UNITED STATES ENVIRONMENTAL PROTECTION AGENCY

                             WASHINGTON. D.C. 20460

                                July 31, 1985
                                                                      OFFICE OP
                                                                  THE ADMINISTRATOR
Hon* Lee M. Thomas
Administrator
U. S. Environmental Protection,
  Agency
401 M Street, S. W.
Washington, D.C.   20460

Dear Mr. Thomas;

The Environmental Engineering Committee of the Science Advisory Board was
asked by the Office of Water to review a report entitled "A Probabilistic
Methodology for Analyzing Water Quality Effects of Urban Runoff on Rivers
and Streams*"  The Committee has completed its review, and Is pleased to
forward its report.

The Conmittee believes strongly that statistically-based approaches to water
quality management are an important tool for the decision-maker, and commends
the Agency for supporting the effort under review.  The method described is
technically sound, but only for the specific applications for which it was
developed.  The Committee has serious concerns about apparent Agency Interest
in using the aproach la situations for which It is not technically suitable.
The Cotraalttee does not believe that the technique, as it now exists, should
be extrapolated beyond the purpose and application area for which It was
developed without appropriate additional development and verification, nor
should It be used by individuals who do not fully understand the approach
and the assumptions Inherent therein.

If you have any questions, or should you wish any further action on our part,
please call on us.

                                               Sincerely,
                                               Raymond C. Loehr
                                               Chairman, Environmental
                                                 Engineering Committee
cc: E, Longest                                 Science Advisory Board
    C. Myers
    D. Athayde
    S.- Tuller                                            '
    T. Barawell
    E, Southerland                             Norton Nelson
    T, Yosle                                   Chairman, Executive Committee
                                               Science Advisory Board

-------
                           REPORT
                      on the review of








         "A PROBABILISTIC METHODOLOGY FOR ANALYZING




WATER QUALITY EFFECTS OF URBAN RUNOFF ON RIVERS AND STREAMS"
                           by the
            Environmental Engineering Committee




                   Science Advisory Board




           U* S. Environmental Protection Agency
                         June, 1985

-------
I. EXECUTIVE SUMMARY

The Environmental Engineering Committee believes strongly that the Agency
should consider certain water quality phenomena in a probabilistic manner.
The Agency is commended for supporting the effort under review and similar
activities, and is encouraged to pursue other efforts to increase its capa-
bilities to deal with random environmental phenomena.

In its review of a report entitled "A. Probabilistic Methodology .for Analyzing
Water Quality Effects of Urban Runoff on. Rivets and Streams", the Committee
agreed to address, and report to the Agency on, two general questions;

     A.  Is the technique scientifically valid and adequate for the specific
application for which it was developed?

     B.  Is it appropriate for further or broader application?

The technique presented in the report was developed for advective systems,
i.e., flowing fresh-water streams.  It was developed for nonpoint source pol-
lutants which are conservative (such as total dissolved solids - TDS) or are
reacting substances characterized by first-order kinetics (such as biochemi-
cal oxygen demand - BOD).

No serious flaws were found in the model, but some clarifications and im-
provements are suggested!  The Committee believes that the technique is sci-
entifically valid and acceptable forthe purpose and conditions for which it
was developed and presented in the report„

The Committee has serious reservations about the potential for indiscriminate
use of the technique for purposes and applications beyond its current purpose.
In order for the existing technique, as documented in the report, to be appro-
priate for use in water quality situations involving many commonly encountered
real-world aspects, considerable additional development and validation would
be required.  Included would be capabilities to deal with non-conservative se-
quential reactants (such as dissolved oxygen or some toxic substances or nu-
trients).  The technique as it exists is not applicable to tidal or estuarine
systems, nor to lakes.  Aspects such as these are frequently encountered with
both point and nonpoint source pollutants.

The Committee does not believe the technique, as it now exists, should be ex-
trapolated beyond the purpose and application area for which it was developed
without appropriate additional development and verification.  This could en-
tail substantial additional effort.  Furthermore, the Committee believes
there is cause for great concern if a valid technique with deceptively attrac-
tive simplicity in its ease of application (such as the one described) is
pushed beyond its capabilities.  This is, of course, true for any mathemati-
cal modeling, simulation or predictive procedure.  The potential exists here
due to the technique's having been developed specifically for the Nationwide
Urban Runoff Program (NURP), but being under active consideration for adoption
and use in other problem areas such as waste load allocation and toxic/pesti-
cide fate and effects.

-------
                                     -2-
II. INTRODUCTION

There appears to be increasing Interest in the environmental community in
the fact that many natural phenomena have strong random components, and that
environmental regulations should explieity acknowledge this fact.  The Water
Planning Division, Office of Water Planning and Standards, as a part of the
Nationwide Urban Runoff Program, has been developing statistical methods
for the analysis of urban stormwater for some time.  These methods have
had limited use and have not, in general, addressed water quality impacts,

The Environmental Engineering Committee of the Science Advisory Board was
asked, in a memorandum from the Office of Water, to review a. report entitled
"A Probabilistic Methodology for Analyzing Water Quality Effects of Urban
Runoff on Rivers and Streams", dated February 15, 1984» and prepared by the
Office of Water.

The principal Issues for review were:

     A.  Is the technique described in the report scientifically valid and
adequate for the specific application for which it was developed?

     B.  Is the technique appropriate for further or broader application?

The Committee agreed to accept the task* and organized a Subcommittee, ia-
cluding several consultants with special expertise in the area of stochastic
modeling* chaired by Dr. Benjamin C. Dysart III (see Subcommittee roster,
Appendix A), to conduct the review.  The Subcommittee and its consultants
met on November 26, 1984, and on February 25, 1985.  In the course of its
review, the Subcommittee examined the report (together with an earlier report
prepared by the Agency's Office of Water Planning and Standards entitled "A
Statistical Method for the Assessment of Urban Stormwater"), written com-
ments by four individuals, and had an in-depth review of the report by Agency
staff and their consultants (Dr. Dominic DiToro and Mr. Eugene Drlscoll)
who prepared the report under review.  The SAB consultants, Dr.  Mitchell
Small and Dr* Barry Adams, were asked to prepare written comments, and these
are attached as Appendices B and C,
III.  CONCLUSIONS

All natural processes or phenomena such as * hydrology and water quality have
probabilistic or stochastic elements*  However, the majority of the mathema-
tical models used today are essentially deterministic in structure, i* e.,
they are based on the concept of conservation of mass and momentum, and
attempt to simulate explicitly, using direct solutions of equations, the
processes that transport and transform material and energy in the natural
system.  Statistical models, on the other hand, do not explicitly simulate
natural systea processes, but rather provide estiaates of the value(s) of
output (dependent) variables, given the values of input (independent) vari-
ables in the system and the statistical relationships between them.  These
models allow the direct consideration of random phenomena, such as streamflow

-------
                                     -3-
or rainfall, as well as the performance of traditional treatment processes
and other engineered control systems.  They also reflect as yet undefined
mechanisms.

They are limited, however, in that any statistical relationship derived from
a given set of data for a natural system reflects the particular spatial
arrangement and natural system processes existing when the data were collected.
For any significantly different system, new data mist he obtained and new
statistical relationships developed,

The technique presented in the report was developed for advective systems
(freshwater flowing streams), and for nonpoint source pollutants which are
conservative or singular-reacting substances, e.g. total dissolved solids
(TDS), suspended sediment, biochemical oxygen demand (BOD), and bacteria.

The Committee concludes that:

     A*  The technique isscientifically valid and acceptable for the purpose
and conditions for which It was developed and pregented in the report.

No critical flaws were found in the technique as presented in the report*  No
substantial statistical problems were found.  Several conclusions were reached
by the the Conudttee and its consultants (see Appendices B and .C)*  These are
summarized as follows:

          1. fhe methodology presented in the report is acceptable for con-
             servative and first-order reactive substances in flowing fresh
             water streams.  In such systems, the assumption of instantaneous
             and perfect mixing is customarily made; the Committee believes
             that this is an acceptable practical approximation in the vast
             majority of cases.

          2. While some reviewers contended that a simulation modeling ap-
             proach was more appropriate than, the probabilistic model, the
             Committee believes that these two approaches should be consi-*-
             dered as alternatives, the selection, of which would depend on
             the nature of the problem and the question addressed in a spe-
             cific application.

          3* In response to questions raised about the mathematical vali-
             dity of the methodology, the Committee is satisfied with the
             approximations concerning log-normality, the mean recurrence
             interval, and the uncertainty analysis.  This comment is not
             meant to preclude the examination of alternative procedures
             as described in the reviewers' and consultants' comments.

          4. In response to the question concerning incomplete or inadequate
             field data, or data which are thought to be incomplete or in-
             adequate, the Committee recognizes extrapolation from other
             locations of similar characteristics as a common and accepted
             practice.

-------
     B.  The technique, as it is described in Che report, should not be
extrapolated beyond the purpose andApplication area for which it was devel-
oped without substantial additional development and verification*

The Committee believes that a probabilistic methodology has a. very important
role and contribution, as a modern water quality management tool for regula-
tion, planning and facility design.  In order for the technique presented in
the report to be appropriate for use in other water quality assessments,
other capabilites would have to be included.  Among these would be the capa—
bility to deal with non-conservative sequential reactants such as dissolved
oxygen and some toxic substances and nutrients, with tidal or estuarine
systems» and with multiple-source pollutant inputs within a stream reach.

Furthermore, the Cotamittee believes that there is cause for great concern if
a valid technique with deceptively attractive simplicity in its ease of appli-
cation (such as the one described in the report) is pushed beyond its capa-
bilities.  In this case, the potential exists due to the technique's having
been developed specifically for the Nationwide Urban Runoff Program but is
apparently under active consideration in other EPA offices for adoption and
use as is in problem areas such as waste load allocation and toxics/pesticides
fate and effects*

The question of whether it would be worthwhile to modify, extend, and verify
the technique to cover additional application areas is a resource question
for the Agency more than a scientific one that can be addressed by this Com-
mittee.  The Committee believes the potential surely exists to extend the
technique to cover more applications.  This remains to be done, however, and
someone must decide to devote the necessary resources to the effort if the
technique is to realize any potential beyond its existing limited capability.
These resource requirements could be quite extensive*

-------
                                                                   APPENDIX A
                      U.S. ENVIRONMENTAL PROTECTION AGENCY
                             SCIENCE ADVISORY BOMB
                       ENVIRONMENTAL ENGINEERING COMITTEE

                     PROBABILISTIC METHODOLOGY SUBCOMMITTEE


                                    CHAIRMAN      •  .
Dr. Benjamin C» Dysart, III
Environmental Systems Engineering Department
Clemson University
Clemson, SC 29631

                                    MEMBERS

Mr. Richard A. Conway
Corporate Development Fellow
Union Carbide Corporation
P. 0. Box 8361 (770/342)
South Charleston, WV  25303

Dr. Raymond C. Loehr
Civil Engineering Department
8.614 ECJ Hall
University of Texas
Austin, TX  78712

Dr. Donald J, O'Connor
Professor of Environmental Engineering
Environmental Engineering Science Program
Manhattan College
Manhattan College Parkway-
Bronx » m  10471

Dr. Charles R. O'Melia
Professor of Environmental Engineering
Department of Geography and Environmental
 Engineering
The Johns Hopkins University
Baltimore, MD  21218
                                  CONSULTANTS
Dr. Barry Adams
Department of Civil Engineering
University of Toronto
Toronto, Canada M5-S1A4

Dr. Mitchell Small
Department of Civil Engineering
Carnegie-Mellon University
Schenley Park
Pittsburgh, PA  15213

-------
                                                                          APPENDIX  B

           REVIEW  OF PROBABILISTIC  METHODOLOGY FOR ANALYSIS OF
                   WATER QUALITY IMPACTS OF URBAN RUNOFF
                                  Mitchell J.  Small
                             Carnegie-Mellon University
                                 December 21,  1984
 In general,  I  find  the probabilistic model appropriate  for the purpose  and  set-of
conditions  for  which it was designed,  A discussion  of  pertinent technical  details is
provided in the final section of this review,  I did find a clear-cut error in  the  initial
equations given  for  the  mean recurrence interval (MRI).   However, the approximate
equation is  correct.  That  is,  it  approximates   the  corrected  version  of  the MRI
equation provided in my final section.   As  such,  the  results presented ir» the report
are essentially correct.  I also examine a number of the issues raised in the  review
of Kahn.   But  first,  I'd like to  address some  of the broader  issues  relative  to  the
model's applicability and possible  extension.

 The  authors present the  model  as a tool  for  initial assessment.   The  conceptual
nature  of the model limits  it  to  that purpose,  and the  limitations  are highlighted in
the report  and in the  reviews  of Southeriand, Barnwell and  Roesner.   There  is always
a temptation to  push tools as easy to use  as  this  into more detailed applications,
waste-load allocation, etc.   The  reviewers appear to be wary of this.  The first issue
that should be raised, however,  is not  the tool,  but  the  generality  and  importance  of
the question it  addresses:  the  formulation of probabilistic  water  quality criteria  for
nonpoint source  impacts.   Should  we pursue  more  precise specification  of  water
quality exceedence frequencies, as is done with air pollution? Is a 10  year  MRI value
appropriate to  evaluate nonpoint  source  contributions?  If  so,  then  it is up  to  the
working  engineers  and planners  to  devise  appropriate analysis  techniques to predict
exceedence frequencies for their  specific  water system arid problems.  This  could
involve  direct,  deterministic  simulation  using  inputs from  a   long-term  period  of
record;  stochastic  simulation  using  synthetic  flow  and   quality   inputs  generated
consistent  with observed  or regional patterns; or direct  analytical techniques  such as
developed  in the NURP report.  As indicated  in the reviews, many analysts are more
comfortable  with simulation techniques,  in  part  because  random  process  properties
(temporal  and  spatial  persistence, correlation,  etc.)  are  inherently  included.   There
should not be  an official  push to require any one  particular approach.  As engineers
receive more exposure  to  the probabilistic method  and recognize  its ease  of use,
some may wish  to  use it as  a  complement to their  more detailed  evaluations.  This
will  provide working examples  which  will  serve to  either build  confidence  in  the

-------
method,  or  pinpoint  its  weaknesses  and  limitations.   This  may be  stimulated if
analysts  required  to evaluate a  10 year concentration find the  simulation  approach
too  cumbersome.   The  point is, let the problem  and need be better specified ,  .  ,
      1
this  will  drive the selection  of tools.

 As noted by Southerland and iarnwell. direct short-term concentration impacts from
stormwater runoff are rare.   Indeed it Is 'interestrng-that the three test cases cited in
the NURP report  lead  to a conclusion  of no  significant  impairment.   Problems  which
do  occur,  such  as  lake  or estuary  nutrient  problems, long-term bottom  sediment
buildups,  or  coastal  zone  bacterial   problems  also  have  important  probabilistic
features.  Research  to develop analytical methods  for  these problems should, I  think,
be  encouraged.   The  resulting models may again  be  limited to  initial assessments,
but could provide useful  complements  to more rigorous modeling  approaches.

 Another  issue  of  general  concern  involves  the  role   and  characterization   of
uncertainty  in this,  and  other similar  models.  The  uncertainty computed  directly in
the current model is representative  of  inherent temporal variability.  Some  storms are
small,  some  are  large.   Sometimes  the  upstream  flow  is  high,  sometimes  low.
Similarly for  upstream and  runoff  concentrations.   This variability  is represented  by
fitted or estimated  distribution functions.   The parameters  of these distributions are,
however,  also  unknown, reflecting  uncertainty  of  another  sort.    In a narrow  sense
this  uncertainty  may   be   represented  using  confidence  intervals  for  means   or
variances estimated from  observed data,  however  in  reality it  reflects  our  larger
uncertainty in the overall structure  and representation of the problem.   This scientific
uncertainty  is somewhat subjective, but may be represented  in the model and  is  (in
principle) reduceabie.

 The NURP  report  addresses   scientific-parameter  uncertainty   in  its  sensitivity
analyses for stream-runoff correlation  and  upstream  concentrations (Ch, 5).   In recent
years, however, more formal and  complete methods  for sensitivity  and  uncertainty
analysis  have been  developed.  These generally  involve  replication of the underlying
model  with  random  or  stratified sampling  of   the   uncertain  input ' space,  and
examination  of  the  resulting  distribution  of the  output   variabie(s).    The  relative
importance of the different  input parameters is inferred using methods such  as rank-
order correlation  analysis,   A  nice example  of  this technique  is  provided  in  the
attached  article  by  Jaffe  and Ferrara  (Water Research, 18:  1169*1174),   This type  of
analysis  should  be  performed on the  NURP model  to provide a  more complete and
unified picture of model  sensitivity. An alternative approach is to allow  users of  the

-------
methodology  the  ability to perform rapid and  exploratory  sensitivity  analysts  with
interactive graphic software.

 Technical Review

 The  only major  technical problem I  was able  to uncover  }s an  error  In the  initial
equations for the MRI,  Eqs 3-34 and 3-35,  The correct forms arenas follows;- •••*•'•••- •
             >  c   } - 1 - Pr(e   < < }W                      <34a)

   MSI   =     —r"rT"Ty.*l^                                C35a)
Eq. 3S is  a proper first  order  approximation for  Eq. 3Sa {though not for Eq. 35,  which
yields absurd results).   The use of  Eq, 36  is no longer  necessary,  however, because
Eq. 35a will not blow-up, as  did Eq. 35,  when  used with  a  calculator or in  a  BASIC
program.   The  approximation is  close enough, particularly  for long  return periods,  so
that none  of the  results presented  to  date tneed  be redone.   However, for  future
work, Eq. 35a is preferable.

  !  will address  now some of  the  points  raised by  H  Kahn,   Eq*  3-10 is correct
because  high values  of  D corresspond to tow values of $ ,  so that the  95 percentiie
D  corressponds to the  5 pereentile $ ,  The legends  on some of the probability plots
are indeed  reversed  from usual  convention,  showing the probability of  being greater,
rather than less  than, the indicated  value.   I don't expect this to  provide too  big a
problem, however, for  most users.   The  use of  the term  "arithmetic moments" is a
matter of  semantics designed to differentiate  between the  direct  sample  moments
and geometric (log)  moments which are also used.   Additional  clarification  of the
useage could perhaps be provided in  the. report,

  The suggestion  to  use a  beta  distribution for 0 is interesting,    I believe that the
beta  result  is  only precisely  true  when  Q^and  Qs  are  independent  and gamma
distributed  with  the  same  scale parameter  (though  not  necessarily the  same  shape
parameter).  Neither  of  these  limitations  is appropriate for the problem at hand, but
the beta distribution is  sufficiently   flexible over the  range  0-1  so that it  probably
stili provides a good approximation.  It  is  not clear though that it  will  lead  to  either
considerable simplification of the calculation procedure,  or a more  accurate estimate
for the Co distribution  finally computed  (particularly in  the  numerical scheme,  which

-------
seems to be the method of choice).   1 guess  someone needs  to  push the calculation
through to  see if  this  is  the  case.   Similarly  for  the logistics model,  I can  see
where if -tnD is  assumed to have a  logistics  distribution,  then $ corressponds to the
cdf  measure.  'But  I  was  unable to follow  this  through  for  computation  of  the
moments  of $ t or for other direct use in the  model.  !n general, I  am satisfied with
the level  of validation  provided  for  the  computational procedure in the NURP report
and the recent paper of  D5  Tore  (ASCE* June, 1984).   The  authors of the report may,
however,  wish to present some  distribution plots for $ to see how  large is the
deviation   from  lognormality,   and  how   iwweh probability  is  predicted  in  the
'impossible' range, $ >  1.

  I believe  Kahn  is correct in  noting the  sensitivity  of confidence  intervals  in  the
calculated MRI.   I also believe  that some of the "nonstatistical"  factors  which  I raise
above (when  discussing scientific uncertainty) are  most  critical in  determining  the
appropriate  range of this sensitivity,

  To  summarize  then,  I  find the  direct  model  largely appropriate  and  scientifically
credible  for the  purpose for  which   it  was  designed.  A number  of more genera!
issues are raised in addressing  the model's broader applicability.'

-------
                                                               APPENDIX C
                               IEVIEW OF

            A PROBABILISTIC METHODOLOGY FOE THE ANALYSIS OF

                 WATER QUALITY EFFECTS OF UB1AN RUNOFF

                       B.J. Adams, Ph.D., P. lag.
                     Department of Civil Engineering
                         University of Toronto
                             January 1985'
      This review is provided at eke request of the Probabilistic Methods

Subcommittee/Environmental Engineering Committee/Science Advisory Board of

the U.S. Environmental Protection Agency.  Hie review addresses both the

original documene "A Probabilistic Methodology for Analyzing Water Quality

Effects of Urban Runoff pn livers and Streams", Office of Water, U.S. EPA,

Washington, D.C*» 15 February 1984, and comments on the methodology contained

in memoranda from leary D. Kahn (21 July 1984), Elizabeth Sontherland

(undated), Thomas 0, Barnwell, Jr. (29 May 1984) and Larry A. Soesner (11

July 1984).

-------
                                  -2-
BACKGEOUND

      There are some issues concerning methodologies and models which require

addressing before evaluating the probabilistic methodology.


Model Function	

      For Che purposes of this discussion, a distinction is made between
               4
two pronounced differences in the function or role of models.

A crude distinction is as follows;

(i)  Physical Models - models which focus an relating cause and effect in
                       terms of the physical variables of the phenomena
                       involved in the processes being modeled.

(ii) Decision Models - models which focus on relating cause and effect in
                       terms of the decision variables of the alternatives
                       used to control the processes being modeled.

As the function of a model is a. major determinant if its formulation, the

formulation of a model directly defines its use:  a decision model is less

appropriate for explaining cause and effect at the level of the detailed

physical phenomena being described in the problem; a physical model is

less appropriate at the level of, evaluating the cos ^effectiveness o£

control alternatives for solving the problem.

            Examples of commonly encountered physical models are SW£M III,

QUAL II, BECEIV II, etc.  while examples of decision models for evaluating

water quality control alternatives are ST01M, HPS and analytical

probabilistic models.


Model Operation

            Physical models and decision models are "operated" under different

conditions*  Physical models simulate details of the process, including

particular water quality phenomena, under only very specific conditions

-------
                                  -3-
(eg» 7Q10 conditions).  Thus,, the system behaviour is known In

detail but only far very limited conditions.  On the other hand, decision

models simulate a more aggregate response of the process over the complete

spectrum of conditions (eg, the entire probability distribution of flow

conditions).  Thus, the system behaviour is known in less physical detail,

but ie is more generally know over a  wide range of conditions,

            In a very coarse way, the difference between physical and

decision models is the followingi

fhysicsl models describe the response of the system in terms of the details
of the water quality transformations for surrogate conditions.

Decision models describe the response o£ the system in terms of surrogate
water quality transformations for a complete range of conditions.

In this way, physical models are generally limited to the analysis of

specific events while decision models analyze a continaina of evencs*  A

tradeoff in model selection is apparent:  more detailed analysis of behaviour

for less general conditions or. less detailed analysis of behaviour for more

general conditions,

(For a more detailed discussion of event simulation deficiencies, see the

attached paper on "Design Storm Pathology" by Mams and Howard).

-------
THE DECISION PROBLEM



            The decision problem at hand contains a series o£ elements as



follows;'  ISPUT-— DR4IHAGE SYSTEM -*- OUTPUTS -*-B£dI?lR-^-WATES QUALITY-*- USE.



The inputs are meteorologic and ara taken as given.  The drainage system



Is comprised of a catchment and engineered components of the catchment



(conveyance, storage, treatment, etc. components).  The outputs are point



and non point discharges categorized by variable time series of water and



pollutant masses.  The receiver is the stream, river, lake or estuary



accepting the point and noupoint discharges.  The combined characteristics



of the receiver and discharges determine the ia-situ water quality which



ultimately affects the level of beneficial use associated with the receiver-.



            The decision problem is to determine the appropriate level of



beneficial use associated with the receiver and, simultaneously, the



engineering and management measures to be deployed on_the catchment in



order to achieve that level of use.



            It has been, widely accepted that water quality criteria may



be used as surrogate measures of the degree to which level of use is



achieved.  It is thus common practice that decisions regarding the



engineering and management of the catchment4are made on the basis of predicted



in-stream water quality.  Similarly, the magnitude and frequency of point



and nonpoiat discharges (the outputs  of the catchment) may be used in



turn as surrogate measures of the resulting in-stream water quality.  In



this way, the engineering and management measures may be evaluated on the




basis of what happens at the "end-of-the-pipe".



            The question is what level of information is aost appropriate



to the decision at hand?  Is it preferable to have much information on what

-------
happens at the end-of-the-pipe, even though this is a. surrogate for in-stream



water quality •which Is in turn a surrogate for level of beneficial use; or



little information on resulting water quality; or even less information on



level of beneficial use?  Clearly t as we. move farther "downstream1* in the



process from input to use, we gain in. detail but lose in scope.  At some



paint, the quality of information from modeling efforts most be judged.



For a given application there* is an optical tradeoff between detail and



scope of information.



            For example, if two Mutually exclusive engineering alternatives



for pollution control are available at similar cost and if one of the



alternatives results in pollutant1 mass discharges with a smaller mean and



variance than the other, it is clearly preferable.  Ho "water quality model-



ling". Is required to make the decision'.  A valid decision may be made on, the



basis of wbat happens at the end-of-the—pipe.  This esaunple is admittedly



simplistic but it makes the point: the information provided by modeling



efforts should be appropriate to the decision being made,






EVALUATION OF THE METHODOLOG*



            It is in the above contest that the probabilistic methodology



must be evaluated.  The models of the probabilistic methodology should not



be viewed as simply hydrologic models, water quality models or even statis-



tical models; rather, they should be viewed as decision models which contaia



elements of hydrology, water quality and statistics.



            Thus, die adequacy of the probabilistic methodolog7 must be



evaluated with respect to the following:



Physical adequacy - How well do the models represent the physical phenomena



                    (hydrology, hydraulics, water quality transformations,



                    etc,) that they attempt to describe?

-------
                                  -6-
Mathematical/statistical adequacy - How good are the underlying simplifying



                    assumptions made in the model development for reasons



                    of fccactability?  Are the mathematical assmfiptiotts



                    consistent with the physical processes and are the



                    derivations correct?



Functional adequacy - Is the Information provided by the methodology



                    appropriate to the decisions made with the Information?



                    Is there a good balance between the scope and detail of



                    the information?  Do users have confidence in the



                    technique?



            In essential point to consider is that these different measures



of adequacy cannot be applied in isolation.  The model evaluation criteria.



must be applied together.






Physical Adequacy



            The basis of the probabilistic methodology is that probability



distributions of upstream flow/concentration and urban runoff flow/concentra-




tion are transformed to a firobability distribution of downstream concentra-



tion.  The transformation is described by a simple mass balance.  The major



question regarding physical adequacy of this model are:  how realistic are



the data needs of the model and how realis tic is the mass balance as a



representation of the receiver?



            The data needs are the statistics (mean» standard deviation)



of the ups tream flow and concentration and the urban runoff flow and



concentration.  These data requirements are minimal, and it is safe to say



that alternative evaluation methodologies would require much more data.



Hence, these data needs are judged to be quite realistic.

-------
                                  -7-
     1 •  A potential problem exists in the estimation of the statistics of



urban runoff flow/concentration.  The methodology relies on deriving these



statistics from field measurement.  The evaluation of the methodology mist



ask whether tdieae field measurements are generally available-  If not, can



such statistics be generalized from data at other measurement sites?  Even



if field data are available, they are applicable to the current condition



of the catchment.  If alternative engineering or management measures for



urban runoff control are to be evaluated by the methodology, then such



Measures would change the statistics of urban runoff flow/concentration.



To use the probabilistic methodology far the evaluation of alternative con-



trol measures, a satisfactory method of predicting the statistics of urban



runoff flow/concentration resulting from these measures must be found.  Here



again, a reasonable conclusion is that methods for predicting these statis-



tics may not be perfect but.other methodologies with the same objective would



encounter at least similar if not greater problems.




            The second question regarding the physical adequacy of the method-



ology is the adequacy of the mass balance as a representation of the receiver.



The two major problems arising from this simplification are the absence of



reactions in the receiver and the assumption of instantaneous and perfect



mixing.  Both problems have been identified in the memoranda of Southerland



and Barnwell.



            Three approaches can be taken regarding the absence of reactions



in toe receiver:  (i) build appropriate reaction into the model, (ii) re-



strict tie use of the model to "conservative" substances and (iii) impute



the quantitative effects that reactions would have on the model results based



on ao reactions.  The first approach may indeed be possible; however, it may



be undesirable in its detraction from the simplicity of the modal.  The

-------
                                  -8-
second approach may be undesirably by virtue o£ nothing being  truly conserva-



tive in this seme.  Substances which may be conservative in the water  column



may be reactive with respect: to sediment or biota.  The third  approach  may



have merit if the effects of the reaction are generally known  and particularly



if they are monotonic.  For example} it may be sufficient to know if  the



model results are always conservative and if so, than to what  extent  are-



they conservative.



            A similar approach may be taken regarding tie assumption,  of



instantaneous and perfect mixing.  It may be adequate to live  with the  assump-



tion with the understanding of which side the error lies and the extant of



the error.



            These remarks on. the physical adequacy of the methodology must be



taken in the context of the function of the methodology.  It is reiterated



that the physical adequacy cannot be judged in Isolation from  die role  of



the information produced by the aioctels in decision, making.  Ie is my  opinion



that the proposed probabilistic methodology Is appropriate, In terms  of its



physical adequacy, for its intended purposes in this sense.






Mathenatieal/5 tatlatical Adequacy



            The general questions concerning the mathematical/statistical



adequacy of the probabilistic methodology are as follows:



  i) are the derivations correct in a mathematical sense?



 ii) are the distribution assumptions of the input (Q »C ,Q .C^) and $



     appropriate?



iii) are there drawbacks associated with the numerical solution requirements



     of part of the methodology?



 iv) are there other model outputs required (such as confidence llmts  on



     predicted MRl's as suggested in the Kahn memorandum)?

-------
                                  -9-
These questions are addressed in  the same order below.



            1 have found the derivations to be correct except  for  the  deduc-



tion of equations  (34) and  (35) from equation (33) on page 3-13 as noted  in



the review by Small.  However, the estimation for MEI in equation  (36)  is



correct.



          •  The assumption  of distribution function form is a  constant



problem*  Since no real empirical distribution obeys any theoretical dis-



tribution, the selection, of theoretical distribution form Is always a



compromise.  Although the issues  raised by Kahn are quite legitimate,  there



are two-factors in favor of the log-normal distribution assumption:



(i) many natural processes  are well-approximated by log-normal distributions



and (ii) the field data in  the report strongly suggest log-normal  distribu-



_tions.  Since I would find  it difficult to justify alternative distributions*



I would support tiie log-normality assumptions concerning Q ,C  ,Q   and  C  in
                                                          s  s  r       r


lack of evidence to the contrary.



            The assumption  of the log-no reality of $ is more obscure.   The



assumptions appear to have  no significant disadvantage based on the results



presented in the report.  However, I would agree that some experimentation



with other distribution function  forms for $ would be useful,



            The two computational techniques offered by the methodology



consist of an approximate method  of moments which yields closed-form



analytical solutions and a  numerical integration procedure which requires



fewer simplifying assumptions.  I am confident that a user of  the methodo-



logy could rapidly gain an  appreciation for the limits of applicability of



the approximate method by running both procedures in tandem.   I would  think



that the closed-form solutions would be appropriate for stany applications.



Even for those applications where this is not the case, the numerical

-------
                                -10-
precedures are not overly complex.  I would judge diem, to be an order of



magnitude less complex than operating simulation models as suggested in the



Roesner memorandum.



            Since there is uncertainty in parameter estimates osed in the



methodology j additional model output reflecting this uncertainty would be



useful.  The calculation of confidence intervals on ME1 as suggested by



Kahn would undoubtedly enhance the utilization of the methodology.





Functional Adequacy




            The general questions concerning the functional adequacy of



the probabilistic methodology are as follows:



  (i) does the information provided fay the probabilistic methodology strike



      a good balance between an adequate representation of the processes



      being modelled and an adequate scope for the problems being addressed,



      and as such does the methodology lend itself to a good decision



      support model?



 (ii) will users adopt die methodology?



These questions are addressed in the same order below.



            I believe that the methodology Is an appropriate model for



decision support.  The results of the methodology are quite credible and




the validations contained in the report are conviiiciiig.  The information



contained in Figures 4-6a & b is not only extremely useful but also easily



attained.  I would think that information of this kind would go a long way



to -improving decision making in the practice of urban runoff control.



            The probabilistic methodology is not as familiar to most



analysts as more conventional approaches such as simulation; however, I



believe that this should not act as a deterrent to the acceptance of the

-------
                                 -il-
methodology.  This process of user acceptance could be greatly enhanced with



"side-by-aide" comparisons with simulation model output.  As the profession



adapted to simulation modelling from pencil and paper methods, so it should



adapt to analytical models,






ALT1BNATI7ES TO THE PROBABILISTIC METHODOLOGY



            Hie only serious alternative to analytical models such as those



contained In the probabilistic methodology is continuous simulation, as



indicated in the Eoesner memorandum. •  Soesner quite properly points out that



simulation models have the advantages of not requiring many o£ the simplifying



assumptions of analytical models and being able to use historical meteorc-



logic data directly.  However, dies £ features also represent disadvantages.



Continuous simulation models require massive computer code and massive amounts



of meteorologic data.  This has meant mainframe computers and magnetic tapes



of data ^shich act ma significant deterrents to continuous simulation*  Models



such as STOSM have been available for about two decades, yet little vide-



spread practical use has been made of them.  Although such models will become



more accessible when ported to microcomputer systems, analytical models will



still have advantages.-



            In the future, analytical and simulation models should be used



together.  For the present, although continuous simulation is not widely-



used, this form of analysis is essential for Intelligent decision making in



urban runoff control practice.  Analytical models such as those proposed ia



tie probabilistic methodology can fill this void.

-------