closely to those in the atmosphere.  Further,

multiple bags can be run on the same day with known amounts of clean

air and/or emitted compounds added to some bags.  The known

additions can simulate the effects of emission control strategies

and broaden the range of conditions studied.  Comparison of the

predictions of a mechanism with the data from captive-air

experiments provides tests of the mechanism under conditions close

to those in the atmosphere.

     Secondary pollutants, such as ozone (Do), are formed in the

lower  atmosphere by a complicated sequence of chemical reactions

which involve precursors, such as hydrocarbons and NO  .  Chemical

mechanisms, sets of chemical reactions, have been assembled to

describe this complex chemistry.  Coupled with descriptions of

transport and dispersion, chemical mechanisms are used in

atmospheric models to develop strategies for reducing the

concentrations of the secondary pollutants by controlling emissions

of the precursors.

     Because atmospheric  chemistry is  so complex and incompletely

understood, chemical mechanisms are by necessity condensations of

and approximations to what is actually occurring in the atmosphere.

As a consequence, there is not one uniformly accepted  chemical

mechanism but rather a number of different mechanisms, and as new

information on reactions  becomes available new mechanisms are

developed  (Killus and Whitten, 1981a;  Atkinson et al., 1982; Lurmann

et al., 1986; Stockwell,  1986).  Comparisons of different mechanisms

have shown that they give qualitatively similar results.  However,

the comparisons have also shown that quantitative predictions of

emissions reductions needed to meet the National Ambient Air Quality

Standard for 03 can be substantially different from different

mechanisms. (Jeffries et al., 1981; Carter et al., 1982a; Dunker et

al., 1984; Leone and Seinfeld, 1985; Shafer and Seinfeld, 1986)

     This difficulty has focused attention on the process employed

to evaluate mechanisms to determine how well they represent what

occurs in the polluted troposphere.  Jeffries and Arnold (1986) have

recently reviewed the history of how mechanisms are developed and

tested.  In the past, mechanisms have been evaluated by comparing

their predictions with the results of smog chamber experiments.

Such experiments attempt to simulate conditions in the atmosphere,

but the concentrations of the precursors used are generally higher

than found in the atmosphere and the mixtures of hydrocarbons used

represent only the major species found in the atmosphere.

Additional concerns have arisen regarding the effects of chamber

surfaces on the observed gas-phase chemistry. (Killus and Whitten,

1981b; Carter et al.,1982a, 1982b; Dunker et al., 1984; Leone and

Seinfeld, 1985; Shafer and Seinfeld, 1986).  The sources of

disagreement between mechanisms and chamber data, errors in the

mechanism versus errors in the data, very often are not clear.

     This report addresses the question of how to improve the

process for evaluating chemical mechanisms, and two suggestions are

made.  The first suggestion is to  use uncertainty analysis when

comparing a chemical mechanism to  chamber experiments, and the

second suggestion is to conduct captive-air experiments.  To date

these tools have rarely been applied in evaluating mechanisms for

the polluted troposphere.  The next section describes how

uncertainties arise in chemical mechanisms and what numerical

techniques are available to calculate uncertainty limits for the

predictions of the mechanisms.  This is followed by a section

discussing captive-air experiments and why they are a valuable tool

in determining whether chemical mechanisms adequately represent the

atmosphere.  The final section gives concluding remarks.
                        Uncertainty Limits
     A chemical mechanism is evaluated by using it to simulate

chamber experiments and then comparing the predictions of the

mechanism for the concentrations of different species with the

experimental measurements.  Both the measurements and the

predictions, however, have uncertainties associated with them.

Furthermore, the agreement between the measurements and the

predictions is never perfect, and the question then arises whether

the differences seen are significant.  If the differences are

significant, the chemical mechanism does not adequately represent

the data.  If the differences are not significant, the uncertainties

in the experimental data and/or in the results of the chemical

mechanism must be decreased to make a more precise evaluation of the

mechanism.  Uncertainty limits have often been placed on the data

acquired in chamber experiments.  However, very little work has been

done to place uncertainty limits on the results of chemical

mechanisms for simulations of the troposphere in general and chamber

experiments in particular.  This is in contrast to the study of

stratospheric chemistry where efforts have been made to place

uncertainties on the results of simulations (Stolarski et al., 1978;

Ehhalt et al., 1979; Stolarski and Douglass, 1986).

     There are several sources of uncertainty in a chemical

mechanism.  First, some important reactions may be completely

unknown and therefore not included in the mechanism.  Second, for

those reactions included in the mechanism, the rate constants are

not known precisely but rather are known to varying degrees of

accuracy.  Third, the products or the distribution of products for

some reactions are in question.  In addition, when a chemical

mechanism is used, initial concentrations must be supplied for all

the species in the mechanism.  Some of these initial concentrations

may not be measured, and those that are measured will have

uncertainties associated with them.  Important examples encountered

in simulating chamber experiments are the uncertainties in the

initial HONO concentration, the background OH radical flux, and the

aldehyde photolysis rates.

     The first source of uncertainty cited above, complete ignorance

of an important reaction, cannot be treated in an uncertainty

analysis.  One must at least suspect that a reaction occurs, know

the reactants, and have some estimate of the rate constant.  The

other sources of uncertainty can be treated, however.

     The different sources of uncertainty cause uncertainties in the

predictions of the chemical mechanism which vary with time in a

given simulation, vary from one simulation to another, and vary from

one chemical species to another.  A schematic representation of the

effect of errors in initial concentrations and rate constants on the

results of a mechanism is shown in Figure 1.  The points denote

measurements of the concentration C(t.) of some species in a chamber

experiment, and the vertical bars denote the uncertainty associated

with each measurement.  The curves represent results obtained with a

chemical mechanism for simulations of the experiment.  In practice,

one begins the simulation with the measured initial concentrations,

CQ , and uses in the chemical mechanism the approximate rate

constants, k  , measured in other experiments.  This produces the

calculated concentration c(t; CQ ,k ) as a function of time t (upper

curve).  If, however, the true initial concentrations, GO, and the

true rate constants, k, were known, these could be used in the

chemical mechanism to produce the concentration c(t; cQ,k)  (lower

curve).  The difference between the two curves represents the effect

of errors in the initial concentrations and rate constants on the



predictions of the mechanism.

     Quantifying the effects of possible errors or uncertainties in

cJ" and km .requires two steps.  First, the probability distributions

of cnm and km, P(com) and P(km), must be estimated.  Because the

true values cfl and k are not known, estimating the uncertainties in

c0m and km (and thereby estimating P(CQ ) and P(k )) necessarily

requires some subjective judgments.  The uncertainties in the

initial concentrations can be estimated based on knowledge of the

experimental techniques and the reproducibility of the measurements,

and these estimates are probably best made by the experimenters

themselves.  Estimates of uncertainties in rate constants are

available in some reviews of kinetic data  (Atkinson and Lloyd, 1984;

Demore et al., 1985).  Such reviews are very useful in that they

provide the most impartial, least subjective assessments of the

uncertainties in rate constants.

     The second step in the uncertainty analysis is calculating the

             2                                       2
variance 0"(t)  in the predicted concentrations.  fl"(t)  is defined  by
               <7(t)2 = <  [c(t; c0m,km) - ]2  >      (1)

         f (c0m,km) > = / f (c0m,km) P(c0m) P(km) dc0m dkm
for a function f.  The integration here is over all variables for

which uncertainties are considered.  If uncertainties in a large

number of rate constants and initial concentrations are included in

the analysis, calculating 
must be simulated anew for each of the random variations in CQ  , k

to obtain c(t; cQm,km).  Furthermore, ff(t)  will vary from one

chamber experiment to another, so the entire analysis will have to

be redone for each experiment.
     The second approach is the Fourier method  (Cukier et al.,

1978).  Here, cQm and km are varied simultaneously in a systematic

fashion and the resulting variations in c(t; cQm,k ) analyzed by

Fourier series to evaluate the multidimensional integrals.  The

method was originally developed for the sensitivity analysis  of

complex models, and Falls et al.  (1979) have applied the method to  a

chemical mechanism.  While calculations of cr(t)  have not been

reported, this quantity can be obtained by the  Fourier method.  The

Fourier method is more efficient  than the Monte Carlo method, but

the computing time required can still be large.

     The third approach is the sensitivity coefficient method in

which gradients of c(t; CQ ,k ) with respect to CQ  and k   are

calculated and employed to evaluate the multidimensional integrals.

Ehhalt et al.  (1979) have used this method, again, to place

uncertainty limits on the predictions of stratospheric ozone

depletion.  To calculate the sensitivity coefficients, Ehhalt et al.

(1979) made small perturbations in the variables cQm, km, one at a

time, and applied finite difference formulas.   A more accurate and

efficient method for calculating  the sensitivity coefficients, the

decoupled direct method, has been described by Dunker  (1984).

McCroskey (1985) and the Acid Deposition Modeling Project  (1986)

have used the decoupled direct method in calculating uncertainty

limits for simulations of some chamber experiments.  The sensitivity

coefficient method requires less computer time than the Monte Carlo

and Fourier methods and thus appears better suited for routine

application to chamber experiments.  The sensitivity coefficient

method does require care in treating cases with large  uncertainties

in GO , k , and for very large uncertainties may require extension

beyond present capabilities.

     Uncertainty limits corresponding to one standard  deviation can

be placed on c(t; c~  ,k ) using 

Figure 2.  Uncertainty limits  (dashed curves) for the  concentration

predicted by a chemical mechanism (solid curve) using  measured

(approximate) initial concentrations and rate constants.

that a necessary condition for good agreement between simulation

results and chamber data is that the uncertainty limits one standard

deviation wide overlap.

     Several additional remarks regarding uncertainty analysis

should be made.  First, if a complete analysis taking into account

the uncertainties in all the rate constants and initial

concentrations cannot be made, due perhaps to the complexity of the

mechanism or lack of information on uncertainties for some rate

constants, a partial analysis can still be useful.  For example, an

analysis taking into account only the uncertainties in the chamber

photolysis rates and the surface sinks and sources will show whether

discrepancies between the simulation results and the chamber data

can be explained by uncertainties in the chamber operating


     Second, an uncertainty analysis must be done with some care

since the uncertainties in the rate constants and initial

concentrations may not all be independent.  As an example, the NCL

photolysis rate in chamber experiments is often measured by the

photostationary-state method  (Wu and Niki, 1975).  This method gives

                        [NO] COg]
                                 k/                            (3)
where k-m and k0m are the measured rate constants for the reactions
       1       A
           N02 + hi/ — > NO + 0

NO + 0  — > N0
                           2 +
According to Eq.  (3), the uncertainty in k«  will cause uncertainty
in k~ , and the correlation between the uncertainties  in these  two

rate constants should be taken into account  in  specifying P(k..  )  and

in calculating ff(t)2  (Eqs.  (1) and  (2)).
     Finally, when applying  a  chemical mechanism,  one  often  compares

concentrations of the same species  obtained  in  different

simulations, e. g.,  comparison of 0_  concentrations  from  simulations

with different emission rates.   In  such  situations,  the variance  of

the concentration difference should be calculated  because it is

likely that  it will  be less  than the  sum of  the variances for the

separate simulations.  That  is,  one expects

            var[c(t)-c'(t)] < varCc(t)] +varCc'(t)]

where c(t) and c'(t) are concentrations obtained in different

simulations.  The reason for this is that the uncertainties in c(t)

and c'(t) will be correlated because the same rate constants with

the same errors will be used in both simulations.  There is hence a

tendency for errors to cancel when forming the difference c(t)-

c'(t).  Since the uncertainty limits are obtained from the

variances, using varCc(t)] + varCc'(t)3 to determine the uncertainty

limits for c(t)-c'(t) will likely give limits that are overly

pessimistic (unnecessarily wide).  The better approach, using

var[c(t)-c'(t)], has been employed in studies of stratospheric ozone

depletion, where differences in ozone concentrations are examined

(Stolarski et al., 1978; Ehhalt et al., 1979; Stolarski and

Douglass, 1986).
                        Captive-Air Studies
     In  a  captive-air  experiment,  ambient  air  in  an urban  area is

 loaded into  a Teflon bag  in  the  early morning, between  6 AM and 10

 AM.  The bag is  then irradiated  by sunlight throughout  the day,  and

 the  concentrations  of  the precursors and the secondary  pollutants

 are  measured periodically.   Such experiments are  similar to but have

several advantages over traditional smog chamber studies.

     The first advantage is that an ambient mix of hydrocarbons at

ambient concentrations is used in captive-air experiments.

Traditional chamber experiments, however, use mixtures containing

only the major hydrocarbon species found in the atmosphere,

generally at higher than ambient concentrations.  Second, the

photolysis rates in captive-air experiments correspond very closely

to those in the atmosphere because thin Teflon film is virtually

transparent to solar radiation (Kelly, 1981a).  The photolysis rates

in traditional chamber studies are not as close to atmospheric rates

because the spectrum of artificial light sources does not exactly

duplicate the solar spectrum.  Finally, captive-air experiments can

be run in several bags on the same day with known additions made to

some bags.  The known additions broaden the range of conditions

studied and can mimic the effects of control strategies.  Table I

describes some of the additions which can be made and their

relationship to strategies for reducing 0, concentrations. Changes

in concentrations of the precursors have, of course, been made in

traditional chamber experiments, but, again, the hydrocarbon mix to

which the perturbations are made is a simplification of the ambient

hydrocarbon mix.

     Considering these advantages, it is proposed here that captive-

air experiments can help bridge the gap between traditional smog








                     ••  -o





                                     «    (4
                                     bO   U
                                     C    O
                                     (4    h
                                    Jd   -O
                                     U    X

                                      X «H
                                    o    o
                          • H
•4-1  43
 O   W
     • H
 0)  *O
•H   e
 CO  -H





                                         ^   3
                                          0    t->
                                         43   43
                                                          ;    c4
                                              • H





• H

i— 1



f— 1

chamber experiments and the atmosphere.  This is not to suggest that

traditional chamber experiments should be discontinued; they are

valuable in building and evaluating chemical mechanisms.  However,

comparison of the results of a mechanism with data from captive-air

experiments provides an evaluation of the mechanism under conditions

as close as possible to those under which it will be applied.

     There are disadvantages to captive-air experiments as well.  If

experiments are run in several bags simultaneously, the surface

effects in each bag must be characterized or the surface effects

must be small and the expected variability from bag to bag must be

determined.  Otherwise one cannot determine whether differences seen

between experiments conducted in different bags are due to different

precursor concentrations or different surface effects.  Second, it

can be difficult or impossible to control some parameters in

captive-air experiments, such as light intensity, temperature, and

humidity.  As a result, the failure rate and cost of captive-air

experiments are higher.

      Lastly, since ambient air contains a complex mixture of

hydrocarbons, careful measurements of individual hydrocarbon species

must be made at the start of, and preferably throughout, the

experiments if they are to be used to evaluate chemical mechanisms.

This is the most significant disadvantage of captive-air experiments

because it is impossible to identify all the hydrocarbons in ambient

air.  In simulating the experiments, then, it is necessary to make

assumptions regarding the fraction of hydrocarbons which are

unidentified.  However, this problem must be faced because it is

also encountered when using a chemical mechanism in an air quality

model for simulations of an entire urban area.  If a significant

fraction of the hydrocarbons is unidentified, then a comparison

between the results of the mechanism and the data will provide a

test of the mechanism combined with the assumptions on the

unidentified hydrocarbons.

     Several captive-air studies have been conducted to date.  The

most ambitious study was done by Grosjean et al. (1982) in Los

Angeles in the fall of 1981.  Both large  (70,000 1) and small (4000

1) Teflon bags were used, and individual hydrocarbon species were

measured in the captured air.  Results from the large bag were

compared to results from one small bag to check the effect of

chamber size on the results, and additions of NO or NCL were made to

the other small bags.  Predictions of the Atkinson et al.  (1982)

chemical mechanism and the EKMA (Empirical Kinetic Modeling

Approach) mechanism  (Dodge, 1977) for CL, NO, N02, hydrocarbons, and

aldehydes were compared to the observed concentrations.

Unfortunately, experimental problems and  uncertainties in  the flux

of  OH radicals from  chamber surface reactions prevented a  rigorous

evaluation of the chemical mechanisms with the data.

     Kelly (1981b, 1985) has conducted captive-air studies in

Houston (summer, 1977) and Detroit (summer, 1981) .  In these

experiments, small (500 1) Teflon bags were used, and the total

concentration of nonmethane hydrocarbons, rather  than individual

hydrocarbon species, was measured in the captured air.  An extensive

set of additions to different bags was done; clean air, NO, NCL,

butane, ethene, and propene alone and in combination were added.

Using default estimates of the relative amounts of different

hydrocarbon species, predictions of the carbon bond III mechanism

(Killus and Whitten, 1981a) and the EKMA mechanism (Dodge, 1977) for

CL were compared to the observed concentrations.  While this

approach provides an indication of the validity of the mechanisms,

detailed measurements of hydrocarbon species are  necessary for a

rigorous evaluation.  As part of the Southern California Air Quality

Study in the summer of 1987 (Blumenthal et al., 1986), General

Motors Research plans additional captive-air experiments in Los

Angeles in which individual hydrocarbon species will be measured.

This data should allow more thorough testing of chemical mechanisms.
     It has been recognized  for  some  time  that  a  critical  element of

an air quality model  is  the  chemical  mechanism.   In  designing

strategies for reducing secondary pollutants such as 0^,

inaccuracies in the chemical mechanism lead to inappropriate and

ineffective strategies.  It has also been recognized that evaluating

chemical mechanisms to determine how well they represent the

transformations actually occurring in the atmosphere is not a simple

task. A review of past procedures for developing and evaluating

mechanisms (Jeffries and Arnold, 1986) and consideration of the fact

that current mechanisms predict different control requirements

(Jeffries et al., 1981; Carter et al., 1982a; Dunker et al., 1984;

Shafer and Seinfeld, 1986) indicates that improvements are necessary

in how mechanisms are evaluated.

     As described above, uncertainty analysis and captive-air

experiments are two tools which can aid in evaluating mechanisms.

Uncertainty analysis can quantify the agreement between chamber data

and the predictions of a mechanism and show whether there is a

fundamental disagreement between the data and predictions.

Uncertainty analysis can also place error bounds on the predictions

of a mechanism in an application to the atmosphere.  Further,

uncertainty analysis can indicate how the error bounds can best be

reduced.  That is, the rate constants, product yields, and initial

concentrations contributing most to the uncertainty in the

predictions of the mechanism can be identified.  Captive-air

experiments provide tests of a  mechanism under conditions a step

closer to those  in the atmosphere than traditional smog chamber

experiments.  Because the complexity of captive-air experiments is

greater, greater effort is required to achieve the same quality of


     Uncertainty analysis and captive-air experiments are not the

only tools which can improve the procedure for evaluating

mechanisms. They are, however, tools which have received little

attention to date.

Acid Deposition Modeling Project (1986) Preliminary evaluation
     studies with the regional•acid deposition model (RADM),
     National Center for Atmospheric Research Technical Note TN-

R. Atkinson, A. C. Lloyd, L. Winges (1982) An updated chemical
     mechanism for hydrocarbon/NO /S00 photooxidations suitable for
                                 x   *
     inclusion in atmospheric simulation models. Atmos. Environ. 16,

R. Atkinson, A. C. Lloyd  (1984)  Evaluation of kinetic and
     mechanistic data for modeling of photochemical smog. J. Phys.
     Chem. Ref. Data 13, 315-444.

D. Blumenthal, J. Watson, D. Lawson (1986) Southern California air
     quality study suggested program plan. Sonoma Technology Inc.
     Report 10 95050-605.

W. P. L. Carter, A. M. Winer, J. N. Pitts, Jr.  (1982a) Effects of
     kinetic mechanisms  and hydrocarbon composition on oxidant-
     precursor relationships predicted by the EKMA isopleth
     technique. Atmos. Environ. 16, 113-120.

W. P. L. Carter, R. Atkinson, A. M. Winer, J. N. Pitts, Jr.  (1982b)
     Experimental investigation of chamber-dependent radical
     sources. Int. J. Chem. Kinet. 14, 1071-1103.

R. I. Cukier, H. B. Levine, K. E. Shuler  (1978) Nonlinear
     sensitivity analysis of multiparameter model systems. J.
     Comput. Phys. 26, 1-42.

W. B. Demore, J. J. Margitan, M. J. Molina, R. T. Watson, D. M.
     Golden, R. F. Hampson, M. J. Kurylo, C. J. Howard, A. R.
     Ravishankara  (1985) Chemical kinetics and photochemical data
     for use in stratospheric modeling. California Institute of
     Technology Jet Propulsion Laboratory Publication 85-37.

M. C. Dodge  (1977) Combined use of modeling techniques and smog
     chamber data to derive ozone-precursor relationships. U. S.
     Environmental Protection Agency EPA-600/3-77-001.

A. M. Dunker (1984) The decoupled direct method for calculating
     sensitivity coefficients in chemical kinetics. J. Chem. Phys.
     81, 2385-2393.

A. M. Dunker, S. Kumar, P. H. Berzins  (1984) A comparison of
     chemical mechanisms used in atmospheric models.  Atmos.
     Environ. 18, 311-321.

D. H. Ehhalt, J. S. Chang, D. M. Butler  (1979) The probability
     distribution of the predicted CFM-induced ozone depletion. J.
     Geophys. Res. 84, 7889-7894.

A. H. Falls, G. J. McRae, J. H. Seinfeld  (1979) Sensitivity and
     Uncertainty of Reaction Mechanisms for Photochemical Air
     Pollution. Int. J. Chem. Kinet. 11,  1137-1162.

D. Grosjean, A. C. Lloyd, R. J. Countess, F. Lurmann, K. Fung  (1982)
     Captive air experiments in support of photochemical kinetic
     model  evaluation. Environmental Research and Technology
     Document No. P-A764-500,

H. E. Jeffries, J. Arnold (1986) The science of photochemical
     reaction mechanism development and evaluation. U. S.
     Environmental Protection Agency, Workshop on evaluation and
     documentation of chemical mechanisms used in air quality

H. E. Jeffries, K. G. Sexton, C. N. Salmi (1981) The effect of
     chemistry and meteorology on ozone control calculations using
     simple trajectory models and the EKMA procedure. U. S.
     Environmental Protection Agency EPA-450/4-81-034.

N. A. Kelly (1981a) The Characterization of Fluorocarbon-Film Bags
     as Smog Chambers. General Motors Research Laboratories
     Publication GMR-3735.

N. A. Kelly (1981b) An analysis of ozone generation in irradiated
     Houston air. J. Air Pollut. Control Assoc. 31, 565-567.

N. A. Kelly (1985) Ozone/precursor relationships in the Detroit
     metropolitan area derived from captive-air irradiations and an
     empirical photochemical model. J. Air Pollut. Control Assoc.
     35, 27-34.

J. P. Killus, G. Z. Whitten  (1981a) A new carbon-bond mechanism for
     air quality simulation modeling. Systems Applications Inc.
     Report No. 81245.
J. P. Killus, G.  Z. Whitten  (1981b) Comments on  "A smog  chamber  and
     modeling study of the gas phase NO -air photooxidation  of
     toluene and  the  cresols". Int. J. Chem. Kinet.  13,  1101-1103.

J. A. Leone, J. H. Seinfeld (1985) Comparative analysis of chemical
     reaction mechanisms for photochemical smog. Atmos. Environ 19,

F. W. Lurmann, A. C. Lloyd, R. Atkinson (1986) A chemical mechanism
     for use in long-range transport/acid deposition computer
     modeling. J. Geophys. Res. 91, 10905-10936.

P. S. McCroskey (1985) Direct differential methods for kinetic
     parameter estimation. Carnegie-Mellon University, Master's

T. B. Shafer, J. H. Seinfeld (1986) Comparative analysis of chemical
     reaction mechanisms for photochemical smog  II. Sensitivity of
     EKMA to chemical mechanism and input parameters. Atmos.
     Environ. 20, 487-499.

W. R. Stockwell (1986) A homogeneous gas phase mechanism for use in
     a regional acid deposition model. Atmos. Environ. 20, 1615-

R. S. Stolarski, D. M. Butler, R. D. Rundel  (1978) Uncertainty
     propagation in a stratospheric model 2. Monte Carlo analysis of
     imprecisions due to reaction rates, 83, 3074-3078.

R. S. Stolarski, A. R. Douglass  (1986) Sensitivity of an atmospheric
     photochemistry model to chlorine perturbations including
     consideration of uncertainty propagation. J. Geophys. Res. 91,

C. H. Wu, H. Niki  (1975) Methods for measuring NO,, Photodissociation
     Rate:  Application to Smog Chamber Studies. Environ. Sci.
     Technol.  9, 45-52.

                       Discussion Following Dunker Presentation
  Sexton:  In the discussion of calculating uncertainty, you started with an
"if" statement, that if we knew all the reactions properly and all the products
then you could go on and discuss the uncertainty from the kinetic, the rate
constant, then yesterday Roger mentioned that we actually seem to know less
about the proper reactions and the products than we do in the kinetics. Do you
have any suggestions for how to deal with, or how to calculate, or estimate the
uncertainty, not knowing  the chemistry in the products correctly?

  Dunker: Consider a situation where you are developing a chemical mechanism
and a particular reaction is known to proceed via two pathways to two different
products, A and B, but the relative amounts of these products produced is  not
well known.  In such a situation, you can write a single reaction and assign a
best guess for the stoichiometric coefficients of the two products.  Then, you
would place fairly wide uncertainty limits on the two stoichiometric
coefficients and determine the effects of these uncertainties on predicted
concentrations obtained from the chemical mechanism. (In carrying out the
uncertainty analysis, it is  important to remember that the stoichiometric
coefficients for products A and B in this case are correlated. That is, if 80%
of the reaction proceeds to product A, then 20%  must proceed to product B; if
30% of the reaction proceeds to product A, then  70% must proceed to product B;
and so forth.) If the uncertainty analysis shows that  the results of the
chemical mechanism have low uncertainty despite the high uncertainty in the
stoichiometric coefficients for products A and B, then you need not worry about
this reaction further.  However, if the uncertainty in the stoichiometric
coefficients causes large uncertainty in  the results of the chemical mechanism,
then you would have to flag  this reaction as  one which someone should
investigate in detail.

The uncertainty analysis can be used to determine where it is most useful  to
concentrate additional experimental effort.  By ranking the contributions of
different rate constants and stoichiometric coefficients to the overall
uncertainty in the predictions of the chemical mechanism, you can see which are
the critical reactions to study.  The uncertainty analysis can be done at
various levels of detail. I think the simplest  analysis would be  to consider
just those chamber effects which are not well characterized, place
uncertainties on these effects, and see what  the uncertainty analysis tells
you. If it tells you  that the  uncertainty  limits on the predictions of the
chemical mechanism for the chamber experiments are large, then, before you do
anything else, you should  go  back and carefully  characterize the chamber.
Thus, an uncertainty analysis may show  that it is impossible to make reliable
predictions with a chemical  mechanism  without further experimental work.

  Jeffries: Alan, would you estimate what it might cost to do,  let's say a
seven or eight simultaneous bag captive air study for producing enough
reasonable data. By seven or eight bag, I mean you're looking at seven or
eight things at  once.

  Dunker: I  am trying to remember how much the program carried out by ERT in
Los Angeles in 1981 cost.  I believe the cost  was  $200,000  - $250,000, and,
apart from inflation, that is probably a good estimate of the cost of a
captive-air study.


  Jeffries: Strikes me, looking at what happened, that may not be enough?

  Lloyd:  I agree.  Some of the problems we had indicated that it wasn't.

  Lurmann:  I think for $400,000, you should be able to do some very good
runs. That would be, the number of runs that were done were on the order of 30,
total 30 days 20 to 30 days.  You are getting paramount data, in that, if
you're  doing  one run with the ambient air plus  seven or lots of different bags,
you've got alot of individual runs on that.  If everything worked right, you'd
have on the order of 200 or more different conditions. That would be alot of
data, if everything worked.

  Dodge:  I don't think anyone will disagree that irradiating authentic
atmospheric samples is not a preferred route to  go. I mean, obviously that's
considerably  better than irradiating surrogate mixtures, but it just seems from
the past work that's done, that its just frought with experimental
difficulties.  You're dealing with concentrations that are  extremely, well the  .
results are going to be dominated by surface effects,  at those concentrations,
the surface effects are variable from run to run, bag to bag. And, I just
can't see it really working.   Past studies have demonstrated that the data is
just not of sufficient quality that you can run test mechanisms.

  Dunker: If the bags are loaded during the 6-9 AM period near  the center of
a city, the concentrations of the precursors should be high enough to conduct
meaningful experiments. In  the experiments conducted by ERT in Los Angeles, the
initial  concentration of the nonmethane hydrocarbons (NMHC) exceeded 2.0 ppmC
on  15 of 23 days. On four of those days the initial concentration of NMHC was
5.0  ppmC or higher.  For the experiments done by Nelson Kelly in Houston, the
initial  concentration of NMHC exceeded 1.5 ppmC  on 9 or 25 days, and for the
experiments he did in Detroit the initial NMHC exceeded 0.7 ppmC on 5 of 12
days.  On some days the initial concentrations may be so  high that it is
actually desirable to dilute the  contents of  the bag during the course of the
day to simulate  the dilution which occurs in the atmosphere as a result of the
increase in the mixing height.

If surface effects in the bags are believed to be a problem, one can compare
results from a large bag with results from smaller bags.  This was done, for
example, in the ERT study in which one 70,000 1 bag was used along with several
4000 1  bags.  I think that most of the experimental problems encountered in
conducting captive-air experiments are the  same ones encountered in operating
an outdoor smog chamber.  You have to characterize  the chambers and you have to
make careful measurements  of light intensity.  What I see as the big problem in
captive-air experiments is that you must make detailed hydrocarbon measurements
during the runs. This data is essential to simulate the experiments with a
chemical mechanism and  to evaluate the chemical mechanism using the

  Dodge:  The concentrations that are used in the outdoor chambers are usually
considerably  higher than  even your  LA data met. Take the UNC chamber, it's
extremely large when you're talking about working with Teflon bags and
surface/volume ratio.

  Bufalini:  In following what Marcia was just suggesting, we've been involved
for a number of years doing detailed hydrocarbon analysis and looking over our
data as well as Washington State University data, it's not terribly unusual to
run into about 20% of the organics as being unidentified. You've got a large
number of unknown  peaks as Marcia has suggested.  Our approach has been to send
it through scrubbers, in some cases we haven't been able to identify  either
aromatic, parafinic, or olefinic, though it's not exactly clear  if they're
oxygenated. I/we might possibly miss them, and we know approximately what
range they're in, whether they are C8's or C9's or C6's, but still there are a
number of peaks that are still unidentified. 20% is not unusual and I don't
see how you're going to handle them in your model exactly, I mean you could
make some sort of approximation of the carbon bond as  to what they are, but I
still think there's an  amount of uncertainty there.

  Dunker:  I agree that measuring the hydrocarbon species is difficult and it
may not be possible to measure all the species occurring  in the atmosphere.
That is why I  feel that making the hydrocarbon measurements is the major
problem in captive-air experiments.  What we  should remember, though, is that
we  are trying to create chemical mechanisms for application  to the real world.
If we cannot identify 20% of the hydrocarbons occurring in  the atmosphere, then
we  will have to make some assumptions, either explicit or implicit, when
applying any chemical mechanism or model to simulate the real world.  For
example, the simplest assumption is  that the unidentified hydrocarbons have the
same relative amounts of alkanes, alkenes, and aromatics as the identified
hydrocarbons.  If we conduct captive-air experiments and apply whatever
assumptions we choose to make for  the unidentified hydrocarbons, we will have
tested whether the chemical  mechanism plus the auxiliary assumptions for the
unidentified hydrocarbons adequately represents the chemistry in the real
world.  If the agreement with the captive-air experiments is poor, then using
the chemical mechanism in a sophisticated, three-dimensional atmospheric model
with the same assumptions for the unidentified hydrocarbons will very likely
not produce good results.  Conversely, if we do not evaluate chemical
mechanisms using captive-air experiments, we will not have checked an important
set  of assumptions, namely those regarding the unidentified hydrocarbons.

  Dodge:  Alan, I think in almost all cases that you would be able to compare
because you've got so many uncertain parameters.  You can vary their
distribution of hydrocarbons in the  mechanism because you have that 20 to  40%,
you've got all the uncertainty in the wall effects, etc., etc.  You've got
enough adjustable parameters that you probably could get a good agreement but
I'm not sure you could use that as a test discriminator although you're
probably correct that you could use  that to sort of lock in a default
hydrocarbon concentration.

  Dunker:  If a significant fraction of the hydrocarbons cannot be identified,
then the captive-air experiments will only provide a test of the chemical
mechanism together with the auxiliary assumptions for the unidentified
hydrocarbons and not a test of the chemical mechanism by itself.  As you
suggest for such a situation,  the experiments could be used to develop the best
auxiliary assumptions for the unidentified hydrocarbons. One does not have
unlimited  degrees of freedom in developing the auxiliary assumptions, however.
That is, whatever assumptions one develops for the unidentified hydrocarbons
should be  applied uniformly to all captive-air experiments at a particular
location. If a  substantial number of captive-air experiments  are done, it may


still be possible to conclude that a particular chemical mechanism with its
best auxiliary assumptions for the unidentified hydrocarbons performs better
than another chemical mechanism with its best auxiliary assumptions.

  Jeffries: The real difference is that the two PPMC in the chamber's present
13 compounds which are well characterized, can be analyzed with high precision,
and can be followed in great detail.  I would have to agree with Marcia in the
sense, and Joe too, that when we  try automobile exhaust work in our chamber, we
couldn't find 20% of the hydrocarbon most of the time, and it turns out that if
you look at the distribution and try to match that against all kinds  of stuff,
there is enough flexibility in the modeler's choices about where he's going to
put that 20% that he can make the model fit.  It's not as sensitive a test to
test the model against automobile exhaust as it is to test it against
individual pure compounds where you know in detail what's going on.  That's the

  Dunker: I agree that in standard chamber experiments  using a known mix of
hydrocarbons you can accurately follow the  concentration of each hydrocarbon.
Furthermore, I am not advocating discontinuing such experiments; I think we
should continue  them.  I  also think we should  face the fact that a surrogate
mix of hydrocarbons is not the same as what occurs in the atmosphere and deal
with that  problem.  If 20% of the hydrocarbons in the atmosphere cannot be
identified, we should use captive-air experiments to test the chemical
mechanisms together with whatever assumptions we intend to make  regarding the
unidentified hydrocarbons.

  Jeffries: I would agree that you would be very unhappy if it turns out  that
within the range of adjustments,  you couldn't  fit the captive air experiments.
But fitting the captive air experiments doesn't distinguish one from another.
By not fitting it, clearly  you would be able to  distinguish one from  all the
rest that might be  able to fit it, so in  a sense, I agree with you it's a
necessary  but not sufficient condition.

  Whitten: I think that the captive air experiments provide a worthwhile
supplement to all the other types  of experiments, but I don't think they
preempt a lot. I think in the future, a very  important part of these
experiments should include a careful characterization of the chamber surfaces.
There are a certain battery of tests that are  being developed in various places
(I'm going to talk about that later today) that provide some information and
this battery of tests could be included in the thing.  A half step removed from
captive air experiments are two different types of experiments. One is that we
put genuine  automobile exhaust into a fairly well characterized chamber.  I
think Harvey Jeffries was a little optimistic in saying that 20% is
unidentifiable. Our experience was sometimes a little more than that.
Especially with some fairly reactive compounds that were difficult  to
unequivocally identify with because of analytical problems.  Another type of
experiment that has been done in Sydney, Australia, where they did a rather
careful analysis of the mixture of hydrocarbons and they  put together an urban
mix, which to the best of their ability, has manufactured all the compounds
that they were able to identify.  They had as many as 80 or  90.

  Jeffries: They actually went to the individual factories and acquired
samples of solvents and mixtures  and  the whole deal, so they were able to  put
together a mixture of three or four hundred identifiable individual compounds.


  Whitten: And then they did smog chamber experiments with this very high
number of compounds and they also used those experiments to test some chemical
mechanisms and I think that would probably be a little better than your
suggestion of holding back 20%.  I think going to an entirely different
facility with  different  analytical techniques and a different type of system,
perhaps would be a better test of a mechanism.  That way they're not being used
in the same facility. It is really important that we have data from a group of
smog chamber facilities rather than just one or two.  It provides much more
justification and scientific satisfaction, whatever the word is, for believing
that the mechanisms really work in the atmosphere as well.

  Dunker:  I  agree that chamber experiments with complex surrogate mixtures
will provide good tests  of chemical mechanisms, and I agree that we  need
chamber data from more  than one or two facilities.  I view captive-air
experiments as a bridge between traditional chamber experiments and the
atmosphere.  Captive-air experiments can be criticized as dealing with a very
complicated mix of hydrocarbon species with no hope of identifying and
measuring all the different species.  On the other hand, I do not feel that we
can develop chemical mechanisms solely from chamber experiments using
well-known hydrocarbons and then make the leap to atmospheric modeling.
Somewhere in that leap we will have made assumptions regarding the mix of
(unidentified) hydrocarbons found in the atmosphere, and those assumptions will
never have been tested.

  Whitten: You make a good point there.  There were some experiments done at
the University of Santa Clara where there was a problem with emissions from oil
vents in the San Joaquin  Valley and less than 10% of those hydrocarbons have
been identified*  It consisted of alot of hydrocarbons greater than CIO but
yet, right after  the vent,  the concentration in the air was maybe 20,000 parts
per million.  So it was easy to fill a cylinder with the emissions from these
vents and  it was totally unknown as to what these hydrocarbons were. These
hydrocarbons were then injected into a smog chamber, subjected to smog  chamber
experiments and then compared with other known compounds  to at least get  a
relative feeling for the reactivity of this emission source.  Even though it
had hundreds of compounds  that we couldn't identify, that made it  possible to
include that emission in atmospheric simulation. I think  it's very important
to have these types of captive air experiments where you can't identify the
hydrocarbons. I might  add that if you could identify them, I'm not sure  we'd
know what the  individual chemistry of these unknown compounds would be.  That's
even more scary, but yet  we can watch them in a smog chamber and  see what their
relative reactivity is to other  known things. That's an important, useful
piece of information.

  Demerjian: I had a comment that gets back to some of the things that  I
talked  about yesterday.  Alan, you had something on there about comparison with
smog chamber data, that  if you continue to accept it, you could show that the
walls were dominating. I guess one of the problems I have with that is that's
sort of one of these kind  of statements where, well if no one goes and looks,
then how will they ever discover that there's all these problems and therefore
we continue  to have this  cycle that goes on. My feeling is when I was talking
about this sort of aggressive approach into trying to  really understand what's
going on with chambers was  across the board, including even these captive air
experiments which, granted, would be more complex  to characterize just because


of the fact that you're bringing in different ambient conditions in a smaller
volume of air.  But there is the chance, but I guess my feeling is that what's
needed is a very well established procedure in terms of how one actually looks
at these effects, how you characterize them.  One that's done in a way that
gives the people using it information, a very  clear understanding of not only
the sources of these things but also the potential sinks and as well as the
fact of the variability that one might expect, given the type of chamber, if
it's an indoor system, one has obviously more control has on the system.  If
it's an outdoor system, one has less control and one has to understand how it
all factors in to this process.  It seems to me if there's not going to be an
aggressive approach to try to get at those things, then we continue to go
around in circles chasing our tail.  We've all acknowledged that there's a
problem but no one seems to want  to take hold of it. It's like, if we  were
sitting around here planning a manned space flight program, we wouldn't be
talking about hot air  balloons, okay?  And, sometimes I get the feeling that
that's what we're doing here. We're talking about using a technology that
maybe needs to be worked on in a  way that we start to get a handle  on some of
these problems. We know what the issues are, and yet we don't seem to ever try
to get a hold of what they are and solve them.  I don't care how many times you
keep doing these experiments over  and over again, if they still have the
inherent sources of error that we're starting to believe are there and we don't
seem to want to take the effort to  look at and understand it, then I  don't see
how you can expect to proceed.  As far as I'm concerned, it's not enough to say
that the captive air experiment is a complicated problem that has this many
things.  I think we can better look  at what  the problems are  and let's go see
if we can get a handle on that. Otherwise, we'll go  no place. We'll continue
to stay ten years behind.

  Dunker: I agree with you completely on the importance of characterizing
chambers.  I may not have stressed that in my talk, but I feel it is extremely
important.  Bill Stockwell has done a few calculations in which he put
uncertainty estimates on the chamber parameters he felt were not well known,
carried through an uncertainty analysis, and  found large uncertainties in the
concentrations of some species predicted by his chemical mechanism. Unless
chambers are carefully characterized, it will  be difficult or impossible to use
the experimental data to discriminate between chemical mechanisms. I think
that characterization of a chamber must be a continual process. That is, you
do a series of characterization runs, then do some hydrocarbon/NOx runs of
interest, then return and do the characterization runs all over again.  At
present I do not believe we know whether or how a  series of hydrocarbon/NOx
runs will change the surface characteristics of a chamber, and so
characterization runs must be scheduled periodically as part of any
experimental program. My one concern is that understanding in chemical detail
what is occurring on the chamber surfaces may be very difficult, and the
important processes may vary from one chamber to another.  We may always be
forced to use an empirical (but hopefully accurate) treatment of surface

  Demerjian:  At least if there were some standard set of procedures that
could be  carried out through all of them, that would be one thing you could use
to make comparisons in terms of surface characteristics.

  Bradow:  In  the general area, one of the things that Alan talked about was
the question about extending the ranges of experimentation.  One of  the places


where I think most of the models, what we're trying to apply to models these
days involves, multi-day incidence, long-range transport, regional ozone
problems, high ozone to NOx multiples, very low NOx values, all those kind of
conditions seem to be almost inaccessible to the current measurements.  I don't
know that this is  necessarily the case, but certainly these very high
hydrocarbon/low  NOx experiments are very hard to do.  Seems to me, I support
what Ken says, I  think there needs to be someway to either expand the range of
experimentation or range of experimental techniques to accomplish these very
real conditions that are very, very important in terms of the application of

  Barnes: I take  it that no chamber walls have been characterized.  Do you
agree with that?

  Carter: Not completely characterized.  There have been attempts to

  Jeffries: You can  predict the effect but you don't know how to explain it.

                        Paper 7
Development and Testing of a Surrogate Species Reaction
  Mechanism for Use in Atmospheric Simulation Models
                 William P. L. Carter
        Statewide Air Pollution Research Center


                          William P. L. Carter

                 Statewide Air Pollution Research Center
                       University of California
                     Riverside, California  92521

    In this presentation, I shall discuss the first phase of a two-phase
program we have carried out to develop, test, and adapt an alternative
chemical mechanism for use in EKMA models.   This program was carried out by
Roger Atkinson and myself at the Statewide Air Pollution Research Center
(SAPRC), and by Fred Lurmann and Alan Lloyd at Environmental Research and
Technology (ERT).   The EPA contract monitor for this program is Marcia
Dodge, and I also wish to acknowledge her contributions to this work.  The
first phase of this program was carried out primarily at SAPRC and the
second phase was carried out at ERT.

    In the first phase of this program, a detailed chemical mechanism for
the reactions of representative alkanes,  alkenes, aromatics,  and their
major photooxidation products was developed, using the mechanism of
Atkinson, Lloyd and Winges (ALW) (Atkinson et al, 1982) as the starting
point.  This detailed mechanism was tested against the results of over 490
environmental chamber experiments carried out in the outdoor chamber at the
University of North Carolina (UNC), and in two indoor and one outdoor
chambers at the Statewide Air Pollution Research Center (SAPRC) in
Riverside, California.  Testing this mechanism required the development of
a consistent chamber characterization model for the four different
chambers.  This is, I believe,  the first time that a single reaction
mechanism has been tested against such a large and comprehensive data base
of chamber experiments with a consistent set of assumptions regarding
chamber effects.  This discussion will concern primarily the testing aspect
of the mechanism,  but I will also briefly summarize the work done on
updating and extending the chemical mechanism.

    The second phase of this program involved primarily the adaptation of
the mechanism developed and tested in the first phase for use in AQSM
models.  In this phase, the detailed mechanism was condensed, using the
"surrogate species" approach,  so it is more appropriate for use in AQSM
models.  The sensitivities of control requirements,  predicted using the
EKMA technique, to input data were also assesed.   Finally,  the appropriate

procedures and defaluts for using this mechanisms were established and
documented.  This phase of the program is the subject of Fred Lurmann's
presentation, so it will not be discussed further here.

Major Features of the Updated Mechanism

    The chemical mechanism which was developed and tested in thes study
includes detailed reaction schemes for representative alkanes,  alkenes,  and
aromatic hydrocarbons, and for their major oxygenated and organic nitrate
products.  It includes explicit reaction schemes for all compounds for
which single component environmental chamber data are available.  This
required adding compounds in this mechanism which were not included in the
ALW mechanism, such as, for example, 1-butene and 1,3,5-trimethylbenzene,
as well as several individual alkanes which are not represented explicitly
by ALW.

    The detailed mechanism developed in this program is intended to serve
as a "master mechanism",  against which more condensed mechanisms (which may
be more practical to use in AQSM's) can be compared.  This is based on our
belief that the best approach is to use as accurate and detailed a
mechanism as possible when testing against chamber data, so what is being
tested is the chemistry and not the lumping techniques, and that the best
approach for testing lumping techniques is to compare predictions of lumped
models against the detailed mechanism.  This is why the mechanism included
explicit representations of the species in the chamber experiments used to
test it.  In addition, where practical, other chemical approximations were
minimized.  For example,  the ALW mechanism ignored reactions of organics
with 0( P) atoms and NO, radicals, which limited the range of validity of
that model.  These reactions were included in this mechanism.  On the other
hand, condensation techniques which do not include significant chemical
approximations, such as "lumping" of complex reaction schemes into a single
overall process which has the same overall effect, were employed to keep
the mechanism down to a managable size.  In addition, reactions which are
obviously of negligible importance under any reasonable range of
atmospheric or environmental chamber conditions were excluded from the

    Although known chemical approximations were minimized in this
mechanism, the mechanism had to represent processes whose details are
unknown.  Other than perhaps the representation of chamber effects
(discussed later), the greatest area of uncertainty concerns the ring
opening processes in the aromatic photooxidation mechanism, and the nature
and reactions of the highly reactive products formed.  Recent laboratory
data have indicated that previous mechanisms for these processes are
incorrect, and that we actually know less than we thought we did
previously.  In such cases, the development of this mechanism was based on
the philosophy that "if you have to represent something in your model that
you don't understand, simpler is better".  This is our philosophy with
regard to representation of chamber effects (discussed below),  and it
applies equally well to our representation of the formation and reactions
of uncharacterized aromatic ring-opening processes.  Thus, rather than

attempting to devise some type of detailed speculative mechanism for these
processes, the unknown aspects of the aromatic photooxidation mechanism was
represented in a parameterized manner, with two surrogate species used to
represent the reactions of the of the (probably) many uncharacterized
ring-opened products, and with the yields and photolysis rates of these
products being adjusted to fit the chamber data.  This is obviously an
unsatisfactory situation, but until more detailed and quantitative
mechanistic and product yield data are available, we really have no other

    Another area of chemical approximation which current models in practice
cannot avoid is the representation of the peroxy + peroxy radical
reactions, which become important in the absense of NO  or, under some
conditions, at nighttime.  Test calculations we have carried out have shown
that neglecting these reactions entirely is not an acceptable approximation
under those conditions.  However, representing these explicitly in the
model not practical, because of the many different types of peroxy radicals
involved, and the fact that such a representation would require all the
possible cross-reactions to be included.  Thus, in this model, we employed
an technique for representing these processes in a relatively efficient
manner, which requires the addition of only a few reactions and species to
the model.  I will not discuss the details of this method further here,
except to note that is a similar type of approach to that employed by
Lurmann et al. (1986) in the ADOM mechanism.

    Another area of approximation which for all practical purposes is
necessary is the representation of the many different types of by- and
poly-functional oxygenated products which are expected to be formed in the
reactions of the higher alkanes (Carter and Atkinson 1985).  These were
represented by a limited number of species in the model, according to a
modified "surrogate species" approach.  For example, "RCHO",  or
propionaldehyde, was used to represent the reactions at aldehyde groups of
such species, and "MEK", or methyl ethyl ketone, was used to represent
reactions at other carbonly or -OH groups.  This is in line with the
representations used in other models, and a more detailed representation of
the reactions of these bi- and poly-functional model would significantly
increase the complexity (and level of speculation) of the model without
significantly affecting its predictions.

    I should point out that the current version of the mechanism,  which was
used for the calculations whose results I will summarize in this
presentation, is different in some respects than the mechanism documented
in detail in our published report on Phase I of this program (Carter et al.
1986a).  The current version of the mechanism has a somewhat more efficient
representation of the peroxy + peroxy reactions than employed in the
previous model,  though this does not have a substantative effect on the
results of model predictions.   The most substantative difference is that
the methods used to represent the unknown aspects of the aromatic
photooxidation mechanism was modified to represent them in a more efficient
manner.  These changes are documented in the Phase II report on this
program (Lurmann et al. 1987).   However, in most other respects,  the

mechanism whose results are discussed here (and in Fred Lurmann's
presentation) is the same as that documented in detail in our previous
report (Carter et al. 1986a).

Philosophy Used in Testing Mechanism

    When testing this mechanism against the chamber data, and in some cases
modifying the mechanism as a' result of the fits obtained, we attempted to
adhere to the following philosophy, which we believe is appropriate in
carrying out such a program.

    o Where practical, all reactants in the experiments were represented
explicitly.  This is because we are testing our best estimate of the
chemical processes involved, and not techniques for approximating them.  As
indicated previously, testing approximation techniques can be done much
more appropriately (and with a higher degree of sensitivity) by model
calculations using the detailed mechanism as the standard.

    o A consistent set of assumptions was used in modeling all runs.  The
same chemical mechanism was used for all runs, and a self-consistent set of
assumptions regarding chamber effects were employed.  If, during the
process of model development,  we made any change to the chemical mechanism
or how we assigned values for chamber-dependent parameters, then all the
runs were re-calculated.  I might note that complete re-calculations of all
the runs were done -several times throughout this program.

    o No run-to-run adjustment of parameters was done in order to improve
the fits for individual runs.   The input data for modeling an individual
run was based entirely on experimental measurements (or in some cases the
recommendations of the experimentalists) and on assignments of chamber
dependent parameters for groups of runs.  [By "groups of runs" in this
context we mean all runs carried out in a given chamber, or (in the case of
light characterization for the UNC chamber) all runs charied out in a given
chamber in a given year.]  This approach might be questioned by other
modelers, since in reality some chamber-dependent parameters may in fact
vary from run-to-run in a manner that cannot be determined a-priori.
However, in our opinion, run-to-run adjustments of values of unknown
parameters invalidate the entire purpose of using chamber data for model
testing, since such a procedure might well result in disguising systematic
errors in the chemical mechanism being tested.  If parameters do indeed
vary from run-to-run, this variability will show up as variabilities in the
quality of the fit of the model to the experiment, where average or
"typical" values for the variable parameters are used in the model.  If
there are enough experiments,  than an examination of the pattern of these
discrepancies might reveal whether there are systematic problems with the
mechanism, or whether the only problem is variability of conditions.
Fortunately, in many cases, there are now enough experiments that such an
approach based on distributions of fits, rather than fits to individual
runs, can profitably be employed.

    o In line with the philosophy that "simpler is better" when
representing effects that are not understood, the various poorly-understood
chamber dependent parameters were represented in this model as simply as
posssible, with the absolute minimum number of parameters necessary to
describe the data.  Limiting the number of parameters also greatly
simplified the process of determining the sets of values which best fits
the data.  When using almost 500 experiments in a 1-2 year mechanism
testing program, a detailed analysis of the type of multi-parameter chamber
effects model which will result in the very best possible fit for each
individual experiment is obviously not practical.  Even if it were, at best
the resulting speculative model would give the illusion that we understand
the details of processes which in fact are poorly characterized, and at
worst it may (if there were enough adjustable parameters) have the effect
of canceling out errors in the chemical mechanism in the simulations of the
chamber experiments.

    o Only runs with needed data missing were rejected for use for model
testing.  The types of data considered to be essential were determined
prior to carrying out the test calculations, and included light
characterization data in outdoor chamber experiments, spectral distribution
data in SAPRC Evacuable Chamber (EC) runs, and initial concentrations of
all reactive species in the experiment.  No run was rejected from the model
evaluation statistics based on poor fits alone.

    o Although we attempted, as far as possible, to base the model on
a-priori estimates of unknown mechanistic and chamber dependent parameters,
in some cases adjustment of these parameters in order to attain acceptable
fits  could not be avoided.  However, such adjustment was only done on a
global basis, and using only the most appropriate type of runs for a
particular parameter.  In the case of chamber effects parameters, these
were adjusted based on fits to characterization runs or types of runs which
are the most sensitive to the values of these parameters.  In the case of
mechanistic parameters, adjustment was done based on simulations of single
component - NO  - air runs only.  No adjustment of parameters was done to
improve the fits of the model to results of experiments containing mixtures
of organics.

Summary of Parameters Aidusted in the Course of Model Testing

    The mechanistic and chamber dependent parameters which were adjusted
during the course of the development and testing of this model, and the
types of experiments used as a basis for this adjustment, are as follows:

    Mechanistic Parameters

    o The yields and photolysis rates for the species used in the model to
represent the uncharacterized ring-opened products formed in the reactions
of the aromatic hydrocarbons were adjusted based on the model simulations
of selected the benzene, toluene, m-xylene, and the 1,3,5-trimethylbenzene
- NO  - air runs carried out in the SAPRC indoor chambers.  Runs with these

compounds in outdoor chambers were not used because of greater
uncertainties in characterizing the conditions of such runs,  and the much
greater computational requirements which would be involved in using such
runs in our nonlinear optimization program.   The results obtained in the
optimizations based on the indoor chamber experiments were evaluated by
comparing the adjusted mechanism against the results of the UNC toluene and
o-xylene - NO  - air, and were found to be generally satisfactory in that
they did not indicate large systematic discrepancies in the simulations of
the outdoor chamber runs.

    o The radical yields in the ozone + propene and the ozone + isobutene
reactions were adjusted based on fits to propene or isobutene - NO  - air
experiments carried out in the SAPRC Indoor Teflon Chamber (ITC).   Despite
many studies of ozone - alkene reactions, the available data are still
inadequate to determine unambiguously this essential aspect of the
mechanisms for alkenes other than ethene.  The ITC was employed in
adjusting these yields because, of the two indoor chambers, it has the
lower chamber radical source, making such runs more sensitive to radical
input from homogeneous reactions.   The radical yields in the ozone +
propene reaction had to be adjusted downward from our initial estimate, but
it is not greatly different than those used in the ALW mechanism.   The
adjustments of the ozone + isobutene mechanism, which is totally unknown,
is based on simulations of a single experiment, and thus the value obtained
must be considered to be highly uncertain.

    o The photodecomposition quantum yield in the photoysis of methyl ethyl
ketone was adjusted, based on simulations of a very limited number of UNC
chamber experiments.  Previous models have assumed a unit quantum yield,
but that results in significant overprediction of the reactivity observed
in these runs.  However, in view of the limited number of experiments
employed, and the uncertanities in light characterization in outdoor
chamber experiments, the results must be considered to be highly uncertain.

    o Based on analogy with reactions in the n-butane photoxidation system,
it was initially estimated that the OH + butene reactions should result in
approximately 10% organic nitrate yields.  Assuming this resulted in
consistent overpredictions of reactivity in simulations of indoor chamber
experiments,  so this nitrate formation was removed from the model.
However, it should be pointed out that this is based on a limited number of
runs which were not well fit in any case, and the model significantly
overpredicted the reactivity in the few UNC 1-butene experiments,
regardless of which option was used.

    o The photooxidation mechanism of alkanes with chains of four or more
carbons involve the formation of OH-substituted peroxy radicals, which may
react with NO to form alkyl nitrates.  The nitrate yields in the reactions
of the OH-substituted peroxy radicals are not known, though they are known
in the case of unsubstituted peroxy radicals formed in the initial OH +
alkane reactions.  If it is assumed that the nitrate yields in the
reactions of OH-substituted radicals are the same as in unsubstituted
radicals, as has generally been the case in previous models for the higher
alkanes, we found that the model had a consistent tendency to overpredict

the reactivity in n-hexane, n-heptane, n-octane, and n-nonane - NO  - air
runs.  Therefore, the model was modified to assume that the reactions of NO
with OH-substituted peroxy radicals does not involve the formation of
nitrates.  Since this is a radical termination process, the resulting model
predicts greater reactivities for these alkanes, which are generally more
in line with the results of the chamber experiments, at least on the
average.  However, as seen later, the model is the most variable in the
simulations of the alkane - NO  - air experiments, which is attributable to
the sensitivity of these runs £o the variabilities in the chamber radical
source.  Therefore, any conclusions based on fits of model simulations to
results of alkane - NO  - air experiments are subject to a relatively wide
degree of uncertainty,Xand the possibility that the initially estimated
model might be more correct cannot be totally ruled out.
    Chamber Dependent Parameters

    The chamber-dependent parameters which were adjusted consisted of the
NO  offgassing rates, which were adjusted for each chamberto fit the
results of the acetaldehyde - air irradiations,  and the background NO
conversion rate (represented by OH -> HO. in the model), which was adjusted
for based on fits to the pure air irradiations.   The estimate that initial
HONO levels are probably minor in Teflon film chambers was also made based
on results of preliminary modeling studies.  The other chamber-dependent
perameters were obtained or estimated based on analysis of results of
characterization experiments done using these or similar chambers,
characterization data taken during the experimental runs which were
modeled.  The derivation of the various chamber-dependent parameters used
in this model will be discussed below.
Chambers Whose Data Were Used for Model Testing

    As indicated previously, data from four chambers, operated by two
different research groups, were used for model testing.  All four chambers
have been operated for a number of years in experimental programs designed
to obtain data for this purpose.  The major characteristics and
distinguishing features of these chambers are as follows:

    The SAPRC Evacuable Chamber (EC") is a rigid-walled 5800-liter indoor
chamber.  It is routinely evacuated between experiments.  This chamber is
also thermostated, and experiments can be carried out at varying
temperatures, though most of the EC experiments modeled in this study were
carried out at "303 K.  It uses a. Pyrex-filtered Xenon-arc "solar
simulator" as a light source.  Its interior walls consist of FEP Teflon
coated aluminum, and it has quartz windows on either end, though on the end
opposite to the light source aluminum external reflectors are employed to
enhance the light intensity inside the chamber.  An air purification system
supplies purified air for the experiments employing this and the other
SAPRC chambers.  The purified air used in EC experiments is usually
humidified to "50% RH.  The major characteristics of this chamber, and

experimental procedures employed for most of the EC experiments modeled in
this study, are described in a previous EPA report (Pitts et al. 1979.)

    The SAPRC Indoor Teflon Chamber (ITC) consists of a replacable
"5800-liter flexable bag constructed of 2-mil thick Teflon film which is
held in an aluminum frame.  The bag is replaced periodically, typically
every few months,  or be.tween programs involving different types of
experiments.  The light source consists of banks of blacklights on either
side of frame holding the Teflon bag.  Although not thermostated, the
temperature in the chamber is held relatively constant at "300 K by means
of a cooling system which is employed when the lights are turned on.  The
same air purification system as employed in the EC is used to supply air
for this chamber,  which is also usually humidified to "50% RH.  The major
features of this chamber, and the the experimental procedures employed, are
discussed in various recent SAPRC reports (Carter et al., 1984, 1985,
    The SAPRC Outdoor Teflon Chamber (OTC) consists of a replacable "50,000
liter flexable bag also constructed of 2-mil thick film which held on a
frame with a network of ropes.  The reaction bag is replaced after
approximately 1-2 months of use.  Natural sunlight is used as the light
source.  The chamber has no temperature control.  The chamber is covered by
an opaque tarp between experiments and when the reactants are being
injected, and the tarp is removed to begin the irradiation.  A typical
experiment begins at "0900 PST and ends at "1500 PST, and the bag is
usually covered overnight in multi-day runs.  The bag can be optionally
divided in half, to allow simultaneous irradiation of two mixtures, though
it can also be operated in the undivided mode.  Dry purified air is used in
most experiments with this chamber.  This chamber, and experimental
procedures employed, are discussed in several SAPRC reports (Carter et. al,
1985, 1986b).

    The UNC Outdoor Chamber is constructed of 5-mil thick FEP Teflon film
held on a wooden framework.  This chamber has two sides, each "150,000
liters in volume, allowing two different mixtures to be irradiated
simultaneously.  This is the largest of the chambers whose data are used in
this study.  Unlike the SAPRC Teflon chambers, the Teflon film walls are
rarely replaced.  Natural sunlight is used as a light source.  Reactants
are usually injected before sunrise on the morning before the experiment,
and the irradiation begins when the sun rises.  The chamber is located in a
rural area, and relative clean, unpurified ambient air is used in this
chamber.  In some, but not all, experiments the humidiy inside the chamber
is reduced using dehumidifiers to reduce condensation of moisture on the
wall.  This chamber, and the operating procedures employed, are described
in various UNC reports (e.g., Jeffries et al. 1982, Sexton et al,  1987).
Derivation of the Chamber Characterization Model

    The testing of a chemical mechanism with chamber data requires
appropriate specification of chamber-dependent parameters to represent


effects which may vary from chamber to chamber, and depend on the
conditions of specific experiments.  The types of effects and chamber
dependent input which are represented in the chamber characterization model
used when our mechanism was tested are as follows:

     o Intensity, spectral distribution, and (for outdoor runs) time
       variation of the light source

     o The magnitude of chamber radical source, and its dependences
       on light intensity and N0?

     o Initial levels of nitrous acid (HONO),  if any

     o Ozone dark decay rates

     o NO  offgassing rates
     o Rates of heterogeneous hydrolyses of N«0- and of N0_

     o Excess NO oxidation rates, caused (presumably) by background
       or contaminant reactive organics.

In this portion of this presentation, I will summarize how these various
effects are represented in our chamber characterization model, and indicate
how the various parameters employed were derived.  For a more detailed
discussion of this, interested persons should consult our Phase I report
(Carter et al. 1986a).

    Light Source Characterization:  Indoor Chambers

    Characterization of the light source for modeling indoor chamber
experiments requires a knowledge of its intensity and its relative spectral
distribution.  Runs without such information were not used for model
testing.  In both indoor chambers (the SAPRC EC and the SAPRC ITC),  the
light intensity is obtained from results of NO,, actinometry experiments,
which are carried out periodically.  In the SAPRC EC, the spectral
distribution is measured during the course of most runs.  For this testing
program, the runs were divided into groups based on when the runs were
carried out, the lamp employed, and similarities in measured spectral
distributions, and the average spectral distribution for each group was
used for all runs in the group.  The spectral distribution in the EC varies
over time (with the most recent runs having significantly less UV
intensity), and EC runs carried out before the spectral distribution was
routinely monitored were not used for model testing.  In the case of the
SAPRC ITC, the spectral distribution of the blacklights was measured "3
times in separate experiments, and does not appear to change significantly
with time.  One spectral distribution was used in modeling all ITC runs.

    The spectral distribution of the ITC, and representative spectral
distributions for the EC are shown on Figure 1.  For comparison, the z-40
solar spectrum is also shown.  The degradation of the EC spectral
distribution over time, and the relatively low intensity in the ITC at

10      380      410
 WAVELENGTH  (Microns)
Figure 1.  Representative Spectral Distributions for Various Light
          Sources, Normalized to Give the Same NO^ Photolysis Rates.
          1 - SAPRC solar simulator output around the time of run
          2 - SAPRC solar simulator output around the time of run
          3 - SAPRC ITC blacklight output
          4 - Tropospheric,  ground level solar spectrum for zenith
              angle - 40  .

wavelengths above "390 nm (the wavelength region responsible for alpha
dicarbonyl and NC>3 photolysis), is apparent from this figure.

    I might note that although the ITC spectral distribution at the higher
wavelength region is not particularly representative of sunlight, these
differences have provided us useful information regarding the wavelength
regions which affect the photolyis of the aromatic ring opening products.
Aromatic runs carried out in the ITC are much more reactive than predicted
by models which assume that the photoreactive products are alpha
dicarbonyls (as assumed in most previous models) or compounds which
photolyze in similar wavelength regions, indicating that these unknown
products must photolyze at lower wavelengths than do the alpha dicarbonyls.

    Light Source Characterization:  Outdoor Chambers

    Characterization of the time-varying light intensity and spectral
distributions of outdoor chamber runs is much more difficult, and subject
to much more uncertainties than is the case for indoor chamber experiments.
We were not able to carry out as comprehensive an analysis of this in the
limited time frame of this program as was really required, nor did we have
the benefit of the recent work Harvey Jeffries has done in this area.
Therefore, I do not think our model for light characterization in outdoor
chambers should be taken as the last word on this subject, and, at least
with regard to light characterization for the UNC outdoor chamber, is
probably superceded in some respects by Harvey's analysis.  However, in
evaluating the performance of our model in simulating outdoor chamber runs,
and understanding the uncertainties involved, it is important to understand
how we derived our light characterization model when simulating these runs.

    The photolysis rates used when modeling outdoor chamber runs are
obtained by multiplying together the following factors:

     o The theoretical photolysis rate for the reaction, calculated
       using the absorption and quantum yields assigned for it
       in the homogeneous chemical mechanism, calculated using
       Peterson's (1976) "Best Estimate" Actinic fluxes.  This
       is a function of (a) the clock time;  (b) a clock correction,
       discussed below; (c) the latitude of the chamber; and (d) the
       date of the experiment.

     o The ratio of the experimental UV-radiometer measurements to
       calculated UV-radiometer values which correspond to Peterson's
       (1976) theoretical light intensity and spectral distribution.
       The calculated UV values were based on the spectral response
       of the radiometers ysed and on emperically adjusted direct ys
       scattered radiation factors.  This is a function of (a) the UV
       data from the run; and (b) all the factors listed above, which
       affect theoretically calculated photolysis rates.

     o A constant factor which relates the calculated z-0 UV radiometer
       readings to the theoretically calculated z-0 NCL photolysis
       rates.   The value used, which is 152.9 milliwatt cm"   min,


       is derived from published results of simultaneous UV and N0_
       photolysis rate measurements.  The net effect of this factor
       is to relate the NO,, photolysis rates calculated for the
       outdoor chamber runs to those experimentally measured in the
       atomsphere, by means of UV data.

     o A correction factor for errors in calibration of the UV
       radiometer, where applicable.  This is a function of (a) the
       chamber employed and (b) the date of the experiment.

     o A correction factor for the differences in light intensity
       inside and outside the chamber.  Although there may be
       (and, for the UNC chamber, probably are) wavelength and
       zenith angle dependences for this factor, these were ignored
       in this model, since at the time it was developed, no reliable
       information concerning this was available.  This is then a
       function only of the chamber employed in this model.

    The clock correction indicated above is due to the fact that the clocks
in neither chamber are not always totally accurate.  This error is
manifested by the peaks in the UV and TSR readings on clear days being at
different times than theoretically calculated.  These corrections which, as
expected, tended to be constant for runs done around the same time, were
obtained by fitting the shapes of the observed and calculated UV and TSR
curves.  This correction also takes into account corrections for longtitude

    An analysis of the UV data from the UNC chamber runs carried out over
the years indicate a need for a UV correction factor for certain years.  A
measure of the overall light intensity in a given run, as indicated by the
run's UV data, is the extrapolated z-o clear sky UV intensity, which is
obtained by curve fitting the experimental clear-sky UV data to calculated
UV values, and then extrapolating the calculted values to z-0.  This
factors effects of zenith angle on light intensity.  A plot of the
calculated z-0 UV factors against the date of the run for the UNC and OTC
chamber runs on our data set is given in Figure 2.  These factors can be
seen to vary, and tend to be higher later in the year, but for the UNC
values for 1982 and later runs and for the 1983-1984 OTC runs, they are
reasonably consistent with each other, and on the average are near the
theoretically expected value of 70.2 milliwatt cm" .  However, the values
for previous years in the UNC chamber tend to be anamolously low.

    These results suggest probable systematic calibration errors in the UNC
UV instrument during this period.  This is confirmed by Harvey Jeffries
more recent and comprehensive analysis of the UNC UV data, and the history
of the UV instrument employed.  The correction factors we employed in our
model simulations of pre-1982 UNC chamber experiments were estimated by
ratioing the averages of the June-September extrapolated z-0 UNC UV vaules
to those from the 1982 - 1984 runs.  These are: 1.30 for 1978; 1.31 for
1979; 2.06 for 1980; and 1.24 for 1981.  These factors are reasonably
consistent with the corrections factor recently recommended by Jeffries.







 SB   ••

 40   ••

 30   ••

 20   ••

 10   ••
             UK uv FRCTw

                  A  1971
                  X  1979
                  »  1980
                  0  1961
                  +  1982
                  D  1983

        ore uv FACTOR
           *  1983-84
                             0  „
                                         DAY OF YEAR
Figure 2.   Plots of Extrapolated Z-0 UV Radiometer Readings  (in milliwatts
            cm"  ) Against Day of the Year for UNC  and SAPRC Outdoor
            Chamber Runs Modeled in This Study.

However, Jeffries estimates are based on a more comprehensive analysis and
should be used in any future modeling program.

    The problem of relating the light intensity inside the UNC chamber to
the intensity outside (which is measured by the UV radiometer) probably
represents a greater uncertainty in the light characterization for that
chamber.  Jeffries' recent analysis indicate that the walls of this chamber
attenuate light in a wavelength-dependent manner, and that the reflective
floor enhances light in a manner which probably depends on zenith angle.
However, we did not have the benefit of this analysis at the time we had to
develop our UNC light characterization model, and essentially no useable
data were available concerning these effects.  [The N0~ photolysis rate
measurements made inside the UNC chamber by Saeger (1977) are too scattered
to be of any use in this regard.]  For lack of better data, and in line
with our philosophy that "simpler is better" in representing what one
doesn't understand, we assumed that the opposing effects of light
attenuation by the walls and the enhancement by the reflective floor
exactly cancel, and thus no correction factor was used.  However, future
modeling of UNC chamber runs should re-evaluate this, based on the results
of Jeffries' recent evaluation, and any new data that might be obtained in
this regard.

    The uncertainties in estimating the inside vs outside light intensities
for the SAPRC OTC appears to us to be somewhat less, because (a) the SAPRC
OTC does not have a reflective floor (the chamber is over a green
indoor-outdoor carpet) and (b) the Teflon walls are changed periodically,
which should prevent light-absorbing contaminants from building up.  Thus
we assume that the only effect that needs to be taken into account is the
attenuation of the by the walls, and that this attenuation does not depend
on wavelength.  A correction factor of 0.84 was estimated based on the fact
that maximum NO. actinometry values obtained from measurements made
underneath the chamber (with the light passing in and then out of the
chamber) are "70% of the theoretically expected values.  Assuming that the
light reaching the actinometer is suppressed by the same factor when it
goes in as when it goes out, this corresponds to the assumed correction
factor of 0.84, the square root of 70%.  Zenith angle effects, if any, are
ignored.  These estimates are obviously also subject to uncertainty, and
may need to be re-evaluated.

    At this point, I might note that of all the runs used for model
testing, the UNC formaldehyde irradiations are probably the most affected
by the uncertainties in our outside chamber light characterization model.
This is because these runs are driven by formaldehyde photolysis, which is
affected more uncertain lower wavelength radiation, and because in the UNC
experiments, the run is initiated by the pholysis of formaldehyde at very
high zenith angles, when the uncertainties in the light characterization
are the greatest.  These uncertainties are less in the SAPRC OTC runs
because the irradiations begin later in the day when the sun is higher, and
the uncertainties in light characterization are less.  Indeed, the results
of our model simulations of UNC formaldehyde - NO  - air irradiations
suggest, though do not prove, that there may be problems with our UNC light
characterization model.  Concentration-time plots for selected species in


                   HCHO - NO, - AIR RUN     AUB273B
                                           *  EXPERIMENTAL
         1.2     POIHNLOCHTfle
         1.1 •
         •.§ .
                              HOURS AFTER MIDNIGHT
Figure  3.  Experimental  and Calculated Concentration - Time Plots for
          Selected Species in UNC Formaldehyde - NO  - Air Chamber Run
          AU0279B                                X

                     HCHO -  NOK - AIR RUN
         §.72,.    0*,  MO, H03
                               HOURS  AFTER MIDNIGHT
Figure 4.   Experimental and Calculated Concentration  - Time Plots for
           Selected Species in UNC Formaldehyde -  NO  - Air Chamber Run
           JL2381B.                               X

two such runs are shown on Figures 3 and 4.  Figure 3 is more typical of
most of the UNC formaldehyde runs on our data set, and indicates a
significant overprediction of the initial rate of formaldehyde photolysis,
and a corresponding overprediction of the initiation of ozone formation.
The run shown on Figure 4 is somewhat atypical in that this is the only run
in our data set were the model almost fits the formaldehyde decay rate
(though not the initiation time for ozone formation).  However it is
interesting in that it is the same run that Harvey shows as being
perfectically fit by his chamber characterization model.  It is my
understanding that he used the same formaldehyde chemistry as in our model.

    However, the problems we have in fitting the UNC formaldehyde runs do
not prove that our UNC light characterization model is necessarily invalid.
Formaldehyde is a difficult compound to handle experimentally and to
analyze reliably on a routine basis.  As shown by the model fits discussed
later, there may be problems with the UNC formaldehyde analysis technique,
since the UNC formaldehyde data are consistently higher than model
predictions in simulations of chemical systems where the model predictions
are consistent with SAPRC and Unisearch formaldehyde measurements.  Also,
the model has the opposite discrepancy in the simulations of the UNC
acetaldehyde experiments, where it has a tendency to underpredict the
aldehyde consumption rate.  The simulations of the other UNC runs do not
indicate the types of consistent problems we see with the formaldehyde
runs, but these runs in most cases are are less sensitive to the assumed
light characterization model.

    Representation of the Chamber Radical Source

    The existance and importance of chamber-dependent radical sources is
now reasonably well established, so I will not go into a detiled discussion
of this here.  In this context, by "chamber radical source", I mean
radicals from sources other than from HONO formed from the dark N0«
hydrolysis reaction (which will be discussed separately with the other
known heterogeneous reactions) and other than from HONO which may be
initially present in the experiments.  In our model, the chamber radical
source is represented by N02- independent process, which is represented as a
"pure" OH source (since we ao now know what the correct representation is)
                 hv --> OH,

which is assumed to occur in all four of the chambers.  In the SAPRC EC,
there is evidence that the chamber radical source has an NO- dependence
(beyond that expected due to N02 hydrolysis),  and this is represented in
the model by the same type of process as used to represent N00 hydrolysis,
i.e.,                                                        2

           N00 + hv --> 0.5 HONO +0.5 Wall NO
             2                                x

where the radical source is due to the rapid subsequent photolysis of HONO.


There is no clear evidence for such an N0? dependence for the chamber
radical source in the other chambers,  so this was not used for the chambers
other than the EC.

    Other modelers represent the chamber radical source as formaldehyde
offgassing, but this is not assumed in our model.  We find that if this is
assumed, then the rates of NO to N0? conversion in NO  - air
characterization runs are consistently overpredicted.  However, I might
note that in terms of model simulations of most other runs, it does not
really make much difference whether the chamber radical source is
represented as formaldehyde offgassing, or as a "pure" OH source as assumed
in our model.  This is because the excess NO to NO,, conversions caused by
the formaldehyde offgassing radical source model is usually minor compared
to the NO to N0? conversions caused by the reactive organics present.

    The radical source values used in the model simulations whose results
are presented here are shown on Table 1.  That table also indicates how
these values were derived.  In all cases, the radical source was assumed to
be proportional to the NO. photolysis rate, and thus the radical sources
shown on that table are given as ratios of radical input rates to the N0?
photolysis rate.
Table 1.  Radical Source Values Used for Model Testing, and a Summary
          of Their Derivation.
Chamber   RS/kl (ppb)
  EC      0.6 (N02-0.2)
 ITC      0.15-0.6
         (0.3 typical)

 OTC      0.2  - 0.25
Derived from radical source vs N09 regression
of Carter et al (1982), derived from tracer -
NO  - air runs at 50% RH and 303 K.  The value
shown is for a typical NO- value of 0.2 ppm.

Derived from averages of radical input rates
obtained from tracer - nox - air runs carried
out in the same reaction bag.  Varied from bag
to bag.  No NO,- dependence could be determined
and was ignored.

Original RS/kl estimate of 0.3 based on average
values for ITC and OTC (with similar surface).
Not inconsistent with results of UNC n-butane

     Initial Nitrous Acid

     If nitrous acid is  initially present  in chamber  runs,  its photolysis
may  provide a non-negligible  source of radicals.  Initial  nitrous  acid,
which is not measured experimentally  in any of  the chamber runs  in our  data
set  (with  the possible  exception of a few EC NO  - air  irradiations)  has
been used  in previous modeling  studies as an adjustable parameter  whose
value is adjusted  from  run-to-run  to  optimize the fits.  In this model,
such run-to-run adjustment is not  done, but instead  the initial HONO  is
assigned based on  the chamber employed.   In the case of the SAPRC  EC, HONO
has  been measured  in a  number of tracer - NO  - air  experiments  (Carter et
al.  1982,  Pitts et al,  1983), and  the results suggest that under the
conditions of the  EC runs modeled  here, the initial  HONO is approximately
7% of the  initial  N02>  This  is assumed in this model when simulating EC

     For the runs in the other three chambers, we assumed in this study  that
the  initial nitrous acid is negligible.   This is based  on  the fact that
studies at SAPRC (Pitts et al,  1984)  indicate that HONO formation  from  NO-
is much slower in  reactors made of Teflon film, as is the  case with the
ITC, OTC,  and UNC  chambers, than it is in the SAPRC  EC, and thus lower
initial HONO levels are expected.  Preliminary  model simulations of ITC and
OTC  runs indicate  no need to  assume initial HONO, nor did  we see any
evidence that we needed to assume  this in simulations of experiments  in the
UNC  chamber.  In the case of  the UNC  chamber, particularly in runs where
humidity is not well controled  and water  condensation occurrs on the  walls,
the  possibility that initial  HONO  (or HONO offgased  at  various times  in the
middle of  the experiment) is  playing  a role certainly cannot be ruled out.
However, the inclusion  of such  effects in our model  simulations requires a
detailed run-by-run analysis  which was not practical given the number of
runs which had to  be modeled, and  the amount of time available.

     NO  Offgassing Rates

     The need to include some  type  of  representation  of  NO   offgassing in
chamber effects models  is evidenced by the fact that ozone,  and in some
cases PAN, is formed in environmental chamber runs where no  NO  is
injected.  This is was  represented in our chamber effects  model by the
following  1-parameter process

                    hv  —> N02

where K^ is a zero order rate constant which is assumed to be proportional
to the N02 photolysis rate.   Based on fits to results of acetaldehyde - air
irradiations,  the values of the proportionality constant,  k^/k ,  used in
our simulations were 0.5 ppb for the  SAPRC EC;  0.15 ppb for  the ITC; 0.1
ppb for the OTC;  and 0.3 ppb for the UNC chamber.

    The use of this simple model is preferred over over more complex
representations because it requires only one adjustible parameter,  and
because we have presently no way of knowing whether a more complex


representation is necessarily any more accurate.  However, it should be
noted that this simple model does not work very well in simulating the
acetaldehyde - air irradiations in the outdoor chambers, since in general a
different value of k^/k- fits the ozone yields than fits the PAN data.  An
intermediate value between that that best fit PAN,  and that that fit ozone,
was used in this model.

    It should also be noted that the assumption that the NO  is offgased in
the form of NO- is arbitrary, and the possibility that it is being emitted
as NO, or even as HONO, cannot be ruled out.  Indeed, this effect may well
be related to the chamber radical source, as recently suggested-jby the SAI
modelers.  The similarity in the values of the kng/k- and the k /k..
parameters independently derived for the various chambers suggests that
such a connection might exist.  However, no connection between NO
offgassing and the chamber radical source was assumed in this model.

    Rates of Heterogeneous Hydrolysis Reactions

    The heterogeneous hydrolyses of NO- and N^O,. are also represented in
our chamber effects model.   The former process contributes to chamber
dependent radical input, while the latter amounts to a NO  sink which may
be important in affecting maximum ozone yields.  The rates of both of these
processes have been measured in the SAPRC evacuable chamber and in a "4300
liter indoor chamber constructed of the same 2-mil Teflon film as the SAPRC
ITC and OTC.  The rates observed in the "4300-liter Teflon chamber were
used to estimate their rates in the ITC, OTC and the UNC chamber based on
the assumption that these rates are approximately proportional to the
surface/volume ratio.

    The heterogeneous hydrolysis of NO- is represented in out chamber model
as the following overall process:

                NO, --> a HONO + (1-a) Wall NO  ,
                  £                           X.

where a is the HONO yield,  which is based on the experimental data of Pitts
et al (1984) for the EC and "4300 liter Teflon chamber.  The HONO yield
parameter for the ITC, OTC, and the UNC chamber was assumed to be the same
as that observed in the "4300 liter Teflon chamber.  The data of Pitts et
al (1984) indicate that this process is first order in N0_, and the
representation used in this model is consistent with this.  The values  of
these parameters used in the model for the different chambers are
summarized on Table 2.

Table 2.  Parameters Used in the Representation of Heterogeneous
          Hydrolysis in the Model
Chamber    Rate Constant   HONO
              (min  )      Yield

            2.8 x 10
            0.5 x 10
1.4 x
0.9 x
0.5      Data of Pitts et al (1984)

0.2      Measured by Pitts et al (1984) for a
0.2      4300-liter Teflon film chamber.  K
0.2      assumed to be proportional to the
        surface/volume ratio
    Plots of the N.O,. hydrolysis rates observed in the EC and the "4300
liter Teflon chamber against water concentration tend to be linear with
non-zero intercepts (Tuazon et al. 1983).  Thus N^CL hydrolysis in this
model is represented by two overall processes:
                N.Cv —> Wall NO
                 £» 3            X

          N.O, + H00 --> Wall NO
           252             x
The values of the k  and
Table 3.           a
                            parameters used in our model are summarized on
    It should be noted that the representation used for the N^O,. hydrolysis
shown on Table 4 is based on the assumption that the hydrolysis of N-0S to
gas phase HNO,, observed by Tuazon et al. (1983) is a homogeneous, and not
a chamber -dependent, process.  This is in turn based on the assumption the
heterogeneous hydrolysis would form HNO, on the walls, and once HNO. is on
the walls, it stays there.  Thus the rate constants shown on Table 4 are
derived from the difference between the total N_0,. loss rates observed by
Tuazon et al (1983) and the observed formation rates of gas phase HNO..
(The N.05 hydrolysis process forming gas phase HNO- is included in our
homogeneous inorganic transformation model, and is thus not part of the
chamber effects model.)  However, more recent data obtained at SAPRC
indicate that this assumption that formation of gas phase HNO- necessarily
indicates a homogeneous reaction may be erroneous , and that tnis aspect of
the model may need to be re-evaluated (Atkinson et al. 1986).
    Ozone Wall Loss

    Finally, the model must take to account the fact that ozone is lost due
to dark decay on the walls.  Fortuantely, ozone dark decay rates have been

Table 3.  Kinetic Parameters Used in the Representation of Heterogeneous
          N20  Hydrolysis in the Model
k .
(min )
< -*> • -^
(ppm mm )

4.7 X 10
     7.2 X 10
2.5 X 10
1.6 X 10
0.9 x 10
0.5 X 10
0.3 X 10
0.2 x 10
Data of Tuazon et al (1983)

Measured by Tuazon et al for a
4300-liter Teflon film chamber.
Rates assumed to be proportional
to the surface/volume ratio.
measured for all four of the chambers whose data are used in this study.
The results of these experiments indicate that this process is a simple
first order loss process,  and this is how it is represented in this model.
The range of experimental values for the ozone dark decay rates,  and the
values assumed in our model, are given in Table 4.   The experimental ozone
decay rates are somewhat variable, and the dependences of the ozone dark
decay rates on such factors as temperature, humidity, and light intensity
are not known.  However, the ozone decay rates in these chambers,
particularly those constructed of Teflon film, are sufficiently low that
these variabilities and uncertainties are not regarded as major problem
problems in our chamber characterization model.
Summary of Chamber Runs Used for Model Testing

    We will now turn our discussion to the results of the model simulatons
of chamber experiments based on the chemical mechanism and the chamber
effects model we have described.  As indicated previously, a large number
of chamber experiments were used for this purpose, making this, we believe,
the most comprehensive testing of a single model ever carried out as a
single study.  Almost 500 chamber experiments, carried out in the four
chambers, were modeled.  The experiments ranged in complexity from pure air
and NO  - air irradiations and other characterization runs,  to runs with
single organics, runs with various simple and complex mixtures, and finally
runs consisting of irradiations of auto exhaust.  A summary of the types of
environmental chamber runs which which were used for model testing is given
in Table 5.

Table 4.  Ozone Loss Rates Used in the Model, and Ranges of Experimental
          Values From Which They Were Derived
Loss Rate..Used
    (min  )
Range of Experimental Values
        (min  )






- 1.
- 2.
- 2.
x 10
x 10
_ ^£

* 10-4
x 10 *
x 10
(Aug 5-6
(Apr 15-

, 1979)
16, 1981)
Table 5.  Summary of Environmental Chamber Runs Used for Model Testing
Auto Exhaust

         Charger and Volare

         Propene and Mixtures

                 488 Total
                                                    EC   ITC   OTC   UNC
Single Organic - NO

Known Mixtures

n- Butane
C_+ alkanes
Other Aromatics
2 to 5 Component
6+ Component






                                                   117   102



    The specific UNC chamber runs which were modeled were chosen by Marcia
Dodge, our EPA contract monitor, and Harvey Jeffries, the leader of the UNC
chamber group, based on their evaluation of what is a comprehensive set of
appropriate UNC runs to use in this program.   Other than the lack of the
recently-completed UNC methanol substitution runs, whose data were not
ready for us in time for inclusion in this study, these runs are, I
believe, a relatively comprehensive and representative set of the best of
the UNC chamber experiments whose data have been processed.

    The specific SAPRC chamber runs modeled consisted of almost all of the
SAPRC runs which were carried out for the purpose of model testing and
which are considered to be sufficiently well characterized for this
purpose.  The main exception are runs which contain compounds which are not
represented in this model,  such as biogenic organics, naphthalenes,  long
chain alkenes, and several heteroatom-containing organics.  The set of
SAPRC runs also includes results of our recently completed methanol
substitution ITC and OTC experiments (Carter et al. 1986b), and several
other recent ITC runs. These were not included in the set of runs discussed
in our Phase I report (Carter et al. 1986a).
Summary of Results of Model Testing

    The performance of this model in simulating the results of these
experiments is summarized on distribution plots given in Figures 5-12 for
the various groups of runs (other than the characterization runs).   These
figures show the distributions of the discrepancies in the model
simulations of various types of experimental observations, where perfect
fits are indicated by zero discrepancies (points in the middle of the
distribution plots), underprediction by the model indicated by negative
discrepancies (on the tops of the plots),  and overprediction by positive
discrepancies, on the bottom of the plots.  (For such a large number of
experiments, showing distribution plots such as these is a much more
practical method of summarizing model performance than showing fits in
concentration-time plots for individual runs.   Other ways of showing the
performance of the model in simulating large numbers of runs given in
figures included with Fred Lurmann's presentation.)

    For all types of runs,  distribution plots for fits to the maximum ozone
yield, and to a measure of the average rate of ozone formation and NO
oxidation, are given.  The latter quantity is defined as the one half of
the maximum change in ([0_] - [NO]), divided by the time required to
achieve that change, and is used to measure the "timeing" in the
experiment.  In addition, distributions of fits to yields of PAN,
formaldehyde, and in some cases other aldehydes are shown in runs where
these are formed and measured, and distributions of fits to reactant
organic half lives are shown for runs containing measured organics which
react sufficiently rapidly for its reaction rate to be a useful measure of
model performance.

    I will not give a detailed discussion of the performance of the model
in simulating all of these types of runs.  Such a discussion is given in
our Phase I report (Carter et al. 1986a), and it can be seen from Figures 6
- 14, and from some of the plots given in Fred Lurmann's presentation, how
well the model is performing in simulating the various types of runs.
However, the following points regarding  the performance of the model in
simulating the various types 'of runs are noted:

    o The fits of the model to the propene runs (Figure 6) are interesting
since there are a large number of such runs carried out, and, except for
the ozone - propene reaction, the chemistry is believed to be reasonably
well characterized.  It can that although there are are runs which are not
well fit by the model, and the model tends to overpredict ozone in the ITC
runs and underpredict it in the OTC runs, in terms of fits to ozone, NO
oxidation/ozone formation rates, and propene half lives, the model does
reasonably well on the average.  This model was developed primarily on
results of our simulations of SAPRC experiments, and was not adjusted in
any way to fit the UNC runs (which I had never modeled prior to this
program), and thus the fact that the fits to the UNC runs is of similar
quality as fits to the SAPRC EC runs is  encouraging.  The quality of the
fits to the UNC runs appeared to have no relationship to the "light
quality" in those runs; some runs on perfectly clear days were not fit
well, and vice-versa.

    o The fits to PAN in the propene runs is of much lower quality in the
fits to ozone and the "timeing" parameters.  A very wide distribution in
the quality if fits is obtained for all  four chambers, with the model
having a consistent tendency to overpredict PAN yields in the outdoor
chambers, though not in the indoor chamber runs.  Similar wide
distributions are obtained in simulations of other types of runs, such as
the complex mixture runs shown in Figure 12, though in that case, there is
no consistent tendency for the model to  underpredict or overpredict PAN in
the outdoor chamber runs; it just has a very wide distribution in the fits,
with PAN yields in many runs being fit very poorly.

    o The quality of the fits to the formaldehyde yields is the only area
where there is a consistent difference between the UNC chamber and the
SAPRC chambers in terms of model performance.  The fits in the propene and
ethene systems are interesting in this regard, because in both chemical
systems the formaldehyde yield is reasonably well established, and in both
systems relatively high yields of formaldehyde are expected.  The model has
a relatively wide distribution in discrepancies of fits to formaldehyde
yields in the SAPRC propene and ethene runs, but there does not appear to
be any consistent tendency to overpredict or underpredict these yields.  On
the other hand, the model consistently underpredicts the formaldehyde
yields observed in UNC chamber runs when measured by the formaldehyde
monitoring technique usually employed in that chamber.   However,  in a few
UNC propene runs, formaldehyde measurements were made by Unisearch
Associates, inc. using a different technique, and in those cases,  the
results were in excellent agreement with model predictions.   These results
can be explained if the standard UNC formaldehyde monotering technique


tends to give high values when in routine use in chamber runs.   I know that
technique was found to give good results in recent special intercomparison
studies, but this does not mean that there may not be problems when it is
used routinely.  However, modeling by itself certainly cannot prove that
this is necessarily the case.

    o I should also point out the model's consistent tendency to
overpredict acetaldehyde yields in the ITC experiment.  In this case, I
believe it is most likely a problem with the data.  We have had problems
recently with acetaldehyde calibrations, and in this case, I definitely
think the model is more reliable.

    o The ethene runs (Figure 7) are interesting in that in both the EC and
the UNC chamber, there are two groups of runs, carried out around the same
time.  For each chamber, the model tended to overpredict reactivity in all
runs in oue group, and to underpredict the reactivity in runs in the other.
(The model gave good fits to the reactivities in the two ethene runs
carried out in the ITC.)  This illustrates how apparent variability in run
conditions can have an effect on the performance of the model.   It also
illustrates the potential for misleading results if only a limited number
of runs of a given type are used for model testing, and if all the runs
were carried out in the same chamber around the same time.  If one looked
at only one of those five groups of ethene runs (two each in the EC and the
UNC chamber, and the ITC runs), one may conclude that the model is
consistently "fast" or "slow" (depending on which group),  but looking at
all five groups together suggests that there is probably no consistent

    o Of all the types of runs modeled, the alkane - NO  - air runs have
the widest distributions of the fits to ozone yields and NO oxidation and
ozone formation rates.  This can be attributed to the extreme sensitivity
of model simulations of these types of runs to the assumed magnitude of the
chamber-dependent radical source.  Since the actual radical source tends to
vary from run to run, but the model simulations used a single set of
assumptions regarding it for all runs carried out in a given chamber, this
variability results in a variability in the qualities of the fits of the
model predictions to the experimental data.

    o A large number of runs with complex, but known, mixtures were modeled
in this study, and the fits to these runs are summarized on Figure 12.  As
with the simulations of the propene and other types of runs for which there
are relatively large numbers of experiments, there is a distribution of
discrepancies, but there was no large tendency for systematic discrepancy.
The distributions of fits for PAN and formaldehyde yields are much wider
than is the case for NO oxidation and ozone formation rates,  or reactant
half lives.

    o Most of the complex mixture runs carried out in the SAPRC ITC and OTC
are runs which employ a single 8-hydrocarbon surrogate, which was designed
by the modelers at SAI to represent urban emissions.   Many of these runs
were carried out as part of our study of the effects of methanol
substitution, and thus some of these runs had one third of the hydrocarbon


surrogate replaced by methanol or 90% methanol + 10% formaldehyde.  There
is no significant difference in the model performance in simulating the
runs with methanol than in simulating those with only the hydrocarbon
surrogate.  I should also note that the EC seven hydrocarbon surrogate runs
employed a variety of different relative amounts of the reactants, so they
represent more than one single mixture.

    o Many of the ITC and OTC complex mixture runs modeled were multi-day
irradiations.  In general, we found that if the model could satsifactorily
simulate the results on the first day, it could usually also simulate the
results for the subsequent days.  If it performs poorly on the first day,
then there is little chance that the subsequent days can be adequately
simulated.  The distribution plots on Figure 13 show only the fits to the
day 1 results.  An example of how well the model can perform in simulating
a two-day, dual chamber, OTC run. is shown on Figure 14.

    (o The PAN data shown for Side 1 on Figure 14 are also interesting.
The PAN is observed to rapidly increase after t - 6 hours, when the chamber
is covered for the night.  This dark PAN formation on Side 1 is reasonably
well simulated by the model, and is due to the reaction of NO- radicals
with acetaldehyde.  This illustrates that these NO, + aldehyde reactions,
which are generally of negligible importance in the daytime,  can be
important at night, and must be included in multi-day models.)

    o Figure 13 shows that the fits of the model to the ozone yields and
the NO oxidation and ozone formation rates in the UNC auto exhaust runs is
of comparable, or even slightly better, quality as the fits of the model to
the UNC complex known mixtur.e runs.  This is reassuring, since auto exhaust
runs are obviously important for control purposes, and represent the most
complex type of mixture studied.  These fits were obtained with no
adjusting of the mechanism or modifications of the assumptions made
regarding the composition of the exhaust mixtures in order to optimize the
fits.  Fred Lurmann was responsible for the speciation of the auto exhaust
mixture.  He did an analysis of the data given in the UNC report on these
auto exhaust experiments (Jeffries et al. 1985), and the analytical data
obtained during the experiments, and came up with compositions for me to
use in the modeling of each of these runs.  I used the compositions Fred
gave me without modifications, and obtained the fits shown on the figure.

Possible Sources of Poor Fits of Model Prediction to Experiment

    The results of this evaluation indicate that there are many cases where
experimental results are not well fit by the model prediction.  While in in
most cases the discrepancies appeared to be random in nature, there are
cases where apparent systematic discrepancies are observed.   From the point
of view of evaluating a mechanism for use in AQSM's, it is clearly the
systematitic discrepancies which are of the greatest concern.  Possible
sources of systematic discrepancies are as follows:

    o Errors in the homogeneous photochemical reaction mechanism

    o Inappropriate representations of chamber effects or light

    o Systematic measurement errors,  especially of initial reactant
      concentrations or of characterization data

    o analytical interferences

    Determining whether errors in the photochemical reaction mechanism
exist is clearly the primary goal of this testing program, since such
errors would invalidate the AQSM predictions which use this mechanism.
While errors in the chamber characterization model would not necessarily
affect the validity of AQSM predictions,  they could give an indication that
a correct mechanism is incorrect, or, worse, that an incorrect mechanism is
correct.  This is why it is important we make an effort to better
understand the fundamental physical and chemical processes which cause such
chamber dependent effects.

    The possibility that in some cases systematically poor fits of model
precictions to experimental data may be due to problems with the data
rather than the model also cannot be ruled out.   In my experience,  in cases
where the causes of gross systematic discrepancies in model simulations of
a series of experiments have subsequently been determined, it has more
often than not been due to problems with the data.  Indeed, modeling has
frequently been shown to be a useful tool in revealing the existance of
experimental problems.  Probable problems with the UNC formaldehyde data
and known problems with the SAFRC acetaldehyde data are examples of where
the experimental data are more suspect than the model predictions.
However, it is the experimental problems that we don't know about,  rather
than those discussed in this presentation, that are of the greatest
concern.  While environmental chamber experimentalists make every effort at
quality control, such experiments involve a very large number of variables,
and totally comprehensive quality control is a monumental, if not
impossible, task.  Significantly inproved quality control beyond that
currently practiced would require a much greater commitment of funding and
personnel to enviromental chamber experimental groups than is presently the

    Random discrepancies are of concern primarily because they limit the
level of precision to which the model can be tested.   In addition,  if there
are significant sources of random discrepancies that cannot be eliminated,
then a large number of experiments is required for adequate model testing.
The most likely sources of such discrepancies, which I suspect might be
significant in many of the cases observed in this study, include the

    o Random uncertainties or errors in specifying initial reactant

    o Run to run variation in uncontrolled chamber-dependent effects


    o Inadequate characterization or control of run conditions

    o Incomplete or inappropriate chamber effects model

    Improved quality control when carrying out chamber experiments would
reduce some, but not all, of these sources of random discrepancy.  For
example, even if all errors in the analytical and characteriziton data were
somehow eliminated, there may still be uncontrolled variabilities in the
chamber radical source.  This would result in a corresponding variability
of in the fits of model predictions to experimental results, particularly
for chemical systems which are highly sensitive to this effect.  Reducing
this type of source of variability requires understanding what causes and
effects such as the chamber radical source, and then either (a) developing
experimental procedures or chamber design modifications to eliminate this
effect or (more likely) make it consistent and predictable, or (b)
developing a quantitative understanding of all the experimental parameters
which influence the effect, and using this to develop a predictive model
for it.  Either alternative requires a major experimental research program.
Conclusions and Recommendations

    I will now briefly summarize what I believe are some of the conclusions
which can be drawn from this study regarding the performance of this model
and the quality of the available chamber data set, and give several
recommendations concerning future model testing efforts.  These are not the
only conclusions and recommendations which can be derived from the  results
of this study; other conclusions and recommendations, particluarly with
regard to the problems with using this and other mechanisms in AQSM's, will
be discussed by Fred Lurmann in his presentation at this workshop.

    Conclusions Regarding Current Model Performance and Data Quality

    This study showed that a model based on our current understanding of
the atmospheric chemistry and our best estimates concerning chamber effects
is, for most types of experiments, usually able to simulate ozone yields
and formation rates to within "30%.  However, there are many experiments
which are now well simulated by this model, with most of the discrepancies
being apparently random in nature.  However, there were cases where
apparently systematic discrepancies were observed as well,  indicating
possible systematic problems with the mechanism, our chamber
characterization model, or the experimental data.  With the exception of
the simulations of the formaldehyde runs and the formaldehyde data, there
did not appear to be any large difference in the performance of the model
in simulating the UNC or the SAPRC runs.  This is gratifying, since prior
to this program I had no experience with modeling UNC chamber chamber
experiments, and those aspects of the model which were adjusted were
adjusted primarily based on fits of the model to results of SAPRC runs.

    The many instances of wide, apparently random distributions of
qualities of fits of model simulations to current environmental chamber
data means that fits of a model to a single "typical" run, or limited
groups of runs, is of limited significance.  Only if the data set for a
compound or mixture is sufficiently large or varied can we obtain a
meaningful measure of model performance.   This measure is obtained on the
basis of statistics of fits of the model to the data set as a whole, and
not of the results of the simulations of a single experiment.

    The available chamber data set is sufficiently large and varied to give
us a meaningful test of the model for a number of compounds and mixtures.
With regard to single compounds, this is the case for propene, toluene and,
to a lesser extent for several other alkenes and aromatics.  It is not the
case for aldehydes, ketones,  butenes and the higher alkenes,  any of the
alkanes,  and aromatics other than toluene and m-xylene.  In some of these
cases, the inadequicy of the available data set is not due to the lack of
runs, but the fact that the available runs are sensitive to poorly
controlled or characterized variables.  Two examples of this are the alkane
- NO  - air and the formaldehyde - NO  - air runs.  In the case of the
    X                                X
alkane runs, their sensitivity to the variable chamber radical source
results in a wide distribution in the quality of fits of the model to the
data, and thus a much larger number of runs is required for adequate
assesment of model performance than is the case for compounds which are
less sensitive to this.  In the formaldehyde case, runs with this compound
are highly sensitive to uncertainties in light characteristics, and also
the analysis of formaldehyde is subject to considerable uncertainty.  In
this case, it is difficult to asses the extent to which experimental
problems contribute to the qualities of the fits observed, and it is
doubtful that more experiments carried out using the same procedures as
employed previously would significantly improve the situation.

    In the current environmental chamber data set, measurements of ozone,
NO, and the easily measured and rapidly reacting hydrocarbons give the most
reliable indication of model performance.  The utility of other types of
experimental measurements in the current data set in assesing model
performance is much more uncertain.  For example, the current model does
not perform nearly as satisfactorily in predicting PAN yields as it does
for ozone, with much wider ranges of apparently random discrepancies being
observed.  This wide distribution of fits suggests, though it obviously
does not prove, that that there may be problems with the quality of PAN
data in the current data set.  In addition, more confidence in the quality
of aldehyde data obtained during chamber experiments is required before
these can be used as a reliable indication of model performance.

    To some extent, the utility of the existing environmental chamber data
base for testing the mechanism can be inproved by further characterization
of the conditions under which those experiments were carried out.  The
recent work in studying the light characteristics in the UNC chamber, and
our previous work in characterizing the radical source in the SAPRC EC are
two examples of this.  To the extent that further studies could improve the
characterization of the existing chamber data base, these should be

encouraged, since this is probably the most cost effective means to improve
the utility of the data base used to test photochemical models.

    However, in order to obtain a significantly more comprehensive test of
the current chemical mechanisms than was carried out in this study, use of
environmental chamber data with more complete characterization and control
of chamber-dependent conditions than is possible with the present data set
is required.  This requires, I believe, studies aimed at determining the
most appropriate types of experimental techniques and chamber surface
materials to employ, as well as development of improved analytical
techniques.  For example, fundamental studies of the types of surface
reactions which contribute to chamber effects should be considered as part
of this effort.  Finally, the process of carrying out environmental chamber
experiments for use in model testing needs to be given a much greater
commitment of manpower and other resources than previously has been
possible.  This obviously requires a major research program, and may not be
consistent with the current funding levels and priorities of the EPA.
However, this is required if any major improvement in our ability to test
photochemical mechanisms is desired.

    Recommendations for Future Model Testing

    This study illustrates how we believe photochemical reaction mechanisms
should be tested.  In particular, the same mechanism should be tested under
a wide variety of conditions with a consistent set of assumptions without
run to run adjustment.  Model evaluation using "typical" runs, or limited
sets of runs, or "evaluations" with run-to-run adjustment of parameters to
optimize the fits, should no longer be considered to be acceptable for
documenting model performance.  Using a limited number of runs to evaluate
a model might have been appropriate in the past when there were only
limited amounts of data, but now we have a large enough data base of
chamber experiments, and sufficiently low cost computational capabilities,
that this should no longer be necessary.  While there might be some areas
of disagreement with what we assumed concerning certain chamber effects,
and certainly additional or improved methods of evaluating the model
performance could be derived, we believe that the general principles and
philosophy employed when carrying out this study serves as an appropriate
model for future studies.

    It is also important that the chemical mechanism which is tested
against the chamber data be a detailed mechanism based on our best estimate
as to the correct chemistry.  In particular, we do not believe it is
appropriate to use environmental chamber data to test lumped or condensed
mechanisms.  Use of mechanisms which are unnecessarily approximate
introduce additional and unnecessary uncertainties regarding the accuracy
of the model, and the effects of such approximations could be more
unambigously tested by comparison against model calculations using less
approximate models.  If we obtain poor fits in simulations of chamber runs
when using lumped mechanisms, we do not know whether they are due to errors
in the chemistry (or the chamber effects model) or to use of an
inappropriate lumping technique.  Alternatively,  if we obtain good fits,
there is always the possibility that errors in the chemistry are being


cancelled by errors in the lumping technique employed.  The more
appropriate approach is to first test the detailed mechanism against the
environmenal chamber data, and then use model calculations to explore
effects of various lumping or condensation techniques, with the detailed,
tested mechanism serving as the standard.  This is the approach used in
this program.

    However, this study should not be taken as the "last word" in the
evaluation of this mechanism.  The assumptions we made regarding chamber
and light characterization should be evaluated by other modelers and
chamber experimentalists, and possible areas of imprivement and alternative
sets of assumptions recommended.  Then this (and other) model should be
re-tested based on the improved or alternative sets of assumptions.  At a
minimum, the model should be re-tested against the UNC chamber runs based
on the new information we have regarding the light characterization in that
chamber.  It could also be re-evaluated based on alternative sets of
assumptions concerning the chamber radical source, such as those employed
by the modelers at SAI.  Examining how differing assumptions regarding
chamber and light characterization affect the model fits would give us a
more precise idea of which of these amount to the most critical
uncertainties in terms of model testing.

    We recommend that other chemical mechanisms which are currently
considered to be at the state of the art, such as the most detailed version
of the latest "Carbon Bond" mechanism, be tested against these same data
using the same set of assumptions.  Although various "Carbon Bond" models
have at various times been tested against a large number of experiments,  as
far as I am aware none of these mechanisms has been tested in a systematic
manner against such a large and varied data set without run-to-run
adjustment of parameters.  If such tests have been carried out,  they have
not been documented.  The results of such tests would allow us more direct
comparison of the performance of the different models in simulating the
available chamber data than is presently possible.

Atkinson, R.,  A. C. Lloyd, and L. Winges (1982),  Atmos.  Environ., 16.

Atkinson, R.,  E. C. Tuazon, H. Mac Leod, S. M.  Aschmann, and A. M. Winer
       (1986), Geophys. Res. Lett. 13.,  117.

Carter, W. P.  L.,  R. Atkinson, A. M.  Winer and J. N.  Pitts, Jr. (1982),
    Int J. Chem Kinet., 14, 1071.

Carter, W. P.  L.,  A. M. Winer, R. Atkinson, M.  C. Dodd,  W.  D. Long,
       and S.  M. Aschmann (1984), "Atmospheric Photochemical Modeling
       of Turbine Engine Fuels.  Phase  I.   Experimental  Studies,"  Final
       Report to the U. S. Air Force, ESL-TR-84-32,  September.

Carter, W. P. L. and R. Atkinson (1985), J. Atmos.  Chera.,  3, 377.

Carter, W. P. L.,  M. C. Dodd, W. D. Long, and R. Atkinson (1985),
    "Outdoor Chamber Study to Test Multi-Day Effects", EPA-600/3-84-
       115, March.

Carter, W. P. L.,  F. W. Lurmann, R. Atkinson, and A. C. Lloyd (1986a),
       "Development and Testing of a Surrogate Species Chemical Reaction
       Mechanism," EPA-600/3-86-031, August.

Carter, W. P. L.,  W. D. Long, L. N. Parker, and M.  C. Dodd  (1986b),
       "Effects of Methanol Fuel Substitution on Multi-Day Air Pollution
       Episodes,"  Final Report on California Air Resources Board Contract
       No. A3-125-32, April.

Jeffries', J. E. , R. M. Kamens, K. G. Sexton, and A. A. Gerhardt  (1982),
       "Outdoor Smog Chamber Experiments to Test Photochemical
       Mechanisms", EPA-600/3-82-016A, April.

Jeffries, J. E., K. G. Sexton, T. P. Morris, M. Jackson, R. G. Goodman,
       R. M. Kamens, and M. S. Holleman  (1985), "Outdoor Smog Chamber
       Experiments Using Automobile Exhaust", EPA-600/3-85-032.

Lurmann, F. W., A. C. Lloyd, and R. Atkinson (1986), J. Geophys.  Res.
       91, 10915.

Lurmann, F. W., W. P. L. Carter, and L. A. Coyner (1987) "A Surrogate
       Species Chemical Reaction Mechanism for Urban-Scale Air Quality
       Simulation Models.  Volume I - Adaptation of the Mechanism,
       Final Report on Phase II of EPA Contract No. 68-02-4104.

Peterson, J. T. (1976) "Calculated Actinic Fluxes (290 - 700 nm)  for Air
    Pollution Photochemistry Applications", EPA-600/4-76-025, June.

Pitts, J. N., Jr., K. Darnall, W. P. L. Carter, A.  M. Winer, and R.
       Atkinson (1979), "Mechanisms of Photochemical Reactions in Urban
       Air", EPA-600/3-79-110, November.

Pitts, J. N., Jr., R. Atkinson, W. P. L. Carter, A. M. Winer, and E. C.
       Tuazon (1983), "Chemical Consequences of Air Quality Standard
       and Control Implementation Programs", Final Report to the
       California Air Resources Board Contract No.  Al-030-32, April.

Pitts, J. N., Jr., E. Sanhueza, R. Atkinson, W. P.  L. Carter, A.  M.
       Winer, G. W. Harris, and C. N. Plum (1984),  Int. J.  Chem.  Kinet.,
       16, 919.

Saeger, M. (1977), "An Experimental Determination of the Specific
       Photolysis  Rate of Nitrogen Dioxide", Masters Thesis, University
       of North Carolina, Chapel Hill, NC.

Sexton, K. G.,  H. E. Jeffries, J.  R.  Arnold,  T.  L.  Kale,  and R.  M. Kamens
        (1987), "Validation Data for Photochemical Mechanisms",  Report
        on EPA Cooperative Agreement 811673,  January

Tuazon, E. C.,  R. Atkinson, C. N.  Plum,  A.  M. Winer, and J.  N.  Pitts, Jr.
       (1983),  Geophys Res. Lett., 10, 953.

Figure 5.  Summary of Fits for the Aldehyde and Ketone - NO  - Air Runs,

                  EC   ITC/OTC  UNC      EC    UNC      UNC       UNC
(Calc-Expt) (ppm)
(model low)
< -0.33
0.33 - -0.27
0.27 - -0.21
0.21 - -0.15
0.15 - -0.09
0.09 - -0.03
0.03 - 0.03 -
0.03 - 0.09
0.09 - 0.15
0.15 - 0.21
0.21 - 0.27
0.27 - 0.33
> 0.33
(model high)




     Maximum Ozone
Average d([03] - [N0])/dt
(model low)
< -1.00
-1.00 - -0.82
-0.82 - -0.64
-0.64 - -0.45
-0.45 - -0.27
-0.27 - -0.09
-0.09 - 0.09 -
0.09 - 0.27
0.27 - 0.45
0.45 - 0.64
0.64 - 0.82
0.82 - 1.00
> 1.00










(model high) (0 - OTC) (A - Acetone, M - MEK)

Figure 5.  Summary of Fits for the Aldehyde and Ketone - NO  - Air Runs
            EC     UNC
            EC     UNC
(Calc - Expt)/Expt
                       Reactant Half Life
-0.50 -
-0.41 -
-0.32 -
-0.23 -
-0.14 -
-0.05 -
 0.05 -
 0.14 -
 0.23 -
 0.32 -
 0.41 -

Figure 6.  Summary of Fits to Propene - NO  - Air Runs.

(Calc-Expt) (ppm)
(model low)
< -0.33
-0.33 - -0.27
-0.27 - -0.21
-0.21 - -0.15
-0.15 - -0.09
-0.09 - -0.03
-0.03 - 0.03 --
0.03 - 0.09
0.09 - 0.15
0.15 - 0.21
0.21 - 0.27
0.27 - 0.33
> 0.33
(model high)
(model low)
< -1.00
-1.00 - -0.82
-0.82 - -0.64
-0.64 - -0.45
-0.45 - -0.27
-0.27 - -0.09
-0.09 - 0.09 --
0.09 - 0.27
0.27 - 0.45
0.45 - 0.64
0.64 - 0.82
0.82 - 1.00
> 1.00
(model high)

Maximum Ozone

* ***
* *
** **** ******
**** . . ** . **
******* * ***
* * **
* *

Average d([03] - [N0])/dt

* ** * ****
** - * - ** - ********
**** ** * *******
*** ** * **


Figure 6.  Summary of Fits to Propene - NO  - Air Runs (continued)
(Calc - Expt)/Expt
    Maximum PAN
(model low)
< -0.50
-0.50 - -0.41
-0.41 - -0.32
-0.32 - -0.23
-0.23 - -0.14
-0.14 - -0.05
-0.05 - 0.05 --
0.05 - 0.14
0.14 - 0.23
0.23 - 0.32
0.32 - 0.41
0.41 - 0.50
> 0.50
(model high)
(Calc - Expt)/Expt
(model low)
< -0.50
-0.50 - -0.41
-0.41 - -0.32
-0.32 - -0.23
-0.23 - -0.14
-0.14 - -0.05
-0.05 - 0.05 --
0.05 - 0.14
0.14 - 0.23
0.23 - 0.32
0.32 - 0.41
0.41 - 0.50
> 0.50
(model high)
















                            Propene Half Life

Figure 6.  Summary of Fits to Propene - NO  - Air Runs (concluded)
(Calc - Expt)/Expt
Maximum Acetaldehyde
(model low)
< -0.50
-0.50 - -0.41
-0.41 - -0.32
-0.32 - -0.23
-0.23 - -0.14
-0.14 - -0.05
-0.05 - 0.05 --
0.05 - 0.14
0.14 - 0.23
0.23 - 0.32
0.32 - 0.41
0.41 - 0.50
> 0.50


(model high)
(Calc - Expt)/Expt Maximum F
(model low)
< -0.50
-0.50 - -0.41
-0.41 - -0.32
-0.32 - -0.23
-0.23 - -0.14
-0.14 - -0.05
-0.05 - 0.05 --
0.05 - 0.14
0.14 - 0.23
0.23 - 0.32
0.32 - 0.41
0.41 - 0.50
> 0.50






(model high)


           U -

                            HCHO Data

Figure 7.   Summary of Fits for Ethene and Butene - NO  - Air Runs.
Calc-Expt (Calc-Expt)

(ppm) Maximum Ozone /Expt d( [0_ ] - [NO] )/dt
< -0.33
-0.33 - -0.27
-0.27 - -0.21
-0.21 - -0.15
-0.15 - -0.09
-0.09 - -0.03
-0.03 - 0.03 -
0.03 - 0.09
0.09 - 0.15
0.15 - 0.21
0.21 - 0.27
0.27 - 0.33
> 0.33




< -1.00
-1.00 - -0.82
-0.82 - -0.64
** -0.64 - -0.45
-0.45 - -0.27
-0.27 - -0.09
-0.09 - 0.09 -
*x 0.09 - 0.27
* 0.27 - 0.45
* 0.45 - 0.64
* 0.64 - 0.82
0.82 - 1.00
> 1.00

X - X
*•** **

(Calc-Expt) (Calc-Expt)


- ***


/Expt Reactant Half Life /Expt Maximum Formaldehyde
< -0.50
-0.50 - -0.41
-0.41 - -0.32
-0.32 - -0.23
-0.23 - -0.14
-0.14 - -0.05
-0.05 - 0.05 -
0.05 - 0.14
0.14 - 0.23
0.23 - 0.32
0.32 - 0.41
0.41 - 0.50
> 0.50




< -0.50
-0.50 - -0.41
-0.41 - -0.32
-0.32 - -0.23
***x -0.23 - -0.14
-0.14 - -0.05
-0.05 - 0.05 -
* 0.05 - 0.14
0.14 - 0.23
* 0.23 - 0.32
0.32 - 0.41
0.41 - 0.50
> 0.50

**x *
**x - *xx


(* - Ethene, x - Butenes)



Figure 8.  Summary of Fits to the n-Butane - NO  - Air Runs
Calc - Expt Maximum Ozone Calc-Expt Avg. d( [0_] - [NO] )/dt
(model low)
< -0.33
-0.33 - -0.27
-0.27 - -0.21
-0.21 - -0.15
-0.15 - -0.09
-0.09 - -0.03
-0.03 - 0.03 -
0.03 -. 0.09
0.09 - 0.15
0.15 - 0.21
0.21 - 0.27
0.27 - 0.33
> 0.33






* < -1.00
* -1.00 - -0.82
-0.82 - -0.64
* -0.64 - -0.45
* -0.45 - -0.27
-0.27 - -0.09
-0.09 - 0.09 -
* 0.09 - 0.27
0.27 - 0.45
* 0.45 - 0.64
* 0.64 - 0.82
0.82 - 1.00
> 1.00







(model high) (0 - OTC) (0 - OTC)

Figure 9.  Summary of Fits to the C..+ Alkane  - NO   -  Air
                                   J             X
                  23-DMB  ISO-C.   n-Cc    n-C,    n-C-.     n-CQ
                          ISO-C*       5678
                                            n-  Cr
(Calc-Expt) (ppm)
              Maximum Ozone
(model low)
< -0.33
0.33 - -0.27
0.27 - -0.21
0.21 - -0.15
0.15 - -0.09
0.09 - -0.03
0.03 - 0.03 -
0.03 - 0.09
0.09 - 0.15
0.15 - 0.21
0.21 - 0.27
0.27 - 0.33
> 0.33
(model high)






Average d([03] -  [N0])/dt
(model low)
< -1.00
1.00 - -0.82
0.82 - -0.64
0.64 - -0.45
0.45 - -0.27
0.27 - -0.09
0.09 - 0.09 -
0.09 - 0.27
0.27 - 0.45
0.45 - 0.64
0.64 - 0.82
0.82 - 1.00
> 1.00
(model high)




: - EC,




U - UN(









Figure 10,   Summary of Fits to the Aromatic  -
                      NO   - Air Runs,
- Xylenes -
(Calc-Expt) (ppm)
              Maximum Ozone
0.33 -
0.27 -
0.21 -
0.15 -
0.09 -
0.03 -
0.03 -
0.09 -
0.15 -
0.21 -
0.27 -
0.03 -
                                   **  .
                                                ** -
(model low)
< -1.00
1.00 - -0.82
0.82 - -0.64
0.64 - -0.45
0.45 - -0.27
0.27 - -0.09
0.09 - 0.09 -
0.09 - 0.27
0.27 - 0.45
0.45 - 0.64
0.64 - 0.82
0.82 - 1.00
> 1.00
(model high)




Average d([03] -  [N0])/dt
                                                                  ** .

Figure 10.  Summary of Fits to the Aromatic - NO  - Air Runs (concluded)
                Benzene   	 Toluene 	
                 ITC      EC       ITC  UNC
           - Xylenes -
           EC  ITC  UNC
(Calc - Expt)/Expt
Maximum PAN
(model low)
< -0.50
0.50 - -0.41
0.41 - -0.32
0.32 - -0.23
0.23 - -0.14
0.14 - -0.05
0.05 - 0.05 -
0.05 - 0.14
0.14 - 0.23
0.23 - 0.32
0.32 - 0.41
0.41 - 0.50
> 0.50
(model high)





                                         *** .

Figure 11.  Summary of Fits to the Two - Five Comomponent Mixture

(Calc-Expt) (ppm)
(model low)
< -0.33
-0.33 - -0.27
-0.27 - -0.21
-0.21 - -0.15
-0.15 - -0.09
-0.09 - -0.03
-0.03 - 0.03
0.03 - 0.09
0.09 - 0.15
0.15 - 0.21
0.21 - 0.27
0.27 - 0.33
> 0.33
(model high)
(Calc-Expt) /Expt
(model low)
< -1.00
-1.00 - -0.82
-0.82 - -0.64
-0.64 - -0.45
-0.45 - -0.27
-0.27 - -0.09
-0.09 - 0.09
0.09 - 0.27
0.27 - 0.45
0.45 - 0.64
0.64 - 0.82
0.82 - 1.00
> 1.00
(model high)

Ml V A f^ A 1 1^ AT^O O — — — •» — O^l^O 1* M "f "V ^1 1T"OG — .• — — -.
nixed A.LKenes ----- vjcnet rnxuuxea
Maximum Ozone

** *
** **
* * ****
. * . ** . ** . ******QOO - ***
rC'X ff'XfC *fC 'rCfCyC'fCfCrCfC « yf ^*00 *ffff
**** **** **

Average d([03] - [N0])/dt

* *
**** *
** *** 00000 ********
- * - - ****** - * *****
* ** *** * ****
******* **

(0 - 1986 Runs)

Figure 11.  Summary of Fits to the Two - Five Comomponent Mixture
            Runs (concluded).
                    Mixed Alkenes
                    EC        UNC
Other Mixtures
(Calc - Expt)/Expt
Maximum PAN
0.50 -
0.41 -
0.32 -
0.23 -
0.14 -
0.05 -
0.05 -
0.14 -
0.23 -
0.32 -
0.41 -



                        - •**
  (model high)
            *            :
            **00         : ***
           (0 - 1986 Runs)

Figure 12  Summary of Fits to the NO  - Air Runs Employing Complex
                    EC      "SAI" 8 HC Surrogate
                7 HC Surg     ITC          OTC
                                     Misc. Mixes
(Calc-Expt) (ppm)

 (model low)
0.33 -
0.27 -
0.21 -
0.15 -
0.09 -
0.03 -
0.03 -
0.09 -
0.15 -
0.21 -
0.27 -
0.03 -


            Maximum Ozone

  (model high)
Average d([03] - [N0])/dt
-1.00 -
-0.82 -
-0.64 -
-0.45 -
-0.27 -
-0.09 -
0.09 -
0.27 -
0.45 -
0.64 -
0.82 -

0.09 -


* :XXX*
* .-xxxxxxxxxx
****** : XXXX
******* :x*
* :
(X - 2 runs)




Figure 12.   Summary of Fits to the NO  - Air Runs Employing Complex
            Mixtures (continued).
7 HC Surg
(Calc - Expt)/Expt
(model low)
< -0.50
-0.50 - -0.41
-0.41 - -0.32
-0.32 - -0.23
-0.23 - -0.14
-0.14 - -0.05
-0.05 - 0.05 - ***
0.05 - 0.14 ***
0.14 - 0.23 ***
0.23 - 0.32 **
0.32 - 0.41
0.41 - 0.50
> 0.50
(model high)
(Calc - Expt)/Expt
(model low)
< -0.50
-0.50 - -0.41
-0.41 - -0.32
-0.32 - -0.23 *
-0.23 - -0.14 *
-0.14 - -0.05 **
-0.05 - 0.05 - ***
0.05 - 0.14 **
0.14 - 0.23 *
0.23 - 0.32
0.32 - 0.41
0.41 - 0.50
> 0.50 *
(model high)

"SAI" 8 HC Surrogate UNC
ITC OTC Misc. Mixes
Maximum PAN

• "fcfc^ff^fff ^fr^t
: **
** ; ******
"^T • "JFJt'fc'fC iff
• °Jc$c'$c'4c'k'$f
* ; **** ****
«. - • 'ffTk*Jttff'fp*fc _
* TAnfr^t^f^ik^t^r
* ; **** *
* Tt^lr^fTfrifr^t
* ; ********** ******

Maximum Formaldehyde

^f • *fc*fc"ff
*ff • ifrVnfr^r^r
* ; **********
: ***********
'k'ff'yf * 'ff^fc'fc'^c'^f^e'fe^f'^
*** : *******
- *** - ; **
: ******
* ;*
* :
****** : ****


Figure 12.  Summary of Fits to the NO  - Air Runs Employing Complex

            Mixtures (concluded).

                  7 HC Surg
      •SAI" 8 HC Surrogate

       ITC            OTC
(Calc - Expt)/Expt
Propene Half Life
0.50 -
0.41 -
0.32 -
0.23 -
0.14 -
0.05 -
0.05 -
0.14 -
0.23 -
0.32 -
0.41 -


Figure 13.  Summary of Fits to the UNC Auto Exhaust Runs
Calc - Expt Maximum Ozone Calc-Expt Avg. d( [0_ ] - [NO] )/dt
(model low) (model low)
< -0.33
-0.33 - -0.27
-0.27 - -0.21
-0.21 - -0.15
-0.15 - -0.09
-0.09 - -0.03
-0.03 - 0.03 -
0.03 - . 0.09
0.09 - 0.15
0.15 - 0.21
0.21 - 0.27
0.27 - 0.33
> 0.33
< -1.00
** -0.82
* ** -0.64
*** * -0.45
**** * -0.27
**** . **** -0.09
** 0 . 09
* 0.27
0.64 •
- -0.82
• -0.64
- -0.45
- -0.27
- -0.09
0.09 -
> 1.00

*** ******
****** . *****
* **

(model high) (model high)
Calc - Expt Ethene Half Life Maximum PAN Maximum Formaldehyde
(model low)
< -0.50
-0.50 - -0.41
-0.41 - -0.32
-0.32 - -0.23
-0.23 - -0.14
-0.14 - -0.05
-0.05 - 0.05 -
0.05 - 0.14
0.14 - 0.23
0.23 - 0.32
0.32 - 0.41
0.41 - 0.50
> 0.50

^fTfe 'JC'JC'ic
*** *** *
** . ****
* **


(model high)




xxxxxx *******


(X - 2 runs)

                                    .§7  TROPEHE
                                         ,H_ n-BUTHHE

                         ELAPSED TM ChourO
Figure 14.
       Experimental and  Calculated Concentration  - Time  Plots
       for  Selected Species  Observed in  the  Surrogate  -  NO
       Air  Run OTC-242                                    X
            * - Experimental  data,  side A.   (33% Methanol
               substitution  surrogate)
             - Model calculation,  side A.
            x - Experimental  data,  side B.   (Base  case  surrogate)
       -  -  - - Model calculation,  side B.

                       Paper 8
Evaluation and Intel-comparison of Chemical Mechanisms
                    Fred Lurmann
     Environmental Research and Technology, Inc.

                         Fred Lurmann
                          ERT, Inc.
                  975 Business Center Circle
                   Newbury Park, CA  91320
                         Presented At

                   The EPA/ASRL Workshop on
       Evaluation/Documentation of Chemical Mechanisms

     Various aspects of atmospheric chemical mechanism evaluation
and intercomparison  are  discussed  in  this  paper.   The  topics
covered include  the methods  used  to  test   chemical  mechanisms
against environmental chamber data,  results from recent mechanism
evaluations, and  uncertainties   in  the  modeling and  evaluation
process.  Also presented are some preliminary mechanism intercom-
parison results for  the  Carbon  Bond IV and  Carter  et  al.  (1986)

Mechanism Evaluation

     The general methods for developing  and evaluating atmospheric
chemical mechanisms were  formulated by  the  pioneering  workers in
this field  (see e.g. Niki,  Daby,  and  Weinstock 1972);  Demerjian,
Kerr, and Calvert  1974; Hecht,  Seinfeld,  and Dodge  1974).   While
the kinetic and mechanistic data base, and the number of environ-
mental chamber experiments have greatly expanded in recent years,
and testing  procedures  have  been  refined,   the  basic  methods
recommended by the early workers are still followed today.   These
procedures consist  of   developing detailed   chemical  mechanisms
that consist of  elementary  reactions  and testing  the  mechanisms
against the best  available  smog chamber experiments using  a con-
sistent protocol (i.e., without  arbitrary run-to-run adjustment of
uncertain parameters).    More recent  studies  (Whitten  and  Hogo
1977; Carter et al. 1979;  Atkinson et al. 1982; Killus and Whitten
1983; Carter et al. 1986)  have shown the importance of testing the
mechanism according  to  the  heirarchy  of species, testing against
data from chambers with different radiation intensity and spectra,
testing against a  large number of experiments, and testing with a
consistent model  of the  chamber  effects (such  as  heterogeneous
wall reactions leading the offgassing  of radical  species and NOX).

     Examples of  mechanism  performance  data  obtained  with  a
recently developed detailed  mechanism are  shown in  Figures 1 -6
(from Carter et al. 1986).  The figures show the predicted versus
observed maximum ozone  concentrations in six  different  types of
chamber experiments carried out  in  two  indoor  chambers (SAPRC EC
and 1TC) and two  outdoor  chambers  (UNC and  SAPRC).   The experi-
ments involved irradiation of NOX with individual alkenes, indivi-
dual alkanes, individual aromatics,  mixtures of organics from two
groups, complex  mixtures   of  organics  from  three  groups,  and
automobile exhaust (UNC only).    A total of 490 chamber runs were
employed in the evaluation.  The results  show that the mechanism
is able  to  predict the  maximum  ozone  within ±30%  in  the  vast
majority of cases.  The  noticeable  exception  to  this  result are
the predictions  for   NOX   -  alkane   experiments   which are  very
sensitive to the assumed  chamber radical  source strength.  Table
1  shows the average bias  and error  in the maximum  ozone for the
different groups  of   runs employed  in  this  evaluation.   These
results show that  the average  bias  in  the  ozone  predictions is
small, usually less than  10%,  and usually positive.   The average
error is typically 20%  to 30%.  The  performance  results for the
organic mixtures are particularly important because the mechanisms
are generally applied  to complex  mixtures  in  atmospheric modeling.
Overall, the average  bias  and  error for the  mixtures are +4% and
±24%.  Statistics  for timing of the  ozone peak and rate  of NOX
oxidation are quite similar.

     In developing the  Carter  et al. mechanism,  many variations
in mechanistic  parameters were  investigated by  testing  against
chamber data.  One of the more interesting  and important results
was that while alternate  mechanistic  assumptions   affected  the
average bias for a group of runs (i.e.,  making the mechanism more
or less  reactive),  the alternate  assumptions had  only  a  minor
affect on the  average error.   This  suggests  that numerous mecha-
nisms can fit the data reasonably well,  so it may be difficult to
distinguish between mechanisms  solely  on   the   basis  of  their
performance relative  to  chamber data.   It  also suggests  that
there are basic uncertainties in the mechanism, chamber data, and
chamber characterization which  presently limit the performance of
the mechanism  in  this  type   of testing  program.    No   one  has
carefully tried to estimate the uncertainties of the data employed
in this modeling system, however, I have made  some rough estimates.
N02 photolysis rate are  known  to within  ±10%  in  indoor  chambers
and no better  than ±20  in outdoor chambers.  Many rate constants
have uncertainties of ±20  or more  (see Atkinson  and Lloyd 1984;
Atkinson 1986).  Product  coefficients for the organic reactions
often have  uncertainties  of ±20  or more.   Uncertainties  in the
chamber concentration data,  used for initial  concentrations and
for comparison purposes, are  also significant for organics (±10%),
PAN (±20%),  and  oxygenates (±30%).    Lastly,  the  chamber radical
source strength  could easily be  uncertain  to ±50%  or more.  If

these uncertainties are realistic, which I believe they are, then
it is  not  surprising  that  the mechanism  performance evaluation
results show ±20%  to  ±30%  error on  the  average.  This  level  of
error in the output is totally consistent with  the  level of errors
in the input data.

     If regulatory agencies need mechanisms that  are substantially
more accurate and  precise  than the present  generation  of mecha-
nisms, then  higher  quality  chamber  experiments  and   improved
mechanistic and  kinetic-data must  be obtained.   The technology
is generally available to  conduct  higher quality chamber experi-
ments.  For example,  the uncertainty in  the N0£ photolysis rates
can probably be  reduced  substantially by  continuous in  situ k1
measurements in the experiments.  DNPH/HPLC methods are available
for obtaining high quality data for oxygenated species.   The dual
organic tracer technique could be employed in many experiments to
derive OH  radical  concentrations   and   improved radical  source
strengths.  FTIR could be  employed to measure nitrous acid early
in the  runs  to  get  better  estimates  of  its  role  initiation.
Thus, there is clearly substantial room for  improving the quality
of the data used in the evaluation process.

Mechanism Intercomparison

     A sufficient number of chemical mechanisms had been developed
by 1980 to allow researchers to begin  intercomparing the mecha-
nisms.  Carter et  al.  1982,  and Jeffries et al.  1981  showed that
while the  mechanisms  predict  different,  but   still  reasonable
results for  chamber  experiments,   they  predicted  substantially
different VOC control  requirements when used  with the  Empirical
Kinetic Modeling Approach (EKMA).   This  topic was explored further
by McRae et  al.  1983; Leone  and  Seinfeld 1984,  and  Schafer and
Seinfeld 1985 using slightly more up-to-date mechanisms.  Results
are presented here which employ even more  recent versions of the

     Figures 7  show a comparison  of the  ozone  predictions from
the Carbon Bond  IV  mechanism and  from  the  Carter  et  al.  1986
mechanism for simulations  with 1   ppmC  of urban VOC  and initial
NOX levels 0.05,  0.1,  0.167  ppm.  The figure  shows that the Carbon
Bond IV  mechanism  consistently  predicts  higher ozone  than  the
Carter et al. mechanism, and that the discrepancy  tends to  increase
as the initial VOC/NOX ratio decreases.   Figure 8  shows the NO,
N02, and 03 profiles for a similar simulation carried out with an
initial VOC/NOX  ratio of  3.   This  shows  that  both  mechanisms
appear to oxidize  the NOX  at the  same  rates,  however,  the ozone
prediction is 20%  higher on the second  day  with the  Carbon Bond
IV mechanism.  Figures 9 and  10 compare the Oh,  H02,  HN03, h202,
toluene, and ethene profiles  predicted  by the two  mechanisms for
the initial VOC/NOX =  3  case.  The agreement on these species is
generally better than that  found for ozone.

     Another set of simulations was carried out with the same two
mechanisms in the OZIPM model  (Version  3 - Whitten and Hogo 1986).
An EKMA analysis was performed for nine cases, representing urban
conditions with emissions, an  ozone design  value  of 0.24 ppra,  an
initial mixing height  of  250  m and three  maximum mixing heights
(500, 1000, and 1500 m), and three initial VOC/NOX ratios (6, 10,
and 20).  The OZIPM predicted VOC control requirements to achieve
compliance with the 0.12 ppm standard  are  shown  in Table 2.   For
the 1000 m maximum mixing height cases, the results of the Carter
et al. and CBM-IV mechanisms are  29% and  41%  at  VOC/NOX - 6, 58%
and 68%  at VOC/NOX  =  10,  and  80% and  85%  at  VOC/NOX =  20,
respectively.  These differences are surprisingly large when you
consider that the mechanisms are based on  the  same basic kinetic
and mechanistic  data  (i.e.,  Atkinson  and  Lloyd  1984)  and  have
been testing against many  of  the  same smog chamber experiments.
The differences are almost as large as  the differences identified
five years ago by Jeffries et al.  1982.  Thus, while the accuracy
and precision  of  the  mechanisms  may  be  improved  in  the  new
generation, the problem  of different  mechanisms  giving signifi-
cantly different  VOC   control  requirement  predictions  is  still
very much with us.

                  TABLE 1

RUN TYPE               BIAS (%)*      ERROR (%)*
Formaldehyde            -1              19
Acetaldehyde           -26              26
Other Carbonyls         +4              44
All Carbonyls              -5              25

Ethene                  +2              18
Propene                 +3              18
Butenes                 +4              34
All Alkenes                +3              21

Butane                 +31              67
Branched Alkanes       +34              49
Long-chain Alkanes     +83              84
All Alkanes               +46              69

Benzene                 +3               5
Toluene                +11              24
Xylenes                 -9              16
Mesitylene             -11              21
All Aromatics              +1              19

All Single HC Runs        +12              33
Simple Mixtures        +10              35

Mini Surrogates        +10              22

Full Surrogates         +3              23

Auto Exhaust           -11              15

All HC Mixtures            +4              24

All Run Average            +7              28
*  Positive bias indicates model overprediction.

             TABLE 2


                         Chemical Mechanism
Mixing Height


              O   ETHENE
              +   PROPENE
              «   1 -BUTENE
              A  TRANS-2-BUTENE
              x   ISOBUTENE
                   i    i    r   T^   iiir
                  0.2      0.4      0.6      0.8
                       ObMrved Maximum Ozon« (ppm)
                                           i   i   i
                                           1       1.2
«   1-BUTENE
X  X'
      i   i   i   i   i  i   i   i   i   i  i   i   r  r   r  r  i
   0     2      4     6     8     10    12    14    16    18
           Ob«erv«d d( [03] - [NO] )/dt (ppb/min)

   tredicted versus observea maximum ozone concentrations ana
   timing parameter for alkene  - iNOx chamber  experiments.










  0.1 -
               O   BUTANE
               +   2.3 DJMETHYLBUTANE
               o   N-ALKANES
               A   ISOALKANES	
+ n
                      0.2        0.4        0.6
                        ObMrvtd Uoximum Ozorw (ppm)
    "    5H
         1 -
                                   2.3 DIMETHYLBUTANE;
                               •   N-ALKANES
                               A   ISOALKANES
            Observed d( [03] - [NO] )/dt  (ppb/min)
2.  Predicted versus observed maximum ozone concentrations and
    timing parameter for alkane - NOX chamber experiments.

       0.3 i
       0.1 -1
          O   BENZENE
          +   TOLUENE
          «   XYLENE
          A   MESITYLENE
                      0.2          0.4           0.6
                      Obt«rv«d Maximum Ozone (ppm)
            O   BENZENE
Figure 3.
             2     4     6     8    10   12   14   16
                Observed d( [03] - [NO] )/dt  (ppb/min)

        Predicted versus observed maximum ozone concentrations and
        timing parameter for aromatic - NOX chamber experiments.

                      Simple  Mixtures

      1 -

  <-     8-

^     7"

^     6-

5     5^



      1 -
            0  Mixed from 1 group

            4  Mixed from 2 groups

                  Observed d( [03] - [NO] )/dt (ppb/min)
Figure 4.  Predicted versus observed maximum ozone concentrations

           and  timing parameter for chamber experiments with simple

                    Surrogate Mixtures
   O EC 7-Component
   •f ITC A OTC a-Compon«nt
     UNO Multi-Component.
   A ITC 4-Componont
   x UNC 3-Compon«nt
                  0.2   -   0.4      0.6       0.8
                     Observed Maximum Ozone (ppm)







1 -
           O  EC 7-Component
           *  ITC 4 OTC 6-Component
           «  UNC Multi-Component.
           A  ITC 4-Component
           x  UNC 3-Component
                  Obeerved d( [03] - [NO] )/dt (ppb/min)
Figure 5.  i'redicted versus  observed maximum ozone concentrations
          and timing parameter for chamber  experiments with complex
          organic mixtures  and

                       UNC  Auto  Exhaust
                     0.2        0.4        0.6       0.8
                       ObMrv«d Maximum Ozon« (ppm)
Figure b.
          Obwrved d( [03] - [NO] )/dt (ppb/min)
Predicted versus  observed maximum ozone and a timing parameter,
average d([03]-lNO])/dt for auto exhaust-NOx smog chamber
experiments conducted  in four chambers.

   Mechanism  Comparison  at VOC/NOx = 6,20
               10  11  12  13  14  15   16   17  18
                        "Tim* (hour)
                        — Carbon Bend IV
    Mechansim  Comparison  atVOC/NOx=10
                                               VOC/NOx -1C
               i  i i i i  i i i I  i i i  I i i i  i i i  i i i i  i i i  i
               10  11  12  13  14  15  16   17  18
                            TTm« (hour)
                                B«nd IV
Figure 7. Ozone predictions from the SAPRC/ERT and Carbon bond IV
        chemical mechanism for case with initial VOC to NOX ratios
        of b, 10, and 20.

    Mechanism  Comparison  at  VOC/NOx  =  3
                               NO * NO2
          i i i  i i  i i  i i   i  i i
          10 12 14 18 18 2O 22
i  i i
2  4
 i ii ii i  iii i  i
10 12 14 16 18 2O
 Figure 8.  NO, NC^, and ozone predictions from the SAPRC and Carbon Bond
          IV chemical mechanisms for a case with an initial VOC to
          ratio of three.

    Mechanism  Comparison  at VOC/NOx  =  3
        8 10 12 14 16 18  20 22  0  2  4  6  8  10 12  14  16 18 20
        8  10 12 14 16  18 20 22  0   2  4  6  8  10  12 14 16 18 20
                                Carbon Bond IV
Figure 9.  OH and h02 radical predictions  from Che 3APRC and Carbon Bond
         IV chemical mechanisms for a case with an initial VOC to
         ratio of three.

    Mechanism  Comparison  at VOC/NOx  =   3
                              HNO3 & H2O2
                                   HNO3   / /
                                           /' '
      "  i i  i i  i i  i i  i i  i i  i i  i i  i i  i i  i i  i i  i
        8  10 12 14 16 16 20 22 0  2  4   6   8  10 12 14 16 16 20
                             To1u«n« A Ethan*
        6  10 12 14 16 18 2O 22 0   2   4   6   8  10 12 14 18 18 20
                - — SAP9G
Carbon Bond IV
Figure 1U.  HM}}, H202,  toluene, and ethene predictions from the SAPRC
          and Carbon Bond IV chemical mechanisms for a case with an
          initial VOC to NOx ratio of three.


Atkinson, R. Chem. Rev.,  85, p. 69, 1986.

Atkinson, R., A. C. Lloyd, J. Phys. Chem. Ref. Data, 13, pp. 315-
444, 1984.

Atkinson, R., A. C. Lloyd, L. Winges, "An Updated Chemical Mecha-
nism for Hydrocarbon/Nitrogen Oxides Photooxidations Suitable for
Inclusion in Atmospheric Simulation Models", Atmos. Envir. ,  16, pp.
1341-1355, 1982.                                           ~

Carter, W. P., A. C. Lloyd,  J. L.  Sprung,  J.  N. Pitts, "Computer
Modeling of  Smog  Chamber  Data:   Progress  in  Validation of  a
Detailed Mechanism for the Photooxidation of Propene and n-Butane
in Photochemical Smog," Int. J. Chem. Kinetics,  11 ,  p. 45, 1979.

Carter, W. P. L. ,  A.  M.  Winer, J.  N. Pitts,  "Effect  of Kinetic
Mechanism and Hydrocarbon  Composition on Oxidant-Precursor Rela-
tionships Predicted by the EKMA Isopleth Technique," Atmos. Enyir.,
]_6, pp. 113-120, 1982.

Carter, W. P. L.,  F, W. Lurmann, R. Atkinson, A. C.  Lloyd, "Develop-
ment and Testing of a  Surrogate  Species  Chemical Reaction Mecha-
nism," Final  Report EPA  Contract  No.   68-02-4104,  ASRL,   U.  S.
Environmental Protection Agency,  Research Triangle Park, NC, 1986.

Demerjian, K. L. ,  J.  A.  Kerr, J.  G. Calvert,  "The  Mechanism of
Photochemical Smog Formation,"  in Adv. in Environ. Sci. and Tech.,
4, John Wiley and Sons, NY, 1974.

Hecht, T., J. Seinfeld, M. Dodge,  "Further Development of Genera-
lized Kinetic  Mechanism   for  Photochemical  Smog," Environ. Sci.
Tech., B, p. 327-339,  1974.

Jeffries, H.  E. ,   K.   G.  Sexton,   C.  N.  Salmi,  "The  Effect  of
Chemistry and  meteorology  on  Ozone  Control  Calculations  Using
Simple Trajectory Models  and the  EKMA  Procedure," EPA-450/4-81-
034, U.  S.  Environmental  Protection  Agency,   Research  Triangle
Park, NC, 1981 .

Killus, J. P.,  G.  Z.  Whitten,  "Effects  of Photochemical Kinetics
Mechanisms on Oxidant  Model Predictions," EPA-600/3-83-111 , U.S.
EPA, RTP, 1983.

Leone, J. A.,  J.  H.  Seinfeld,  "Evaluation of Chemical Reaction
Mechanisms for  Photochemical Smog, Part II:  Quantitative Evalua-
tion of  the  Mechanisms,"   Final  Report  CA  810184,  Atmospheric
Sciences Research  Laboratory,   U.  S.  Environmental  Protection
Agency, Research Triangle Park, NC, 1984.

Niki, H. , E. Dady, B. Weinstoctc,  "Mechanisms  of Smog Reactions,"
in Photochemical  Smog and Ozone  Reactions,  Adv. in Chemistry
Series. 113, p. 13, 1972.

Whitten, G.  Z.,   ri.  riogo,  "Mathematical Modeling  of  Simulated
Photochemical Smog,"  EPA-600/3-77-011 ,  (J.  S.  Environmental Pro-
tection Agency, Research Triangle Park, NC,  1977.

A. Background Document - Need for Chemical Mechanism Documentation
                           Kenneth Sexton and Harvey Jeffries

B. Background Document - The Science of Photochemical Reaction
                           Mechanism Development and Evaluation
                           Harvey Jeffries and Jeffrey Arnold

C. Workshop Agenda

D. Workshop Attendees

               Appendix A
Need for Chemical Mechanism Documentation
    Kenneth Sexton and Harvey Jeffries

                               Need For
              Chemical Mechanism Documentation

                  Kenneth Sexton and Harvey Jeffries
                       Department of Environmental
                         Sciences and Engineering
                        University of North Carolina
                           Chapel Hill, NC 27514


Documentation is needed from both the mechanism developer and the mechanism
users to assure that implementation of a chemical mechanism is error-free. Docu-
mentation from the developer serves as guidance, and documentation from the user
assures that no mistakes were made in its implementation, and that all assumptions,
methods, and data are properly identified.

   Implementation of a chemical mechanism usually requires more than represent-
ing a set of reactions  in a model for simulation of the chemistry. The mechanism
developer, consciously or not, uses the chemical model with certain methods,  as-
sumptions and conditions.  Without guidance from the developer,  users may use
other methods, make other assumptions, with other conditions. The mechanism
developer might disagree with these choices,  resulting in denial from the devel-
oper that the chemical mechanism was properly implemented. Implementation of a
chemical mechanism, therefore, requires that knowledge of the supporting methods,
assumptions, and conditions, as described or suggested in guidance from the devel-
oper, be demonstrated or perhaps implemented, as well as the mechanism. If a user
wishes to deliberately implement a user-modified version of a chemical mechanism
or if he wishes to use  it in a manner which  does not conform to guidance from the
developer, then the user should carefully identify the modifications (to either chem-
istry or procedures for use) as part of the description of "the chemical mechanism
as modified by...".

Implementation Problems

There are several problems which  can hamper implementation of a given chemical
    • communication problems resulting from lack of detailed specification and

                                                        Lack of Detailed Specification
    • confusion between the mechanism and other computer programs used to
      support it or use it.
    • problems encountered when using a different computer/system and/or sup-
      porting code due to unrecognized assumptions; and
    • less obvious problems which may result in misuse and predictions different
      than those which would be generated by the mechanism developer.
We will discuss  each of these below.

Lack of Detailed Specification
Many of the problems encountered are due to the lack of a clear description/definition
of "the mechanism" by the developer. This, in turn, affects how it is implemented,
used, modified,  and reviewed by others in different situations.

    Proper implementation and demonstration of proper implementation requires
adequate documentation by both developer and the user if the work and results are
to be adequately reviewed.

    Ideally, proper  implementation implies that the mechanism implemented is as
published or distributed personally by the developer or the EPA. Versions obtained
from another user may be outdated or modified  incorrectly.  Formal written guid-
ance with examples are also needed to insure proper implementation.  These are
often lacking, or incomplete,  or are only obtainable from the developer. There are
often different versions of a given mechanism in use. The "version problem" results
from continuing development and mechanism changes.  Ideally, one should obtain
the chemical mechanism and guidance directly from the developer or the official dis-
tributor (such as the EPA). Guidance has changed after a mechanism was adopted.
Examples should be used to verify that implementation of both the mechanism and

    The important assumptions, limitations, procedures, or conditions which are by
definition of the mechanism developer an inherent part of the mechanism, should be
identified. These might include approach and structure, treatment of light, temper-
ature, pressure, water, radical initiating sources  (general or specific to a chamber
or an atmosphere), empirical chemistry, background reactivity, composition treat-
ment, treatment of secondary functional groups of hydrocarbons, and rate or stoi-
chiometry modification with composition variation (state whether to do so or not).
For example, some  mechanisms ignore temperature effects and give a rate constant
evaluated at some  temperature (often 298K) and more commonly A-factors and
activation energies  are used.  In some mechanisms, photolytic rates are  expressed

Using Different Computer Systems
as ratios to the NO2 photolysis rate and in others the rates are calculated explicitly.
The sources of rate constants including photolytic rate constants should be identi-
fied. Quantum yields, cross-sections, and light intensity assumptions as a function
of wavelength should be given.

    The developer should indicate which reactions  are thought of as explicit and
which are simplifications.  He should describe which reactions can be updated as
new kinetic information becomes available and which are to be left "as is"  because
of other compensating chemistry.

Confusion About Support Programs

Confusion and errors arise from the concept of "the mechanism" and some larger
model which may use it.  Also the distinction should be made between the theoretical
description of  the chemical mechanism, and any computer code currently used to
represent and implement the representation, or code used to support it. The next
section discusses some of the problems which can be avoided with the awareness of
these concerns.

Using Different  Computer Systems

An awareness of the purpose and assumptions of supporting programs for a given
chemical mechanism is necessary in implementation on a different computer/system
or with different programs. Depending on the type of example simulations which
a new user may be trying to duplicate, different support programs or procedures
will be used.  One of the main problems in reproducing example simulations  are
the dependence on supporting programs which have characteristics which may be
difficult to reproduce or may have their  own peculiarities or errors.  For example
different programmers have implemented the "Characteristic Curve Mixing Height
Profile" for estimating dilution of an air parcel. Different numerical techniques can
be used to generate time-varying emission rates for input to a photochemical system
from hourly averaged emission estimates. Calculation of photolytic rates from even
the same  source of input  data can  be affected by the algorithm and numerical
methods used. Only by duplicating these routines or by using the specific output
of these routines can the examples be reproduced accurately.

    The descriptions of supporting programs (both the original and new) including
their purpose,  assumptions, and limitations should be given. Ideally the listings of
all code should be available. This is to allow  for comparison to assure that the code
is error-free, can be executed, and adheres to the developers guidance.

                                                     Using Different Computer Systems
    Often the "appearance" of the mechanism must change when it is implemented
and solved with different  code on a different  machine.  Such changes allow for
several types of problems.  Examples are needed to allow the new user to verify
proper implementation.

    Often there are insufficient examples provided by the developer to assure a new
user of proper implementation or proper use of the chemical mechanism and/or
programs used. Often only printouts (no plots) are provided with results listed for
only a few  large time- steps. Enough data should be printed to allow for careful
comparison with the timing of important chemical events.

    Mechanisms appear or can  be obtained in different forms. Some forms are not
clear or may be incomplete descriptions.  Mechanisms are often not implemented
in the form that would be used in a journal publication. Journal articles  often
do not provide enough information to reproduce the examples shown.  Sometimes
the mechanism representation file is distributed. This mechanism file read by the
computer might appear very different.  Sometimes the simulation printout lists a
representation of the  mechanism. Space  limitations might require the developer
to round-off stoichiometry, rate constants, A-factors and activation energies.  The
reactions might be rearranged to conform to the syntax requirements and limitations
of a computer program which reads it. Reactions might be condensed to save space.
Rate constants might be evaluated at some given temperature or for some assumed
water concentration.

    Clear unabbreviated listings of the chemical mechanism are therefore useful in
implementing by a new user. Otherwise, new users may proceed making their own
choices and assumptions for aspects not addressed in an incomplete discussion or de-
scription of a chemical mechanism model and thus create a different representation
than the mechanism authors intended.

    When a particular algorithm is essential to the application of the mechanism, for
example, a  particular way to treat the time variation in hourly averaged emissions,
problems may arise. Different machine/code environments can present problems for
either the direct transport of code between machines/systems or for the translation
of listings of code to new code.

    These effects may not be obvious.  Very often there are differences between
compilers on different machines. The compiler or system treatment of numerical
errors may  be different. There may be machine limitation differences (eg numerical
accuracy).  Compiler or language differences may cause translation problems. Poor

A Proposal	

coding practices such as dependence of "tricks" (taking advantage of a particular
odd trait of a language or compiler)  may cause translation problems. An exam-
ple would be that some compilers perform an auto-zero of uninitialized variables.
Problems of this type highlight the need for simulation examples from the devel-
oper.  The examples provided and used for comparison should demonstrate and
test all portions of the code.  Ideally, the source code (at  least listings) should be
provided, with  code documentation of high standards. The code documentation
should include description of above concerns and identification of dependencies on
peculiarities of compiler and  canned  procedures.  References  should be given for
sources of algorithms and code fragments.

   Given the possible causes of disagreement in attempting to reproduce an exam-
ple simulation, there needs to  be a determination as to what is considered "Accept-
able Accuracy".

The Original  Errors Problem
Sometimes errors are made by the developer which are subsequently detected by
another user. There is  often a problem with determining which errors can be cor-
rected by a user without "changing the mechanism". Publication and distribution
errors have been found. Typographical errors by the user are most common during
implementation. Typographical errors in species names, for example, can  be very
difficult to  detect.  Bad copies or wrong versions or examples are sometimes dis-
tributed by the developer.  Implementation and misuse errors by new users often
result from insufficient  guidance from the developer.

A Proposal

The chemical mechanism developer should provide a guidance  document which in-
cludes examples.  This  document should begin with a  written general  description
of the chemical mechanism including the approach and type of representation. As-
sumptions and  limitations inherent in the mechanism formulation should be iden-
tified  to help explain the general use procedures. Describe which "species" in the
mechanism represent the different chemical  species in atmosphere. Identify which
species and chemistry  are explicit.  Identify which reactions  if any are modified
(rate or stoichiometry)  by composition and explain how with examples, identifying
assumptions.  Explain how water, oxygen, pressure, M, light, and temperature are
treated and represented and incorporated into the mechanism (rate constants). All
reactions that are dependent on temperature, water, and light, should be identified.
Identify the source of photolytic rates and explain treatment of light. Outline pro-

                                                                     A Proposal

cedures for determining light for different locations or conditions.  Explain how (if
allowed) to update the mechanism if photolytic rates need to be changed later,

    For implementing a given mechanisms representation into new users code, the
new user must have a listing of the mechanism, preferably directly from the devel-
oper (recently)  which was also used to generate the example also provided.  This
should be included in the guidance document. To minimize chance for typographi-
cal  errors by a new user,  the mechanism should ideally be provided in a computer
file  in ASCII which would allow others to read directly.

    The new user should  obtain example simulations, preferably from the user at
the  same time the mechanism is supplied. Examples should be as complicated as
the  intended use (ie try to show examples using variation of several parameters, and
options). The examples should provide prediction results with plots and printouts.

    The examples should  provide initial conditions (both original "real" and for the
mechanism) including chemical species, light, temperature, and water, especially if
these  effect rate constants.  Descriptions and examples of input condition calcula-
tions from composition speciation and non-speciation (ie from detailed HC analysis
and from none or incomplete HC speciation) must  be described. The compounds
that are considered non-reactive by the mechanism developer should be identified.
If these assumptions affected the rate constant of a reactions, these should be  iden-
tified  so that changes can be performed by the new user (if allowed by definition).
Any assumptions about default compositions should be included for emissions, sur-
face transport,  and aloft  materials.  This would include aldehydes, hydrocarbons,
NO, NO2, O3, etc.

    If heterogeneous processes are included in the mechanism, these should be  iden-
tified  and for which types  of  simulations these are  intended.  Any guidance on
adjustment of these rates for different conditions should be given.

    The new user  should be able to duplicate the results provided in all of the
examples. Inability to duplicate the results obviously indicates an  implementation

    An outline of an example guidance document is given in Table 1.

A Proposal	

            Table 1. Example Outline of Guidance Document

    1. Structure of Chemical Mechanism
          A. General Description
                 i. intended use
                       > basic input
                       > basic output
                ii. representation schemes
                iii. basis
                       > sources of reactions and kinetic information
                       > reaction simplification techniques
                              o condensation techniques
                              o approximation techniques
                              o verification
                       o techniques for rate constant approximations
                iv. assumptions and limitations
                v. disclaimer
          B. Inorganic chemistry
                 i. explicit  chemistry
                       > rate constants
                       > assumptions and omissions
                       > references
                ii. simplifications
                       > assumptions
                       > user adjustments
                       > references
                iii. adjustments for heterogeneous processes
                       > representations and assumptions
                       > measurements and experimental data required
                       > procedure for estimation of heterogeneous rates
          C. Organic chemistry
                 i. explicit  chemistry

                                                                 A Proposal
                   >  rate constants
                   >  assumptions and omissions
                   >  references
            ii. simplifications
                   >  assumptions
                   t>  user adjustments
                   o  references
            iii. natural background reactivity
                   >  assumptions
                   >  representation
            iv. adjustments for heterogeneous processes
                   >  representations and assumptions
                   >  measurements and experimental data required
                   >  procedure for estimation of heterogeneous rates
      D. Light and Photolytic Rates
             i. Assumptions
            ii. Treatment
      E. Temperature
             i. Assumptions
            ii. Treatment
      F. Deposition
             i. Assumptions
            ii. Treatment
2.  Representation of Chemical Reactions
      A. syntax
      B. rates and activation energy units
      C. light dependent reactions
      D. water dependent reactions
      E. oxygen assumptions
      F. "M" and  pressure dependent reactions
      G. description of file of mechanism representation
3.  Photolytic Rates

A Proposal
         A.  Theory, approach and assumptions
         B.  values needed for mechanism species and units
         C.  Method of Calculation
         D.  Sources of data
         E.  Procedures for Determination of Photolytic Rates
         F.  Description of programs and data files
         G.  examples
    4. Determination of Initial Concentrations
         A.  NOx
         B.  Hydrocarbon Calculations
                 i. Types of input  data
                       > detailed site-specific analysis: criteria for HC analysis to
                         be considered complete
                       > partial analysis: guidelines for estimating composition
                         of non-measured VOC
                       t> no detailed data and default compositions
                ii. Methods of Calculation
                iii. Sources of Data
                iv. Procedures for ...
                v. Description of programs and data files
                vi. examples
         C.  03
         D.  CO
         E.  other compounds
         F.  examples
    5. Hydrocarbon Calculations for Reaction Modifications
         A.  Types of modifications
                 i. rate constants
                ii. stoichiometry
         B.  Method of Calculation
         C.  Sources of Data
         D.  Procedures for ...

	A Proposal

       E.  Description of programs and data files
       F.  Examples
 6. Suggestions for Support Programs
       A.  Numerical Solution of Chemical Kinetics
       B.  Emissions entrainment
       C.  Meteorology (dilution)
       D.  Sun's Position
       E.  Photolytic Rates
 7. Example Simulations
       A.  Smog Chamber
       B.  Urban Atmosphere
 8. Listings of Programs
 9. References
       A.  for citations made above
       B.  for all available literature that describes the scientific basis and testing
          of the reaction  mechanism.

A Proposal         	

    Guidance from the EPA for the proper general use  of photochemical models
should be developed. "Use" would include documentation by both mechanism de-
velopers and users.  Such documentation would require demonstrating the ability
to reproduce an example and a full explanation of all assumptions and  prepara-
tory calculations. This is needed to assure proper implementation, use, and review.
Guidance, policy and definitions of model terms and procedures need to be devel-
oped. The EPA should define the format of a "guidance document" to be written
for a chemical  mechanism and a "user modeling documentation report".  The for-
mat of the guidance document should include at least one example calculation for
every part of the mechanism input  and  use. The documentation reported by the
user should cover the same concerns of the mechanism developer.  If the  user fol-
lowed the suggested guidance then the documentation required should actually be
easily prepared.  User specific input data of course should be listed with  example

    Chemical mechanisms and photochemical models are implemented with com-
puters because of the complexity of the problem. The implementation of chemical
mechanisms and photochemical  models  in general would be much more success-
ful if those in this field agree to conform to the documentation standards used in
computer programming.

    Official policy and guidance  on  modeling, the requirement of a guidance doc-
ument from the  developer,  documentation from the mechanism/model user, and
insistence of good documentation in general could form the basis of a model qual-
ity assurance program which would help  assure proper implementation of chemical
mechanisms and models.  Some important policy questions pertaining to imple-
menting chemical mechanisms are given in the next section.

    EPA  should have responsibility  of being the official source of chemical mech-
anisms and photochemical models.  Also the EPA should have responsibility for
reviewing and determining the adequacy of guidance for  each. Ideally, part of the
review of modeling for EPA would be that EPA personnel could reproduce the re-
sults, ideally with their own code implementing the chemical mechanism with their
own support programs.

                                                     An Example of Current Guidance
                                                       and Mechanism Documentation

 An Example of Current Guidance
    and Mechanism Documentation

It  is interesting to examine a recently guidance document for a new mechanism
developed for the EPA and to compare it to the outline presented above and some
of the ideas presented here.

    "Guidelines for Using OZIPM-3 with CBM-X or Optional Mechanisms, Volume
I," is mostly an instruction manual for using OZIPM as the title states. It starts
with a concise overview of the model. Definitions are given and the technical basis
of the model is discussed. Necessary calculations are explained, the mechanism is
listed, and example simulations are provided. The explanation of how the programs
work and the code are presented in Volume II. However, some important items seem
to be missing.

   The document explains that the photolytic rate constants are represented with
a multiplication factor to the NOj photolytic constant, and that the user can change
these. However, there is no explanation as to the source of any of these rates, making
it  difficult for a user to determine how much one should  change  these to attain
another absolute rate. There is also the note that these rates have been increased
to represent an altitude of 600 meters.  These new rates are not supported by any
additional discussion or references.  Since a user may be interested in modifying
these rate constants,  it would be helpful to know what assumptions and techniques
were used to estimate the existing values.

   The mechanism structure and chemistry are only discussed in a few paragraphs.
Several references are given to earlier reports for a more complete discussion of the
development of the mechanism, technical issues, and associated calculations. It is in
some of these earlier reports that show the simulation of smog chamber  experiments

   It appears that at least in this case, several reports are needed to document and
describe fully a chemical  mechanism and how to use it in a photochemical model.
It illustrates the fact  that a listing of a mechanism is incomplete documentation for
implementation in a new  environment.

Questions for Discussion
Questions for Discussion

    1. Does the mechanism developer have complete control on the definition of the
      mechanism and how it is used?
    2. Does the validation database or procedure become part of the definition or
      description of a chemical mechanism ? Does it effect the use of the mechanism
      ? Should it limit the application ?
    3. What aspects of a  mechanism or model can be "updated" or "corrected"
      (with more scientifically defensible information) and by whom? What level
      of documentation of these changes should be required? What should EPA's
      response be to a SIP to modeling performed using adjusted photolytic rates
      from different assumptions of actinic flux from different assumptions of albedo
      or cloud cover?  Is  the user using  "the validated mechanism" once such a
      change is made?
    4. What changes can  be made to  "validated" chemical mechanisms/models
      without invalidating?  Are there different answers to this question from  the
      scientific and policy and regulatory groups?
    5. Should guidance for model use recognize and distinguish between regulatory
      modeling vs research modeling and provide different guidance for  each?
    6. Are  there circumstances in which it is  satisfactory for a model/mechanism
      user to "correct" known "incorrect" or  "scientifically unsound" aspects of a
      mechanism before its use?

                  Appendix B
The Science of Photochemical Reaction Mechanism
          Development and Evaluation
              [Strawman Document]
       Harvey Jeffries and Jeffrey Arnold

       The Science of
Photochemical Reaction
Mechanism Development
      and Evaluation
 Harvey Jeffries and Jeffrey Arnold

     Department of Environmental
       Sciences and Engineering
     University of North Carolina
       Chapel Hill, NC, 27514
       (919) 942-3245/966-5451
          November 17, 1986

            Prepared for
    U.S. Environmental Protection Agency
            Workshop on
     Evaluation and Documentation of
      Chemical Mechanisms Used in
          Air Quality Models
          December 1-3, 1986


The role of this document is to stimulate discussion. To accomplish this goal, we have
     >  defined the scientific process, both in a general sense and more specifically for the field of
        air pollution modeling;
     >  provided evidence that our view is accurate;
     o  argued the implications of this view;
     >  made recommendations for continuing the advancement that has occurred over the last 15
To accomplish the first step we have relied extensively on the work of Thomas Kuhn, a science
historian and Professor of Philosophy at the  Massachusetts Institute of Technology, whose work
has been described by the journal Science as "A landmark in intellectual history."

    We believe that dialogue within the community of scientists practicing in the field of air pol-
lution modeling and testing is needed to plan an action to counter a developing preception, one
that questions the efficacy of modeling for EPA's needs and  thus has a potential impact for the
support the field has enjoyed for nearly two decades. We believe that  a clarification of process
would be beneficial to the community. Clarification must include examination, and examination
might led to new insights that  could produce new processes, or perhaps to a reaffirming of method-
ology upon which progress had been made.  In any- case, a strategy for EPA operations  in the ap-
plication of models is necessary to avoid a dichotomy of explain-predict.

    It is hoped that you will help in this process  by examining your own philosophies, comparing
and contrasting them with others in our rather small community. And that you then come pre-
pared to discuss and explore the critical issues in an open and amicable manner.

    Perhaps EPA will learn something that will allow them to:
     >  improve the design and structure of their programs;
     >  maximize the benefits  (increased knowledge) from all of our efforts;
     >  encourage the continued participation in their programs; and
     >  promote cooperation and communication.

1   The Need for Science in EPA Models                                1

2   Nature of Science                                                    8

    Textbook Science	    8

    AKuhnianView	    8

    A Psychological Interpretation	   16

    The Rest	   19

3   The New Paradigm                                                 20

    The Beginning of Explanation	   20

    Niki, Daby, and Weizwtock	   23

    The Practice of Normal Science	   25

    Early Normal Science Work   	   27
      First Class Work  	   27
      Tools Needed to Advance the Theory	   27

4   First Explanation                                                   29

    Demerjian, Kerr, Cahrert	   29
      Mechanism Development	   30
      Realism  	   31
      DKC's Rules	   33

5   First Reformulation                                                41

    Hecht, Seinfeld, Dodge	   41

6   Second Explanation                                                  50

    Bui-bin, Hecht, Whitten	   50

    Late 1970s Explicit Models	   55

    The Beginning of the First Crisis	   57

7   Second Reformulation                                                58

    The Dodge Mechanism	   58

    Falls and Seinfeld, McRae Mechanism	   60

    Carbon Bond Mechanism	   63

    Atkinson, Lloyd, Winges Mechanism	   67

    A Contrast	   69

    Mechanisms Compared	   69

8   Response to Uncertainties                                            73

    Whit ten's Hierarchy of Species for Testing	   73

    Carter's Chamber Radical Experiments	   79

9   Third Explanation                                                    82

    Atkinson and Lloyd Review	   82

    Leone and Seinfeld Master Mechanism	   84

10   The Essential Tension                                               88

    Potential and Actualization	   88

    Explanation and Prediction   	   93

    Testing and Fit	   95

    Verification and Vindication	   95


11   Methodologies                                                       98

    Mechanism Vindication  	   98

    Classes of Scientific Work	   99

    Agreement	  100

  References                                                             105

 The Need for Science
 in EPA Models
The Government Accounting Office recently told Congress,1 that the Environ-
mental Protection Agency uses air pollution models "as tools in administering
the Clear Air Act ... and often [models] are the only option available." For EPA
then, models are more than useful representations of selected real-world events.
According to the GAO, however, the foundations of these air  quality models "are
based on assumptions, approximations, and judgments, all of which affect a model's
accuracy, validity, and reliability."  Furthermore, EPA's most  reliable models, ac-
cording to the GAO, "estimate actual pollution concentrations within a range
of minus 50% to plus 200%." The  GAO said EPA officials indicated that "more
precise results are unlikely because of the limitations in the science."J The GAO
implied that improvement was possible, however, by quoting the American Meteo-
rological Society to the effect that  "the scientific community must improve model
physics, calculation techniques, and model input in order to obtain more precise
results from models."

   These GAO  'findings' led the House Subcommittee on Oversight to ask EPA
Administrator Lee Thomas to determine "which model is  the most reliable," and
to "explain the range of reliability, and identify the uncertainties of each."  Fur-
ther, the subcommittee asked Thomas: "When is the 'scientific community' going
to make the model improvements mentioned and what is the timetable?"*

   Though these questions are directed specifically to the science in air pollution
modeling, they lead on to policy issues, and thus to the overarching regulatory
function of EPA. Basil Dimitriades3 said that EPA was disturbed by the fact that

 The Need for Science
 in EPA Model*
models of "insufficiently documented validity" were being used to make costly
control strategy decisions. Further, even the "relative validities" of the various
models were difficult to document. This led Dimitriades to say that  "it is imper-
ative that model-type guidelines issued by EPA be of well-documented utility and
validity."  He added, "EPA is committed to expend the necessary resources to
meet this  requirement."  By examining and ranking approaches to evaluation—
not the models themselves—the EPA workshop described by Dimitriades is part
of that effort.

    The majority of the GAO comments were about air quality simulation mod-
els, especially dispersion models with little or no atmospheric chemical transfor-
mation processes. EPA has had a long-standing cooperative agreement with the
American Meteorological Society to review and recommend procedures to eval-
uate the performance of air quality models.4 Nothing in this  work, however, has
ever addressed chemical transformation. While the new "Guideline On Air Qual-
ity Models (Revised)"5 includes discussion on accuracy and uncertainty of models
and makes recommendation on model use, it only  contains a  recommendation to
use the Urban Airshed Model or city-specific EKMA. It directs the user's atten-
tion to a number of user's guides for these models, yet none of these discusses the
"utility, reliability, and uncertainties"  in the various available chemical mecha-

    The focus of our discussion is  the chemical component of those air quality
models intended to simulate situations in which the air pollution problems are
those caused by chemical compounds formed in the atmosphere. In the  past,
these models have been ones of ozone formation: OZIPM/EKMA models, the Ur-
ban Air Shed Model, and the Regional Oxidant Model. When alternative chem-
ical mechanisms of ozone formation have been used in air quality models, they
generally have given different predictions.6'7 Non-scientists who must by law use
these models want to know "Which prediction is right?" Because the definition
of adequate model testing—and this is one of our goals—has not been sufficiently
addressed, the policy maker is not alone; even the scientist working in the field
has difficulty dealing with this question. And unresolved issues in the use of these
air pollution models are no longer restricted to ozone formation chemistry, but
now become even more important as this chemistry is taken over for new applica-
tions. Increasingly, secondary aerosol formation and acid rain chemistry models
are being  requested for regulatory use. For the most part, these latter models also
must include the chemistry leading to ozone and other oxidant formation, because
our current understanding suggests that the formation processes of all the sec-
ondary pollutants are interlocked through the hydroxyl-hydroperoxy cycle.

                                                              The Need for Science
         	in EPA Model*	

    Thus, chemical models are being extended to new applications (new models
are being developed), and our past experience suggests that this might be the
case for some time.  Successful model guidelines then must independently eval-
uate every mechanism's fit to our current understanding, yet they must balance
this fit with the power to predict new events. This environment of different mech-
anisms and conflicting predictions is a classic confrontation of science theory and
practice; it is the everyday workplace of the scientist. While the scientist engaged
in this modeling activity has clear motives for his actions, such actions are often
not well understood by the policy maker or regulator who merely wants a tool to
conduct his business.  He asks, "When is the scientific community going to make
the changes that will give me an accurate, valid, and reliable model and what is
the timetable for  this effort?"

    This question naturally arises because one of the things the public expects
from science is  the capacity to explain things and concurrently to enable at least
scientists to predict what will happen in various combinations of circumstances.
When done 'right' in the minds of the public, the explainer not only derives the
explanation correctly, but he also gets others to understand and accept his deriva-
tion. One normally explains a phenomenon by citing the mechanism that brought
it about or sustains it, where the mechanism itself is understood by subsumption
under causal, or at least time-dependent laws.8 At present, there  is no single, gen-
erally accepted chemical mechanism to describe the atmospheric transformation
of precursors into the regulated and potentially regulated air pollutants. To many
people this is a frustrating situation. After more than a decade and a half of air
pollution modeling research, we still cannot predict  ozone to most people's sat-
isfaction.  This  sense of frustration with conflicting and contradictory results felt
by atmospheric scientists—as well as the public—has led some administrators at
EPA to question the policy of reliance on atmospheric models for predictive use.
Some have even doubted the necessity of further attempts to resolve the divisive
modeling questions within the scientific community  that created them.  Throwing
good money after bad, they seem to be saying, makes even  less sense in science
than it does in  economics, and so EPA perhaps should not seek a solution to the
problems arising from models by funding additional model research.9

    Dimitriades3 suggested that resolution of some of these  issues could be ob-
tained, at least for EPA's purposes, by having the scientific community critically
examine and rank approaches to evaluating the accuracy of gas phase chemical
mechanisms. To stimulate discussion, he suggested that there were at least three
different approaches to evaluating the accuracy of mechanisms:
     • we could compare model predictions against field data;

 The Need for Science
 in EPA Models	

    • we could compare model predictions against smog chamber data;
    • we could evaluate the accuracy of basic kinetic data and construct the
      most detailed mechanism that theory allows—this mechanism must be
     . close to the truth and we should accept it at face value because of concep-
      tual and practical problems of the experimental approaches.
But before such evaluations could be applied to models currently in use or un-
der consideration at EPA, Dimitriades suggested, evaluations themselves must
be "critically examined and ranked." Identifying in this examination all the fac-
tors that determine the  reliability, comprehensiveness, and precision of each of the
three approaches would  insure, from EPA's perspective, that the model evaluation
process would be amenable to the scientists and useful to the policymakers. We
believe that the 'scientific community' of modelers will welcome this examination
and evaluation of approaches. In fact, many have said that such an examination
is overdue and EPA's effort to provide it are applauded. Examining approaches
and determining their influential factors, however, is not equivalent to ranking the
models themselves.  And it is here, in the use of these to-be-established evaluation
procedures, that the differing perspectives  of the scientist and the policymaker
become of critical importance.

   From EPA's perspective, the best outcome of this process of examination and
ranking would be the development of a "standard mechanism evaluation  pro-
cedure (s)" and the definition of a "standard data base" needed for performing
the evaluation.  It should be recognized, however, that such an outcome may not
be useful from the scientist's perspective. That is, the scientist is concerned to
understand the world and to extend the precision and scope with which it has
been ordered. If intense scrutiny of this world reveals pockets of apparent disor-
der, these challenge him to a new refinement of his observational techniques or
to a further articulation of his theories so as to extend the precision and  scope
of his understanding.10 Often the theories-(and so the mechanisms) that arise in
this process must be in complete contradiction to the ones that have gone before
them. This is so because the new theory must be more successful  precisely where
the old theory failed. A new theory  may necessitate new experimental evidence
and a radically (so to speak) different mechanism. I believe that this is how sci-
ence progresses.  This has certain  implications for the concept that there may  be a
single mechanism evaluation procedure and its concomitant database suitable  for
all present and future theories and mechanisms.  Such a concept ignores the evolu-
tionary character that has brought about the very success it seeks to measure.

                                                              The Need for Science
	in EPA Models

    If mechanisms, necessarily different, require necessarily different  evaluations,
then how are we ever to determine rules? In fact, as will be discussed below,
some science historians believe that a lack of a standard interpretation or of an
agreed reduction to rules will not prevent a 'scientific community' from conduct-
ing research that is acceptable  to the community. This condition often contributes
directly to rapid and wide-ranging progress in our science. This is true, however,
only as long as the relevant scientific community accepts without question the par-
ticular problem-solutions already achieved.  Thus rules can become important
wherever the underlying guiding principles of the science are successfully chal-
lenged or even where the community only perceives that the principles are weak-

    If we have arrived at the point where rules are now required to sort theories
and mechanisms, then something fundamental in the science of air pollution mod-
eling is under challenge. This challenge arises, we believe, from two situations: 1)
as a result of the inability to resolve the differences among competing explana-
tions of the chemistry of paraffins and  olefins; and 2) as a result of the failure to
offer acceptable explanations of the aromatics chemistry. These are failures not
only of the experimental arts,  but also of theoretical practice. For these reasons,
Dimitriades was certain that achieving his goal would "hinge upon the resolu-
tion of complex scientific issues." He said, "This is the normal result of the fact
that evaluation of  chemical mechanisms is fraught with unresolved issues, most
of which, in the lack of convincing experimental evidence, have acquired a sub-
jective, even philosophical character."  This, he suggested, is likely to lead to an
unavoidable clash  between those believing in the unchallengeable validity of real
world data and thus for whom experimental evaluation is  the only option avail-
able, and those believing that  the real  world is hopelessly  complex and thus advo-
cate that theory shall be used as  "gospel." As we shall see, however, the nature of
the problem is not quite this black and white.

    I agreed to assist Dimitriades by describing and discussing various approaches
to model evaluation. Upon reflection on the issues, however, I realized that much
of what we want to discuss amounts to a "meta-modeP of chemical mechanism
modeling, that  is,  "What is  it  that modelers do when they develop and  then test
a model?"

    To begin this discussion though, it  is necessary to agree on the definitions of
terms we all use in developing, describing, and testing our theories and mecha-
nisms. For the substance of this paper let us agree on these

 The Need for Science
 in EPA Modeii
evidence:  empirical generalizations, restricted in their application, arrived at
      through observation, and intended to prove or provide grounds for belief;
      easily refuted with other evidence. Example:  the sun rising.
theory:  views about the ultimate structure and lawlike qualities of the world.
      They make some claim for  permanence, and do not admit simultaneous
      consideration of other proposals; that is, they are mutually exclusive. Ex-
      ample: the theory of evolution.
hypothesis: tentative candidates for theory;  something not proved but taken for
      granted for the purpose of argument or inference. Example: electrons are
      in orbits.
auxiliary statements: simplifying assumptions made to allow application of a
      theory to  a given situation. Example: To apply Newton's theory of univer-
      sal gravitation we might assume for the purposes of calculation: no bodies
      exist except the sun and the earth; these exist in a hard vacuum; they are
      subject to no forces except  mutually induced gravitational forces.  From the
      conjunction of the theory and these auxiliary  statements we can predict
      Kepler's observations.
model: a set of assumptions about some object or system drawn from exist-
      ing theory, hypothesis, auxiliary statements, and experimental evidence.
      Though a model may be associated entirely with a single theory, it makes
      no claim for isolated permanence; that  is, it presents itself as one set of de-
      scriptive simplified assumptions in a universe  of many such sets.  Example:
      Bohr's model of the atom.
mechanism: That inner structure or composition attributed by a model to the
      object or system it describes, intended  to explain various properties exhib-
      ited by that object or system.  Example: the mechanism attributed by the
      Bohr model to the hydrogen atom to explain radiation of discrete wave-
      lengths emitted  when hydrogen is excited.
truth: correspondence with experience, facts, or reality (whatever these may be).
      It need not be a statement  about true-belief or about permanence, and it
      is not to be confused with a) coherence theory of truth in which consis-
      tency equals truth; b) instrumentalist theory of truth in which usefulness
      equals truth; or  c) evidence theory of truth in which "know to be true" is
      taken for  "true."
facts: the  assertion of something  as existing or done; (facts are sometimes things).

                                                              The Need for Science
 	in EPA Models

    I will present, for discussion purposes, an abbreviated history of how I under-
 stand the process by which mechanisms have been "developed and verified" over
 the last 15 years by citing books, journal papers, progress reports, EPA reports,
 conversation, and even personal experience. From this description of practice,
 I hope that we might obtain insight into the essential underlying assumptions,
 methodology, and philosophy of our field, and that we might find a common-
 alty of approach and discover particularly revealing examples of practice. Some
 philosophers of science believe that scientists work from patterns of behavior
 (mental constructs) "acquired through education and through subsequent expo-
 sure to the literature often without quite knowing or even needing to know what
 characteristics have given their methodologies and  approaches acceptable status
 in the community that they practice in. If they have learned such abstractions at
 all, they show it mainly through their ability to do successful research."lo This
 is not to say that there are specific "rules of the game," but it is meant to imply
 that there are in every field a set of recurrent and quasi-standard illustrations of
 various theories in their conceptual, observational,  and instrumental applications.
 By studying these and by practicing with them, the members of the correspond-
 ing community learn their trade.

    Because all modelers whom I know consider themselves to be scientists, this
 "meta-model of modeling" must bear some kinship to current philosophies of sci-
 ence. Therefore it will be useful to look first at work in this area for insight and
 guidance. The application of these more general descriptions of scientific activ-
 ity can provide a framework—and provide us with a ready-made vocabulary—to
 organize our thoughts on the process of mechanism model development and evalu-
 ation rather than on the content of the modeling work. I am hopeful that in turn
 we can use these outlines from other sciences and from the philosophical reflec-
 tions on them to advance both the content and practice in our own  field. This
 goal is admittedly beyond drawing up guidelines for model evaluation, but I hope
 to show that questions about the nature of modeling and testing strike to some-
thing of core importance in all of the sciences. These are especially  important in
our field  where models are "more than useful representations of real-world events"
 and are "tools for regulation."

                                     These are the days of miracle and wonder,
                                     and don't cry,  baby, don't cry.  Don't Cry.
                                                                 Paul Simon
                                                            Boy in the Bubble

 Nature  of Science
Textbook Science

Prior to about 1960, a textbook image of science involved some combination of
the following points (taken whole from Hacking's book Scientific Revolutions14):

   1. Realism. Science is an attempt to find out about one real world. Truths
      about the world are true regardless of what people think, and there is a
      unique best description of any chosen aspect of the world.
   2. Demarcation. There is a pretty sharp distinction between scientific theories
      and other kinds of beliefs.
   3. Cumulative. Although false starts are common enough, science by and
      large builds on what is already known.
   4. Observation-theory distinction. There is a fairly sharp contrast between
      reports of observations and statements of theory.
   5. Foundations. Observation and experiment provide the foundations for and
      justifications of hypotheses and theories.
   6. Theories have a deductive structure and tests of theories proceed by deduc-
      ing observation-reports from theoretical postulates.
   7. Scientific concepts are rather precise, and the terms used in science have
      fixed meanings.
   8. The unity of science. There should be just one science about the one real
      world. Less profound sciences are reducible to more profound ones.  Psy-
      chology is reducible to biology, biology to chemistry, chemistry  to physics.

A Kuhnian View	Nature of Scitnce

A Kuhnian View

In 1963, Thomas Kuhn, trained as a theoretical physicist, but later a professor
of History and Philosophy of Science at MIT, published The Structure of Scien-
tific Revolutions10 in which he presented an alternative picture of science. This
work has been described as one of the three essential works in the field and has
itself revolutionized current philosophical thought about the nature of science.
Again from Hacking's book, I quote his summary of Kuhn's alternative view  of
   A. Normal science and revolution.  Once a specific science has been individu-
      ated at all, it characteristically  passes through a sequence of normal science-
      crisis—revolution—new normal science. Normal science is chiefly puzzle-
      solving activity, in which research workers try both to extend successful
      techniques, and to remove problems that exist in some established body of
      knowledge. Normal science is conservative, and its researchers are praised
      for doing more the same, better. But from time  to time anomalies in some
      branch of knowledge get out of  hand, and there seems no way to cope  with
      them.  This is a crisis. Only a complete rethinking of the material will suf-
      fice, and this produces revolution.
   B. Paradigms. A normal science is characterized by a 'paradigm' J(PAIR-uh-
      dime), a pattern]. There is the  paradigm-as-aehievement. This is the ac-
      cepted way of solving a  problem which then serves as a model for future
      workers.  Then there is the paradigm-as-eet-of-shared-values.  This  means
      the methods, standards, and generalizations shared by those trained to
      carry  on the work that models itself on the paradigm-as-achievement.  The
      social unit that transmits both  kinds of paradigm may be a small group
      of perhaps one hundred or so scientists who write or telephone each other,
      compose the textbooks,  referee  papers, and above all discriminate among
     „ problems that are posed for solution.
   C. Crisis. The shift from one paradigm to another through revolution does
      not occur because the new paradigm answers  old questions better. Nor
      does it occur because there is better evidence  for the theories associated
      with the  new paradigm than for the theories found in the old paradigm.
      It occurs because the old discipline is increasingly unable to solve pressing
      anomalies. Revolution occurs because  new achievements present new ways
      of looking at things, and then in turn create new problems for people to
      get on with. Often the old problems are shelved or forgotten.

Nature of Science	A Kuhnian View

   D. Incommensurability, Successive bodies of knowledge, with different paradigms,
       may become very difficult to compare. Workers in a post-revolutionary pe-
       riod of new normal science may be unable even to express what the earlier
       science was about.  Since successive stages of a science may address dif-
       ferent problems, there may be no common measure of their success—they
       may be incommensurable.

   E. Noncumulative science.  Science is not strictly cumulative because paradigms—
       in both senses of the word—determine what kind of questions and answers
       are in order.  With  a new paradigm old answers may cease to be important
       and may even become unintelligible.

   F. Gestalt switch. 'Catching on' to a new paradigm is a possibly sudden tran-
       sition to a new way of looking at  some aspect of the world. A paradigm
       and its associated theory provide different ways of 'seeing the world.'

    The contrast between  the image presented in points 1-7 and Kuhn's points
A-F is not so much  in specifics as in a different concept of the relation between
knowledge and its source.  Kuhn said,
      Traditional discussion of scientific method have sought a set  of rules that would per-
      mit any individual who  followed them to produce sound knowledge. I have tried to
      insist, instead, that, though science  is practiced  by individuals, scientific knowledge
      is intrinsically a group product and  that neither its peculiar  efficacy nor the manner
      in which it develops will be understood without reference to the special nature of the
      groups that produce it.

    Thus Kuhn's terms deal more directly with the scientist than their science,
with the theorists as a group than with their stated theories.  He has examined
science and its community. He  says, for  example,
      The hypotheses of individuals are tested, the commitments shared by his group being
      presupposed; group commitments, on the other hand, are not tested, and the process
      by which they are displaced differs drastically from that involved in the evaluation of

Reflecting on his book 10  years later, Kuhn said,15
      The term "paradigm" enters in close proximity, both  physical and logical, to the
      phrase "scientific community."  A paradigm is what the members of a scientific com-
      munity, and they alone,  share. Conversely, it is  their  possession of a common para-
      digm that constitutes a  scientific community of a group of otherwise disparate men.

          One thing that binds the  members of any scientific community together and
      simultaneously differentiates them from the members  of other apparently  similar
      groups is their possession of a common language or special dialect.  These essays
      suggest that in learning  such a language, as they must to participate in their com-
      munity's work, new members acquire a set of cognitive commitments that are  not, in


A Kuhnian View	            Nature of Science

      principle, fully analyzable within that language itself. Such commitments are a con-
      sequence of the ways in which the terms, phrases, and sentences of the language are
      applied to nature, and it is its relevance to the language-nature link that makes the
      original narrower sense of "paradigm" so important.

    But how is this community formed and maintained? Hacking said,
      The paradigm-as-achievement is commonly taught, not by giving axioms and making
      deductions, but by giving examples of solved problems and then using exercises in
      the textbook to get the apprentice to catch on to the method of problem solution.
Often this early 'catching on' results in the first of many intellectual Gestalt switches
that a scientist experiences  in his career and prepares him to expect such events
throughout his work.  Many of you may well .recognize the form of your own train-
ing in science in Kuhn's paradigm-as-achievement definition. For me, Kuhn's ex-
planation of the history of science education  is striking valid.

    But mere identity of a community's model and the world of events is not the
aim of science. Kuhn described the aim of science as "to invent theories that ex-
plain observed phenomena and that do so in  terms of real objects, whatever the
latter phrase may mean."J5 Kuhn therefore  is a realist.  For him, however, it is
      a reality for which we construct different representations. Realism  is  compatible with
      incommensurability, for representations arising from attempts to answer different
      problems need not mesh well  with each other—perhaps the world is too complicated
      for us to get one comprehensive theory.  Even if our theories are plural and incom-
      mensurable, we could still think of them aiming at different aspects of one totality.

For this reason, Kuhn has been described as  a "constructionist," one who con-
structs many realities to account for many views.  Hacking says,  "... the spirit of
Kuhn's approach runs counter  to  the unity of science.  There is a plurality rather
than  a unity in representations of the world,  and successive representations ad-
dress different problems which  need have very little common subject matter."
But a careful reading of the passage cited above reveals  that Kuhn has  already
dealt with this problem and so refuted  the potentially disastrous consequences of
scientific plurality; what seems plural, he says, are only aspects of one totality.
We are comfortable with Kuhn's delineation of aspects of one reality if only be-
cause it means that the last 15 years of air pollution modeling work  (in which we
have  taken many different assumptions and arrived at contradictory conclusions)
have  not been describing different realities. If Hacking were right and Kuhn's
view  necessitated  construction  of many realities, then scientific conclusion could
never be made. So if different mechanisms are describing only  different aspects
of one reality, then even completely independent evaluation must say something
about the mechanism, the theory  that generated it, and the fit of both to reality.


Nature of Science	A Kuhnian View

Note that I do not yet accord any weight to comparison of evaluations, only state
that independent evaluations would be valid measures of all mechanisms' fit to
the one reality.

    Because, as we have seen above, a mechanism often reflects only the new the-
ory (hypotheses and auxiliary statements) that was directly responsible for its
generation, and not much of the more commonly assumed theory, it is  necessary
to examine "theories" before looking  at approaches to evaluation. Kuhn says that
he does "not believe that there are rules for inducing correct theories from facts."
Instead he views theories
      as imaginative posits, invented in one piece for application to nature. And though
      we point out that such posits can and usually do at last encounter puzzles they can-
      not solve, we also recognize that those troublesome confrontations rarely occur for
      some time after theory has been both invented and accepted.
In other words, for Kuhn, theories have a natural obsolescence.  They cannot be
expected or forced to survive past the point where they contribute in a meaning-
ful way to  the puzzles posed by the paradigm that directs any particular scien-
tific venture. It is important to note, however, that Kuhn's definition requires at
least some time for reflection on the theory and that during this time the develop-
ing theory is not directly challenged by troublesome confrontations.  This aspect
of slow theory-growth and protection by the scientists who brought it about has
been explored in great depth by another philosopher.16

    For Kuhn, there is an  "intimate and inevitable entanglement of scientific ob-
servation with scientific theory."15 Continuing from Hacking's book:
      Kuhn rejects a sharp observation-theory distinction because the things that we no-
      tice, and the ways we see  or at least  describe them, are in large part determined by
      our models  and problems. There  is no 'timeless* way in which observations support
      or provide foundations for theory.  The relations between observation and hypothesis
      may differ in successive  paradigms. Hence there is no pure logic of evidence  or even
      of testing hypotheses, for each paradigm, in its own day, helps fix what counts as
      evidence or test.
Just as Kuhn's definition of incommensurability and noncumulative science helped
to justify my claim for independent mechanism evaluation, so here the  recognition
that there  is no observation-theory distinction, leads to significant conclusions.
In this case, it is that the process of mechanism evaluation—and so comparison
of the theory with facts—is  not static, but must evolve along with the  scientific
content of air pollution chemistry.  "

    There are several issues  tied to this notion of evolutionary evaluation the con-
clusions to which have been directing both the theory and practice of modeling


A Kuhnian View	       Nature of Science

working since 1975.  Here I will describe some of these issues—puzzle-solving, ex-
plicit standards, testing, and fit— and point out the effects of their conclusions.
Later (in Chapters 3-9) I will introduce specific instances where decisions about
testing  and fit have considerably changed how modelers model and how they treat
chamber data.  In Chapter 10, I will talk about how these decisions and their con-
sequences have affected progress  in our field. Lastly, I will draw together these
assumptions, definitions, and conclusions into suggestions for  model evaluation I
hope will be the outcome of my work here.

    The business of science, Kuhn says, is puzzle-solving. Though this may sound
demeaning, Kuhn certainly intends no offense; he says that this process has re-
quired the highest scientific talents. In fact, he choose the metaphor because of
the striking similarity between solving jigsaw puzzles and the search for expla-
nations of phenomena and the ways to harness nature. Kuhn contends that this
puzzle-solving activity is a vital part of science so long as the puzzles are interest-
ing and the solutions are accepted by the community. In their day-to-day work,
scientists do not try  to disprove the theories that direct their  research, nor do
they seek unexpected and unpredicted results.  So, according to Kuhn, although
science  proceeds by the extraordinary episodes of shifts of professional commit-
ments, paradigm revolutions, it cannot advance directly from one revolution to
the next. In fact, since Kuhn's structure for science denies linear progression com-
pletely,  it would not admit progress from any paradigm to any other in any di-
rect manner at all. That we have no new paradigm 15 years after the  proposition
of HO -attack does not mean that air pollution science has failed to progress in a
Kuhnian sense. Rather, the puzzle-solving done in those 15 years to expand and
ratify, or modify the theories and auxiliary statements of HO -attack has been the
success  promised by  that paradigm. And it was that  promise, Kuhn would say
and I would agree after looking at the actual documentation of  those early years,
that led us to accept the HO -attack theory in the first place.

    While others have merely noted an asymmetry with regard  to testing theories,
Kuhn takes a special view of the process. The asymmetry can be stated as:  it
is a logical truism that a scientific theory cannot be shown to apply successfully
to all its possible instances, but  it can be shown to be unsuccessful in particular
applications. Some have used this concept to suggest that scientists directly test
theories and reject them on the basis of experiment. Karl Popper, for example, in
Logic of-Scientific Discovery written in 1959, said
     A  scientist, whether theorist  or experimenter, puts forward statements, or systems of
     statements, and tests them step by step. In the field of the empirical sciences, more

Nature of Science	A Kuhnian View

      particularly, he constructs hypotheses, or systems of theories, and tests them against
      experience by observation and experiment.

Kuhn says Popper is historically mistaken, and that his mistake "misses just that
characteristic  of scientific practice which most nearly distinguishes the-sciences
from other creative pursuits."  Kuhn says
      There is one  sort of "statement" or "hypothesis" that scientists do repeatedly subject
      to systematic test. I have in mind statements of an individual's best guesses about
      the proper way to connect his own research problem with the corpus of accepted sci-
      entific knowledge. ... He may conjecture that a newly discovered spectral pattern
      is to  be understood  as an effect of nuclear spin.  The next steps in his research are
      intended to try out  the conjecture or hypothesis.  If it passes enough or stringent
      enough tests, the scientist has made a discovery or has at least resolved the puzzle
      he had been set.  If  not, he must  either abandon the puzzle entirely or attempt to
      solve it with  the aid of some other hypothesis.  ... Tests of this sort are a standard
      component of what  I have called normal science. In no usual sense, however, are
      such  tests directed to current theory. On the contrary, the scientist must premise
      current theory as the rules of his game.  His object is to solve a puzzle, preferably
      one at which others have failed, and current theory is required to define that puzzle
      and to guarantee that, given sufficient brilliance, it can be solved.  ... the practi-
      tioner must often test the conjectural puzzle solution that his ingenuity suggests.
      But only  his  personal conjecture is tested.  If it fails the test, only his own ability,
      not the corpus of current science  is impugned. ...  in the final analysis it is  the indi-
      vidual scientist rather than current theory  which is tested.

    There comes  a time, of course, when a paradigm's potential for suggesting
new and interesting puzzles has been fully actualized. At times of such paradig-
matic crisis, Kuhn claims, scientists are reluctant to give up the theories and their
auxiliary support statements which have allowed them to make progress. In fact,
it is my experience that scientists often work hardest to protect a theory (or even
a paradigm) that is devoid of potential.  In such cases and even throughout the
practice of normal science when theories and paradigms have reached their ob-
solescence, scientists must confront the anomalies left unexplained (even maybe
unpredicted) by the old for only the hope  of progress with the new. Kuhn is quite
clear, though, that such change is  never brought  about by the'direct refutation
of theory by physical evidence. He even extends this argument to apply to the
practice of all normal science.  Thus, Kuhn denies that scientists practice "falsifi-
cation" of theory—showing an instance wherein the theory fails in an attempted
application, and hence to argue a reason to compel assent from any member of
the relevant professional community as to the theory's falseness. He says,
      Where a whole theory ... is  at stake, arguments are seldom so apodictic.  All experi-
      ments can be challenged, either for their relevance  or their accuracy. All theories can
      be modified by a variety of ad hoc adjustments without  ceasing to be, in their main
      lines, the  same theories. It is important, furthermore,  that this should be so, for it


A Kuhnian View  	Nature of Science

      is often by challenging observations or adjusting theories that scientific knowledge

    Kuhn explicitly realizes that a Gestalt switch is needed to accept a new the-
ory, but is difficult for scientists to make. He suggests that scientists, "though
they may begin to lose faith and then to consider alternatives, do not renounce
the paradigm that has led them into crisis." They do not treat anomalies as counter-
instances.  If an anomaly cannot be assimilated within the existing theory,  then
the scientist deletes the problem (discounts its existence). And the shift is even
more difficult to make if the theory under attack is actually the paradigm, for
once a theory has achieved the status of a paradigm, it is declared invalid only
if an alternative candidate is available to take its place.  Kuhn declares that "No
process yet disclosed by  the historical study of scientific development at all re-
sembles the methodological stereotype of falsification by  direct comparison with
nature." Instead, Kuhn  asserts  that
      ... the act of judgment that leads scientists to reject a previously accepted theory
      is always based upon more than a comparison of that theory with the world.  The
      decision to reject one  paradigm is always simultaneously the decision  to accept an-
      other and the judgment  leading to the decision always involves comparison of both
      paradigms with nature and with each other.*
                                                        Truths turn into dogmas
                                                  the moment they are disputed.

Nature of Science	A Piychologieal Interpretation

A Psychological Interpretation

Some philosophers and scientists strenuously object to Kuhn's description of sci-
entific progress by saying that it is far too  psychological—that it gives the indi-
vidual more power and importance than the theory.  For them Kuhn has said that
science is only temporary agreement within the community of scientists. In fact,
Kuhn's image of how we practice science is very similar to descriptions arising
in sociological fields, for example, one "meta-model" for human communication
and interaction is neuro-linguistic programming (NLP).17 This model was devel-
oped to explain how some therapists are more successful than others at getting
people to change. It was described by Gregory Bateson as "making explicit  the
syntax of how people avoid change." According to this system, we live in one real
world. We do not function directly in this  world, however, but through a series
of "maps" of the world that we  develop from our experiences (of course, not all
maps are verbal). Maps, like all models, are created by three general process:
    •  Generalization. To  simplify our representation, we treat objects that  are
      different in the real world as  if they were the same in our map.
    • Deletion. To simplify our representation further, we omit from our map
      objects or details in the real world that we think are not important or we
      do not have a representation in our  map of objects in the real world of
      which we have no awareness.
    • Distortion. To fit some facts into our map, we modify the representation of
      real world objects so as to be consistent with our map. Something must be
      distorted to make any map.

   A possible consequence of this mapping is that one confuses  a useful map
with the  territory it represents and acts as if his map were the world. Success-
ful maps  are more likely to bring about this mistake precisely because they are
more useful  to the map maker.  A person with a highly distorted map or other-
wise impoverished representation (usually formed under stress and without ade-
quate adult  information) often has difficulty coping with the real world; he never
sees the options available to him, and so sometimes acts in what seems to others
as inappropriate ways.

   For example, a person complains that he is not worthy of love and that his
wife does not love him.  Since he was not loved in his childhood and since he is
not loved now, it must be  that he is not worthy of love (generalization) and  he
can prove that because his wife  does not love him (the map  is real).  When it is

A Ptychological Interpretation	Nature of Science

pointed out to him that frequently his wife has been observed making loving re-
marks to him, he replies that he does not recall any such statements (deletion).
When they are pointed out in real-time, he then distorts what he hears by saying,
"But she is not sincere—she only wants something."

    Maps more free from distortion, those sufficiently rich in detail, provide op-
tions precisely in those places where others have failed to include all the facts.
Such facts, present but not accounted for by some, allow the person with a rich
map to chose his response more appropriately, and so to determine the outcome
of an event in a way that is more pleasing to him.

    A person having difficult life-experiences can often benefit by modifying their
maps. It is rare that this person, without careful training, would be capable of
"stepping outside" of his map, and thus he usually requires the help of a thera-
pist.  According to the NLP model, a successful therapist first deduces the struc-
ture of a person's map through his actions and statements (a form of Kuhn's
puzzle-solving). Next the therapist creates situations that cause the person  to
confront the flaws in their maps (non-functional generalizations, deletions, and
distortions)  and eventually to resolve the discrepancies. The map adjustment is
usually accomplished by a 'Gestalt switch' that suddenly allows the person to see
solutions to his problems.  Successful therapy thus results in an richer map and
more options for coping in the world, and therefore, to have more successful out-
comes in real-life situations.

    Kuhn's concept of the  paradigm can be easily and usefully translated into
these terms.  Kuhn's map is scientific realism, the idea that the world is unitary
and explainable.  Progress  is made within a paradigm, Kuhn would say, in one
sense by the scientist who  generalizes information about hypotheses and obser-
vations; this is problem solving. In a second sense, the same researcher may also
group individual representations to generalize more precisely about their behavior
or content: this is the basis for prediction. These together constitute the business
of normal science.

    Similarly,  normal science requires the scientist to delete information and nar-
row his field if he is to make progress at all. Kuhn says,
      Though science clearly grows in depth,  it may not grow in breath as well. If it does
      so, that breath is manifest mainly in the proliferation of scientific specialties, not in
      the scope of any single specialty.

Indeed, Kuhn would require that the working scientist  delete from his map all po-
tentially interfering information. He calls narrowmindedness a "characteristic" of


Nature of Science	A Psychological Interpretation

scientific communities; it is the concern with detail that ultimately allows com-
parison of theory and observation on the level necessary to solve the puzzles of
normal science.

    The third map-generating process, distortion, is yet more powerful and more
interesting in two ways: 1) as an essential, directing behavior that creates  the
map; that is, some things must be distorted to make any map; and 2) as a func-
tional result of the misapplication of the first two processes as when overgeneral-
ization or excessive deletion results in a map untrue to its territory.

    In the first sense, distortion allows the scientist to  explain arrestingly novel
events  by distorting their place (significance) on his map (in his paradigm); thus
allowing him to solve puzzles and thereby strengthen the paradigm (map).  In the
second, destructive sense, distortion prevents a scientist from making the break to
a new paradigm even when the existing one is repeatedly bettered at explanation
and prediction. He has generated his map in response to input taken directly or
indirectly from the  actual world of events. This map has been successful at the
puzzle-solving science has required and this success has re-affirmed his commit-
ment to the map. And, as in therapy, this is the sort of distortion that  requires
a Gestalt switch for the individual believer to abandon his map. Such distortion
is especially disastrous to a practicing scientist in a community undergoing crisis
because at that point he is in danger of mistaking his map, his distorted map, for
the world he tries to explain.  The new map, the old believer claims, cannot be
closer to the world of events because his existing map  is that world. And he be-
lieves in that identity because his map has been validated through repeated suc-
cess in explanation  and prediction.  Thus, in the terms of our comparison, Kuhn's
scientific realism  has become in the distorted map both scientific and physical re-

The Rest	Nature of Science

The Rest

Now that we have agreed on terms and laid the philosophical groundwork for a
discussion of what it means to theorize, test, and fit reality we can next show
where these terms and constructs fit in a history of air pollution modeling  and
evaluation. It was necessary to define these terms and ideas because they are ones
that we all use everyday. Because of that familiarity they have implicit meanings
for each reader which must be reconciled if we are to have meaningful conversa-
tion.  Similarly, it is necessary now to  recount a history of mechanism develop-
ment using the terms and concepts described so that the philosophy I have de-
scribed will have a real connection to our own science. Lastly, it will be necessary
to draw out of this interpretation of our past just how we might proceed.
                                         It is the customary fate of new truths
                                                          to begin as heresies
                                                    and end as superst/t/ons.
                                                              —T.H. Huxley

The New  Paradigm
To understand the paradigm-as-achievement we must understand what has been
achieved. Kuhn said that science historians (or scientists themselves) can agree
in their identification of a paradigm without agreeing on, or even attempting to
produce a full interpretation or rationalization of it.  What follows must be the
product of my own generalization, deletion, and distortion. Gerald Weinberg,
writing18 about his profession, system analysis, said,
  "If we study history even for a short while—long enough to read more than
   one source—we learn perhaps the most important lesson:

                There is no one truth about what happened."

Some of you, no doubt, will have a different specific image of what happened;
other works, not mentioned here, will have more relevance for you. Nevertheless,
I think the historical examples I have selected illustrate many of the essential fea-
tures of the science of model development, therefore, they are useful for our pur-

The Beginning of Explanation

Prior to 1970, the science of air pollution chemistry was lacking in major respects;
it was in a pre-paradigm stage. In the period 1961-1970, there was no generally
accepted view about the exact causes of the chemical transformations that pro-
duced secondary air pollutants. The primary source of theoretical information
was Phillip Leighton's 1961 book Photochemistry of Air Pollution.19 Many re-
searchers have commented on how surprised they were to find that Leighton had
discussed some topic that later became central to the paradigm. Kuhn says,
     In the absence of a paradigm or some candidate for paradigm, all of the facts that
     could possibility pertain to the development of a given science are likely to seem

Th« Beginning of Explanation	        Tht N«w Paradigm

      equally relevant. As a result, early fact-gathering is a far more nearly random activ-
      ity than the one that subsequent scientific development makes familiar. Furthermore,
      in the absence of a reason for seeking some particular form of more recondite infor-
      mation, early fact-gathering is usually restricted to the wealth of data that lie ready
      to hand.  The resulting pool of facts contains those accessible to casual observation
      and experiment together with some of the more esoteric data retrievable from  estab-
      lished crafts like medicine. Because of this ... technology has often played a vital
      role in the emergence of new sciences.

    Some of this pool of facts included Calvert and Pitt's book Photochemistry
and the smog chamber work of Paul Altshuller and his colleagues. Atkinson ft al.
suggested20 that, in the case of air pollution chemistry, the technological source of
information was combustion chemistry.  Chain reaction theory was developed in
1934 and the hydroxyl radical (HO) was proposed as an  important intermediate
in high temperature oxidation systems.

    In the air pollution field, a major anomaly arose at the juncture of theory and
observation when it was noted that the  high rates of conversion and the  large O3
production demonstrated in smog chambers could not be accounted for using the
prevailing chemical theories. Although the air chemistry textbooks and papers
were full of chemical reactions, rate constants were not commonly given  and reac-
tions  rarely appeared in clusters of more than three or four. Only a few  air pol-
lution scientists could successfully argue the consequences of a chemical reaction
sequence that was longer than three steps. Their arguments sometimes were so
confusing that the listeners were totally lost and unable  to engage in discussion.

    Along with a large amount of other potential radical chemistry, Leighton had
described HO  and had  postulated several  reactions:  "Hydroxyl radicals are pro-
duced by the photodissociation of hydrogen peroxide and possibly by the pho-
todissociations of nitrous and nitric acids." Further, he even had suggested the

                           HO2 + NO	> HO  + NO2

in the middle of a list of four other potential HO source  reactions, saying, how-
ever,  "Unfortunately, little may be said  as to the rate at which hydroxyl radicals
are produced in polluted air."  Leighton had described what we now know to be
key reaction in alkoxyl radical reactions,

                          CH3O  + 02	> HCHO + HO2

Th« N«w Paradigm	The Beginning of Explanation

as being "purely speculative."  He had concluded that "... certainly there is no
evidence that the reactions of hydroxyl radicals with nitric oxide or nitrogen diox-
ide are of any importance in air." Thus the HO radical  was only one of several
ideas, part theory, part observation competing with each other in an effort to
explain  early smog chamber results. As such they had the suggestive force of a
pre-paradigm theory or observation, but not the coercive power of a paradigm, al-
ready accepted by many. Nevertheless, for a young researcher entering the field,
this was hardly a cause to investigate HO' reactions.  As was said above, "the
things we notice, and the ways we see or at least  describe them, are in a large
part determined by our models."

   As another example, the first photochemical smog mechanism for use in a
computer program that I recall seeing was Lowell Wayne's  15 or 16-step mecha-
nism which appeared in  one of the blue covered Air Pollution Foundation Reports
about 1967.  This mechanism did not use HO , and to account for the observed
rates, Wayne assumed, as others had, that organic radicals reacted with oxygen
to made ozone,

                           RO2 + 02 —-> RO + Os

Although this approach allowed Wayne to predict Os production, it was met with
little acceptance, because the reaction must be endothermic,19 and tt... many of
the proposed steps are not elementary reactions,  and they have little or no physi-
cal meaning."21

   Using Kuhn's terms, there  was essentially a "revolution" during 1969-70—'a
new way of seeing the world'—and it has led to today's normal science in chemi-
cal mechanism development. Greiner22 made the  first determinations of the rate
constants for HO with alkanes  in 1967.  Based on his results, he postulated that
these  reactions would be important in photochemical smog chemistry. Two groups
soon after put forth the supposition that it was a chain process involving HO -HO2
that was responsible for  the "excess" rate of hydrocarbon consumption and NO
oxidation. In a paper presented at "Chemical Reactions in Urban Atmospherers"
in Michigan in 1969, Heicklen,  Westberg, and Cohen23 "suggested that HO', pos-
tulated  as intermediates in the reaction of hydrocarbons with NO, serve as chain
carriers  to oxidize NO to NOj in the presence of CO."  At the same meeting in a
discussion, Bernie Weinstock (who, according to Joe Bufalini, was a combustion
chemist at Ford Motors), described how he calculated the chain length in propy-
lene/NOx smog chamber experiments performed by Altshuller. Less than  a year
later,  Steadman, Morris, Daby, Niki, and Weinstock described "The Role of OH


Niki, Daby, and Weinstock    	The New Paradigm

radicals in Photochemical Smog Reactions."  Others, too, were rushing to print
with ideas based on the HO -chain concept (see the next chapter, for example).
The rapid conversion to the "HO -paradigm" suggests that Gestalt switches were
made quickly by many scientists practicing in somewhat different fields. And the
faith required to make such a switch was- great because no one had ever deter-
mined that HO  radicals existed in the ambient air.

    The mere recognition that an HO -chain was involved, however, was not suffi-
cient to serve as a paradigm.  Demerjian, Kerr, and Calvert*J say, "The first  at-
tempts to describe the individual chemical steps important for the irradiation of
mixtures containing propylene, NO, and NOj in air were reported by Westberg and
Cohen (1969) and Wayne and Ernest(1969)."  The former, consisting of 71 reac-
tions, was an Aerospace Corp. publication, and the latter, consisting of 40 ele-
mentary reactions, was presented at an APCA meeting. Demerjian et al. go  on to
say, "However some of the 'elementary' reactions  ... involved such extreme struc-
tural rearrangement or unfavorable energy requirements for reactants to form
products, that they can not be important in real  systems." Therefore their so-
lutions to the puzzle were not acceptable to the larger community and they were
not candidates for a paradigm.

Niki, Daby, and  Weinstock

By 1970-71, the HO -HOj chain concept was sufficiently developed that Niki, Daby,
and Weinstock had produced  a paper, "Mechanisms of Smog Reactions," that
solved the puzzle in an acceptable manner. This paper, therefore, received wide
circulation2* before eventually being published in 1972.K

    Over 150 elementary reactions had been examined and 60 elementary reac-
tions were selected for the mechanism of about 40 species. The mechanism was
generated under the HO -chain theory. But the theory at the heart of this para-
digm needed auxiliary assumptions for the model to explain adequately and  pre-
dict correctly. So the mechanism proposed was HO  attack on the  propylene dou-
ble bond, leading to the formation of an hydroxy-alkyl radical which subsequently
reacted with oxygen to produce an hydroxy-alkyl peroxy radical.  This radical re-
acted  with NO producing  NOj, HO2, and aldehydes. Photolysis of the  aldehydes
supplied a continuous  source of new H02 radicals which in turn  produced HO' rad-
icals.  Hydroxyl radicals were  lost continuous by reaction with NOj to form nitric
acid. Other radical-radical reactions also terminated the chain. The only source
of O3 in this mechanism was the photolysis of NOj followed by the reaction of O +

The New Paradigm	Niki, Daby, and Wtin«tock

0; to make Oj. Each reaction had an assigned rate constant supported by multi-
ple literature citations. The work included 123 references.  That these rate con-
stants values were so prevalent in the literature is yet another proof that the field
was in a pre-paradigm stage.

   The model was applied to a propylene/NOx smog chamber experiment and
could account for the principle features of propylene chemistry. Further, the model
was used to illustrate the effects of changes in model inputs and experimental
conditions and a mechanistic interpretation was given for each outcome. This
fulfills the explanation function of a paradigm. These examples of the model in
practice helped to establish the basic acceptance of the mechanism concepts.
"Theory" had been used to construct a complex description that not only ex-
hibited the right kinds of response, it was consistent with a large body of older
scientific information, and it had the "feel" of a good solution (i.e., there was a
certain sense of congruity or recognition). Thus the model as theory might also
predict accurately.  Its predictions only had to resemble the smog  chamber results,
they did not have to fit it perfectly. Further, problems that could attract a range
of researchers were clearly obvious from the work.

   This work exemplified the new paradigm-as-achievement. Kuhn says,
      We must recognize how very limited in both scope and precision a paradigm can
      be at the time of its first appearance. Paradigms gain their status because they are
      more successful  than their competitors in solving a few problems that  the group of
      practitioners  has come to recognize as acute. To be more successful is not, however,
      to be either completely successful with a single problem, or notably successful with
      any large number. The success of a paradigm... is, at the start, largely a promise of
      success discoverable in selected  and still incomplete examples.
Putnam26 says of the new paradigm, "It is important that the application—say,  a
successful explanation of some fact, or a successful and novel prediction—be strik-
ing; what this means is that the success, is sufficiently impressive that scientists—
especially young  scientists choosing a career—are led to try to emulate that suc-
cess by seeking further explanations, predictions, or whatever on the same model."

   Prior to the use of such computer models, theoretical air pollution chemistry
was mainly descriptive. After 1971, we had a clear  example of how it could be
quantitative. We had in the paradigm of HO -chain and its attendant auxiliary
hypotheses, coupled with the solution possibilities of the computer, the potential
to explain, with increasing precision, exactly how atmospheric chemical processes

Th« Practice of Normal Science.  	The New Paradigm

    As a methodology, computer simulation could make clear the consequences
of a tremendous series of "what if" hypothesis in a very short time. It aided in
the initial articulation of the theory.  Most of this early exercise consisted of ex-
ploring the theoretical consequences of basic choices in the theoretically possible
chemistry. It was not concerned with direct comparison of model prediction with
observations (e.g., smog chamber data) so early emphasis was on explanation
and not on testing. In fact,  much of the chamber data were not in  an appropri-
ate form to compare with the computer predictions. Older descriptive concepts
lost meaning (e.g., tabulations of relative reactivity of hydrocarbons for making
ozone). New information was in demand: rate constants, reaction pathways, and
chamber operating conditions. The paradigm suggested new tools were needed
(new chambers), new instruments, and it offered direct guidance on what experi-
ments should be the most revealing.  It also suggested what new 'facts'  would be
critical  (rate constants for specific processes). That is, theory began to drive ex-
periment and vice versa.

The Practice of Normal Science.

Kuhn says, "Normal science consists  in the actualization of that promise [of suc-
cess offered by the paradigm], an actualization achieved by extending the knowl-
edge of those facts that the paradigm displays as particularly revealing, by in-
creasing the extent of the match between those facts and the paradigm's predic-
tions, and by further articulation  of the paradigm itself."

    In other words, normal scientific research  is directed to the articulation of
those phenomena and theories that the paradigm already supplies and is not con-
cerned with calling forth new sorts of phenomena nor is it normally concerned
with inventing new theories. By focusing attention upon a small area, the para-
digm forces scientists, Kuhn says, "to investigate some part of nature in a detail
and depth that would otherwise be impossible." This attention leads to increas-
ingly finer adjustments and additions to the theory and its auxiliary assumptions.
Eventually, it leads to facts that cannot be explained by the theories—anomalies.
A sufficient number of anomalies, coupled with an alternative explanation, leads
to crisis and possibly to revolution if a new theory is selected  to form the basis
of a new paradigm. So far the HO -chain paradigm has served air chemistry well.
Intense  scrutiny has been directed to questions of both the theory and practice.
Much detailed work has been in the area of testing the auxiliary hypotheses gen-
erated under the HO -chain paradigm.

The New Paradigm	The Practice of Normal Science.

   Kuhn says that during the time the paradigm is successful the problems of
normal science occur in three classes—determination of significant fact, matching
of facts with theory, and articulation of theory. Kuhn claims that these exhaust
the literature of normal science, both empirical and theoretical. Because of the
complex  interaction of theory and observation, it is difficult to separate these ac-
tivities. Descriptions of the classes follow:
    • The first  class is the production of facts (experimental work) that the par-
      adigm has shown to be particular revealing of the nature of things. By us-
      ing these  facts in solving problems, the paradigm has made them worth
      determining both with more precision and in more situations (e.g., mea-
      suring reaction rate constants).  A small part of theoretical work consists
      of using theory to predict factual information of intrinsic value. This, how-
      ever, is often viewed as "engineering" work and not of much real interest to
      scientists  (e.g., producing astronomical ephemerides, making predictions of
      air pollution control requirements). It is rarely published in journals.
    • The second class of experimental work is that class of factual determina-
      tions that can be compared directly with predictions from the paradigm
      theory. These often demand theoretical and instrumental approximations
      that severely limit the agreement to be expected (e.g., measuring the am-
      bient air concentration of HO ; wall effects in smog chambers). The second
      type of theoretical work  is the manipulation of the theory so as to produce
      predictions that can be confronted directly with experiment. These pre-
      dictions may or may not have any intrinsic value. The purpose of these
      activities  is to display a new application of the paradigm or to increase the
      precision  of an application that has already  been made. Precision in theory
      application concerns the type and nature of the simplifying assumptions,
      testing assumptions of presumptive generalities, or eliminating  anything
      that is likely to  restrict the agreement to be expected with actual  measure-
    • The third class of empirical work is that undertaken to articulate the para-
      digm theory, resolving some of the residual ambiguities and permitting the
      solution of problems to which it had previously only drawn attention.  This
      work includes the determination of quantitative laws. Sometimes, in the
      early applications of the paradigm, this work includes exploration experi-
      ments to see how the paradigm fits in some new area of application (e.g.,
      aromatics chemistry). These might be useful elaboration of the various
      possibilities and distinguishing among them. Theoretical problems of par-
      adigm articulation are often those of re-formulation both to achieve clari-
      fication and to allow extension of application (e.g., creation of condensed


Tool* Needed to Advance the Theory	The New Paradigm

      mechanisms for use in air shed models). A large share of the theoretical
      problems of paradigm articulation, however, are accounting for changes
      arising from the empirical work undertaken to articulate the paradigm
      (e.g., revising and updating mechanisms with new rate constants, path-
      ways, refining auxiliary assumptions). Often, when working in this class of
      work, a researcher will use both facts and theory to adjust and modify the

Early Normal Science Work

First Class Work
As you are aware, a large amount of first class (in both senses)  experimental work
occurred in the 1970's. Chief among this work was  the determination of the rate
constants of HO with almost every hydrocarbon.  In addition, the rate constants
of the most critical inorganic reactions were determined repeatedly with increas-
ing precision. How did we know which ones were the most critical? The models,
based on the HO -paradigm told us!

Tools Needed to Advance the Theory
The mathematical expressions arising out of reaction kinetics present significant
solution difficulties. No analytically solutions of these complex systems are pos-
sible, although some have been  tried for simpler systems.  Therefore, numerical
integration methods  are needed. For the kinetics systems  arising in air pollution
chemistry and which are described as "stiff," most of the numerical methods are
unstable—that is, the so-called  solution  produced by the method, which is only
an approximation, rapidly diverges from the true solution and the predicted val-
ues may approach infinity. In the early 1970's these problems limited the effec-
tiveness of computer simulations and were surely a  major  difficulty in the  work
described in the next chapter.

    In 1971, however, William Gear, a computer scientist, described a stable and
efficient method and produced a FORTRAN program suitable for the practical
numerical solution of large kinetic reactions systems.27 The wide-spread availabil-
ity (after about 1974) and efficiency of code using the "Gear" technique allowed
many air pollution researchers to take up computer modeling. That is, general
computer codes were produced that  accepted "chemical reactions" and initial
species concentrations as input and produced tables of species concentrations
verses time as their output.  The user of these programs did not have to know
anything about how the method worked and the codes were robust enough that


The New Paradigm	TooU Needed to Advance the Theory

numerical difficulties rarely arose. The researcher was therefore  free to think only
about the collection of "chemical reactions" that were used with these "turn the
crank" solution methods.

   These Gear codes are still too demanding of computer resources for use  in
three-dimensional grid air shed simulations, however, and no single solution  ap-
proach has dominated these programs. Thus simulations of smog chambers  can
be very complex and detailed (because they use the Gear codes), while even the
simplest three-dimensional grid atmospheric simulation that couples transport
with chemistry requires quite large computer resources (memory and time).  Be-
cause the chemistry in these latter simulation accounts for  about 80% of the com-
puter time, for practical applications, simpler chemical representations must be
used. This has created a certain kind of problem for the air pollution modeling
field: a difference between explain and predict.
                                     He wasn 't the first to explore the region,
                                                  but he returned with maps
                             that could  be followed with confidence by others.
                                                 —The Mechanical  Universe,

 First Explanation
The air pollution modeling literature, including books, journal articles, and EPA
reports can be divided into two broad categories: those works whose primary aim
was explanation of the transformation process, and those works whose primary
aim was prediction of atmospheric change.  Initially these were products of two
different sets of practitioners; later the two goals were integrated into one body of
practice, but with two expressions. I will conveniently describe the development
and testing of models by using these two categories as a framework in which to
present the process.

Demerjian, Kerr, Calvert

The first well-known example of normal science work following the adoption of
the new paradigm was "The Mechanism of Photochemical Smog Formation* by
Ken Demerjian, Alistair Kerr, and Jack Calvert21 which appeared  in 1974. This
was a large piece of work, totaling 262 pages, and although it did not appear in
its complete form until 1974, it  had actually been developed over the time period
1970 to 1972 (the manuscript was submitted to the editor in 1972). Alistair Kerr
was a summer visitor in 1970 and '71 to Jack Calvert's group at Ohio State where
Ken Demerjian was a PhD student. Demerjian's Master's Thesis, completed in
1970, had been "Computer Simulation of Smog Chamber Studies." Because of
the publisher's delay in printing the article, and similar to the Niki et al. case,
many of the active workers in the tropospheric chemistry area were sent preprints
of the article. In addition, a series of articles based on  the full manuscript ap-
peared in Environmental Letters in 1972 and 1973.  Probably of more significance
in shaping the careers of many workers in the area of chemical modeling, however,
was the "School on the Fundamental Chemical Basis of Reactions  in the Polluted
Atmosphere" which was held at the Battelle Conference Center in Seattle on June

Fint Explanation	Mechanitm Development

18-29, 1973 in which Jack Calvert and Ken Demerjian played key roles. Xeroxed
copies of the manuscript were essentially used as a 'textbook' and a computer link
to the Battelle computers allowed Ken Demerjian to offer a 'hands-on,' first-time
simulation experience for many of the hundred or so attendees.  There can be
little doubt that this work shaped the early paradigm that was  "limited in both
scope and precision" into a full example of how to practice science.  For this rea-
son, it guided a lot of people who wanted to start modeling—it  became a 'bible.'
The work embodies a whole range of explicit and implicit commitments about
how science should be practiced. We can only touch on a few here.

    In the introduction Demerjian, Kerr, and Calvert (DKC) said that the pur-
poses of their work were
      ... to evaluate the various alternative mechanisms and reaction rate constants pro-
     posed in view of the best kinetic data in hand today.

      ... to quantify our own thinking and that of our colleagues.

      ... to identify some of those potentially important reactions in the mechanism of
      photochemical smog formation for which there was insufficient basic  kinetic data to
      allow for  a realistic judgment of their importance.

      ... to predict theoretically the effects of atmospheric composition, as yet  experimen-
     tally unobserved, on the rates of formation of specific secondary products of some

      ... to stimulate the research activities related to this area of chemistry.

    I think it  lends support to  Kuhn's description of scientific activity to  note
that not only  do the goals and programs  set out by Demerjian et ol. fit well  into
what  Kuhn has called the classes of normal work, but that the very terms them-
selves echo Kuhn's.  Truly, as Kuhn has said, it is not necessary for researchers
to agree on a  paradigm, or even on its existence, to be influenced by it. The rel-
evance of Kuhn's description is clearly evident in this case at least.  Identity of
terms insures, I believe, that Kuhn's careful description about scientific progress
and radical change have at least nominal relevance in our specific science. And
if they have nominal relevance, then we can profitably look to them to sort out
problems of theory, theory-testing, and observational fit. These  are  the very prob-
lems that EPA has raised for this workshop.

Mechanism Development
The process of mechanism development was clearly described by Demerjian, Kerr,
and Calvert,


Mechanism Development                      	First Explanation
      W« feel that the only real hope of defining the reactions which are important in the
      real atmospheres lies in the elucidation of the simpler smog chamber experiments.

      We will consider first the simplest possible photochemically active polluted atmo-
      spheres.  Then  the effects ... of adding ... reactants common to the real atmosphere
      will be determined.

      We attempt to explain the product formation and reactant disappearance of a par-
      ticular smog chamber study with respect  to a rational series of elementary chemical
      reactions. Given the initial conditions and the rate constants for the elementary re-
      actions involved, the computer will generate by a numerical solution the reactant and
      product concentrations versus time curves for the kinetic mechanism proposed.

      In recent  simulations ... more complete and seemingly accurate reaction rate schemes
      have been employed. However, as with all such complicated systems for which accu-
      rate reaction rate data are not available for many of the reactions, a reasonably large
      degree of chemical intuition has been employed in the choice of rate constants.  In
      reality many of the chemical reactions which are believed to be involved in photo-
      chemical smog  have been studied experimentally, and their rate constants must be
      employed in kinetic  matching attempts; they should not be considered as variables
      which  can be adjusted outside of the experimental error limits in the attempted fit
      by computer.  On the other hand there are many reactions of potential importance in
      these systems for which no experimental data or very limited information exists. In
      these cases it is necessary  to estimate the rate constants 'theoretically,' avoiding any
      completely arbitrary choice of values.  Such estimates must be  consistent  with the
      present knowledge of chemical kinetics and thermodynamics.

      Although we have made what we believe to be reasonable estimates for the rate con-
      stants  for these reactions,  these estimates may be in considerable error.  Reliable
      values for them are necessary either to confirm our suggestions or to force a reevalu-
      ation of the importance of some portions of the smog mechanism outlined here.

      —  without certain  key  information among the smog chamber data, it is impossible
      to develop a reaction scheme which is completely free from adjustable parameters
      which  are varied in an arbitrary fashion to match the kinetic simulation to the ex-
      perimental data. Simulations which are built on such  flexible schemes cannot add
      significantly to the  quantitative scientific understanding of the  atmospheric reactions.

      ...  there is often relatively poor reproducibility between chamber product rate stud-
      ies run by different laboratories, and the results of one individual study  may be
      rather inaccurate and misleading.  We have attempted to overcome this problem and
      to develop a consistent set of chemical reactions  related  to photochemical smog sys-
      tems through the treatment of several very different systems studied by  a variety of

Firrt Explanation	Realiim

Let us begin our discussion of the philosophy behind DKC's statements by exam-
ining the manner in which theories are seen to relate to reality. This topic is a
subject of current debate; for example, a conference on scientific realism was held
at UNC, Greensboro in 1982.  There are two basic philosophical positions (with a
great deal of intravariability):
     > instrumentalism or antirtalism posits that the nonobservational portions
       of even "true" theories are not really assertions about the world, but are
       devices of some sort for codifying relations between observable variables.
       This becomes the claim that  any theory, however well confirmed and widely
       accepted  in scientific practice, might well be and might eventually be  re-
       vealed to be false.9
     > realism posits that physical theories make meaningful assertions about the
       real world (i.e., the terms in  a theory are genuinely referential); they are at
       least approximately true.  The approximate truth of a scientific theory is a
       sufficient explanation of its predictive success. Theories which  are regarded
       as genuinely explanatory can be used to predict facts other than those for
       which they were first invented (or they can be modified in a simple way
       to do this) because the entities that are described by the theories really

    In its most common form, realism is the belief that "science makes possible
knowledge of the world beyond its accessible, empirical manifestations."20

    The antirealist sees major objections to the realist's position:
      Whatever continuity may be discerned in the growth of empirical knowledge,  theo-
      retical science has been radically discontinuous. Scientific views about the ultimate
      structure and law like organization of the world have frequently been overthrown  and
      replaced by  incompatible views. Much of the discarded science-was, for an apprecia-
      ble time, eminently successful by the standards we employ in assessing current sci-
      ence. The inference seems inescapable that the evidence available to support current
      science is by nature unreliable and systematically underdetermines what ought to be
      believed about the world beyond our experiences. Scientific theories, however well
      secured by observation and experiment are inevitably fallible.  Nor is there any ba-
      sis for expecting  the future evolution of scientific standards and methods to provide
      a more secure foundation for scientific knowledge. For methodological developments
      that have occurred thus far, whatever improvements they have generated at the level
      of human interaction with nature, have failed utterly to resolve the basic dilemma of
      the underdetermination of theory.20

DKC'i Rule*	Fir»t Explanation

    The second major area of objection to realism concerns "What kind of ex-
planation is truth?"* That is, what kind of mechanism is truth?  How does the
truth of a theory bring about, cause or create, its issuance of successful predic-
tions? Priefly, "the truth of a theory is no more likely a mechanism for sifting out
false predictions the theory might have made than it is likely as the mechanism
by which the theory made  true predictions."8

    An instrumentalist is of course an instrumentalist about all theories. A re-
alist, however, can be a realist about some theories (those which he believes are
true)  and an instrumentalist  about others (which he believes to be useful, though
not true).29 In much of the scientific literature the term  "model"  has been used
for the latter sort of theory.  (See Box 'Theories and Models' for example).

    This latter position  has been stated in the form:
      If you believe that guessing based on some truths is more likely to succeed than
      guessing pure and simple, then if our earlier theories were in large part true and
     if our refinements of them conserve the true parts, then guessing on this basis has
     some relative likelihood of success.

DKC's Rules
With that background, let  us examine Demerjian, Kerr, and Calvert's rules and
rationale for mechanism development and testing. These include:
    1. Mechanism reactions must be elementary reactions.
      These are believed to be the actual steps in which the transformation takes
      place. This representation is most commensurate with kinetic measure-
      ments and thermodynamics.  Every other representation involves an ap-
      proximation  (e.g., steady-state, no competing reactions, etc.)  that may
      limit the adequacy of the representation under some conditions. Direct
      generalized reactions "provide no clue as to the real intermediates or tran-
      sient species  involved."
    2. Rate constants must be based on experimental determination of rate con-
      stant or must be  estimated using thermodynamic techniques.
      Failure to use all known information to eliminate possibility of mere "curve
      fitting" results in a description for which there is "no evidence of the unique-
      ness or the correctness of the  chosen mechanism."
    3. The mechanism must explain  smog chamber data.
      If we can  not explain the simpler systems in smog chambers we  have no
      reason to  infer that we have correctly chosen among the alternative mecha-
      nisms and rate constants.


Pint Explanation	DKC's Rules

    4. The understanding of smog chamber data is best revealed by starting with
      simple systems and increasing the complexity of the hydrocarbon compo-
      nents stepwise.
      Starting with simple systms focuses attention on particular aspects of the
      theory.  Otherwise there is "no evidence of the uniqueness or the correct-
      ness of the chosen mechanism."
    5. Rate constants of important reactions that were estimated 'theoretically'
      must be determined experimentally.  This, in turn, may force a reevalua-
      tion of the smog mechanism.
      Theory can predict only an approximate value for the rate constant. The
      application of thermodynamic theory to calculate a rate constant requires
      numerous auxiliary assumptions which contribute to an uncertainty in the
    6. When there are "missing" chamber experimental information (e.g., pho-
      to lytic rates), the experimental  data must be supplemented by a rational
      process for estimating the missing data.
      Smog chamber data inevitably have missing information that must be esti-
      mated by the theorist. This should be done in as mechanistic  a manner as
      possible, but may involve "fitting"  exercises to make final choice. This, of
      course, contributes to uncertainty,  leading to the next requirement.
    7. Must model more than one chamber. The basic core of the model must
      not be changed from species to  species simulation or chamber to chamber
      This derives from a long standing heuristic of observation science. An indi-
      vidual chamber and experiment may be inaccurate and misleading, so use
      repeated observations by different investigators using different tools.

    From this abstraction of DKC's approach, we can deduce that DKC are, in
some important respects, realists. They believe that there are'"real intermedi-
ates" ; that there are unique and correct mechanisms for the transformations; that
some mechanisms are "seemingly accurate"; that there is a close correspondence
between their theoretical constructs and the real world so that it is possible to
"predict theoretically the effects of atmospheric composition, as yet experimen-
tally unobserved, on the rates of formation" of interesting secondary  products—
even if they had to construct "imagined"  reactions and rate constants from the-
ory.  They criticize their more instrumentalist colleagues who adjust models to fit
chamber data by saying,

DKC'i Rule*                                                               First Explanation
                                   'Theories and Models'

 We all learned Boyle's Gas Law,
                                      PV = constant

 and found out that, although useful, that it was not really correct for real gases. Boyle's law
 can be derived from a 'model' of gases which supposes  that gas molecules are point masses.
 Later we learned that Van der Vaal's Law,

                                (P +    )(V - 6) = constant
 was better, but still not quite correct, for real gases. Van der Vaal's law can be derived from a
 'model' of gases which supposes that gas molecules are elastic spheres with real volume.

     The best basic theoretical knowledge about real gases derives from the calculation of the in-
 termolecular potential between gas molecules from quantum mechanics by solving the Schrodingei
 equation. In practice, this is virtually impossible, and such calculations have been made only for
 the very simplest molecules. In the absence of such  complete calculations, it is necessary to
 approximate the potential by a semi-empirical formula, containing one or more adjustable con-
 stants, which is chosen on the basis of physical plausibility and mathematical convenience.'0

     Theoretical laws are not generalizations of observations, but are postulates of a theory, and
 are tested  indirectly by comparing the consequences of the theory, cur a whole, with experimental
 or observational facts.  Anomalies, by themselves,  are insufficient reasons to reject theories.29
      It is evident that a good computer match of smog chamber rate data obtained using
      arbitrarily chosen rate data for the extremely complex systems involved in photo-
      chemical smog is no evidence of the uniqueness or the the correctness of the chosen
      mechanism.  In fact, conclusions and predictions based on the extrapolation of such a
      mechanism to conditions not employed experimentally can be completely misleading,
      since the actual intermediates  involved and their competitive reaction steps may not
      be included.

The implication being that their method does not suffer from this problem, be-
cause they have used their fundamental  belief in chemical kinetics and thermo-
dynamics to calculate rates 'theoretically.' In so  doing they opposed their funda-
mental belief to  the "large degree of chemical intuition" that had been employed
by others in the "choice of rate  constants."

    There is a  remarkable omission in the DKC work: there is no description of
the methodology for choosing the 490 elementary reactions that made up their
mechanism from the thousands  that they recognize must occur. There are phrases
such  as "rational series of elementary chemical reactions" and  "complete reaction

Firat Explanation	DKC's Rules

scheme" and "seemingly accurate" schemes, and an expression of "hope of defining
the reactions".  It is clear that a near complete literature review was undertaken
(there were 219 references, some of which were in turn reviews). A great number
of the reactions and their measured rate constants used by DKC came from this
review.  DKC were also clear that, for rate constants that had not been measured,
they had made theoretical estimates. In the latter case, there are two types of re-
actions: 1) those that have been observed to occur in the laboratory, but which
have not had quantitative measurements allowing rate calculations; and 2) those
that have never been observed but are supposed to occur, that is, which were hy-
pothesized to exist to explain some other observation. In this case, we need to
distinguish two sub-classes, the first is when the reaction does occur in the real
world (i.e., it was a correct guess) and the second  is when the reaction does not
occur in the real world (i.e.,  it was a wrong guess).

    While DKC implied that they needed smog chamber data to evaluate "alter-
native mechanisms and reaction rate constants proposed," they  did not explain
how this occurred. Recall the statement "We feel that the only real hope of defin-
ing the reactions which are important in the real atmosphere lies in the elucida-
tion of the simpler smog chamber experiments." They did not make explicit a
strategy for, to use Kuhn's terms, "elaboration of the various possibilities and to
distinguish among them."  In the discussion of each reaction, they occasionally
use a practice of making choices on the basis of mostly qualitative comparisons
of alternative prediction pathways but mainly with kinetic experimental data and
not with chamber data (e.g., this path can not be  important because we do not
observe its product). The incommensurability between the smog chamber data
available to DKC and the needs associated with the theory articulation, however,
may have prevented much more, so this strategy was not articulated in their ap-
proach (see below). DKC also did not offer a description of the  methodology for
the elaboration of the theoretical possibilities they considered.

    There is a third class of reactions whose rate constants have not been deter-
mined. These are reactions that occur and have not been observed in the labo-
ratory because no one has ever known to look for them. The theory never said
that they are likely to exist or be important. Thus the only way they are likely
to be found is through serendipity, a common enough happening in science. Note,
however, that this requires a broadminded observer, one who does not discount
the observation as "noise"—everything that is not predicted by  the theory.  The
possibility of important undiscovered reactions must present a double bind for one
who is trying to build a "complete reaction scheme": on the one hand, he will not
delete anything from the scheme because it might  be important  in some way he


DKC'f Rules   	First Explanation

might not yet understand, and on the other, he probably has deleted hundreds of
reactions that are important! Coming to grips with this situation has got to be
uncomfortable for those seeking a "complete reaction scheme" for the "truly sci-
entific understanding of air chemistry."

    How should a scientist deal with what was described above as the "basic dilemma
of the underdetermination of theory?"

    Another observation that can be made about the DKC approach is that it ex-
hibits an asymmetry for the treatment of experimental fact. Contrasting points
6 and 7, with  point 5, it is evident that failure to fit smog chamber data is  not
a sufficient reason to reject a mechanism. However, the experimental measure-
ment of a rate constant that changes the value already being used in a mecha-
nism may require a complete re-evaluation of the mechanism.  By this analysis,
chamber data constitute neither a sufficient nor necessary condition for testing
theory-generated mechanisms. That is, success in modeling chamber data does
not measure a mechanism's accuracy, and failure in modeling chamber data is not
acceptable for mechanism modification.  In this situation, the conclusions (based
on the definition and significance of "sufficient and necessary") about chamber
data fits by mechanism predictions is that no mechanism is ever dependent in any
way on  chamber data, regardless of how the mechanism's authors construe  their

    I think this position troubles few modelers.  That is,  modelers have tradition-
ally explained their reluctance to use chamber data by pointing to the massive
uncertainties inherent in any such measurement system.  This I have suggested
is in part due  to their basic realist philosophies. Such an explanation though  is
somewhat awkward when we recognize that the foundation that modelers take
as unimpeachable— kinetic rate constants —are themselves measurements, of-
ten conducted in a highly distorted environment compared to the ambient air,
and are subject to problems similar to ones that confront chamber data. It is true
that chamber  data are artificial, but I would contend that they might be consid-
ered less artificial than the conditions used to produce rate constants. In addi-
tion, chamber data, used as an environment to test real and non-real world situ-
ations, are distorted in the first sense [see Chapter 2]  so that their usefulness can
fit with the paradigm.  They are less distorted in the second sense—that is, as a
consequence of generalization and deletion that  led to map-making. Kinetic rate
data, however, are distorted in both the first and second senses: in the first just
as chamber data are, and in the second sense precisely because the model itself

First Explanation	DKC'» Rules

determines which rates are important and which are not, and therefore, may de-
termine which rates are measured.

    I believe that it is just easier to juxtapose the fact to the theory for the rate
constant, but the fact of the smog chamber is just as real and should carry similar
weight.  The problem appears to be the difficulty of confronting theoretical detail
with the more macroscopic fact from the smog chamber system. Furthermore, as
Kuhn says, scientists who  have spent their life's work with  a particular paradigm
can be quite comfortable not treating anomalies as counter-instances, and, there-
fore, are very reluctant to  modify their theories.

    In spite of their carefully worded effort to avoid it, DKC were somewhat hard
on the smog chamber experimenter compared to the laboratory kinetist. The
straight interpretation of what was said is:
    > we had significant and recurring problems simulating smog chamber data,
      mostly due to incomplete information provided by the smog chamber team;

    > this required that we  guess about certain key information; therefore,
    > we could not develop  a mechanism free from arbitrarily adjustable parame-
      ters; therefore
    > simulations using these mechanisms cannot  add significantly  to scientific
      understanding of atmospheric  reactions.
And so  "chamber-bashing" was added to the modeling folklore.

    On the other hand,
    > numerous kinetists did not measure numerous rate constants or even deter-
      mine the existence of  key reactions; therefore,
    > we had to make 'theoretical' estimates of reactions that did not exist; there-
    > we could not develop  a mechanism free from arbitrarily adjustable parame-
      ters; therefore
    > simulations using these mechanisms cannot  add significantly  to scientific
      understanding of atmospheric  reactions.

DKC'i Rules                                                           First Explanation
                          'Famous Theoretical Hate Constants.'

 While the use of thermodynamic techniques to calculate rate constants can be very useful and
 sometimes represents the only way to 'guess,' this methodology does not carry a guarantee of
 valid prediction, For example,

                             HO2 + NO	. NO2 + HO

 was calculated to have a rate
                                 200   ppm'^min"1

 and DKC said this value was "consistent with all of our simulations." Later experimental mea-
 surements place the true rate at
                                12000 ppm~2-min~i

 The following cases  illustrate that it  is possible to 'thermodynamically' estimate rates for reac-
 tions of entities  that do not exist (e.g., HC(0)02) and for reactions that do not occur, as with
 PAN+ NO. For  a complete discussion of the PAN reaction see box 'Explicit Mechanism Formu-
 lation' later in this essay.
 HC(0)02 + NO -  N02 + HC(0)0  9.1E2 DKC'i reaction 51a
 PAN + NO - 2 N02 +  CH3C(0)0  1.6E-1 DKC's reaction S3b
    The real issue here is the nature of Kuhn's second class of normal scientific
work, that is, the work in experimental areas to produce facts that can be com-
pared directly with theory and the manipulation of theory so as to produce pre-
dictions that can be confronted directly with experiment. Without close juxtapo-
sition of the experimenter and the theorist, experimental measurements are not
likely to include all the input needed to attempt a quantitative simulation of the
rate data.  The experimenter does not just randomly measure everything, he mea-
sures what he thinks is important according to his understanding of the system.
If the theorist produces a new explanation, this will of necessity accord a differ-
ent importance to different facts. Without new measurements suited to this new
importance, the theorist must construct the needed "missing" facts if he wishes
to use the existing  data to test his theory. This is in small part what was recog-
nized by Kuhn when he described the incommensurability of successive bodies of

    Likewise, one experimenter may  have a different understanding of the im-
portance of any explanation than another experimenter, and so may use differ-
ent methodologies. In the evaluation of models this means that the model testing
ability of a particular set of data may be  somewhat incommensurate with the par-
ticular needs of the theorist while other aspects of the same data may fit other


First Explanation	DKC's Rules

needs of the theory and the auxiliary assumptions quite satisfactorily. This, then,
is a particularly clear example of another way that underdetermination of theory
has troubled air pollution chemistry.

    This need to combine experimental and theoretical work was recognized by
DKC in the last paragraph of the conclusions to their work,
      We hope that the reader has become convinced that there is a real benefit to be
      gained by both parties in the close association between  smog chamber experimenters
      and those who model such systems in terms of theoretical mechanisms. A truly sci-
      entific understanding of the chemical changes which occur in the various complex
      chemical mixtures which constitute our polluted atmosphere may never be possible.
      However our only hope of achieving this end lies in the  understanding of the simpler
      smog chamber experiments. Eventually the knowledge gained from these systems
      will allow the atmospheric  scientist to exercise the  necessary  degree of chemical so-
      phistication in computer simulation of atmospheric reactions  so that he can develop
      scientifically sound and useful predictive atmospheric  models and control criteria.

    Actually the science has  advanced as suggested by DKC and the original the-
ory augmented by strong auxiliary assertions has greatly improved the explana-
tion of air chemistry. Likewise, the the chamber data have advanced, and today's
best chamber data can offer  so many counter-instances for major parts of DKC's
mechanism that there  would be little chance anyone would judge it an adequate
representation of the truth. But in that that was itself one of DKC's goals, we
can easily see that their approach certainly must be considered a significant, and
primary contribution to the paradigm for the development of explanatory models.

    But problems do remain. How else, other than  the use of smog chambers, can
we produce the  detailed facts that must be compared unambiguously and directly
with the model  (theory+auxiliary statements) while preserving as much of the
whole chemical transformation process as possible to  allow for  the evaluation of
significant over-generalization and deletion? How (or  even do)  such comparisons
deal with the question  of the underdetermination of theory? And if the uncertain-
ties associated with  chamber data will always cause the model to under determine,
then should we abondon the whole set of them?
                          ... important characteristics of maps should be noted.
                                        A map  is not the territory it represents,
                         but, if correct, it  has a similar structure to the territory,
                                              which accounts for its usefulness.
                                                               —A. Korzybski,
                                                             Science and Sanity

 First  Reformulation
Hecht, Seinfeld, Dodge

As its title, "Further Development of Generalized Kinetic Mechanism for Photo-
chemical Smog," suggested, the 1974 paper by Tom Hecht, John Seinfeld, and
Marcia Dodge31 (HSD) was an effort to articulate further the paradigm's ma-
jor theory by a reformulation to allow extension of its application to the atmo-
sphere (Kuhn's third class of normal science work). While a prime objective for
DKC was to have  "a complete reaction scheme" and thus to explain, the prime
objective for HSD  was "that the mechanism predict the chemical behavior of a
complex mixture of many hydrocarbons, yet that it include only a limited degree
of detail" which would allow the mechanism to be applied to the "prediction of
both smog chamber reaction phenomena and atmospheric reaction phenomena."
Thus they sought  "a reasonably rigorous, yet manageable, description of the pho-
tochemistry of air  pollution." This work, therefore, in its conception and defini-
tion, set out toward a different end than the DKC work: DKC sought scientific
explanation; HSD  wanted prediction.

   In part I suppose, because it would be difficult to argue from a position -of
'completeness of representation' in their case,  HSD said,
     A kinetic mechanism, once developed, most be  validated.
Reading this, I realized that the word "validate" never appeared in the DKC
work! Continuing  from HSD,
     [Validation] is commonly conceived as consisting of two parts: validation in the ab-
     sence of transport processes and validation in their presence. In practical terms we
     are speaking, respectively, of comparison of the model's predictions with data col-
     lected in smog chamber experiments and with data collected at contaminant moni-
     toring stations in an urban area.

First Reformulation	Hecht, Seinfeld, Dodge

    Something is not clear about this argument.  We understand why the chem-
istry must be 'validated' in the absence of transport—this situation sharpens the
point of comparison between the theories  of transformation and the species mea-
surements. Hecht32 said
      In  a system  as complex as the atmosphere, isolation and characterization of the
      chemical processes that occur are virtually impossible.  Thus, two fundamental ap-
      proaches have been devised for controlled study of the chemistry of smog formation:
      irradiation of known concentrations of pollutants in a  large reactor (smog chamber)
      and determination of the rates and mechanisms of elementary reactions thought to
      occur in polluted air. ... /In the chamber,] one is unable to gain insight into the de-
      tails of the chemical transformations taking place:  only the macroscopic effects of
      the overall chemical  process is observed. On the other hand, detailed information
      concerning the rate and course of a single  reaction, such as that obtained in a kinet-
      ics study, reveals little concerning the nature of the overall smog formation process.
      Indeed, one  has difficulty knowing whether all the  important reactions have been

          Kinetic  simulation provides a means of comparing the results of these  two ex-
      perimental approaches for investigating the chemistry of smog formation.  ... kinetic
      simulation takes the results of elementary  reaction studies  as input data and pro-
      duces results of a form identical to the data obtained in smog chamber experiments
      as  output. By comparing the predictions of the mechanism with actual chamber
      data, one is  able to determine the degree of agreement between the two experimental

Therefore, unlike in the DKC approach, Hecht tt al. recognize a priori that dele-
tion is a significant problem in mechanism construction and they accord equal
consideration to both kinetic data and smog chamber data.

    With regard to the need to validate in the presence of transport processes,
however, the wording in the HSD paper might be taken to imply that transport
processes somehow modify the chemistry beyond the changes  that effects the con-
centrations of react ants. What is the real need to validate in the atmosphere?
Hecht said in the quote above that it was  virtually impossible to isolate the chem-
ical process in the atmosphere. In any case, HSD finessed the question by saying
they were only comparing "predictions and experiment  based on smog chamber

    After listing a series of mechanisms (including the DKC) developed by others,
HSD made an important point:
      For none of  the specific  mechanisms listed  above has there been reported a program
      of  validation over a range  of initial reactant concentrations. Thus all five can only be
      considered at this point as detailed chemical speculations.

To be clear on this point, Hecht said,


Hecht, Seinfeld, Dodge	  Firet Reformulation

      ... [Mechanism] uses are contingent upon the accuracy of the mechanism.

          Smog chamber data constitute a standard against which the accuracy of pre-
      dictions of a kinetic mechanism can be measured. But because of wall effects and
      other operating characteristics  of the system, the observed time-varying chemical dis-
      tribution in the chamber cannot be accounted for entirely in terms of the reactions
      occurring in the gas phase. Thus, the use of chamber results as a standard or refer-
      ence for judging the accuracy of a mechanism must be carefully qualified.
He went on to say that because a  chemical mechanism represents the chemistry
that presumably would be observed in an infinitely large, uniformly irradiated,
well mixed reactor and because  a typical smog chamber does not meet these cri-
teria, then, before predictions of the mechanism and chamber observations can be
compared, terms must be added to the mechanism to represent the characteristics
of the chamber that differ from  ideality.

    Further elaboration of the nature of validation was presented by HSD. For
      The significance of the validation results for a kinetic mechanism is to a large degree
      dependent upon the diversity and reliability of the experimental data base.
Hecht32 said that the rates of numerous reactions were being determined in a con-
tinuing fashion and that speculative reactions were being scrutinized, and  known
reactions were being confirmed. He added
      In such a dynamic scientific environment, it would be presumptuous to  suggest that
      any given set of reactions and rate constants is the chemical mechanism for smog
A bit later in his discussion he said that most changes in mechanisms come about
because of the need to "replace overly general representations with more detailed
descriptions and to substitute sound experimental measurements—as they become
available—for speculative estimates."

    HSD made a clear statement about what they meant by diversity in the cham-
ber data:
     •  a variety of hydrocarbon  systems of high  and low reactivity  had to  be in-
     •  the experimental data had to be for single reactants  and for mixtures;
     •  the initial conditions had to cover a broad range of HC-to-NOx ratios.
The latter condition was seen as especially critical for control applications in ur-
ban  atmospheres.

    By reliable data base, HSD said they meant:


First Reformulation	Hecht, Seinfeld, Dodge

    • certain experimental variables had to specified accurately;
    • potential chamber wall effects had to be quantified;
    • experimental measurements had to be specific; accurate, and repeatable.
An interesting remark made by HSD with regard to the data used by them was
"in spite of /..some problems..], the data are in general  reproducible, were care-
fully taken, and are as suitable as any currently available for validation purposes."
This was a recognition of a problem that will always plague modelers. The cham-
ber and laboratory data require 2-3 years to be produced, be refined, and be
communicated. New models, therefore, are likely to incorporate ideas that are
of newer origin than those that drove the design of the chamber and laboratory
experiments. The irony, of course, is that when the data can be made available,
they often conflict remarkably and repeatedly with the model predictions. I think
that the ability of chamber data to present sincere tests 2-3 years after they were
observed—and observed sometimes for different purposes—confirms the impor-
tance of the testing for which HSD called. That  such testing confirms 'validation,'
though, has not proved convincing.

    By validation of the mechanism, HSD meant following the process of:
    1. obtaining estimates of the various input parameters to the mechanism:
           > the reaction rate constants,
           > parameterized stoichiometric coefficients,
           > initial concentrations of reactants,
           > chamber operating parameters-dilution rate, light intensity, etc.
    2. carrying out sensitivity studies for these parameters—i.e., establish the
      effect of controlled variations in the magnitude of the various parameters
      on the concentration-time profiles for major species;
    3.* predicting concentration-time profiles for the various reactant mixtures
      using the initial conditions;
    4. comparing predictions with experiment results by plotting both predicted
      and measured concentrations on the same plot.
    5. assessing the "goodness of fit" and identify the sources of uncertainty
    6. reaching a conclusion with regard to the mechanism's adequacy for the
      particular application.
I had not read this paper for nearly 10 years, but its effects had undoubtedly
shaped some of my work. In studying the paper for this essay, I was surprised to

Heeht, Seinfeld. Dodge      	Firtt Reformulation

find how accurately it describes my current practice:  I used this exact procedure
in recent tests of the Carbon Bond Four and the RADM mechanisms.

    As to the mechanism development process, Hecht32 provides more insight
than the HSD paper. He says:
      The primary criterion used to decide whether a reaction involving reactants present
      in photochemical smog should be included in a chemical mechanism is the rate of
      the reaction. ... To be sure, the complete kinetic analysis of a reaction  entails more
      than the measurement of the rate  constant:  the mechanism and products of the ele-
      mentary reaction must be determined as well.

While many of the reactions in his  mechanism were thought to be well character-
ized, Hecht recognized that there were significant exceptions. He said,
      Complete omission or inaccurate specification of products of key reactions, along
      with inaccuracies and uncertainties in the values of many rate constants, are prob-
      ably the causes of a major portion of the disagreement between predictions and ex-
      perimental data.  Resolution of these types of uncertainties is a slow  and painstak-
      ing process. In the absence of experimental measurements, provisional rate constant
      values can be estimated and elementary reactions and reaction mechanisms can be
      hypothesized on the basis of thermochemical principles (e.g. Benson). However, be-
      cause of the large uncertainty associated with most estimates, we depend mainly on
      experimental studies  to justify the changes introduced into the mechanism.

That is, while experimental work was underway, Hecht would hypothesize a par-
ticular reaction mechanism and rate constants so as "to augment the agreement
between the predictions and the data for the smog chamber"; this hypothesized
mechanism must be tested extensively.  He said
      The development of the mechanism during this project followed two complemen-
      tary paths.  The first approach involved testing the quality of predictions in relation
      to smog chamber data. By adding, modifying, or deleting reactions and then vary-
      ing the values of rate constants within their  bounds of uncertainty, we sought to in-
      crease the accuracy of the prediction of the mechanism over a wide range of organic
      compound-to-NOx ratios.

          The second approach focused on quantifying the influence of uncertainties in
      the values of individual reaction rate constants on the predictions of  the mechanism.
      ... [The uncertainty/ index reflects the importance of obtaining more accurate rate
      constant measurements as a means of reducing the uncertainties in the predictions of
      the model. In addition, ... we were able to point out insensitive reactions that can
      possibly be deleted from the mechanism without significant loss of accuracy.

    Along the latter lines, HSD introduced some compression techniques that are
still in use in  today's mechanisms.  For example, alkyl radicals are assumed to be
converted to peroxyalkyl radicals exclusively and thus never need specifically en-
ter into the mechanism, and acylate radicals (RC(O)O ) are assumed to decompose


Firtt Reformulation	Hecht, Seinfeld, Dodge

exclusively to from alkyl (hence peroxyalkyl) radicals and C02, thus they do not
have to be included in mechanism, and so forth. These steps by themselves re-
move much detail from the mechanism without any significant loss of accuracy
and therefore are still used today, even in so-called explicit mechanisms.

   HSD listed a number of criteria that had to be met by a mechanism suitable
for atmospheric prediction:
   A.  it must rigorously treat the inorganic reactions; t.e., explicit elementary
       chemistry must be used for these reactions;
   B.  the detail associated with specific hydrocarbon mechanisms, i.e., ones in
       which each species in a reaction represents a distinct chemical entity, is
       not practical in developing a mechanism for the atmosphere; therefore a
       "lumped" representation  must be used, i.e., "one which contains certain
       fictitious species that represent entire classes of reactants";
   C,  on the other hand, the mechanism structure must not depend upon difficult-
       to-specify stoichiometric coefficients or parameters; such parameters ad-
       versely affect the "confidence in the predictions of the mechanism for cases
       not explicitly validated";  therefore, the stoichiometric coefficients and  mech-
       anism structure must be derivable directly from the underlying chemistry
       of each elementary reaction.
   D.  rate constants must be related to  elementary reactions; i.e.,"While the ki-
       netic mechanism is written in a general fashion, we have striven to formu-
       late it in such a way that all important features of the detailed chemistry
       are retained. Thus our goal has been to include each elementary reaction
       thought to contribute to the overall smog kinetics."
   E.  the validation value of the rate constants must be within the range of val-
       ues recommended by a consensus of investigators;
   F.  the representation of the  atmospheric hydrocarbon mix must not be over-
       simplified; i.e., the reactivity  and change in reactivity must be preserved
       in the generalized representation;  therefore, the mechanism must be "suffi-
       ciently detailed  to distinguish among the reactions of the various classes of
       hydrocarbons and free radicals"; the mechanism must include in its specifi-
       cation a method for how  to represent the complex atmospheric mixture in
       the input.

   The HSD mechanism was a conscious attempt to produce a representation
that contained carefully described, nevertheless fictitious entities for the purpose
of simplicity.  Look  back at the box "Theories and Models."  Molecules are not


Hecht, Seinfeld, Dodge 	First Reformulation

point masses, yet we find that treatment a useful fiction. By generalizing the im-
portant reactions and deleting the unimportant reactions, the HSD mechanism
became much more understandable in a qualitative sense. It certainly is easier to
understand the overall transformation processes using this mechanism than, for
example, using the 47 pages of reactions in the DKC mechanism.

    Clearly the emphasis of the  HSD mechanism is on prediction and less on  ex-
planation.  If one really wanted to know, for example, why substituting one HC
for another in a mixture did not change the O3, the HSD mechanism as an ex-
planation is fairly useless.  One would have to say, "Because the stoichiometric
coefficients for the HC that was substituted were the same as those of the original
mixture."  Not much of an explanation! This is, I believe, part of the reason  for
the requirement expressed in point C above:  the stoichiometric coefficients and
mechanism structure must be derivable directly from the underlying chemistry
of each elementary reaction. This keeps HSD in line with DKC and with  a real-
ist viewpoint in that there was a common belief that the fundamental chemistry
is the same for test,  explain, predict. From this point of view, HSD were  a new
tack on the same course. It is the extension of the fundamentals, however, that
separates HSD from DKC. The  last sentence of point F illustrates this separation:
HSD start with  an input representing the "complex atmospheric mixture" and
then include in their model exactly what DKC had not even finished describing!

    After comparing their model predictions with chamber data, HSD said
      ... the data and  predictions are not always in good agreement for all species over
     the full period of irradiation. These discrepancies can be  attributed to at least  five
      [sic..only four listed] possible sources of uncertainty:
The sources were:
    •  The mechanism may be incomplete.
    •  The lumping  process may introduce error.
    •  There are uncertainties in the experimental data used for validation.
    •  Not all the rate constants and pathways are known with a high degree of

And HSD conclude:
     As a consequence of these uncertainties, we have not yet reached the point in model
     validation where we are in a position to assess the "goodness" of the proposed mech-
     anism, or for that matter, to draw unequivocal qualitative conclusions regarding its
     merits. Yet, the mechanism appears capable of predicting the concentration-time
     behavior of a variety of reactant systems over a wide range of initial conditions.

Firat Reformulation	     Hccht, Seinfeld, Dodge

    Did this mean that the mechanism was validated? To be validated means33 to
have  "strength or force from being supported by fact," to be "well grounded on
principles or evidence," and therefore, to be "able to withstand criticism or ob-
jection." Because of uncertainties in both the mechanism and the chamber data,
HSD  were not willing to say that the mechanism was quantitatively accurate, but
they were willing to say that the mechanism was valid for "predicting ... behav-
ior."  That is, it was capable of predicting external appearance or action. This
means that HSD were asserting that the mechanism was  satisfactory for  predic-
tion,  but not satisfactory for explanation.

    HSD were making a bold claim!  They were announcing to the "scientific com-
munity" that they had reasonable enough agreement between theory and facts
that subsequent use of their model could be believed as to the predictions of the
external appearance or action of smog chemistry.

    How could they  do this? Kuhn gives us an explanation in his "The Function
of Measurement in Modern Physical Science."1S While describing tables of the-
oretical predictions verses experimental measurements which are supposed show
agreement,  he says,
      ... scientists  seek not- "agreement"  at all, bat what they often call "reasonable agree-
      ment." Furthermore, if we ask  for a criterion of "reasonable agreement," we are lit-
      erally forced  to look in the tables themselves. Scientific practice exhibits no consis-
      tently applied or consistently applicable external criterion. "Reasonable agreement"
      varies from one part of science  to another, and within any part of science it varies
      with time.

         For example, in spectroscopy, "reasonable agreement" means agreement in the
      first six to eight  left-hand digits in the numbers of a table of wavelengths.  In the
      theory of solids,  by contrast, two-place agreement is often considered very good in-
Kuhn further says that the -actual function of measurement must be sought in the
journal literature, which displays not finished and accepted theories, but theories
in the process of development. When a journal article is accepted by the profes-
sion as showing "reasonable agreement" it becomes the definition of "reasonable
agreement." Kuhn says that is why the tables are there.  By studying the tables
and graphs a reader learns what can be expected of the theory. An acquaintance
with  the tables is part of an acquaintance with the theory itself. "Without the
tables," says Kuhn,  "the theory would be essentially incomplete."

    The quantitative comparison of theory  and observation is Kuhn's second type
of normal scientific work and he claims  that this work has required the highest


Heeht, Seinfeld, Podge	Tint Reformulation

scientific talents to invent apparatus, reduce perturbing effects, and estimate the
allowance to be made for those that remain. In so far as their work is quantita-
tive, most scientists do this type work most of the time.  He says, "Its objective
is, on the one hand, to improve the measure of 'reasonable agreement' character-
istic of the theory in a given application and, on the other, to open up new areas
of application and establish new measures of 'reasonable agreement' applicable
to them." This is exactly  what HSD did.  Since there was no refutation of their
statement by the 'scientific community', there was "reasonable agreement."

    We shall see this process at work many times in this brief history.

Second  Explanation
Durbin, Hecht, Whitten

Until about 1975, mechanisms had been compared with smog chamber data that
were, in effect, "left over" from the pre-paradigm stage. That is, the data were
not produced using the new insights provided by the HO -paradigm. As described
by Kuhn, this resulted in an incommensurability between the old facts and the
new theory because the experimental work before modeling practice began was
directed at different goals and needs than the facts needed to compare with mod-
els. The pre-paradigm experimenter had not measured things the HO -paradigm
modeler considered from his new prospective to be critically important. In addi-
tion, as Kuhn suggests,15 without some expectation as to what the results "ought"
to look like, the experimenter was satisfied with what he got.

   In the period 1972-75, the Statewide Air Pollution Research Center (SAPRC)
at the University of California at Riverside (UCR) under the leadership of Jim
Pitts,  and with funds from  both the state of California and the U.S. EPA, de-
signed and built a chamber facility specifically to produce data for model vali-
dation. By 1975, the first of these data became available to modelers and a new
round of improving the characteristic  "reasonable agreement" between theory
and experiment began. Meanwhile, the sensitivity analysis work done by Hecht et
a/, provided guidance to  kinetists as to what reactions were critical, and so, new
kinetics information was  also becoming available. In addition, the stratospheric
ozone  and SST issues generated a large interest in kinetic data needed to produce
credible models of the stratosphere; much of this work was directly applicable to
the air pollution models, especially for the inorganic reactions. As described In
the previous chapter, however, the production time for such "facts" is 2-3 years
and therefore there is always a lag between new mechanism concepts and experi-
mental data.

Durbin, Hecht, Whitten	Second Explanation

    The first victim of this new data was the generalized HSD mechanism.  Whit-
ten, in the introduction to his  1977 EPA report34 said,
      Some problems were encountered, however, in attempting to apply the [HSD] mecha-
      nism to situations other than ... the range of concentrations and hydrocarbon mixes
      upon which it was based.

          These problems stemmed largely from the use of parameters that were based
      on smog chamber data, not on the fundamental chemistry. Incorrect fundamental
      chemistry and chamber-dependent phenomena could be compensated for or masked
      by these parameters.  No single set of parameters would fit all smog systems, and
      there was little theoretical  guidance for adjusting the parameters for systems for
      which no experimental data existed.

If you think  you have read that before, you are right. Demerjian, Kerr, and Calvert
made essentially the identical statement (quoted on page ?? of this text).

    Durbin, Hecht,  and Whitten (DHW)  suggested an alternative in their 1975
report.35 That is,
      Ideally, a kinetic mechanism would be simply the assemblage of results from kinet-
      ics studies of all reactions that occur.  In  reality, not all of the  reactions that could
      occur have been studied, and often orders of magnitude of uncertainty may be as-
      sociated with those that have been studied. In addition, the number of possible
      reactions in smog is very large. Hence, a complete mechanism is neither practical,
      because  it would include an enormous number of reactions, nor feasible, because the
      needed kinetic information  is  unavailable. The best one can do to obtain a closed
      kinetic system of chemical equations and rate constants is to use available kinetic
      information and methods for  estimating other  information, to draw  analogies, and
      to make other simplifying assumptions. The kinetic mechanism that results may be
      fairly accurate, albeit simplified, description of reality.

      The mechanism's complexity is then dictated by the criterion for "important."

DHW therefore compared the explicit mechanism's predictions  with detailed prod-
uct yields from chamber experiments (the new data) to determine if they have all
the "important" reactions. The power of the explicit mechanism, Whitten later
explained, was
      Because explicit mechanisms are based on the fundamental chemistry, a poor fit be-
      tween predictions and measurements for a given species can sometimes be traced to
      uncertainties in chemical reactions or inaccuracies in smog chamber experiments. For
      example, poor fits between  predictions and measurements for some propylene/NOx
      experiments in the evacuable  chamber at  (UCR led DHW] to hypothesize that the
      intensity from the UV light source in the chamber was decreasing more  rapidly at

Second Explanation      	     	         	   	Durbin, Hecht, Whitten
                              Explicit Mechanism Formulation

  The following example  is taken from Durbin, Hecht, and  Whitien's 1975 EPA Final Report.  It
 illustrates typical reasoning used in constructing explicit mechanisms.

     PAN Chemistry

     The formation of peroxyacetylnitrate (PAN), and its  homologs, occurs by another radical-
 radical reaction.  Reaction (44),

                         CH3(O)C02 + NO2	> CH3(O)CO3NO2                       (44)

 was discussed in  last year's final report. It was speculated there that PAN might hydrolyze on
 the walls of the UCR chamber. Although this undoubtedly could occur, an analogy to N2Os
 suggests that gas phase collisional destruction could be several orders of magnitude faster than
 surface reactions  under  ambient conditions. Thus, we presently propose that  PAN may undergo
 a thermal decomposition reaction, resulting in the rupture of the peroxy and carbon-carbon

                                PAN-^- NO3 + C02 -I- CH3O2                          (45)

 Based on data contained in Benson (1968) and Domalski  (1971), this reaction is exothermic  by
 about 14 kcal/mole.

     The occurrence of Reaction 45 is supported by the experiments of Schuck et al. (1972)
 and recent PAN decay experiments in the Riverside chamber.  In the former study, PAN was
 found to oxidize  NO to  NO2. The reaction was first order in PAN and seroth order in NO. The
 ratio of C02 produced  to PAN consumed was nearly 1.  The ratio of NO2 formation to PAN
 consumption was approximately 2 in a nitrogen atmosphere, but was much greater than 2 in
 an  oxygen atmosphere.  This is further evidence for the occurrence of Reaction 45, followed by
 NO3  + NO —* 2NO2 in the nitrogen atmosphere, and  the same reaction plus CHsOj -f NO —*
 NO2 + CH3O and CH3O  + O2 -» HCHO + HO2 and HO2 -I- NO -* N02 + HO  in  an oxygen
 atmosphere.  Schuck et al.'s rate constant,  2.06 x 10~2 min"1, is 10 times that obtained  from the
 half-lives observed in the UCR chamber. This difference may  be due to additional wall decom-
 position  in Schuck's reactor. The Riverside half-lives of 5.7 ±0.1 hours  in the light  and  5.5 ± 0.4
 hours in the dark provide further confirmation that  PAN does not photo-decompose at an appre-
 ciable rate (Leighton, 1961).

     Continued on next  box.
      short wavelengths than at long wavelengths.  Subsequent measurements on replace-
      ment light sources  at UCR were consistent with this hypothesis.
For an example of uncertainties in  chemical reactions see the box 'Explicit Mecha-
nism Formulation.'

Durbin, Hecht, Whitten                                     	Second Explanation
                         Explicit Mechanism Formulation, continued

  The N20r, analogy referred to by DHW in the box above was

                                   N2O5	> NO2 + NO3

 In 1977, based on detailed kinetic experiments, Hendry and Kenley published26 what is now
 known to be the correct  reaction for PAN decomposition:

                                PAN	> NO2 + CH3(O)C02                            (45)

 which is just the reverse  of its formation reaction.  Thus, the OO-N bond is the weaker one,
 not the CO-ON bond as  was suggested in the DHW analysis.  Sometimes analogy works and
 sometimes it does not.

     Subsequent understanding of the PAN system suggested that wall decomposition in Shuck's
 reactor had little to  do with his measured life-times.  This example of a  counter-instance to a
 theory-based hypothesis is consistent with Kuhn's description of how  scientists often deal with
 such  anomalies.

     Hendry and his  co-workers were more explicit than most modelers in their 1978 EPA re-
 port*7 saying,  "During this year, the following four major developments  in laboratory data set
 our effort apart from the earlier modeling programs."  These were:  1) The observations that

                               HO2 + NO2	•- MONO -I- 02

 does  not occur at a significant rate.  2) A much higher rate constant  for

                                HO2 + NO	> NO2 -I- HO

 than  had been previously used  (e.g., 12000 vs.  2000). 3) That PAN is thermally labile. And 4)
 A breakthrough in the modeling of toluene has resulted from data on the initial products of the
 reaction of HO  and  toluene under conditions applicable to the atmosphere.
    Whitten described the advantage of the explicit mechanisms as follows:
      The explicit mechanisms predict smog chamber data better than the HSD mech-
      anism, and without any adjustments  of parameters  they fit a much wider range
      of concentrations than does the HSD mechanism. They provide more detailed in-
      sight into the smog formation process.  Because they are not as empirical, there is
      a theoretical justification for applying them outside the range of concentrations and
      hydrocarbon mixes used in smog chamber experiments.  Furthermore, if chamber-
      dependent reactions are removed and appropriate atmosphere reactions are added,  an

Second Explanation	Durbin. Hecht, Whitten

     explicit mechanism can be used as a component of a regional air pollution simulation
This is equivalent to saying the the explicit mechanism is, for the most part, "real
and therefore it can be believed to map correctly the real world. This is also a
move toward the position of Demerjian, Kerr, and  Calvert.

   The mechanism developed by DHW and described in their 1975 report35 was
used by Dodge as the basis of the EPA control strategy calculation method. This
mechanism and the re-formulations produced by Dodge to use it for prediction
will be described in the next chapter.  That work overlapped the mechanism de-
velopments described next.

Late 1970« Explicit Model*.	        Second Explanation

Late 1970s Explicit Models.

In his 1977 EPA report, which was mostly concerned with developing explicit
mechanisms for  propylene and n-butane, Whitten says,34
      The objective of [this work] is to produce simulations based on scientific knowledge
      that also fit the measurements.  How well this objective was achieved can best be
      seen by examining the many figures in this report that show UCR measurements and
      our simulations.  The overall fit  for most species is good.

     As an example of Whitten's concern with fitting, I will use his  discussion of
his propylene model for the UCR data,
      In this report we present what we consider our best simulations of these runs, using
      the same overall mechanism throughout and varying only the light spectrum and
      initial MONO concentration.

           In our final fitting  procedure we began with propylene runs EC-95, EC-96, and
      EC-121, because detailed spectral measurements were available for these runs.  We
      used our photolysis constant program to calculate the carbonyl photolysis constants.
      We then varied the percent of ozonide formation in the mechanism until the total
      maintenance source of radicals from aldehydes, propylene-O atom reactions, and
      OBone-propylene reactions gave  a good fit to the UCR data. Note that the percent
      of osonide formation  is only a fitting parameter. If the quantity is measured, many
      other parameters or added reactions  could be used as fitting parameters. Some of
      these are as follows:
           o  Reactions of  ozonides with radicals to act as sinks, or reactions that produce
              or release radicals.
           >  A slower or faster rate  constant for the oione-propylene reaction.
           >  Larger or smaller quantum yields for carbonyl photolysis
           >  Fundamental changes in the mechanism, such as PAN chemistry, NOx loss
              chemistry,  and  aldehyde formation.

           During  the fitting procedure for propylene runs, we discovered that the photoly-
      sis constants calculated from the spectral data for runs EC-95 and EC-121 differed.
      Of course, it was still possible to produce good fits by arbitrarily varying the pho-
      tolysis constants in the simulations of these runs. ... Thus, we had to lower radical
      production from  the ozone-propylene reaction by setting the tuning parameter,  the
      fraction of ozonide formation, to a high value, 0.33.

           /And so forth.}

Because he was  so concerned with fitting the measurements, at that time Whitten
was often accused of being too much an instrumentalist and too little a realist.58

Second Explanation	Late 1970i Explicit Models.

    To understand the dilemma, however, let us contrast Whitten's discussion
with that of Carter  et  al.39 for the same data and time period (i.e., late 1977),
      We present here  a detailed model for the N Ox-air photooxidation of n-butarie and/or
      propene which is consistent with  most of the extensive smog chamber data obtained
      at SAPRC for propene and n-butane, and detailed comparisons between model pre-
      dictions and experimental data are shown.  These efforts to produce a satisfactory
      model for the n-butane + propene systems have led to the identification of a num-
      ber of experimental problems and mechanistic questions in need of further study, and
      these are discussed here.  ...

          Fits were usually attained to within ±20% or better ...

          The good fits to experimental data were attained only after adjusting several
      rate constants or rate constant ratios related to uncertainties concerning chamber
      effects or the chemical mechanism. The largest uncertainty concerns the necessity to
      include in the mechanism a significant rate of radical input from unknown sources
      in the smog chamber. Other areas where fundamental kinetic and mechanistic data
      are most needed  before a predictive,  detailed propene + n-butane-NOx-air smog
      model can be completely validated concern other chamber effects, the O3  •+• propene
      mechanism, decomposition rates of substituted alkoxy radicals, primary quantum
      yields for radical production as a function of wavelength for aldehyde and ketone
      photolyses, and the mechanisms and rates of reaction of peroxy radicals with NO
      and NO2.

In his "Normal Science as  Puzzle-solving" chapter, Kuhn says,
      If it is to classify as a puttie, a problem must be characterized by more than an
      assured  solution. There  must also be rules that limit both the nature of acceptable
      solutions and the steps by which  they are to be obtained.

          If we can accept a considerably  broadened use of the term 'rule'—one that will
      occasionally equate it with 'established viewpoint' or with 'preconception'—then the
      problems accessible within a given research tradition display something much like
      this set  of puzzle characteristics.  The man who builds an  instrument to determine
      optical wave lengths must not be satisfied with a piece of equipment that merely
      attributes particular numbers to particular spectral lines.  He is not just an explorer
      or measurer.  On the contrary, he must show, by analyzing his apparatus  in terms of
      the established body of optical theory, that the numbers his instrument produces are
      the ones that enter into theory as wave lengths. If some residual vagueness in the
      theory or some unanalyzed  component of his apparatus prevents his completing that
      demonstration, his colleagues may well conclude that he has measured nothing at all.

    I believe that within the 'scientific community' there is something subtly more
acceptable about how Carter said essentially the same thing that Whitten  said.
Whitten appears to have 'knob-twiddled' while Carter used 'scientific uncertainty.'

The Beginning of the Firat Criai»	Second Explanation

   While some in the modeling community may have found subtle differences
between these two, the conclusions reached by many scientists and poKcy makers
in the air pollution community after being exposed to such discussion was "We
understand nothing at all!"  For example, even in 1984, Leone and Seinfeld said,
"It is our feeling that with the current controversy regarding the magnitude of
chamber radical sources, it is difficult to consider any mechanism validated simply
because it can simulate chamber data."

The Beginning of the First Crisis

So there was an impression  that one group of modelers  were adjusting .the model
in such a way that  it was not clear that the values were the correct ones to enter
the theory, while the other group  of modelers had a vagueness in their theory and
a unanalyzed component in their  apparatus.

   This perception began to undermine the promise that had come with the
adoption of the HO -chain paradigm and to dampen the exhilaration that arose
from the initial successes of its application.
                              There's no sense being precise about something
                       when you don't even know what you are talking about.
                                                     —John von Neumann

Second Reformulation
The Dodge Mechanism

While the field was turning to explicit mechanisms for their explanatory powers
there was a pressing need for practical prediction. To deal with this situation,
Marcia Dodge produced an interesting re-formulation of the existing paradigm.40
She used the explicit mechanism for propylene and butane developed by DHW
and Dimitrlades' BOM automobile exhaust smog chamber data to produce an
isopleth diagram that was taken to be representative of the  "worst case" ambient
polluted urban atmosphere. If that seems like a major extension of the paradigm
to allow an application, it was. The formulation of this concept developed in the
interaction between Dimitriades' desire to extend his smog chamber results to the
atmosphere and Dodge's acute awareness of the limitations of the 1975 models.

   To highlight the latter point, Dodge had commissioned the Durbin,  Hecht,
and Whitten (DHW) report. In this report, DHW had presented forty pages of
chemical kinetics discussion to support their formulation of the propylene/butane
mechanism. They said that smog chamber studies done  at UCR and at Battelle
"serve as the data base for validating the kinetic mechanisms." These consisted of
five propylene and four butane chamber runs from the UCR chamber, and five
propylene runs in the Battelle chamber.  DHW had described the comparison
of prediction and measurements for the propylene experiments as being "fairly
good" for NO, NOj, and Oj, while carbonyls were "consistently low" and PAN pre-
dictions were "some times high and some times  low." They  remarked that  "The
chemistry of PAN is still largely unexplored, and this is very  likely a large part of
the problem." They said,
     The butane mechanism was considerably  less successful than the propylene mecha-
     nism, even though they  are similar.

The Dodge Mechanism	 Second Reformulation

         Of course, the traditional scapegoats, uncertainties of surface and photqjysis
      reactions, can also be blamed for the disparity between model predictions and exper-
      imental results. Although they probably exacerbate the problem, they are not the
      sole culprits.

         Although the accuracy was not very good, the propylene mechanism was still
      able to follow the behavior of each species.

         The mechanism's application utility was demonstrated in a study of hydrocarbon
      reactivity and ozone formation. Thus the mechanism can be a useful tool to investi-
      gators of photochemical air pollution.
Thus the claims were less bold this time. Nevertheless, there was the suggestion
that the mechanism might be useful in  prediction, if not entirely acceptable for

    Dodge accepted this proposition. She chose the DHW propylene/butane mech-
anism as the starting point for her own modeling tests.  I think, however, that
Dodge was skeptical  about both the utility of the DHW formulation and the qual-
ity of the nearly 10-year old BOM data until she was successful in fitting the data
with  the mechanism.  Kuhn says,15
      When measurement is  insecure, one of the tests for reliability of existing instru-
      ments and manipulative techniques mast inevitably be their ability to give results
      that compare favorably with existing theory.

    With a very reasonable set of assumptions about the photolytic rates and the
chamber characteristics of the BOM chamber runs,  Dodge was able to achieve a
"goodness  of fit" between the predictions and the data that was less (ca. 10%)
than  the BOM replicate run agreement (ca. 15%) while only using  a single ad-
justable parameter: the relative concentration of the initial butane and propylene.
Dodge discovered that the reactivity of auto-exhaust in the BOM chamber could
be well represented by the DHW propylene/butane  model by assuming that  the
initial NMHC in the auto-exhaust was represented as 75% butane and 25% propy-
lene.  Thus, the mechanism—now known as the Dodge Mechanism—was able to
predict ozone formation, but could not explain how it occurred.  That is, the  ozone
formation  in the BOM chamber certainly did not occur because of  reactions  ex-
clusively for butane and propylene. Dodge was careful to avoid the word  validate;
it was only used once in the whole paper and then to refer to the chamber data,
not the model.

    Dodge  had not validated this kinetics mechanism, she had vindicated its use
in atmospheric applications. Vindication means35 "a fact or circumstance that
justifies" and justify  means "to show an adequate reason for  something done."


Second Reformulation	Falls and Seinfeld, McRat Mechanism

Vindication does not carry that sense of "well grounded on principles" which is
associated with validation; it just implies that one can assert a claim. So, based
on her ability to predict the ozone concentrations in the BOM chamber, and on
the fact that there were mechanistic representations of processes in the model,
Dodge asserted that she could adjust the model conditions to be "representative
of the polluted urban atmosphere." This she did.

    This methodology proved to be such a useful strategy  that OAQPS has con-
sidered using the Dodge Mechanism 10 years after it was formulated and vindi-
cated.  In view of the enormous success EPA has had with this strategy, and its
demonstrated scientific  utility, one wonders about the utility of vindication verses
the hope of validation.

Falls and Seinfeld, McRae Mechanism

In their 1978 paper,41 Falls and Seinfeld (FS) continued the approach used by
Hecht, Seinfeld,  and Dodge.  That is, they constructed a mechanism containing
fictitious entities and stoichiometric coefficients for representing the chemistry
of these entities.  FS described two  approaches for atmospheric mechanism de-
    >  surrogate mechanism—a mechanism in which organic species in a partic-
       ular class, e.g., olefins, are represented by one or more members of that
       class, e.g., propylene;
    >  lumped mechanism—a mechanism in which organic species  are grouped
       according to a common basis such as structure or reactivity.
FS  concluded that "because of the  computational requirements associated with
calculating chemistry and transport simultaneously, surrogate mechanisms are
inappropriate for use in atmospheric  models that include an adequate treatment
of meteorology."  FS further concluded that in  1978,  "insufficient information is
available to include aromatics species"  in the mechanism under development, and
therefore, the mechanism only included olefins, paraffins, and aldehydes.

    With regard to mechanism formulation, FS said,
     In view of the complexity of the HO  and O3 reaction paths for individual hydro-
     carbons, the major problem in developing a lumped mechanism is representing the
     chemistry of a mixture  of alkanes,  alkenes, and aldehydes. We choose to partition
     the organic species into five classes: ethylene, higher olefins, paraffins, formaldehyde,'
     and higher aldehydes....

         Because both rate constants and mechanisms differ among the species that  com-
     prise a lumped class, lumped  mechanisms, by necessity, involve parameters that must


Falls and Seinfeld, McRae Mechanism                                        Second Reformulation
      be chosen within theoretically defined bounds to provide the best fit of data and pre-
      dictions. The values of such parameters cannot be defined precisely . . . The art of
      developing a mechanism is to deal with the inherent uncertainties in an optimal
      manner, while adhering to certain important constraints, such as maintaining car-
      bon, nitrogen, and oxygen balances —

          We have opted to concentrate the inherent uncertainty of the lumped mechanism
      in the RO chemistry. . . .

                     RO - >aHO2 + (1 - a)RO2 + 0HCHO + -yRCHO

      Thus, the lumped olefin-HO and paraffin-HO  reactions are written as producing
      RO2 radicals that,  when reacting with NO give NOj and a lumped RO radical. The
      stoichiometric coefficients obey the following restrictions 0 < a < 1, 0 < £ < 1, 0 <
    FS were very careful not to use the word  validate when describing the tests
of their mechanism; instead they evaluated the mechanism using smog chamber
experiments. Evaluation means33 "to determine the worth of; to appraise."  FS
      In evaluating a mechanism, the customary procedure  is to compare the results of
      smog chamber experiments, usually in the form of concentration-time profile, with
      simulations of the same experiment with the proposed mechanism.  In all mecha-
      nisms a sufficient number of experimental unknowns exist such that the predicted
      concentration profiles can be  varied somewhat  by  changing rate constants (and per-
      haps mechanisms) within accepted bounds.  The inherent validity of a mechanism can
      be judged by evaluating  how  realistic the parameter values used are and how well
      the  predictions match the data. Since the mechanism will generally not be able to
      reproduce every set of data to which  it is applied, two factors must be considered:
      identification of the major sources of  uncertainty, such as inaccurately known rate
      constants or mechanisms of individual reactions, and evaluation of so-called "chamber
      effects", phenomena peculiar  to the laboratory system in which  the data have been

    Recalling that  validity  means "strength or force from being supported by
fact,"  in effect, FS said, "The strength of a mechanism that arises from being
supported by fact can be appraised by determining how well the parameters con-
form to 'real' values and how much 'reasonable agreement' there is between the
model predictions and the  facts." In this particular case, I find it hard to accept
this assertion.  There are no  "real" values of the parameters in  the chemistry of
the chamber; these are fictitious entities. Therefore, such parameters cannot be
specified  before hand for a particular  mixture, but must be obtained by trial and
error fits to the data, that  is, they are empirical parameters and are thus limited
to the range of conditions used to select their values.  For example, FS said, "For


Second Reformulation	Falls and Seinfeld, McRae Mechanism

the propylene/butane system, good simulations results were obtained when the
RO decomposition step was taken to give RCHO and HOj exclusively."  In other
words, in spite of the accepted fact that propylene yields one HCHO for  every HO
+ propylene reaction, the FS model fits the chamber data best when this fact is
ignored even when large concentrations of propylene were present.

    With objection to FS's first criterion, I looked to the second, reasonable agree-
ment between prediction and data.  For demonstration of "reasonable agreement"
FS simulated
     >  three n-butane experiments,
     >  eight propylene  experiments,
     >  five propylene/n-butane experiments.
The evaluation  data base barely exceeded the number of smog chamber experi-
ments  used to test the  original Hecht, Seinfeld, and Dodge mechanism four  years
before.  There were no  plots of data and predictions shown in this paper, only a
table of 03, NO2 and PAN maximum value predictions and chamber values were
given as demonstration of "reasonable agreement". Failure to use all the chamber
data available (more than 44 experiments) and to use a higher level of reason-
able agreement criterion meant that FS had applied insincere tests to their  model.
Perhaps this is why the paper ended with
     Although the revised mechanism contains significant new information on rate con-
     stants and some mechanisms, there still exist, in addition to aromatic chemistry, key
     areas of uncertainty that require experimental study.

    While in the FS paper there was no claim or assertion of adequacy  for pre-
diction, this did not prevent McRae from later claiming  in Chapter 8 of "Mathe-
matical Modeling of Photochemical Air Pollution"42 that the FS mechanism was
selected because it "has been validated against a wide range of smog chamber ex-
periments." " Furthermore, in McRae's description of the mechanism, aromatics
had been added as a lumped species and their chemistries were represent by one
                          HO  + ARO	>R02 + RCHO

With similarly little justification or explanation, the values of the parameters for
RO were assigned a = 1, 0 — 0.5, 7 = 0.5..

    The choice of 0 and f has a certain physical significance as was described by
Killus  in discussing a similar reaction in the Demerjian Photochemical Box  Model

Fallt and Seinfeld, McRae M«chani«m	Second Reformulation

mechanism. It is the only choice that preserves carbon balance. Killus pointed
out that the cycle

                                     NO2  + R02     +   CO2
                                     NO2  -I-         RO
                                     HO2  + 0.5HCHO +   0.5RCHO

only preserves mass balance if the coefficients in the last reaction are exactly 0.5!
This is because of  the infinite series:

                1 = 0.5 + 0.25 + 0.125 + 0.0625 + 0.003125 + ...

If 7 was larger than 0.5, then the system would gain carbon mass each cycle.
This un-realistic production of R02 was recognized by Falls, McRae, and Sein-
feld43 but they suggested with no further elobrations that it might have a small
effect when the initial HC-to-NOx ratio was large and "other modes of radical pro-
duction besides RO were available."

    Seven SAPRAC smog chamber experiments using multi-component urban
mixtures were simulated with the McRae version of the FS mechanism (which
was later called the CalTech Model Mechanism).  Plots were shown for these sim-
ulations. There was no  demonstration of the accuracy of representation of the
different HC classes in the mechanism (certainly the new aromatics chemistry
was not justified).  Nor was there any attempt to address compensating errors
among the HC classes  (e.g., olefins under-reactive and aromatics over-reactive).
The same set of stoichiometric coefficients were used for all mixture compositions
which clearly violated the formulation concept described by FS and quoted above.
No evidence was presented to show that the McRae version of the FS mechanism
could adequately represent the daily composition change expected in urban simu-
lation conditions. Again the model had been insincerely tested, now even ignoring
its own  call for multifaceted explanation and justification.

   At the same time  the mechanism testing was  described, McRae et  al. were
      Model validity refers  to the essential correctness of the model in terms of its repre-
      sentation of the basic chemistry and physics as well as to its accuracy of numerical
      implementation  as  measured by adherence to certain necessary conditions such as
      conservation of mass.
The CalTech Model was criticized by Killus and Whitten in part on the basis
that it did not adhere to conditions of conservation of mass.


Second Reformulation	Carbon Bond Mechanism

Carbon Bond Mechanism

To deal with the need to have a 'practical' mechanism to predict in the atmo-
sphere, Durban, Hecht, and Whitten suggested in their 1975 report that a new
approach, different from that of HSD, and based on explicit chemistry be devel-
oped.  They said,
      After validation, the explicit mechanism can then serve as a basis for rederiving a
      generalized mechanism and for checking the accuracy of lamping techniques. We
      should emphasize that the explicit approach is an interim step. We hope that it will
      serve to further clarify smog kinetics and thus to lay the groundwork for the more
      practical generalized approach.
Whitten later (1977) added, "A condensed version of the explicit mechanism
would combine the advantages of a basis in elementary chemical reactions and
speed of computation."

    In describing the condensation method—which was called the carbon bond
approach—Whitten said,
      But the most important difference is the scientific basis: whereas the HSD mecha-
      nism was of necessity somewhat empirical, -the carbon bond mechanism is derived
      from the fundamental chemical reactions in smog as represented by explicit mech-
      anisms.  Thus the carbon bond mechanism is a condensation of our understanding
      rather than a parameterization of our uncertainty.

    The basic idea put forth by Whitten was that a practical predictive mecha-
nism ought not be created directly. He suggested that we have no evidence that
such a mechanism is likely to succeed in prediction in applications that exceed its
validation data base; in fact, we have evidence that this approach does not work.
On the other hand,  according to Whitten's approach, an explicit mechanism is
supposed to  be based on scientific theory. And because it has a scientific repre-
sentation for each important process thought to occur in  air, he believes we have
a basis for believing that it can function adequately outside the range of its vali-
dation database, because, these mechanistic representations are  supposed to have a
true correspondence  to the real world.  Therefore, if we base our practical predic-
tive mechanism on the explicit mechanism, we can, in turn, model this correspon-
dence to the real world with fewer entities. The basis for testing the latter mech-
anism is the  explicit mechanism from which it was  derived, not the real world.  A
potential dilemma here is that the compressed mechanism may fit  the real world
data better than the explicit mechanism, but Whitten would have to say, "for the
wrong reasons."

    Whitten, Hogo,  and Killus state in their 1980 paper,44


Carbon Bond Mechanism	Second Reformulation

      The carbon-bond mechanism is a generalized kinetic mechanism in which most car-
      bon atoms with similar bonding are treated similarly, regardless of the molecules in
      which they occur—

          The carbon-bond mechanism is termed a "condensed" mechanism because it is
      a condensation of explicit mechanisms developed to simulate smog chamber experi-
      ments that were initiated with only one or two hydrocarbon species present—

          With the exception of aromatics chemistry for which no explicit mechanism ex-
      ists, the reactions in the carbon bond mechanism represent essential features of four
      explicit mechanisms developed by Whitten and Hogo.**

          ... the aromatic oxidation reactions are assumed to produce the same products
      as the olefin oxidation reactions ... The mechanism for slow double bonds is an in-
      terim solution  to a difficult problem, that of aromatics oxidation.

The carbon  bond concept eliminates the concept of average molecular weight be-
cause carbon atoms undergo reaction independently, so there is no stepwise oxida-
tion of a hydrocarbon chain and there is explicit carbon mass balance.

    Whitten tt a/, said,  "The carbon bond mechanism was developed and vali-
dated with the predictions of explicit mechanisms for  smog chamber experiments
performed at UCR during 1975 and 1976." This was true for the paraffin and
olefin portion of the mechanism.  For the aromatics portion, the model was tuned
directly against chamber data and Whitten  tt al. said that this portion of the
mechanism was "more properly an empirical mechanism than a condensed one."
A table of 44 SAPRAC chamber simulations was presented, listing four measures
of model  agreement with data. Fifteen plots showing model agreement with data
were also shown.

    Whitten et al. made a strong point with regard to the comparison of the mech-
anism with data,
      The carbon bond mechanism is intended to produce predictions similar to those of
      explicit  mechanisms in less computing time. Consequently, the carbon bond mecha-
      nism is  properly validated with data from simulations of smog chamber experiments
      using explicit mechanisms rather than with data from the smog chamber experiment
      itself. In other words, the standard for judging the carbon bond  mechanism is the
      explicit  mechanism. A fit between  the predictions of the carbon bond mechanism
      and  experimental data that is closer than the corresponding fit for an explicit mecha-
      nism must be regarded as fortuitous.

In his grant  report, however, Whitten said,
      Periodic updating of generalized mechanisms such as the CBM it to be preferred to
      continuous updating.  Changes in one reaction may require compensating changes
      in other reactions to maintain the overall predictive accuracy of simulations using


Second Reformulation	Carbon Bond Mechanism

      the mechanism.  Consequently, after a change the mechanism should be tested with
      an entire set of smog chamber data to ensure that no special problems have been
      created that would create difficulties in atmospheric applications. The costs of such
      testing makes it desirable to test the effects of several changes  at once.

So here he seems to be suggesting that direct testing against chamber data is
used to ensure that no special problems have occurred in modifying the  CBM
mechanism. I would assume from the former statement, that explicit mechanisms
would have to be re-done, and the CBM re-derived, not that the CBM would
be adjusted against chamber data. I believe that what actually happens is that
Whitten initially creates the CBM from explicit mechanism, but then  makes sub-
sequent adjustments directly to the  CBM, with the implicit assumption that these
adjustments really reflect changes to the explicit mechanism. Therefore the two
mechanisms, explicit and CBM, are slightly out of step most of the time. This is
actually substituting the model of the model for the model.

    A recurring problem for the evolution  of CBM to CBII  to CBIII was the un-
known nature of the aromatics chemistry.  Whitten said,
      Our need for a mechanism that adequately simulates the fate of the aromatics pre-
      cludes the option of waiting for an  accurate explicit description of the chemistry.
      Since a validated explicit aromatics mechanism similar in accuracy to those available
      for propyiene and butane does not exist, our efforts to produce a condensed kinetic
      mechanism for aromatic compounds must be viewed as conjectural.

          We analysed the explicit mechanism proposed by Hendry as an explicit descrip-
      tion of toluene oxidation chemistry. The explicit mechanism does not satisfactorily
      describe ozone formation and limitation. Thus, we are forced to rely on empirical
      relationships combined with our best speculations as to the true nature of the chem-
      istry involved.

By  the 1980 final report, ARO chemistry had been modified several times  to ac-
count for new information.
      Approximately 73% of the oxidation pathway leads to glyoxal photolysis in both the
      explicit and CBII mechanism.  One  notable difference has been a prediction, using
      the new chemistry, that the  addition of aromatics to a mixture of olefins at high HC-
      to-NOX ratios would suppress osone formation.

In spite of the successful prediction  of the suppression effect of adding high con-
centrations of toluene to propylene-NOx systems, we now know that the aromatic
chemistry in CB-II or CB-III is not  correct; we still do not  know the correct chem-

    In his EPA report Whitten said, "... results suggest that more of the uncer-
tainty in  a CBM simulation of a smog chamber experiment was caused by defi-
ciencies in knowledge of smog chemistry or inaccuracies in smog chamber data


Atkinton, Lloyd, Winges Mechanism _ Second Reformulation

than by the approximations and assumptions on which the CBM is based."  What
has been claimed here is that CBM adequately reproduced the explicit mecha-
nism.  We should carefully note that the explicit mechanism may not  itself be an
adequate model of the data. Also, you should remember this paragraph when we
later discuss the Leone and Seinfeld rationale for producing their "master mecha-
    In their conclusion in the 1980 paper, Whitten et al. said,
      We believe that the carbon bond mechanism offers advantages in realism and ease
      of application. These advantages include: carbon mass conservation, elimination of
      assumptions concerning average molecular weights, reduction of the span of rate con-
      stants covered by each grouped species,  and reduced computing requirements.

         ... we believe that presently available photochemical kinetic mechanisms are
      sufficiently developed  for most applications to the atmosphere.

    We must be careful to interpret the  term "realism" used above correctly.  Cer-
tainly the carbon bond mechanism is  not a realistic representation of atmospheric
chemistry — there are no OLE, PAR, ARO,  or CARB species in the atmosphere! The
carbon bond mechanism is a model of a model and therefore it suffers any inac-
curacies and imperfections of the original model plus additional generalizations,
deletions, and distortions caused by its own modeling process. Conceptually at
least, the latter should be easy to understand  and access because both models are
available for testing and comparison.  We should also recognize that the modeling
of a model process is similar to the original model construction in that pieces are
assembled to make an image of the whole. Each explicit mechanism may be de-
veloped independently of each other, and while each may work on the relatively
small range of chamber conditions used  to develop it, the combined effect of all
pieces must be tested using experimental mixtures that vary the composition in
ways  that approximate the urban atmospheric variations. Even at the conclusions
of such carefully  advancing testing, only those reactions described in the explicit
chemistries would have been examined.  And so the real world' (and all the reac-
tions  that could or do exist)  still awaits.

Atkinson, Lloyd, Winges Mechanism

Roger Atkinson,  Alan Lloyd, and Linda Winges (ALW) in their 1982 paper45
present- another alternative to the need for a practical  predictive mechanism:
      It is hoped that an intermediate approach such  as this, which is in-between a "very
      lumped" mechanism and a multi-hundred reaction explicit mechanism,  may pro-
      vide a better  "fidelity" as regards the predictions of time-concentration profiles  for


Second Reformulation	Atkinson, Lloyd, Wingta Mtehanigm

      a wide range of organic-NOx photooxidations. Additionally, it is hoped that this
      intermediate  approach will be compatible with the more recent detailed ambient
      atmospheric analyses, and sufficiently flexible to be readily used for explicit single
      hydrocarbon/NOx irradiations simulations.
The "very lumped" mechanisms referred to above were the ELSTAR model; the
McRae or CalTech model; and the Carbon Bond models.

    ALW said, "This chemical model is based on kinetic, mechanistic  and product
data from the literature, and the reaction rate constants and mechanisms are gen-
erally consistent  with the recent evaluations of NBS and CODATA for the simpler
species, and with the recent evaluation of Atkinson et a/., 1981 for the more com-
plex organics." In addition, ALW made extensive use of the Carter simulations in
1979 explicit mechanisms for  propylene and n-butane and of Atkinson's 1980 aro-
matic simulations.  As we shall see in the next chapter, the evaluation of Atkinson
et al, referred to  in this paper would have a significant  influence on future mecha-
nism development.

    The ALW mechanism had twelve HC classes: two types of alkanes; three types
of alkenes; three  types  of aromatics; three types of aldehydes; and one type of ke-
tone. The low reactivity in each class was an explicit compound, e.g., propane,
ethylene, formaldehyde, benzene, while the higher classes tended to be average
representations of several species.

    The mechanism was "tested"—validate or evaluate  was never used—against
21 UCR smog chamber irradiations.  These tests included: one each of six individ-
ual species butane, ethene, t-2-butene, propene, toluene, m-xylene; a single four-
alkene mixture irradiation; three seven-hydrocarbon mixture irradiations; seven
multicomponent  hydrocarbon irradiations; and four hydrocarbon-NOx-SOj irradia-
tions.  A continuous HO radical source that varied with each run was used in the
simulations. Fifty-seven comparisons of data with prediction plots were included
in the paper.  ALW said,
      As can be seen, [from the plots] the fits are generally very good, being typically
      within ±20% of the  experimental values for 03, intermediate carbonyls and PAN.
         In general, the fits of calculated to experimental data for the evacuable chamber
      multihydrocarbon-NOx-air irradiations were good, as may be expected from the gen-
      erally excellent fits of the single hydrocarbon-NOx-air irradiations.  For the All Glass
      Chamber, the fits were again judged to be reasonably good, since the hydrocarbon,
      initial NO and NO2 and O3 profiles are accurately predicted.  ...

Meehani«nM Compared	Second Reformulation

         Thus, the reaction mechanism has been tested against 21 smog chamber irradia-
      tions, yielding good fits between the experimental and calculated time-concentration

    No assertions were made based upon the last sentence quoted above. In the
conclusions, ALW said,
      The new mechanism has been tested extensively against a variety of smog chamber
      data for single hydrocarbons, binary, seven component and surrogate hydrocarbon
      data.  The mechanism gave good agreement between  predicted and measured time-
      concentration profiles for all measured species.  [A set of simple reductions] produces
      a viable chemical mechanism for inclusion into air quality simulation and reactive
      plume models.

    What is a "viable chemical mechanism" ?  Viable means33 "able to live; specif-
ically at that stage of development that  will permit it to live  and develop under
normal  conditions." If we take 'live and develop' to mean  'function1 then we can
interpret the conclusion to mean "The mechanism  is at a stage of development
that will allow it to function under normal conditions in an air quality simula-
tion." Compared to some other modeler's.conclusions this can be seen as a mod-
est statement.

A Contrast

I am somewhat surprised at how the outcomes of testing the  mechanisms dis-
cussed in this chapter were described as being so good and at the strengths of
claims made by the authors for effectiveness in atmospheric prediction in contrast
to the "uncertainty" and lack of explanation attributed to the detailed explicit
mechanisms in the last  chapter.

Mechanisms Compared

By 1980, EPA policy makers were aware that they  were depending on  "known
incorrect chemistry" in the Dodge Mechanism (yet it predicted correctly, didn't
it?), and with so many  mechanisms making claims  of goodness of prediction, they
wanted  to know how to treat new State  Implementation Plan revisions that used
different mechanisms.

    Until this time, each mechanism had only been tested  by  its authors and had
never been compared with each other. Jeffries, Sexton, and Salmi  (JSS) at UNC

Second Reformulation	Mechanising Compared

were given a contract6 to "examine the effects of input choices and modeling meth-
ods likely to be used in 'Level II' and 'Level HP-type SIP applications on the pre-
dicted ozone precursor control requirements." A Level II analysis required a sim-
ple trajectory model incorporating effects of pathway, emissions, mixing height
changes, entrainment from aloft, and chemistry be used to simulate days of high
Oj formation.

   Four different chemical mechanisms were examined by JSS (Dodge Mecha-
nism, Carbon Bond II Mechanism, Demerjian Box Model Mechanism, CalTech
Mechanism).  In setting the scope of work, it was argued by EPA that, because
the Dodge Mechanism had been successful in simulating the BOM data  base, the
other three mechanisms would likewise have to be tested with the same  data base.
First, the BOM data base was the only reasonably complete auto-exhaust data
base. Second, because of the empirical nature of the 75:25 butane:propylene split,
the Dodge Mechanism could not be used to simulate other HC mixture systems.
Yet, the newer mechanisms all made  at least an implicit claim that they could
simulate various HC mixtures, including urban atmospheric mixtures, which EPA
assumed were well represented by dilute auto-exhaust such as that in the BOM

   The 31 BOM irradiations were simulated with each of the four mechanisms.
A single consistent set of chamber assumptions were used for all mechanisms
(e.g., the same photolytic rates, the same auto-exhaust composition, but uniquely
represented in each mechanism). From a plot of observed Oj vs. predicted Os for
each mechanism, it appeared that each mechanism showed "reasonable agree-
ment" within ±25%, although the CalTech mechanism had several low predic-
tions. Comparison of cross-sections of isopleth diagrams with all four mechanisms
plotted on one cross-section, however, revealed differences among the mechanisms.
For example, for the CalTech mechanism, predictions for higher NOX were very
low compared to the other mechanisms and to  the experimental data. Inspection
of the CalTech mechanism revealed several major differences in its use of coeffi-
cients for olefin and aromatic products compared to the other mechanisms. The
Carbon  Bond and Dodge mechanisms' predictions converged at low HC,  while the
Carbon  Bond and Demerjian mechanisms' predictions converged at high HC. JSS
said, "Compared to the experimental points, however, it would be difficult to say
which mechanism is a better description of the BOM results." They further con-
cluded that "the 'noise' in the experimental results is greater than the differences
shown among the Dodge, Demerjian, and Carbon Bond II mechanism simulations
which have large differences in construction assumptions." And finally,

Mechaniam* Compared	Second Reformulation

      Given the range of uncertainty in the BOM data (e.g., Oz maxima and initial alde-
      hydes) and the sensitivity of mechanisms to necessary assumptions, it is concluded
      that the Demerjian, Dodge, and Carbon Bond II mechanisms were all capable of pro-
      viding adequate descriptions of the 03 formation in the BOM chamber; the Carbon
      Bond II mechanism is probably slightly better at describing the other measured sec-
      ondary products.

    This was the good news. Now for the bad. In atmospheric simulations, the
models predicted significantly different control requirements for the three days
on which the meteorological conditions were judged to be reasonably represented
(i.e., of the 10 RAPS days selected by EPA, the trajectory calculation methods
were so poor on seven of the days that CO could not be reasonably predicted us-
ing a meteorology and emissions-only trajectory model, and thus a comparison
of chemistry predictions for these  days would have been meaningless).  In some
cases for the three good days, the atmospheric Os isopleth diagram produced by
the Carbon Bond and CalTech mechanism did not allow an  EKMA-type solution
to the control requirement. That is, HC-to-NOx ratio line for the  day did not in-
tersect any isopleth at all! This was a failure of the EKMA approach itself. For
the conditions in which the EKMA approach did produce a solution, the differ-
ent mechanisms exhibited very different AOS/AHC for the  different atmospheric
simulations and therefore they predicted very  different HC  control requirements:
the range of control requirements  was from 15% to 70% HC reduction needed de-
pending upon the day, the mixing height profile,  and the mechanism used. The
most important factor was the day with its O3 aloft assumption; the second most
important factor was the choice of the mechanism.

    JSS conclude, in part,
      The chemical mechanisms used in this study are probably inadequate descriptions
      of the real  atmospheric processes because they were not developed by comparison
      with data bases which included the type and range of variations which apparently
      occur in the atmosphere.  That is, none of the mechanisms have been tested against
      data bases  in which the hydrocarbon composition,  including the aldehyde mole frac-
      tion, had been systematically varied, and in which  dynamic dilution and injection
      of new reactants had occurred, and in which dynamic addition of O3 such as from
      entrainment aloft had occurred, or in other factors similar to atmospheric processes
      had been used. These data bases do not exist ... It will require years to  develop
      the needed  databases to insure that the chemical mechanisms are giving adequate
      descriptions of the atmospheric chemical processes.

    In a Kuhnian interpretation, this would be equivalent to saying that an in-
sufficient amount of the second class of normal scientific work had been done to
justify the extension of the paradigm to the atmosphere. That is, the "solution

Second Reformulation	Mechanism* Compared

to the puzzle" had been forced—yet, until this point, accepted by the commu-
nity. JSS also suggested that the method used by EPA to 'correct' the model's
predictions when they did not agree with the real world, was a major cause of the
prediction problem. That is, EPA used the mechanisms' predictions in a relative
way, assuming that they would perform better that way. In fact,  JSS said it made
them perform worst.

    It is little wonder then, that the JSS study had an impact on the  air pollu-
tion community.  Reviewers at the EKMA conference in 1981 were so  affected by
its implications that they suggested that EPA abandon the modeling  approach.
These folks surely felt that the modeling paradigm had failed to provide puzzles
with solutions. According to Kuhn, this should have led  the modeling commu-
nity to "either abandon the puzzle entirely or attempt to solve it  with the aid of
some other hypothesis." Yet  several research groups were attracted  to the puz-
zle presented by the fact that the mechanisms could apparently fit chamber data,
and still  they gave different atmospheric predictions.<6i<7'48i<9>5°  Some researchers
sought explanations in the mechanism details as to why they should give different

    For example, in their EPA report6 Leone and Seinfeld state,
     These mechanisms are all based on the same body of experimental rate constant
     data, and each mechanism has been evaluated against data from various  smog cham-
     ber facilities.  In each mechanism, the detailed atmospheric chemistry has been greatly
     simplified by a process referred to as lumping. However, because this lumping has
     been carried out in different ways, no two mechanisms are the same.
Their approach was therefore to compare each mechanism with "the detailed at-
mospheric chemistry," as represented by a fully explicit mechanism, as a way of
"understanding the fundamental reasons for the discrepancies."

    I believe, based on the conjunction of "the same body of experimental rate
constant data" and "the detailed atmospheric chemistry  has been simplified,"
from the quote above that there was a presupposition to the Leone and Seinfeld
hypothesis. They presupposed that if asked  to produce a "full explanation," the
model developers would have produced the same explicit mechanism, and there-
fore it was only the various lumping methods that  were the cause of different
predictions. I believe  that this presupposition was  incorrect and that  further-
more, it led them to make the same mistake whoes cause they were seeking. It
was not lumping that was responsible for the mechanisms not being the same.
Even the incomplete and brief history presented here shows that  "the same body

Mechaniinu Compared                                               Second Reformulation

of experimental rate constant data" only represents a small part of the knowl-
edge needed to construct a mechanism—the modeler's creativity and ingenuity
are fully operational in constructing his mechanism. A great deal of what appears
in the mechanism is the modeler's personal conjecture; if not, mechanisms could
be constructed by computers.  Leone and Seinfeld presumed that there was a sin-
gle explicit mechanism available in the "body of experimental rate constant data"
that represented the truth (they say  "the detailed atmospheric chemistry"), and
that this represented all that modelers had started with in their model construc-
tion. From the subsequent actions of Leone and Seinfeld, it is clear that for them
this truth was represented by information given in the Atkinson and Lloyd sur-
vey (which we will discuss in Chapter 9). Thus, they confused the Atkinson and
Lloyd survey of kinetics information with a mechanism. Leone and Seinfeld there-
fore created yet another mechanism using their own 'creativity and ingenuity' and
called it the "master mechanism"—their version of the truth against which oth-
ers were measured.  Their solution to different mechanisms predicting different 03
for the same input was to change all  the mechanisms to be consistent with their
"master mechanism."  Then the difficulties mostly go away.
                                                    What is true by lajnplight
                                               is not always true by sunlight.

Response to  Uncertainties
At the end of Chapter 6, Second Explanation, I suggested that the late 1970's
brought model development and testing to what might be called the HO -paradigm's
first crisis—the impression that some modelers were adjusting their models in
such a way that it was not clear that the values they produced were the correct
ones to enter into the theory, and that other modelers had a vagueness in their
theories and an unanalyzed component in their apparatus. I suggested that this
perception began to undermine the promise that had come with the adoption of
the HO -paradigm.

   The two major modeling groups responded to these impressions in different
ways. Whitten at SAI put forth principles of model development and testing that
"clarifies the sources of uncertainty in simulations" and therefore restricts the
amount of 'knob-twiddling1 possible. And Carter at UCR performed a large num-
ber of chamber simulations directed at explaining the large HO background source
in the SAPRAC EC chamber.

Whit ten's Hierarchy of Species for Testing
In his 1982 EPA report, Whitten said, "A principal goal in computer modeling of
smog chemistry is to develop a set of reactions and rate constants that provides
the closest possible agreement between simulations and measurements for a series
of experiments."  That is, the basic concept is to decrease the margin of "reason-
able agreement" between theory and facts. Whitten then described this process
   A. Use measurements or estimates  for all important reactions, products, and
      rate constants known or expected to occur in the system of interest, to for-
      mulate a kinetic mechanism.  While published data on reactions and rate
      constants should be used where possible in constructing the mechanism, we

Response to Uncertainties	Whitten'i Hierarchy of Species for Testing

       should also recognize that all mechanisms contain hypothetical reactions or
       estimated rate constants.
    B. Estimate the physical conditions that applied for the experiments per-
    C. Combining these two, simulate the smog chamber experiments using a
    D. Modify the mechanism by adding reactions, products, and rate constants
       within their limits of uncertainty until satisfactory agreement between
       simulations and measurements is achieved. However, there are many con-
       straints that must be met:
           >  common reactions must have the same rate constants in all  experi-
           >  chamber-dependent effects should be consistent.
           >  precursor decay must be simulated correctly.

    I believe that Whitten's most important concept was that a complex mecha-
nism should not be constructed in one operation, that is, the building of a mech-
anism should proceed in a stepwise fashion, and the order of building is given by
the concept of a hierarchy of chemical species. That  is, each species in oxidant
formation can be assigned to a hierarchical level on the basis of the number of
HC/NOx systems in which it occurs, with the most ubiquitous species occupying
the lowest level. Thus CO is lower than HCHO, which  is lower than CHSCHO. A
slightly different version of Whitten's hierarchy is shown in Figure 1.

    Whitten said,51
      The value of the hierarchical concept is the principle it suggests in validating ex-
      plicit mechanisms for a series of HC.  In brief, one should validate mechanisms for
      the species at .the lowest level  first. ... Validating mechanisms in the stepwise order
      suggested by the hierarchical levels clarifies the sources of uncertainty in simulations.

          In validating each successive mechanism, one can focus attention on the added
      reactions and  rate constants because the other reactions and rate constants have
      already been validated. Following this procedure reduces the possibility that a com-
      plex mechanism, such as that for propylene, contains errors that compensate for each
      other in simulations of a set of smog chamber experiments.  ... Another  advantage
      for this  approach for validating each mechanism is that it minimizes the possibil-
      ity of fortuitous agreement between simulations and measurements. A valid kinetic
      mechanism, unlike a mere curve-fitting exercise, should give reasonable predictions
      when used in  applications such as atmospheric modeling that are outside the range
      of conditions and smog chamber experiments for which it was developed.

Whitten'i Hierarchy of Speciei for Testing	Response to Uncertainties

    In following this guideline, one would first make sure he could simulate ex-
periments that only contained the inorganic species, e.g., NOx-air and NOx-air-
CO experiments. In the most ideal case, one would use NOx-hydrogen peroxide
(H2O2)-air systems in which the photolysis of HjOj served as a radical source to
further test the inorganic reactions.  Then, without any changes in rate constants
or reaction content for the already tested part of the mechanism, he would add and
adjust  only those reactions and rate constants needed to simulate NOx-HCHO-air
and NOx-HCHO-CO-air experiments.  If, in simulating these latter experiments, dif-
ficulties were encountered  that required the addition of new inorganic reactions,
say for example, 0-atom reactions that might have been lumped out, then the
NOx-air and NOx-air-CO experiments would have to be simulated over again to
verify that no fortuitous agreement between simulations and measurements had
occurred.  Next, one would simulate  experiments of NOx-PAN-air without adjust-
ing the HCHO, NOX, and CO mechanism. PAN makes HCHO, so all the reactions of
HCHO and the inorganic reactions are needed to simulate PAN correctly. Likewise,
PAN must be corrected simulated to  successfully simulate CHsCHO. The mecha-
nism is thus built-up in a stepwise fashion, one new species at a time, and only
allowing the new chemistry to be adjusted.  If the mechanism fails to fit a given
species, then  either the kinetics data for this species is lacking or some error ex-
ists in the experimental data for  this species, but,  since all  lower species data have
already been simulated successfully, it is most likely just this species  that is in error
and not the whole mechanism.

    We should contrast this approach with how the original HO -paradigm smog
chamber data were produced: propylene and n-butane were the first species for
which new smog chamber  data were available.  We have already seen that sev-
eral models were only tested on a few propylene and a few  n-butane experiments.
Propylene is quite high in  the species hierarchy. To model  propylene successfully,
you have to model not only the HO'  and Os attacks on propylene, you have to
model  acetaldehyde, PAN, formaldehyde, and the inorganic  reactions correctly.
How did these modelers know that they did not have off-setting errors between
say the HO -propylene rate constant  and the photolysis rate of HCHO? The answer
is "They did not."  It was  not until after the original explanatory models 'failed,'
that smog chamber runs using the higher species's products as initial reactants
were performed, thus allowing a hierachial development approach to occur.

    In KuhnV terms, testing by Whitten's hierarchy of species method clarifies
the juxtaposition of fact and theory, it  narrows somewhat the macroscopic facts
of the smog chamber so that they more clearly confront the theory.  In other words,

Responat to Uncertainties	Whitten'» Hierarchy of Specie* for Testing

it leds to an improvement of "the measure of 'reasonable agreement' characteristic
of the theory in a given application."

    I  think, however, that modelers, who think of themselves as good puzzle solvers,
while expressing support  for this concept, do not use it as described. That is,
a) having "solved" the  chemistry of, say a particular chamber's wall-effects once
and having moved on to successfully model HCHO, CH3CHO, and propylene, and b)
upon  being told by the chamber operator that new evidence is available on wall
effects which suggest that the technique used by the modeler to represent these is
"not quite right,"  the modeler is not interested in solving the puzzle again, (that
is, in  starting the  whole cycle over again  and re-testing the current mechanism in
successive stages). Instead, assuming that, since the original model  worked there
cannot be any real problems with the current version, and so he just makes a
"patch" to the existing mechanism and proceeds to solve some new  puzzle, which
is more fun anyway.  In addition, the modeler has used the previous mechanism
to produce several hundred simulations and graphs for his report  and is three-
fourths  through the  total project time and money—he literally can not afford to
repeat all the tests!  He also can not conclude that his model, the  one that he is
sticking with, may be wrong. Thus, the more testing that is done with a given
mechanism, the less  likely it is that a flawed mechanism will be changed, espe-
cially if the flaw is in a lower hierarchical species  such as formaldehyde, the prod-
uct of all organic oxidation.  The very success of the mechanism rules against its
being adjusted in  spite  of potential anomalies in the lower hierarchial species.

    Whitten must have been thinking along these lines when he said,
      In computer modeling studies  such as this one, many ideas are  tried, and large quan-
      tities of computer output are produced. In the description of the activity that  pro-
      duced the current closest agreement between simulations and observational data, the
      implicit conclusion is that the steps taken were both unique and necessary.  However,
      experience  has shown that equally close agreement is possible from several combi-
      nations of adjustments  to physical conditions and mechanisms.  Hence, the conclu-
      sions presented here  must be qualified with the caveat that the results are subject to
      change in accordance with new data and  further modeling efforts.

    Whitten also offered guidance on how to proceed within a given species simu-
lation as well. He  said that his  approach was based on the following:
    • The first measurement that must be reproduced with acceptable accuracy
      are those related to the consumption of the initial precursors. Ozone de-
      velopment and other manifestations of the experiment must depend on the
      products that result from decay of the precursor hydrocarbons and NOX.

Whitten's Hierarchy of Species for Testing         	   	            Response to Uncertainties
        Figure 1. A Hierarchy of Species


Respons* to Uncertaintiei	Whitten'i Hierarchy of Species for Testing

      Good agreement between measurement and simulated ozone concentra-
      tions, coupled with poor agreement for hydrocarbon decay, is indicative of
      compensating errors  in the kinetic mechanism. Errors that compensate one
      another under the conditions of a particular smog chamber experiment are
      not likely to do so under atmospheric conditions.
    • In simulating a series of experiments in the same chamber, chamber-dependent
      effects must be treated consistently. If a light source is assumed to emit
      progressively lower amounts of short wavelength radiation over a period of
      several months, the photolysis rate constants for the series of experiments
      must diminish in accordance with the order of performance of the experi-
      ments. Arbitrary adjustments for such effects must be avoided.

    Whit ten's clear understanding of the cause-and-effect in the mechanisms leads
to this advice. First the HC and NO must be converted at the correct rates to the
intermediate products that  then must maintain the reactivity of the system to
produce the final products,  usually PAN, HN03, Oj, CO, and some aldehydes and
ketones. As we have seen, however, "acceptable accuracy" and "good agreement"
are ill-defined concepts. More importantly, in the SAPRAC EC chamber, obtain-
ing correct rates for the initial HC lost rate requires assuming a relatively  large
"unknown" chamber source of HO.  Thus, an 'ad hoc' technique was needed to
obtain "good agreement" between theory and data. This brings us to the second
response to the crisis.

Carter'* Chamber Radical Experiment*	Response to Uncertainties

Carter's Chamber Radical Experiments
In Carter tt o/.'s 1979 mechanism paper,35 they said, "The largest uncertainty
concerns the necessity to include in the mechanism a significant rate of radical in-
put from unknown sources in the smog chamber."  In the UCR models this input
was 0.2 to 0.7 ppb/min of HO ; this is 0.2 ppmV of direct reactivity in a simula-
tion (i.e., 0.2 ppmV out of say 0.5  ppmV of toluene could be converted into rad-
icals throughout a 400 minute irradiation by an 'ad hoc' specification). To com-
plicate the problem, different modelers using the UCR EC data were represent-
ing the needed radical source in different ways: Falls and Seinfeld and Whitten
used only initial HONO as a radical source;  the UCR group were using a constant
source of HO, and Hendry tt al.37 were using a combination of the two methods.
Initial HONO will decrease rapidly in the first 30-60 minutes  and will make little
contribution later in the experiment, whereas the constant HO source will make a
relatively smaller contribution initially, but will make a much larger contribution
later in the experiment.  Carter tt  a/.52  said
      Clearly, aspects of the photochemical mechanism relating to radical initiation and
      termination processes cannot be unambiguously validated using  smog chamber  data
      until this presently poorly characterised radical source is elucidated.

    In three short pages, Carter tt al. presented their 'fit' to a series of chamber
experiments designed to investigate the radical source. They said, "detailed de-
scriptions of the experimental procedures will be given in a subsequent publica-
tion."  The radical source was described as

                   radical flux = *!(0.30 + 2.9[N02])   ppb/min

They went on to say,
      It is significant to note that this equation predicts radical fluxes very similar to those
      used in previous computer  modeling studies form this laboratory, thus further vali-
      dating the chemical models developed in those studies.

         The results reported here show conclusively that radical input from unknown
      sources is an important process in smog chamber systems, and that initial HONO
      is, at most, a minor contributor to this process for typical chamber irradiations of
      es6-h duration. Thus it is clear that photochemical smog models validated against
      chamber data assuming only initial HONO as  the radical source must be Devaluated.
      However, it is probable that use of a constant radical source during an irradiation
      is also an oversimplification, and it is clearly essential to understand the chemical
      and physical processes producing this effect in environmental chambers, and whether
      similar processes also occur in the atmosphere.

    In comments on the Carter et al. paper, Killus and Whitten" first argued
that a large number of factors had not been addressed in the work and that Carter


Response to Uncertainties                                 Carter's Chamber Radical Experiments

et al. had over-estimated the magnitude of the radical flux.  In addition, Killus
and Whitten refuted the assertion quoted above,
      Carter tt al. call for a "revaluation" of only those  modeling studies, like ours, which
    •  have used initial HONO. Carter et al. do not recommend revaluations of those mod-
      eling studies which  have used a constant radical source, even though they do ad-
      mit this is "probably an oversimplification." We suggest that a revaluation of prior
      modeling efforts using a chamber source that varies continuously with NO2 concen-
      tration would result in a much larger discrepancy than can be subsumed  under the
      word "oversimplification."

    This exchange was followed by an extensive paper by Carter et al.M This pa-
per had undergone a number of revisions caused by reviewer comments and suc-
cessive refinements in the data treatment procedures.  The final radical flux in the
EC was
                  radical flux = Jfci(0.39 + 1.37[NO2])  ppb/min
which is nearly a factor of two less dependence on NO] concentrations compared
to the first equation proposed above. Most importantly, Carter et al. found that
the indoor and outdoor Teflon chambers, and new Teflon bags produced data
that suggested much lower radical sources (only 30-40% of the EC chamber) and
showed little dependence upon NOX.

    Killus and Whitten55 again argued that some  of the UCR data were suspect
because of other chamber background or newly characterized wall processes such
as the heterogeneous formation of HONO from NO2 as reported by Sakamaki et

    The following year, Pitts et al.37 confirmed the results of Sakamaki et al., and
extended the work to both the atmosphere and other UCR chambers such as the
indoor Teflon chamber.  They confirmed by in situ spectroscopic measurements
that all chambers had  a continuous conversion of NO2 to HONO, but that the rate
and gas phase yield  of HONO was different in the EC and the Teflon chambers
with the Teflon chambers being less reactive than both the SAPRAC and NIES
EC-type chambers, that the formation in the EC chamber was partially a func-
tion of it previous history,  and that this process was a factor of 10 less than the
source determined by Carter et al. for the EC chamber. Pitts et al. say that
      These facts strongly suggest that the interaction between NO:  and water  and the
      chamber walls, which are responsible for dark HONO formation, could also be in-
      volved in the production of radicals during the irradiations.  However, the details of
      this process remain  elusive at present.

Carter's Chamber Radical Experiments	Response to Uncertainties

    In their latest work, Carter ct a/.58  show that their preferred treatment for
the EC chamber is still to use a large radical source that depends upon N02(e.g.,
for ki = 0.3 and [NO2J=0.5, the source would be 0.44 ppb/min plus 0.1-4 ppb/min
from NOj-f walls), but to use much smaller radical sources that do not depend
upon NO2 for the UCR Teflon chambers and the UNC Teflon chamber (e.g., 0.09
to 0.06 ppb/min plus 0.014 ppb/min from NO2+walls).

    Thus, Carter is comfortable that he can at least predict the radical source
even if he can not completely explain it. He showed that there  are experimen-
tal alternatives, that is, other chambers with lower  radical formation rates that
can be used to check results obtained in the EC chamber, thus  the models do not
have to ultimately rest upon "unknown sources."
                                The image has a. certain dimension, or quality,
                       of certainty or uncertainty, probability or improbability,
                              clarity or vagueness. Our image of the world is
                 not uniformly certain, uniformly probable, or uniformly clear.
                           Messages, therefore, may have the effect not only of
                 adding to or reorganizing the image.  They may also have the
                      effect of clarifying it, that is, of making something which
                         previously was regarded as less certain more certain,
               or something which was previously seen in a vague way, clearer.
                                  Messages may also have the contrary effect.
                                                       —Kenneth Bounding

Third  Explanation
Atkinson and Lloyd Review

What started as a review article to support the development of the Atkinson,
Lloyd, and Winges mechanism in 1982, became in and after 1984 the standard
for all explicit mechanisms in the third phase of explanation.

   Although the review article was referenced in the ALW mechanism paper
as "in press 1981," it did not actually  appear until 1984. According to Marcia
Dodge (EPA paid publication costs), the Journal of Physical Chemistry Reference
Data had difficulty publishing a 330 page article in one issue. During the delay in
both review and in finding an opening in the journal, new and important infor-
mation became available which required revisions, and so publication was delayed
for these revisions. Typesetting such a complex manuscript—and proofing the ac-
curacy of the numerical information—was extremely difficult. In fact, only the
first two pages and a few tables were typeset, the rest were photocopied from a
typewriter-style printer. Some modelers had pre-print versions of the review at
least two years before it was published.

   The purpose of the review was to perform "a critical evaluation  of the rate
constants, mechanisms, and products of selected atmospheric reactions of hydro-
carbons, nitrogen oxides, and sulfur oxides in air" using the literature through
early 1983. In the Introduction, the authors said,
     It is evident that, along with laboratory investigations, ambient atmospheric mea-
     surements, and computer modeling development, there  must also be a continuing
     effort to critically evaluate the rates, mechanisms, and  products of the relevant chem-
     ical reactions and to  update these evaluations as new experimental data become
     available. Such a continuing evaluation for stratospheric reactions has been very suc-
     cessfully carried out mainly through NBS and more recently NASA. However, such
     evaluations, although dealing very thoroughly with  the inorganic reaction systems,

Atkinson and Lloyd Review	                        	Third Explanation
      generally consider only the relatively few organics of specific interest to stratospheric
      problems. ...

          Thus there is a critical need for a comprehensive evaluation of the gas phase
      chemistry associated with the more  complex organics encountered in photochemical
      air pollution systems.

          	The results of this evaluation are intended to aid modelers in developing
      chemical mechanisms to describe smog chamber experiments and to ensure that
      the best available data are used in the modeling study and to limit any tendency
      to "curve  fit" the data.  While we consider this report to be as current as possible,
      we believe it is very  important that  the report be updated at regular intervals and
      expanded  to other chemical compounds so that maximum benefit to  model develop-
      ment is provided.

          ... the majority of this report deals with reactions which have not been previ-
      ously covered,  but which are critical for model development. ... estimates are needed
      if the modeling technology is to improve.  Thus, in the more complex chemistry ...
      much use  has been made of thermochemical arguments and analogies with other
      related systems, since in many cases no unambiguous experimental data are avail-
      able. Hence in such cases only discussions of related chemistry  and suggested reac-
      tion rates, pathways, and mechanisms can be presented, rather  than  firm recommen-
      dations. These suggested rates and  mechanisms should be regularly evaluated and
      upgraded  as new experimental data  becomes available.

In our philosophical terms, Atkinson and Lloyd were saying,
   "Here are the auxiliary statements, including some speculations, that are
    needed to  connect the theory of  air pollution chemistry  to the facts as we
    know them in 1983?

    It is also important to identify what was not said by Atkinson and Lloyd.
They did not say, "Here is a mechanism for explaining smog chamber data."  No
simulations were included. There is  no assertion as to completeness or adequacy
of the description. This can only be  done by comparison of the theory with facts
and a demonstration of "reasonable  agreement." As Kuhn said, without tables
(or in chamber modeling, the plots)  comparing the theory and the data, "the the-
ory would be essentially incomplete." The Atkinson and Lloyd review can be
seen as representing a first class effort to do Kuhn's first class of normal science
work, that is,  the production of both empirical and theoretical information that
the paradigm  has shown to be particularly revealing of the nature of things. It  is
necessary work that must be done before any significant second class of normal
scientific work can be undertaken, that  is, the direct comparison of factual deter-
minations with predictions from the  paradigm theory, in the same way that smog

Third Explanation	L
Leone and Seinfeld Master Mechanism	Third Explanation

    We do not  have the time or effort to evaluate the master mechanism with
    smog chamber data. But the modeling community will never accept the
    simple assertion of theory for a mechanism so we are going to do a token
    evaluation by showing agreement between predictions from our mechanism
    and two (and after review, 11) irradiation experiments with 7-component
    HC mixtures. On the one hand, we have real doubts that mechanisms can
    be validated  with chamber data, and on the other, you can believe our
    mechanism because with  no adjustments it fits the chamber data.

    Within a few months of finishing the EPA report, LS published a paper59 on
the toluene mechanism that was a part of the "master mechanism." In this paper
LS used  comparisons of predictions with chamber data "to evaluate this mecha-
nism." Although they  mentioned the "controversy in previous attempts at sim-
ulating aromatic hydrocarbon" systems due to chamber derived radical sources,
they did not express doubts that they could evaluate their mechanism with such
data as they had suggested might be the case in the EPA report. They accepted
Carter's fit to the radical source in the EC chamber without modification.

    Nine toluene and two cresol experiments were simulated. No experiments
were simulated for any of the theoretically predicted fragmentation products (e.g.,
glyoxal, methyl glyoxal, pyruvic  acid);  in other words, a hierarchical testing scheme
was not followed and the mechanism was used to fit a few pieces of data for the
whole system (e.ff., NOX, Os, PAN, HC-decay).  In the paper, 39 plots of model pre-
dictions  and experimental data were given. From these one can ascertain a some
measure of "reasonable agreement" for the major species only.

    LS conclude,
     The chemical mechanism for toluene photooxidation which is developed here contains
     the latest available kinetic and mechanistic data,  and has been extensively tested
     against smog chamber data. The new mechanism gives good agreement between the
     predicted and observed concentration-time profiles for all the major species in both
     toluene-NOx  and  toluene-benzaldehyde-NOx experiments ... The agreement between
     theoretically  predicted and  experimentally observed oxidant yields is especially good.

          The new mechanism has been used to gain insight into areas of toluene pho-
     tooxidation which are not completely understood.  At present the major uncertainties
     in the mechanism center around the methyl glyoxal and benzaldehyde photolysis
     products, the exact ring fragmentation pathways, and the aerosol forming ability
     of some of-the oxidation products. Although further experimental information on
     any of these  items would be very useful and should be given a high priority, we feel
     that the mechanism presented here can be used confidently in its current form, and
     can also serve as  a useful starting point toward the development of accurate lumped
     mechanisms for modeling gas phase aromatic hydrocarbon photooxidation.


Third Explanation	Leone and Seinfeld Master Mechanism

    To recommend something confidently means,33 "with strong assurance; with-
out doubt or wavering of opinion." I have great difficulty reconciling t&is conclu-
sion with the position taken a few months before in the EPA report, "In fact, it
is our feeling that with the current controversy regarding the magnitude of cham-
ber radical sources, it is difficult to consider any mechanism validated simply be-
cause it can simulate chamber data." Apparently LS somehow resolved their con-
cern over the magnitude of chamber radical sources, so that it became possible
for them to recommend use of a mechanism validated with smog chamber data
without doubt or wavering of opinion.

    Yet, if this conclusion is correct, then the rationale Leone, Flagan, Grosjean,
and  Seinfeld present in their  1985 paper60 contradicts it. In this paper, Leone tt
a/, say,
      The  SAPRAC chamber has been a prime facility for generating data for evaluating
      photochemical reaction mechanisms.  However, it has been shown that reactions car-
      ried  out in this chamber are influenced  by a significant source of free radicals from
      chamber walls.  For this reason,  it is useful to continue to evaluate our understand-
      ing of toluene chemistry with data taken from another smog chamber. In this work
      we present the results of such an evaluation using experimental data from a recently
      constructed outdoor smog chamber facility that has been carefully characterized with
      regard to all of the important chamber parameters, including wall radical sources.
So in spite of their confident  recommendation there must have been lingering
doubt—that is they wanted to conclude that the mechanism was correct, but they
were not sure that it was because of the chamber radical source. As we shall see
shortly, there was reason for doubt, but it had nothing to do with the chamber
radical source problem.

    In this 1985 paper, simulations of six toluene-NOX outdoor irradiations were
shown. A total of 23 plots showing model predictions and experimental data were
presented. The fits of model  predictions to observations  (data included NOX, Os,
PAN, CO, HC-decay, benzaldehyde, and cresol) were as good as those for the UCR
EC chamber experiments.

    At the end of the paper, after the quality of the model had been demonstrated,
Leone tt al. said,
      Since the mechanism shown in the Appendix closely simulates the data from both
      the SAPRAC evacuable chamber and our outdoor chamber, one might conclude that
      it represents a fairly accurate description of the actual toluene photooxidation pro-
      cess. However, Tuaion et al. have recently reported yields of methyl glyoxal from
      toluene that are much smaller [10-17%] than those predicted by our mechanism
      [72%] or, for that matter, by any of the existing toluene mechanisms.

L«on« and Seinfeld Matter Mechaniim	Third Explanation

      ... Even though the mechanism developed in Leone and Seinfeld has now b*eit eval-
      uated successfully against experimental data from two environmental chambw facili-
      ties,  there is concern because this mechanism is not consistent with some very recent
      experimental findings concerning dicarbonyl yields. However, no reasonable sequence
      could be established that makes use of these new data, and, at the same time, gives
      reasonable predictions of the observed concentration-time profiles in toluene-N Ox
      irradiations. Thus, it is clear that additional experimental work is needed —

    Return to Chapter 4 and read the anti-realist objections. We have an exam-
ple here of "Scientific theories, however well secured by observation and exper-
iment, are inevitably fallible." While I know  that there are methodological im-
provements we possibly could use to avoid the types of problems encountered by
Leone tt al., I, too, do not know how "to resolve the basic dilemma of the under-
determination of theory."  Their conclusion was that additional experimental work
was needed.

    Could it be that other parts of the "Master mechanism" might suffer a fate
similar to the toluene part?

    In fact, there are hints that the ethylene chemistry described in the  Atkinson
and Lloyd review is incomplete (underdetermined).  First, in July of 1984, Lur-
man, Lloyd, and Atkinson updated  the information in the review and updated
the ALW  mechanism for use in ADOM. They tested this updated-ALW mecha-
nism  with a series of 15 irradiations from the EC chamber comprising single HC as
well as mixtures. They were able to simulate successfully all but two irradiations
with  half  the Carter wall source strength.  The two irradiations were ethylene ex-
periments (at 2 and 4 ppmC) and these required 1.5 times the Carter wall source
strength and 10 ppb HONO to speed up the simulations  so as to have the timing
and O3 yields fit the data. In addition the HCHO predictions were not only poor,
they were biased opposite on the two runs. Lurman et al. said,
      ... the requirement for the higher wall source  strength to simulate the experimental
      NO2 and ethene data suggest the chemistry of ethene may not be  as well understood
      as once thought. ... the ethene chemistry  incorporated into the mechanism may be
      slower than the actual chemistry and further research is needed on the fate of the
      product species.

    In UNC's testing of the SAI explicit ethylene, Carbon Bond X, and  Carbon
Bond Four mechanisms, we independent came to exactly the same conclusions as
Lurman et al..  To fit our outdoor irradiations,  Whitten used large wall  radical
source strengths and unreasonable values of initial HONO. When we repeated the
simulations with values that were consistant with our chamber characterization
experiments, the simulations were too slow and un-reactive. In addition, HCHO is


Third Explanation	Leone and Seinfeld Matter Mechanism

greatly underpredicted. We have good agreement between the model and mea-
surements for a nighttime Os + ethylene experiment, including HCHO.  -

    In his 1986 study,58 Carter modeled six EC ethylene irradiations and six UNC
ethylene irradiations.  The variability in HCHO predictions compared to the data
was the worst of any of his 400 simulations. He .tended to underpredict UNC
HCHO measurements by 50%.  He suggested that our instrument was subject to
interferences and was therefore over-reporting the HCHO data. In the summer of
1986, an intercomparison study of HCHO was conducted at  the UNC chamber us-
ing four different methods. Preliminary results suggest that the standard UNC
method does not over-report HCHO data, and specifically in the case of ethylene

    Thus there is increasing experimental evidence that we not know all of the
chemistry of ethylene.  Atkinson and Lloyd said,
   ".. .there must be a continuing effort to critically evaluate the rates, mech-
    anisms, and products of the relevant chemical reactions and to update
    these evaluations as new experimental data become available."

    I would suggest that this is also part of a way to ensure that the needed ex-
perimental data  do become available. If we all believe that there are no puzzles
to solve in the ethylene chemistry, then we have no cause to look for interesting
and exciting problems there.

The Essential Tension
In the beginning I said that from a description of practice in our field I hoped
that we might obtain insight into the essential underlying assumptions, method-
ology, and philosophy of our field, and that we might find a commonalty of ap-
proach or discover particularly revealing examples of practice. I think you can
agree that much of this goal has been achieved. The title of this chapter is stolen
from one of Kuhn's books wherein he referred to the essential tension of scientific
tradition and change. Our brief history here reveals sets of essential tensions that
form the crux of the issues that  we should consider at this workshop.  These are
discussed here.

Potential and Actualization

We adopted a new paradigm around 1970. This paradigm has guided both the-
oretical and experimental development. Like most new paradigms, in  the begin-
ning it was mainly one of potential—a promise of success discoverable in selected
and still incomplete examples. The paradigm leads to a criterion for choosing
problems that, while the paradigm is still taken for granted, can be assumed to
have solutions.  That is, the paradigm must be a source of puzzles, the solutions
of which add to the scope and precision with which the paradigm can be applied.
Puzzles are that special category of problems that can serve to test ingenuity or
skill in solution. Puzzles must have an assured solution (which itself may have no
intrinsic importance). Kuhn says, "One of the reasons why normal science seems
to progress so rapidly is that its practitioners concentrate on problems that only
their own lack of ingenuity should keep them from solving." Puzzles must have
rules that limit both the nature of acceptable solutions and the steps  by which
they are obtained.

Potential and Actualization	The Essential T«n»ion

    The paradigm-as-shared-values is a source of commitments for the scientists
that share it. These include:
     •  Explicit statements of scientific law and about scientific concepts and theo-
       ries. While these continue to be honored, such statements help to set puz-
       zles and to limit acceptable solutions.
     •  A multitude  of commitments to preferred types of instrumentation and to
       the ways in which accepted instruments may legitimately be employed.
     •  Quasi-metaphysical commitments that tell  scientists what sorts of entities
       the universe  does and does not contain, what ultimate laws and fundamen-
       tal explanations must be like, and what research problems should be.
     •  A commitment to be concerned to understand the world and to extend the
       precision and scope with which it has been ordered, therefore, says Kuhn,
       "to scrutinize, either for himself or through colleagues, some aspect of na-
       ture in great empirical detail. And if that scrutiny displays pockets of ap-
       parent disorder, then these must challenge  him to  new refinements of his
       observational techniques or to a further articulation of his theories."
Kuhn says, "Normal science is a highly determined activity, but it need not be
entirely determined by rules." There are rules for moving the chess pieces; there
are few rules for playing a game of chess.

    Kuhn's analysis leads him to say, "Normal science consists in the actualiza-
tion of that promise [of success offered by the paradigm], an actualization achieved
by extending the  knowledge of those facts that the paradigm displays as par-
ticularly revealing, by increasing the extent of match between those facts and
the paradigm's predictions, and by further articulation of the paradigm itself."
These are the three areas into which Kuhn says all journal literature of  normal
science can be classified.  (There is extraordinary science—those transitions be-
tween paradigms, but few scientists ever get to participate in such events.  Those
of us who were practicing in the 1970's are lucky to have seen the adoption and
growth of our own paradigm, and we know how exciting that time was.)

    As normal science advances, the promise of the paradigm is actualized and
the paradigm itself is refined, reformulated, and further articulated. This enter-
prise seems to the outsider  (one who does not share the commitments described
above) to be an attempt to force nature into the preformed and relatively inflexi-
ble  box that the paradigm supplies. The areas investigated by normal science are
minuscule; the enterprise has drastically restricted vision. Kuhn says, "But those
restrictions, born from confidence in a paradigm, turn out to be essential to the

Th« Essential Teniion	Potential and Actualitation

development of science. By focusing attention upon a small range of relatively es-
oteric problems, the paradigm forces scientists to investigate some parr of nature
in a detail and depth that would otherwise be unimaginable."  Thus the juxta-
position of our theoretical constructs and the observations associated with them
leads to an extension of the precision and scope of the ordering of the world.

   This progress of the paradigm, the practice of normal science—by its very na-
ture (the finer and finer attention to detail)—encounters anomalies that require
distortion to include, or deletion to ignore.  As these mount, the promise offered
by the paradigm begins to fade—the paradigm ceases to be a source of puzzles
with assured solutions. Scientists committed to the paradigm begin to realize
that they may fail to solve a puzzle within  the confines of the paradigm, but, for
a long time, this is likely to be seen as an individual failure, not a failure of the
corpus of current science. Individuals respond to this  developing crisis in different
ways: some  leave the field, some stop solving problems, some escalate the distor-
tion and the deletion and force the system  to work anyway; these actions stresses
the community.  Ultimately the paradigm becomes non-functional, it ceases to be
a source of solvable puzzles. The scientist committed to understanding the world
must then move on to another paradigm or quit. In some rare cases, the promise
of a paradigm may be fully actualized, that is, all the puzzles may be successful
solved (this  happened in optics, for example).  In this  case, the science becomes
a craft, and  scientists, committed to increasing the understanding of their world,
move  on to other more productive areas. I  suspect that the EPA policy makers
would be very happy if air pollution modeling became a craft.

   Functioning with maps that are highly  distorted or that suffer from serious
deletion must ultimately be painful. This leads to either resolving the crisis and
modifying the map or to non-functional behavior from those using the maps.

   And this brings us to our current situation. I believe that we need to face the
anomalies, not force the fit, and, on the one hand,  tighten the standards for "rea-
sonable agreement" while on the other, making it a little safer to take a risk and

   Because, I believe, the standards for acceptance of agreement between fact
and theory have been set too low by the current modeling community, easy moves
occur from one viewpoint to another. I have* described the mechanism of "rea-
sonable agreement" in the history above. In Kuhn's view, in so far as their work
is quantitative, most scientists attempt to improve the measure of "reasonable
agreement"  characteristic for a theory  in a  given application and to open up new


Potential and Actualitation	The E«»«ntial Tension

areas of application and establish new measures of "reasonable agreement" ap-
plicable to them. He has shown, however, that there is no consistently applied or
consistently applicable external criterion for "reasonable agreement" and has sug-
gested that reasonable agreement is what is accepted by the community; usually
this is what is accepted for journal publication that is not refuted in a  very short

    We have seen in the example here, that for modeling, the measure of "reason-
able agreement" is often established by claim and is based on a ridiculously lim-
ited set of evidence.  If the reviewers of the papers are not themselves modelers,
this evidence may seem convincing, or if they are modelers, they may object—
but not too strenuously—because they see themselves in the same position with
respect to publishing their own modeling results!  Some modelers have gone else-
where or have found other avenues of expression, for example, EPA reports. These,
however, are always associated with funding and contractual commitments that
confound the science and the peer review process. These too are most likely to
"confirm" or "support" the model developed.  In its present situation, EPA is of-
ten helpless when it comes to enforcing a measure of quality assurance with re-
gard to the characteristics of "reasonable agreement." Therefore, here too, we
find low standards accepted for reasonable agreement.

    On the other hand, we must be cautious not to set the standards too high, for
then "no one satisfying the criterion of rationality would be inclined to try out
the new theory, to articulate it in ways which showed its fruitfulness or displayed
its accuracy and scope."1S Kuhn says, "What  from one viewpoint may seem the
looseness and imperfection of choice criteria conceived as rules may, when the
same criteria are seen as values, appear an indispensable means of spreading the
risk which the introduction or support of novelty always entails."

    Therefore, I believe that we must  work as a community to reestablish an ex-
plicit recognition of a set of rules. These rules, in the words of Kuhn "limit both
the nature of acceptable solution and the steps by which they are to be obtained."
If we take 'limit' not to mean restrict, but rather to define, then these rules are
not unnecessary burdens on science.  Also the interpretation of limit as "define"
explains why rules can lie  unexpressed below the surface of regular scientific ac-
tivity: definitions, as we have seen, must often be taken for granted if they are
to function at all. These definitionary rules do not define what the state of the
science, or even what the paradigm, is. But rather, they define whether or not
solutions fall within the paradigm, require modification or distortion of the par-
adigm, or fall completely outside the paradigm. Such rules can change  in direct


The E««entiai Tension	Potential and Aetualitation

response to the input of puzzle-solutions, but clearly they are not infinitely flex-
ible. Frequently in times of crisis, more solutions are ruled out of the paradigm
than are included within it. When the rules function in this way to define the
paradigm more and to prevent its unnecessary distortion, the rules can save a
still-functional paradigm and work goes on.

    When the rules have strengthened and protected the paradigm in these ways,
they—now accepted by consensus in the community—recede again from explicit
recognition and indirectly guide the community toward interesting puzzles in need
of solution.  Thus the  dcfinitionary rules can be changed in response  to the increas-
ingly refined solutions proposed within the paradigm, but they direct research best
only when they are sufficiently accepted by the community to be implicit.

    Sharing risk  is a traditional approach in times of crisis.  This I believe is one
of the reasons the Atkinson and Lloyd review has acquired such dominance: using
it not only avoids having to repeat or re-examine work, it is safer than sticking
your own neck out; the user of the Atkinson and Lloyd review does  not have to
defend it. It has become a kind of standard. A problem with standards, as EPA
well knows, is that they become outdated. The force that the Atkinson and Lloyd
review has exerted on the field clearly suggests one obvious action EPA can un-
dertake to resolve some of the conflicts: make a commitment to, and provide sup-
port for, regular review processes like the Atkinson and Lloyd work.

    But agreement as to what constitutes kinetic knowledge is not sufficient. There
are other categories of empirical and theoretical knowledge that must be agreed
upon. We have described this agreement, this standard, as  "reasonable agree-
ment," defined as the fit  of various kinds of evidence measured according to con-
sensus, or in some cases, unchallenged assertion.  And this in part has been the
problem: solutions have been asserted and accepted that later have been shown
to have been established  on weak evidence. When this becomes ordinary, then
crisis is not far away.  That is, we want to say that "reasonable agreement" is one
type of rule that needs to surface in times of paradigmatic crisis.

    For science to advance at all there must be reasonable agreement between
three sets of evidence:
    1. between theory and observation;
    2. between different sets of observations; and

Explanation and Prediction	The Essential Tension

    3.  between explanation and prediction.
Without agreement, no detailed comparison could be made between the three sets
of evidence: no puzzle-solutions could be evaluated. And the puzzles most mean-
ingful in maintaining or challenging a paradigm are exactly those that lie on the
interfaces of the three sets listed above.

    Thus the first sense in which reasonable agreement determines scientific progress
Is in defining what sorts of evidence can be compared.  That is to say, that by
defining what qualifies as evidence of theories, observations, explanation, and
predictions, the community establishes reasonable agreement for quantitative
precision. Once evidential standards of reasonable agreement have been recog-
nized, the puzzles we have described as at the interfaces of theory and observa-
tion, between observations, and between explanation and prediction can be de-
fined. When  solutions to the problems are offered, the second function of rea-
sonable agreement is activated: agreement for better quantitative and qualita-
tive precision to descriminate among the solutions. Again, this function is deter-
mined  by consensus within the community and again as a standard it varies di-
rectly with the progression of the paradigm from potential to actualization. Here
again, commitments from EPA to support a regular process to examine reason-
able agreement in the three sets of evidence above will be beneficial  to the model-
ing community.

Explanation and Prediction

That we are in, or approaching a crisis, brought about by forcing actualization
over promise, I believe can be seen in the tension between explanation and predic-
tion revealed in the brief historical examples present here.

    Looking back at the work that has been done in the last 15-years, it appears
that, first, modelers attempt a rather full explanation of the observations in smog
chambers using new kinetics data and producing new hypotheses about path-
ways and important reactions. It is no surprise, according to Kuhn,  that in the
end, all these theories+auxiliary statements  have "agreed with the data" more
or less.  If there was not the chance that the puzzle had a solution, it would not
have been undertaken. And, no one is going to conclude that he failed to solve
the puzzle he set for himself, even if he has to discount facts. If we see demon-
strations of "reasonable agreement" at all, we only see those that can be the basis
of a claim. Yet other modelers often make very different conclusion from the ev-
idence  presented to show "agreement with the data" and so doubt the model's

The Ewntial Teniion	Explanation and Prediction

verisimilitude. Second, the mechanism, that more or less agreed with the data be-
comes the source of a simpler description, which some compare not with reality,
but with the original model—the idea being that the original model is  the correct
explanation—while others compare with a smaller set of chamber data and also
announce "agreement with data."  History shows, however, that both the original
modeler and other modelers find "pockets of apparent disorder" which "challenge
him to a new refinement of his observational techniques or to a further articula-
tion of his theories." and they soon present us with a new model.10 Because, I
think, the evidence presented for the  former model is  too weak, this new challenge
immediately undermines the basis of  belief in both the explanation and the pre-
diction of the former model.  This  can be seen to result from a process  of "false
actualization" and insufficient vindication of the work accomplished.

    Viewed by one outside  the scientific process, the situation described above
closely associates in daily experience  with charlatanism or outright fabrication—
"Yeah, that's the ticket!  Or, ah, maybe ..." Viewed from a historical perspective,
the theories associated with the HO -paradigm are new theories. Kuhn  says most
new theories do not survive.  And  when they do, much work, both theoretical and
experimental is  required  before the new theory can display sufficient accuracy and
scope to generate widespread conviction.  Before a group accepts it, a new the-
ory has been tested over time by the  research of a number of men, some working
within it, and others within a rival. Such a mode of development  requires a deci-
sion process which permits rational men to disagree, and if a scheme attempts to
eliminate this mode of development it will not produce science.15

    So if there is to be science in photochemical modeling, EPA has to  learn to
live with a process whereby rational men may disagree.  Coupled with this recog-
nition are other characteristics of these members of professional scientific groups.
Kuhn  identified several striking characteristics,
      The scientist must be  concerned to solve problems about the behavior of nature. In
      addition, though his concern with nature may be global in extent, the problems on
      which he works must be  problems of detail. More important, the solutions that sat-
      isfy him may not be merely personal but must instead be accepted as solutions by
      many. The  group that shares them may not, however, be drawn at random from
      society as a whole, but is rather the  well-defined community of the scientist's pro-
      fessional compeers. One  of the strongest,  if still unwritten, rules of scientific life is
      the prohibition of appeals to heads of state or to the populace at large in matters
      scientific. Recognition of the existence of a uniquely competent professional group
      and acceptance of its role as the exclusive arbiter  of professional achievements has
      further implications. The group's members, as individuals and by virtue of their
      shared training and experience, must be seen 
Verification and Vindication	The Eatential Tension

     shared some such basis for evaluations would be to admit the existence of incompat-
     ible standards of scientific achievement.  This admission would inevitably rave the
     question whether truth in the sciences can be one.

    Thus while there clearly may be a role for EPA in affirming standards, it
must be one of supporting actions of the scientific community as a group.

Testing and Fit

So we can agree that in times of paradigmatic crisis, standards for definitinal
rules and for reasonable agreement  must be made explicit, and that with these
tools the paradigm is strengthened  or successfully challenged. But how are stan-
dards employed? And if we really believe that examination of these standards
leads to useful guidelines, just what are the lines guiding? The answer is testing.

    There are at least three different fields for testing and they coincide with the
three sets of evidence listed above for the standards of reasonable agreement.
That is, theory must be tested with observation, observations must be tested
together, and explanation must be tested with prediction. The 'tests' in each of
these areas are actually the puzzles set by the paradigm or a theory within the
paradigm for the practice of normal science and we have seen several examples
in the history. Here we are concerned with how tests of these three areas deter-
mine the use of any particular theory, the model generated in response to the
theory, and the mechanism that underlies the model. I shall emphasize that, al-
though  a model and its mechanism derive from common theory and are its repre-
sentatives for testing, successful testing of any one component does not convey
even the hope of success to the others. Thus it is entirely possible that a use-
ful theory is entirely unsuccessful in model and mechanism generation and so is
untestable—and thus does not become the subject of normal science and will re-
main a mere speculation until it does become testable. More importantly, how-
ever, I would argue that models and mechanisms which seem to be well-tested
may in fact have not been sincerely tested at all,  and that in any  case—pass or
fail, sincere or not—such tests say nothing about the theory that  the models and
mechanisms represent.   Then what can testing mean?

Verification and Vindication

We have seen that scientists apparently do not test the paradigm theory when
testing the theories' models, mechanisms, and auxiliary statements. Failure to fit
is assumed to be a problem in the mechanism, or the auxiliary statement, not in

The Eaatntial Tension	Verification and Vindication

the the theory itself. Successful fitting, on the other hand, means justJ,hat.  What
about next time? In testing you see, we rely on inductive logic. Inductive logic is
simply the belief that the world will function in the future as it has in the past,
based on our best experience of it. In that it relies on the uniformity of nature,
such reasoning is not objectionable on its face, but it is by definition invalid for
proof.  This is because the assertion that the world is uniform is made and weakly
confirmed solely through observation—and we have seen that to a large extent,
our observations are filtered through our expectations, through our theories  so to
speak. The sun rises in the East each morning. "You  mean it is not moving and
we are?"

    To the historian, at least, it makes little sense to suggest that verification or
validation is establishing the agreement of fact with theory. Invoking realism also
has its problems. For example, there is the belief that a model that offers expla-
nation has inherent ability to predict because its ability to explain is taken as
proof that it is truth, that the entities in the model have a true correspondence
with the real world. No support accrues to realism by showing that realism  is a
good hypothesis for explaining scientific practice. Need we take a good explana-
tory hypothesis to be true?  A complete acceptance of realism commits one to an
unverifiable correspondence  with the world.  We have many examples of validated
mechanisms that are no longer used. Is today's scientific  environment any less
"dynamic" than when Tom Hecht said "In such a dynamic scientific environment,
it would be presumptuous to suggest that any given set of reactions and rate con-
stants is the chemical mechanism for smog formation."?

    Does this mean  that all models are just opinion? If we insist upon the criteria
of certitude, incorrigibility, and immutability, then yes. If we relax those criteria
and recognize that these opinions we can affirm on the basis of evidence and rea-
sons that have sufficient probative force to justify our claim at the time  that the
opinion affirmed is true. We must stress at the time because we must be prepared
to have the opinion we now claim to be true on the basis of evidence and reasons
now available turn out to be false or in need of corrections when new evidence
and other reasons come into play."

    We may agree that induction can not be "validated," that it is logically im-
possible to show that  the conclusion of an  induction follows logically from the em-
pirical evidence. Nevertheless, we can perhaps "Vindicate" induction.  That is,
establish a form of argument "If any method of predicting the future works, then
induction works." Or if induction does not work, then no method will work.  In
our terms, this translates to being able to make the statement


Verification and Vindication	The Essential Tension

   "If this model does not explain or predict in this situation then I do not
    know how to produce another model that does. Therefore, we judge that
    it is sensible to adopt this model for this use at this time."
What is required to make such a statement?

Mechanism Vindication

One purpose of this workshop is to describe a method that would allow a modeler
to make a vindication statement about his model and would allow the modeling
community and model users to have confidence that it was true.

   I believe that some of the elements of this method would be:
   A.  We could believe that a mechanism is the best  description we can obtain
      about an atmospheric chemistry at this time when:
         1) there is a high degree of confidence in the existence and behavior
           of the majority of the entities that are in the mechanism—meaning
           they have been measured and characterized by more than one re-
           search group and the community has reached a consensus on the
           information, or
         2) in the case of explicit mechanisms, there is confidence that the re-
           maining entities described in the model exist and behave as de-
           scribed; this requires acceptable processes for
               > elaboration of the alternatives;
               > distinguishing among alternatives;
               > estimating the consequences of good  and bad choices;
               > making unique predictions that can be verified experimen-
         3) in the case of simple models, there is confidence that the simple
           model agrees with both the explicit model and data.
   B.  There is consistence among the various forms of evidence:

Cl»i»e» of Scientific Work	Methodologies

          1)  the mechanism agrees with chamber data and the agreement is es-
             tablished in a stepwise manner;                       *
          2)  different chambers produce data that agree;
          3)  the mechanism not only explains data, it predicts correctly in situa-
             tions not used in its development.
   C. Before a mechanism is declared suitable for use in atmospheric simulations,
      it has been tested over the range of conditions expected to occur in the
      atmosphere, e.g., mixture composition (both static and dynamic), large
      dilution, dynamic injection, entrainment of material from aloft.
   D. Others.

Classes of Scientific Work

Normal scientific work associated with mechanism development and testing com-
    1. Production of significant facts that the theory of HO -chain attack has shown
      to be particularly revealing of the nature of things.
          A.  The determination of the existence, the rate, and the mechanism of
             elementary reactions of importance to the initiation, propagation,
             and termination of the chain.
          B.  The organization and evaluation of elementary reactions and the
             theoretical estimation of the critical reactions that have not yet
             been determined in A.
    2. Production of factual determinations that can be compared directly with
      predictions from mechanisms based  on the HO -chain theory, and  the ma-
      nipulation of the theory or the production of auxiliary statements to allow
      the theory to make predictions that can be confronted directly with mea-
          A.  Producing smog chamber experimental data in a form suitable for
             the direct comparison with predictions.
          B.  Creating mechanisms that adequately describe the experimental sit-
          C.  Producing simulations and comparing predictions and observations.

    3. Production of empirical fact to articulate the HO -chain theory, to resolve
      some of the residual ambiguities, and accounting for changes in auxiliary
      statements and basic theory that arise from the empirical work.


Methodologies	Agreement

         A.  Performing chamber characterization experiments; quality assurance
             measures, intercomparison of different instrumental techniques; field
             studies; making measurements of predicted key intermediates.
         B.  Formulation of new model programs and approaches; testing of pre-
             diction methods, making unusual or "risky" or unique predictions
             that can be tested.
         C.  Comparisons of predictions from different mechanisms; comparisons
             of observations from different chambers.

    We have seen that essentially none of these activities moves forward without
the others.


The form that a science puzzle often takes is to deduce the form of auxiliary state-
ments that, when combined with generally accepted theory will show "reasonable
agreement" with facts or finding ways to produce facts that can be shown to be
in "reasonable agreement"  with predictions.

    On the empirical side of the puzzle, it has been a long standing heuristic that
multiple measurement systems must be used to verify the facts and to estimate
the accuracy  and bias of their statement.  This means that  different chambers
must be used to test models, perhaps we should even say different chamber groups.
At times there may need to be directly  related experimental work in different fa-
cilities to obtain answers needed to resolve some of the puzzles—6.9., UCR needs
to compare their HCHO measurement method with, say the  EPA cartridge system,
so that the differences in reported HCHO between  UCR and UNC can be resolved.

    On the theoretical side of the puzzle and in the case of models, I suggest that
there is contamination among the classes of knowledge that are used to construct
the model. For example, I might identify these classes  as:
    > accepted theory
    > new theory
    t> quasi-theory
    > current hypothesis
    t> speculation

Agreement	Methodologies

    Having understanding and agreement as to what parts of the model belong to
which class of knowledge and having a demonstration as to the sensitivity of the
model to these parts is essential in judging its verisimilitude. This process was
more common in the earlier days of model development, suggesting that recently
it has been assumed that there is some implicit agreement for recent models as to
what is accepted as true. We have already suggested that Atkinson and Lloyd's
review has been accepted by some as defining accepted knowledge and thus, they
assert, can be used without defense. We know, however, that Atkinson would be
the first to deny the status attributed to his now 3- to 4-year old work.

    Therefore, reviews and classification of knowledge, both empirical and the-
oretical, are needed on a regular basis which would be similar to the  reviews of
rate constants, but which must include all available experimental data and all
current hypotheses in use in all candidate models. To be acceptable to the sci-
entific community, these reviews must be conducted by the group of scientists
participating in the creation activities. EPA's role should be one of facilitator—
arranger, publisher, and disseminator of results.

    We have a good example of a review for the types of evidence described under
IB  above. I know of no examples of reviews of the types of evidence  described
in 2A and 3C above. That is, a review effort is needed to assess the agreement
among different chamber groups. Another review is needed to assess the degree
of agreement among different modeling groups for representations of  known and
hypothesized entities in the mechanisms.

    Kuhn's analysis of how scientist adopt a new  theory strongly suggests that
the comparison of a single model with facts will never be the source of rejecting
that model. Therefore, at least two models must  be compared with facts at the
same time to achieve progress. We have already seen, however, that each model
helps fix what counts as evidence or test.  This means that the resolution of this
comparison may need specific facts;  such facts often require 2-3 years to produce.

    A final type of regular review, then, would be the comparison of models com-
pared to chamber data.

    It should be our goal not to seek multiple representations of reality—for ex-
ample, it is not the Carbon Bond Mechanism, nor should it be the Carter, Atkin-
son, Lloyd, Lurman mechanism—instead, it should be
     > 'What is the  best representation of what we know with a high degree of
      certainty?', and


Methodologies	Agreement

    > 'What is an adequate representation for what we probably know?' and
    > 'How creative can we be about what we are guessing at?'

 1    "Improvements Needed in Developing and Managing EPA's Air Quality Models," GAO/RCED-
      86-94, April 1986.

 2    Letter from Chairman John Dingell to EPA Administrator Lee Thomas, June 9,  1986.

 3    Letter from Basil Dimitriades to participants in workshop on Evaluation and Documenta-
      tion of Chemical Mechanisms Used in Air Quality Models, July 29, 1986.

 4    Fox, D.  "Judging Air Quality Model Performance,"  Bull. Amer. Metero. Soc., 62, pp
      599-609, 1981.

 5    "Guideline  On Air  Quality Models (Revised)," EPA-450/2-78-027R, Office of Air  Quality
      Planning and Standards, U.S. Environmental Protection Agency, Research Triangle Park,
      NC, 1986.

 6    Jeffries,  H.E,  K.G.  Sexton, C.N. Salmi,"The Effect of Chemistry and  Meteorology on
      Ozone Control Calculations Using Simple Trajectory Models and the  EKMA Procedure."
      EPA-450/4-81-034, U.S. Environmental Protection Agency, Research Triangle Park, NC.

 7    Leone, J.A, J.H.  Seinfeld, "Evaluation of Chemical Reaction Mechanisms for Photo-
      chemical Smog, Part II: Quantitative Evaluation of  The Mechanisms," Final Report
      CA 810184, Atmospheric Sciences  Research Laboratory, U.S. Environmental Protection
      Agency, Research Triangle Park, NC.  1984.

 8    Levin, M. "What Kind of Explanation is Truth?" in Scientific Realism, ed. J. Leplin,
      University of  California Press, Berkeley, 1984.

 9    Hidy, G.M, "Future Needs in Atmospheric Research and Modeling," Final Workshop Re-
      port, ASRL, U.S. Environmental Protection Agency, Research Triangle, NC, 1986.

10    Kuhn, T.S., The Structure of Scientific Revolutions, University of Chicago Press,
      Chicago, 1962.

11    Feyerabend, P.K., "Explanation, Reduction, and Empiricism," in Minnesota Studies in the
      Philosophy  of Science, ed H. Feigl and G. Maxwell,  Minneapolis, 1962.

12    Atchinstein, P., Concepts of Science: A Philosophical Analysis, John Hopkins Press, 1969.

IS    Popper, K., Objective Knowledge Oxford  Press, 1972.

14    Hacking, I.  Scientific  Revolutions, Oxford University Press, NY, 1981.

15    Kuhn, T.S., The  Essential Tension, Selected Studies in Scientific Tradition and Change,
      University of  Chicago Press, Chicago, 1977.

16    Lakatos, I, in Minnesota Studies in the Philosophy of Science, Minneapolis, 4972.

17    Handler R., J. Grinder, The Structure of Magic, Science and Behavior Books, Inc., Palo
      Alto, CA,  1975.

18    Weinberg, G., Rethinking System Analysis and Design, Little, Brown and Company,
      Boston, 1982.

19    Leighton, P.A. Photochemistry of Air Pollution,  Academic Press, NY, 1961.

20    Atkinson, R., K.R. Darnall, A.C. Lloyd, A.M. Winer, J.N. Pitts, "Kinetics and Mech-
      anisms of the Reactions of the Hydroxyl Radical with Organic Compounds in the Gas
      Phase," in Advances in Photochemistry, 11, John Wiley and Sons, 1979.

21    Demerjian, K.L., J.A.  Kerr, J.G. Calvert, "The Mechanism of Photochemical Smog For-
      mation," in Adv. in Environ. Sci. and Tech., 4, John Wiley and Sons, NY, 1974.

22    Greiner, N.R., J. Chem. Phys., 46, p 3389,  1967

23    Heicklen, J., K. Westberg, N.  Cohen, "The Conversion of NO to N02 in Polluted  Atmo-
      spheres," Symp. Chem. Reactions in Urban Atmospheres, Research Laboratories, General
      Motors, Warren, Michigan, Oct. 1969.

24    Niki, H., E. Daby, B. Weinstock, "Mechanisms of Smog Reactions," Ford Motor Co.
      Technical Report, ca. 1970.

25    Niki, H., E. Daby, B. Weinstock, "Mechanisms of Smog Reactions," in Photochemical
      Smog and  Ozone Reactions, Adv. in Chemistry Series, 113, p 13, 1972.

26    Putnam, H. "The 'Corroboration' of Theories' in Scientific Revolutions, ed. I. Hacking,
      Oxford University Press, 1981.

27    Gear, W. The Numerical Solution of Initial Value Problems, Printice Hall, 1971.

28    Leplin, J., "Introduction," in Scientific Realism,  ed. J. Leplin, University of California
      Press, Berkeley, 1984.

29    Smart, J.J., Between Science and Philosophy, Random House, NY, 1968.

30    The Encyclopedia of Physics,  2ad Edition, ed. R.M. Besancon, Van Nostrand Reinhold
      Co. NY, 1974.

31    Hecht T., J. Seinfeld, M. Dodge, "Further Development of Generalized Kinetic Mecha-
      nism for Photochemical Smog,".Environ. Sci. Tech., 8, p 327-339, 1974.

32    Hecht,  T. M.K. Liu, D.G. Whitney,  "Mathematical Simulation of Smog Chamber  Photo-
      chemical Experiments," EPA-650/4-74-040, U.S.  Environmental Protection Agency, Re-
      search  Triangle Park, North Carolina, 1974

33    Webster's New Twentieth Century Dictionary, Unabridged, Second Edition, Simon and
      Schuster, New York, NY,  1983.


References	.	

34    Whitten, G.Z., H. Hogo, "Mathematical Modeling of Simulated Photochemical Smog,"
      EPA-600/3-77-011, U.S. Environmental Protection Agency,  Research Triangfc  Park, NC,

35    Durbin, P., T.A. Hecht, G.Z. Whitten, "Mathematical Modeling of Simulated  Photochem-
      ical Smog," EPA-650/4-75-026, U.S. Environmental  Protection Agency, Research Triangle
      Park, NC, 1975.

36    Hendry, D.G., R.A. Kenley, J. Amer. Chem. Soc. 99, p 3189, 1977.

37    Hendry, D.G., A.C. Baldwin, J.R. Barker, D.M. Golden, "Computer Modeling of Simu-
      lated Photochemical Smog," EPA-600/3-78-059, U.S. Environmental Protection Agency,
      Research Triangle Park, NC, 1978. .

38    Jeffries, H., personal recollection.

39    Carter, W.P., A.C. Lloyd,  J.L. Sprang, J.N. Pitts, "Computer Modeling of Smog Cham-
      ber Data: Progress in  Validation of a Detailed Mechanism for the Photooxidation of
      Propene and n-Butane in Photochemical Smog," Int. J.  Chem. Kinetics 11 p  45, 1979.

40    Dodge, M,  "Combined Use of Modeling Techniques  and Smog Chamber Data  to Derive
      Ocone-Precursor  Relationships," in Proceedings of Inter. Conference on Ozone, Raleigh
      NC, 1977.

41    Falls, A.H., J. Seinfeld, "Continued Development of a Kinetic Mechanism for Photochemi-
      cal Smog,"  Environ. Sci. Teeho. 12, pp 1398-1406,  1978.

42    McRae, G.J., W.R.  Goodin, J.H. Seinfeld, "Mathematical Modeling of Photochemical Air
      Pollution,"  State of California Air Resources Board  EQL Report NO.  18, 1982.

43    Falls, A.H., G.J.  McRae, J.H. Seinfeld, "Sensitivity  and Uncertainty of Reaction Mech-
      anisms for Photochemical Air Pollution," Inter. J. Chem. Kinetics, XI, pp.  1137-1162,

44    Whitten, G.Z., H. Hogo, J.P. Killus, "The Carbon Bond Mechanism: A Condensed Ki-
      netic Mechanism of Photochemical Smog," Environ. Sci. and Techno., 14, pp. 690-700,
    "  1980.

45    Atkinson, R., A.C. Lloyd, L. Winges, "An Updated  Chemical Mechanism For  Hydrocar-
      bon/Nitrogen Oxides Photooxidations Suitable for Inclusion in Atmospheric Simulation
      Models," Atmos. Envir. 16, pp.  1341-1355, 1982.

46    Killus, J.P., G.Z. Whitten, "Effects of Photochemical Kinetics Mechanisms on Oxidant
      Model Predictions," EPA-600/3-83-111, U.S. EPA, RTF, 1983.

47    Carter, W.P.L., A.M.  Winer, J.N. Pitts, "Effect of Kinetic  Mechanism and Hydrocarbon
      Composition on Oxidant-Precursor Relationships Predicted  by  the EKMA Isopleth Tech-
      nique," Atmospheric Environment^, pp. 113- 120,  1982.

48    Leone, J.A., J.H. Seinfeld,  "Evaluation of Chemical  Mechanisms For Photochemical Smog:


      Part II. Quantitative Evaluation of the Mechanisms," Final Report, EPA Cooperative
      Agreement 810184, 1984.

49    Dunker, A.M., S. Kumar, P.H. Berzins, "A Comparison of Chemical Mechanisms Used in
      Atmospheric Models," Atmo. Environ. 17, 1983.

50    Dennis, R.L., M.W. Downton, "Evaluation of Urban Photochemical Models for Regula-
      tory Use," Atmo. Environ.lS, pp. 2055-2069, 1984.

51    Whitten, G.Z., "The Chemistry of Smog Formation: A Review of Current Knowledge,"
      Environment International 9, pp. 447-463, 1983.

52    Carter, W.P.L., R. Atkinson, A.M. Winer, J.N. Pitts, "Evidence for Chamber-Dependent
      Radical Sources: Impact on  Kinetic Computer Models for Air Pollution," 7n.tr. J. Ckem.
      Kinet.  13, pp. 735-740, 1981.

53    Killus,  J.P., G.Z. Whitten, "Comments on Paper 'Evidence for Chamber-Dependent Radi-
      cal  Sources," Intr. J. Ckem. Kinet.lZ pp.  1101, 1981.

54    Carter, W.P.L., R. Atkinson, A.M. Winer, J.N. Pitts, "An Experimental Investigation of
      Chamber-Dependent Radical Sources,"  Intr.  J. Chem. Kinet. 14, p. 1071, 1982.

55    Killus,  J.P., G.Z. Whitten, "Background Radical  Initiation Processes in Smog Chambers:
      A Perspective," System Applications Paper,  San  Rafael, CA, 1982.

56    Sakamaki, F., S. Hatakeyama, H. Akimoto, "Formation of Nitrous Acid and Nitric  oxide
      In the Heterogeneous Dark Reaction of Nitrogen  Dioxide and Water Vapor in a Smog
      Chamber," Intr. J. Ckem. Kinet. 15 pp. 1013-1029, 1983.

57    Pitts, J.N., E.  Sanhueza, R. Atkinson, W.P.L. Carter, A.M. Winer, G.W. Harris, and
      C.N. Plum, "An Investigation of the Dark Formation of Nitrous Acid in Environmental
      Chambers," Intr. J. Ckem. Kinet.16, pp. 919-939, 1984.

58    Carter, W.P.L, F.W. Lurmann, R. Atkinson, A.C. Lloyd,  "Development and Testing of a
      Surrogate Species  Chemical  Reaction Mechanism," Final Report EPA Contract No. 68-02-
      4104, ASRL, U.S.  Environ. Prot. Agency, RTP, NC, 1986.

59    Leone,  J.A.,  J.H. Seinfeld, "Updated Chemical Mechanism for Atmospheric Photooxida-
      tion of Toluene," Intr. J.  Ckem. Kinet.16, pp. 159-193, 1984.

60    Leone,  J.A.,  R.C.  Flagan, D. Grosjean, J.H.  Seinfeld, "An Outdoor Smog Chamber and
      Modeling Study of Toluene-N0x Photooxidation," Intr. J. Ckem. Kinet.17, pp. 177-216,

61    Adler, M.J. Ten Philosophical Mistakes MacMillan Publishing, NY, 1985.

   Appendix C
Workshop Agenda

                                                         Appendix C
         Workshop on Evaluation/Documentation of Chemical Mechanisms
          Sponsored by EPA's Atmospheric Sciences Research Laboratory
                    Research Triangle Park, North Carolina
The Raleigh Inn
Raleigh, North Carolina
December 1-3, 1986

 9:00 AM     Registration

 9:30 AM     Call to Meeting

 9:35 AM     Welcome-Introduction of
              Dr. Courtney Riordan

 9:45 AM     Opening Remarks

 10:00 AM     Introductory Remarks by
              Workshop Chairman

 10:15 AM     Presentation


 12:00 NN     Lunch

 1:30 PM     Presentation


 2:30 PM     Presentation


 3:30 PM     Break

 4:00 PM     Presentation

Workshop Chairman and
Moderator: R. Atkinson

B. Dimitriades, EPA

A. Ellison, EPA

C. Riordan, EPA

R. Atkinson, UCR

H. Jeffries, UNC
J. Tikvart, EPA
K. Demerjian, SUNY
R. Atkinson, UCR
 Evening:     Steering Committee meets to formulate questions and organize
              discussions for the next two days.

                                                          Appendix C

 9:00 AM     Presentation


 10:00 AM     Presentation


 11:00 AM     Presentation


 12:00 NN     Lunch

 1:30 PM     Presentation


 2:30 PM     Presentation


 3:30 PM     Presentation


 4:30 PM     Report on Steering Committee
              Deliberations and Plans

              Continuing Discussions

 5:00 PM     End of Workshop


 9:00 AM     Final meeting of the Steering Committee
M. Gery, SAI
A. Dunker, GM
W. Carter, UCR
F. Lurmann, ERT
G. McRae, Carnegie Mellon
G. Whitten, SAI
R. Atkinson

    Appendix D
Workshop Attendees

                                                         Appendix D
                            Workshop Attendees
Dr. Roger Atkinson
Statewide Air Pollution Research
University of California-Riverside
Riverside, CA 92521
Dr. Marcia Dodge
Atmospheric Sciences Research
  Laboratory (MD-84)
U.S. Environmental Protection Agency
Research Triangle  Park, NC  27711
Dr. Mike Barnes
Atmospheric Sciences Research
  Laboratory (MD-57)
U.S. Environmental Protection Agency
Research Triangle Park, NC 27711
Dr. Alan Dunker
Environmental Sciences Dept.
General Motors Research Laboratories
12 Mile and Mound Roads
Warren, MI  48090-9055
Dr. Ronald Bradow
Atmospheric Sciences Research
  Laboratory (MD-59)
U.S. Environmental Protection Agency
Research Triangle Park, NC 27711
Dr. Jack Durham
Atmospheric Sciences Research
  Laboratory (MD-57)
U.S. Environmental Protection Agency
Research Triangle Park, NC  27711
Dr. Joseph Bufalini
Atmospheric Sciences Research
  Laboratory (MD-84)
U.S. Environmental Protection Agency
Research Triangle Park, NC  27711
Dr. Alfred Ellison, Director
Atmospheric Sciences Research
  Laboratory (MD-59)
U.S. Environmental Protection Agency
Research Triangle Park, NC  27711
Dr. William Carter
Statewide Air Pollution Research
University of California-Riverside
Riverside, CA 92521
Dr. Kenneth Demerjian
Atmospheric Sciences Research Center
Earth Science Building - Room 324
State University of New York-Albany
1400 Washington Avenue
Albany, NY 12222
Dr. Basil Dimitriades
Atmospheric Sciences Research
 Laboratory (MD-59)
U.S. Environmental Protection Agency
Research Triangle Park, NC  27711
Dr. Michael Gery
Systems Applications, Inc.
101 Lucus Valley Road
San Rafael, CA  94903
Dr. Daniel Jacob
Department of Earth and Planetary
Harvard  University
29 Oxford Street
Cambridge, MA  02138
Dr. Harvey Jeffries
Dept. of Environmental Sciences and
School of Public Health
University of North Carolina
Chapel Hill, NC 27514

                                                        Appendix D
                            Workshop Attendees
Mr. William Keith
Office of Research and Development
U.S. Environmental Protection Agency
401 M Street, SW
Washington, DC  20460
Dr. Courtney Riordan
Office of Research and Development
U.S. Environmental Protection Agency
401 M Street, SW
Washington, DC  20460
Dr. Alan Lloyd
Environmental Research and Technology
975 Business Center Circle
Newbury Park, CA  91320
Dr. Fred Lurmann
Environmental Research and Technology
975 Business Center Circle
Newbury Park, CA  91320
Mr. Edwin Meyer
Office of Air Quality Planning and
  Standards (MD-14)
U.S. Environmental Protection Agency
Research Triangle Park, NC  27711
Dr. Gregory McRae
Department of Chemical Engineering
Carnegie Mellon University
Shenley Park
Pittsburgh, PA  15213
Dr. Hiromi Niki
Science Research Laboratory
Ford Motor Company
P.O. Box 2053
Dearborn, MI  48121-2053
Dr. Deran Pashayan
Office of Research and Development
U.S. Environmental Protection Agency
401 M Street, SW
Washington, DC  20460
Mr. Kenneth Schere
Atmospheric Sciences Research Laboratory
U.S. Environmental Protection Agency
Research Triangle Park, NC  27711
Dr. Kenneth Sexton
Dept. of Environmental Sciences and
School  of Public Health
University of North Carolina
Chapel Hill, NC 27514
Dr. Jack Shref f ler
Atmospheric Sciences Research Laboratory
U.S. Environmental Protection Agency
Research Triangle Park, NC  27711
Mr. Joseph Tikvart
Office of Air Quality Planning and
  Standards (MD-14)
U.S. Environmental Protection Agency
Research Triangle Park, NC  27711
Dr. Gary Whitten
Systems Applications, Inc.
101 Lucus Valley Road
San Rafael, CA 94903