United States Office of Air Quality EPA-450/4-79-025
Environmental Protection Planning and Standards December 1979
Agency Research Triangle Park NC 27711
__
SERA The Use of Photochemical
Models in Urban Ozone
Studies
-------
APPLICATION OF PHOTOCHEMICAL MODELS
Volume I
The Use of Photochenical Models in Urban Ozone Studies
prepared by
Association of Bay Area Governments
Hotel Claremont
Berkeley, California 94705
in association with
Bay Area Air Quality Management District
San Francisco, California
Lawrence Livermore Laboratory
Livermore, California
Systems Appli cati ons, Inc.
San Rafael, California
prepared for
U.S. Environmental Protection Agency
Office of Air Quality Planning and Standards
Research Triangle Park, North Carolina 27711
EPA Project Officer: John Summerhays
Contract No. 68-02-3046
Final Report, December 1979
-------
PREFACE
This document is one of four volumes intended to provide information
relevant to the application of photochemical models in the development
of State Implementation Plans. The reports are particularly directed
toward agencies and individuals responsible for preparation of
non-attainment plans and SIP revisions for ozone. The four volumes are
titled as follows:
Application of Photochemical Models
Volume I - The Use of Photochemical Models in Urban Ozone
Studies .
Volume II - Applicability of Selected Models for Addressing
Ozone Control Strategy Issues
Volume III - Recent Sensitivity Tests and Other Applications
of the LIRAQ Model
Volume IV - A Comparison of the SAI Airshed Model and the
LIRAQ Model
This work is to a large extent based on the photochemical modeling
experience gained in the San Francisco Bay Area in support of the 1979
Bay Area Air Quality Plan. The following individuals made significant
contributions to this work:
Association of Bay Area Governments - Ronald Y. Wada
(Project Manager)
- M. Jane Wong
- Eugene Y. Leong
Bay Area Air Quality Management District - Lewis H. Robinson
- Rob E. DeMandel
- Tom E, Perardi
- Michael Y. Kim
Lawrence Livermore Laboratory - William H. Duewer
Systems Applications, Inc. - Steven D. Reynolds
- Larry E. Reid
The authors wish to express their appreciation to John Sunmerhays, EPA
Project Officer in the Source Receptor Analysis Branch of OAQPS, for his
thoughtful review and comments on earlier drafts of this report.
-------
TABLE OF CONTENTS
Page
1. INTRODUCTION 1
BACKGROUND 1
OVERVIEW OF THE MODELING PROCESS 3
The Modeling Process for Ozone Plan Development ... 4
Schedule Considerations 8
Resource Requirements 9
Comparison of Rollback Modeling Concepts with
Deterministic Modeling Concepts 12
REFERENCES 16
2. MODEL SELECTION 19
SELECTION CRITERIA 19
THE SELECTION PROCESS 21
GUIDELINES FOR SELECTION OF GENERAL MODELING APPROACH ... 22
Urban Areas with Extensive Monitoring Networks .... 24
Urban Areas with an Average Number of
Monitoring Stations 25
Urban Areas with Limited Monitoring Networks 25
3. INPUT DATA COLLECTION AND MODEL INPUT PREPARATION 27
METEOROLOGICAL AND TOPOGRAPHICAL DATA 30
The Prototype Day Approach 30
Assessing Data Requirements 31
Sources of Data 32
Field Studies 33
Data Preparation and Preprocessing 33
LIRAQ AND SAI Model Applications 34
n
-------
TABLE OF COKTENTS (Continued)
Page
EMISSION INVENTORY DATA 35
Grid Selection 36
Overview of the Inventory Effort 37
Spatial and Temporal Distribution of
Stationary Source Emissions 39
Temporal Distribution of Stationary
Source Emissions 41
Stationary Source Projection Methods 42
Motor Vehicle Emissions 44
Hydrocarbon Species Allocation 45
AMBIENT AIR QUALITY DATA 47
TREATMENT OF INITIAL AND BOUNDARY CONDITIONS 52
Methods for Treating Initial Conditions 55
Base Year 55
Future Year Simulations 56
Boundary Conditions 57
Examples of the Treatment of Boundary
Conditions in Baseline Calculations and
Model Evaluation Studies 58
Future Year Simulations 60
Summary 61
SPECIAL FIELD MEASUREMENT STUDIES 62
Meteorological Data 63
Air Quality Data 63
Collecting Supplementary Field Data for LIRAQ .... 64
CALTRANS Field Studies in Los Angeles 66
REFERENCES 67
-------
TABLE OF CONTENTS (Continued)
Page
4. THE EVALUATION OF PHOTOCHEMICAL MODEL PERFORMANCE 69
OUTLINING THE MODEL EVALUATION STUDY 71
Definition of the Size and Boundaries
of the Modeling Region 71
Definition of Spatial and Temporal
Model Resolution 71
Time Period to be Simulated 72
Choice of Meteorological Conditions 72
Determination of the Number of
Meteorological Regimes 73
Performance Measures and Standards 74
MODEL EVALUATION PHASE 76
Adaptation of the Model to the Study Area 76
Collection of Model Inputs 77
Performance of Model Simulations 77
Assessment of Model Simulation Results 77
Rectification of Inadequate Performance 79
EXAMPLES OF PREVIOUS MODEL EVALUATION STUDIES 82
The LIRAQ Model 82
The SAI Airshed Model 83
REFERENCES 90
5. MODEL APPLICATIONS 93
BASELINE PROJECTION 93
EMISSIONS SENSITIVITY ANALYSES 94
CONTROL STRATEGY SIMULATIONS 94
ALTERNATIVE PROGRAM DESIGNS 94
IV
-------
TABLE OF CONTENTS (Concluded)
Page
PRACTICAL ASPECTS OF DESIGNING A MODEL
APPLICATION STUDY 96
MODEL APPLICATIONS IN THE SAN FRANCISCO BAY AREA 99
Baseline Forecasts 99
Emissions Sensitivity Analyses 99
Control Strategy Simulations 99
MODEL APPLICATIONS IN THE DENVER METROPOLITAN AREA .... 105
Baseline Forecasts 105
Sensitivity Analyses 105
SUMMARY OBSERVATIONS 110
REFERENCES Ill
6. MODEL INTERPRETATION 113
THE "WORST CASE" ISSUE 113
IMPERFECT MODEL PERFORMANCE 114
SPECIFICATION OF INITIAL AND BOUNDARY CONDITIONS 116
METHODS FOR MODEL INTERPRETATION USED IN THE
SAN FRANCISCO BAY AREA 116
METHODS FOR MODEL INTERPRETATION USED IN THE
DENVER METROPOLITAN AREA 119
CONCLUDING REMARKS 119
APPENDICES
APPENDIX A PREPARATION OF METEOROLOGICAL INPUT FIELDS FOR LIRAQ
SIMULATIONS IN THE SAN FRANCISCO BAY AREA
APPENDIX B PREPARATION OF METEOROLOGICAL INPUT FIELDS FOR SAI
SIMULATIONS IN DENVER AND LOS ANGELES
APPENDIX C PREPARATION OF EMISSION INVENTORIES FOR PHOTOCHEMICAL
MODELING IN THE SAN FRANCISCO AND DENVER REGIONS
-------
LIST OF FIGURES
PAGE
Figure 1-1 The Air Quality Implementation Planning Process. ... 5
Figure 1-2 Sample schedule for modeling and related plan
development tasks for preparation of a 1982 SIP
revision 10
Figure 3-1 Map of Bay Area showing the region within which 5-km
grid resolution is available in LIRAQ 38
Figure 3-2 Process for preparation of separate, disaggregated
trip end and hot stabilized motor vehicle emission
inventories 46
Figure 3-3 Schematic diagram of grid cell inputs 53
Figure 4-1 Examples from the comparison of available station
observations of oxidant with LIRAQ calculated time
histories of ozone 84
Figure 4-2 Sample scatter diagrams of observed versus
calculated hourly average N02 concentrations,
observed versus calculated station mean
concentrations, and observed verus calculated
station maximum concentrations from the LIRAQ
simulations 85
Figure 4-3 Examples of observed and SAI Airshed model predicted
hourly ozone concentrations (pphm) at various
stations in Denver on 28 July 1976 88
Figure 4-4 Estimate of bias in SAI Airshed model predictions as
a function of ozone concentration 89
Figure 5-1 Sample diagram of potential emissions sensitivity
test cases 95
Figure 5-2 Sample schedule for model application tasks 97
Figure 5-3 Example LIRAQ resul ts--2000 baseline ozone
projections 101
Figure 5-4 Plots of unadjusted and adjusted regionwide and high
hour ozone as a function of % reductions of 1985 HC
emissions 102
VI
-------
List of Figures Continued
PAGE
Figure 5-5 Control strategy testing with the AQMP modeling
system 103
Figure 5-6 Reduction in predicted ozone concentrations (pphm)
at Denver stations due to predicted emissions
changes 108
Figure 6-1 Sample statistical extrapolation to determine "worst
case" ozone in a future year 115
Figure 6-2 LIRAQ verification for two 1973 prototype days using
1975 emissions, based on 16 hourly values at 15 118
locations
-------
LIST OF TABLES
PAGE
Table 1-1
Table 2-1
Table 3-1
Table 3-2
Table 3-3
Table 4-1
Table 4-2
Table 5-1
Table 5-2
Table 5-3
Table 5-4
Estimated resource requirements for photochemical
model application 13
Partial inventory of photochemical models 23
Example levels of detail in data used as input to
photochemical model s 28
Typical variables for emission projection for the
Bay Area 43
Summary of LIRAQ compatible percent emissions by
source type 48
Sample photochemical model performance measures and
standards 75
Summary of SAI experience in evaluating the
performance of the SAI Airshed model 86
Baseline LIRAQ projections for the San Francisco Bay
Area 100
LIRAQ emission sensitivity analysis results
.100
Summary of control strategies tested in the San
Francisco Bay Area 106
Effectiveness of alternative control strategies for
the San Francisco Bay Area 107
-------
1. INTRODUCTION
BACKGROUND
The purpose of this document is to provide technical information on the
use of advanced photochemical models (primarily Eulerian grid models and
secondarily Lagrangian Trajectory models) in the development of State
Implementation Plans. Both Lagrangian trajectory models and Eulerian
grid models are mathematical representations of the physical and
chemical processes that occur in the atmosphere to form ozone*. As
their respective names imply, Lagrangian trajectory models simulate
ozone production within a parcel of air as it follows a prescribed
trajectory, while Eulerian grid models simulate ozone production and
transport in all cells of a regular square or rectangular grid.
Ozone levels well in excess of the National Ambient Air Quality Standard
have been a persistent and pervasive problem. Many of the control
measures potentially needed to attain the ozone standard across the
nation are complex, expensive, and controversial. In contrast,
previously applied techniques for determining control requirements for
precursor emissions are generally acknowledged as unvalidated and of
limited applicability. Thus, there is a need for making available to
State and local air quality planning agencies improved analytical
techniques for evaluating the effectiveness of alternative ozone control
measures.
The history of photochemical modeling has been split into two distinct
areas: research and model development, and model applications. The
state-of-the-art in ozone model development has been summarized by many
individuals in varying contexts. The 1976 International Conference on
Photochemical Oxidant Modeling and Its Control contained several papers
in this area, most notably that by Demerjian (1976). Roth, et al.
(1976) have also performed an in-depth evaluation for the American
Petroleum Institute. Ozone modeling has achieved a high level of
complexity and sophistication. While there continue to be potential
problem areas and uncertainties regarding the suitability of existing
ozone models for addressing specific issues such as long range
transport, with proper application and interpretation of results, their
use in SIP analyses is now encouraged by EPA.
*In this document reference is made to both ozone and oxidants. Ozone
is the primary component of photochemical oxidants in most areas.
Advanced photochemical models simulate ozone production, and the
current EPA reference method for oxidant monitoring measures ozone.
Previously used monitoring methods and models (e.g., EPA's "Appendix J"
rollback model) were based on oxidants. The recent change from a 0.08
ppm (1-hr) oxidant standard to a 0.12 ppm (1-hr) ozone standard places
the standard, the monitoring method, and advanced models on the same
basis.
-1-
-------
In stark contrast, the application of models in support of ozone plan
development has until recently been restricted to the most primitive
forms available. Legislative mandates and the need to develop ozone
control plans have created a demand for improved analytical tools to
assess the effects of alternative control strategies on ozone levels.
Summerhays (1976) has inventoried ozone model applications as of that
date. Those previous applications have not been conducted to
systematically assess alternative ozone control strategies. Since that
time, the Systems Applications, Inc. (SAI) Urban Airshed Model has been
applied in the Denver metropolitan area to support the evaluation of
oxidant control strategies, and the Livermore Regional Air Quality
(LIRAQ) model (see MacCracken and Sauter, 1975) has been applied in
direct support of the development of an oxidant control plan for the San
Francisco Bay Area. As suimiarized by Wada, Leong, and Robinson (1977),
LIRAQ was used to guide the selection of control strategies to attain
and maintain the (former) federal oxidant standard in the region. A
number of problem areas were encountered during the course of the
analysis and were subsequently overcome. Guiding the modeling effort
was an interagency modeling committee including modeling experts from a
broad spectrum of participating agencies. The combined expertise of the
modeling committee made it the forum for discussion and resolution of
problems related to model applications. Successful completion of the
analysis was in large part attributable to the functioning of the
committee.
Much of the guidance presented here takes advantage of the modeling
resources, experience and process developed for the Bay Area Air Quality
Maintenance Plan.
The Environmental Protection Agency has been involved over the past few
years in the development of guidelines for applying air quality models
in varying contexts:
o Evaluating indirect sources (U.S. EPA, 1975a)
o Applying atmospheric simulation models to air quality
maintenance areas (U.S. EPA, 1975b)
o Background information on the procedures, uses, and
limitations of oxidant prediction relationships (U.S. EPA,
1977 and 1978a)
o Guideline on Air Quality Models (non-photochemical) (U.S. EPA,
1978b)
o Procedures for comparing air quality models (U.S. EPA, 1978c)
This document details the use of photochemical models in the development
of ozone control strategies for State Implementation Plans and includes
procedural descriptions directed toward agencies responsible for
preparation of such plans. Particular emphasis is placed in the
document on:
-2-
-------
o planning the total effort
o collecting and preparing meteorological, topographical, air
quality, and emissions data
o assigning initial and boundary conditions
o evaluating model performance
o practical aspects of model application
o interpreting model results
In each area, general discussion is followed by specific examples of
previous model applications to illustrate how a particular problem has
actually been treated. Time and resource requirements are also noted
throughout the document to assist other agencies in developing the
modeling program most appropriate to the problems and constraints in
their regions.
The information is addressed to individuals in planning and/or control
agencies who have a basic familiarity with commonly existing emission
inventory and meteorological data bases, rudimentary modeling techniques
(e.g., EPA's UNAMAP models), and commonly accessible computer
facilities. The document covers those tasks which should be common to
most regional photochemical modeling efforts. Specific machine-related
procedures or other problems that a^e unique to each modeling effort
must be addressed individually by each model user.
OVERVIEW OF THE MODELING PROCESS
Models are representations of our current understanding and information
regarding a process. For photochemical oxidants (ozone), models are
representations of current knowledge and data regarding the origins of
precursor emissions, their transport and chemical transformation, and
the resultant distribution of pollutants in space and time. Modeling
facilitates the evaluation of complex problems through the analysis of
each sub-component of the problem. These components are then assembled
to study the problem in its entirety. Models are thus vehicles for
quantifying complex relationships, such as are embodied in the formation
of ozone and other oxidants.
Air quality models always have and will continue to contain inherent
inaccuracies. This is a reflection of practical limitations of data
acquisition, computer capacity, and an incomplete understanding of
important atmospheric processes. However, air quality problems exist
now; to make rational control decisions which directly address ambient
air quality standards established to protect public health and welfare,
no viable alternatives can be identified to the use of air quality
models. The more relevant and meaningful questions are: How will
modeling results be translated into emission control requirements, and
-3-
-------
does that procedure provide flexibility and opportunities for
"mid-course correction?"
The air quality implementation planning process is schematically
summarized in Figure 1-1. The need for air quality models is a direct
consequence of the definition of air quality goals in terms of ambient
standards. Models serve as the bridge between ambient standards and
source control requirements; they may be broadly defined as expressions
of the relationship between source emissions and ambient air quality.
A second feature illustrated in Figure 1-1 is the continuing feedback
and reevaluation of control strategies required as part of the planning
process. This feature is required in recognition of the dynamic nature
of both air pollution problems and our understanding of how to deal with
them. The key point is that it is not necessary to specify the ultimate
solution to the air quality "problem" via a single analysis undertaken
at a single point in time. Indeed, the process has been designed with
the recognition that it is not realistic to do so. This concept is
fundamental to the air quality planning process.
The Modeling Process for Ozone Plan Development
The process for applying a deterministic model in support of a plan may
be divided into several discrete steps, as follows:
o Model selection - This step consists of two parts; first,
deciding whether to use a photochemical dispersion model; and
second, selecting a specific model. All of the information
presented in this document is in part intended to assist in
the decision of whether to use a photochemical model by
describing what would be involved in such an effort. The
decision should be based on the availability of adequate time
and resources and on the need for credibility that a
photochemical dispersion model generally provides. Section 2
contains specific model selection criteria as well as a
step-by-step selection procedure.
o Data collection and model input preparation - This step
consists of:
- assessing the existing data base and identifying
inadequacies in meteorological and topographical
data, emission inventory data, and ambient air
quality data
- planning and conducting a field study to collect
supplementary air quality and meteorological data
- planning and developing suitable emission
inventories (spatially and temporally disaggregated,
speciated, and projected to future years)
-4-
-------
Define the
air quality standard
Determine the
emissions required
to achieve the standard
Define alternative control
strategies to achieve the
specified emission level
Determine the
air quality effects
of each strategy
Select the most
cost-effective strategy
for implementation
Implement
the control strategy
Monitor compliance
• monitor emissions
• monitor air quality
Determine whether
additional controls
are necessary
Figure 1-1
The air quality implementation planning process
-------
- preparing the input data in model-compatible format
Section 3 and Appendices A, B, end C contain detailed
information regarding each of these steps.
o Model performance evaluation - This step consists of adapting
the model to the study area, performing initial model
simulations based on the input data collected, assessing the
simulation results, and rectifying inadequacies in model
performance. These steps are reviewed in detail in Section 4.
o Model applications - There are a variety of model applications
that may be made depending on the specific issues to be
addressed. Section 5 describes the types of applications that
are useful in the development of an air quality plan. In
addition, a variety of ozone control issues of potential
interest are described in Volume II of this report.
o Model interpretation - Three significant issues regarding the
use and interpretation of modeling results to demonstrate the
adequacy of a given control strategy for meeting the ozone
standard are discussed in Section 6.
Careful planning of the total modeling effort will help the user to make
the best use of available resources. The user must identify the scope
of the effort, the model to be used, desired accuracy and acceptable
model limitations, the existing emissions and aerometeric data for
preparing inputs, and the resources available for collecting
supplemental data. First a general plan should be formulated to define
the anticipated applications, how model performance will be measured,
and what constitutes adequate model performance. Although these
expectations may change during the course of the project it is important
to establish pertinent goals at the outset in order to provide some
basis for making the planning decisions, especially those relating to
data collection and model modification.
After a general plan has been developed, the data requirements are
examined in light of the available data base in order to ascertain the
need to collect additional aerometric or emissions data. The planning
stage must occur far enough in advance of the actual model simulations
to allow the planning and execution of any special monitoring program
that may be necessary to collect supplemental field data. In general,
the requisite amount of input data depends on the model chosen, the
region and conditions of interest, and the desired model performance.
If NEDS emission inventory data are being considered for use, that data
should be carefully checked and updated as necessary to ensure that the
data conforms to the input requirements of the model to be used.
Once the input files have been constructed and the model has been
appropriately modified, the first simulation is carried out. The
results from this simulation are analyzed as described in Section 4. If
model performance is judged unacceptable, then either the input data
must be modified (perhaps by reanalyzing the available data or
-6-
-------
collecting additional data) or the model must be modified and the
simulation rerun. This process is repeated until satisfactory results
are obtained. Additional evaluation runs will not only supply
information on how the model performs under different conditions but can
also point out weaknesses in the data or the model. If suitable
performance cannot be achieved, it may be necessary to select another
model or to make an assessment of the consequences of using an
improperly verified model in the intended applications study.
Once adequate model performance has been verified, model applications to
address the control issues related to the air quality plan may proceed.
All previous steps have been conducted in preparation for this phase.
Here, it is important to have clearly identified the issues to be
addressed, and how the model is to be employed in each case.
Modifications to input file (i.e., emission inventory files, initial and
boundary conditions) to simulate the effect of various controls should
be thoroughly documented,, and duplicate files stored as a contingency
against the need to re-create the simulations at a future time (e.g.,
during the review of the plan).
Of crucial importance in the execution of each step is the necessity for
interagency participation and peer review of the assumptions,
procedures, and judgments made. There are two factors that contribute
to this necessity. First, as previously mentioned, air quality models
always have and will continue to contain inherent inaccuracies, such
that perfect performance cannot be expected. This means that the way in
which model results are interpreted can substantially influence
subsequent control decisions. Second, models tend to be highly complex,
such that they are viewed suspiciously by non-modelers as being "black
boxes" over which there is very little control. Under these conditions,
it is desirable to establish the credibility of the results by setting
up an open process by which technical staff from a variety of
organizations can participate in making the variety of technical
assumptions, judgements, and interpretations which must invariably be
made along the way.
The participation and review process should be accompanied by the
preparation of formal, detailed do ;umentation at each step. This
contributes to the openness of the process, facilitates review by
interested parties, and can minimize any subsequent controversy
regarding the technical basis of the resulting plan.
In the San Francisco Bay Area, the interagency participation and review
process is centered in a "modeling committee". The committee meets
roughly once per month (or more often, as necessary) to review and
discuss various aspects of the modeling work as they progress, and also
functions as a vehicle for coordinating the transfer of data between
agencies. Modeling experts from the following organizations participate
on the committee:
o Association of Bay Area Governments (lead non-attainment
planning agency)
-7-
-------
o Bay Area Air Quality Management District (primary user of the
LIRAQ photochemical model)
o Metropol itan Transportation Commission (responsible for
transportation inputs to the modeling analysis)
o Lawrence Livermore Laboratory (authors of the LIRAQ model)
o California Department of Transportation
o California Air Resources Board (State agency that must review
and adopt the plan)
o Environmental Protection Agency, Region IX (EPA regional
office that must review and accept the plan)
o Systems Applications, Inc. (a private photochemical modeling
firm)
All technical memoranda and issue papers related to the modeling work
are reviewed in draft form by this committee before being widely
disseminated.
It should be understood from the beginning that this type of open
process substantially reduces efficiency in the preparation of the
modeling analysis. (However, such a process should also minimize any
technical disagreements that might otherwise occur during the plan
review period.) Technical objectivity and a spirit of cooperation among
the participating agencies are required so that analyses can be
completed in a timely manner.
Schedule Considerations
Of the five steps in the modeling process previously listed, the
collection and preparation of input data will generally be the most time
consuming. No clear relationship between the quality and quantity of
input data, and the quality of modeling results has yet been
established. The quality or credibility of a given result is largely a
subjective assessment at this time*. Therefore, considerable latitude
exists in deciding how much time and monetary resources to devote to the
modeling analysis as a whole, and to each step of the analysis. If the
ozone problem in a given area is of such a magnitude that the control
strategies to be considered are expected to be costly and controversial,
then it is crucial that the modeling analysis supporting the plan be as
complete and rigorous as possible.
*EPA is sponsoring efforts to develop systematic model performance
evaluation methods, as well as additional analyses to determine the
sensitivity of model performance to input data of varying quality and
quantity. See Section 4 and Volume III of this report for additional
discussion.
-8-
-------
Figure 1-2 is a sample schedule for preparation of a modeling analysis
for a 1982 State Implementation Plan revision. This schedule is
presented to illustrate the phasing of tasks in relation to one another.
Actual time required for any specific application may vary from that
shown depending on a number of factors, i.e., the time required, if any,
to collect supplementary field data, the number of cases to be simulated
and the expected turnaround time at the computer facility, etc. It is
important to establish the boundaries of the schedule in the context of
the development of the total plan. Estimates should be made for time
required to prepare the plan, state and local plan review, modification
and approval, and submission to EPA. Depending on the significance of
the controls recommended and the institutional complexity of the region,
the plan review and approval process may range from 6 months to well
over a year.
Resource Requirements
Successful application of a photochemical air quality simulation model
requires personnel with several skills and fields of expertise. The
discussion below is not meant to imply that a large number of people are
required; indeed, in some instances, only a few are all that are needed
if they possess the requisite experience and understanding.
Project Manager. Some form of overall project management or supervision
is required for coordination of activities such as data acquisition and
verification, codification and inspection of model inputs, computer
simulation, analysis of results, and reporting. The project manager
should have an understanding of all facets of the air quality modeling
program, including an appreciation of how the results of the study are
likely to be interpreted and used.
Meteorologist. Able assistance in the field of boundary layer
meteorology is an indispensable part of the modeling program. The main
responsibility of the meteorologist is to determine which of the
available algorithms for computing wind fields, mixing depths, and
diffusivity fields are most appropriate for the application at hand. In
some cases, it may be advantageous for the meteorologist to develop new
algorithms for preparation of the meteorological inputs. In addition,
he must check the resultant meteorological inputs for accuracy and
consistency with physical intuition.
Atmospheric Chemist. Knowledge of atmospheric chemistry is central to
the modeling effort. An atmospheric chemist should be called upon as
necessary to estimate hydrocarbon species present as initial and
boundary conditions. Also, the chemist is involved in the analysis and
interpretation of the simulation results.
Computer Speci ali st. The computer specialist must be thoroughly
experienced with the computer system on which the model is to be
implemented—typically one of the large-scale computer systems such as
the IBM 360/370, CDC 6600-7600, Univac 1108/1110, or Honeywell 6000
series. For any particular system, the specialist must be familiar with
the system design, data file storage, tape handling procedures,
-9-
-------
CJ
oo
CD
CO
c\i
CO
CD
c
O
O
o3 9 a
~
CL c ~
cd CO £
CD
O
C. C
CO O
§ TO
a CD
o
CO
25
C
CO C
2 c 2
* .9 ro
„ t) co
3 CO Q. Q.
S 2
¥ g
> — c
CO TO CO S
>
CD
CO
O TD Q.
O C Q.
_l CO CO
CO CO
> CO
O CD
^— e i
s?
CO
dec: c
CD CO CO
,. U> CO i; (/)
S?St3 « 0
0) CD CO t t
01 C Q. Q) C
c ~ 0
E co o
<« CD
CO
c
o>
v> 5
-
2 CO c R
S £ .9 £ co
o -
S 0 CO -Q
c §
O Q. E
CD c O
c/> E o
O)
-10-
-------
structured programming, and the appropriate program language.
Additionally, the specialist should be familiar with computer plotting
hardware and software (if such is available) because much of the model
output is more easily understood when presented graphically.
Air Pollution Control Engineer/Emission Inventory Specialist.
Familiarity with the source inventory being used, its weaknesses and
inherent assumptions in its derivation are also of importance. This
expertise is of particular importance when simulating the effects of
alternative control strategies.
Data Analyst. The data analyst has two main responsibilities. First,
he inspects the accuracy and consistency of the raw air quality,
meteorological, and emissions data to be used in the data preparation
programs. This person also participates in the review of the model
inputs prior to their use in the simulation. Thus, it is highly
desirable that the analyst have an understanding of meteorology,
atmospheric chemistry, and air quality modeling. Second, he interprets
the modeling results. Although some of the models do provide software
for generating graphical displays and output statistics, it is possible
that additional analysis of the simulation results will be desired.
TechnicalIllustrator. Because the results of the modeling analysis are
likely to be distributed formally, it may be worthwhile to prepare
graphs or diagrams presenting certain of the more significant results of
the study. In that case, the services of a technical illustrator may be
desirable.
Labor and computing costs for the modeling effort are estimated on the
basis of past experience acquired by the organizations participating in
this study. The estimates necessarily presume that the individuals
performing the modeling analysis already possess a sound understanding
of atmospheric photochemical modeling in general and of the specific
intended model in particular. In the absence of this experience,
additional time will most likely be required to acquaint the user with
the model and to develop the requisite level of experience in its use.
The ranges in the estimates are rather large for several reasons,
including:
o Variability in the skill and experience of the individuals
using the model.
o Variability in the complexity of the application (e.g.,
unusual meteorological or emissions patterns).
o Variability in the scope and depth of the proposed emission
control strategies.
o Variability in the amount of existing data (e.g.,
transportation patterns, emissions, growth projections, air
quality, and meteorology) and the extent to which existing
data have been consolidated into a modeling data base.
o Variability in the time available with which to plan and carry
out the modeling study.
o The extent to which the study area has been modeled
previously.
-11-
-------
o The required accuracy of the modeling analysis.
Table 1-1 summarizes the estimates of the ranges in level of effort (in
weeks) required to perform the various tasks in the modeling program.
In estimating these figures, we assume that model performance would be
evaluated for two different meteorological scenarios. Also included are
estimated ranges in the computing time required to accomplish the
various tasks assuming the usage of a grid model, such as LIRAQ or the
SAI Airshed Model.*
The variability in the ultimate cost of preparing a model for use
depends largely on whether an adequate data base exists for the urban
area. When adequate data are available, model preparation and
performance evaluation can normally be carried out for about $75,000 to
$150,000. However, if an emissions inventory must be compiled and if a
short term intensive field program is mounted to collect additional air
quality and meterological data, then the overall costs may run as high
as $500,000 to $750,000. We note that the cost of even more extensive
supplemental emissions inventory and aeromatric monitoring programs
(such as EPA's Regional Air Pollution Study in St. Louis) can exceed
several million dollars. In view of the resources required to perform
the modeling analysis, thorough planning of all phases of the modeling
program is indispensable.
Comparison of Rollback Modeling with Deterministic Modeling
There are several key differences between deterministic photochemical
models (i.e., Lagrangian trajectory models and Eulerian grid models) and
rollback models which, if not understood at the outset, could lead to
confusion later in the planning process.
Rollback models require as input aggregated regional emission
inventories for a baseyear and future years, and a "design value"
highest expected oxidant measurement in the base year. With this
information, the percentage control of hydrocarbon emissions required to
attain the standard is computed based on the ad hoc presumption of a
linear relationship between hydrocarbon emission reductions and
reductions in maximum ambient ozone levels.
The technical basis for rollback models suffers from major deficiencies.
Among those deficiencies are the following:
0 The role of NOx emissions in the formation of oxidants is
ignored.
o Varying photochemical reactivities for different organic
compounds are ignored.
^Computer times assume usage of CDC 7600 computer.
-12-
-------
TABLE 1-1. Estimated Resource Requirements for Photochemical Model Application
Activity
PROGRAM DESIGN
Episode Selection
Control Strategy Review
Identification of data
requirements
Prescription of model
evaluation procedures
DATA ACQUISITION
Routine aerometric data
Readily available emission
1nventory
Development of a new
Inventory
Special studies—aerometric
and emissions inventory
Acquisition of new aerometric
data
Data base management
INPUT DATA PREPARATION
Mixing depths
Wind fields
Solar radiation
Initial and boundary
conditions
Hydrocarbon speciation
Emissions
Miscellaneous
Total computing resources
needed for data preparation
MODEL EVALUATION
MODEL APPLICATION
MODEL INTERPRETATION
REPORTING
TOTALS**
Level of Effort
(weeks)
2-4
1
1
1-2
1-2
2-4
50-200
1-2
100-250
6-24
4-6
4-24
2
8-12
1
4-6
4-6
—
3-16
5-40
6-16
8-12
64-625
Estimated Cost
for Labor
(dollars)
2,400-4,800
1,200
1,200
1,200-2,400
1,200-2,400
2,400-4,800
60,000-240,000
1,200-2,400
250,000-500,000
7,200-28,000
4,800-7,200
4,800-28,000
2,400
9,600-14,400
1,200
4,800-7,200
4,800-7,200
—
3,600-19,200
6,000-48,000
7,200-19,200
9.600-14.400
84,000-948,400
Computing
Requirements*
(hours)
—
--
—
--
—
1-2
—
--
1-3
--
—
--
--
--
--
--
2-6
3-8
2-10
0.5
__
10-30
•Assuming the usage of a COC 7600 Computer.
**The lower figures are for areas with extensive existing data bases that are easily
converted to appropriate model inputs; the higher figures are for areas where
extensive additional data must be collected to provide a good data base for model
application.
-13-
-------
o The air quality effect of control strategies which result in
non-uniform emission reductions over a given region cannot be
evaluated.
o Rollback models are difficult to validate or verify.
The modeling process associated with deterministic models is
considerably more complex than the rollback procedure. Such models
compute ozone concentrations for selected days on an hour-by-hour basis
at specific locations. They have been designed to directly address
those fundamental characteristics of the ozone problem which are ignored
by rollback models:
o The role of NOx is explicitly described in a chemical kinetic
model.
o Differing reaction rates for different classes of hydrocarbon
compounds are explicitly described.
o Spatial and temporal aspects of the problem and of alternative
controls can be evaluated.
o Partial verification is possible by simulating a number of
days in the historical record.
Key concepts associated with the use of deterministic models are as
follows:
The Prototype Day Concept - Deterministic models are mathematical
representations of the physical processes which lead to the formation
and transport of pollutants with time scales ranging from a few minutes
to a few days. This level of resolution means that instead of the base
year required by rollback, a base or prototype day is of greater
significance. The performance of a deterministic model is determined by
simulating several historical days in order to verify that it can
reproduce the ozone levels that were measured on those days. Having
been verified on specific days, it is then necessary to assume that the
models can forecast pollutant concentrations under different emission
scenarios using the same meteorological conditions which occurred on the
prototype days.
The basic assumption is that the meteorological conditions which lead to
adverse levels of air pollutants are recurring phenomena which are not
expected to be changed in the future by long term trends. While this
assumption is true in the sense that high pollution days have common
characteristics, it may also be expected that no future day will ever be
precisely the same as a past day. Since the precise characteristics of
future days cannot be predicted, it must be assumed that the historical
day(s) used are reasonably representative of what will occur in the
future.
The Worst Case Concept - The worst case concept is the result of
interpretations of ambient air quality standards, and the rollback
-14-
-------
models were designed to be responsive to this requirement. In
attempting to apply the concept to the more complex modeling process for
deterministic models a number of difficulties arise.
First, there must be a mechanism for relating the historical day or days
selected for modeling to worst case conditions in the the future. From
a technical viewpoint it is extremely difficult to predict the
conditions which will lead to worst case ozone levels, especially if
future emission patterns will be significantly changed from existing
emission patterns. The quantity and distribution of pollutant emissions
in a given area may change dramatically over time as a result of
implementation of control programs, and overall growth patterns. Those
conditions which led to worst case ozone levels in the past may not lead
to worst case levels in the future.
Second, the differences between "worst" case, "second-worst" case, and
"third worst" case may not be solely due to differences in
meteorological conditions. For example, these differences may be due to
quirks in the emission inventory for specific days. These factors are
difficult to document and account for, hence the inventory used is
usually a typical weekday emission pattern.
Third, the worst case for one monitoring location is not necessarily the
worst case for other locations. To consider worst case conditions at
all locations would require many more validation days (at least one for
each location), with a concomitant spiraling of time and budget
requirements.
These difficulties are not encountered under the rollback procedure
because the rollback model is not sensitive to the various factors of
concern. In Section 6, methods for addressing this problem are
discussed.
The Emissions Target Concept - Because the output of deterministic
models is in terms of air quality parameters (concentrations of
pollutants) there is no need to define an intermediate "emissions
target." The effects of alternative control strategies on air quality
can be tested directly, and because the effects of NOx and differing
hydrocarbon reactivities are included, it is probable that a number of
different emission levels will result in the same maximum ozone levels.
Thus, the concept of a single emissions target developed under the
rollback procedure is no longer applicable. Emissions targets can be
expressed in a variety of ways (different mixes of precursor emissions
and different spatial and temporal patterns), each of which could result
in attainment of the ozone standard.
It should be clear from the proceeding discussion that a subtle change
in the concepts and perspectives regarding the role of models in the
planning process must occur if a deterministic model is to be applied.
Those previous concepts that are unique to rollback modeling have tended
to become imbedded into the way many individuals perceive the modeling
process for ozone, and should be discarded if a deterministic model is
to be used.
-15-
-------
REFERENCES
Association of Bay Area Governments, Bay Area Air Pollution Control
District, and Metropolitan Transportation Commission, "Draft Air Quality
Maintenance Plan," Appendix C - Results of the LIRAQ Emissions
Sensitivity Analysis, December 1977.
Demerjian, K. L., "Photochemical Air Quality Simulation Modeling:
Current Status and Future Prospects," International Conference on
Photochemical Oxidant Pollution and Its Control, September 12-17, 1976,
Raleigh, North Carolina, Proceedings: Volume II, January 1977,
EPA-600/3-77-016.
MacCracken M.C., and G.D. Sauter, "Development of an Air Pollution Model
for the San Francisco Bay Area, Final Report to the National Science
Foundation," Volume I, October 1, 1975, Lawrence Livermore Laboratory,
UCRL-51920.
Roth, P.M., et al., "An Evaluation of Methodologies for Assessing the
Impact of Oxidant Control Strategies," prepared for American Petroleum
Institute, August 24, 1976, Systems Applications, Inc., EF76-112R.
Summerhays, J. E., "A Survey of Applications of Photochemical Models,"
International Conference on Photochemical Oxidant Pollution and its
Control, September 12-17, 1976, Raleigh, North Carolina, Proceedings:
Volume II, January 1977, EPA-600/3-7-016.
Wada, R. Y., E. Y. Leong, and L. H. Robinson, "A Methodology for
Analyzing Alternative Oxidant Control Strategies," Journal of the Air
Pollution Control Association, Vol. 29, No. 4, pp. 346-351, April 1979.
U.S. Environmental Protection Agency, "Evaluating Indirect Sources,"
Volume 9 of Guidelines for Air Quality Maintenance Planning and Analysis,
Office of Air Quality Planning and Standards, Research Triangle Park,
North Carolina, Revised, September 1978, EPA-450/4-78-001. (a).
U.S. Environmental Protection Agency, "Applying Atmospheric Simulation
Models to Air Quality Maintenance Areas," Volume 12 of Guidelines for Air
Quality Maintenance Planning and Analysis, Office of Air Quality Planning
and Standards, Research Triangle Park, North Carolina, 1975. (b)
U.S. Environmental Protection Agency, "Uses, Limitations and Technical
Basis of Procedures for Quantifying Relationships Between Photochemical
Oxidants and Precursors," Office of Air Quality Planning and Standards,
Research Triangle Park, North Carolina, EPA-450/2-77»021a, November 1977.
U.S. Environmental Protection Agency, "Procedures for Quantifying
Relationships Between Photochemical Oxidants and Precursors: Supporting
Documentation," Office of Air Quality Planning and Standards, Research
Triangle Park, North Carolina, EPA-450/2-77-021b, February 1978. (a).
-16-
-------
U.S. Environmental Protection Agency, "Guideline on Air Quality Models,"
Office of Air Quality Planning and Standards, Research Triangle Park,
North Carolina, EPA-450/2-78-027, April 1978. (b).
U.S. Environmental Protection Agency, "Workbook for Comparison of Air
Quality Models," Office of Air Quality Planning and Standards, Research
Triangle Park, North Carolina, EPA-450/2-78-028a, and EPA-450/2-78-028b
(appendices), May 1978. (c)
-17-
-------
2. MODEL SELECTION
The first step in the process for applying a photochemical model is
model selection. This step should ordinarily precede the collection and
preparation of input data, despite the fact that most deterministic
photochemical models have similar data requirements. Differences in
input data requirements and format from one model to another could
result in added costs and schedule delays if the input data is prepared
first.
SELECTION CRITERIA
In considering the selection of a model, the potential user must weigh
the advantages of greater credibility and capability against the
disadvantage of greater cost, time and personnel requirements. Specific
considerations important when deciding on a model are:
Desired Capabilities
o Physical and chemical phenomena of importance - The model
should ideally be capable of simulating the physical and
chemical phenomena that are known or suspected to be important
determinants of the ozone problem in the region.
o Previous model evaluation results - Most of the models that
are available have been previously applied to one or more
urban areas. Review of past performance may provide valuable
insight with regard to the weaknesses of a given model, and
will also provide some perspective on what may be expected for
model performance.
o Control strategies to be tested - If it is known in advance
What control strategies are to be tested, this may have a
bearing on the model selected. For example, if selective
control of hydrocarbon species is contemplated, or if relative
effectiveness of elevated point source versus ground level
area source emissions control is an important issue then
certain models are better equipped than others to address the
issues of concern. Volume II contains additional information
in this regard.
Resource Constraints
o Availability of resources to collect appropriate input data -
Input data requirements will vary from one model type to
another and from one region to another. For example, Eulerian
grid models require a more complete specification of the flow
field than Lagrangian trajectory models, and produce a more
complete representation of air quality across a region as a
result. A given region may exhibit a greater or lesser degree
of variability in its flow field depending on topography,
-19-
-------
presence of bodies of water, and other factors. The greater
the complexity the more data required to represent the
conditions. Thus, input data availability and,the potential
for additional data acquisition are important considerations
in model selection.
o Cost - The primary costs associated with the application of a
photochemical model are due to input data collection and
preparation, model execution, and diagnosis of model problems.
In each case, both staff and computer resources can be
significant. Resource limitations can effectively preclude
application of the more sophisticated models.
o Computer constraints - The accessibility of appropriate
computing facilities can be an important consideration. For
example, it would be difficult to use the LIRAQ model on any
computer other than the CDC 7600. The current version of the
SAI model is operating on CDC and UNIVAC systems. Some
program modifications would be necessary to implement the
model on other computer systems. Other models may have
comparable constraints. In addition, many government-operated
computer systems are not set up to accommodate the substantial
computing demands of photochemical models. As a minimum, the
conversion of any advanced model to operate on a computer
system other than the system on which it is currently operated
will involve additional costs and time delays.
o Personnel constraints - The capabilities of the model user in
terms of future model applications may be important. Since an
in-house capability to use the model in future applications is
generally desirable, the potential for existing or future
personnel to be trained in model application, troubleshooting
and model interpretation may be an important consideration in
deciding on the degree of complexity of the model to be used.
o Schedule constraints - Schedule constraints are usually
dictated by regulations specifying deadlines for submittal of
plans. A tight schedule may preclude the conduct of field
studies, may place substantial pressure on the preparation of
input data, and may force added expenditures for computer time
and consultants.
The criteria listed are not independent of one another. For example,
deficiencies in input data could be overcome by conducting special field
studies. Such studies tend to be expensive, time consuming, and may
involve costs comparable to or greater than the model application
effort. Consultants can be used to supplement existing user staff, or
in lieu of user staff in order to minimize personnel and schedule
constraints; again at some additional cost. Selection of an appropriate
computer facility can make a substantial difference in "turnaround time"
for model runs, but can also mean added costs.
-20-
-------
Conversely, it is often the case that resources for a given plan
development effort are fixed in advance, and thus the feasibility of
implementing a given model may be limited by the magnitude of resources
allocated.
THE SELECTION PROCESS
As previously stated, model selection involves weighing the user's needs
against available resources. In general, grid models provide the
greatest capability, credibility, and are the most expensive to apply;
EKMA is least expensive and least credible; and trajectory models are in
between. (See Volume II of this report for additional discussion.) The
process for selection of a model consists of a number of specific steps:
(1) Identify the physical and chemical phenomena that are
either known or suspected to be important determinants of
the ozone problem in the region (e.g., typical wind flow
patterns during high ozone episodes, significant sources
of background ozone, presence and behavior of elevated
inversion layer(s), presence of photochemical aerosol,
geographic extent of high ozone levels, etc.).
(2) Identify available air quality, meteorological, and
emissions inventory data--number and location of
monitoring stations, length of historical record, and
ease of access and processing of such data by computer.
(3) Identify (if possible) important aspects of potential
control strategies that might require specific modeling
capabilities. (See Volume II for discussion of examples
such as selective hydrocarbon species control, long range
transport, elevated vs. low-level emissions control,
etc.).
(4) Inventory alternative computer facilities available,
including the potential for use of large remote
facilities (it is typical 3f governments to require a
great deal of red tape and prior approval before any
non-government operated computer facility can be used).
(5) Evaluate resource constraints for the total model
application effort.
(6) Estimate the minimum number of model runs needed,
including model performance evaluation and strategy
evaluation runs (detailed suggestions for this estimate
are given in Section 5).
(7) Prepare a preliminary schedule for plan development to
obtain a sense of time available for each task. This
schedule may consist of guesses, but will still be a
valuable exercise when evaluating the feasibility of
application of alternative models.
-21-
-------
(8) Identify the most promising models through literature
review. The models identified are to be subjected to more
intense investigation.
(9) Contact EPA and/or the developers of the models selected
for further investigation to determine:
o availability of the code and user's manual
o recent updates/modifications
o computer requirements
o other questions that may have arisen during
the literature review
If it is anticipated that extensive use of consultants
will be made, this contact may take the form of a request
for proposals and subsequent proposal evaluation.
(10) Array the advantages and disadvantages of each model with
respect to the selection criteria. It is likely that no
model will be able to satisfy all of the criteria and
some tradeoff decisions will be necessary (e.g., a
typical tradeoff is cost versus technical precision).
(11) Select the model.
Table 2-1 is a partial inventory of alternative photochemical models,
indicating the model developer, where it has been previously applied,
and appropriate literature references. Additional information is
contained in Volume II. The contact point within EPA for information
regarding the models is:
Chief, Source Receptor Analysis Branch
Monitoring and Data Analysis Division
Office of Air Quality Planning and Standards
Environmental Protection Agency
Research Triangle Park, North Carolina 27711
(919) 541-5391
GUIDELINES FOR SELECTION OF GENERAL MODELING APPROACH
For urban areas, the appropriateness of a given modeling approach is
often most constrained by data availability. For the purpose of this
discussion, three levels of extent of monitoring networks providing data
are defined: extensive, average, and limited. Examples of such
categories of monitoring networks are:
-22-
-------
Table 2.1. Partial Inventory of Photochemical Models
MODEL
EKMA
DEVELOPER
U.S. EPA. OAQPS
Research Triangle Park, N.J.
PREVIOUS APPLICATIOHS
Many applications, no
documented verification
available
REFERENCES
o U.S. Environmental Protection Agency,
"Uses, Limitations, and Technical
Basis of Procedures for Quantifying
Relationships Between Photochemical
Oxldants and Precursors," Office of
A1r Quality Planning and Standards,
Research Triangle Park, North
Carolina, November 1977,
EPA-450/2:77-021a.
o U.S. Environmental Protection Agency,
"User's Manual for Kinetics Model and
Ozone Isopleth Plotting Package,"
Office of Air Quality Planning and
Standards, Research Triangle Park,
North Carolina, July 1978,
EPA-600/8-78-014a.
Lanqrangian Trajectory
ELSTAR
Environmental Research S
Technology, Inc.
Boston, Massachusetts
St. Louis
TRACE
Pacific Environmental
Services,
Santa Monica, CA
Los Angeles
SAI TRAJECTORY
EuleHan Grid
LIRAQ
Systems Applications, Inc.
San Rafael, California
Lawrence Livermore
Laboratory,
Livermore, CA
Sacramento
San Francisco
St. Louis
SAI AIRSHED
IMPACT
System Applications, Inc.
San Rafael, California
Science Applications, Inc.
La Jolla, California
Los Angeles
Denver
St. Louis
Sacramento
Sacramento
MADCAP
Science Applications, Inc.
La Jolla, California
San Diego
o Lloyd, A.C., et. al., "The Adaptation
of i Lagranglan Photochemical Air
Quality Simulation Model to the St.
Louis Region and the Regional Air
Pollution Study Data Base,"
Environmental Research and Technology,
Inc., prepared for U.S. Environmental
Protection Agency, Meteorological
Assessment Division, Research Triangle
Park, North Carolina, December 1978.
oDrivas, P.J. and L.G. Wayne,
"Validation of an Improved
Photochemical A1r Quality Simulation
Model," Pacific Environmental
Services, Inc., Pub. No. TP-016, Santa
Monica, California, March 1977.
o Drivas, P.J., "TRACE (Trajectory
Atmospheric Chemistry and Emissions)
User's Guide," Pacific Environmental
Services, Inc., Pub. No. TP-016, Santa
Monica, California, November 1977.
o Meyers, T.C., et al., "User's Guide to
the SAI Trajectory Model," Systems
Applications, Inc., San Rafael,
California, SAI Report No. EF 79-3,
prepared for U.S. Federal Highway
Administration, January 8, 1979.
o MacCracken, M.C., et al., "The
Livermore Regional Air Quality Model:
1. Concept and Development," Journal
of Applied Meteorology. Vol. 17, pp.
255-272, March 1978.
o Duewer, W'.H., et al., "The Livermore
Regional Air Quality Model: II.
Verification and Sample Application in
the San Francisco Bay Area," Journal
of Applied Heterploqy. Vol. 17, pp.
273-311, March 1978.
o Reynolds, S.D., L.E. Reid, and T.W.
Tesche, "An Introduction to the SAI
AIRSHED Model and Its Usage," Systems
Applications, Inc., San Rafael,
California, December 1978, SAi Report
No. EF 78-53R.
0 Fabrick, A., et al., "Point Source
Model and Development Study," Science
Applications, Inc., Westlake Village,
California, prepared for California
A1r Resources Board and the California
Energy Resources Conservation ana
Development Commission, March 1977.
(Appendix C is the User's Guide to
IMPACT.)
o Sklarew, R.C., "Verification of the
MADCAP Model of Photochemical Air
Pollution in the San Diego Air Basin,"
presented at the American
Meteorological Society Annual Meeting,
Reno, Nevada, January 1979.
-------
Monitoring
Network Urban Area
Maximum St. Louis
Common Houston, Philadelphia
Minimum Denver
These different levels of data base quality are described in more detail
in Section 3.
In the discussion that follows, guidelines are suggested for the
selection of an appropriate modeling approach for use in different types
of situations. For each situation, the general spectrum of measurements
that might comprise the data base are described as well as the factors
of importance that might influence the model selection process — such as
the likelihood of importing high concentrations of precursors, relative
contributions of natural hydrocarbon sources to pollutant loading, and
potential for characterizing flow patterns in the area. Recommendations
are then made concerning the modeling approach that might be most
suitable for use in the particular application.
Urban Areas with Extensive Monitoring Networks
For areas where substantial resources for data collection are available
(or where extensive supplementary data collection has already occurred),
it is appropriate to select one of the Eulerian grid models that
includes photochemistry for an evaluation of a control strategy. The
use of such a model offers several advantages. It permits one to
examine the effects of changes in:
o The spatial and temporal distribution of NOx emissions,
hydrocarbon emissions, or both;
o The reactivity of the hydrocarbon emissions;
o The background concentrations;
o The contaminants transported into the area.
If the control strategy focuses on emissions from an isolated or a small
area source, it may be advantageous to apply a trajectory-based
(Lagrangian) model; this type of model is comparatively easy to apply,
and the results it generates are likely to be less expensive and as
appropriate as those obtained using a grid-based model. However,
trajectory models are best applied in areas free from significant
horizontal variations in the concentration field, and in which there are
no significant wind shear effects. If the proposed control strategy
would be likely to cause uniform changes in the regional emissions
distribution, then EPA's EKMA model should be considered as well. In
general, however, Eulerian grid models should be considered for use as
the primary evaluation tool.
-24-
-------
Urban Areas with an Average Number of Monitoring Stations
A less extentisve and more common data base compilation effort will
result in a lower level of confidence in the accuracy of the modeling
results. Nevertheless, with the use of judicious estimates for
unavailable input information (or ranges in the input variables),
meaningful concentration estimates can be made using either grid or
trajectory model s. In situations where only the grid or trajectory
models contain the detail required for the strategy evaluation, these
models are a particularly important means of estimating the impacts on
precursor and ozone concentrations of transport, complex topography, or
proposed changes in the spatial distributions of emissions.
As the availability of data decreases and the physical character of the
areas and control strategies become simpler, use of the EPA EKMA
isopleth model becomes more attractive. If the strategy considers
control of only an isolated point source or of emissions in a limited
area, a reactive plume model or a trajectory model would be a possible
candidate for use. If the impact of transport and changes in the
spatial distribution of emissions were relatively unimportant, the EKMA
model would again be a candidate. EKMA would also be appropriate if the
proposed emissions changes were spatially and temporally uniform, with
no change in hydrocarbon reactivity.
Urban Areas with Limited Monitoring Networks
Urban areas with limited monitoring networks and for which supplementary
data are not collected are likely to experience significant problems in
attempting to evaluate control strategies. To assess the effects of
either the changes in the spatial distribution of emissions or the
reactivity of hydrocarbon emissions in regions that exhibit complex flow
behavior or variable upwind concentrations, a grid or trajectory model
should be used. The data required for the evaluation and exercise of
these models are greater in both breadth and quantity than the data
normally available. In a case where the control strategy to be
evaluated includes neither severe changes in the spatial distribution of
emissions nor changes in the reactivity of hydrocarbon emissions, and
where the area has relatively uncomplicated flow patterns and upwind
boundary conditions, then the EKMA model would be appropriate.
-25-
-------
INPUT DATA COLLECTION AND PREPARATION
Deterministic models require a substantial amount of input data in order
to simulate the production and transport of ozone and its precursors.
The purpose of this section is to outline the various types of data
required and to describe specific considerations in the preparation of
the data for model use.. The data have been divided into three
categories: meteorological and topographical data; emissions inventory
data; and air quality data. Meteorological and topographical data are
used to describe the flow and dispersion of pollutants; emissions
inventory data are the principal inputs that are varied to test the
effectiveness of control alternatives in reducing ambient pollutant
concentrations; air quality data are used to determine initial and
boundary conditions for the model and to evaluate model performance.
This section concludes with guidance in two areas related to input data
preparation: the specification of initial and boundary conditions, and
the conduct of special field studies.
The information is presented on a general level, since each model will
have unique requirements for input data format that are more
appropriately contained in their user's guides. Emphasis has therefore
been placed on acquainting the potential user with the types and
quantities of data required, potential sources of data, and the
significance of the data to the modeling effort. Table 3-1 delineates
three example levels of detail in data for photochemical modeling.
These levels represent data that are prepared for input to a
photochemical model, and may include data from both existing monitoring
systems and supplemental field data. The "maximum practical level"
corresponds to the most extensive data base currently available or
potentially available given present funding constraints and the state of
the art in photochemical modeling. In many respects, the RAPS (Regional
Air Pollution Study) data base in St. Louis is an example of this
category. A data base with a "minimum acceptable level" of detail might
be adequate for some modeling purposes, but numerous assumptions would
have to be invoked in preparing model inputs. For example, to estimate
mixing depths over a city where upper air temperature soundings are
unavailable, one might assume that the vertical temperature gradient
measured at some nearby city reflects conditions in the city of
interest. Between these two levels of detail lies the "commonly-used
level," which includes those data bases that have been employed in
previous photochemical modeling studies. This does not suggest,
however, that such a data base is well suited to model evaluation and
application. Some of the measurements generally lacking or in short
supply in such data bases may be ones to which model performance is
quite sensitive. At both the minimum and commonly used levels, a field
observation program to supplement the available data base is highly
desirable for improving model performance and minimizing modeling
uncertainties. Specific judgments regarding the nature of additional
data to be collected will depend heavily on the resources available and
the existing data base in each region; Table 3-1 may be useful as an aid
in identifying data needs in each specific case.
-27-
-------
TABLE 3-1. EXAMPLE LEVELS OF DETAIL IN DATA USED AS INPUT TO PHOTOCHEMICAL MODELS
Input
Maximum Practical Level
Commonly Used Level
Minimum Acceptable Level*
Mixing depth
Wind fields
Solar radiation
Boundary and
conditions
initial
Continuous monitoring of
mixing depths with acoustic
sounder at one or more
locations
Several (5-20) vertical
temperature soundings
throughout the day at various
locations within the modeling
region
Numerous surface temperature
measurements recorded hourly
at various locations
throughout the modeling region
One or more instrumented
towers providing continuous
measurements of the mixed
layer thermal structure
Numerous ground-based
monitoring stations reporting
hourly average values
Frequent upper air soundings
at several locations
throughout the modeling region
Continuous upper level
measurements on one or a few
elevated towers
Wind, inversion, temperature,
and terrain data used as input
to a 3-D numerical model
yielding a mass conserving
wind field
Several (5-10) UV pyranometers
located in the region,
continuously recording UV
radiation levels
Vertical attenuation of
radiation at a few locations
several times daily determined
by aircraft observations
Spatial (3-D) insolation
fields determined by
interpolation of measurements
Hourly species concentrations
extrapolated and interpolated
throughout the region using
data from the extensive
ground-based monitoring
network; airborne data also
available; hydrocarbon mix
obtained from gas
chromatographic analyses at
several times during the day
A few (2-5) temperature
soundings at different times
of the day at one or two
locations
Several surface temperature
measurements recorded at
various locations throughout
the modeling region
Interpolation from
ground-based monitoring
network and limited (1-5)
number of upper level
soundings at one or two
locations
Resultant wind field rendered
mass consistent by
divergence-free algorithm
One or two UV pyranometers;
insolation assumed constant
over the region
Vertical attenuation estimated
empirically as a function of
aerosol mass
Hourly concentrations
extrapolated and interpolated
using data from several
ground-based stations;
hydrocarbon mix obtained from
gas chromatographic analysis
at one or two stations one or
a few times during the day or
on similar days, limited
airborne data available
Single daily temperatun
sounding at an airport withi
or nearby the region bein
model ed
A few (1-3) surfact
temperature measurements wit
which to estimate tempora
variation
Interpolation from limitei
(2-5 stations) routine surfac
wind data; theoreticall
derived vertical profit
assumed
No radiation measurement
available; estimate
theoretical values based o
the solar zenith angle
Attenuation not accounted for
Hourly concentrations
extrapolated and interpolate*
from a minimal routine
monitoring network; eithei
hydrocarbon mix assumed or
average value obtained from ,
compilation of available dat<
taken in a similar area
No data on concentrati or
variations aloft
28
-------
Input
Maximum Practical Level
Commonly Used Level
Minimum Acceptable Level*
Stationary source
emissions
Hydrocarbon species
distribution of
emissions
Mobile source
emissions
Air quality data
for evaluating
model performance
Separate gridded Inventories
for point and area stationary
sources; characterization of
organic composition, and
NO/N02 emissions rates for
major sources; diurnal and
seasonal variations in nominal
emissions rates for each major
source type based on
questionnaires or stack
sampling
Mix obtained from gas
chromatographic analysis of
samples collected throughout
the region, particularly near
large sources or from stack
samples
MOBILE-1 emissions factors
used in conjunction with local
vehicle age distribution;
corridor-by-corridor VMT,
including peak and off-peak
speed distributions, vehicle
mix, and traffic data for
intrazonal trips
Spatial and temporal
distributions of cold starts
inferred from actual traffic
and demographic data
Cold start factors applied
grid by grid when calculated
mobile source emissions
Hourly averaged species
concentrations for NO, N02,
03, NMHC, CO, and particulates
from an extensive ground-based
monitoring network
Lumped, gridded Inventory for
stationary sources; no species
fractionatlon; seasonal and
diurnal variation in regional
emissions for each pollutant
Mix obtained from standard
emissions factors (AP-42)
together with a detailed
source inventory, supplemented
with one or two gas
chromatographic analyses
MOBILE-1 emissions factors,
assumed vehicle mix, and
intrazonal VMT; estimated peak
and off-peak speeds, fewer
traffic counts available for
verification, VMT available
for fewer major arterials
Cold starts temporally
resolved using traffic
distribution; no spatial
resolution or spatial
resolution only from estimates
of driving patterns
Hourly averaged concentrations
of NO, N02, 03, NMHC, CO, and
particulates from several
ground-based stations
Lumped stationary source
emissions Inventory for the
region as a whole; limited
Information on the percentage
of each source type; no
temporal variation
Mix assumed or obtained from
available data compilation,
either for the city of
interest or some similar area
Gridded VMT, emissions factors
estimated from 49 state mix,
and average (FDC) driving
profile; assumed regional
speed distribution
Cold starts as a fixed
percentage of all
driving—traffic data are not
detailed enough for spatial
resolution of cold starts;
cold starts estimated from
demographic data
Hourly averaged concentrations
of NOx, 03, THC, and CO from a
minimal routine monitoring
network
*Using data at this level of detail necessitates numerous assumptions.
29
-------
METEOROLOGICAL AND TOPOGRAPHICAL DATA
The meteorological data requirements of air quality models vary widely
depending on the sophistication of the model and the application. For
example, for some applications of a simple Gaussian plume model, only a
single surface wind velocity and an estimate of stability may be
required. In more complicated applications, information on the depth of
the mixed layer and wind values updated at specified time intervals may
be required. Lagrangian models, which follow a parcel of air, generally
require more detailed information on surface flow fields. Finally, the
Eulerian (grid-based) models require the most sophisticated time- and
space-dependent meteorological input, fields.
This document is primarily concerned with the grid-based models. The
meteorological data collection and preparation for these models
involves:
o assessing model requirements for accuracy and spatial and
temporal resolution of meteorological variables,
o locating data sources,
o planning and conductng supplemental field studies where
required,
o acquiring the topographic, wind, temperature and other
meteorological measurements,
o selecting prototype meteorological conditions (normally from
the set of days for which supplemental field data are
available),
o selecting the gridded region,
o generating properly formatted inputs for preprocessors, and
o examining output fields and modifying the inputs to the
preprocessors in an iterative manner until the output fields
are thought to be realistic.
The Prototype Day Approach
For air quality planning we attempt to simulate days in a recent base
year on which high ozone levels were observed. Once satisfactory model
performance is achieved for such days we then assume that the model can
forecast ozone concentrations for different future emissions scenarios
under the same meteorological conditions that occurred on the validation
days. This assumption is based on the expectation that a) similar
adverse meteorological conditions will re-occur in the future, and b)
those conditions that now produce high ozone levels will do so in the
future. Ideally, one would like to identify the adverse conditions
associated with future emissions patterns; however, this is difficult at
best to accomplish based on current methods.
-30
-------
Preparation of meteorological inputs for sophisticated grid-based
photochemical models is very labor-intensive, time-consuming and
therefore expensive. For example, preparation of meteorological input
files for simulation of one 24-hour period for LIRAQ in the San
Francisco Bay Area may require up ;o three months of effort of an
experienced meteorologist. Therefore, simulations are ordinarily
performed on a small number of prototype periods ranging in duration
from several hours to a few days. Because of the expense, it is
important to select very carefully those prototype days that are to be
used in model ing.
The selection process often involves making some trade-offs between the
desirability of a day for simulation and the amount of meteorological
information available for that day. For example, if field studies were
performed in a base year, they may not have included the day with the
highest or second highest, or most widespread ozone levels during that
year. (Because the ambient air quality standard for ozone is
interpreted in terms of the worst days, it would be desirable to
simulate the worst or second worst day of a given year.) The
meteorological data base for the worst or second worst day may not be
sufficiently detailed to ensure adequate model performance. In such
cases a less desirable day with a good data base would be preferred,
provided that there is subsequent analysis to determine a worst case
scenario as discussed later in Section 6.
Assessing Data Requirements
One of the first steps in planning air quality simulations is to assess
meteorological data-gathering and processing requirements. This step is
crucial: first, because decisions will have a large impact on the cost
of the project (e.g., how large and expensive a field study is
necessary?), and second, if data requirements are misjudged, the model
may fail to produce acceptably realistic simulations.
The grid-based models require time-dependent, two or three-dimensional
fields of wind velocity and mixing height throughout the simulation
period. These fields are specified by inputing limited surface wind and
temperature observations and even more sparse wind and temperatures
aloft into special preprocessing programs. Because the space/time
density of wind and temperature measurements is always limited, various
assumptions must be made about the structure of the atmosphere aloft for
input to the preprocessors. Preprocessors are designed to produce
physically consistent time-dependent mixing height fields and wind
velocities for each grid cell in the trodeled region.
The required density, in time and space, of meteorological inputs to the
preprocessor programs depends upon: a) the computational requirements
of the codes needed to perform adequate interpolation, extrapolation or
smoothing from limited available measurements, b) the geographical
complexity of the region to be modeled, c) the complexity of atmospheric
circulations on those days chosen for simulation, and d) the model's
sensitivity to variations in the meteorological variables. In a region
with flat, relatively uniform terrain a comparatively sparse network of
-31-
-------
observations may be adequate to characterize the region's wind and
mixing height fields. On the other hand, in a geographically complex
region like the San Francisco Bay Area, where winds are channeled by
terrain and are subject to land/sea breeze reversals, and where land/sea
effects produce large horizontal gradients in mixing height, a large
body of observational information must be collected, analyzed and
formatted so that preprocessors can produce realistic characterizations
of the wind and mixing layer dynamics.
Initial assessments of meteorological data requirements should be made
by a group of individuals with expertise in a) the functioning and input
requirements of the model and preprocessor codes, b) local air quality,
c) the small- and mesoscale local atmospheric circulations --
particularly during air pollution episodes, and d) any past or ongoing
local research or field studies that might yield insights into the
nature of wind and stability fields. Such a group would be competent to
address such questions as:
o What meteorological conditions characterize local air
pollution episodes?
o Which days should be chosen for simulations?
o What grid network is needed to track the pollutant cloud?
o What is the reliability of local meteorological measurements?
(including evaluation of instrument exposure and averaging
times)
o Is the network of local surface observing stations
sufficiently dense?
o Are data from the nearest upper-air measurement station
representative of the area?
o Is the available data base, together with knowledge of local
conditions, adequate to prepare model inputs or are additional
field measurements required?
Sources of Data
Topographical data can be obtained from the United States Geological
Survey (U.S.G.S.). Average elevations of specified grid compartments
can be determined from U.S.G.S. contour maps. Data for the San
Francisco Bay Area for LIRAQ were taken from a U.S.G.S. California
topography tape that gave mean elevations for 1x1 minute sub-quadrants
that were interpolated to a 1 km grid.
Surface measurements of meteorological variables may be available from a
variety of sources - at least in the metropolitan centers. Local
meteorologists should be consulted when performing a survey of available
local data. Surface measurements of wind velocity and temperature are
often collected at major airports and military installations, by air
-32-
-------
pollution control agencies, colleges and universities, by various
Federal, state and local governmental agencies and by some public
utility and other private companies. Weather data from many sites
throughout the United States are placed in archives by and available
from:
National Climatic Center
Environmental Data and Information Service
National Oceanic & Atmospheric Administration
Federal Building
Asheville, NC 28801
The scarcity (in time and space) of upper-air wind and stability
measurements is usually the most troublesome problem facing the analyst
who is attempting to prepare meteorological input fields to
sophisticated grid-based photochemical models. The National Weather
Service (NWS) operates a nationwide network of upper-air stations that
are spaced approximately 250 to 500 miles apart. Soundings of wind,
temperature and humidity are made twice daily, at 12:00 and 00:00 GMT.
In many areas the NWS upper-air network is the only source of scheduled
observations aloft. In some areas, however, routine upper-air
measurements (from pilot balloons, radiosondes, instrumented aircraft,
towers, or acoustic sounders) may be made by universities, some
governmental regulatory agencies or research organizations.
rf
Field Studies
When the available routine meteorological observations in a given region
are judged to be inadequate to construct two- or three-dimensional, time
dependent wind and mixing-height fields, special field observation
programs may be needed. Typically such observations, designed to fill
the most important gaps in the existing observation network, include
supplemental measurements of winds and temperatures aloft obtained from
pibals, radiosondes, instrumented air:raft and acoustic sounders. They
should also include additional surface wind, temperature, and solar
radiation measurements in the more data-sparse or topographically
complex areas. Occasionally a field experiment is performed to study a
specific phenomenon; an example of this was the fluorescent particle
tracer experiment conducted in the Bay Area to study vertical mass
transport across the inversion interface (MacCracken and Sauter, 1975).
Further discussion of field studies is presented later in this section.
Data Preparation and Preprocessing
After prototype days have been selected and all the meteorological and
air quality observations have been gathered, several preprocessing tasks
must be performed to convert the observations into a format compatible
with model input requirements. First, it will probably be desirable to
key some of the data into computer data files to facilitate
error-checking, conversion of units, preparation of inputs for
preprocessor codes and for input to post-processor (model performance
-33-
-------
evaluation) codes. A detailed discussion of the data management system
that was set up for LIRAQ applications in the Bay Area can be found in
MacCracken, 1975.
Grid-based models generally require that wind velocity, mixing layer
depth and other variables (e.g., solar flux density) be specified for
each grid cell in the modeling region. Obviously it is not practical to
measure the variables in each cell, so one or more preprocessors are
employed to compute values for each cell from relatively sparse
observations. Preprocessors generally perform interpolation,
extrapolation and, sometimes, smoothing. They may also calculate wind
and other quantities aloft based on a few observations and specified
assumptions regarding the vertical distribution of wind and stability.
The choice of a particular preprocessing scheme will depend upon the
structure and complexity that one wants to attribute to the
meteorological fields. More specific comments on preprocessors used for
the LIRAQ and SAI models are given in the next section.
Once a preprocessing methodology is decided upon, the necessary input
fields must be prepared from the set of observations. Interpolation
schemes often tend to give unrealistic values in data-sparse portions of
the grid. In order to prevent this, it is sometimes necessary to supply
fictitious values at some locations. The required number, positioning
and magnitudes of the fictitious values are largely determined by
experience. It may be necessary to modify the inputs and rerun a
preprocessor several times before satisfactory output fields are
obtained -- fields that are ready for input to the air quality model.
LIRAQ and SAI Model Applications
The methodologies for preparing meteorological input fields for
grid-based models will vary because models are different and because
each region that is to be modeled is unique. Thus, although general
guidance can be given, specifics must be defined in the course of the
actual model application in a particular area. In order to provide
insights into some of the problems that are encountered and how they are
treated, the following discussion highlights some of the methods and
assumptions used in LIRAQ simulations in the San Francisco Bay Area and
in the SAI Urban Airshed Model applications in Denver and Los Angeles.
The selection of prototype days in all three regions relied heavily on
those days for which supplemental meteorological field data were
available. Prototype day selection often involves deciding whether to
simulate a day with "ideal" conditions (such as the day with the year's
highest or second highest ozone concentration) or a day with more
complete meteorological information.
The choice of the gridded modeling region also involves tradeoffs.
Because of computer limitations, the number of grid cells that can be
handled is limited. LIRAQ simulations in the Bay Area were performed
with a 20 x 20 grid of 5 km elements. In some of the SAI simulations in
Denver and Los Angeles, each region was subdivided into 900 two-mile
squares. The modeling region must always be selected carefully to
-34-
-------
ensure that important source and receptor areas are included, that
boundary effects are minimized, and that flow reversals on a particular
prototype day do not transport the pollutant "cloud" out of the region
and then return it later.
Preprocessing of inversion base height fields is best tailored to the
region that is to be modeled. Overall, the preprocessing of these
fields was relatively uncomplicated in the SAI/Denver studies and
complex in the LIRAQ/Bay Area studies. For example, in the Denver
simulations the depth of the mixed layer was assumed to be uniform in
space. In the Los Angeles studies the depth of the mixed layer was
assumed to be a function of distance from the ocean. In the Bay Area,
because of the influences of the ocean, bays and adjacent mountain
ranges, relatively complicated inversion base height fields were
constructed.
Methodologies for generating wind fields can also be tailored to a
specific region. In the Denver, Los Angeles, and Bay Area simulations
surface wind observations were fed into an interpolation scheme to
obtain a field of surface wind values throughout the gridded region. In
Denver, winds aloft were set equal to surface values in one set of
simulations. In later simulations a more sophisticated approach that
accounts for local convergence or divergence effects was used. In the
Los Angeles studies three different methods were used to calculate the
three-dimensional wind field. Two of these produced reasonable results.
In the Bay Area the MASCON mass-consistent atmospheric flux model was
employed (Dickerson, 1978). The MASCON code was also used to generate
two-dimensional fields of atmospheric transmissivity from Eppley
pyranometer measurements taken at nine locations. Appendix A presents a
more detailed summary of the meteorological inputs used in the LIRAQ/Bay
Area simulations. Appendix B presents the same kind of example
information for the SAI studies in Denver and Los Angeles. These
detailed examples are included to enhance the reader's understanding of
how the data have been prepared in the past, and are not necessarily
intended as guidance for future efforts.
EMISSION INVENTORY DATA
A multitude of guidelines documents, computerized systems, and reference
works have been prepared to assist the development of emission
inventories. The most recent and relevant guidance is being published
by EPA under the title:
Procedures for the Preparation of Emission Inventories
for Volatile Organic Compounds
Volume II Emission Inventory Requirements for
Photochemical Air Quality Simulation Models
(EPA-450/4-79-018)
-35-
-------
That report contains the basic guidance necessary for preparation of
emission inventories for photochemical models. The material presented in
this section is intended to complement previously published guidance in
this area by focusing on the aspects of inventory preparation that are
of key importance to planning applications of photochemical models.
The general problem for source inventory builders for photochemical
models is to develop procedures and techniques to adapt an existing
source inventory to photochemical model format. As a minimum, this
requires a change from the normal basis of tons/day, annual average, by
county, to a basis of gm/sec, hourly average, by grid square for a
specific day. In general, it should not be assumed that an adequate
inventory already exists.
The purpose of this section is to summarize general procedures and
techniques for spatial and temporal distributions and projection
methods. Emphasis is focussed on the San Francisco Bay Area experience,
and the following areas are covered:
o Grid selection
o Spatial distribution methods for stationary sources
o Temporal distribution methods for stationary sources
o Projection methods for stationary sources
o Spatial and temporal distribution and projection methods for
motor vehicles
o Hydrocarbon species allocation.
In addition, detailed example descriptions of inventory preparation
methodologies used in the San Francisco Bay Area and in the Denver
metropolitan area are contained in Appendix C.
Grid Selection
A primary planning consideration for photochemical modeling is the
determination of the grid system. It is extremely important to develop
an appropriate grid system at the start so that all the source emission
estimates are referenced to one grid system.
The area covered by the emission grid should be large enough to 1)
include all major emission sources in the region within the grid, 2)
include the receptor areas where highest pollutant concentrations are
expected, 3) encompass areas of future industrial and residential
growth, 4) include as many ambient pollutant monitoring stations as
possible for model performance evaluation, and 5) include key features
of the prevailing meteorology in the region.
-36-
-------
Conversely, the size of the individual grid cells should be as small as
possible in order to maximize the spatial resolution of the impact of
individual sources. However, smaller grid-cell size requires additional
effort to collect the necessary emission data and to develop appropriate
techniques to allocate source emissions into the grid cell level.
A standard grid system utilized as the coordinate system for
photochemical modeling is the Universal Transverse Mercator (UTM)
system. This system is used in the NEDS and EIS emission data systems
to reference all point source locations. Other common grid systems are
the latitude and longitude system, and the State Plane Systems.
However, it is very important to use a uniform coordinate system,
because changing from one coordinate to another is not straightforward.
For example, changing from state plane coordinates to UTM coordinates is
a difficult procedure involving computerized calculations.
A typical grid size in a number of urban areas where photochemical
models have been run is about 80 km by 80 km, with grid cells typically
being about 2 x 2 km. A typical LIRAQ grid is shown in Figure 3-1 for
the San Francisco Bay Region; the grid is 100 km by 100 km and consists
of 5 x 5 km grid cells.
Overview of the Inventory Effort
There are three different types of inventories that may be prepared for
control strategy planning purposes:
o Prototype day inventory - This is the inventory applicable to
an average day in the historical year that includes the
prototype meteorological days being modeled, and is to be used
in the model performance evaluation. Ideally, the inventory
should correspond to the prototype day(s) being modeled;
however, in practice tailoring the inventory for specific days
is possible for very few source categories (e.g., electric
utility boilers). Seasonal adjustment factors may be used.
o Baseline projection inventories - These are inventories
appropriate to an average day in future years that are
projected assuming that no new controls are implemented beyond
those already adopted. For example, the effects of continued
implementation of the federal motor vehicle control program
should be included in these inventories as estimated in EPA's
MOBILE1 computer code. Appropriate future years for baseline
inventories might be 1982 and/or 1987 to address attainment
requirements of the Clean Air Act, and possibly 2000 to
address long term maintenance requirements.
o Control strategy inventories - For each alternative control
strategy to be tested, a separate inventory should be produced
that includes the estimated effects of the strategy on
emissions in each future year of interest.
-37-
-------
4300
O>
C
IE
500 520
540 560 580 600 620
UTM coordinates (easting) — km
4090
660
Figure 3-1. Map of Bay Area showing the region within which.
5-km grid resolution is available in LIRAQ.
38
-------
The prototype day inventory is needed to check model performance against
ambient monitoring data. The baseline inventories are useful for
determining what future air quality will be and whether additional
control programs beyond those already adopted are necessary to meet the
ozone standard. The control strategy inventories are useful for
simulating the effects of alternative strategies and whether they will
result in attainment and/or maintenance of the ambient ozone standard.
Each of these inventories must be disaggregated spatially into grids,
temporally (usually by hour of the day), and by species (usually into
two or more hydrocarbon classes and NO and N02).
The preparation of these inventories requires an investment of
substantial manpower and computer resources above and beyond that
normally devoted to the preparation of emission inventories. Estimates
of costs to produce a complete set of inventories may vary from $20,000
to $250,000 depending on the size of the region, the completeness of
existing data bases and computer codes in place, and the skills of
available staff.
A final note before proceeding into more detailed discussions is that as
all of these inventories are being developed, care must be taken to
ensure consistency from one inventory to the next. The methods used to
disaggregate the inventories should be the same for prototype day,
baseline, and control strategy inventories. Energy assumptions, growth
assumptions, etc. should also be consistent. In this way, the
differences from one inventory to the next will in fact reflect the
changes that are desired to be measured and evaluated through the
modeling process, rather than artificial differences introduced through
the use of inconsistent methodologies.
Spatial Distribution of Stationary Source Emissions
For the purposes of this document, stationary sources are classified as
point, airport, and area sources. Point sources are generally
considered to be those that emit at least 100 tons per year of air
pollutant from a stack or group of stacks. Area sources are
individually small emissions that are impractical to consider as
separate point or line sources. It should be noted that the value 100
tons per year is somewhat arbitrary, but it is in common use; smaller
cutoff levels for point sources would improve the inventory for modeling
purposes. Airports are included as a separate category because
emissions from airports are neither spread over an entire county nor
concentrated in a single stack.
Point Sources. Point source data are available nationally from the NEDS
and EIS emission systems and locally from air pollution control agencies
in the modeling region. Regardless of the data source to be used, it
should be specifically reviewed to determine whether the data for each
source is up to date and conforms with the input requirements of the
model to be used.
-39-
-------
In general the major point sources present no serious problems in
spatial distribution. One can pinpoint their locations with a street
map, and read their UTM coordinates with a USGS map to the nearest 0.1
km. Major point sources are carried as separate listings and may be
further divided according to stack height. The stack height categories
.differ from one model to another. The SAI model injects the emissions
into an appropriate layer based on the estimated plume rise for each
individual point source while LIRAQ treats only two categories,
"surface" and "elevated" categories, depending on whether the stack
height is greater or less than 100 feet (30.5 meters).
For preparing photochemical modeling inventories the following
information for point sources is very useful and should be included in
point source listings.
o Location of UTM grid coordinates
o Source type
o Annual emission rates for each pollutant
o Normal operating hours
o Seasonal distribution of emissions
For models that have vertical resolution of pollutants the point source
inventory should include stack parameters: stack height, stack
diameter, exit gas temperature, and gas flow rate.
Airports. Since airports also have known locations, some inventories
treat airports as point sources while others treat them as an
independent source category as a matter of convenience. Because some
landing and take-off emissions are spread out within the mixing layer,
it is more appropriate that airport emissions are distributed over
neighboring grid squares in addition to the several grid squares where
the airport is located.
Area Sources. For air quality modeling, aggregated area source
emissions must be distributed in some fashion over the geographical
region of interest.
The simplest approach would be to distribute the emissions uniformly
over a given study area, but this approach might result in a substantial
fraction of area source emissions being distributed over undeveloped
areas (e.g., mountain ridges, bay, marshlands, and undevelopable lands,
etc.). The next level of sophistication would be to distribute the
emissions according to population. This would be a great improvement,
but some major flaws remain. Most census data concern residential
population only and would misplace the many non-major point sources
which operate in industrial and commercial areas. A more accurate
distribution can be achieved, however, if source activities can be
correlated with some sub-group of population, or with categories of
employment or land use.
-40-
-------
Due to lack of appropriate procedures and data, a commonly used spatial
distribution method for area sources in many modeling studies is the
spatial distribution based on population or uniformly gridded emission
factors for area sources. In most metropolitan areas, local agencies
have some sort of land use data. Land use information is normally
available by land use categories (e.g., residential, commercial and
industrial, etc.) and by certain planning zones. From these land use
data, one can set up an allocation procedure that involves determining
the relationship of each type of land use to each emission source
category.
A cross classification of source categories with land use data can be
fairly time consuming, depending upon the detail of the land use data
available in a given region. However, the use of surrogate indicators
derived from land use categories is often the best available procedure
for apportioning regional area source emissions to individual grid
cells. (See Perardi, et al., 1979, for more details.)
Temporal Distribution of Stationary Source Emissions
In order for photochemical simulation models to adequately predict
hourly ozone and other pollutant concentrations, typical hour-by-hour
emissions estimates are needed at the grid cell level. Several basic
approaches can be used for providing the temporal detail needed in the
inventory used in a photochemical model. The most accurate and exacting
approach would be to determine the emissions for specific sources for
each hour of the time period being modeled. This approach is generally
most applicable for point sources. An alternate approach is to develop
typical hourly patterns of activity levels for each source category and
then apply these to the annual or seasonally adjusted emissions to
established hourly emissions. This approach is more appropriate for
area sources.
Normally, the photochemical air quality model, and therefore the
emission inventory, is applied during the season of the year in which
weather is most conducive to ozone formation; for most locations, this
means the summer months. Similarly, emissions are usually chosen to
represent the day of the week on which pollutant-emitting activities are
at a maximum, normally an average weekday. In this case the prototype
days would be summer weekdays. Seasonal adjustments should be made
first, to change the aggregate source inventory from an annual average
basis to a summer weekday, typical of the ozone season.
Airports. Aircraft emissions generally are divided into three inventory
classifications: commercial carriers, military, and general aviation.
Commercial carriers are most important from an emission viewpoint, and
fortunately these are also the best documented. Comprehensive schedule
books keep up-to-date listings of commercial flights to all the major
airports of the world with arrival and departure times and aircraft
type. Temporal resolution factors for commercial airports in a given
area could be compiled based on this kind of data.
-41-
-------
Area Sources. The hourly distribution of area sources can be based on
diurnal variation coefficients (percent of total daily emissions emitted
in each hour) for each source classification. These coefficients could
be compiled by source inventory engineers familiar with the types of
sources in each classification.
Stationary Source Projection Methods
Because of the need to determine whether a given area will achieve the
ozone standard, or remain in compliance in future years, the
inventorying agency must develop emission projections for future years
for use in the photochemical models.
There are many approaches for making emission projections; perhaps the
simplest and most frequently used technique is extrapolation or
regression where a series of historical data points are projected into
the future in some simple linear fashion. Here the historical emissions
estimated are correlated to any number of variables (e.g., population,
fuel use, etc.). What results is an equation which represents the
"best" correlation of the variables to changes in emission rates for use
in projecting future emissions. In essence, it is assumed that the
factors which have apparently accounted for historical emissions will
continue to do so in the future in the same manner. To forecast
emissions using this technique requires a forecast of all the variables
assumed to be important in accounting for historical emission trends.
Another emission projection approach is the use of surrogate variables.
This procedure assumes that a given pollution source category can be
accurately projected by projecting some related variable. For example,
increases in aircraft emissions might be based on projected increases in
passenger air travel. Similarly, hydrocarbon emissions from internal
combustion engines might be projected on the basis of forecasts of
population growth.
Frequently, very little data is available to assist in making future
projections. In these cases, engineering or scientific judgment can be
used. (See Leong and Wada, 1978.)
There are two EPA publications which deal with methodologies and
procedures for estimating emissions in the future years: the Regional
Emission Projection System (REPS) (see Booz-Allen and Hamilton, 1974),
and the Guideline for Air Quality Maintenance Planning and Analysis,
Volume 7: Projecting County Emissions (U.S. EPA, 1975). In general,
the methodologies and data bases employed or recommended in these
documents provide very crude estimates of projected emissions, which
should be carefully reviewed before being used in a photochemical model
application.
-42-
-------
The basic emission projection methodology employed in the San Francisco
Bay Area is the use of surrogate variables. For each of 107 source
categories, related variables are defined and necessary data for
projecting related variables are studied. This procedure assumes that a
given pollution source category can be accurately projected by
projecting some related variable or variables. For example, future
emissions from aircraft in the Bay Area were estimated from projected
air passenger travel, aircraft type and emission factors in a given
future year. Table 3-2 presents an example of some source categories
and related variables used for projection.
TABLE 3-2
TYPICAL VARIABLES FOR EMISSION PROJECTION FOR THE BAY AREA
SOURCE CATEGORY
VARIABLES USED FOR PROJECTION
Petroleum Refining
Chemical Processing
Industrial Coating
Petroleum Evaporation
Other Organic Compound
Dry Cleaning
Domestic Combustion
Industrial Combustion
Power Plant
Aircraft
Oil supply
Oil supply and population
Population
Oil supply and gasoline demand
Population and other factors
National trend
Population
Natural gas and other factors
Fuel usage, natural gas usage
and coal usage
Aircraft operations and air
passenger miles
-43-
-------
Motor Vehicle Emissions
Inventory guidelines previously referenced have described alternative
methods for preparing disaggregated mobile source inventories. The
method best suited for planning purposes involves the use of previously
developed computerized transportation models. Most states and/or major
metropolitan areas use such models to assist their transportation
planning and decision-making processes. They have the capability of
simulating the effects of a variety of (though not necessarily all)
transportation controls. Use of these models in cooperation with
transportation planning agencies will ensure proper projection and
spatial and temporal disaggregation of motor vehicle emissions.
The standard system used nationwide for forecasting regional travel
patterns and highway and transit needs consists of a series of models as
follows:
o Trip Generation Model
o Trip Distribution Model
o Modal Split Model
o Network Assignment Model
Two key outputs from these models for air quality evaluation are: (1) a
trip table indicating the number of daily trips originating and ending
in each traffic assignment zone*; and (2) an "historical record" file
that indicates average daily traffic volumes and average speeds on a
geographically coded highway network.
A number of steps are required to convert the output from the
transportation models into gridded, hourly emission estimates suitable
for input to a photochemical model:
o Appropriate motor vehicle emission factors must be obtained.
o There must be a computer code that reads the transportation
data and applies the appropriate emission factor.
o The resulting emissions must be assigned to the appropriate
grid, and disaggregated to each hour of the day, and possibly
by species depending on the photochemical model to be used.
o The data must be arranged in the appropriate format for input
to the photochemical model.
*Traffic assignment zones are cells of irregular size and shape which
subdivide the urban area or study region to the degree of detail
necessary for transportation modeling.
-44-
-------
A decision should be made as to whether trip-end emissions (emissions
due to cold starts, hot starts, and hot soaks) should be treated
separately or lumped with hot-stabilized emissions. It is evident that
the spatial distribution of trip-end emissions would be somewhat
different from hot-stabilized emissions, although whether the
differences are significant enough to merit separate consideration is
unknown at this time.
Figure 3-2 illustrates the process for preparing separate inventories
for trip end and hot stabilized emissions. EISA's MOBILE! code (U.S.
EPA, 1978) for computing motor vehicle emission factors is exercised a
number of times to produce separate hot stabilized and trip end emission
factors for a given future year at different average speeds and ambient
temperatures. Hot stabilized emission factors are then applied to the
historical record file output from the transportation models to compute
average daily emissions on each highway segment. These data are then
disaggregated into hourly portions and assigned to the appropriate grid
squares covering the region to be modeled. The standard Federal Highway
Administration codes SAPOLLUT and SAPLSM (U.S. DOT, Federal Highway
Administration, 1976) may, with only slight modification of the emission
factors, be used to compute emissions and assign them to grids as
stated.
Trip end emission factors are applied to the trip table, and the
resulting emissions are assigned to traffic assignment zones. To
convert these data into gridded emissions, a method for overlaying a
regular square grid onto the system of irregularly shaped traffic zones
must be developed. In its most straightforward form, this overlay would
consist of a conversion, table that lists the percentage of each traffic
zone lying within each grid. (This table may also be used to allocate
intrazonal trip emissions to grids.) Other methods are possible, and
since there are no standard programs for preparing gridded inventories
from data on a traffic zone basis, the model user is left to exercise
judgment and ingenuity in the preparation of the inventory.
After both hot-stabilized and trip end emission data files are prepared,
they may be merged into a single motor vehicle emissions file.
The simpler alternative of using composite trip end and hot stabilized
emission factors eliminates the use of the trip table as well as the
need for a method to convert traffic zone information to grids. (As
long as intrazonal trip emissions are a small component, they may be
represented by adding small fictitious links to the network.) In Figure
3-2, this alternative would consist solely of the left half of the flow
chart using composite emission factors rather than hot stabilized
emission factors.
Hydrocarbon Species Allocation
Because air quality models attempt to simulate complex photochemistry,
hydrocarbon emissions must be distributed into various species classes.
In addition, NOx emissions have to be distributed into NO and N02.
-45-
-------
Motor vehicle
emission factors
(mobile 1)
Transportation
model
Hot
stabilized
emission
factors
(gms/mt)
Average weekday
highway link
traffic volume
& speed data
Trip end
emission
factors
(gms/trip)
Zone-to-
zone
trip table
Emissions
calculation code
(e.g., SAPOLLUT)
Emissions
calculation
code
Link-to-grid
emission allocation
code(eg.,SAPLSM)
Zone-to-grid
conversion code
Hourly
gndded
emissions
Hourly
gndded
emissions
Emissions input
file for
photochemical
model
Figure 3-2
Process for preparation of separate, disaggregated
trip end and hot stabilized motor vehicle emission
inventories
-46-
-------
The typical procedure for allocating hydrocarbons to species classes is
(1) to assume that the total hydrocarbon emissions from each type of
source contain a certain percentage of each class of compound, and (2)
to apportion the total hydrocarbon emissions by these percentages and
thus determine emission totals of each species class by source category.
Each photochemical model requires specific types of volatile organic
compound (VOC) input data from an emission inventory, and this VOC
information must be in proper format for use by the model in this
computations of chemical reactions. The EPA report, "Volatile Organic
Compound (VOC) Species Data Manual," by KVB Engineering, Inc.
(EPA-450/3-78-119) may provide useful data in this regard.
For example, the LIRAQ model contains a 48-reaction chemical mechanism
with three main classes of hydrocarbons, designated HC1 (mainly alkene),
HC2 (mainly alkanes, simple aromatics, ethers, alcohols), and HC4
(mainly aldehydes, some ketones and some aromatics). Table 3-3
summarizes the percentage breakdown of hydrocarbon into the three LIRAQ
classifications by source type as used in the San Francisco Bay Area.
Such estimates should in general be made in consultation with personnel
familiar with the composition of organic emissions and the treatment of
atmospheric chemistry in the model of interest.
AMBIENT AIR QUALITY DATA
This section discusses the basic air quality monitoring data that are
needed to set initial and boundary conditions and evaluate the
performance of photochemical models. The discussion is followed by two
examples of efforts used to gather data for performance evaluation.
The evaluation of a photochemical model's performance primarily consists
of comparing the ambient pollutant levels predicted by the model with
the actual ambient pollutant levels measured by a monitoring network.
The purpose of this comparison is to determine the accuracy of the model
predictions. Theoretically, the most accurate way to achieve this
comparison is to insure that the spatial and temporal characteristics of
the predicted and measured ambient concentrations are similar. For
example, if the model output is expressed in terms of 1-hour ozone
concentrations averaged over a grid area of 1 square kilometer, then the
ambient data should match this as closely as possible.
The following ambient air quality data and related information are
needed to evaluate the performance of a photochemical model:
o measured hourly average ambient concentrations of HC, NO, N02,
and 03 (and in some cases CO and S02);
o the location (geographical coordinates and altitude) of each
monitoring station at which the measurements were made;
o the analytical methods that were used to measure the ambient
levels of each pollutant.
-47-
-------
TABLE 3-3
SUMMARY OF LIRAQ COMPATIBLE PERCENT EMISSIONS BY SOURCE TYPE
Source Type HC1 HC2 HC4
Petroleum Production - 100%
Refinery Operations 23% 77%
Underground Storage 15% 85%
Auto Filling Tanks 27% 73%
Fuel Combustion 38% 30% 32%
Waste Burning Fires 22% 40% 38%
Heat Treated Coatings 20% 80%
Air Dried Coatings 14.4% 85% 0.6%
Petroleum Based Dry Cleaning 2% 98%
Dry Cleaning - syn ' - 100%
Degreasing TCE 100%
Degreasing III-T - 100%
Rotogravure Printing 7% 93%
Flexigraphic Printing 21% 78% 1%
Rubber, Plastic, etc. Mfg 36% 51.5% 12.5%
Pharmaceuticals 27% 70% 3%
Misc. Solvents 17% 80% 3%
Gasoline Exhaust 22% 71% 7%
Gasoline Evap (Mobile) 24% 76%
Diesel Exhaust 18% 60% 22%
Gas Turbine (Jet) 33% 53% 14%
Piston Aircraft 26% 66% 8%
-48-
-------
For any given area, the above data usually can be obtained from a number
of sources:
o local air pollution control district;
o state air quality agency;
o Federal monitoring programs;
o universities and colleges;
o private industry;
o private research organizations.
The maximum amount of ambient air quality data that are compatible with
model output should be obtained from each source.
Once the initial data-gathering exercise has been completed, the data
should be screened prior to use in evaluating model performance.
The purpose of this screening is to ensure that:
o pollutants measured by different networks (e.g., local, state,
Federal, private, etc.) were measured with comparable
analytical methods;
o each station has enough data to obtain a statistically
representative sample of the true pollutant levels in the air.
The screened data quite possibly may need to be adjusted before they can
be directly compared to model output. For example, many monitoring
systems measure total oxidant (of which ozone comprises a large
fraction), while many models predict only ozone. The total oxidant air
quality data would thus have to be corrected to represent only ozone.
Furthermore, monitoring systems usually measure total hydrocarbons (THC)
and non-methane hydrocarbons (NMHC), whereas photochemical models
predict hydrocarbon levels by reactivity class. Some correction factor
would have to be applied to the ambient hydrocarbon data to allow
comparison with the model results. Even when prepared with the utmost
care, such comparisons are difficult due to the high uncertainty in the
monitoring data for NMHC.
The air quality data should next be examined to determine if they are
adequate to verify the model. Specifically, the following issues should
be considered:
o is the spatial and temporal resolution of the ambient data
adequate to characterize initial conditions and to simulate
the diurnal cycle in oxidant production?
o are data available to determine vertical concentration
profiles of pollutants, and to determine boundary inflow
through the top of the mixed layer?
-49-
-------
o are there adequate data collected at stations outside the
model domain to characterize boundary inflow?
If the existing data are inadequate to evaluate model performance and/or
specify model inputs, then field studies should be employed to gather
the needed air quality data. These field studies should also measure
meteorological parameters concurrently with the ambient air pollutant
concentrations.
The following paragraphs discuss two examples of collecting data to be
used in evaluating the performance of photochemical models. The first
example concerns the data collection effort used to evaluate the
performance of LIRAQ in the San Francisco Bay Area, and the second
example illustrates the monitoring that was undertaken to evaluate the
performance of the Denver Air Quality Model (early version of the SAI
Model) for Denver.
The data gathering effort for evaluating the performance of LIRAQ was
aimed at collecting ambient 03, HC, NO and N02 monitoring data from the
BAAQMD monitoring network, a NASA aircraft monitoring program, and
Barringer spectrometry studies. (These are described in greater detail
later in this section.) These data were screened to ensure that they
represented measurements taken with the same (or similar) analysis
methods and that each station had enough data to formulate
representative statistics. All of these data represented point
measurements, and thus were not strictly comparable with the LIRAQ
output, which is in terms of average concentration in a 1- by 1-, 2- by
2-, or 5- by 5- kilometer box (MacCracken, et al., 1975); however, since
better data were not available, the point measurements were used to
evaluate the model. (The problem of comparing point measurements with
spatially-averaged model predictions is common to all photochemical
models currently in use.)
After screening, the ambient hydrocarbon and oxidant data had to be
adjusted to facilitate comparison with LIRAQ output. The ambient
hydrocarbon levels predicted by LIRAQ were broken down into three
reactivity classes, but the ambient levels measured by the monitoring
stations were reported as either total hydrocarbons or non-methane
hydrocarbons. The model output was modified slightly so that the
predicted ambient hydrocarbon levels were reported in terms of a
"typical" hydrocarbon type that could be compared with the ambient
measurements. The pre-1975 measured ambient oxidant levels were
artificially high due to inadequacies in the instrument calibration
procedures used at the time. All the affected oxidant data were
multiplied by 0.8 to correct the error (MacCracken, et al., 1975). The
ozone levels predicted by the model were compared with the adjusted
oxidant data.
-50-
-------
Even though the BAAQMD (and other groups) had compiled an extensive
ambient air quality data base, there was still insufficient data to
develop and evaluate the performance of LIRAQ; consequently, field
studies were undertaken to gather the needed data.* Specifically, NASA
aircraft and Barringer correlation spectrometer field studies were
performed to gather additional data to supplement the existing BAAQMD
monitoring network. The NASA aircraft data consisted of measured
ambient levels of CO, oxidant, 03, NO, N02, and S02, in addition to
concurrent measurements of several meteorological parameters. The data
provided vertical profiles of pollutant levels over both urban and rural
sites (MacCracken, et a!., 1975). Although these data were point
measurements (and hence not strictly comparable with LIRAQ output), they
did provide valuable information on the vertical profile of the
pollutants. This information was used to evaluate the performance of
LIRAQ in generating vertical pollutant profiles. The Barringer
Correlation Spectrometer Study was conducted in order to obtain
spatially-averaged ambient pollutant concentration data that could then
be compared with the spatially-averaged LIRAQ output. Using mobile
monitors, the total N02 level (ppm) in a straight line path was measured
in combination with fixed surface monitors measuring total sulfur and
total NOx (MacCracken, et al., 1975).
The Denver Study (Anderson, et al., 1977) conducted during 1975-1976
contained another example of supplemental monitoring that was undertaken
to obtain data for model evaluation. The spatial coverage of the
ambient air quality data collected by the existing monitoring network
was inadequate to evaluate the performance of the model.
The data used to evaluate the performance of the DAQM came from four
sources:
o the existing ambient air quality monitoring network operated
by the Air Pollution Control Division (APCD) of the Colorado
Department of Health;
o supplemental ambient air quality monitoring stations operated
by the Colorado Highway Department during 1975 and 1976;
o the meteorological monitoring network operated by the Colorado
APCD;
o supplemental meteorological monitoring stations operated by
the Colorado Highway Department, the National Weather Service,
private concerns, and others.
Each of these data sources will now be discussed.
*Much of the field work was originally undertaken to gather data for
model development; this data was later used to evaluate the performance
of LIRAQ.
-51-
-------
The existing Colorado APCD ambient air quality monitoring network
consisted of six stations measuring ambient HC, NO, N02, 03, S02 and CO
levels; only HC was monitored at all six stations during the time of the
study (1975-1976).
The supplemental ambient air quality monitoring network consisted of
four stations in 1975 and tjiree stations in 1976. Only 03 was monitored
at the supplemental stations. The supplemental network consisted of one
mobile and three fixed stations in 1975 and one mobile and two fixed
stations in 1976. In 1975, the supplemental monitoring was done on
predicted "bad" ozone days; approximately 20 days of data were
collected. In 1976, the monitoring system was run continuously for 90
days during the peak ozone season.
The meteorological monitoring was conducted at four of the six ambient
air quality stations. Wind speed and direction at 10 meters above the
ground were monitored at the four stations.
The supplemental meteorological monitoring was conducted by a variety of
sources. The exact number of contributing stations varied daily, with a
maximum of about 11 stations supplying data. No meteorological
variables were monitored at the supplemental ambient air quality
stations.
TREATMENT OF INITIAL AND BOUNDARY CONDITIONS
The treatment of initial and boundary conditions can play a major role
in determining the results of an air quality model simulation,
especially for ozone/oxidant. Primary pollutants are introduced to the
grid model domain* through three model elements of comparable
magnitudes: initial conditions, inflow through the boundary, and
emissions. The actual and relative magnitudes of these three elements
will be a function of detailed model inputs (e.g., boundary
concentration, emissions rate) of location within the model domain, and
of model structure. For example, Lagrangian models generally assume no
flow across horizontal boundaries, but are influenced by initial
conditions and entrainment at the top of the air column. Because
individual trajectories are often run for only a few hours, the relative
importance of initial conditions usually be greater for a Lagrangian
model than for an Eulerian model.
Initial conditions completely determine the pollutant content of the
model at the initial time, and dominate the loading of pollutants for
some time thereafter. Figure 3-3 illustrates a single element on an
upwind boundary of an Eulerian grid model. The initial mass loading of
pollutant in a given model element (assuming uniform concentration over
the cell) will be given by the product of the initial mass concentration
in the element and the initial volume of the element.
*The volume of space to which a model is being applied, usually defined
horizontally by the extent of the grid, and vertically by the ground
and the height of the mixed layer.
-52-
-------
C 0)
at x;
fl) 4->
O 4J U-l
O O
CJ *O
£ (1)
•H en E
4J tn -H
3
»-i o
o a) •—i
en 4J o
3 -H .Q 4->
o Q) «J c
V4 Q)
4J
fl)
0)
VH j^!
•H 4J
4J 10
C
-H
it) h
X! «J
-d
C
3
V4 0)
U4>
M O
-H O
10
u-i O
O 1-4
M-4
0-
,
jj i—i
c
4> *O
01
4J X
O -H
0)
.C
4J
3
Q.
O)
CO
s_
CD
Q
O
•a
O)
_
o
•H
4J
2 <*6
4-1 O
C -f
0) <
o
o
e o 0
o x;
0) 4J
H E
•^ -H C
. 0) 4J -H
'° CJ C
•O XI O
•H 4-> -H
V4 4J
0** -A^ (u
(0 N
•H M (0 •
4J K) ->H (T
fS -O 4J C
•U C -H -H
C 3 G G
,<
ai
s-
-53-
-------
Inflow across the model boundaries will provide a pollutant source given
by the product of the volume influx times the concentration at the
boundary integrated over the time period simulated. Inflow often
dominates the pollutant mass input near the inflow boundaries.
Emissions provide a source of pollutants given by the integral of the
emissions rate over the time period simulated. Emissions will usually
govern mass input near strong sources, but may not be the primary source
of pollutant a few kilometers away from major sources.
The relative importance of these three terms will vary greatly as a
function of time and location. Emissions will usually dominate the mass
inputs of primary pollutants in central cities; however, this will often
not be the case for the modeling grid if it is structured to consider
ranges comparable to those involved in the formation and transport of
photochemical oxidants.
In a representative 0400 hours to 1800 hours LIRAQ simulation of Bay
Area air quality, emissions accounted for about 3/5, initial conditions
accounted for about 1/10, and boundary influx accounted for about 3/10
of the total input of hydrocarbons and NOx to the model domain. For CO,
emissions accounted for about 1/5, initialization for 1/10, and boundary
flux for 7/10 of the total mass input. Although these numbers would
change if the meteorological scenario, grid domain, model structure or
assumed boundary and initial concentrations had been changed, they are
representative of a smoggy San Francisco Bay Area day. Other areas and
other models might receive relatively more mass input from initial
conditions and less from boundary flow, since early morning mixing
depths are often very shallow in the San Francisco Bay Area, and a
strong sea breeze is often observed even on smoggy days, giving rise to
a significant boundary flow.
Just as initial conditions completely determine concentrations at the
beginning of a model simulation run, boundary fluxes will often
determine the mass loading near the upwind boundaries of Eulerian
models. This happens because the upwind boundaries are often in regions
of low to moderate emissions, and even a modest inflow velocity can
carry a large volume of air through a grid cell boundary.
It might seem that this discussion overstates the importance of a few
ppb of pollutants introduced as background values on initialization or
through boundary fluxes. Indeed, relative to ground level measurements
of primary pollutants in central urban areas, background levels are
usually unimportant. There are exceptions; for example, for more than
half of a 100 km x 100 km grid centered on the city of St. Louis in a
recent LIRAQ calculation (at 1500 hours on a high oxidant day), mean
layer CO was within 15% of the background value used (100 ppb), the mean
layer hydrocarbons were less than 3 times the assumed background of 30
ppb, and mean layer NOx was less than twice the asssumed background of 4
ppb. The case is qualitatively similar in the San Francisco Bay Area,
especially in suburban high oxidant areas. Although concentrations of
hydrocarbon and NOx are not particularly high in many suburban
locations, the influence of moderately low concentrations of
-54-
-------
hydrocarbons and NOx on ozone is still significant, at least in the
LIRAQ model. This was also established in sensitivity studies carried
out with the 1975 version of the LIRAQ model and described in the LIRAQ
documentation (Duewer et al., 1978). These studies found that the
difference between using very small boundary concentrations and boundary
conditions slightly larger than those now employed was a factor of
nearly 1.6 in the average ozone content at measuring stations (a larger
factor if averaged over the entire domain).
Methods for Treating Initial Conditions
Base Year
Ideally, in treating base year scenarios the initial conditions are
based on measured ambient concentrations with the measurement stations
giving a fairly uniform coverage of the entire model domain, and
providing extensive data aloft. In practice, base-year initial
conditions are usually based on relatively sparse measurements taken
almost exclusively at the surface and primarily near urban centers. As
a result, concentrations aloft and away from urban centers must be
estimated from the limited available direct measurements, from
conjectures about concentrations at the domain boundary and aloft, and
from some interpolation/extrapolation scheme. This is largely a
consequence of the great difficulty and expense involved in accurately
measuring pollutants at concentrations one to two orders of magnitude
below standards, or of measuring anything aloft. An adequate sampling
program would be expected to cost hundreds of thousands of dollars if
run for even a few days.
In LIRAQ applications, a vertical profile based on emission rate,
deposition velocity, mixed layer depth, and wind speed is combined with
surface observations to yield mean layer concentrations above
measurement stations. Concentrations at boundaries are estimated based
on the larger of global background or lowest measured value. If data
had been available that indicated higher concentrations, higher
concentrations would have been adopted, but the available data were of
the character of upper limits. Currently in LIRAQ applications, a
Gaussian weighting scheme is used to interpolate initial concentrations
over the grid; the interpolation uses measured values and boundary
conditions at the horizontal boundaries. Finally, concentrations in
grid cells that are above the inversion are set to the values estimated
at the upper boundary.
In the case of the SAI Urban Airshed Model a similar procedure has been
used except that it is a multi-layer model. The mixed layer is assumed
to be well mixed, the concentrations above the mixed layer have been set
to an assumed background value (or available measurements aloft, if any,
have been used), and the concentrations assumed at the horizontal
boundaries have not been used. In order to avoid extrapolation of urban
center concentrations into rural areas, synthetic stations are
occasionally inserted into outlying areas and assigned background
concentrations. In areas (such as Los Angeles) where measurements show
high concentrations of pollutants aloft or at the boundaries, high
-55-
-------
concentrations have been used in modeling efforts.
In the case of LIRAQ applications, initial concentrations usually
provide on the order of 10% of the total input of most primary
pollutants for a typical 14-18 hour run. For some trajectories in
trajectory model calculations, initial conditions will provide nearly
all of the input of primary pollutants. This will be particularly true
if the trajectories are run for short periods. In application of a
multi-layer model, if the maximum height of the model were to be
maintained at a constant value, and the mixing height varied within
that, initialization would occur over a much deeper layer, and initial
conditions would usually provide a larger fraction of the total mass
input to the model. Of course, boundary flow would be correspondingly
reduced in importance.
Since the photochemical models use multiple classes of reactive
hydrocarbons, measured or estimated nonmethane hydrocarbon
concentrations (NMHC) must be converted into alkene-like, alkane-like,
aldehyde-like, etc. hydrocarbon concentrations. The choices made here
rarely have any directly appropriate data to guide them and can have
factor-of-ten effects on the effective reactivity of the hydrocarbons
entering the model through initial conditions. The factors needed to
assign NMHC to model classes should be strongly affected by the local
emissions inventory, the significance of biogenic emissions (which
should provide a few ppb of very high reactivity materials) and long
range transport (which should provide primarily low reactivity
materials). The breakdown of NMHC into model hydrocarbon classes can be
expected to differ from model to model, from region to region, and even
from day to day for a given region. The ideal resolution of this
problem would require a comprehensive and detailed analysis of the
actual organic species contained in ambient air samples in the region of
interest, preferably on the days to be modeled. However, a significant
improvement in the available data is definitely feasible for a cost
likely to be of the order of ten thousand to fifty thousand dollars.
Future Year Simulations
For future year simulations, no measured initial conditions are
available, and initial concentrations must be forecast from emissions
changes, baseline observations, and known background concentrations. In
cases where the measured base year initial concentrations are
substantially above background concentrations, a reasonable hypothesis
is that the measured concentrations reflect controllable local
emissions. However, if the base-year initial concentrations approximate
global background concentrations, future year initial concentrations
will not likely be much different from base-year initial concentrations
(unless emissions are to be dramatically increased).
Because initial concentrations measured in the San Francisco Bay Area
have generally been well above global background, in LIRAQ applications
the initial concentrations have been scaled with emissions. This has
been done in one of two ways. In some cases, the initial concentration
field developed by the model was multiplied by the ratio of future
-56-
-------
emissions averaged over the grid and day to emissions on the prototype
day similarly averaged. In what may be a slightly superior alternate
methodology the station observations have been scaled by the same ratio,
the initial boundary concentrations left unchanged, and a new
concentration field developed. In the first method, all initial
conditions are scaled with emissions, even when they may be dominated by
global background concentrations. The virtue of the second method is
that initial conditions are not scaled when they are in regions where
the initial conditions are determined by the assumed background values.
In the application of the SAI model to Denver the initial concentrations
were near background levels, and no scaling was used.
For both of these examples, the estimation of future initial
concentrations of precursors may be expressed by the following formula:
C(future) = C(bkgd) + E(future) x [C(prototype day)-C(bkgd)]
E(prototype day)
If background concentrations are small relative to the observed initial
concentrations on the prototype day being modeled, then future initial
concentrations become scaled by the ratio of emissions. If initial
concentrations are close to background levels on the prototype day, then
little or no changes should occur for future year simulations. The
situation becomes more complicated if the background contribution to the
observed initial concentrations includes pollutants transported from a
nearby area not included in the modeling grid.
One possible method of reducing the problem of specifying initial
conditions would be to perform extended (i.e., multiday) simulations.
Before this approach is adopted, one must establish that the model
adopted can satisfactorily represent such a multiday period, and that
the available data permits a satisfactory representation of flow
patterns aloft as well as at the ground.
Boundary Conditions
Most models use the concentration computed just inside the outflow
boundary as an outflow boundary condition, and outflow boundary
conditions present no difficulty. The problem is in the treatment of
concentrations aloft (when the inversion base is rising) and at the
upwind boundaries. In the ideal case, observational data are available
to provide information about the concentrations at the model boundaries.
However, in practice, very few useful data are ever available. In part,
this is a result of the difficulty involved in making measurements
aloft, and the tendency for most air quality measurements to be made at
locations where standards are expected to be violated. However, a more
pervasive problem is that most instruments designed for air quality
measurements are either improperly calibrated or insufficiently precise
when concentrations are at the ppb level. Carbon monoxide provides a
simple example of both faults. In many air quality monitoring programs,
-57-
-------
carbon monoxide concentrations are rounded to the nearest 1000 ppb.
When carbon monoxide has been measured using systems carefully designed
to operate at levels below 1 ppm, the lowest concentrations measured
over the north Atlantic or north Pacific have been about 100-150 ppb.
Such concentrations would contribute about 1/2 the total mass loading to
a typical model simulation over a 10,000 km2 grid in an area like San
Francisco. However, when the RAPS program in St. Louis adapted CO
monitors for operation at sub-ppm levels, they frequently reported
concentrations of 50 ppb, a default value recorded whenever the
measurements were 50 ppb or less. This is almost certainly a reflection
of improper calibration (R. A. Rasmussen, private communication). The
situations for reactive hydrocarbons, NOx, and S02 are generally worse
than for CO since the concentrations are generally much lower, and in
the case of hydrocarbons, corrections must be made for 1,300-1,700 ppb
of methane.
In practice, boundary conditions have occasionally been defined based on
air quality measurements taken near the boundaries of the modeling
region. When this was done in the early 1970's, the boundary flows
often almost completely dominated the problem. In more recent
applications much lower concentrations have generally been used.
However, even at concentrations representative of the marine boundary
layer or the upper troposphere, CO boundary fluxes would usually be
comparable to model emissions over a 100 km x 100 km grid. The
situation with respect to non-methane hydrocarbons and S02 is less clear
because true background concentrations of non-methane hydrocarbons and
S02 are not well established. In the case of NOx there is evidence that
in the marine boundary layer, and presumably in the upper troposphere,
NOx concentrations are in the range of 0.01 to 0.15 ppb (Noxon, 1978;
McFarland et al., 1978). At these levels NOx boundary flow is
negligible relative to emissions. However, there is ample room to doubt
whether concentrations that low would be seen in continental air over
the United States although very low NOx concentrations might be
appropriate as western boundary fluxes in the San Francisco Bay Area.
This discussion does not imply that extremely low boundary
concentrations are necessarily correct. Either reci rcul at ion or
long-range transport might lead to high concentrations of anthropogenic
materials at model boundaries, and, at least occasionally, natural
events may also be important (e.g., volcanoes can be major sources of
S02, forest fires major sources of particulate and organic materials,
normal biogenic and geogenic organic emissions may also be significant
in some areas). The point is that the difference between two sets of
fairly low concentrations at the boundary can potentially have a
significant effect on model predictions of secondary pollutants at
downwind sites.
Examples of the Treatment of Boundary Conditions in Baseline
Calculations and Model Evaluation Studies
In LIRAQ applications, the inflow boundary conditions have been treated
with an expression of the form
-58-
-------
where Cb is the inflow boundary concentration, Co is a minimum
background concentration, p is a parameter between zero and one, and Ci
is the concentration just inside the boundary. Co and p are specified
for each of five boundaries (E, W, N, S, Top). The rationale for this
expression is that if Ci is low, the adjacent areas just beyond the
boundary would usually also have low pollutant concentrations reflecting
background levels; however, if Ci is large with respect to background,
the boundary region likely either has significant emissions, or has been
influenced by air advected from an emissions-dominated area. Also, in
the case of secondary pollutants, air just outside the boundary should
have experienced similar photochemical conditions to that just inside.
In either case, pollutant concentrations in air from just outside the
domain boundary would more likely be above background if concentrations
just inside the boundary are elevated. The precise form was chosen to
give a continously differentiable function for convenience in the
mathematical treatment of the model. In the earlier LIRAQ developmental
efforts p values of 0.5 to 0.95 were usually used (Duewer, et al.,
1978). This suffers from the disadvantage that a p of 0.95 causes the
concentration in a region experiencing no sources except rapid advection
from an upwind boundary to eventually reach 3.2xCo. In contrast, a p of
0.5 would only cause the concentration to reach 1.15 Co. In current
applications of the LIRAQ model all p's are recommended to be set at 0.3
except p for ozone aloft, which has been set at 0.85. This reflects the
observed behavior of ozone in the San Francisco Bay Area, where ozone
often displays its maximum concentrations aloft.
The Co values must reflect local conditions. In the San Francisco Bay
Area carbon monoxide observations are seldom reported as less than 1.0
ppm, even at rural stations, although some suburban stations report 1.0
ppm most of the time on the days that have been simulated by LIRAQ. For
this reason Co was set to 1.0 ppm (1000 ppb) for carbon monoxide,
although a Co of 100-200 ppb would be easier to justify. Hydrocarbons
are divided into three classes for LIRAQ: HC1, a high reactivity class
similar to alkenes; HC2, a low reactivity class similar to alkanes; and
HC4, a photoreactive aldehyde-like class. Representative Co values used
in LIRAQ applications would be 10 ppb for HC1, 50 ppb for HC2 and 1 ppb
for HC4. Go's for NO and N02 have often been set at 4 ppb, S02 at 10
ppb, and ozone at 10 ppb in the boundary layer, 25 ppb aloft. Species
other than those listed have been assigned very low Go's of 0.004 ppb.
The SAI model has a substantial variety of possible methods for
generating boundary conditions. In most applications, boundary
conditions have either been based on measurements or have reflected
"background" values. When the boundary conditions are based on
measurement they have reflected the spatial and temporal variability of
the measurements; when assigned background values, they are constant in
space and time. Background values recently recommended by SAI (Reid and
Reynolds, 1979) are NO = 1.0 ppb, N02 = 2.0 ppb, 03 = 40 ppb, HN02 = 0.1
-59-
-------
ppb, H202 = 0.01 ppb, S02 = 10 ppb, CO = 100 ppb, olefins = 0.4 ppb,
paraffins = 35 ppb, aldehydes = 5.0 ppb, and aromatics = 6 ppb.
Phenomena such as long-range transport and recirculation of natural
emissions may lead to elevated concentrations aloft and at boundaries;
thus there is no general upper limit to the concentrations that should
be used where there is no measured data even when emissions are low in
the boundary region. However, measured concentrations in remote
tropospheric areas may provide a lower bound for the boundary
concentrations. For populous rural areas (e.g., the northeastern and
midwestern U.S.), these lower bounds will likely be too low. Near more
isolated areas such as San Francisco or Denver, they may be approached.
Neither the LIRAQ Go's nor the SAI model background values reflect
probable remote tropospheric concentrations (e.g., clean marine air or
upper troposphere). If remote tropospheric values were to be estimated
from the best available data, CO concentrations would be 100-200 ppb
(Seiler, 1975), NO concentrations would be 0.004 - 0,020 ppb (McFarland
et al., 1978; Drummond 1979), and N02 concentrations would be
0.010-0.100 ppb (Drummond, 1979, Noxon, 1978). Remote tropospheric S02
concentrations are likely to be below 0.4 ppb (Georgii, 1978), while
ozone concentrations of 20-60 ppb seem representative of clean air
(Rasmussen, 1976) (ozone and S02 concentrations probably increase with
altitude because of deposition on surfaces). Background concentrations
of hydrocarbons are, as yet, not reliably known. Concentrations of
6-1600 ppb of non-methane and non-aldehydic C2-C6 hydrocarbons have been
reported (Robinson et al., 1973). Recent estimates of alkenes are of
the order of 1-5 ppb, alkane hydrocarbons 5-100 ppb. While no direct
estimates of aldehydes are available, model calculations (of the free
troposphere) suggest 0.1-1 ppb of CH20 from methane oxidation, and total
aldehydes of 0.5-5 ppb. H202 seems likely to be present at 0.1 to 1
ppb, while HONO is unlikely to be above 0.03 ppb (Wuebbles et al.,
unpublished work, 1979).
One method of estimating boundary concentrations might be to run "nested
models"; i.e., to first run a larger scale model and use the predictions
to generate boundary conditions for a smaller scale model. While
conceptually attractive this concept has received only rather limited
testing thus far.
Future Year Simulations
If boundary conditions reflect natural background concentrations or low
level emissions from areas unlikely to experience significant growth or
emissions reductions, there is no reason to adjust boundary conditions
in future year applications, and indeed, boundary conditions have
usually not been adjusted in model applications. (In the case of LIRAQ
applications the portion of the boundary concentration that reflects the
concentration just within the boundary is automatically adjusted by
variations in that concentration as they involve emissions changes).
-60-
-------
However, if emissions from another area, or from developing rural areas,
play a significant role in determining either baseline or future
boundary concentrations, then a satisfactory treatment of boundary
conditions should reflect projected emissions in nearby areas (nearby
might be as far away as several hundred km for some pollutants, e.g.,
CO, 03 and S02). Unfortunately because of the large amount of effort
involved, a detailed treatment will not likely be attempted except under
extraordinary circumstances.
Summary
In summary, initial and boundary concentrations can have an influence on
model-computed results for suburban and downwind areas that approaches
the importance of emissions, and can have a significant influence on
calculations of secondary pollutants even in central urban areas.
Moreover, this is true even when the initial and boundary concentrations
are of the order of a few to one hundred ppb. On the other hand
boundary and initial concentrations will rarely if ever cause or make a
major contribution to violations of standards for primary pollutants.
The problem with respect to concentrations approaching the standards is
largely limited to secondary pollutants.
In the absence of a special field data collection study, there are
almost no data available that are really useful in guiding the choice of
boundary conditions or of conditions aloft. Data obtained with normal
air quality monitoring instruments are so imprecise that they are often
difficult to interpret.
Although most current model applications use lower background values
than were used even two or three years ago, currently used
concentrations are not low enough to be a negligible factor in model
calculations. Moreover, except for carbon monoxide, there is little
reason to believe that even the lower concentrations used in model
applications are as low as those representative of a remote tropospheric
air mass. Conversely, there is also little reason to believe that the
air 10-100 km from a representative urban center should be very close to
remote tropospheric air in its composition. Thus, the specification of
initial and boundary conditions remains a major problem in the
verification and application of air quality models, and all proposed
solutions to the problem are of an essentially ad hoc nature.
The problem of choosing initial and boundary conditions should be
considered when the choices of model domain and meteorological
conditions are made. If a city is at the western edge of a populous
region, it is likely that boundary concentrations will be lower and more
easily estimated so that calculations will be more reliable if the winds
are from the west. If a city is embedded in a megalopolis, a
large-scale model is likely to be required at least for species like
ozone that require several hours for formation from their precursors.
-61-
-------
SPECIAL I:IELD MEASUREMENT STUDIES
The purpose of a special field study is to augment the existing
emissions and aerometric data gathering efforts toward the end of
providing an adequate data base for preparing model inputs, evaluating
performance, and examining alternative control strategies. Because
provision of model input and evaluation data is not the normal objective
of routine emissions and aerometric monitoring, one invariably
encounters a data base that is sparse in one or many components when
applying a model to an urban area. As might be expected, there is wide
variation in the number and type of measurements made in urban areas in
the United States. All cities have at least a few surface wind
stations, for example, but the amount of data on upper level winds and
atmospheric stability varies greatly. The number of air quality
monitors likewise varies and, to a degree, reflects the predominant air
quality concern in each city. Oxidant monitoring in St. Louis and Los
Angeles is extensive, whereas in Las Vegas, for example, concern seems
to focus on carbon monoxide. Routine measurements are not made of
concentrations of hydrocarbons by species or pollutant concentrations
aloft, though these measurements are made during some special field
studies. Finally, emissions inventories for various cities range from
region-wide estimates of nominal emissions rates for various pollutants
(in the NEDS format) to detailed, gridded inventories that disaggregate
emissions according to individual sources or source categories.
A field observation program can be expensive and care must be taken in
its planning and coordination to ensure that the desired information is
obtained. The costs and benefits of long-range vs short range
scheduling are an important consideration. For example, suppose one
assigns a very high priority to obtaining supplemental data on the worst
air pollution day of the year. This could be guaranteed by taking
measurements every day during the smog season - but at very high cost.
Or, one might schedule field programs long in advance for several days
during the peak smog season in the hope that the worst day happens to
fall on one of those days. A third alternative might be designed to be
implemented on very short notice - perhaps no more than a day or two.
Short notice would be necessary because adverse conditions cannot be
forecast reliably more than a day or two in advance. This approach,
while having a fairly high probability of success, also has inherent
drawbacks. First, because the worst day cannot be known in advance, the
field program will have to be implemented on several of the most adverse
days. Second, the requirement for deployment of equipment and personnel
on short notice will preclude the use of some observation systems that
require long lead times for set-up. Personnel costs per unit time will
also tend to be higher.
The following sections discuss some of the possible components of a
special field measurement study; previous studies carried out in the San
Francisco Bay Area and Los Angeles are also summarized.
-62-
-------
Meteorological Data
Supplementary meteorological data gathering efforts are commonly
required to upgrade the data base available from routine monitoring
activities. As previously indicated, measurements of upper air
conditions frequently are sparse and do not provide enough information
for satisfactory interpolation or for use in windfield or mixing depth
models. Moreover, in situations where special conditions prevail, such
as topographical influences on air flows, routine measurement networks
of surface conditions can prove to be inadequate to supply necessary
detail in the input data.
Wind data (speed and direction) are normally available from surface
measurement networks. However, additional observations may be needed to
adequately characterize wind flows in outlying or suburban areas or in
regions where significant terrain features may substantially alter flow
patterns. The principal supplementary activity, as pointed out above,
will be the collection of vertical wind soundings to obtain upper air
wind data. Such measurements can be effected by releasing and tracking
pilot balloons, or pibals, at various locations and times of the day.
An alternative method that can be used for measurements up to a few
hundred feet is to mount instruments on fixed towers.
Radiosondes are instrumented balloons that are released and tracked to
give information on the temperature structure of the atmosp'here,
pressure gradients, relative humidity as a function of elevation, and
wind speeds and direction. However, because they are considerably more
expensive to deploy than pibals, their use is primarily in connection
with the first three variables mentioned and only incidentally for wind
measurements. Aircraft are also used to measure vertical temperature
structure and pressures; these data are used to infer mixing depths
throughout a region. An aircraft has the advantage that it can be flown
according to a fixed pattern, such as a vertical spiral, and thus make a
measurement at a predetermined location. Another method for measuring
the depth of the inversion layer is by means of acoustic soundings,
which are made from ground-based stations. Finally, tetroon releases,
when tracked from the ground or from aircraft, can be used to follow air
parcel trajectories. They have the advantage over pibals that, because
they follow isentropic surfaces, they can be tracked for longer
distances.
Air Quality Data
Most locations will have an existing network for collecting air quality
data; a routinely collected data base, however, may fail to satisfy the
needs of a particular study. The data may be deficient in terms of
their spatial or temporal coverage, lack of upper air measurements, or
lack of detailed information on some pollutants of interest.
Should the existing network be judged too sparse, it may be supplemented
by additional fixed stations or by mobile stations. Mobile stations are
probably more cost-effective in a supplementary data gathering effort
because they are much more adaptable and can be converted to other tasks
-63-
-------
after the study has been completed. Sample hydrocarbon analyses can be
carried out at a central location if bag samples are collected and
transported to the analysis facility.
Routine networks will often fall short of model requirements with
respect to the collection of boundary condition data. These data must
be collected in outlying areas and aloft. Boundary condition
requirements vary according to the region being modeled and the
conditions on the day being modeled. If large pollutant fluxes are
present on the upwind boundary, boundary conditions will be important;
if air is coming from an area of minimal concentrations (e.g., the
Pacific Ocean in the case of San Francisco), boundary conditions may not
be needed in much detail. Initial conditions also are important and may
require supplementary data collection. Accurate specification of
initial conditions can be the key to a satisfactory simulation.
Routine monitoring networks frequently do not provide concentrations of
particular pollutants of interest. The analyses that are almost
invariably lacking are the concentrations of individual hydrocarbon
species, knowledge of which is necessary for a photochemical model. The
different hydrocarbon species undergo different reactions in the kinetic
mechanism, and the course of the overall chemical transformation is
highly dependent on the hydrocarbon distribution. Thus, it frequently
is necessary to upgrade the analytical capabilities of existing
monitoring stations during special field studies.
In summary, it is almost invariably necessary to supplement routine data
collection by conducting a special study. The actual extent of the
additional activities must be determined after comparing the existing
data base with the data needs of the model. Costs of this supplementary
data collection are difficult to estimate in general. Miedema et al.,
(1973) surveyed monitoring costs in the early 1970s, and their report
gives capital and operating costs for many instruments and types of
measurements. However, it must be recognized that their estimates have
been overtaken by inflationary cost increases; present-day costs will be
substantially higher. An updated monitoring cost survey is in
preparation (Lutz, 1979), and a report may be forthcoming in 1979 giving
more current figures.
Even in the most widely monitored areas the routinely collected data are
usually insufficient to characterize the basic model inputs with the
desired spatial and temporal resolutions. Determining the spatial
variation of the mixing depths and pollutant concentrations aloft will
almost always require a special field study. The quantity of data
collected and the duration of the field study should depend on both the
available resources and the complexity of the modeling region. The
following sections describe two field studies that were undertaken to
supply data for airshed modeling.
Collecting Supplementary Field Data for LIRAQ
Data bases of meteorological information and pollutant concentrations
adequate for applying the LIRAQ model to San Francisco were not
-64-
-------
available, even though the BAAQMD had, for a number of years, operated
one of the most extensive networks of meteorological and air quality
monitoring stations in the nation. Even with the addition of data from
other existing sources in the Bay Area, it appeared that the resulting
data base would be inadequate. Accordingly, a field data collection and
processing effort was carried out. In addition to data from the
existing surface measuring stations of the BAAQMD and other agencies,
data were collected by temporary surface stations deployed during
selected field intervals and by an aircraft operated by the NASA Ames
Research Center (ARC) over predetermined flight paths. The field
program also included a fluorescent tracer particle study and a
correlation spectrometer study, each designed to test some of the
assumptions on which LIRAQ was based.
Field Observations at Temporary Sites. To augment a single routine
meteorological rawinsonde site at Oakland International Airport, a group
of sites was selected for radiosonde and pibal observations of vertical
profiles of wind, temperature, and humidity, using portable equipment.
The California State University at San Jose, under subcontract to the
NASA-Ames Research Center, was responsible for setting up and operating
these portable sites. The University agreed to provide four teams for
the primary purpose of taking vertical soundings and making wind
observations. Generally, one team operated the rawinsonde equipment at
the University, and other teams performed mobile activities in the
field. One of the teams operated a mobile radiosonde van on loan from
NOAA/National Weather Service and doubled as a pibal team. The
remaining teams generally took observations and all of the teams took
surface meteorological observations. In addition to these teams, three
portable wind observation stations for specialized observations and the
services of one or two mobile vans were available from the BAAQMD.
Observations by Instrumented Aircraft. The bulk of vertically resolved
air quality and meteorological data needed for model development was
provided by an instrumented twin-engine Cessna aircraft flown by the
ARC. Vertical soundings were taken by the aircraft within the polluted
surface layer at specified sites and along climbing and descending
flight paths between the sounding sites. Parameters sampled included
wind speed and direction, temperature, dew point, ozone, total oxidant,
oxides of nitrogen, carbon monoxide, sulfur dioxide, and hydrocarbons.
Sampling was instantaneous at 20 second intervals.
Barringer COSPEC Studi es. During the first year of the project, in
conjunction with the other field observations, a program of special
studies involving the use of the Barringer correlation spectrometer
(COSPEC) was carried out. The program involved obtaining measurements
of the integrated burden of nitrogen dioxide over a straight-line path
in combination with fixed-point surface measurements of total oxides of
nitrogen dioxide and total sulfur.
The purpose of these studies was to provide spatially-averaged
concentration data for use in testing the spatially-averaged assumptions
embedded in the LIRAQ model.
-65-
-------
Flourescent Tracer Studies. In the second project year, a set of
specialized fluorescent tracer experiments was designed and carried out
to examine the questions of transport through the interface between the
marine air and the overlying inversion layer. That information was
necessary to assess the assumptions made in the submodel for determining
a mass-consistent wind field.
In conjunction with each of the experiments (which were not conducted on
normal field experiment days), temperature soundings were taken on a
regular basis by volunteer aircraft of the Oceanic Society in order to
define the structure of the inversion layer.
No tracer was found to penetrate the lower interface of the inversion
layer during either of the two experiments. These results neither
confirm nor reject the hypothesis of transport through the inversion
interface, but they do reinforce the widely held assumption that such
penetration is unlikely, especially when the inversion height is not
changing. Meteorological and time constraints unfortunately prevented
experiments under inversion destruction conditions such as breaking
waves and thermal ablation of the inversion layer. The original design
concept had incorporated experiments of that type.
CALTRANS Field Study in Los Angeles
The California Department of Transportation (CALTRANS) has evaluated the
SAI model using data collected on 26 June 1974. The data available for
this day were determined to be insufficient to fully describe the
three-dimensional wind field and the distribution of pollutants aloft,
so a study was undertaken to collect additional data during the 1975
smog season to use in characterizing the aloft wind and pollutant
patterns. This study was a multiagency cooperative effort involving
CALTRANS, the CARB, air pollution control districts in five counties,
NOAA, the EPA-Research Triangle Park, the National Environmental
Research Center-Las Vegas, the U.S. Navy, and the University of
California at Los Angeles.
The sampling program consisted of the following elements:
o Instrumented aircraft flights.
o Pibal launches (for winds aloft) throughout the basin.
o A network of mechanical weather stations sited to obtain an
adequate representation of the surface wind field for the
region.
o Solar radiation measurements both above and below the
inversion layer and covering the modeling region.
o Air quality data from all the measuring sites in the
regions.
To reduce expenses, sampling was performed only when the San Bernardino
County APCD predicted an episode that would exceed 690 ug/m3 (35 pphm)
oxidant. Although the APCD predictions were not always accurate, this
procedure did result in some very satisfactory data sets. The total
cost of obtaining these data, including editing and entering them into a
computer, was about $150,000. No model evaluation runs have yet been
done using these data.
-66-
-------
REFERENCES
Andersen, G.E., S. R. Hayes, M. J. Hillyer. J. P. Killus, and P. V.
Mundkur, "Air Quality in the Denver Metropolitan Region 1974-2000."
Preared for the U.S. Environmental Protection Agency, Region VIII by
Systems Applications, Incorporated, under contract number 68-01-4341, May
1977.
Booz-Allen and Hamilton, "Regional Emission Projection System," prepared
for U. S. Environmental Protection Agency, EPA-450/3-74-051, 1974.
Dickerson, M. H. (1978), "MASCON - A Mass Consistent Atmospheric Flux
Model for Regions with Complex Topography," J. Appl. Meteor., Vol. 17,
No. 3, pp. 241-253.
Drummond, J. W., private communication, 1979.
Duewer, W. H., M. C. MacCracken, and J. J. Walton, "The Livermore
Regional Air Quality Model: II. Verification and Sample Application in
the San Francisco Bay Area," J. Appl. Meteor., 1_7, 273-311, 1978.
Georgii, H. W., "Large Scale Spatial and Temporal Distribution of Sulphur
Compounds," Atmos. Environ., 12_, 681-690, 1978.
Leong, E. Y. and R. Y. Wada, "Emission Inventory Projections: Hindsight,
Insight and Foresight," proceedings of the Air Pollution Control
Association Specialty Conference on Emission Factors and Inventories,
Anaheim, California, November 13-16, 1978.
Lutz, D. (1979), Environmental Protection Agency—personal communication
to Systems Applications, Incorporated.
MacCracken, M. C. (1975), "User's Guide to the LIRAQ Model: An Air
Pollution Model for the San Francisco Bay Area," UCRL-51983, Lawrence
Livermore Laboratory, Livermore, California.
MacCracken, M. C., and G. D. Sauter, editors, "Development of an Air
Pollution Model for the San Francisco Bay Area, Volume I." Final report
prepared for the U, S. Energy Research and Development Administration by
Lawrence Livermore Laboratory under contract number W-7405-Eng-48,
October 1, 1975.
McFarland, K, D. Kley, W. C. Kuster, A. L. Schmeltekopf and J. W.
Drummond, "NO, 03, JN02, and CO Measurements Made in the Equatorial
Pacific Region," paper given at the 1978 Fall American Geophysical Union
Meeting.
Miedema, A. K., et al. (1973), "Cost of Monitoring Air Quality in the
United States," EPA-450/3-74-029, Environmental Protection Agency,
Research Triangle Park, North Carolina.
Noxon, J. F., "Tropospheric N02," J. Geophys. Res., 83, 3051-3057, 1978.
-67-
-------
Perardi, T. E., et al., "Preparation and Use of Spatially and Temporally
Resolved Emission Inventories in the San Francisco Bay Region, Journal of
the Air Pollution Control Association, Vol. 29, No. 4, pp. 358-364, April
1979.
Rasmussen, R. A., "Surface Ozone Observations in Rural and Remote Areas,"
J. Occupational Medicine, 18, 346-350, 1976.
Reid, L. E. and S. D. Reynolds, "The Conversion of LIRAQ Inputs to SAI
Airshed Model Inputs," SAI Report EF79-50, April 1979 (Draft).
Robinson, E., R. A. Rasumussen, H. H. Westberg, and M. A. Holdren,
"Non-urban Non-methane Low Molecular Weight Hydrocarbon Concentrations
Related to Air Mass Identification," J. Geophys. Res., 7£, 5345-5351,
1973.
Seiler, W., "The Cycle of Atmospheric CO," Tell us, 26, 116-135, 1974.
Tesche, T. W. (1978), "Evaluating Simple Oxidant Prediction Methods Using
Complex Photochemical Models (Monthly Technical Progress Narrative No.
1)," EM78-14, Systems Applications, Incorporated, San Rafael, California.
U. S. Department of Transportation, Federal Highway Administration
"Special Area Analysis - Part 4. Special Area Pollution," prepared by L.
R. Seiders, Comsis Corporation for FHWA Urban Planning Division,
Washington, D. C., August 1973.
U. S. Department of Transportation, Federal Highway Administration,
"SAPOLLUT/SAPLSM User's Guide," FHWA Urban Planning Division, HHP-23,
Washington, D.C., October 1976.
U. S. Environmental Protection Agency, "Guidelines for Air Quality
Maintenance Planning and Analysis, Volume 7: Projecting County
Emissions," EPA-450/4-74-008, Research Triangle Park, North Carolina, May
1975.
U. S. Environmental Protection Agency, "Mobile Source Emission Factors,"
Office of Transportation and Land Use Planning, Washington, D. C.,
EPA-400/9-78-005, March 1978.
Wuebbles, D. J., W. H. Duewer, R. Tarp, and J. S. Chang, unpublished
recent calculations, 1979.
-68-
-------
4. THE EVALUATION OF PHOTOCHEMICAL MODEL PERFORMANCE
The purpose of this section is to consider those efforts involved in the
assessment of photochemical air quality model performance. In Section
2, a number of different photochemical models were identified that might
potentially be employed in the preparation or revision of State
Implementation Plans (SIPs). Given this variety of models, there is a
need for an adequate understanding of their performance, both in a
relative sense, compared with other models, and in an absolute sense,
for a particular application. This understanding will enable the user
to choose an appropriate model to ensure that it performs adequately in
the intended application.
In the past, model performance was usually evaluated by the model
developer in the course of an application or a special evaluation study.
Although air quality model usage has been increasing over the last
several years, there are no EPA-published guidelines describing how such
studies should be carried out. Recognizing this deficiency, EPA has
recently commissioned a study to examine performance measures and
standards as well as general procedures for conducting a performance
evaluation study. One of the resulting reports, by Hayes (1979),
considers alternative means of measuring model performance and possible
ways in which performance standards might be established. A companion
report by Hillyer, Reynolds, and Roth (1979) presents a generalized,
step-by-step procedure that may be used to evaluate the performance of
an air quality simulation model. The discussion in this section adopts
many of the concepts presented by Hayes (1979) and Hillyer, Reynolds,
and Roth (1979). The reader is referred to these reports for background
and supplemental information.
Some of the terminology used to describe the model evaluation process
may require clarification. The phrases "model validation" and "model
verification" are frequently used to designate the process of comparing
model predictions with suitable observations. In this section, however,
the terms "validation" and "verification" are avoided in referring to
the evaluation of a model. Instead, the more general phrase "model
performance evaluation" is used since it is more representative of the
process that this section describes. The "validity" of a model is taken
to be a concept defining how well model predictions would agree with the
appropriate observations, given a perfect specification of model inputs.
That is, validity relates to the inherent quality of the model
formulation. The term "verification" is reserved to describe a
successful (or positive) outcome of the model evaluation process.
The examination of model performance is motivated by two factors:
First, the model treatments of various physical and chemical phenomena
usually involve a number of approximations; second, the information
provided as inputs is often subject to considerable uncertainty. The
principal sources of these uncertainties include (Seinfeld, 1977):
-69-
-------
o Wind velocity components.
- Uncertainties in available wind speed and direction
measurements.
- Inadequate number of nonrepresentative locations of
measurement sites (especially aloft).
-Approximations associated with wind field analysis
techniques used to prepare model inputs.
o Source emissions and removal functions.
- Inaccurate specification of source location.
- Inaccurate estimation of plume rise.
- Errors in emission factors for stationary sources.
- Inadequate representation of actual driving characteristics
in the federal mobile emissions test procedures.
- Errors in emission factors and vehicle miles traveled for
mobile sources.
- Inaccurate characterization of temporal variations.
- Inadequate parameterization of pollutant removal mechanisms.
o Chemical reaction mechanism.
- Omission or inadequate characterization of chemical reaction
steps.
- Uncertainties in measurement or specification of reaction
rate constants.
- Inaccurate characterization of temperature effects.
- Uncertainties associated with categorizing species into
reactive groups (such as paraffins, olefins, aromatics,
etc.).
o Initial and boundary conditions.
- Inadequate spatial characterization of the concentration
field on the upwind boundary of the region.
- Inadequate characterization of concentrations aloft.
o Numerical solution methodology.
- Computational errors associated with the use of finite
difference methods.
- Computational errors associated with other numerical
techniques.
In addition, it should be noted that the air quality data employed to
judge model performance are subject to error. Instrumental errors can
occur, or the spatial character of the measurement may not be directly
comparable to that of the model predictions (e.g., a point measurement
may be compared to a spatially averaged model prediction). Thus, there
are a number of sources of error in even the most sophisticated
photochemical models. Some of the simpler techniques, such as
trajectory models, can be derived from the same governing equation as
grid models with appropriate assumptions, and so they are subject to as
many, if not more, sources of error than are the more sophisticated
techniques. The purpose of the model evaluation study is to test the
entire model, including formulation, available aerometric and emissions
data, and model input preparation procedures, to obtain some
quantitative measure of model performance. Hi 1st (1978) suggests in a
recent report the additional need to evaluate the performance of
specific components of a model. For example, special studies could be
performed to assess the plume rise algorithm included in a Gaussian
-70-
-------
point source model. These types of investigations will be especially
important in the initial applications of a model because they may aid in
the identification and rectification of inadequate treatments of
atmospheric processes.
OUTLINING THE MODEL EVALUATION STUDY
The initial outline for a model performance evaluation can be divided
into two parts:
o Specific definition of the scope of the study;
o Selection and design of performance measures and standards
(i.e., performance evaluation criteria);
Defining the extent of required verification entails addressing the
following issues: size and boundaries of the modeling region, model
resolution, pollutants to be included, time period to be simulated, and
type and number of meteorological regimes.
Definition of the Size and Boundaries of the Modeling Region
The appropriate modeling region, a function of the model and of its
application, should contain all areas v.'ith significant population and
those where the peak concentrations are expected to occur. Trajectory
models are additionally constrained by the distances over which the
basic model concept is valid. Grid models are constrained by the number
of grid cells that can be accommodated in the available computer core
storage. These cells must be judiciously arranged to cover the area of
interest. If the entire area cannot be accommodated by the host
computer, the grid cell size can be adjusted, though this adjustment of
course influences the resolution of the model.
Definition of Spatial and Temporal Model Resolution
Spatial resolution (i.e., spacing between grid points) is related to
region size and the number of grid cells into which the region is
divided. In addition to the amount of computer storage, the number of
grid cells is constrained by time limitations. For example, though the
SAI Airshed Model has no inherent limitations on the number of grid
cells, doubling the number of cells slightly more than doubles the
necessary computer time.
In addition, the spatial resolution of the model is influenced by the
spatial resolution of the input data. Assuming the existence of a
uniformly spaced wind measurement network and invoking sampling theory
concepts, Lamb and Seinfeld (1973) argue that, theoretically, a model
cannot resolve features in the initial concentration field on a scale
smaller than one-half the distance between the monitoring stations.
Furthermore, the model cannot resolve features in the predicted
concentration field on a scale smaller than the resolution of the
emission inputs. In previous photochemical model applications, the grid
-71-
-------
specification has usually been more heavily influenced by the resolution
in the available emissions data than by the spacing between the wind
stations. This situation does not appear unreasonable when one
considers that re-creation of the actual flow field is not as important
in the evaluation of future emissions patterns as it is in the attempted
simulation of an historical air pollution episode. In general, the grid
spacing should be sufficiently fine to resolve the expected
characteristics of the concentration field as well as the important
features of the physical and chemical atmospheric processes. The
horizontal grid spacing employed in pre/ious grid modeling studies has
ranged from 1 to 10 kilometers.
Grid models are best suited for predicting the concentrations of
species, such as ozone, that have a one-hour air quality standard.
Averaging times of up to 24 hours can be accommodated with grid models,
but longer time intervals would require too much computer time to be
practical. If the one-hour averages calculated by the model are to
reflect actual conditions during that hour, the input data should also
have a comparable temporal resolution. This requirement is discussed in
more detail for each input variable in Section 3.
Time Period to be Simulated
The simulation must include the induction period for the formation of
photochemical pollutants. To minimize the effects of the initial
conditions, the simulations should be started prior to the morning rush
hour. In some cases pollutant carryover from the previous day is
important, making a multiple day simulation desirable. The length of
the simulation depends on the time period over which significant
pollutant concentration levels are observed or predicted, e.g., from
early morning to late afternoon.
Choice of Meteorological Conditions
In general, performance is evaluated by applying the model to one or
more historical air pollution episodes. If adequate performance is
achieved, the model can then be assumed ready for use in calculating the
effects on air quality levels of changing emissions. Ideally, the
model's ability to predict the effects of emissions changes should be
directly assessed. However, historical data bases suitable for
evaluating model performance usually cover only a limited span of time
over which emissions have not significantly changed. A limited check on
this aspect of model performance can be obtained by carrying out
simulations for both weekdays and weekends. (However, the value of this
check is limited by the generally low quality of weekend emissions
inventories.) In addition, as models are applied to several urban areas
having varying emissions intensities, some information will be obtained
on the model's prediction capabilities under different emissions
conditions. Until control strategies are actually implemented and their
effects on air quality are measured, model performance must necessarily
be evaluated in the context of present emissions conditions.
-72-
-------
An air pollution episode suitable for use in a control strategy
evaluation should possess three main attributes. First, it should not
place undue "stress" on the model such as would be the case for episodes
characterized by extremely complicated wind flows or unusual
ground-level pollutant concentration distributions. The objective is to
select an episode for which the model can be expected to perform well;
use of an unduly complicated episode increases the likelihood of
degraded model performance and attendant uncertainties about the meaning
of the strategy evaluation simulations. Second, the episode should
stress the control strategies: if the aim is to evaluate the effects on
the exposure and dosages experienced by the human population of
controlling point source emissions in a city with many point sources in
its eastern portion, an episode with predominantly westerly winds should
not be chosen. Third, the episode should represent worst-case
conditions: the meteorological and emissions conditions should be
conducive to the occurrence of upper percentile pollutant
concentrations.
Selection of an episode is a screening process. First, historical data
are searched to define periods of from one day to several consecutive
days during which elevated pollutant concentrations were observed,
preferably at several monitoring locations. Then, the extent of data
available, from both the routine monitoring network and special studies,
is determined. Those episodes for which serious data gaps occur are
deleted from further consideration. Finally, the remaining candidate
episodes are examined individually to select the most promising
episode(s).
Determination of the Number of Meteorological Regimes
Often it may be advantageous to consider the selection of more than one
episode. If high concentrations are observed under very different
meteorological regimes, it may be necessary to test alternative
emissions control strategies under the various worst-case regimes. Of
course, the number of episodes that can be analyzed is limited by the
resources available to prepare data bases and to carry out the required
model evaluation exercises. Two basic considerations in selecting the
number of regimes to be included in an evaluation study are:
o Specification of a sufficient number to enable
characterization of model performance.
o Selection of regimes that will be appropriate for use in the
applications studies.
In general, two to six different meteorological scenarios are
recommended. A final control strategy should be assessed using at least
two or three different "worst-case" meteorological conditions to
determine compliance with the NAAQS. We refer the reader to Section 6
for a further discussion of the choice and usage of "worst-case"
conditions.
-73-
-------
Performance Measures and Standards
No measured quantity is equivalent to a photochemical model's
concentration predictions; therefore, no absolute measure of model
performance exists. The best available comparison is between station
measurements and grid averaged predictions. That is, it is necessary to
assume that station measurements are representative of the average
concentration throughout the grid. Although there is no officially
accepted method of comparing observed and estimated concentrations,
several techniques are commonly used, including calculating correlation
coefficients, percent differences, and standard deviations.
Hayes (1979) has carried out one of the most systematic considerations
of how to measure model performance to date. Table 4-1 lists the
measures and standards from this study that are thought to be most
useful for photochemical models. Hayes (1979) has identified three
objective rationales for setting model performance standards, viz:
o Health effects
o Control level uncertainty
o Guaranteed compliance.
The first of these contains the premise that computed considerations
should not deviate from the true values sufficiently to result in
significantly underestimated health effects. In the second rationale,
uncertainties in the percentage of emissions control required should be
held to tolerable values. The third rationale implies that compliance
with air quality regulations should be "guaranteed," that is, that
uncertainty in the model's predictions should be biased in a
conservative direction.
Unfortunately, these rationales, which state desirable goals for model
evaluation, cannot be applied at the present time because there is no
objective basis for setting the required uncertainty levels. It is
therefore recommended that a "pragmatic/historic" rationale (Hayes,
1979) be employed to evaluate the performance of photochemical models.
In this method, the performance standard is that the model must perform
at least as well as in recent evaluation efforts carried out for a
similar urban area and model application.
Trajectory models present a special case in determining performance
measures and standards. Because trajectory models yield predictions
along a trajectory path, special attention must be given to how the
computed and measured values are to be compared. Two techniques used in
previous model performance evaluation studies include:
o Running the trajectory near stations of interest and
estimating the "measured" concentration at the trajectory
location through interpolation of the available observations
(Eschenroeder, Martinez, and Nordsieck, 1972).
o Defining a set of trajectories that will arrive at a
monitoring station location throughout the period of interest.
-74-
-------
Table 4.1. Sample Photochemical Model Performance Measures and Standards
Performance
^Attribute
Accuracy of the
peak prediction
Performance Measure
Ratio of the predicted station peak to the
measured station peak (could be different
stations)
Difference in timing of occurrence of station
peakt
Performance Standard
Limitation on uncertainty in aggregate health
impact and pollution abatement costs*
Model must reproduce reasonably well the
phasing of the peak—say +1 hour
Absence of systematic
bias
Average value and standard deviation of the
mean deviation about the perfect correlation
line, normalized by the average of the
predicted and observed concentrations,
calculated for all stations during those
hours when either t.he predicted or the
observed values exceed the NAAQS
No or very little systematic bias at
concentrations (predictions or observations)
at or above the NAAQS; the bias should not be
worse than the maximum bias resulting from
EPA-allowable calibration error (-8 percent
is a representative value)
Lack of gross error
Average value and standard deviation of the
absolute mean deviation about the perfect
correlation line, normalized by the average
of the predicted and observed concentrations,
calculated for all stations during those
hours when either the predicted or the
observed values exceed the NAAQS
For concentrations at or above the NAAQS, the
error (as measured by the overall values of
the average and standard deviation of the
mean normalized absolute deviation) should
not be worse than the error resulting from
monitoring instrumentation error
Temporal
correlation!
Temporal correlation coefficients at each
monitoring station for the entire modeling
period and an overall coefficient averaged
for all stations
At a 95 percent confidence level, the
temporal profile of predicted and observed
concentrations should appear to be in phase
(in the absence of better Information, a
confidence Interval may be converted into a
minimum allowable correlation coefficient by
using an appropriate t-statistic)
Spatial alignment
Spatial correlation coefficients calculated
for each modeling hour considering all
monitoring stations, as well as an overall
coefficient average for the entire day.
At a 95 percent confidence level, the spatial
distribution of predicted and observed
concentrations should appear to be correlated
* These measures are appropriate when the chosen model is used to consider questions involving pollutants subject
to short-term standards. They are most important when the pollutant is also received.
t These may not be appropriate for all regulated pollutants in all applications.
derived based on pragmatic/historic experience should be employed.
When they are not, standards
Source: Hayes (1979).
-75-
-------
When the air parcel is within a certain distance of the
station location (say 1 to 5 kilometers), then the predictions
can be time averaged and compared with the observations. In
some instances, it may be necessary to combine the results
from several trajectory runs to calculate a time-averaged
prediction commensurate with the measured values (Wayne,
Kokin, and Weisburd, 1973).
These two techniques may be combined in evaluating the performance of a
trajectory model. In general, the second scheme cited above is
preferable for establishing the model's performance characteristics
since it is not subject to the added uncertainties introduced by the use
of the interpolation procedure.
In summary, regarding both grid and trajectory models, there is a clear
need for further studies to establish suitable model performance
measures and standards.
MODEL EVALUATION PHASE
The fourth and final phase of the model evaluation effort entails the
adaptation of the model to the study area, the collection of
supplemental data, the preparation of inputs, the exercise of the model,
the analysis of the results, and the rectification of performance
deficiencies. The following sections discuss the most important topics
related to these issues.
As part of the model selection task, the issue of the availability of a
suitable computer was raised. At this point, it should be emphasized
that achieving successful runs of the programs on a computer for which
they have not been previously implemented can be very time consuming,
possibly requiring assistance from the model developer. To ensure that
the programs are transferred correctly, a test case should be supplied
by the model developer or past users to be employed to test the
programs. The programs should be up and running on the user's computer
before any modifications are made.
Adaptation of the Model to the Study Area
Model modifications fall under two categories: modifications to
accommodate the modeling region and alterations in the model algorithms.
The first of these types of modifications may simply require routine
alterations to the codes. For example, in some models the array
dimensions must be changed for each modeling region of a different size
or shape. This is usually a fairly simple and straightforward process.
The second type of modification can be somewhat more complex. The best
available model may not treat all the atmospheric processes as the user
would like. In addition, the available procedures for preparing model
inputs may not be the most suitable for use in the new study area.
Thus, it may be necessary to modify an existing algorithm in the model
or, perhaps include a process not treated by the model. For example,
-76-
-------
the particular plume rise algorithm contained by the model may not be
considered to be the best one for the intended application. A
better-suited algorithm can be coded and inserted in the model as a
replacement for the existing algorithm. To make this kind of
modification requires a thorough knowledge of the computer code and may
require the assistance of the model developer. After the modifications
are made, the program should be carefully tested to determine whether
the new algorithm is working properly and to assure that no other
operations have been changed accidentally.
Collection of Model Inputs
The methods of obtaining the data for specifying the model inputs are
described in detail in Section 3. As the data are gathered, they must
be checked for consistency and reformatted for use with preprocessor
programs, which then use the collected data to construct the input files
for the main simulation program.
The conversion of the raw data into the simulation program input files
is one of the most critical tasks in the evaluation study. The
preprocessor programs should make the maximum use of the data, and the
output from these programs should be carefully reviewed to ensure that
the results meet the user's expectations. Since the data preparation
process is very dependent not only on the quantity of data gathered, but
also on the procedures employed to estimate the inputs, a good
understanding of the algorithm and assumptions used by the data
preparation programs is essential to prepare a set of adequate model
inputs.
Performance of Model Simulations
Once the input files have been prepared it is a simple task to exercise
the simulation program. The model results should be saved on a
permanent file (e.g., a tape) so that performance measures can be
estimated and subsequent computations can be carried out by the user.
If the run is successful, other analyses may be of value, such as dosage
based on the distribution of pollutant concentration levels and the
population, concentration isopleths, areal extent of the region for
which concentrations exceed the NAAQS, and concentration frequency
distributions.
Assessment of Model Simulation Results
The first step in analyzing the results of the model simulation is to
compare each computed performance standard. If the measure meets the
standard, the user may proceed with some confidence that the model will
produce reliable results for the intended applications.
If a performance measure fails to meet a performance standard in some
respect, further analysis is indicated, as outlined below. It should be
noted that even if the measure does meet the performance standard, this
type of analysis will give much useful information about model behavior;
therefore, the analysis is strongly recommended in all cases.
-77-
-------
The analysis of the model evaluation results should center on the
differences between computed and observed pollutant concentrations, that
is, the residuals. These residuals can arise from three sources:
o Errors in the input data (emissions data, meteorological data,
initial and boundary conditions).
o Errors in the formulation of the model (approximations made in
modeling pollutant transport or chemical transformations).
o Errors in the air quality measurements used for comparison.
Estimates of the precision of various aerometric measurements can be
determined by replication of the measurements and by the application of
standard statistical techniques. The accuracy of the measurements can
be characterized through an analysis of instrument calibration data.
Nonzero residuals can also result from the extrapolations and
interpolations necessary to generate a complete model input data set
from insufficient data. The errors introduced by these extrapolations
and interpolations are somewhat more difficult to quantify than are
instrumental errors. In addition, station exposure to localized biases
such as near roadway sources of NO may unduly influence the
measurements.
Discrepancies introduced by shortcomings in the model's formulation are
difficult to evaluate because there is no "true" model of the relevant
atmospheric processes for comparison. Also, in light of the
uncertainties of the input and comparison data, error due to model
formulation cannot be isolated from the total modeling process.
Careful analysis of the residuals can yield much useful information
about the model, even if quantitative statements about sources of error
cannot be made. Several different ways of analyzing residuals can be
informative [see, for example, Koch and Thayer (1971)]. Plots of
residuals against time of day can reveal systematic biases, which might
result from an inadequate kinetic mechanism in a photochemical model.
Dependence of the magnitude of residuals on concentration might indicate
that a monitor is poorly located to detect a large area-wide
concentration level, or that the wind field inputs for the model
incorrectly represent the transport of a plume from a large point
source. Differences between residual dependencies for primary and
secondary pollutants can be used to infer deficiencies in the kinetic
mechanism or dispersion processes. The examples given here are only a
sampling of possible situations; this type of analysis should be guided
by the particular situation and a knowledge of the various technical
features of the model. We refer the reader to the report by Liu et al.
(1976), which includes an evaluation of the performance of three
photochemical models based on an analysis of residuals.
Other statistically based analysis methods that can be used to study the
evaluation results are referenced below:
-78-
-------
o Scatter plots of observed and computed considerations
(Anderson et al., 1977, Duewer et al., 1978, MacCracken et
al., 1975)
o Correlation between observed and computed results (Reynolds et
al., 1979, Duewer et al., 1978, MacCracken et al., 1975)
o Nonparametric tests of location to indicate possible bias of
computed concentrations relative to observations (Lehmann,
1975).
Rectification of Inadequate Performance
When the model performance evaluation indicates less than satisfactory
performance, diagnostic studies should be carried out to obtain evidence
of deficiencies in the model's formulation or its inputs. The model
user may need to perform these diagnostic analyses at three points:
o During the initial model adaptation to the study region and
preparation of inputs.
o Just after the first unsuccessful model exercise.
o Immediately prior to carrying out the final simulation.
In general, the emphasis of the requisite analyses for each of the three
stages is somewhat different.
When the model is initially being adapted to the study region, the user
is dealing directly with issues of the adequacy of the data base for use
in preparing model inputs. After the user has selected a proposed input
preparation algorithm and has exercised the appropriate computer
program(s), the resultant input file must be examined to determine
whether the atmospheric phenomena represented on the file are adequately
characterized. For example, estimating the surface wind field using an
inverse distance weighted interpolation scheme can sometimes yield a
"jump" in the wind velocity at those locations near the radius of
influence of a monitoring station. To correct for this unrealistic wind
behavior, it may be appropriate to change the radius of influence input
to the program, to employ a smoothing algorithm, or possibly to select
an alternative wind field preparation technique. Each file prepared for
input to the model must be scrutinized carefully. Previous experience
in using the available preprocessor routines is often useful in the
anticipation of problems that are likely to occur in using particular
algorithms in conjunction with sparse observational data. This
experience will be helpful in reducing the number of iterations required
to achieve a reasonable initial set of model input files.
The first set of results obtained from a model are sometimes found to
represent inadequately the actual observed concentration patterns in the
region. Thus, diagnostic analyses are needed to uncover the causes of
these discrepancies. For example, after the first simulation of a
smoggy day in Sacramento, Reynolds et al. (1979) found that the
-79-
-------
magnitudes of the computed ozone concentrations were in general
agreement with the observed values, but the predicted ozone cloud was
not in the appropriate location. Suspecting a possible problem in the
wind inputs, they reexamined the wind input file and the observed
meteorological data. An assessment of the single available pibal
sounding in light of the predicted concentration results suggested that
they should not have ignored the significant directional shear indicated
in the sounding. In preparing the initial wind input file, they
considered wind speed changes with height but not directional changes.
Upon revising the wind inputs and exercising the model, they obtained
significantly improved agreement between computed and observed ozone
values.
Throughout the simulation, the behavior of both primary and secondary
pollutants may suggest ways of improving the model results. Comparison
of the diurnal cycles of the measurements with the model predictions may
indicate that the initial or boundary conditions have been set too high,
that the estimates of emissions are too low, or that the atmospheric
stability has been incorrectly specified. These problems can generally
be discovered by reassessing the input data, and an appropriate change
can then be made that improves the results. If the improvement is still
insufficient to bring performance up to standard, the improved results
should be reanalyzed according to the same procedure that was used
initially, until the desired results are obtained.
However, results obtained by adjusting model inputs to bring calculation
closer to observation should only be accepted as evidence of
satisfactory model performance if the input adjustments can be justified
as a consequence of independent data (e.g., wind data not used
initially, remeasurement of emissions composition, etc). Simulations
based on the adjustment of some input within its range of error (to
improve fidelity) should be accepted only if the adjustment is general
(say a reaction rate constant) and should not be used in model
evaluation if the adjustment is day specific (say a wind speed or a
boundary concentration).
If model results are found to be inadequate for a particular day, then
selection of another day may yield improved model performance. The
available data base may not be suitable for describing complex
meteorological phenomena; a day with simpler meteorological conditions
may be better suited for testing the model's performance.
In the development of any photochemical model, many approximations are
invoked to derive a "workable" computational procedure. Some of these
approximations may degrade model performance. For a particular
combination of inputs, a set of approximations that under most
conditions does not significantly affect the results, may generate
errors that are unacceptable. If this situation can be identified, it
may be possible to correct it. However, making changes in the computer
code requires a thorough understanding of that code. Most users will
not have enough experience with models to make code changes, so if a
model deficiency is discovered the model developer should be consulted.
Another possibility would be to consider the usage of a potentially more
-80-
-------
suitable model, if available, though time and resource constraints may
preclude this alternative.
In summary, the diagnosis of model performance problems is generally an
iterative procedure: A set of results is analyzed to determine the
likely causes of the discrepancies, the inputs of the model are revised
to reflect better the actual physical and chemical atmospheric
phenomena, and then the model is exercised again.
To carry out the appropriate diagnostic analysis, the study team should
have a thorough knowledge of the model's formulation, the computer
codes, and the pertinent atmospheric and emissions phenomena, as well as
experience in applying this knowledge to the resolution of model
performance problems. Without these capabilities, model users are
likely to pursue inappropriate directions leading to inefficiencies that
will raise project costs and delay completion of a project. If the
users do not possess the requisite skills, then appropriate EPA modeling
experts or private consultants should be contacted to aid in the
diagnosis.
It is important that model inputs be modified within their range of
uncertainty. Since discrepancies between computed and measured values
can result from errors in both the model's formulation and its inputs,
one should not attempt to cover up formulation inadequacies through
selective use of model inputs. Inadequate model performance is an
important finding, possibly indicating a need to further refine one or
more components of the model. Alternatively, inadequate model
performance may suggest that the atmospheric phenomena that actually
occurred on the simulation day cannot be sufficiently characterized by
the available observational data.
Before the model is used in the application study, satisfactory
performance should be demonstrated. In some cases it may not be
possible to verify the model with the available data base and resources,
yet circumstances may dictate that, even though unverified, the model
must be used anyway. For example, the analysis may be subject to a
deadline imposed by governmental regulations. Perhaps the model
represents the current state of the art and no improvements are possible
without a major research effort. Whatever the reason, if the model is
used, the previous analyses may pinpoint some of the deficiencies in the
model that caused it to fail to meet the required performance standards.
These deficiencies should be fully detailed in the account of its use.
The use of "correction factors" to calibrate model results should be
carefully considered and reviewed by all entities participating in or
reviewing the plan. It is unclear that such calibration factors can be
validly applied to subsequent control strategy results. Thus, the
stress should be placed on discovery and correction of discrepancies
between computed and measured values, not on calibration. Additional
discussion of this issue is contained in Section 6.
-81-
-------
EXAMPLES OF PREVIOUS MODEL EVALUATION STUDIES
Some of the most extensive photochemical model performance evaluation
studies have been carried out using the Livermore Regional Air Quality
(LIRAQ) Model and the SAI Airshed Model. This section provides a brief
overview of these model performance evaluation efforts.
The LIRAQ Model
The LIRAQ Model was evaluated by researchers at the Lawrence Livermore
Laboratory (Duewer, MacCracken, and Walton, 1978), who applied it to the
San Francisco Bay Area, the region for which it was developed.
Determination of the modeling region and the date and time to be
simulated were based on the types of future studies that were expected
to use the model. For applicability to selection of future control
strategies, the modeling region would have to include all the major
source areas. Since the photochemical version of LIRAQ can have at most
20 x 20 grid squares, the size of the grid squares was set at 5 km x 5
km so that the main population and industrial areas would be included
while maintaining reasonable spatial resolution.
The simulation period chosen was sunrise to sunset on two days in 1973:
26 July and 20 August. The former was one of the highest oxidant days
of the year, whereas the latter had relatively clean air. The period
from sunrise to sunset covers the typical cycle of atmospheric events,
starting with the morning emissions of primary pollutants, proceeding
with the buildup of secondary pollutants, and concluding with the
cessation of photochemical reaction processes when the sun sets. The
1-hour-average concentrations were calculated for comparison with the
station measurements.
The LIRAQ model evaluation study was coordinated with the Bay Area Air
Quality Management District (BAAQMD, formerly the BAAPCD) and other
local government agencies, which were responsible for collecting most of
the input data. Section 3 briefly describes how the data were collected
and used to prepare the input files. Data collection and data
preparation were coordinated so that the input file programs were
designed to use all the available data and not to require information
that was not available. This feature would not apply if LIRAQ were used
for another study region.
Since LIRAQ was originally developed for the San Francisco Bay Araa,
model adaptation to the study area was carried out as part of the
developmental efforts. The only input variable that was adjusted as
part of the model evaluation was the boundary conditions. The almost
complete lack of data for specifying the boundary conditions motivated a
sensitivity study of the effect of the boundary conditions on model
performance. In that sensitivity analysis, three sets of boundary
conditions were used for each day. The results showed that the effect
of the boundary conditions is significant; one set produced better model
performance in terms of a variety of statistical measures.
-82-
-------
The simulation results were compared with the station measurements of
NO, N02, total hydrocarbons, CO, and 03 (MacCracken and Sauter, 1975;
Duewer, MacCracken, and Walton, 1978). Examples of the types of
analyses performed are shown in Figures 4-1 and 4-2. In the plots for
each station, the qualitative measure of model performance is easier to
interpret than some of the statistical measures, but the quantitative
measures are necessary for comparison of different evaluation studies.
The scatter plots of the mean and maximum concentrations at each station
provide another indicator of model performance.
Since the first evaluation study of -IRAQ, the model has been improved,
particularly by the inclusion of an updated chemical mechanism. This
improved model has been evaluated using the same data base used in the
first effort. Subsequently, the model has been used to determine
possible control strategies for meeting the NAAQS as part of the
revisions being proposed to the State Implementation Plan (SIP) for the
San Francisco Bay Area.
The SAI Airshed Model
The SAI Airshed Model has been applied to five U.S. cities, including
Denver, Los Angeles, Las Vegas, St. Louis, and Sacramento. Table 4-2
summarizes the number of days simulated for each city and provides a
list of relevant references. The discussion below is limited to the
Denver evaluation effort.
Denver Evaluation Studies. Three studies of the air quality in the
Denver metropolitan area have been carried out using different versions
of the SAI Airshed Model. This discussion is limited to the
investigation that was done for Region VIII of the EPA (Anderson et al.,
1977). That study evaluated model performance and analyzed the
sensitivity of 1985 and 2000 air quality to different growth and
transportation scenarios using data collected by the Colorado
Departments of Health and Highways.
The daylight hours of three days were simulated for the evaluation
study: 29 July 1975, 28 July 1976, and 3 August 1976. To collect
additional meteorological data, a special field study was conducted on
29 July 1975. No additional data were collected for the other two days.
The data that are routinely collected in the Denver area include wind
speed and direction at 10 to 15 stations, two radiosondes per day at
Stapleton International Airport, and air quality measurements at about
10 locations. The emissions were calculated from data collected by the
Air Pollution Control Division of the Colorado Health Department and
from a highway model developed by the Colorado Highway Department.
Section 3 describes the data gathering efforts in more detail.
Unlike the LIRAQ evaluation application, the SAI Airshed Model had to be
modified specifically for the Denver application, since the model was
applied first to the Los Angeles air basin. These modifications
included relatively straightforward changes to the DIMENSION and FORMAT
statements in the computer programs. The latest version of the SAI
-83-
-------
20
o
10
.._ . • » A
DBU -- Burllngame
«*• I K-M C Ti-
tle • • I 4-M f K>
•• cm > I !««••«
b. DCO -- Concord
I'D •• i «•« • m-m
20
f 15
O
M
O
10
f ArD CMAMMM*
C. DFK — rrenont
TrMCCirT • •
n* « 1*1
<•••••
(to «• f «•« > «-«e
•• pnai • i i»ii-«c
o
5 10
C
CO«CL>TI« oatrr • t.n
t. DUG -- Los Gatos «««•« »>n • ••
ICM 1 «*M I «•••«
ITB «t« > *'• I «€•«•
1 tone: I
f. OHV « ttountaln
F
!!?•
0400 0800 1200 1600 0400 0800 1200 1600
Time (Pacific Standard) Time (Pacific Standard)
Source: Duewer, MacCracken, and Walton (1978).
FIGURE 4-1 EXAMPLES FROM THE COMPARISON OF AVAILABLE STATION OBSERVATIONS
OF OXIDANT WITH LIRAQ CALCULATED TIME HISTORIES OF OZONE
-84-
-------
i
3
I2
i
0
i
20
18
b.
12
X
2
£ 8
10
I
c.
20
fr
I
0 S 10
Observed
concentrations (pphm)
I
OSR
I
234 567
Observed mean (pphm)
1 I I
o DSJ
OORC
OR.MO
i
i
i y i
i
4 6 I 10 12 14 1C 18 20
Observed Mxlnum (pphm)
Source: Duewer, MacCracken, and Walton (1978)
FIGURE 4-
-------
cu
I
I
0)
oo
co
>
O CU
rv fv
C
O
4->
I-
=j
00
T3
C
ITS
cu
.C
(J
CO
O)
c r^-
O cn
+-> t— i
3
CQ it)
00
[-•»
cr>
o
r— i— CO •—
(D o r*»^ ^3
o. cn
Ol
LU
C
•p-
o
5
•r»
U
CL
X
LU
I— I
•^
CO
l+-
o
X
t-
3
00
•
c\j
•
•S
o
2_
t.
o
c
o
Q.
OO
tn
c
cu
o
—
VO ^O ^O
f^^ r*^ ^**
cn cn cn
in
3t
cn
JC
M-
o
c
o
•^
C/)
>
•r-
Q
O
eo
S-
o
f— «^ r—
O D- O
O LU O
C\J CVJ CM
CU
0
s-
cu
c
cu
• — •
IO VO
cn cn
1—4
»— »
>
C
o
•r—
cn
eu
o;
< t—
Cu O
LU O
CO CM
CO 00 CO
i — r-. i —
cn cn cn
cu
cn cn
nj ••-
CD 3
CU O
> —I
00
CU
(O
oo
-86-
-------
Airshed Model can be applied to almost any region without such
modifications.
At the time of the Denver project, the Carbon-Bond Mechanism that is now
incorporated in the SAI Airshed Model had just been developed. Since
the Carbon-Bond Mechanism had not previously been used in the model, the
simulation program and the preprocessor programs were adapted at that
time to handle this chemistry routine and the species associated with
it.
As is common in most applications, almost no data were available for
specifying the boundary concentrations. Since the Denver modeling area
had few upwind sources, the boundary concentrations were set to
relatively low background levels. Once the program "bugs" and the
keypunching errors were identified and removed, the model calculations
agreed satisfactorily with the actual observations. Examples from some
of the analyses performed for this study are given in Figures 4-3 and
4-4. Figure 4-3 shows the calculated and measured ozone concentrations
for each station as a function of time. This information is the basis
for all of the other statistical analyses. Figure 4-4 presents a way of
estimating the model bias; it indicates that the model tends to
underpredict at low concentrations, but the results show almost no bias
at higher concentrations. Further analyses showed that discrepancies in
the model calculations are comparable to the expected error in the
observations due to measurement errors.
- 87-
-------
—0 -0 PARKER RD
f6 7 8
7 * ?
9 10 11 12 1 2 34 5 6
Tff TT T? T I T 4? ? 7
of D.y, by
Source: Anderson et al. (1977).
FIGURE 4.3 EXAMPLES, OF OBSERVED AND SAI AIRSHED MODEL PREDICTED
HOURLY OZONE CONCENTRATIONS (pphtn) AT VARIOUS
STATIONS IN DENVERON 28 JULY 1976
.-88-
-------
I t
+ 1.0r-
u
c
o
3
§
t-
u
o
t.
01
a.
§
c
o
c.
c
Root-Mean-Square Ozone Concentration (ppnm
1/2
l(Observed)2 + (Predicted)2
-2.0
Source: Anderson et al. (1977).
FIGURE 4-4
ESTIMATE OF BIAS IN SAI AIRSHED MODEL PREDICTIONS AS
A FUNCTION OF OZONE CONCENTRATION
-89-
-------
REFERENCES
Anderson, G. E., et al. (l'977), "Air Quality in the Denver Metropolitan
Region: 1974-2000," EPA-908/1-77-002, Systems Applications,
Incorporated, San Rafael, California.
Donnelly, D. E. (1978), "Oxidant Model Applications: Denver," 57th
Annual Transportation Research Board Meeting, January, 1978, Washington,
D. C.
Duewer, W. H., M. C. MacCracken, and J. J. Walton (1978), "The Livermore
Regional Air Quality Model: II. Verification and Sample Application in
the San Francisco Bay Area," J. Appl. Meteorol., Vol. 17, No. 3, pp.
273-311.
Environmental Protection Agency [EPA] (1976), "Quality Assurance
Handbook for Air Pollution Measurement," EPA-600/9-76-005, Research
Triangle Park, North Carolina.
Eschenroeder, A. Q., J. R. Martinez, and R. A. Nordsieck (1972),
"Evaluation of a Diffusion Model for Photochemical Smog Simulation,"
CR-1-273, EPA-R4-73-012a, General Research Corporation, Santa Barbara,
California.
Hayes, S. R. (1979), "Performance Measures and Standards for Air Quality
Simulation Models," EF78-93R, Systems Applications, Incorporated, San
Rafael, California.
Hillyer, M. J., S. D. Reynolds, and P. M. Roth (1979), "Procedures for
Evaluating the Performance of Air Quality Simulation Models," EF79-25,
Systems Applications, Incorporated, San Rafael, California.
Hilst, G. R. (1978), "Plume Model Validation," EA-917-SY, Workshop
WS-78-99, Electric Power Research Institute, Palo Alto, California.
Houghland, E. S. and N. T. Stephens (1976), "Air Pollutant Monitor
Siting by Analytical Techniques," J. Air. Pollut. Control Assoc., Vol.
26, p. 51.
Koch, R. C., and S. D. Thayer (1971), "Validation and Sensitivity
Analysis of the Gaussian Plume Multiple-Source Urban Diffusion Model,"
EF-60, GEOMET, Incorporated, Gaithersburg, Maryland.
Lamb, R. G. and J. H. Seinfeld (1973), "Mathematical Modeling of Urban
Air Pollution—General Theory," Environ. Sci. Techno!.. Vol. 7, pp.
253-261.
Lehmann. E. L. (1975), Nonparametrics: Statistical Methods Based on
Ranks, (Hoiden-Day Incorporated, San Francisco, California).
-90-
-------
Lui, M. K. , et al., (1976), "Continued Research in Mesoscale Air
Pollution Simulation Modeling: Vol. I. Assessment of Prior Model
Evaluation Studies and Analysis of Model Validity and Sensitivity," EPA
600/4-76-016A, Systems Applications, Incorporated, San Rafael,
Califo^nia.
MacCracken, M. C., and G. D. Sauter, eds. (1975), "Development of an Air
Pollution Model for the San Francisco Bay Area," UCRL-51920, Vol. 1,
Lawrence Livermore Laboratory, Livermore, California.
Ott, W. R. (1977), "Development of Criteria for Siting Air Monitoring
Stations," J. Air Pollut. Control Asoc.. Vol. 27, p. 543.
Reid, L. E., et al. (1979), "Adaptation of the SAI Airshed Model for
Usage with the Regional Air Pollution Study (RAPS) Data Base," EPA
68-02-2429, Systems Applications, Incorporated, San Rafael, California.
Reynolds, S. D. et al. (1979), "Photochemical Modeling of Transportation
Control Strategies, Vol. I. Model Development, Performance Evaluation,
and Strategy Assessment," EF79-28, Systems Applications, Incorporated,
San Rafael, California.
(1973), "Further Development and Validation of a
Simulation Model for Estimating Ground Level Concentrations of
Photochemical Pollutants," Systems Applications, Incorporated, San
Rafael, California.
Roth, P. M., et al. (1971), "Development of a Simulation Model for
Estimating Ground Level Concentrations of Photochemical Pollutants,"
71-SAI-21, Systems Applications, Incorporated, San Rafael, California.
Seinfeld, J. H. (1977), "Current Air Quality Simulation Model Utility,"
Department of Chemical Engineering, California Institute of Technology,
Pasadena, California.
(1972), "Optimal Location of Pollutant Monitoring
Stations in an Airshed," Atmos. Environ., Vol. 6, p. 847.
Tesche, T. W., and C. S. Burton (1978), "Simulated Impact of Alternative
Emissions Control Strategies on Photochemical Oxidants in Los Angeles,"
EF78-22R, Systems Applications, Incorporated, San Rafael, California.
Tesche, T. W., and R. I. Pollack (1978), "Evaluating Simple Oxidant
Prediction Methods Using Complex Photochemical Models," EPA 68-02-2870,
Systems Applications, Incorporated, San Rafael, California.
Tesche, T. W. , C. S. Burton, and V. A. Mirabella (1979), "Recent
Verification Studies with the SAI Urban Airshed Model in the South Coast
Air Basin," Proc. of Fourth Symposium on Turbulence, Diffusion, and Air
Pollution, 15-18 January 1979, Reno, Nevada.
Wayne, L. G., A. Kokin, and M. I. Weisburd (1973), "Controlled
Evaluation of the Reactive Environmental Simulation Model (REM)," EPA
R4-73-013a, Pacific Environmental Services, Incorporated, Santa Monica,
California.
-91-
-------
5. MODEL APPLICATIONS
Model applications to support the development of a SIP can take a
variety of forms. As a minimum, the model should be applied to
demonstrate that a control strategy or set of emission reductions will
result in attainment of the ozone standard. This single application
forms the bridge between the problem, as defined by excesses of the
ambient ozone standard, and the proposed solution. The problem occurs,
however, of how to anticipate the appropriate control strategy. Since
there have been only two previous applications of photochemical
dispersion models in support of air quality plan development, the base
of available experience in this area is limited.
The approach emphasized in this report is the approach used in the San
Francisco Bay Area. It consisted of three types of applications:
baseline projections, emissions sensitivity analyses, and control
strategy simulations. Depending on the specific issues to be addressed
and resource constraints, the degree to which one or another of these
applications is employed will be unique to each modeling program. For
example, it may be more economical to omit or minimize the use of
sensitivity tests if the control strategies to be tested are
well-defined at an early stage in -;he process. Other approaches or
refinements may be developed in the future as more experience is gained
in model applications.
BASELINE PROJECTION
This application establishes the magnitude of air quality problems in
future years in the absence of any additional control programs. It
addresses the question of whether the ozone standard will be met at some
point in the future. For example, the 1977 Clean Air Act Amendments
attainment deadlines of 1982 and 1987 would be likely choices for future
years to be used in the modeling analysis. If the results of the
baseline runs for those years indicate that excesses of the ozone
standard will still occur despite the continued application of all
existing control programs, then the need for additional control programs
is established. If the results indicate that attainment may be expected
sometime between 1982 and 1987, then any additional controls considered
could be focussed on what could be implemented in the near term. If
attainment is indicated in both 1982 and 1987, then there is no need for
a non-attainment plan and attention can be focussed on long-term
maintenance of the ozone standard.
The Clean Air Act also requires long-term maintenance of air quality
standards, and EPA has interpreted this to mean a minimum of ten years
following the expected attainment date. Thus, an additional future year
of interest would be further into the future, such as the year 2000.
The selection of the particular year to be used in this case may depend
more heavily on the availability of data on which the emission inventory
projection may be based.
-93-
-------
EMISSIONS SENSITIVITY ANALYSES
There are numerous sensitivity tests that may be conducted to answer a
variety of questions. One set of tests that would be central to plan
development would be to simulate the effect of various hydrocarbon
and/or NOx emission reductions in order to estimate appropriate targets
for controls. In this instance, it may be useful to set up a diagram of
the tests as illustrated in Figure 5-1. This serves to display the
potential test options for discussion. (If resources are abundant
enough tests may be run such that an EKMA-type isopleth diagram could be
derived which would be unique to the region being modeled.)
The principal advantage of sensitivity tests is that they may be
structured to focus on whatever control issues are important 1n a given
region. Volume II of this report describes the applicability of various
selected photochemical models to a variety of strategic control issues
(e.g., control of elevated vs. low-level emissions, selective control of
hydrocarbon species, long range trensport, effects of spatial and
temporal redistribution of emissions, etc.). In each case, care should
be taken in the construction of the tests, particularly in the setting
of initial and boundary conditions for each run. (See Sections 3 and 6
for additional discussion of initial and boundary conditions.)
CONTROL STRATEGY SIMULATIONS
It is likely that more than one control strategy simulation will be
desirable because (1) there will likely be more than one combination of
control measures that could result in attainment of the ozone standard,
and (2) there may be particular interest in testing the effectiveness of
specific controls. The structure of these simulations is determined
solely by the priorities for each, and the time and resources available
at this stage of plan development. It should be kept in mind that for
plan development purposes, these simulations are the culmination of all
of the previous efforts to collect data, evaluate model performance, and
perform baseline and sensitivity tests. Again, particular care should
be taken in specifying the initial and boundary conditions for these
runs.
ALTERNATIVE PROGRAM DESIGNS
One approach to designing a model application program is to emphasize
the use of sensitivity analyses to assist in the structuring of control
options. For example, analyses could focus on estimating the impact on
ambient ozone levels of pre-specified percentages of hydrocarbon and NOx
emission reduction. These percent emission reductions could be applied
uniformly without regard to location, time, or reactivity of the
emission. This sensitivity study would generate target percent emission
reduction estimates for hydrocarbon and nitrogen oxides emissions. An
emission control strategy could then be designed to result in the given
percent emission reductions, and would then be subsequently tested by
the model.
-94-
-------
c
g
o
o
en
w>
E
0>
o
.a
o
c
(U
o
20 I
40 I
601
801
1001
Baseline
expected
ozone maximum
80
60
40
20
Percent NOx emission reduction
Figure 5-1
sensitivity
-95-
-------
A second approach would emphasize the direct testing of alternative
control strategies. This approach would involve an iterative testing of
various control strategies. For example, the first strategy tested may
be a strategy which meets the emission reduction requirements estimated
by EKMA. If the resulting ozone concentration is below the standard,
then the next strategy to be tested would be less stringent. This
process continues until an optimum strategy that meets the ambient air
quality standard is derived.
The specific approach adopted in a given program will be unique to the
issues and constraints of the total planning effort.
PRACTICAL ASPECTS OF DESIGNING A MODEL APPLICATION STUDY
A sample sequence of tasks for completing each of the model applications
mentioned is shown in Figure 5-2. The sequence is consistent with the
timing required to produce the emission inventory inputs. Baseline
inventories can be prepared first, while inventories that include the
effects of alternative controls require additional development effort
and are therefore done last. In between, the emissions sensitivity
analyses can be done using the baseline inventories, and can provide
useful guidance on how stringent the control strategies should be
designed to be.
The amount of time and resources devoted to each application will depend
on the number of prototype days being used, the number of future years
under consideration, the number of different strategies to be examined,
etc. These parameters will be unique to each region.
The number of model runs required beyond those used to evaluate model
performance may be estimated with the following formula:
N = DxYx(l+S + C)xM
where
N = number of model runs
D = number of prototype days
Y = number of future years under consideration
S = number of sensitivity analysis cases
C = number of control strategy cases
1 is indicated to account for baseline runs
M = arbitrary multiplier to account for wasted runs caused by
errors in input specification or other factors.
For example, if two prototype days and two future years are being used
to evaluate six sensitivity cases and three control strategies, then the
number of useful model runs needed is 40. If M=2, then the total number
of runs becomes 80, In many cases, and especially in the case of
Eulerian grid models, it may not be practical to complete this number of
runs. There are a variety of ways to reduce the number of model runs:
-96-
-------
CO
C
CD
-? CD
QQ o a 2.
C. O)
•• o «
S Q. (0 T3
CO -J5
f^
Q co
"> CO
•£ _ 9> -«
.g o £ «
D) c
E o -
d c Ti D)
CD JO O O
CL Q. E CL
O
>
CD
55 3 S
CD CO CO
^ > CO
o c a. c
_i to co a.
CD
CO ^
CD JO jo
a. n a.
,„ CO CO ^ CO
co£ o go
o> CD co E i=
CO C Q. CD C
< 5 .i ^ 8
en
c
CO >
CD .=
"O to
jl> T5 E JO
CL co "cO -fe
S CO r- C
•S i .2 -S co
0 .§ co % ffl
c?s|-o e>
H O CO 'o
C CO C wZ
9 ~o cB ^
? c ±± o
— CO CO O
o
-
g
O Q. t
C^
0)
5
in
§
.
a
(0
"55
•o
o
JS
I
o
(0
«>
Q.
E
&
CD
o
CO g
ill
o£5
5 a. a>
CD 2
c co
— CO
CD O
CO CD
CO
c
CD
CO
CO
c co
O CD
Eg
CU CO
O)
OJ
CO
k_
CO CO
— CD
O CO
O r-
O CO
J5
-------
o The sensitivity analysis tests may be restricted to a single
future year and a single prototype day (in the example, six
sensitivity runs would be made);
o Control strategies to be tested may be lumped together, and
tests may be restricted to two prototype days (in the example,
D=2, Y=2 and C=3, resulting in 12 control strategy runs);
o The EKMA model or a trajectory model may be employed to
perform sensitivity analysis tests.
Care should be exercised in interpreting the results of the sensitivity
tests if a different model is being employed for that purpose. Due to
their unique construction, each model may provide results which appear
qualitatively similar but may quantitatively differ. The ultimate
usefulness of applying more than one model in the planning effort is
limited due to the likelihood of discrepancies between results. Even
small discrepancies of ten percent can translate into substantial cost
differences in the controls required to meet the standard, thus reducing
the credibility of the plan. Thus, if EKMA is used in conjunction with
a photochemical dispersion model, it should only be used to produce
informal initial estimates of emission control requirements.
The multiplier factor should not be underestimated, for even the most
careful programmer can overlook some minute detail that could invalidate
a run. In the San Francisco example to follow, over seventy runs were
made of the LIRAQ model (at an average cost of $600 per run not
including staff costs), but only thirty were successful. The multiplier
in that case was 2.5.*
Given that there are time and resource constraints, it should be
apparent that very early in the planning process an inventory of desired
model runs should be prepared. The time and resource constraints should
be translated into an estimate of total model runs possible once the
constraints are known and the model has been selected. (As indicated in
Section 2, the constraints may influence the choice of the model to be
used.) The desired runs should then be ranked in priority for
execution. The resulting list may change over the course of the project
due to any number of factors such as computer problems, inaccurate cost
estimates, slipped deadlines, unanticipated political interest in a
particular control, etc., such that the need for flexibility and
periodic review of the list should be anticipated.
*The successful runs were divided roughly equally among the baseline,
sensitivity and control strategy applications described. Greater
emphasis would have been placed on the control strategy simulations if
not for time and budget problems encountered at the end of the planning
program.
-98-
-------
MODEL APPLICATIONS IN THE SAN FRANCISCO BAY AREA
The specific model applications made in the San Francisco Bay Area
included baseline forecasts, emission sensitivity analyses, and control
strategy simulations.
Baseline Forecasts
Forecasts of future ozone trends in 1985 and 2000 were made using
prototype meteorology for July 26, 1973, and emission inventories
projected and disaggregated for these future years. The results of
these model runs are summarized in Table 5-1. A sample isopleth map is
shown in Figure 5-3. The projections indicated that regional ozone
levels are expected to improve between 1975 and 1985, and that the
maximum level would be reduced by approximately 20%. This improvement
was attributed primarily to the effect of continued implementation of
the federal and California motor vehicle emission control programs.
Between 1985 and 2000, due to growth in population, motor vehicles, and
normal urban activities reflected in the projected emission inventories,
ozone levels were projected to deteriorate to about the 1975 levels.
Emission Sensitivity Analyses
To define the emission reductions needed to meet the oxidant standard,
1985 baseline emission levels were systematically reduced and then input
to the LIRAQ model. The results of the sensitivity analysis are
summarized in Table 5-2 and Figure 5-4, and led to two key conclusions
regarding control strategy design:
o Reduction of hydrocarbon emissions alone was more effective in
reducing ozone levels than the combined reduction of
hydrocarbon and nitric oxide emissions (NOx emission
reductions resulted in increased ozone levels within the study
area under the conditions being modeled).
o An approximate 43% reduction in hydrocarbon emissions in 1985
would result in attainment of the 0.08 ppm oxidant standard.
(Subsequent analysis indicated that an approximate 27%
hydrocarbon emission reduction in 1985 would result in
attainment of the 0.12 ppm ozone standard.)
Control Strategy Simulations
The effectiveness of alternative control strategies was determined by
applying a complete land use, transportation, emission and air quality
modeling system as illustrated in Figure 5-5. Each measure in the
strategy was translated into the appropriate variable and parameter
values or into an adjustment of the emissions inventory. The methods
for doing this are described below.
o Technological Controls. Technological controls were tested
with relative ease because they did not involve significant
changes in human activities. Rather, they involved the
-99-
-------
Table 5-1. Baseline LIRAQ Projections for the
San Francisco Bay Area
1975 1985 2000
Regionwioe High Hour Ozone (ppm) .17 .13 .17
(9.5 kms SEE of Livermore)
Ozone at Highest Station .13 .10 .15
(Livermore)
Table 5-2. LIRAQ Emission Sensitivity Analysis Results
% Reduction HC 0 20 40 60 80 40 80
% Reduction NO 000 0 0 20 40
Expected worst case
regionwide high
hour ozone (ppm) .19 .14 .08* .07 .06 .11 .06
*This value was rounded off from an original value of .0846 ppm.
Assumptions: 1) 1985 Baseline Emission Inventory
2) July 26, 1973 Prototype Meteorology
-100-
-------
Figure 5-3. Example LIRAQ Results - 2000 Baseline Ozone Projections
July 26, 1973 Prototype Meteorology (1500 Hours PST)
-101-
-------
Figure 5.4
PLOTS OF UNADJUSTED AND ADJUSTEC?REGIONWIDE HIGH HOUR OZONE AS A
FUNCTION OF % REDUCTIONS OF 1985 HC EMISSIONS
a.
a.
8
o
^
o
I
o
Ol
O>
3
55
I
0.14
0.13
0.12
0.11
0.10
(.189)
: Adjusted High Hour Ozone
: Unadjusted High Hour Ozone
EPA Standard (0.08 ppm)
Natural Background
(0.03-0.05 ppm)
80%
Percent Reduction of 1985 Baseline Hydrocarbon Emissions
*Adjusted values are obtained by multiplying model results by factors
that account for worst-case scenarios and imperfect model performance.
Such adjustments are discussed in Chapter 6.
-------
cu
3
cn
*
IT
•!
*
![
ri.
iiii
-103-
•V
1
I
-------
implementation of improved techniques for reducing the
pollutant emissions resulting from normal human activities.
Such emission reductions were accounted for by applying a
percentage reduction factor to the "emission factors" used in
the emissions models. For example, requiring even more
stringent control of motor vehicle emissions than currently
required was reflected in future motor vehicle emission
factors. This served as input to the emissions calculations
which subsequently were input to the LIRAQ model. Regulations
for controlling volatile organic compounds or for implementing
combustion modifications to reduce nitrogen oxide emissions
from small industrial and utility boilers were handled
similarly.
o Transportation Controls. Transportation controls were tested
through the travel demand modeling system. Depending on the
specific nature of the controls, different approaches to
simulating their effects were taken. For example, the effects
of a general regionwide improvement in transit service were
tested by changing the transit travel time or "wait time" in
the modal split model. This produced an estimate of the
percent of total trips diverted to transit and produced a net
decrease in highway network traffic. Testing service
improvements in specific areas involved changing the transit
network to reflect the improvements. Cost
incentives/disincentives such as a gasoline tax or increased
parking costs were simulated in the modal split model.
o Land Use Controls. The effectiveness of individual land use
control mechanisms could not be tested by the forecasting
system in a straightforward manner. What could be tested were
the ultimate objectives of land use control measures. For
example, one policy goal of land use control for improving air
quality was to halt the outward spread of the metropolitan
area boundaries and redirect future growth into existing
urbanized portions of the region. The effectiveness of
specific mechanisms or tools which might be employed to
accomplish this result (e.g., tax incentives/disincentives,
public facility restrictions, changes in general plans and/or
zoning ordinances) could not be tested by the forecasting
system. Instead, the system was used to test the effect of
accomplishing that "compact development" policy goal on
regional air quality. The land use policy goal in effect
became an assumption for a subsequent reiteration of the ABAC
forecasts. The results of these forecasts were then fed
through the modeling sequence to produce estimates of
resulting air quality. The information thus obtained was used
to evaluate the air quality effects of a more compact
development pattern in the region.
Land use controls or objectives were the most difficult and
time-consuming to forecast. This was due not only to the
difficulties in developing clear statements of the policy
-104-
-------
goals, but also the fact that changes in the ABAG demographic
forecasts necessitated additional runs of the subsequent
travel demand, emissions, and air quality models.
A summary of the control strategies tested with the modeling system is
presented in Table 5-3. The schematic flow diagram of the modeling
system and how alternative strategies or sensitivity analyses were input
are shown in Figure 5.5. The main results of the strategy analysis are
summarized in Table 5-4. The table indicates that substantial
improvements in air quality can be made through the use of source
control technology. The transportation and land use management
strategy, although relatively ineffective in the short term, is shown to
become increasingly effective with time.
MODEL APPLICATIONS IN THE DENVER METROPOLITAN AREA
Model applications in the Denver Metropolitan area also included
baseline forecasts and emissions sensitivity analyses. No actual
control strategy simulations have as yet been made using the SAI urban
airshed model.
Baseline Forecasts
Forecasts of future ozone trends were also made for 1985 and 2000.
Prototype meteorological conditions for two separate days were used in
the baseline forecast. The results of these forecasts are summarized in
Figure 5-6. Hourly ozone concentrations predicted for grid squares
containing monitoring stations are plotted for each year. The area
between results for successive years is shaded, with hatching between
1976 and 1985, and cross-hatching between 1985 and 2000. In general,
reductions are greater in the 1976 to 1985 period than in the 1985 to
2000 period. Reductions of peak levels averaged 44% in the earlier
period and 23% in the later period. (Additional baseline forecasts were
subsequently made for 1982 and 1987.)
Sensitivity Analyses
A series of sensitivity tests were made of the effect of spatial
redistribution of emissions in the Denver region. Emissions were
reduced in various geographic sectors of the region to determine the
impact of emissions from each area on ozone levels. The main conclusion
was that perturbations in emission rates of 25 percent in different
communities had no effect on region-wide ozone concentrations.
Two additional sensitivity analysis cases were run which simulated (1) a
30 percent reduction in both hydrocarbon and NOx emissions and (2) a 65
percent increase in NOx emissions. The conclusions from these tests
were:
o A 30 percent reduction in all emissions resulted in only a 15
percent drop in the predicted ozone concentration.
-105-
-------
fT3
i^
CO
o
u
CO
• 1—
g
ro
s-
u_
c
IB
00
o
.c
4J
c
TJ
QJ
I *
to
o>
r—
to
QJ
•r-
0)
4J
»C
U
ifr *
00
I
%
o
o
l*-
o
£
UJ
1—
r—
00
UJ
REHENSIV
a.
5:
o
0
QJ
to
QJ
J=
QJ
CX
E
O
LO
00
O"i
r-l
>^
CQ
o
.
UJ
oo
:D
Q >-
ea; uj
—i i—
^^ ^ 0
J_l ? •*-* "1
ro JE Qj"-^ —
*- QJ X X
CO 4_> QJ QJ rO
CD QJ QJ
t- -I-1
C 03 CD
-"Z
i- S_
ro ex 3
ex o
QJ
CD u CO
1 — ro c co
ro t_ fl. 3
C 3
o o o
'£,.£ c ro
cu-o ro
OQJ-
.i *J CO
, QJ CO
^ro
O c
,M •"" o:
QJ
OJ > QJ
s- QJ x:
3 "O -M
CO
QJ O -r~
"M
c: E -rj
0 0 3
•r- O r-
4-J O
ro <+- c
l ,A ^J *^" '
0 WJ -g
to u c
c QJ a,
ro 4- si
J_ t- QJ
4-> QJ X
QJ
CD
U
ro
r—
ro
to
X QJ
ro QJ
4J *4-
CD CD
c c
t- i.
to ro
CX CX
i i
QJ U
E O
>'Z*- •
CO
O QJ tO
-C tf^ QJ
QJ -r~ 4->
U ^£
c c c
tO -r-
CO •"
^ ^ CD
00 X c
r-l QJ >
c, A)
t- 'XJ »—
0 S-u
t»- 14- ro
co
r— •
CD O
C 0
•r- CX
^ C
co j_ ro
QJ CX
C r— CO
QJ ro r—
(J •-- 0
4-> O
CJ> C CX
C QJ i.
»p— t. ro
CX QJ 0
O O) t_
J= S- 0
to cm-
QJ QJ
> -C
00 M-
CX
U 3
-C5
QJ 3
4-> U
C
•r-
o >,
O CD
o w
^» +•>
co to
o
4->
CO
C
ro
u
^ >
^-
ro
c
o
•r—
*^"
"O
rovide
ervice.
CL. to
O
cal control measures
the exhaust control
•<- »-
+j
3
QJ
CD
•r—
t/>
•r—
x
QJ
C"
CD O O
O t|-
i— CO
O 4-> QJ
C CX U
SZ QJ -r-
U O >
OJ X
4-> QJ
T5
C
ro
CO
QJ
C
ro
r—
P»
O
o
CX
t-
ro
^*>.
to
3
.Q
ncrease
t— 4
o
QJ
10
•
a>
c
f_
^
QJ
1
CX
ro
i.
rucks (this measure
short term benefits
d all of the land
rtation measures.
^uj P* ^^
to ro °r
OJ QJ g
.~ -a .- ro
to p ^- "Jy
ro i- c co
CD CX O 3
c to u>"O
•'"tO C
QJ c ro
§"~ «- w
N "> . ro >>
r" -^ -^ C
0 l»- •" 4-)
1- •— W- >P-
4J ro ro QJ
c j. J- -o >r>
O 4J 4-> "
O r— m
«-» C OJ m *
£ «tJ "° S^
e ,7 -r- ••- .1- o
OJ "- i. > > O
r~ — -P O L CX
§• ro 2 ° « §
•— i oo -o a. w> >
o o
QJ
^_
u
o
•r—
J3
QJ
>
>rH
to
0)
•M
X
QJ
0)
i-
o
E
a. •
o ^
QJ jj
> to
Q tO
O
^J S_
c ro
Q) QJ
g.*
O QJ
"ttJ +J
>
QJ >>
•vS
*J C
0 O
(O •!-
CX CD
§£!
u
(1)
£5
o
6 +J
'!! O CD
•5^8
^ 4-> CSJ
o
PO
If)
ro
to
UJ
Q£
00
O
O
O
UJ
to
CD
QJ
>
ro
S-
0
C
ro
•o
c
ro
QJ
i- o
QJ
4-> CO
ro "o
Q>
o>
CO
03
QJ
CD
ro '
3
0 U
C
co 03
E
to u_
>» o
to
L.
^M-
co !2
•i- S-
ro 03 _ r—
ro
ro-
ro j_
ro
S 0
QJ p
•- ,_ O
ro
>> co
CD
o en
r- C
OJ
OJ
C 4J
QJ 3
QJ
o ro
E
ro
to
rs
o
QJ QJ
r- QJ >>
ro o ti
ro
OJ
£•-£
gSiS
,_ Q. S_ U
e
o
CO
3
ro
.c
X
QJ
QJ
QJ
CD
C
•— >r- •"- to
•Or— E EC£OJ 0)
< "—• Q) >—I T- CX > Of
O O O
-106-
c t-
O 4**
-------
OJ
s_
•>< o K'S
!
.
Is
*• O «»" *
. ox o*
8>^.fc-
W "Q
1*5*5
41
is
s*I
•13- *
1
tons/
cmUt
tons/day)
twitted
I
torn/day
*
I
I
1
I
C
i
t
I
1
|l
P
*5 M
tarn/day
£
|1
* I
!i
I !
1
t
? •
II
-107-
-------
5 6
6 7
7 8
8 9
9 10 11 12 1
To TT TZ T 2
f
4
I
6
Time of Day, By Hourly Interval
("start hour"!
Lstop hour J
E3 Reduction 1976-1985
Bffi Reduction 1985-2000
(a) Meteorology for 28 July 1976 Assumed
Figure 5-6. REDUCTION IN PREDICTED OZONE CONCENTRATIONS
(pphm) AT DENVER STATIONS DUE TO PREDICTED
FUTURE EMISSIONS CHANGES
-108-
-------
_. . _ „ u , T *. i ("start hourl
Time of O.y, By Hourly Int.rva! [stop ,,„, J
Reduction 1976-1985
(b) Meteorology for 3 August 1976 Assumed
Figure 5-6 (concluded)
-109-
-------
o A 6b percent increase in NOx emissions resulted in up to 60
percent reduction in the peak ozone concentration within the
modeling grid.
SUMMARY OBSERVATIONS
Both the LIRAQ application in the San Francisco Bay Area and the SAI
model application 1n Denver fell somewhat short of the guidance being
offered for model applications. In San Francisco, only one prototype
day was used throughout all model applications while in Denver the
strategy and emission sensitivity tests were severely limited. Efforts
are being made in both regions to improve their respective analyses. In
part, the guidance presented in this section is derived from a
retrospective evaluation of these two model applications.
EPA is sponsoring additional sensitivity studies for both the SAI model
and LIRAQ. (The results of additional LIRAQ sensitivity tests are
documented in Volume III of this report.) The guidance offered here
should be modified as additional experience in model applications is
gained.
-110-
-------
REFERENCES
Anderson, G.E., et al., "Air Quality in the Denver Metropolitan Region
1974-2000", Systems Applications, Inc., prepared for U.S. Environmental
Protection Agency, Region VIII, EPA-908/1-77-002, May 1977.
Association of Bay Area Governments, Bay Area Air Quality Management
District, Metropolitan Transportation Commission, "1979 Bay Area Air
Quality Plan", Berkeley, California, January 1979.
Wada, R. Y. et al., "A Methodology for Analyzing Alternative Oxidant
Control Strategies", Journal of the Air Pollution Control Association,
Vol. 29, No. 4, pp 346-351, April 1979.
-Ill-
-------
6. MODEL INTERPRETATION
Both Lagrangian trajectory and Eulerian grid models are designed to
replicate the physical conditions of the atmosphere and the processes
which affect pollutant concentrations. This approach differs
substantially from previously applied rollback techniques, and leads to
practical problems when attempting to interpret model output with
respect to the ozone standard. The key problems are summarized as
follows:
o The "prototype meteorological days" used in the model
performance evaluation may not be the "worst case" days which
have occurred, particularly if special field studies are
relied upon to collect supplementary data for use in the
preparation of model inputs. Since the plan must demonstrate
attainment of the ozone standard, which in turn is now
statistically defined, some method for relating the
effectiveness of control strategies simulated on the prototype
days to their effectiveness in meeting the ozone standard must
be developed.
o Since no model can be expected to precisely replicate
conditions which occurred on a specific day, it may be
expected that the model's performance will not be perfect.
Therefore, demonstrating ozone levels at or below the ozone
standard on the simulated prototype days does not necessarily
demonstrate that attainment would occur on the actual days.
o The specification of initial and boundary conditions for
simulations of future air quality is a potential problem for
cases in which substantial changes in emission distributions
and/or levels from the validation case are being tested.
THE "WORST CASE" ISSUE
There are three basic options for resolving the "worst case" issue. The
first option is to use a "worst-case" prototype day to evaluate the
model's performance in the first place. This, however, is frequently
not possible due to a lack of adequate meteorological and air quality
data for such a day. The second option is to ignore the issue and
presume that the prototype conditions being modeled are reasonable
representations of worst case conditions. Methods for extrapolating
from prototype conditions to worst case conditions are not well
developed at this time, and may introduce additional uncertainties into
the modeling analysis. If substantial changes in the magnitude and
distribution of precursor emissions are being simulated, it may be
argued that the conditions that led to "worst-case" ozone levels on a
particular day may not have the same effect under such altered emission
conditions. In fact, it is extremely difficult to define in advance
what meteorological conditions would lead to "worst-case" ozone levels
under emission conditions that are substantially altered from historical
-113-
-------
patterns. In short, the second option is to avoid introducing
additional sources of uncertainty into the modeling analysis.
The third option is to develop a relationship between the prototype days
being modeled and the estimated second-worst day. One straightforward
method is illustrated in Figure 6-1. Here, the ozone data for the year
from which the prototype day(s) was selected is plotted as a standard
Larsen-type log-normal distribution. The positions of the estimated
second worst day as well as the prototype day may be found on this
distribution. Using the same slope of the distribution, the
model-predicted ozone level for a given simulation may be used to plot
the future year distribution under the emissions conditions being
tested. The new distribution can then be used to estimate ozone levels
on the second-worst day for that simulation. Here it must be assumed
that the distribution of days remains intact in the future. Under this
assumption, the somewhat elaborate procedure just described reduces to
taking the ratio of the prototype day ozone levels to second-worst case
ozone levels, and presuming that the ratio will remain constant in all
cases simulated.
This option provides a simple method for extrapolating model output to
"worst-case conditions." However, there are a number of conceptual
problems with this method. First, by focusing on the maximum level as
the measure of air quality on a given day, a great deal of the
information being generated by the model is wasted. Second, since the
maxima in the future will likely be displaced in both space and time
from their original positions, it is likely that the position of a given
day on a statistical distribution would also change. Third, portions of
the region which may experience small increases in ozone levels as a
result of a given strategy cannot be accounted for. Other statistical
extrapolation methods could conceivably be developed in the future to
overcome these problems.
It may be seen that there are problems associated with each of the
options for interpreting model output vis-a-vis the ozone standard.
However, a regulatory determination must be made concerning the
acceptability of the plan, and its compliance with the requirements of
the body of existing laws and regulations. This interpretation of model
results is a cornerstone to that determination.
IMPERFECT MODEL PERFORMANCE
There are two basic options for dealing with imperfect model
performance. The first option is to accept the model results as they
are, making no attempt to "correct" or "calibrate" the results according
to ambient monitoring data. If more than one prototype day is being
used, and any systematic bias in model results is eliminated during the
model performance evaluation phase, then no after-the-fact calibration
or correction is necessary. On the other hand, it may not always be
possible to eliminate all systematic bias or to obtain satisfactory
results for several different prototype days, particularly if schedules
are constrained by regulatory deadlines. In such cases, some method
-114-
-------
en
c
c
o
o
O)
o
0>
CT
0)
1
D
I
o
0)
I
0)
T3
2
ao
"jjj h_
2 co
o «>
Q.>
ceo
•C 3
8-
CO
I
g>
LL
•- CD
2§
«> N
J) O
Q-%.
Eni
CO
CO CO
(O o
O 0)
-C Q.
E -
5 o
i 2
x a
o
LU ,
i»
i§
x o CD
I li
w c tl
' - c
Us o §
1 E o UL r:
<0 CD
"D >,
-115-
-------
should be considered for appropriately compensating the results to
ensure that control requirements are not either understated or
overstated. For example, a calibration or correction equation may be
used as was done in the San Francisco Bay Area example described later.
SPECIFICATION OF INITIAL AND BOUNDARY CONDITIONS
The specification of initial and boundary conditions for simulations of
future year conditions is an important problem for ozone modeling,
particularly when testing control strategy cases that would reduce the
simulated emissions and resulting ozone levels to levels at or near the
ozone standard. As emission levels are reduced, the contributions of
pollutant concentrations specified at the beginning of a model run
(initial conditions) and concentrations specified at the boundaries of
the model grid (boundary conditions), become increasingly important to
the resulting predicted ozone levels. /,s previously noted in Section 3,
concentrations of pollutants at the boundaries of a metropolitan area
are usually poorly known quantities in the modeling analysis—there are
very limited data that could be used for model validation on an
historical day, and virtually no data for future simulations that could
act as a guide. The following examples provide some hints for how to
proceed. However, it should be clear that additional research is needed
in this area, and that model users should make these assumptions
carefully and ensure their full documentation.
METHODS FOR MODEL INTERPRETATION USED IN THE SAN FRANCISCO BAY AREA
The LIRAQ application in the San Francisco Bay Area included the use of
an extrapolation to account for worst-case conditions. The existing
prototype days in the LIRAQ library were limited by the days for which
extensive supplemental monitoring data were available. Those days were
not the worst-case days according to historical monitoring data. The
maximum one-hour oxidant level of 0.18 ppm recorded on the day for which
validation results were best (July 26, 1973), was in the top ten for
that year as well as the upper 3 1/2 percent of the overall five year
distribution from 1970 to 1974. The expected maximum for that year,
however, was 0.24 ppm.
The procedure used to obtain a worst-case evaluation involved a
straightforward application of the Larsen model to relate the prototype
day to the worst days recorded for the year. Using the daily regionwide
high hour oxidant measurement to characterize each day, the distribution
of days for the year was developed in the standard Larsen format. By
knowing the position of the days according to the monitoring data, it
was possible to perform an extrapolation of LIRAQ results to determine
an expected maximum oxidant level on the "worse-case day." The
procedure involved the use of the maximum one-hour oxidant concentration
calculated anywhere on the LIRAQ grid to characterize projected air
quality for that day. The resulting adjustment factor was 0.24/0.18 =
1.33.
-116-
-------
An adjustment factor was also developed to compensate LIRAQ results for
an imperfect validation. Although it was amply demonstrated that LIRAQ
exhibited no systematic biases on two different validation days, as
shown in Figure 6-2, resource and schedule constraints limited the
control strategy analysis to a single prototype day. This left some
uncertainty regarding the effect of the control strategies under other
meterological conditions.
The formula for the adjustment factor was as follows:
Cf Cr C
f = s or C = m . Cf
where Cm= regionwide high hour ozone concentration
measured on the validation day
Cv= regionwide high hour ozone concentration
reproduced by the model on the validation day
Cf= regionwide high hour ozone concentration
forecasted by the model under some future
emission scenario
Cs= regionwide high hour ozone concentration to be
computed and compared to the standard.
In other words, the ratio of the measured regionwide high hour ozone
concentration on a given validation day to the model-produced regionwide
high hour pxidant concentration was used to adjust forecasted ozone
maxima. This compensated the forecast for any inherent biases in the
model or input data. On July 26, 1973 the measured oxidant maximum was
.18 ppm, while the model-produced maximum was .17 ppm. The adjustment
ratio was therefore .18/.17 = 1.06.
Finally, the problem of specifying initial and boundary conditions for
future year simulations was addressed as follows:
o Initial conditions for hydrocarbons and nitric oxide were
factored up or down proportionally to the change in the
aggregate regional emission inventory for each pollutant. In
addition, all simulations were initiated in pre-dawn hours to
minimize the initial concentrations of secondary pollutants.
o Since the prototype meteorology being used consisted of a
prevailing onshore wind, flow through the upwind boundary was
assumed to contain background levels of all species both
during model validation and during future year simulations.
-117-
-------
Figure 6-2
LIRAQ verification for two 1373 prototype days using
1975 emissions, based on 16 hourly values at 15 locations.
020
Q.
Q.
~ 0.10
o
T3
O
RMS Error - 0.026 PPM
Correlation Coefficient = 0.77
Sample Size - 226 station-hours
July 26, 1973
• I ,
,::,;•• :•«,;•:
•il'"'!1 '"••
ji!.:-;:
0.10
f
Observed Ox (PPM)
0.20
0.08
Q.
CO
O .0.04
RMS Error -- 0012 PPM
Correlation Coefficient - 0 71
Sample Size - 225 sta hrs
August 20, 1973
• i
« i
I .
I •]
0.04
Observed Ox, (PPM)
008
-118-
-------
o The vertical boundary condition at the base of the temperature
inversion was originally defined to depend partially on the
concentration of pollutants in the grid cell below.
Therefore, as emission levels changed, the boundary condition
at the ceiling would change in the same direction. Since the
degree of vertical transport down from the inversion was
relatively small on the prototype day, no change in this form
of specification was considered warranted.
METHODS FOR MODEL INTERPRETATION USED IN THE DENVER METROPOLITAN AREA
The application of the SAI model in Denver was also constrained by a
limited number of days for which supplemental monitoring data were
available. Statistical analyses of model performance on the three
prototype days indicated no apparent systematic bias in the model
results at high ozone levels (see Figure 4-4). Since two prototype days
were used in subsequent applications, no correction for an imperfect
validation was used.
To estimate second-worst case ozone levels in 1982 and 1987, a
correction factor of 1.21 was developed based on the ratio of the actual
second highest hour ozone level and the high hour measured on the
prototype day. As in the previous SAI study, no correction for
imperfect model performance was made. Initial conditions were adjusted
for future year simulations by applying the ratio of total regional
emissions of each pollutant to the initial early morning ambient
concentration field. Since boundary concentrations were assumed to be
close to natural background values in all simulations, no adjustment of
boundary conditions was made for future year simulations.
The baseline result for 1987, including the worst case correction, was
slightly in excess of the .12 ppm ozone standard. For subsequent
evaluation of control strategies, linear rollback was applied to the
1987 emissions inventory and model-predicted ozone level to provide a
demonstration of attainment. For the 1982 SIP revision, further model
applications are planned to evaluate the efficacy of the total control
strategy.
CONCLUDING REMARKS
This report was prepared with the expectation that advanced
photochemical models will be increasingly used in the development of
State Implementation Plan revisions. The guidance and examples
presented are based on the limited experience currently available.
There are three general problem areas that can be anticipated by
potential model users:
o Modeling resources and expertise are diffuse. While there are
at least three different modeling groups within EPA, and
expertise can be found in various consulting firms,
universities, and certain state agencies, there is no single,
-119-
-------
authoritative and unbiased sour ;e for obtaining expert
assistance in implementing a model applications program. The
potential model user must be able to sort through the
literature, obtain advice from a number of different sources,
and then decide how to proceed. Moreover, each modeling
program will deal with a unique set of conditions, issues, and
constraints which will require the exercise of independent
judgment and coordination with all participating entities.
o There are no standards for model performance. While attempts
are being made to develop modelperformance standards, a
crucial unresolved issue is what to do in the event that a
model doesn't perform up to the prescribed standard? All
procedures for estimating emission control requirements to
meet an ambient standard developed to date involve a model of
the relationship between emissions and air quality. One model
might perform better than another in a given situation, but it
is often difficult to anticipate that in advance, and a
substantial investment of time and resources would be
necessary to find that out. Until substantial additional
experience is gained in model applications under a variety of
conditions, or until some reasonable alternative procedure to
modeling is developed as a default option, this issue will
remain unresolved.
o The validity of model interpretation methods is uncertain.
Worst-case corrections, corrections for imperfect model
performance, and assumptions concerning future year initial
and boundary conditions are complex technical and legal issues
in need of further research. Resource constraints may force
the use of such interpretation methods in many instances. In
each case, the validity of the method should be evaluated in
the context of the total model applications program, and its
relationship to the plan.
Despite the absence of substantive guidance in each of these areas, the
use of photochemical dispersion models will provide the firmest possible
technical foundation for SIP development, and will represent a
substantial improvement over previously used methods. As additional
experience is gained in model applications, the guidance that can be
offered should become more specific and useful to both modelers and
model users.
-120-
-------
APPENDICES
-------
APPENDIX A
PREPARATION OF METEOROLOGICAL INPUT FIELDS
FOR LIRAQ SIMULATIONS IN THE SAN FRANCISCO BAY AREA
The LIRAQ model has been used in several studies of air quality in
the San Francisco Bay Area. Because this area is characterized by complex
and changing flow and mixing-height fields, the preparation of suitable
meteorological inputs to the model is difficult and time-consuming. This
Appendix illustrates some of the factors that had to be considered in order
to prepare the necessary meteorological input fields.
Selection of days was originally largely determined by availability
of data from extensive field studies. Even though the permanent meteorologi-
cal observation network in the Bay Area is quite extensive, the mesoscale
circulations are sufficiently complex that the field studies were judged to
be essential in order to characterize the wind and mixing height fields in
the region. Later, after experience had been gained, it was possible to
develop prototype days for which there were no supplemental field
measurements.
Model Grid Regions. The 170 x 120 km region that can be studied
with the LIRAQ model is shown in Figure A-l. The MASCON meteorological pre-
processor code (see below) is limited to a maximum field of grid elements
of 65 by 65. The whole Bay Area can be treated with a 5-km grid. For
smaller (1 x 1 km or 2 x 2 km) grid elements, the region must be subdivided
into several subregions. The photochemical version of the model (LIRAQ-2)
is limited to a 20 x 20 grid (e.g., a 100 x 100 km area with 5 km grid
elements). The choice of a particular 100 x 100 km subregion depends
upon a) the area whose air quality is to be studied, b) the locations of
the source emission regions that affect the study area, c) the minimiza-
tion of boundary effects, and d) possible flow reversals whereby pollutants
might be transported out of the region and then returned later.
Meteorological Data Sources. Upper wind and temperature data were
available from the National Weather Service (NWS) Oakland upper-air station
(twice daily) and from San Jose State University. Surface data were gathered
from 85 permanent sites on designated field observation days. Most of
these sites provided surface wind and temperature data. The surface obser-
vation sites are operated by many organizations, including the Bay Area Air
Quality Management District (BAAQMD), various airports, the California
Division of Forestry, various military bases and private industry. Solar
radiation measurements were obtained from 9 BAAQMD monitoring stations.
In addition to all of these "quasi-perr.anent" data sources a large body of
supplemental information was collected from special field studies for some
of the prototype days that were developed. Vertical profiles of temperature
and several pollutants^obtained from aircraft spirals at several Bay Area
locations»were very useful in specifying the time- and space-dependent
inversion base height field and upper boundary conditions.
A-l
-------
1000-2000
2000-3000
• > 3000
TERRAIN HEIGHT (ft.)
O Calm
IM/Sec
A_ 5M/Sec
123" 30' 122"
Figure A-l. Observed and synthesized winds, 10:00 a.m. PST, 24 July 1974, used
as inputs to MASCON.
r\~~ t.
-------
The Mass-Consistent Wind Field Model (MASCON) The MASCON computer
code (Dickerson, 1978) generates mass conserving fields of wind velocity and
of inversion base height from observed and analyst-synthesized input data.
The code interpolates irregularly spaced input data to a regular grid and
employs an adjustment scheme based on variational analysis to the mass
fluxes (mean winds within the layer times mixing depths) to ensure agreement
with an appropriate form of the continuity equation.
MASCON was developed because visual interpretation of the meteoro-
logical fields is difficult, especially for the Bay Area's complex terrain
and changing meteorology, and therefore semi -objective procedures were
needed to simplify the analysis. Results are not always consistent with
expectations; consequently application of the mass-consistent data is usually
an iterative process. When a particular analysis reveals areas with unrea-
listic divergence or flow patterns, a reanalysis with additional or adjusted
synthesized wind or inversion data must be performed before proceeding to
simulate air quality.
When a constant density mixing layer is assumed, mass is not allowed
to accumulate or to deplete within a given zone (defined as a single grid
square extending vertical >y from the surface to the spatially and temporal-
ly varying inversion height). This constraint must be satisfied for the
transport calculations, which are based on the flux into and out of a given
zone. If this constraint is not satisfied, the mass gained or lost within
a zone as a result of an imbalance of the net fluxes can invalidate any
calculated transport of diffusion of air pollution concentrations.
Because LIRAQ is developed in a flux form, the appropriate equation
of continuity can be written as (Dickerson, 1978):
) + w = 0,
where H is the height of the inversion base above topography, u and v are
the east-west and north-south components of the mean velocity within the
mixed zone, respectively, and w is a vertical velocity. When observational
data (or data interpolated from observations) are used for h, u, v, and w
in a finite difference form of the above equation, a non-zero residual (e)
appears on the right-hand side. In some cases small values of e can produce
errors in mass fluxes which, for LIRAQ, are large enough essentially to
invalidate the transport, eddy diffusion, and chemical transformation
calculations. Therefore it is necessary to ensure that the e field over
the grid is sufficiently small that the resultant fictitious mass losses
or mass gains are acceptably small.
MASCON is designed to adjust the product of the mean wind and the
height of the inversion base above topography in such a way that the air
mass is conserved and the observational data are changed as little as possible,
The adjusted wind and inversion fields, prepared at three-hour increments,
are then used as input meteorological data for LIRAQ.
A-3
-------
Wind Field Inputs to MASCON. Observational data tend to be
concentrated in the populated areas and to become much more sparce in the
rural regions and at the outer edges of the domain of interest, particularly
over the ocean. Because the model needs wind data at each grid point,
it utilizes a Gaussian weighting (according to distance R of the observation
from the grid square) interpolation scheme in order to develop a regional
wind field from the relatively sparce observed data field. In some regions
such interpolation can lead to unrepresentative winds at some grid points far
from any observed wind vector. Thus, wind vectors at some remote locations
are best synthesized based on subjective meteorological experience and in
consideration of such factors as described below.
One of the first steps in synthesizing wind data is a determination
of the gradient wind over the Bay Area and an assessment of how this should
be reflected in the subinversion layer. In the absence of other information
this gradient wind may be used as an estimate of the wind around the
boundary of the model area. Variations of this estimate over the model area
should be made only while one is simultaneously considering the effects of
topography, the land and sea breeze regimes, and mass conservation in the
subinversion layer.
In order to prepare a reasonable pattern of observed and synthe-
sized winds such as that shown in Figure A-l,the analyst must be aware of
the nature of the mesoscale circulations in the planetary boundary layer
over the San Francisco Bay Area. These flows are influenced by coastal
hills and gaps and by the thermal contrasts that are set up by differential
heating over the ocean, bays and adjacent land areas. On most summer and
early fall days there is an elevated subsidence inversion with a layer of
marine air beneath. In the early morning,thermal gradients are weak and weak
drainage flows prevail out of the coastal valleys. In the afternoon the
marine layer advances onshore, driven by a sea breeze circulation that
results from the pressure difference between cool coastal waters and the
hot interior valleys. For a fully developed sea breeze condition there is
pronounced channeling of the winds through the coastal gaps and into the
interior valleys. Past statistical studies of local climatological flow
patterns (Smalley, 1957 and MacCracken, 1975) were found to be an invaluable
aid in evaluating mesoscale flow characteristics for preparation of synthe-
sized winds.
Inversion Base Height Inputs to MASCON. The topography of the
inversion base height is one of the most important data fields needed by
the LIRAQ model because it influences such factors as mixing depth, wind
profiles, pollutant concentration profiles, and the height to which topo-
graphy controls the flow. Factors to be considered and used as guidelines
in constructing regional inversion-base-height contours (for use as input
to MASCON) from limited amounts of observed data are described here.
Over the adjacent part of the Pacific Ocean, surface temperature
is essentially uniform and thus also is the depth of the marine mixed layer
(and therefore the height of the subsidence inversion). Inland, temperature
gradually departs from temperatures over the ocean as the trajectory
distance increases from the ocean. This difference, due to solar heating
and radiational cooling of the land surface, undergoes a diurnal cycle
with the land temperature exceeding the ocean temperature in late afternoon.
A-4
-------
At this time, if it is not completely destroyed, the inversion will reach
its maximum height. Conversely, the land will cool below the ocean
temperature at night, and a new low-level radiation inversion will form
that will reinforce the subsidence inversion.
Because of radiation effects, elevated surfaces act as heat sources
during the day and heat sinks at night, thus enhancing the diurnal temperature
cycle and its effect on the inversion base height. For modest topographic
features the inversion-base-height contours should tend to parallel topo-
graphy contours unless overpowered by the effect of trajectory distance
from the ocean. Furthermore, when the onshore flow becomes moderate-to-
strong, the air tends to retain the temperature and inversion-base-height
characteristics of the oceanic source region. Thus, in the absence of
other effects, inversion-base-height cfntours will tend to parallel air
trajectories. In general, in regions where there are strong horizontal
gradients in the inversion-base-height or flow fields, the streamlines
and inversion-base-height contours will tend to be parallel.
Before putting sparse wind and inversion base height data into
MASCON, it is necessary to have experienced meteorologist construct fields
of wind velocity and of inversion base height (at three-hour intervals
throughout the period to be simulated) in order to allow additional synthe-
sized data to be provided to the interpolation scheme. Examples of these
fields are shown in Figures A-l and A-2. Because of this need for competent
human intervention in a region as complex topographically as the Bay Area,
the initial processing of wind and inversion base height fields is quite
time-consuming, requiring several person-weeks of effort by an experienced
meteorologist. In a less complex region, this task could be greatly
simplified.
Once the wind and inversion base height fields, at three-hour
intervals, have been prepared and put ^nto MASCON, an interative process
begins. First, MASCON interpolates the wind and inversion base height
fields -for each grid cell averaged over three-hour periods. Graphical
output is also produced, as shown in Figures A-3 and A-4. Next, MASCON
adjusts the fields until the flow fields are mass conserving. Figure A-5
shows the adjusted (mass-consistent) flow field derived from the inversion-
height and wind fields shown in Figures A-3 and A-4, respectively. Finally,
the meteorologist studies the adjusted flow field and if it appears
unrealistic in any way, he modifies the input fields and runs MASCON again.
The procedure may be repeated several times before acceptable mess-
consistent flow fields are obtained.
Preparation of Fields of Atmospheric Transmissiyity. The photo-
chemical module of LIRAQ is very sensitive to the variation of the photo-
dissociation rates. These rates are proportional to a wavelength-dependent
photon flux density. The "clear sky" photon flux density at the earth's
surface was calculated as a function of zenith angle. This value was then
multiplied by a transmission coefficient representing the ratio of Eppley
pyranometer measurements to calculated clear sky transmission. Thus the
transmission coefficient is a measure of solar flux obstruction by clouds
and aerosol.
A-5
-------
San x
Francisco
MAPORTHE
SAN FR4NCIS&O BAY AREA
1000-2000
2000-3000
> 3000
TERRAIN HEIGHT (ft.)
Figure A-2.
123" 30' '22° 30
Inversion base height field (in meters above sea level), 10:00 a.m. PST,
24 July 1974, constructed as a guide for generating input values to MASCON,
A~ u
-------
1300
Figure A-O.
MASCON -genera ted Interpolated field of mean inversion
base height above topography (meters), averaged over
three hours (10:00 to 13:00 PST), 24 July 1974.
A-7
-------
tJOO.
iZOO. 4-
«ico. •»-
»OG.
5?0. MO,
Figure A-4, MASCON-qenerated interpolated field of surface winds,
averaged over three hours (10:00 to 13:00 PST),
24 July 1974.
A-8
-------
«ioo •
100.
fio. »»o. MO too
•I) )u',rrn r i u» r i»t n
Figure A-5. Mass-consistent flow field derived from the interpolated
inversion base height and wind fields shown in
Figures A-3 and A-4.
A-9
-------
The transmission coefficient fields were estimated from pyrano-
meter measurements made st nine San Francisco Bay Area stations operated
by the BAAQMD. These transmission coefficients were averaged for the
three-hour intervals over which the LIRAQ model assumed uniform wind fields.
Analysis over the mountainous regions of the Bay Area was a prob-
lem due to the absence of data there. Because the transmission coefficient
probably increased with elevation, it was decided that the mountainous
regions above 2000 ft and covering a significant area would be bordered by
a contour two to four units higher than the surrounding lower elevation
region. The analysis of the August 20, 1973 maps was further complicated
by the presence of stratus just offshore and over portions of the Bay.
This stratus failed to appear over any of the pyranometer stations, and
consequently no estimates of transmission coefficients typical of such cloud
cover were available. To handle the situation transmission coefficients
representative of stratus overhead for each 3-hour period were determined
by comparison with pyranometer data from San Francisco on a day with
stratus. These were then used in the August 20 analyses. A sample plot
of the regional fields so developed is presented in Figure A-6.
References
1. Dickerson, M.H. (1978), "MASCON - A mass consistent atmospheric flux
model for regions with complex topography," JL Appl. Meteor.,
Vol. 17, No. 3, pp 241-253.
2. MacCracken, M.C., and G.D. Sauter, eds. (1975), "Development of an
Air Pollution Model for the San Francisco Bay Area, " UCRL-51920,
Vol. 2, Lawrence Livermore Laboratory, Livermore, California.
3. Smalley, C.L. (1957), "A Survey of Airflow Patterns in the San
Francisco Bay Area," Preliminary Report, United State Weather
Bureau Forecast Center, San Francisco, California.
A-10
-------
4 .; 4-H
MAP OF THE
SAN FRANCISCO BAY ARE A
WITH 5KMUTM GRID
Strafus: tt ^ 0.40
N ^
,-rv" •( •-,(_
0.60 ""0.64 0.68 0.72
Figure A-6. An example of the transmission coefficient fields prepared
for MASCON input. This analysis represents conditions
observed on 20 August 1973, 13:00 to 16:00 PST.
Source: MacCracken and Sauter, 1975.
A-ll
-------
APPENDIX B
PREPARATION OF METEOROLOGICAL INPUT FIELDS
FOR SAI SIMULATIONS IN DENVER AND LOS ANGELES
The SAI Airshed Model has been used in several air quality
studies in the Denver and Los Angeles areas. Each of these
regions has its own distinct topographical, meteorological and
air quality characteristics. This Appendix illustrates some
of the problems encountered by the modelers and describes how
these problems were resolved. In each region, as experience
with the model increased, it was possible to refine some of
the methodologies to produce more realistic meteorological
input fields and thereby improve model performance.
1. DENVER SIMULATIONS
Selection of days to simulate was based on episode days
on which specialfield studies had been mounted to collect
additional air quality and meteorological data. On one of
these days a frontal system moved through the region, causing
air pollution levels to be lower than expected. The maximum
ozone concentration observed during the summer of 1976 was
27 pphm; however, supplemental meteorological data were not
collected on this day. Instead, three days including a peak ozone level
of 17 pphm were selected because supplemental field measurements were
available for these days.
The modeling grid in the original Denver modeling work
reported by Donnely (1978) was a 30x30 mile portion of the
Denver metropolitan area, using 1x1 mile gridded emissions.
Later, Anderson, et al. (1977) found that simulations using
2x2 mile grid squares produced nearly identical results.
Anderson, et al. also noted that pollutants were sometimes
advected out of the modelin-g region and then returned later
when the wind reversed. To contain the Denver pollutant
"cloud" within the model, later simulations carried out by
Reynolds et al. (1979) used a 60x60 mile area. However, few
meteorological observations were available in the outlying
areas of the expanded grid.
Meteorological data sources included upper-level winds
and mixing depths derived from twice-daily radiosondes from
the National Weather Service (NWS) upper air station at
Stapleton International Airport. The NWS also provided
hourly measurements of surface winds, surface temperature,
B-l
-------
humidity, visibility and cloud cover. Hourly surface winds
were obtained from 24 sites operated by the Colorado Division
of Highways (CDH), the Air Pollution Control Division of
the Colorado Department of Health, the NWS and various private
organizations. Total solar radiation data (hourly averages)
were collected at one CDH site. Hourly surface temperatures
were obtained from 5 monitoring locations.
Wind field inputs to the Airshed Model were estimated
from hourly surface observations. Data from up to 24 surface
monitoring sites were used to construct hourly estimates of
the wind speed and direction for each ground level grid cell.
These wind fields were generated by the inverse-distance-
weighted interpolation scheme described by Liu, et al. (1973).
Because there were relatively few wind observations in some of
the outlying areas of the grid, it would otherwise be necessary
to extrapolate using the measured values in the interior of
the grid region. To maintain some control over this extrapola-
tion process, additional "fictitious" stations were defined
and assigned wind speeds and directions based on an analysis
of the actual observations and the local terrain features.
The resulting surface wind fields were smoothed by replacing
the wind velocity in each grid cell with the vector average
of the wind velocities in the block of nine cells (in the
Denver city area) or 25 cells (in outlying areas) centered at
the cell of interest.
In the original Denver simulations, winds aloft were
set equal to the surface wind field. In more recent simula-
tions, a procedure developed by Killus, et al. (1977) was
employed which produces a wind profile consisting of a mean
velocity that is a function only of height and a component
resulting from local convergence or divergence effects.
Mixing depths were assumed to vary only as a function
of time because there were insufficient data with which to
characterize spatial variations in the mixing depth field.
Hourly values of the mixing depth were established using the
two temperature soundings made at Stapleton International
Airport and available surface temperature observations. During
the predawn hours on both simulation days, the early morning
(5:00 a.m.) temperature sounding indicated the existence of
both a surface and an elevated inversion layer. Based on an
examination of the temperature records, it was estimated that
the surface based inversion was completely eroded by 7:00 a.m.
and 8:00 a.m. on 29 July 1975 and 28 July 1976, respectively.
After these hours, mixing was confined to the height of the
B-2
-------
elevated inversion base. To estimate this height, at each hour
during the day, the surface temperature aloft was extrapolated
at the vertical gradient exhibited in the afternoon temperature
sounding. This methodology predicted that the elevated inversion
layer was broken up by about noon.
Other meteorological variables included hourly relative
humidity and ambient temperature data obtained from the NWS
station at Stapleton International Airport. These values were
used to derive estimates of the water vapor concentration.
Vertical temperature gradients, both below and within the
elevated inversion layer, were derived from the available
radiosonde observations. The temperature at the inversion
base was estimated by extrapolating the surface temperature
aloft at the gradient indicated in the afternoon temperature
sounding. Hourly values of the exposure class (used to estimate
vertical diffusivities) were derived from estimates of the
diurnal variation of solar insolation. The values of photolysis
rates were increased by 15 percent to account for Denver's
elevation of over 5000 ft, above sea level. Finally, the
atmospheric pressure input was set at 0.85 atm.
2. LOS ANGELES SIMULATIONS
Prototype day selection. Two ozone episodes that
occurred in Los Angeles during the summers of 1974 and 1975
provide the basis for carrying out the Airshed Model simulations
for this large metropolitan area. During the 26-29 June 1974
episode, a land-sea breeze regime prevailed—a common occurence
in the South Coast Air Basin (Kieth and Selik, 1977). At night,
pollutants were advected out over the water; with the ensuing
morning sea breeze, some of this material was reintroduced
into the air basin. The 26 June meteorological conditions
resulted in the occurence of high (34 pphm), but not extreme,
peak ozone levels. These conditions are representative of a
"typical" smoggy day in Los Angeles. A simulation was also
made of 27 June, when the highest ozone concentration (49 pphm)
observed during 1974 occurred.
On 4 August 1975, during another high ozone episode,
onshore winds prevailed throughout the entire 24-hour period.
In fact, analysis of the synoptic meteorology and local wind
station measurements revealed persistent onshore winds for the
two days preceding 4 August as well. Selection of this day
allowed evaluation of model performance for a characteristically
different set of meteorological conditions.
B-3
-------
An important consequence of the unusual meteorological
regime on 4 August 1975 is the greater ease of preparation of
the three-dimensional wind fields and initial and boundary
conditions for that day compared with the 1974 episode days
(Tesche and Burton, 1978). To facilitate the specification
of boundary conditions for the 1974 episode days, it was
necessary to include as much of the Pacific Ocean as possible
in the modeling region to enable the treatment of offshore
pollutant transport. For the 4 August simulation day, boundary
conditions reflect essentially "clean" marine air, unperturbed
by local anthropogenic emissions sources. In this case it was
not necessary to include grid cells over the Pacific Ocean
in the modeling region, since there was no significant amount
of offshore transport.
Selecting the modeling region. Before discussing
the procedures used to generatethe various model inputs, we
first consider the definition of the modeling region. At the
time the simulations reported here were carried out, the
computer programs that composed the Airshed Model would only
accommodate a region subdivided into about 900 grid squares.
Gridded emissions data were available for an 80x40 array of
2x2 mile grid squares covering the area shown in Figure B-l .
As a result, it was necessary to model only a portion of the
large 160x80 mile region. After considering the locations of
the highest measured ozone concentrations and the need to
include portions of the Pacific Ocean to treat offshore pollut-
ant transport, a 36x25 array of 2x2 mile grid squares was
selected, encompassing the area indicated in Figure B-l.
B-4
-------
cr>
O)
oo
ex.
UJ
o
X
1—^
s:
-------
Meteorological data sources included upper level
winds at Riverside, El Monte, Los Angeles International Air-
port (LAX), Pt. Mugu and San Nicholas Island. Mixing depths
were determined from radiosondes from LAX, El Monte, Pt. Mugu
and San Nicholas Island, from an aircraft spiral at Riverside
and an acoustic sounder at El Monte. Hourly surface winds
were available from up to 60 stations operated by the county
air pollution control districts, the California Air Resources
Board (CARB), the NWS and the Pacific Missile Test Center
(PMTC) at Pt. Mugu. Hourly surface temperatures and humidity
were available from 15 stations. In 1974, hourly radiation
measurements were made at LAX, downtown Los Angeles and
Riverside. In 1975, radiation was measured at El Monte and
Riverside. Visibility and cloud cover data were available
from seven airports.
Hind fields. To ascertain how model performance is
influenced by the procedure employed to generate wind inputs,
Reynolds et al. (1979) prepared three different wind fields
for the 26 June 1974 simulation day. Starting with the surface
and aloft wind measurements, three different methods were used
to calculate a three-dimensional wind field. Two of these
methods, (1) interpolation using the station values, and (2)
use ofathree-dimensional wind model produced acceptable results.
o Interpolation Technique. Preparation of the three-
dimensional wind inputs required by the Airshed Model involved
several steps. Wind data from 60 locations in the Los Angeles
Air Basin were obtained from the California Department of
Transportation, the CARB, and the local air pollution control
districts. The 34 stations within the modeling region and
the stations just outside of the region were used to construct
a surface wind field using an inverse distance weighted
interpolation scheme as described by Liu, et al. (1973). The
resultant wind field was smoothed to eliminate any sharp
changes in wind speed and direction. The smoothing algorithm
replaces each grid value by the five point average about that
gri d eel 1 .
Winds aloft were estimated using the three wind soundings
made in the morning--at El Monte, Riverside, and LAX--and two
made in the afternoon--at El Monte and LAX. These measurements
taken aloft were used along with the ground-level measurements
to estimate an upper level "synoptic" wind velocity for each
hour. This wind velocity was assumed to determine completely
the wind field above the mixing layer.
For grid cells situated between the surface and the
top of the mixed layer, the winds were determined by interpola-
tion, using the information from the surface field and the
B-6
-------
synoptic wind. In making estimates, it was assumed that the
divergence in the wind is surface-generated and that this
divergence dies out with height, being completely dissipated
above the mixing layer.
o Three-Dimensional Wind Model. The three-dimensional
wind model used in this study if described by Yocke, Liu, and
McElroy (1977) and Tesche and Yc eke (1978). The model
computes horizontal flow fields at up to 10 equally spaced
heights, starting at sea level. These heights are measured
from sea level as opposed to ground level; thus, if a region
contains mountains at any particular level, no wind vectors
will be calculated in those areas. The flow fields for
each level are calculated by specifying the surface tempera-
ture field; surface roughness; average ground elevation for
each grid square; and the winds around the boundary of the
region, estimated from the available wind data.
The wind field obtained from this model is mass consis-
tent and highly influenced by topography. The general direction
of the flow for a particular level is largely determined by
the terrain configuration, surface heating patterns, and
general features of the synoptic level flows. In this study
it was difficult to establish the wind flows aloft for each
hour because instantaneous upper air soundings were made at
a maximum of six stations only once or twice a day.
In the Airshed Model, the z-direction has been trans-
formed so that all heights are measured from the ground rather
than from sea level. This means that the output from the
wind model must be transformed so that the wind vectors are
specified at the nodes of the Airshed Model grid cells rather
than at the equally spaced heights above sea level used by
the wind model. Consequently, some of the features of the
original wind field are lost; most importantly, the transposed
horizontal wind field is not mass consistent. A suitable
vertical wind velocity is calculated within the Airshed Model
at the time the pollutant calculations are carried out which,
in effect, renders the wind field mass consistent. For Los
Angeles the final wind field of this three-dimensional model
reflects the channeling of the wind through the mountain
passes more realistically than does the interpolated wind
field model.
Mixing Depths. Special consideration must be given to
the preparation of mixing depths in Los Angeles because the
behavior of the inversion over that area is complicated by
B-7
-------
the differing heating and cooling rates of the air over the
land compared with that over the ocean. To account for this
complexity, the modeling region was divided into subregions,
as shown in Figure B-l . The subregions were drawn parallel
to the coastline in conformity with the notion that inversion
height is reasonably constant along lines parallel to the
Los Angeles coastline (Edinger, 1959). The data that were
available to estimate mixing depths were morning and afternoon
soundings at LAX and El Monte, a morning sounding at Riverside,
hourly acoustic soundings at El Monte, three daily soundings
at Pt. Mugu, and hourly surface temperatures at these five
locations. These observations were used to estimate hourly
mixing heights for LAX, El Monte, and Riverside.
The LAX values were used for Subregion I, the Riverside
values were used for Subregion IV, and the grid squares along
the border of subregions II and III were given the El Monte
values. The mixing depths for Subregions II and III were
then interpolated using the two closest assigned values.
The mixing depths for Subregion V were also interpolated,
using the closest subregion on the east and the Pt. Mugu
value on the west.
After every grid square had been assigned a mixing
depth according to the above scheme, the mixing depth field
was modified in the vicinity of significant terrain features.
The scaling factor is represented by the ratio of the mixing
depth to the sum of the ground elevation and the mixing depth.
This factor is designed to have a pronounced effect in reduc-
ing the mixing depth at high ground elevations, but a
relatively small influence as the mixing depth decreases.
The values of the mixing depths at measurement sites were
altered so that, upon application of the scaling factors,
the measured mixing depths were obtained.
Once the mixing depth has been determined, the height
of the modeling region can be calculated. The modeling
region consists of six layers of grid cells, with the first
layer being the "surface layer" with a constant height of
60 feet. At any point in the region, the other five layers
are equal to each other in height. For the time period
5:00 to 8:00 a.m. PST, four of these five layers were within
the mixed layer and one was above; for the time period 8:00 a.m.
PST to the end of the simulation, all five of the layers were
within the mixed layer.
B-8
-------
01her Meteoro1og1ca ^Variables. Temperature and
humidity were measured at "five stations within the modeling
region. The observations for each hour were averaged, so
that a single temperature and a single water vapor concentra-
tion were used as inputs to the model. Temperature gradients
both above and below the inversion base, were used to char-
acterize atmospheric stability and to estimate plume rise
from large point sources. These inputs were established
from the available temperature soundings.
B-9
-------
References
1. Anderson, G.E., et al. (1977), "Air Quality in the Denver Metropolitan
Region: 1974-2000," EPA-908/1-77-002, Systems Applications,
Incorporated, San Rafael, California.
2. Donnelly, D.E. (1978), "Oxidant Model Applications: Denver," 57th
Annual Transportation Research Board Meeting, January 1978,
Washington, D.C.
3. Edinger, James G., (1959), "Changes in the Depth of the Marine Layer
Over the Los Angeles Basin," _J. Meteorol.. Vol. 16, No. 3, pp. 219-226.
4. Kieth, R.W., and B. Selik (1977), "California South Coast Air Basin
Hourly Wind Flow Patterns," South Coast Air Quality Management
District, El Monte, California.
5. Killus, J.P., et al. (1977), "Continued Research in Mesoscale Air Pollu-
tion Simulation Modeling - - Vol. V: Refinements in Numerical Analysis,
Transport, Chemistry, and Pollutant Removal," EF77-142, Systems
Applications, Incorporated, San Rafael, California.
6. Liu, M.K., et al. (1973, "Further Development and Evaluation of a Simu-
lation Model for Estimating Ground Le/el Concentrations of Photochemical
Pollutants - - Vol. Ill: Automation o; Meteorological and Air Quality
Data for the SAI Urban Airshed Model, " Systems Applications, Incor-
porated, San Rafael, California.
7. Reynolds, S.D., et al. (1979) , "Photochemical Modeling of Transportation
Control Strategies, Vol. I. Model Development, Performance Evaluation,
and Strategy Assessment," EF79-28, Systems Applications, Incorporated,
San Rafael, California.
8. Tesche, T.W., and C.S. Burton (1978), "Simulated Impact of Alternative
Emissions Control Strategies on Photochemical Oxidants in Los Angeles,"
EF78-22R, Systems Applications, Incorporated, San Rafael, California.
9. Tesche, T.W., and M.A. Yocke (1978), "Numerical Modeling of Wind Fields
over Mountainous Regions in California," Conf. on Sierra Nevada
Meteorology, American Meteorological Society, 19-21 June 1978,
South Lake Tahoe, California.
10. Yocke, M.A., M.K. Liu, and J.L. McElroy (1977), "The Development of a
Three-Dimensional Wind Model for Complex Terrain," Proc. of the
Conference on the Applications of Air Pollution Meteorology,
28 November - 2 December 1977, Salt Lake City, Utah.
B-10
-------
APPENDIX C
Preparation of Emission Inventory For Photochemical Modeling
In the San Francisco end Denver Regions
The LIRAQ source inventory for the San Francisco Bay Area is made up
from four component parts, each part compiled with independent data
sources and techniques. The four components as shown in Figure C-l are:
major point sources, area sources, airports, and mobile sources. Major
point sources include oil refineries, electric utilities, chemical
industry, metallurgy, rock and mineral operations, etc. -- any
stationary source emitting more than 0.1 ton/day or 25 tons/year of any
pollutant. Such sources are listed separately in the existing source
inventory, with information on location, emissions, stack parameters,
operating schedules, and process variability. Area sources, also called
"population-distributed" emissions, include: domestic fuel combustion,
off-road mobile sources, utility engines, and small stationary sources
such as service stations, dry cleaners, small plastic manufacturing,
etc. Emissions are estimated by a variety of techniques including
direct measurement, natural gas use, solvent sales, gasoline sales.
point and resin use, etc. Airports include emissions from commercial.
military and general aviation aircraft from 37 airports in the Bay Area.
Table C-l presents the relative contributions of the four source
components for the 1975 baseline inventory. It is clear that mobile and
area sources are the largest organic emission contributors. These two
categories are also the most complex for emissions estimates end spatial
and temporal resolution. Spatial and temporal distribution techniques
used in the Bay Area study are discussed below, for each component of
the inventory.
Major Point Sources
Data concerning emissions from more than 120 major industrial operations
in the Bay Area are maintained as part of BAAQMD's source inventory.
Data maintained include: UINi coordinates, stack height, actual
measurements of stack gases, operating hour, seasonal variations, and
fuel usage records. Based on these records, the total emissions from
major point sources were distributed into appropriate UTM grid cells.
hajor point sources are carried as separate listings and are further
divided into "surface" and "elevated" categories, depending on the stack
height. (The dividing line is 100 feet.) The elevated major point
sources remain as individual listings in the final source inventory
file, QSOR. Surface major point sources are merged with emissions from
the other three source inventory components.
The hourly and seasonal variations of major point sources are generally
known. Because these are large individual emitters, they are subject tc
intense scrutiny from regulatory agencies. And by the nature of large
operations, they usually maintain reliable internal process records.
Based on operating hours and seasonal variations, hourly emission
distributions for each major point source in the Bay Area were
constructed.
C-l
-------
Figure C-1
Summary schematic of QSOR^file preparation.
rour source inventory componem
Maior Point Sources
located on UTM grid
\ 2
/
Link . 3
Population ^ 4
5 '
Elevated
Surface
Airport
-
Mobile
Irmm
Area
7
N
1 — 7
\
/
wown MIC
Elevated
OSOR
disaggregated
source inventory
1,2 or 5-Km grid,
24-hr variation,
input tor LIRAO
I
QSOR is the code name for the final disaggregated source inventory file
which serves as input to the Livermore Air Quality Model.
C-2
-------
TABLE C-l
1975 Source Inventory for San Francisco
Bay Region by Source Categories
\Jype
SourceN^
CategoryN^
Major Point Source
Area Source
A i rport Sou rce
Stat i ona ry Tota 1
Mob Me Source
Tota 1
Part. Org. NOX S02 CO
( tons/day )
19 119 167 187 74
98 412 101 12 338
9 20 13 1 55
126 551 281 200 467
43 472 399 19 3,869
169 1,023 680 219 4,336
Source: BAAQMD
C-3
-------
Airports
The locations of 57 airports in the Bay Area were identified by UTM
coordinates. Because some landing and take-off emissions are spread out
within the mixing layer, airport emissions were distributed over
neighboring grid squares. For commercial and military flights,
emissions were distributed uniformly over all grid squares within 2
miles of the airport. For general aviation at community airports, a
distance of 1 mile was used.
Aircraft emissions are divided into three source inventory
classifications: commercial carriers, military, and general aviation.
Commercial carriers are most important from an emission viewpoint, and
these are also the best documented. Comprehensive schedule books
provide up-to-date listings of commercial flights to all the major
airports of the world, with arrival end departure times and aircraft
types. For the LIRAQ project, data were compiled on commercial carrier
operations at Say Area airports from the 1975 "North American Air
Guide." Temporal resolution factors for three commercial airports (San
Francisco, Oakland and San Jose) were based on these data.
Operations data for four military air bases in the Bay Area were very
limited and also difficult to predict. Emission estimates were based on
fuel usage data, but actual daily flight schedules were not available.
Basec on available data, hourly operations for four military air bases
were estimated to be 90% during daylight hours and 10% during night
flights.
Finally, general aviation emissions were uniformly distributed over
aayl ight hours for 28 small community airports, and extended to include
some early morning and late evening flights at busier airports. Hourly
traffic counts on airport approach roads were used to check the diurnel
pattern of general aviation.
Area Sources
Among the 107 activity classifications in the BAAQMD source inventory,
58 include some area source contributions. Emissions in 1975 from area
sources comprised about 58% of total particulstes, 29% of organics, 15%
of nitrogen oxides, 6% of sulfur dioxide, ano 8% of carbon monoxide
emitted in the Bay Region.
Ihe method developed for spatial resolution of area sources in the Bay
Region was a "cross classification" technique. The objective was to
postulate functional relationships between source categories and 5
variety of demographic, economic, and land use variables being used as
surrogates. A table of coefficients was compiled to link 58 source
activity classifications (those with area source components) with 19
known employment categories from ABAG's "Series 3 Projections." (See
Hoffman, et al., 1978, and Perardi, et al., 1979).
C-4
-------
The Series 5 projections cover population, housing, employment and land
use in the Bay Area. For the nine counties around San Francisco Bay,
the data are compiled for 440 sub-regional areas termed "zones," which
are made up of one to approximately seven 1970 census tracts. Housing
is recorded by dwelling unit, end population/employment by 23
categories. The information was based on census data, local surveys,
fertility and immigration statistics. A list of the Series I variables
used in this project is provided in Table C-2. Since the Series 3 data
were internally consistent between the 1975 base year and future year
projections, use of this data ensured that the area source emissions
projections and spatial distribution vould also be consistent.
Before they could be used as a basis for are source distribution, the
Series 3 data had to be distributed over the 1-km UTM grid system. This
critical step was accomplished by a combination of manual and computer
techniques. First, regional maps were used to eliminate those grid
squares which are essentially uninhabited. Those areas (bays,
tidelands, marshes, mountains, etc.) comprise about 75% of the total
area. Series 3 variables were then distributed fron1 440 zones to the
remaining grid squares, which total 5000 to 6000 km2 of developed or
developable land. The exact total depends on the year being considered.
A cross-classification table was then developed to link certain types of
area sources with appropriate Series 3 variables. For some source
classifications a direct correspondence could be found. For example,
BAAQMD source category no. 18 "Farming Operations" could be linked with
Series o employment category P7 "AGRI" which includes agricultural
production and services. Similarly, source classification no. 40
"Printing" could be distributed with Series 3 "MFG1" which is printing,
publishing and related industries. In most cases, however, the source
classification did not fit clearly with a single Series 3 variable. For
these cases, professional judgment was employed to produce a multiple
distribution formula, so that area source emissions from a single source
classification could be distributed with two or more Series 3 variables.
For example source classification no. 25 (Degreasers) provides area
emissions of 42 tons/day of organics. These were distributed as
follows: 60% with MF&5 (fabricated metal products), 20% with RET. SERV.
(including auto repairs), 10% with MFG4 (including electrical and
optical equipment), and 10% with OTHER SERV. (including local transit
and transportation services). Excerpts from the classification table
are shown as Table C-3. The percentage values were selected based on
knowledge of local industry ano operating conditions.
Area source emissions were distributed and then totalled for each Series
3 category (for each pollutant). Totals were divided by the known total
population of the category to produce a per capita emission rate. As an
example, for the 394 tons/day of organics for area source distribution,
the total for Series 3 category P9, from all source classifications, was
20.75 tons/day. The total employment population in P9 (printing end
publishing) was 25170, so the per capita emission factor was .00082
tons/day of organics per printing publishing employee. The per capita,
emission rates, for each Series 3 category and each pollutant, were then
used with the known Series 3 population distributions to produce the
C-5
-------
I
U
c
0
1-t
4J
cx
•H
O
in
0)
o
c
0
1-1
4J
cfl
& u
0 i-t
r-l UH
C/3 -H
cn
CO
CO
r-l
U
QJ
r-l
J3 OJ
CO Q
•H cfl
n z
CO
QJ
T3
O
U
QJ
rH
JO
•2
CO
>>
r-l
4J
cn
01
V-<
O
cfl
U
•H
r-l
CX
cx
CO ON
1
4J r*^
O
C -
rJ
rJ M
5 o
o <
r-l t— •
CX, CX,
C
0
1-1
4-1
CJ
cfl
^J
4J
X
QJ
cn
CO
60
"° 60
rH C
1-t 1-1
0 f
m
* .,-4
r*"N ^
V-* *O
V-i 3
cfl CX
3
cr -
60
- c
00 -H
C 4-1
•H C
C i-l
i-l Vw
^J
r-t
CO
t— (
O r^*
rH
z o
M &
CO ON
CX, IX
cn
OJ
-H
V-i
4-1
tn
3
•a
c
«tH
r-t
CO
QJ tO
E u
o
- 3
v-i -a
ai o
a. u
co a.
cx
•a
* 0)
• r^
e -o
0) C
x: -H
0 ^
- X)
• c
r~~ t Cfl
O
u -a
4J C
QJ O
f^ [l^
CO
*
CN
CO
•s
ON
*o
O
O
*
rH
0)
V-i
cfl
cx
cx
CO
f
tn
OJ
i— i
4J
X
QJ
H
ON
CO
*
rH
CO
.
in
CN
CN
CM
sO
O
*j
rH
CX
tn
OJ
C rH
i-l co
i-t H
OJ U
a QJ
1-1 4J
CX Cfl
^ B
/-* 60
0 C
4J 1-t
y TJ
0) rH
1 *^
C 3
O X)
c
^-* *
OJ
e -a
o «
•H V-<
4-J 4-1
(0
4J QJ
I-l rH
O CO
cx in
C rH
Cfl O
V^ i—
H 5
vO
•^
1
»^
<}•
CN CN
^3- in
0 O
o- vn
Z rJ
< O
gg
in so
rH rH
cx, cx,
OJ
CJ
c
fO
u
3
cn
c
•H
*
rH
Cfl
1-t
O
cfl
C
•H
f^.
\O
CO
vO
CN
z
M
Cm
r%.
rH
CX,
U)
QJ
*T-t
U
QJ
i— 1
rH
'i
»
tn
§
0)
tn
tn 3
OJ E
u
•H •
> Q)
1- 0
QJ -H
cn ^
>j
tn Q)
tn cn
QJ
•H U
tn 3
3 -0
CO [d
ON
CO
^j.
CO
CO CN
I~^ 00
r-l CM
J^ ^
ai ai
td td
to to
00 ON
rH rH
cx, cx.
4J
c
01
B
c
QJ
>
0
CJ
CN
&*
1-t
>
0
CJ
o
CN
cx,
en
OJ
u
o tfi
4J 01
cn o
i-t
"O r»
O L
O OJ
u-i w
c
QJ 1-1
cn e
1-t "T3
T3 cfl
C
CO -
J= i-t
O cfl
l-i 60
QJ OJ
C rH
i— t *
co x:
01 rH
C cfl
OJ 01
u a
sO
ON
i— t
ON CO
in
ro O
m co
•
&
td
t/5
•
H c/)
aD
PQ
rH CN
CN CN
DM CX,
60
C
•r-t
V
C
cfl
^n
to -
t-i 0)
i-l OJ
eg i-t
f\ ^j
QJ -H
V-I rH
iH
•• 4-1
QJ 3
CJ
*vH ••
r* *-*
oi cn
cn c M
CO QJ
rH V-l Xi
Cfl U 4-1
c o
o -
tn c «
V-i O 0)
01 -H 4-1
CX 4-1 CO
O 4J
• 3 tn
cn i-i QJ
0) Cn rH
•u C CO
o o QJ
a: u v-i
o
sO
- ON
ON ON
-3"
1
ON r^ in
f— ^S- ON
1 1
m -co
r~. rH ON
-»
CS • SD
r>» r^ sO
rH
O m rH
1^. rH SD
> as :
pcj td
td OO
to
&.
• td ;
Si
co -j-
CN CN
cx, cx
c
OJ
ON
CO
C
CO
C
o
m
cn
co
r-l
u
cn
•o
c
Cfl
•a
c
«
4-1
to
C-6
-------
c
O
—
4-
j
JQ
—
1_
4-
to
—
13
_-
(0
.—
4-
ro
Q-
co
L_
0
vi-
ta
0)
CO
3
0
—
j3
(O
4-
C
O
.—
4-
rt>
O
.—
•4-
.—
to
to
ro
—
0
1
CO
01
O
!_
O
0
x:
4-
e
o
i_
s-
to
4-
Q.
1_
0
O
X
LU
ro
1
O
0
—
-Q
ro
t—
•
tO
c
o
—
co
CO
—
E
0
0
o
1_
3
o
to
ro
0
L.
ro
*+-
O
i
Xl
£
O
u
4-
x_x
co
c
o
•—
CO
co
. —
E
0
0
U
!_
3
O
to
ro
0
L-
ro
0
x:
-t-
t-
O
0
O)
rtj
4-
c
0
O
u
0
Q.
0
x:
4-
-f-
C
0
i/l
0
l_
a.
0
L.
to
0
3
—
ro
>
TJ
0
—
x>
ro
I—
ro
CO
0
—
L_
0
to
T)
0
4-
ro
O
•—
~o
c
—
0 tO
-c CD
4— —
l_
x: o
4- O>
— 0
3. 4-
ro
x> O
0
4- ro
3
xi to
— 0
t_ •—
4- t-
CO 0
— CO
-a
0
_Q
O
4-
--^
c
O •
.— to
4- O
ro ro
— ._ —
0) L. O
ro
ro > 0
o
z:
>
i_
0
CO
.
1_
0
to
—
ro
4-
0
]p
«
Q.
•—
3
CT
0
.
O
O
U
a.
x:
CO
.—
. —
X)
3
Q-
4-
co
0
u
0
U-
co
4-
—
C
;3
.
O
—
en en o o
OJ O
— -
0 0 0
(Nl ITS VO
O O O
CN ~ —
O
O
—
O O
— o
—
o
o
—
o
o
—
• - -- '
— ON CN "^" C^ CN CT\
1 .......
v£) O O^ CN fO O LT\
— CT* "st — —
•
>
CO CD —
C • O) O O
O o ro co t-
•- O i- 0
4- t- o a.
ro ex 4- CT) •>
t- to c en
0 .- C
a. u 4- to —
O — > (0 I- C
t- — o 0 ro co
O) CO O U tO 0 C
c ro to ro — —
— \ tO 0 O 4-
E X> • 3 t- C C
1- O COT? C0>-— 3
roO'-c0L.l_ro
t-L-U-O — Q Q Q_ — 1
oo o\ c^ ~~ co ^o cu r^-
- — — c%4 rO fO hO T}" CO
0
J3
ro
•— •
u
ro
ro
to
0
—
L.
CP
to
cn
—
TJ
C
ro
to
c
o
~^
4-
ro
U
•—
^t-
—
to
to
ro
—
U
0
O
1_
3
O
to
ro
0
l_
ra
oo
to
to
ra
x:
0
—
XI
ro
4-
c
O
. —
4-
ro
u
._
-4-
._
to
to
ro
—
o
i
to
to
0
l_
o
—
—
3
4_
0
x:
1-
ro
-~^
•
—
ro
—
L_
0
to
4-
O
c
to
. —
en
c
~_
L.
0
Xl
e
3
2:
-~^
.
to
0
.—
1_
O
CT)
0
4-
ro
u
•
o
4-
0)
^
4-
C
0
e
>^
o
—
O-
£
0
*>
C
O
•—
4-
ro
—
3
a.
O
Q.
f-
o
to
C
O
_
4-
U
0
— j
O
[_
Q-
K^
CO
0
__
t_
0
to
CD
<;
CQ
<;
X)
0
>
ro
x:
to
4-
c
ro
4-
3
—
—
o
Q-
L
0
x:
4-
O
.
>-
ro
"U
_*:
0
0
5
l_
0
E
E
3
to
ro
1_
o
H-
-.
>.
ra
-a
•-^
to
rr
O
4-
^
to
C
o
.—
CO
to
.—
£
0)
to
o
._
c
ro
O)
U
O
0
U
t_
3
o
to
rO
0
!_
<
U
.
to
0
a>
ro
4-
c
0
o
i_
0
CL
c.
O
._
4-
3
X)
. —
!_
4-
to
. —
-a
0
E
ro
to
0
x:
4-
0
to
3
4-
3
X3
to
0
4-
r0
1_
C
o
_
to
to
._
e
0
4-
c
0
l_
0
•+-
M_
._
TJ
C-7
-------
area source spatial resolution. Results were checked by summing area
source emissions over ell grid squares and comparing the resulting total
with the total area source emissions used as the starting point.
Changes in the percentage values shown in Table C-4 do not change the
amount of area source emissions (as long as the entries sum across to
100%). Only the distribution of the emissions would be changed.
The hourly distribution of area source emissions is based on diurnal
variation coefficients (percent of total daily emissions per hour of the
day) for each source classification. Weighted hourly variation factors
were then produced by multiplying area emissions per classification
times the diurnal variation factors of each classification. The
resulting (normalized) set of factors were then used for temporal
resolution of all area sources emissions.
Mobile Source Emissions
In the San Francisco Bay Area, the highway, or "link" related,
hot-stabil ized emissions were computed using modified versions of the
two Federal Highway Administration computer codes SAPOLLUT and SAPLSM,
previously mentioned. The trip eno related emissions were computed
using programs developed at ABAG. The overall sequence of operation and
input data requirements and sources for both codes are summarized in
Figure C-Z. As shown, each set of programs outputs hourly emissions
that are geographically distributed by one kilometer UTM grid squares.
The two data sets are then merged tor input to the air quality model
(LIRAQ). For input, both codes require transportation data from the
Metropolitan Transportation Commission (MTC) and emission factors from
the California Air Resources Board (ARB). A summary of baseline
transportation data inputs is shown in Table C-5.
The motor vehicle emission factors used were derived through the use of
a California Air Resources Board emission factor program, EMFAC3. This
program was, in turn, based on EPA's Supplement 5* to AP-42, with some
minor modifications.
The EMFAC3 program computed hot-stabilized emission factors for HC and
MOx in units of grams per mile. It provided emission factor estimates
for average route speeds from 5 to 50 mph, ambient temperatures from 20
to 80°F, and any desired mix of cold and hot start operation. Factors
were produced for both a weighted average of four vehicle types (light
duty auto, light duty truck, heavy duty gasoline, heavy duty diesel),
and individully for each vehicle type.
*This supplement has been replaced by "Mobile Source Emission Factors,"
EPA-400/9-78-005, March, 1978.
C-8
-------
T ab I e C-4
AREA SOURCE DISTRIBUTION PERCENTAGES FOR SERIFS 3 ACTIVITY CATEGORIES
HTi rroTTTcT 1—I 7 ' H \' 1—! 10 V Tl 1 'I? ! 11 \ K J i', , ].,' | !/ i T^ , Tl , ^T ; 71 ' ; 77 ] 71 nmiKR LOT
-, j In-, ilf if at Ion [>.,.! 1 yn ' ?-1'.' Tl v i I vi-r •)• >rv- ' •;• r.V "rn',1 "i /• f,] | .y, I-,,T|OI i r] •;'•••• II' 'V ("V ,;n ';••', ",rry ;irT±"rl'VI IV ( >/,
]'-»• MI CAT.
I _ '!! •.. L Chcm^i'^.K A
t/i ;!.; u:u"r7f-b:r."
j ' !i iu i a 1-St out1 , Sard
._)_ .'• U1-3J
i •, i 'i lal^S.T i.! 'U,r I ,
'_•" ' '"'Wai X"1?-!
-II — «<.
_ 4_ -i Oi-"'/-AL.Eic_._Itloc_i
_j 'n_ii t .-..T.iv Mtbt
i -"fh .a
TLniL'iror»r.. jrar. EVAP.
!orflgc lanks -
"_J _l_-j' lj Cpatj^-Wa.ter^jliise.
i Com. f^iluni. c;o3l. . -
iUCU . . 4 .?0..^._1.3_.1(
__J ,VL-i:ic_Jj^Uj:iji_l!uia_
_ J .. i:£ Jj-ljl'dy" i II.IPITV
|_AiiJUiuil-i'J.ld i ir
ArclJcni ,V iTjo.1 ire:
v -F-I:W in
Afi'ic. Tractors -
i
ir.-cl
L rue.!. quip. -
U'-tl
T.-hipE .Carr.o/Pas.
n',liil>'. ,'h ' Itarv
,lTHoiu^r,'LpsJ"
cV Kloi or ^lilps -
nii!.c r
iM I'crryLonts
T,0 .. - 1 .
. Ui 11 1 tv EriRS.
: ' ' . '
50 ; j ! j . 3
\
N
I
y
/
/
z_
Llfifl.
TREA
- — -
r AC
POTN-
;
' SOlltCES
I
20
10
1
20 i
•\ ':
-* r "
IN SI
irrih
j
C LAf
1
10
10
ES
T?m
10
3 1 2 . "i
|
i
| |
1 ' 2 . ^
I
"3 I 10 1 3
U 4- L- — -*-- - -
*— ^- r r _ . . .
I 10
10
*Series 3 categories defined in Table C-2.
Source: BAAQMD
C-9
-------
TABLE C-5. SUMMARY OF BASELINE TRANSPORTATION DATA INPUTS TO
MOTOR VEHICLE EMISSIONS ESTIMATION FOR THE SAN FRANCISCO BAY AREA
PARAMETER YEAR
1965 1S75 1985 2000
AVERAGE WEEKDAY
VEHICLE TRIPS
o hom^based work
o Non-work
o LDV lota!
1,706,983
5,370,480
7,077,463
2,144,693
6,904,098
9,048,791
2,542,951
8,215,373
10,758,324
3,038,406
9,859,449
12,897,855
AVERAGE WEEKDAY
VEHICLE MILES
o HomSbased work 14,055,453 20.1S9.644 23,645,050 30,309,087
o Non-work 27,873,495 40,623,164 52,516,997 73,350,341
o LDV sub-total 41,928,948 60,822,808 76,162,047 103,659,428
c HDV @ 12.8% 5,366,905 7,785,319 9,748,742 13,268,407
o Total VMT 47,295,853 68,608,127 85,910,789 116,927,835
C-10
-------
ARR's EMFAC3 was the basis for computing the mobile source emission
factors.** however, a numb&r of variables, which vary with geographical
location and situation, can affect emissions estimates considerably:
average vehicle speed, ambient temperature, type of vehicle, percentages
of cold and hot start trips ano percen ; of travel by vehicle age (see
Figure C-2). Therefore, localized correction factors reflecting these
variables were developed based on Bay Area conditions.
Speed and ambient temperature correction factors were also developed
from formulas provided in EPA' s Supplement 5 to AP-42. Estimates of
link speeds and the distribution of vehicle types were provided by the
Federal Highway Administration. Assumptions for the diurnal variation
of average ambient temperatures were estimated from average summer
minima and maxima observed at different regional locations. The ambient
temperature correction factor was insensitive to temperatures above
80°F. Vehicle age distributions and pollution control equipment
deterioration rates (provided by EMFAC3) were also incorporated into the
emission factors.
Link Emissions
A coded highway network (for 1975 and updated for 1985 and 2000) and a
transportation model to forecast travel volumes on each link were the
basis for the link emissions calculations. As previously stated, a
modified version of SAPOLLUT (U.S. DOT, 1976) was used to actually
compute the link emissions, given the appropriate link information and
the emission factors. The modified model computed estimates of speed on
the highway network according to the volume/capacity ratio on each link,
for each hour. These speed estimates determined the appropriate speed
correction factor to apply.
Ihe model also provided diurnal traffic distributions and the
distribution of vehicle types on different road types. Five types of
vehicles were examined: light duty autos (LDA), light duty trucks
(LOT), heavy duty gasol1ne-powered vehicles (HDG), heavy duty
diesel-powered vehicles (HDD) and motorcycles. The distribution of
total VMT among the various vehicle types was obtained from ARB as
follows:
Vehicle Type Percent of Light Duty Vehicle
VMT (LDA + LPT)
LDA 86.2%
LOT 13.8%
HDG 8.6%
HDD 4.2%
Motorcycle 0.9%
**These factors were subsequently adjusted to incorporate EPA's draft
Supplement 8 factors (June 1977). The latest revision embodied in
EPA's Mobile 1 (March 1978) was not available in time to be used in
the analysis.
C-ll
-------
Figure C-2
Organization of the motor vehicle emissions code
A. LINK EMISSIONS COMPONENT
B TRIP-END EMISSIONS COMPONENT
MTC Link
Output
ARB / B
s
f
ABAG
VEM
~ >
t
A
L r—-^ AnAr. $~L
nr
j
BAGRID
N
f
Hourly /
UTM '
Gridded
Emlssons
D /ARB
E /SATS
LINK EMISSIONS
Required input data
A — State plane/UTM transform
B — Emission and deterioration (actors for
each model year for 1975, 1985, 2000
C — Motorcycle emission factors, SO? and
particulate emission factors for all veh-
icles, weighted average tor 1975,
1965, 2000.
D — Updated speed correction equations
for LDV. HDV, diesel
E — Percentage truck and motorcycle VMT
by hour and functional road type
TRIP-END EMISSIONS
Required input data.
A — Ongin-deitination trip tables for each
travel model run (including intrazonal)
B — Hourly distribution of trip starts by trip
purpose (or tour soak periods
C — Intrazonal VMT per zone
D —Cold start, hot soak emission (actors
for 1975, 1985, 2000 (weighted aver-
age over vehicle population)
E — Hot-soak period distribution
F — 440 zone/1 km grid conversion
ABAG ~ Association of Bay Area Governments
ABAGVEM = name of computer code with vehicle emission factors
ARB « California Air Resources Board
BAGRID ' computer code to distribute link emissions to grid squares
CALTRANS = California Departnient of Transportation
LIRAQ - Ltvermore Air Quality Model
MTC - Metropolitan Transportation Commission
OSOR - name of aource inventory file for LIRAQ model
SATS * Sacramento Area Transportation Study
UTM - Universal Transverse Mercator coordinate system
C-12
-------
Finally, the modified SAPLSM code read the hourly and total daily link
emissions for the entire network and performed the following:
o converted the plane coordinates of the MTC network to the UTM
(Universal Transverse Mercator) coordinates required for
LJRAQ.
o allocated the link hydrocarbon emissions into three
LIRAQ-defined reactivity classes
o wrote an output file in a format suitable for input to LIRAQ
Trip End Emissions
A separate computer program for estimating trip end emissions produced
hot start, cold start and hot soak emissions by zone and hour of day
(see Figure C-3). It also computed hot stabilized emissions for
intrazonal (i.e., within the zone) trips. The basis for the emissions
computations was a trip table which was developed by the transportation
model. From this table the program computed the number of trip starts
(i.e., origins) and stops (i.e., destinations). The emissions rates
were determined as a function of the parking (or shutdown) time before
and after a trip (for the start and stop tripends, respectively) and
whether the vehicle was catalyst or non-catalyst equipped.
A special study by the California Department of Transportation provided
parking time profiles by trip purpose and by trip origin or destination.
From these profiles more accurate estimates were made of the percent of
trip starts and stops experiencing hot and cold start and hot soak
emissions. The trip purpose generally differentiated between the
work-related and therefore long-term parkers and the non-work, and
therefore, short-term parkers (e.g., shopping, recreational). The trip
end identified whether the vehicle was starting or stopping at the home
zone and thus the likelihood of a long parking period.
In computing trip end emissions, four trip types were considered.
Considering work trips, for example, there are:
o the trip origins at the home (i.e., production) end, where
many cold starts take place in the morning
o the trip destinations at the work (i.e., attraction) end,
where many hot soaks take place in the morning
o the trip origins at the work (attraction) end, where many
evening cold starts occur for the return trip
o the evening destinations back at the home (production) end,
where hot soaks occur.
Note that the terms starts and origins, stops and destinations are used
interchangeably.
C-13
-------
Flowchart of trip-end emissions program.
TRIP END EMISSIONS
INPUT
Trip productions/attractions
for «ach purpose and zone
I
COMPUTE
Composite hot soak emission factor
for production and attraction ends
PRODUCTION END
ATTRACTION END
COMPUTE
Number of trip starts
by hour, purpose, zone
COMPUTE
Hot start and
cold start emissions
COMPUTE
Number of trip stops
by hour, purpose, zone
COMPUTE
Number of tnp starts
by hour, purpose, zone
COMPUTE
Hot soak emissions
COMPUTE
Hot start and cold
start emissions
1
OUTPUT
Total emissions for each pollutant by hour
COMPUTE
Number of trip stops
by hour, purpose, zone
COMPUTE
Hot soak emissions
C-14
-------
DENVER METROPOLITAN AREA
Stationary Sources
A general summary and comments on emissions inventory for the Denver
study are listed in the following section. Table C-6 presents major
assumptions made in estimating emissions from seven stationary sources
categories.
o Emissions data for stationary sources were compiled by the
Colorado Department of Health.
o Emissions data were collected by seven major source
categories: point sources, gasoline service, solvent users,
oil-based paint use, space heating, incinerators, airports.
o hourly emissions estimates were compiled on a 1 x 1 mile grid
covering 900 square miles. However actual model runs were
made with 2 x 2 mile grid.
o Automobiles and point sources are major emission sources in
the Denver Metropolitan region. Mobile sources account for
about 80 percent of the total hydrocarbon emissions, 30 to 40
percent of the total NOx emissions, and 90 percent of the
total CO emissions in the region. Point sources contribute
approximately 15 to 20 percent of the hydrocarbon emissions
and 40 to 50 percent of the NCx emissions.
o SAI's Urban Airshed Model used in the Denver study reouires
emissions of reactive hydrocarbons and nitrogen oxides to
determine the concentrations of photochemical oxidants. The
reactive hydrocarbon emissions are further split into
emissions of aldehydes, olefins, aromatics, and paraffins.
The most recent Urban Airshed model chemistry is a 42-reaction
"Carbon-bond" mechanism with four main classes of
hydrocarbons, designated as single bonos (paraffins), fast
double bonds (olefins except ethylene), slow double bonds
(aromatics and ethylene), and carbonyl bonds (aldehydes and
ketones).
o The "Denver Model" treats two types of sources: "ground
level" sources, whose emissions are injected into the surface
layer of grid cells, and elevated point sources, whose
emissions may be injected into any grid level depending on the
calculated plume rise. Point sources having stack height less
than 50 feet are combined with other ground level sources.
o Hydrocarbon emissions are assumed to be 75 percent reactive
ano 25 percent nonreactive. Nitrogen oxide emissions are
assumed to be 85 percent by weight nitric oxide and 15 percent
nitrogen dioxide.
C-15
-------
Method for Preparing the Niotor Vehicle Emission Inventory in the Denver
Metropol1 tan Area
In the Denver study, trip end emissions were lumped with hot-stabilized
emissions. Mobile sources were separated into twc files, "Auto Link"
for traffic that crosses traffic zone boundaries end "Auto Area" for
intrazonal traffic (traffic that remains within traffic zone
boundaries), the link file was created by the Colorado Division of
Highways using their transportation models. For the years 1974 and 1975
the inputs to the model were derived from traffic count data collected
in 1971. For future years, traffic loadings were estimated from the
Joint Regional Planning Program year 2000 land use plan, the Colorado
Dept. of Highways road projections and the Denver Regional Council of
Governments Empiric Activity Allocation Model for estimating types of
trips. Vehicle emission factors were estimated using EPA AP-42,
Supplement 5, predecessor of MOBILE1. A composite emission factor was
used which combined trip end emissions with hot stabilized link
emissions.
Since the emission of pollutants from automobiles and other traffic are
dependent on the typical cycles of operation, which vary with location
and type of roadway, the Division of Highways classified the links in
the Denver transportation system under eight roadway types and four area
types. For the years 1985 to 2000, however, only five different roadway
types were used in this classification system. Depending on whether the
operating conditions are peak (rush-hour) or off-peak, the Division of
Highways estimated average vehicle operating speeds for each of the
roadway types within each area type, as well as the diurnal variation in
traffic flow for the Denver metropolitan region.
Using the link positions and lengths (inputs to the transportation
models), the estimated average daily traffic (output of the
transportation models), the estimated emission factors and the speed
tables tor the various link types, the Colorado Division of Highways
estimated the Auto Link emissions as a function of the time of day and
location within the Denver highway Planning coordinate system. These
emissions were then apportioned to the appropriate 2 mile square grids
defined for the Denver area for application of the SA1 photochemical
mcdel.
For "Auto Area" emissions, traffic loadings and diurnal cycle were
estimated based on the demographic and geographic characteristics of a
grid square. Emission factors were also derived from EPA-AP-42,
Supplement 5. Once the Auto Area emissions were calculated for each
square and temporally resolved, they were added to the Auto Link
emission inventory.
C-16
-------
REFERENCES
Bay Area Air Quality Management District, "Base Year 1975 Emissions
Inventory, Summary Report," Bay Area Air Quality Management District,
San Francisco, California, 1976.
Bay Area Air Quality Management District, "Base Year 1975 Emission
Inventory - Source Categories Methodologies," Bay Area Air Quality
Management District, San Francisco, California, 1976.
Hoffman, S. R., E. Y. Leong, and R. Y. Wada, "Air Quality Plan
Development in the San Francisco Bay Region: Integrating Complex
Regional Models into the Decision-Making Process," presented at the
American Institute of Planners 61st Annual Conference, September 27 -
October 1, 1978, New Orleans, Louisiana.
E. Y. Leong and R. Y. Wada, "Emission Inventory Projections: Hindsight,
Insight and Foresight," proceedings of the Air Pollution Control
Associstion Specialty Conference on Emission Factors and Inventories,
Anaheim, California, November 13-16, 1978.
Perardi, T. E., M. Y. Kim, E. Y. Leong and R. Y. Wada, "Preparation and
Use of Spatially and Temporally Resolved Emission Inventories in the San
Francisco Bay Region," Journal of the Air Pollution Control Association,
Vol. 29, No. 4, pp. 358-364, April 1979.
U. S. Environmental Protection Agency, "Air Quality in Denver
Metropolitan Region, 1974 - 2000," Environmental Protection Agency,
Region VIII, Denver, Colorado, 1977.
U. S. Department of Transportation, Federal Highway Administration,
"Special Area Analysis - Part 4. Special Area Pollution," prepared by
L. R. Seiders, Comsis Corporation for FHWA Urban Planning Division,
Washington, D. C., August 1973.
U. S. Department of Transportation, Federal Highway Administration,
"SAPOLLUT/SAPLSM User's Guide," FHWA Urban Planning Division, HHP-23,
Washington, D. C., October, 1976.
C-17
-------
TECHNICAL REPORT DATA
(Please read Instructions on the reverse before completing)
« i -\V1 T NO
EPA-450/4-79-025
4 TIM I ANllSUHTITLE
2.
3. RECIPIENT'S ACCESSIOC*NO.
5. REPORT DATE
Application of Photochemical Models in the Development
of State Implementation Plans
.Volume..Ji_..Ihfi_ilSfi. .Qt Photocheniical .Models. An_-Urban Dxiftont.
7 AUTHORI'i)
Wada, Ronald Y., et al.
6. PERFORMING ORGANIZATION CODE
8. PERFORMING ORGANIZATION REPORT NO.
9. PERFORMING ORGANIZATION NAME AND ADDRESS
Association of Bay Area Governments
Hotel Claremont
Berkeley, California 94705
10. PROGRAM ELEMENT NO.
2AA635
11 CONTRACT/GRANT NO.
68-02-3046
12 SPONSORING AGENCY NAME AND ADDRESS
13. TYPE OF REPORT AND PERIOD COVERED
U. S. Environmental Protection Aqencv
Office of Air Quality Planning and Standards
Research Triangle Park, North Carolina 27711
14. SPONSORING AGENCY CODE
15. SUPPLEMENTARY NOTES
16. ABSTRACT
This document describes procedures for application of photochemical models in the
development of State Implementation Plans. Based largely on recent experience
gained in photochemical model applications in the San Francisco Bay Area and in
Denver, the guidance is directed toward potential model users in other ozone non-
attainment areas. The guidance covers the following tasks: model selection;
data collection and model input preparation including meteorological and topograph-
ical data, emission inventory data, ambient air quality data, treatment of initial
and boundary conditions, and special field studies; the evaluation of photo-
chemical model performance; model applications; and interpretation of model results
with respect to attainment of the Federal ozone standard.
KEY WORDS AND DOCUMENT ANALYSIS
DESCRIPTORS
Photochemical Modeling
SIP Development
b.IDENTIFIERS/OPEN ENDED TERMS
c. COSATI field/Group
13. DISTRIBUTION STATEMENT
RELEASE UNLIMITED
19. SECURITY CLASS (This Report)
Unclassified
21. NO. OF PAGES
20 •
22. PRICE
EPA Form Z2ZO-1 (9-73)
-------
>-no
j en 3
3^ ?
CD
C
O O Q.
"la
yi _ _
(COO
0)0)0)
H|"3?
QJ ,-.'
1 p2
"5-*°
0) ~ CP
—• ct
•D<
?; -o§
ill
«
B £ < ft) w
3 O ^ w 5)
n n O -ncn
-------
|