United States Office of Air Quality EPA-450/4-83-O19
Environmental Protection Planning and Standards August 1983
Agency Research Triangle Park NC 27711
Air
&ERA The St. Louis
Ozone Modeling
Project
-------
EPA-450/4-83-019
The St. Louis Ozone
Modeling Project
by
Henry S. Cole, David E. Layland,
Gerald K. Moss, Conrad F. Newberry
U.S. Environmental Protection Agency
Region V, Library
230 South Dearborn Street
Chicago, Illinois 60604
U.S. ENVIRONMENTAL PROTECTION AGENCY
Office of Air Quality Planning and Standards
Monitoring and Data Analysis Division (MD-14)
Research Triangle Park, NC 27711
August 1983
-------
DISCLAIMER
This report has been reviewed by the Office of Air Quality Planning
and Standards, EPA, and approved for publication. Mention of trade names
or commercial products is not intended to constitute endorsement or
recommendation for use.
U.S. Environmental Protection Agency
ii
-------
Contents Page
1.0 Introduction 1
1.1 Organization of the Report 4
1.2 Background 4
2.0 The Urban Airshed Model 7
2.1 Mass Transfer (Advection and Dispersion) 9
2.2 Emissions 11
2.3 Chemistry 12
2.4 Surface Removal Processes 15
2.5 Assumptions and Limitations 16
3.0 Methods and Data Bases 17
3.1 General Approach 17
3.2 The RAPS Data Bases 18
3.2.1 Emissions 18
3.2.2 Air Quality 20
3.2.3 Meteorological Data 21
3.2.4 Selection of Test Days 22
3.3 Preparation of Airshed Inputs 22
3.3.1 The Modeling Domain 22
3.3.2 Input Files 23
4.0 Model Performance Evaluation 27
4.1 Methods 28
4.1.1 Accuracy of Peak/Near-Peak Predictions .... 28
4.1.2 Overall Accuracy and Precision 31
-------
4.1.3 Replication of Patterns 34
4.2 Results of Model Evaluation Analysis 35
4.2.1 Results: Peak/Near-Peak Accuracy Analyses 35
4.2.2 Overall Accuracy/Precision. 39
4.2.3 Replication of Spatial and Temporal Patterns. ... 43
4.3 Potential Sources of Error 50
4.4 Summary of Model Performance 56
5.0 Model Sensitivity 58
5.1 Introduction 58
5.2 Test Days 61
5.3 Uniform Changes in Hydrocarbon Emissions 63
5.4 Uniform Changes in Oxides of Nitrogen Emissions 69
5.5 Uniform Changes ir Hydrocarbon Reactivity 74
5.6 Other Sensitivity Tests 79
5.6.1 Spatial Distribution 80
5.6.2 Temporal Distribution 81
5.6.3 Mixing Height 82
5.6.4 Photolytic Rates , 84
5.7 Discussion 85
6.0 Estimating Control Requirements 89
6.1 Response to Uniform Changes in Emissions 90
6.2 Control Strategy Simulations/Methods .... 91
6.2.1 RACT 91
6.2.2 FMYPC . 92
6.2.3 I/M - 94
6,3 Control Strategy Results 95
6.4 Uncertainties
iv
-------
Page
7.0 Conclusions 104
7.1 Conclusions on Model Performance 104
7.2 Model Sensitivity 105
7.2.1 Findings 105
7.2.2 Recommendations for Further Sensitivity Analyses 107
7.3 Conclusions with Regard to Control Requirements 108
7.4 Errors and Uncertainties 109
7.5 Feasibility of the Airshed Approach Ill
8.0 References 113
-------
1.0 Introduction
Refined photochemical grid models have a number of distinct advan-
tages over the simple techniques currently being used to develop control
strategies for most urban regions. They treat physical and chemical
processes more realistically and completely and allow for a more precise
spatial and temporal resolution of model inputs and outputs. The prob-
lem is that photochemical grid models require more intensive emissions
and aerometric data bases than do simple models such as EKMA. The col-
lection and preparation of data and the model are demanding and require
considerable expertise and resources. Moreover, the typical photochemical
grid model analysis will require 3 years or more.
It is therefore essential for EPA to conduct studies to determine
whether the accuracy and resolution of photochemical grid models warrant
the extensive efforts and resources required for their application.
Such studies will also enable the Agency to identify and resolve problems
and to provide guidance on photochemical modeling. To meet these objec-
tives, the EPA is conducting modeling studies in St. Louis, Tulsa,
Philadelphia, and Denver. The purpose of this report is to present the
findings of the St. Louis Ozone Modeling Study.
The model tested is a three-dimensional photochemical grid model
developed by Systems Applications, Incorporated, known as the Urban
Airshed Model (Airshed).1 This model numerically simulates the effects
of emissions, interurban transport of ozone and precursors, advection,
diffusion, chemistry and surface removal processes on pollutant concentra-
tions in a large number of grid cells.
-------
While several photochemical grid models are available, the Urban
Airshed Model has been tested most extensively. An alternative model,
LIRAQ,2 was also tested for St. Louis; however, preliminary results
indicated that LIRAQ contained technical and logisties problems and was
unable to produce significant ozone levels for St. Louis-3
St. Louis was selected for an extensive test program for several
reasons: (1) it is a moderate-sized city with periodic occurrence of
high ozone; (2) its terrain is relatively simple; (3) it does not lie
immediately downwind of other major urban areas in that ozone influx is
not a major complication and (4) the Regional Air Pollution Study (RAPS)
provided extensive air quality, meteorological and emissions data bases
for Airshed modeling.
It is emphasized that the project is not aimed at prescribing con-
trol requirements for the St. Louis Air Quality Control Region (AQCR);
its purpose is to serve strictly as a demonstration project designed to
increase the Agency's understanding of Airshed modeling and to provide
guidance to potential users of this model. The specific objectives of
the St. Louis Study are as follows:
1. to evaluate the ability of the Airshed Model to reproduce
observed concentrations and patterns of concentration;
2. to determine the model's sensitivity to important input
parameters;
3. to use the model to estimate the degree of control required
to attain the ozone standard and to test the effectiveness of specific
control programs;
4. to determine sources of uncertainty and error in the analyses;
-------
5. to assess time and resource requirements of Airshed applications;
6. to gain first-hand experience with extensive data preparation
and with problems likely to be encountered by users;
7. to compare Airshed performance and control estimates to that
obtained using simpler methods such as EKMA and Rollback.
While most of the objectives are self evident, several require
amplification. Evaluation of the model's performance (objective 1),
required the authors to identify a set of performance measures. We
basically used the procedures recommended by the American Meteorological
Society.4 In some cases we found that some of the recommended measures
were particularly useful while others were difficult to interpret or
not applicable to the case of grid models. Thus, an ancillary objective
for the St. Louis Study is to make recommendations on procedures for
model performance evaluation.
The sensitivity analyses (objective 2) were conducted to establish
the model's response (in terms of ozone concentration) to changes in
various input parameters. These tests are useful for identifying those
variables which result in the most significant change in model output.
Those parameters shown to be critical deserve the greatest attention in
data collection and preparation activities. Conversely, approximations
which save time and resources are warranted for those parameters to
which the model results are insensitive. Secondly, efforts to identify
sources of error should focus on the critical parameters. Similarly,
sensitivity analyses can contribute to an understanding of uncertainty
associated with the value used and (2) the model's sensitivity to the
input. The greatest uncertainty is introduced by those inputs which are
3
-------
difficult to specify accurately and which induce a strong response in
model output. While quantification is beyond the scope of this study,
the report does attempt to identify major sources of uncertainty in
the analysis.
1.1 Organization of the Report
The report is ordered as follows. Chapter 2 is a description
of the Airshed Model, its basic approach, its treatment of physical and
chemical processes, and its underlying assumptions; and limitations.
Chapter 3 outlines the basic approach used in the modeling analysis and
summarizes procedures used to obtain model input data from the emissions
and aerometric data bases. The results of base-case simulations and the
results of the statistical evaluation of model performance are given in
Chapter 4. Chapter 5 presents the results of sensitivity analyses and
Chapter 6 describes the control strategy analysis. Chapter 7 summarizes
the main conclusions of the study and makes recommendations on modeling
procedures and inputs, the need for additional work and when it is appro-
priate, to use advanced photochemical grid models as opposed to simpler
techniques.
1.2 Background
Two organizations; within EPA have contributed to the Agency's
Airshed Model Application for St. Louis. The Environmental Sciences
Research Laboratory (ESRL) of the Office of Rese?rch and Development (ORD)
has had the lead role in the development v.d refinement of photochemical
models and has used the St. Louis 'i.ito. base for base-case simulations and
4
-------
comparisons of computed and observed concentrations. The Monitoring
and Data Analysis Division (MDAD) of the Office of Air Quality Planning
and Standards (OAQPS) has used the ESRL simulations to perform a statisti-
cal evaluation of model performance, has conducted sensitivity analyses
and control strategy simulations using a subset of the ESRL test days.
While ESRL's analysis focuses on the scientific validity of model com-
ponents (numerical diffusion, chemistry, windfields, etc.), the MDAD
analysis focuses on the evaluation and application of the model from a
a regulatory standpoint.
Initial simulations conducted in 1978-1979 resulted in a
strong tendency toward underestimation of afternoon ozone concentrations.
As a result, EPA carried out an extensive program to determine the
causes of poor model performance. Detailed examinations of model com-
ponents, the emission inventory and data preparation methods led to
substantial modifications in these areas. The most significant changes
are (1) the replacement of the Carbon-Bond I chemical mechanism with
the Carbon-Bond II mechanism, (2) correction of the emissions inventory
with the result that the quantity and reactivity of emissions were
increased, (3) layer-averaged photolytic rate constants used in place of
surface-based rate constants and (4) replacement of the numerical advec-
tion routine to eliminate artificial numerical diffusion. A substantial
improvement in model performance resulted from these modifications.
The methodologies and results for the ESRL base-case simula-
tions are detailed in the "Final Evaluation of Urban-Scale Photochemical
-------
Air Quality Simulation Models" by Schere and Shreffler.3 While the current
report summarizes the ESRL work, its primary emphasis is on the performance
evaluation, sensitivity studies and control strategy analysis conducted by
MDAD.
-------
2.0 The Urban Airshed Model
As previously stated, the Airshed Model is a three-dimensional photo-
chemical grid model which numerically simulates ozone and precursor concen-
trations for a large number of grid cells which collectively represent the
urban domain. The purpose of this chapter is to describe the technical basis
for the Airshed Model, its theoretical foundation, its treatment of important
physical and chemical processes and its assumptions and limitations.*
Overview
The Airshed Model is Eulerian, meaning that computations simulate
the time changes in precursor and secondary pollutant concentrations which
take place at given locations, i.e., within grid cells. For each simula-
tion, the user designates the horizontal extent of modeling grid and a uniform
dimension for the grid squares; both remain constant during the course of
simulation. The user also designates the depth of the modeling region and
the width of the vertical layers used; these dimensions, however, are allowed
to vary diurnal!y to account for changes in mixing. In typical simulations,
the modeling region encompasses the entire metropolitan area including the
downwind region which often experiences high ozone concentrations and the
region immediately upwind of urban areas. The choice of grid square size
reflects a balance between the need for resolution and the increase in
cost encountered as the number of grid cells grow. For the St. Louis
application, a 4 km square length was found to represent a reasonable com-
promise. In most applications, the first several thousand meters of the
atmosphere are modeled and the regions above and within the mixed layer are
*More detailed discussions of the Airshed Model are presented in
several reports.1,5
7
-------
each represented by several layers. Figure 1 illustrates the three-
dimensional grid approach used by the model.
The mathematical basis for simulating concentrations of ozone and
precursors is the conservation of mass. The major processes which cause
changes in the amount of mas:; in a given air parcel (cells) are emissions,
transfer of mass into and out of the cell associated with advection and
dispersion and chemical changes in the composition of the cell. The Air-
shed Model solves the following equation (stated in terms of concentrations
rather than as time differentials). For each cell,,
Concentration (t-j) = [Concentration (t-j_-|)] + [j-ate of mass transfer] At
cell volume
[mass emission rate] ^t + [cnemical transformation rate] At
cell volume
This equation is iterated through many time steps. In a typical simu-
lation each time step may represent several minutes. In the above equation
each time step is represented by At where At = t-j-ti_i. The left
hand term in the equation represents the concentration of a specific species
at the end of a time step. This concentration equals the concentration at
the beginning of the time step plus the changes in concentration due to
transfer, emissions and chemistry (right-hand terms) that occur during the
time step. At the beginning of the simulation (t = t0) an initial concen-
tration based on measured or assumed ambient concentration must be designated.
A typical time for starting simulations is 5 i.m.;, this is prior to the
morning rush hour when concentrations ^re still fairly low. This practice
makes the model estimates less sensitive to initial concentrations.
8
-------
While the primary concern may be ozone, simultaneous solution for
precursors and intermediate species as well as ozone is necessary. Pre-
cursors include NO, N02 and organic compounds split into five different
classes corresponding roughly to paraffins, olefins, ethylene, aromatics
and aldehydes. The intermediate species include various types of organic
and inorganic free radicals.
The right-hand terms of the equation are discussed in greater detail
in the following sections. Detailed mathematical expressions are available
elsewhere.5
2.1 Mass Transfer (Advection and Dispersion)
The Airshed Model simulates the transfer of pollutants in the
atmosphere which result from atmospheric motions. Advection refers to a
transfer by the mean wind flow and dispersion to a transfer by atmospheric
turbulence. The effect of both processes is dilution or from the standpoint
of fixed cells a net transfer of mass from cells with higher concentrations
to those with lower concentrations. Advection in the model is a function
of the gradient in concentration (differences in concentration between cells)
and the speed of the flow component along the direction of the gradient.
(Flow perpendicular to the gradient results in no advection.)
The rate of transfer due to dispersion is a function of the con-
centration gradient and the intensity of random or turbulent motions (depar-
tures from mean flow). In the model, the intensities are represented by
horizontal and vertical proportionality constants known as eddy diffusivity
coefficients (Kn and Kv respectively). Because they are difficult to measure
and specify precisely, the approach used in the model is one of approximation.
9
-------
With regard to horizontal pollutant transport, advection tends to be sig-
nificantly more important than turbulent diffusion. This dominance is
strengthened because local concentration gradients near point and line sources
cannot be represented in the Airshed Model owing to the use of a grid with
relatively coarse resolution,. Given the insensitivity of the model to
horizontal diffusion, K^ is set to a constant value of 50 m^/sec. In con-
trast to horizontal transport, turbulent diffusion is frequently the domi-
nating vertical transfer process. Thus, the model uses an individual value
of Kv values as a function of stability class, ground-level wind speed,
surface roughness and the height of the grid cell, all parameters which
affect the intensity of vertical mixing. (Reference 5, Appendix, for
detailed discussion).
The Airshed Model also simulates vertical advective transfer.
Vertical transfer arises primarily from two sources: (1) balancing the
effect of convergence and/or divergence of horizontal winds in order to
achieve a proper mass balance of air, and (2) accounting for changes in cell
height as the mixing height changes. The first source of vertical transfer
must be considered in preparing the wind data for use in the model. For
example, if the horizontal winds are converging in a cell, the wind field
used in the model must have a sufficient vertical component to prevent an
artificial accumulation of pollutants in the cell. The second source of
vertical transfer is an artificial one arising from the mechanics of the
simulation program. As the cell height •'=•=• v-,;, material located at a
given height may be transferred from -ir
-------
Thus far we have discussed the transfer of mass between cells.
The model also estimates the mass which is advected and diffused into the
modeling region from the surrounding region. Thus, an input requirement of
the Airshed Model is a set of concentrations at the margins of the modeling
region. More specifically, the model requires concentration estimates for
all key species for each hour simulated along the sides and at the top of
the modeling region. Transfer into the region is particularly important
where interurban transport of ozone (and precursors) occurs. However, even
the naturally occurring ozone background (on the order of 0.04 ppm) can
have a significant effect on urban photochemistry.
2.2 Emissions
The treatment of emissions is straightforward conceptually: the
emissions into a cell are added uniformly throughout the cell. Thus, the
change in concentration due to emissions simply equals the mass added
divided by the volume of the cell. Ground-level emissions are added into
the lowest level cell. Elevated emissions are added into upper level cells.
This requires that certain stack parameters be input for the major point
sources. The stack parameters are used along with the wind speed to esti-
mate an effective plume height for each major point source according to the
methods of Briggs.^ The emissions are then added uniformly throughout the
(upper level) cell which is calculated to receive the plume.
The Airshed Model is designed to treat emissions of eight pollut-
ants: NO, N02 five classes of organics and CO. NOX emissions must there-
fore be divided into emissions of NO and N02 and organics must be divided into
emissions of five organics classes. These classes are discussed in Section 2.3.
11
-------
A point of comparison between the Airshed Model and Gaussian
dispersion models is the treatment of plume rise. Like most inert pol-
lutant dispersion models, the Airshed Model uses the Briggs^ formulae
to estimate plume rise. However, unlike most Gaussian models, the Air-
shed Model is capable of simulating emissions that rise above the mixed
layer. Although the Airshed Model also disregards any emissions in
plumes that rise above top of the modeling region, the modeling region
in the Airshed Model can extend above the mixing height. It is thus
possible to simulate plumes that rise into an inversion layer between
the mixing height and the top of the modeling region. With diurnal
growth in mixing height, such pollutants may enter the mixed region
and affect surface concentrations at a later time.
2.3 Chemistry
The chemical mechanism is one of the most complex components
of the Airshed Model. The chemistry of ozone production has been the
subject of intensive study for well over a decade, and yet there is
surprisingly little agreement on the list of reactions to use in simu-
lating ozone photochemistry. Thus, this discussion will only provide
a general outline of the photochemistry of ozone production. The
discussion will then outline the major distinguishing features of the
chemical mechanism used in the Airshed Model.
The general process by which tropospheric ozone is formed is
illlustrated in Figure 2a shows the N02-NO-Q3 cycle. These reactions
may be written as chemical equations:
12
-------
a. N02 + sunlight + NO + 0
b. 0 + 02 -»• 03
c. 63 + NO * N02 + 2
Ozone is not emitted in any measurable quantity, and, in fact,
reactions (a) and (b) represent the only significant sources of tropospheric
ozone. However, these reactions are fully reversed by reaction (c). Thus,
NOX by itself will not cause significant concentrations of ozone. The only
way significant concentrations of ozone will occur is for some other species
to oxidize NO to N02 without destroying ozone. This is exactly what is shown
in Figure 2. This may be written as:
d. Organic radicals + NO+N02 + miscellaneous products
The net result of reactions (a), (b), and (d), then, is a recycling of
N02/NO, a modification of the organic species, and the generation of an
excess 03 molecule. That is, organic species permit the ozone formation
step while bypassing the ozone destruction step of the N02 - NO - 03
cycle. The result is a buildup of the ozone concentration. Reactions
(a), (b), and (c) are fast enough that a balance or equilibrium is still
maintained between the concentrations of N02, NO, and 03, but this balance
involves a much higher ozone concentration than would occur without organics.
The above four reactions provide an overview of the chemical proc-
cesses of ozone formation. However, the number of reactions which affect this
process is far larger and the chemical reactions simulated in these models are
a simplified representation of the innumerable reactions that actually occur.
It is more difficult to represent the various reactions or organics
than it is to represent the reactions of NOX. There is substantial agreement
on about 10 to 15 NOX reactions as being the most significant reactions for
13
-------
describing the fate of NOX. However, there is relatively little agreement
as to how best to represent organic reactions. This is primarily due to the
almost infinite number of species and reactions that occur. Moreover, urban
emissions include a wide variety of organic species. Therefore, a simplified
representation of all these reactions is needed.
The typical means of representing significant organic reactions
is to utilize a small number of reactions for about three to five categories
of organics. In contrast to other photochemical models, the categories used
in the Airshed Model represent total numbers of bonds of specific bond types
rather than total numbers of molecules of specific: molecular types. In
other words, the Airshed Model treats individual compounds not as molecular
units but rather as carbon-bond units. Several different carbon-bond types
are recognized: single bonds, double bonds, aromatic bonds, and carbonyl
bonds. (These can be recognized as the identifying bond types for paraffins,
olefins, aromatics and aldenydes.) The double bonds (olefins) are further
broken down into highly reactive and moderately reactive double bonds.
Carbonyl bonds include not only those associated with aldehydes, but also
those associated with ketones and esters. The single bonds are the least
reactive while the highly reactive double bonds are, as their description
implies, the most reactive; the others are of intermediate reactivity.
Another species which can be considered in the Airshed Model is
CO. Carbon monoxide plays only a minor role in ozone photochemistry and
most photochemical models consider it ins- ;-;'i, leant enough to ignore.
Carbon monoxide has more s" gnific:-n* r as u relatively inert gas which may
be used as a tracer. By comparing the CO concentrations estimated by the
model to measurements of CO concentrations, it is possible to obtain a
14
-------
direct assessment of the accuracy of the treatments of emissions, advection,
and dispersion without the complicating influence of chemistry. This can be
quite useful for assessing model performance.
Another feature of the chemical mechanism is the option to consider
temperature effects. Temperature has a significant effect on the rates
at which many reactions occur, although it is unclear what effect it has on
overall ozone production. Unfortunately, there is little known about the
temperature dependence of the rates of many important photochemical reactions.
Nevertheless, the Airshed Model does provide the option to utilize current
knowledge about the temperature variation of the rate constants used in the
chemical mechanism.
2.4 Surface Removal Processes
Ozone, N02 and other pollutants treated in the Airshed Model are
removed from the atmosphere by adsorption, absorption, or chemical processes
at the different surfaces in the region being modeled. The rate of removal is
dependent on the species, the type of surface and the rate of turbulent trans-
fer in the lowest portion of the atmosphere. In the Airshed Model, surface
removal is treated using the concept of deposition velocity; that is, the
uptake of a given pollutant at the surface is proportional to the concentra-
tion in the lowest cell and the proportionality constant is the deposition
velocity. The deposition velocity is estimated by defining a resistance to
mass transport and a resistance to surface removal. The transport resistance
is calculated for each surface cell (for each hour) as a function of surface
roughness, stability and windspeed, while the resistance to surface removal
is a function of pollutant and land use category.''
15
-------
2.5 Assumptions and Limitations
It is important for the reader to understand the major assumptions
and limitations inherent in the Airshed Model since these assumptions limit
the accuracy of the model in replicating observed concentrations and in
estimating control requirements. The model assumes that:
1. The chemical processes affecting ozone in real urban atmos-
pheres can be accounted for accurately with the simplified chemical
mechanism.
2. Subgrid processes are not considered in the current version
of the Airshed Model; since the model assumes instantaneous mixing of
emissions into an entire grid cell, it tends to overestimate the dispersion
rate for line and point sources.
3. Similarly, the degree of vertical mixing is artificially
enhanced by limiting the number of vertical layers (for economic reasons),
four layers were used in the current analysis.
The limitations in 2 and 3 appear to have the effect of suppressing estimated
ozone levels downwind of major NOX sources; this problem is discussed in
Chapter 4.
Additional limitations are associated with the collection and
preparation of data required by the model. For example, the user must con-
struct a three-dimensional windfield for the modeling region—usually from
a limited set of observations. Assumptions are made regarding the speciation
of hydrocarbon emissions from various sources and boundary concentration pro-
files are often assembled from sparse data. These problems are discussed in
subsequent chapters.
16
-------
3.0 Methods and Data Bases
The purposes of this chapter are: (1) to outline the general approach
used in the St. Louis Ozone Modeling Study; (2) to describe the RAPS (Regional
Air Pollution Study) emissions, meteorological, and air quality data bases
used in the analysis; and (3) to describe in brief the data preparation
required to create model input files.
3.1 General Approach
The general approach used in the St. Louis Study is outlined as
follows:
1. Twenty test days from the 1975-1976 RAPS period were selected
for base-case simulations.
2. Necessary input files were prepared from the RAPS data bases.
Meteorological and air quality inputs were derived for the 20 days. Emis-
sions files represent typical summer weekdays for the 1975-1976 period.
3. The Airshed Model was used to simulate air quality for the 20
test days. (Base case simulations)
4. The AMS recommended procedures^ were used to compare the
computed values for the base-case concentrations with observed values.
Most of the comparisons made are day specific. (Chapter 4)
5. Sensitivity analyses were carried out for a 3-day subset of
the 20-day data base. (Chapter 5)
6. Control strategy simulations were conducted using modified
emission inventories for the 3-day subset. Emission inventories used were
adjusted to represent the effects of stationary and/or mobile source control;
meteorology was held constant. (Chapter 6)
As previously stated, several rounds of testing were required to improve the
initially poor model performance. The results used in this study are based
on the fully revised emissions inventory and the 1982 version of the Airshed
Model.^ One other point needs clarification. Given the large uncertainties
associated with population and economic trends in the St. Louis area, we
17
-------
assumed that the levels of activity for stationary and mobile sources remain
constant at 1976 levels for the control strategy cases. For stationary
sources, changes in emissions, reflect federal1./ mandated controls. For motor
vehicles, the distribution and age mix are estimated for 1987. No changes in
automobile numbers, traffic volumes or traffic patterns were considered in
estimating emissions.
3.2 The RAPS Data Bases
The following sections give brief descriptions of the emissions,
air quality and meteorological data bases useri in the study. The dis-
cussion focuses on the data bases used for base-case simulations; descrip-
tions of input modifications for sensitivity and control strategy simulations
are presented in Chapters 5 and 6 respectively.
3.2.1 Emissions
The Airshed Model requires two different emissions files:
one for area sources and a second for major point sources. The area-source
file includes small stationary sources, minor point sources and mobile sources
Emissions from the area sources are all treated as ground-level sources and
are added to the lowest layer of the modeling region. Major point sources
include large industrial sources of hydrocarbons and NOX. (There are about
100 of each in the RAPS inventory). The model requires parameters for major
point sources which allow the calculation of plurne rise. The plume-height
calculations are used to determine the la/e»* ~"r, ..tticfi major point-source
emissions are introduced.
The emissions for both area and point sources are
designated for each hour of the model simulation and allocated into the
18
-------
appropriate species (NO, N02 and five classes of organics). The area-
source emissions are allocated spatially into nearly 2000 grid cells vary-
ing in size from 1 km in areas of dense emissions to 10 km in outlying areas.
Prior to modeling, the emissions from the RAPS inventory were reallocated
into a new grid array consisting of 374 grid squares 4 km on a side. Loca-
tions of the major point sources are retained in both the RAPS and Airshed
Model grids.
A complete description of the information sources and
methods used for the point and area-source emissions is beyond the scope
of this report. (Interested readers may refer to an extensive report by
Littman^ for documentation). Nevertheless, some additional information
is useful. In general, emissions for larger point sources are based on
houly records of process flows or fuel combustion for the specific sources.
However, smaller stationary sources are treated as area sources and emis-
sions from the different categories are obtained using various indirect
methds. For example, NOX emissions from commercial and residential heating
are derived from county-wide sales using allocation factors based on popu-
lation and land-use distribution. Mobile-source emissions are estimated
using transportation models of traffic flow and emission factors estimated
by Mobile I. Roadways are represented primarily as straight lines or links
whose emissions are determined as a function of average daily traffic, hourly
distribution of traffic and speed. Link emissions are then allocated into
the RAPS grid cells.
The RAPS emission inventory has been set up as an automated
archive system by ESRL. This system tracks point emissions not only by
19
-------
source, but also by industrial or process category. This feature was
utilized in the development of emission inventories for stationary source
control.
Tables 1 and 2 show the distribution of HC and NOX emissions
by source type for the St. Louis AQCR. Area sources are summarized in Table
1 and point sources in Table 2. Note that for hydrocarbons, stationary and
mobile (highway plus off-highway'l sources each contribute about half of the
emissions. Stationary source HC emissions are associated primarily with
evaporative source (e.g., miscellaneous solvent use, surface coating,
painting, degreasing). In contrast, nearly two-thirds of the NOX emissions
stem from a single source-type electric power generation. About 25 per-
cent of the NOX emissions are from mobile sources. That a large portion
of NOX is emitted from a relatively small number of power plants with ele-
vated stacks is extremely significant from the standpoint of photochemistry.
During the morning hours much of the NOX emssions will be emitted above
the urban mixed layer; on the other hand, a majority of HC emissions are
emitted within the mixed layer e.nd the HC/NOX concentration ratio tends
to be higher than indicated by "the total emissions.
3.2.2 Air Quality
Measured pollutant concentrations were used (1) to
estimate boundary and initial concentrations used in the model simulations,
and (2) for purposes of comparison with predicted concentrations. Air
quality measurements were obtained largely from the RAPS network of 25
surface stations which operated on a continuous basis during 1975 and
1976. The location of these monitors is shown on Figure 3. Each station
20
-------
was equipped with gaseous pollutant analyzers to measure ozone, NO, NOX
(N02 by difference), hydrocarbons (gas chromotography), total suspended
particulates S02, and CO. As Figure 4 shows, the monitoring stations are
clustered in the urbanized portion of the modeling region and large portions
of the nonurban modeling region have no monitors. Four sites (122, 123,
125 and 125) are located just outside of the modeling region and were used
to obtain boundary concentrations. An aerial monitoring system consisting
of three helicopters complimented the surface network. The airborne data
(total NOX, 03, S02, total hydrocarbons, methane and particulate measure-
ments) were used to depict three-dimensional pollutant distributions in the
urban region, and estimate the flow of ozone and precursors into the model-
ing region (vertical boundary concentration profiles). Documentation of
the surface land airborne measurement program is found in an EPA report by
Strothmann and Shiermeier.9
3.2.3 Meteorological Data
The 25 monitoring stations discussed in the preceding sec-
tion were also used to obtain meteorological measurements including wind
speed and direction (at 10 or 30 m), temperature, dew point, barometric
pressure and solar radiation. Vertical profiles of winds and temperature
were obtained from slow ascent rawinsondes released at 6-hour intervals
with pibals released hourly between radiosonde observations at urban site
141 and at outlying sites 142, 143 and 144. The meteorological data were
used to generate three-dimensional wind fields, to estimate diffusivity
and to estimate mixing heights for each hour of the simulations. (See
Section 3.3 for discussion of data preparation methods. Additional
information on the meteorological network is available in the EPA report).^
21
-------
Analyses of the RAPS aerometric data and on the relationship between
observed concentration and meteorological factors are presented in a series
of reports by ESRL. 10,11,12,13^4
3.2.4 Selection of Test Days
The evaluation of model performance contained in this
paper is based on Airshed simulations for 20 days (11 days from 1975 and 9
from 1976). There were three criteria for selection: data availability,
high measured ozone, and representation of different meteorological regimes.
For all 20 days the maximum nourly ozone concentration for at least one
site exceeded 0.16 ppm (160 apb). Table 3 lists the basic meteorological
conditions and peak ozone coicentrations for the 20 days. Most of the days
were characterized by sunshine and low to moderate wind speeds reflecting
the presence of high pressure systems which are conducive to photochemical
oxidant formation.
Three of the 20 days (Days 159, 195 and 275) were used for
extensive sensitivity testing and control strategy applications. All 3 days
experienced high ozone; 2 of the 3 had moderate wind speed and a third (275)
was characterized by very low wind speeds and mixing heights (stagnation).
3.3 Preparation of Airshed Inputs
3.3.1 The Modeling Domain
The horizontal extent of the .-nodeling region used for
St. Louis is shown in Figure 4. The grid s... -jomposed of 374 squares 4 km
on a side. The outermost rows and c
-------
the surrounding area. Vertically, four layers of cells were used; two
bottom layers for the mixed layer and the two top layers for the stable
air above the mixed layer.
3.3.2 Input Files
Extensive manipulation of the aerometric and emissions
data bases is performed using a set of preprocessors to obtain the input
files required for the Airshed Model. The content of these files is
summarized in Table 4. Several of the files warrant additional comment.
The diffusion break and region-top files describe varia-
tions in the mixing heights and modeling region depths respectively over
the domain during the course of the simulation. In the St. Louis Study,
the region top was designated to be higher than the mixing height for all
hours; this ensures that all of the urban emissions introduced into the mixed
layer will be treated. The diffusion break file requires hourly values of
the mixing height for representative locations within the modeling region.
As previously stated, rawinsondes were released at single urban site and
several rural sites at 6-hour intervals. The resulting temperature profiles
were used to estimate mixing heights at these locations and linear inter-
polation was used to generate hourly values. Spatial interpolations are
made within the Airshed Model.
In order to construct the windfield file, ESRL used a
preprocessor developed by Systems Applications, Inc.15 The preprocessor
assimilates all available surface and upper air winds and produces a
gridded field of U and V wind components for each of the four vertical
23
-------
levels used in the model. The preprocesor utilizes an objective analysis
which: (1) derives an average surface wind vector for the entire net-
work; (2) simulates mesoscale urban circulation patterns using the surface
temperature patterns and basic dynamic equations; and (3) uses measured
upper air winds to define wind profiles. Vertical velocities are calculated
internally by Airshed from the equation of mass continuity. The spatial
averaging of surface winds is done in order to eliminate unrealistically
large horizontal divergence, divergence that will result in excessively
large vertical motions within the model.*
The metscalers file describes the temporal variation of
vertical temperature gradients, stability class, atmospheric pressure,
water vapor concentration, and the N02 photolysis rate constant. Values
for these parameters do not vary spatially in the model. The top concen-
tration file specifies the concentrations of principal species at the top
of the modeled region throughout the simulation. In most cases these con-
centrations will be close to the clear air background values for these
species, although substantial concentrations of 03 can be advected over
the region. Its value is determined from the 03 measurements at the far
upwind surface monitoring sites after the nocturnal inversion has eroded
and the air aloft mixes through the atmosphere to the surface. The range
of inflow ozone concentrations experienced can be seen in the final column
of Table 3.
*A report by Schere^ examines the use ct alternative wind routines for
the St. Louis Application of Airshed,
-------
The air quality, temperature and wind speed files all
require data from the surface monitoring network. Hour averages of observed
species concentrations are objectively interpolated across the model grid to
produce a field of initial concentrations for the Airshed Model. Typically,
this initial field is applied at a near-sunrise time. The temperature file
produces gridded fields of surface temperature at each hour of model simu-
lation. These are required for both the wind and emissions point source
files (for plume rise).
The emissions file for the current study was prepared by
reallocating the hourly area-source emissions from the RAPS inventory (see
Section 3.2.1) into a new grid array (the Airshed Grid) consisting of 374
grid squares, 4 km on a side. Similarly, the point-source file required
a reallocation from the RAPS to the Airshed Grid. Figures 5 and 6 show
the emissions distributions from all sources of THC and NOX, respectively,
allocated into the Airshed Grid. The particular emissions shown are for
the 1-hour period 0800-0900 CST on July 13, 1976, but also characterize
the typical summer, weekday rush hour in St. Louis. Downtown and indus-
trial areas are evident in the central area from the higher emission rates.
The stacks from several large power plants are seen in high cell NOX
emissions in the north central and south central portions of the regime.
Finally, the terrain file describes the spatially vary-
ing types of surfaces in the modeling region. Both surface roughness and
deposition velocities are calculated within the model as a function of the
25
-------
land-use categories in the terrain file.^ (For detailed discussion
of methods used to develop this file, see reference 17). The land-use
distribution is shown in Figure 7.
-------
4.0 Model Performance Evaluation
The purpose of performance evaluation is to assess how well a model
can reproduce observed concentrations using emissions data and meteoro-
logical conditions for the same period as the observed concentrations. In
actuality, it is not only the model that is being tested, but the appli-
cation of the model including the appropriate use of the model and the
validity of the inputs. The objectives of this chapter are to report on
the St. Louis evaluation of the Airshed Model—its methods and its results.
Several concerns are foremost: (1) the accuracy of the model applications,
especially in simulating the peak and near-peak concentrations; (2) the
precision of the estimates; and (3) the ability to replicate observed spa-
tial and temporal patterns. The authors view accuracy as the most critical
of the attributes; however, the precision measures are used to establish
the level of confidence that we have in our estimates of accuracy. An
additional objective is to identify sources of error. The primary focus
of the study is ozone; the full complement of statistical tests described
below were performed for ozone; however, a smaller number of tests were
conducted for CO and N02-
The specific procedures used in this study are those recommended by
the AMS Workshop on Model Evaluation^ as supplemented by Cox.1** Before
discussing these procedures in detail, it is necessary to make several
statements about the data sets used in the analysis and their treatment
prior to statistical analysis. The starting point for the calculation of
performace measures is a table of observed and predicted concentrations
for each day arranged by hour and by monitoring station. An example for
27
-------
one day (Day 226) is shown in Table 5. The predicted concentrations shown
are not the grid square concentrations obtained directly from the model but
are estimates of discrete concentrations for the locations of the monitoring
sites. These estimates were derived by averaging the values from the grid
cells in close proximity to the monitor. The grid values were weighted
according to the inverse of the distance between cell and monitor. Secondly,
it is important to note that the performance measures were calculated sep-
arately for each day due to the large variation of meteorological condi-
tions from day to day. However, several techniques are used to examine the
behavior of the performance measures over the whole 20-day data set.
The methods used to evaluate model performance are discussed in Sec-
tion 4.1, results in Sectiois 4.2 through 4.6. Section 4.7 reports on a
partial analysis made for pollutants other than ozone and 4.8 is a summary
of the main conclusions on model evaluation.
4.1 Methods
4.1.1 Accuracy of Peak/Near-Peak Predictions
From the standpoint of regulation, the most important
aspect of the evaluation is determining how accurately the model reproduced
peak and near-peak observed concentrations. The methods used in the cur-
rent study are outlined in Table 6; these include the recommendations of
the AMS Model Evaluation Workshop4 and the refinements of Cox.1^
Table 6a summarizes the calculations made for each of
the 20 test days separately. Peaks are compared for four different
28
-------
residuals using observed and predicted maximum hourly concentrations.
Residuals are for observed and predicted values which are paired in time
and location (Fl), paired in time but not space (F2), paired in space but
not time (F3), and unpaired in time and space (F4). Fl is clearly the most
stringent measure; a zero residual (no bias) requires that the computed
value for the time and location of the observed peak be identical to that
of the observed peak. In the least strigent measure F4, the model peak
is compared to the observed peak without regard to location or hour. F2
and F3 have intermediate stringencies; for example in F2, the maximum
observed concentration is compared to the highest value predicted for
the same hour but without regard to location.
Several authors have pointed out that the Fl, F2 and
F3 measures have an inherent tendency to show underprediction even when
the model has no actual overall bias.18,19 This tendency is illustrated
in the following simple example. Suppose the observed peak is 200 ppb and
occcurs at 3:00 p.m. at Station A. The model on the other hand predicts
that the peak for the day will occur at Station A and will also be 200 ppb;
however, the time is 4:00 p.m. rather than 3:00 p.m. The predicted value
for station A at 3:00 p.m. is 160 ppb. One can see that residuals Fl and
F2 are both equal to (200-160 = 40 ppb), i.e., the model is interpreted as
underpredicti ng the overall and temporal peak. However, in actuality the
calculated bias is the product of temporal displacement, rather than a fail-
ure of the model to generate sufficient ozone. Similarly, a displacement in
the location of the peak will result in a positive bias (underprediction)
for F3. Only the totally unpaired residual F4, is free from this
29
-------
characteristic. A second problem is that there measures are based on a
single residual and thus are not robust.
To overcome the problems associated with the residuals,
more robust alternative procedures for evaluating near-peak accuracy of
models are also considered. For the Al statistic (see Table 6a), the
peak residuals paired by hour are calculated for all hours and not merely
for the hour of the highest observed value. Similarly, for A2, residuals
of peak values paired by site are calculated for all sites. Thus, as
Tabl e 6a shows, the availability of multiple residuals allows an Al and
A2 bias (
-------
constructed using the standard deviations of observed and predicted
values.*
As evident from Table 5, the F measures are actually a
subset of the A measures. In Section 4.2.1, a numerical comparison of the
two measures is presented.
Table 6b shows the measures obtained to show the behavior
of residuals over the full 20-day test set. For the F residuals, a 20-day
bias (~5 for Fl, F2, F3 and F4) is obtained by averaging the residuals for
the 20 days. For the A residuals, the 20-day bias is the average of the 20
daily bias values. Calculation of precision values (standard deviation)
for the 20-day data base is shown in Table 6b. In addition to the bias and
precision measures, a number of scatter diagrams of observed versus
predicted values were constructed using the different F and A residuals;
respective correlation coefficients of observed and predicted values were
also obtained. The results of the analysis on peak and near-peak accuracy
are reported in Section 4.2.1.
4.1.2 Overall Accuracy and Precision
The measures used to assess overall accuracy and pre-
cision across the full range of concentrations are essentially those
recommended by the AMS Workshop.4 These measures are outlined in Table 7.
*The reader should keep in mind, however, that both the "F" and the "A"
statistics are tied to predictions made at specific stations. Thus,
neither is completely free of the effects of spatial displacement if
such displacement is associated with predicted peaks which were not
precisely at the monitoring sites. The A statistics are freer from
this effect than the F statistics.
31
-------
As this table shows, measure;; of bias and scatter were obtained using paired
and unpaired analyses. In the paired analysis residuals are calculated by
subtracting the predicted concentration for a given hour and site from the
observed value for the same hour and site. For the unpaired analysis,
individual residuals are not calculated; instead the characteristics of
the sample of observed values are compared to the characteristics of the
sample of predicted values. For example, the cumulative frequency dis-
tributions of observed and predicted concentrations are plotted separately
and compared.
For both tlie paired and unpaired analyses, the study uses
a simple graphical approach to combine information on accuracy and preci-
sion. In this approach, the bias values for each day are bracketed by day-
specific 95 percent confidence intervals; this arrangement enables one to
weigh the relative importance of bias and scatter. For example, when the
confidence intervals do not overlap zero we may assume that the bias is
statistically significant and indicates systematic rather than random
error. The methods of constructing the confidence intervals are given
on Table 7. For the paired analysis, the intervals are based on standard
deviation of the residuals; for the unpaired analysis a pooled standard
deviation (from S0 and Sp) is used. It should be noted that the con-
fidence intervals were computed assuming independence of the data. This
assumption however, is not true. We know that soatial and temporal auto-
correlation exists in the cbserved and pr?HTte.J data sets and also among
the residuals. The confidence intervals should therefore be viewed as
32
-------
first approximations which tend to underestimate the true extent of scatter
about the bias.
In addition, to determine how well the model does in an
overall sense, it is important to find out whether the errors fall into any
discernible patterns. For example, is the model more accurate (or precise)
at high concentrations versus low concentrations? Is there a relationship
between bias and time of day or between bias and receptor location (upwind
versus downwind of the urban core)? Is the model more accurate for transport
days versus stagnation days?
Several analyses were conducted to study how bias and
scatter vary with concentration level. The bias/confidence interval graphs
were constructed separately for three subsets of paired residuals: (1) all
values; (2) those for which either the observed or predicted values
exceeded 80 ppb; and (3) those for which either observed or predicted values
exceeded 120 ppb. Secondly, scatter diagrams showing predicted vs. observed
concentrations were plotted to further establish day-specific relationships
between bias and concentration level. For unpaired analysis, day-specific
frequency distributions of observed and predicted concentrations are compared.
To test for diurnal and spatial patterns of error, a
series of diagrams was prepared showing the errors as a function of hour
and site for each of the 20 days. In addition, an analysis was made to
determine whether the ability of the model to predict daily peak observed
concentration varies with meteorological conditions such as wind speed and
mixing height.
33
-------
4.1.3 Replication of Patterns
As stated in Chapter 1, grid models are designed not only
to predict maximum concentrations, but also to predict spatial and temporal
distributions. For example, we would like to use the Airshed Model to fore-
cast how the aerial extent of exceedances (e.g., number of cell/hours with
ozone concentration >0.12 ppm) changes with given control programs. Prior to
using the model in this way, it is necessary to demonstrate that the model
can in fact reproduce spatial and temporal patterns of variation with a rea-
sonable degree of accuracy. As Table 8 indicates, the techniques used to
examine these attributes include (1) visual comparison of time series and
spatial plots of observed and predicted concentration, and (2) computation of
spatial and temporal correlation coefficients between observed and predicted
concentrations. Both of these techniques were used for each of the 20 test
days. One fact to bear in mind is that the correlation coefficients
reflect only the degree of covariation between observation and prediction;
thus, a high correlation coefficient may be computed when the predicted
values contain a large systematic error or bias. Thus, the visual compari-
sons are more useful in determining how well the model does in estimating
the distribution of areas exceeding the NAAQS or other levels. Formulae
for correlation coefficents are found in Table 8. While no attempt was
made to establish confidence intervals about the correlation coefficients
(as suggested in the AMS report), temporal correlation coefficients were
calculated for all monitoring sites and spati?.! -.orrelation coefficients
were calculated for all hours. Thus, tho .^/-^age values and the range of
temporal and spatial correlation \. >:f 'iclents are established for each day.
34
-------
4.2 Results of Model Evaluation Analysis
The basis for all of the model evaluation analyses are matrices
of observed and predicted values for each day; the matrices are arranged
*
by hour and by station. An example (for Day 226) is presented in Table 5.
For convenience, the results are presented and discussed in
three sections corresponding to the three Protocol Tables 6, 7, and 8.
Results of the peak/near-peak accuracy analysis are presented in Section
4.2.1; those of the overall acuracy and precision analysis are given in
4.2.2 and 4.2.3 provides an assessment of the model's ability to replicate
spatial and temporal patterns.
4.2.1 Results: Peak/Near-Peak Accuracy Analyses
The computed values of the performance measures for peak
accuracy are presented in Table 9. The values for each day are given in
Table 9a and the 20-day average residuals and 20-day confidence intervals
are given in Table 9b. All calculations have been made according to the
formulae given in Table 6.
The analysis of the results focuses on a number of ques-
tions: (1) is there an overall tendency towards negative or positive
bias in the peak and near-peak predictions? (2) does this bias vary from
day to day? (3) how large is the random component or scatter of the
errors? (i.e., what uncertainty is associated with our estimates of bias),
and (4) what differences and relationships can we discern in the behavior
of the different residual measures (F4, Al, A2, etc.)? A number of graphs
35
-------
have been prepared to facilitate interpretation of the results and to
answer the above questions.
* Figure 8 is i graphical presentation of the 20-day aver-
age residuals and confidence intervals presented in Table 9. The average
residuals shown are all positive, and most may be interpreted as being sig-
nificantly different from zero; only A4 confidence intervals (based on
p p
S g, S p)* overlap the zero bias line to a small extent. Thus, we can
state with considerable confidence that the model has a systematic tendency
to underpredict peak and near-peak ozone concentration over the 20-day
data set, irrespective of the particular measure used.
Figure 8 demonstrates that the choice of measure used will
have a pronounced effect on one's perception of the degree of bias. The
differences between the values of the different measures are consistent
with conclusions drawn in Section 4.1.1. Estimates of underprediction are
largest for the F residuals because they are based on the time and/or
location of the observed peak; temporal and spatial displacements of the
predicted peaks are constraiaed to introduce positive bias (underprediction)
Only F4 is free from this effect and is shown to have a smaller positive
bias.
Confidence intervals for the unpaired measures F4 and A4 were cal-
culated using two methods. The inner confidence intervals measure the
variability of the cT values over the 20-day period and are based on the
standard deviations of the differences (S H). The outer intervals are used
to determine whether 1$ is significantly different from zero and are based
on S = /^ *~s~
o p.
36
-------
The A residual measures for temporal and spatial pairing have
values of 12 and 14 ppb compared to a value of 48 ppb for both of the cor-
responding F2 and F3 measures. As expected, the Fl measure which is paired
in both time and space to the observed peak has the largest value (61 ppb).
The solid bars represent 95 percent confidence intervals
about the mean residuals based on the variance of the residuals over the 20-
day data set. The intervals on A measures are considerably smaller than those
of the F measures, indicating that the former are more stable indicators of
bias. This is to be expected since the F measures are based on single residuals
for each day, while A measures are more robust averages (Al: based on 15 hours;
A2: based on 21 sites; A4: 25 highest observed, 25 highest predicted values).
Figures 9 and 10 show how the measures vary from day to day.
Figure 9 presents the spatially paired measures. Confidence intervals based
on within-day variation are constructed about the A2 mean residuals.
The A2 measures demonstrate statistically significant under-
prediction for 13 of the 20 days, significant overprediction for 3 days and
near zero bias (confidence intervals substantially overlap zero line) for 4
days. These measures allow us to assess how well the model matches the
observed maxima (regardless of time) for the ensemble of sites. They demon-
strate that for the ensemble of sites, underprediction occurs on most but not
all days and suggest that the extent of underprediction is less serious than
indicated by the F3 measure.*
*The F3 measures are positive (underestimates) for all days. This
finding suggests that the constraints inherent in this statistic mask its
ability to distinguish true model bias. From the F3 measure we can conclude
only that at the location of the observed peak, the maximum model concentra-
tion is underpredicted for all days.
37
-------
Figure 10 shows the values of A4 and F4 measures for each
of the 20 test days. Although both are unpaired, the F4 is based on the
single residual (Omax - P max) while A4 is based on a comparison of the
average of the highest 25 observed values and the average of the highest
25 predicted values.
The behavior of the unpaired measures over the 20-day
data set is similar to that for the spatially paired measures (Fig. 9).
however, two of the F4 measures show negative bias (overprediction) and
several are close to the zero line. However, F4 is a less stable indi-
cator and undergoes more day-to-day variation than does the A4 indicator.
Note also that the F4 values tend to be greater thain the A4 values in an
absolute sense (i.e., where both positive and negative bias appear to be
magnified by the F4 estimator)- The smaller magnitudes of bias for A4
probably result from averaging the predicted and the observed values to
obtain the A4 values.
Model evaluation exercises often include scatter diagrams
of observed and predicted concentrations. It is possible to construct such
diagrams using the observed and predicted concentrations which correspond
to the different measures used to estimate the bias of peak and near-peak
values. Figure lla gives scatter diagrams for the spatially paired measures
A2 and F3 for the 20 test days. Figure lib gives similar diagrams for the
unpaired measures A4 and F4. Correlation ^efficients (obs. vs. pred.) and
centroid position (0, PT a~re also given for each of the four measures. The
diagrams reaffirm the findings discussed previously with regard to the bias
estimates: all of the estimators ind'cate positive bias (underprediction)
38
-------
for the data set as a whole; the F measures show a greater apparent pos-
itive bias than do the A measures. In addition, the A measures result in
a much higher correlation between observed and predicted concentrations;
this result probably stems from the fact that the A measures are averages
and eliminate irregularities associated with individual values. Note also
that while the F4 (unpaired) measure shows less positive bias than the F3
(spatial) measure, the scatter of the F4 measure is greater and its cor-
relation coefficient smaller than those for F3.
4.2.2 Overall Accuracy/Precision
Table lOa gives the bias and confidence intervals for the
unpaired analysis for each of the 20 test days. The first set of columns
gives the results for the entire set of pairs; the second set gives the
results for those pairs in which either the observed or predicted value is
80 ppb.* In the third set, the cut-off level is 120 ppb. In Figure 12, the
bias values and confidence intervals for the three data sets are portrayed
graphically. Figure 13 is identical except that the bias and confidence
intervals are based on normalized data.**
Several conclusions can be drawn from Figures 12 and 13.
The normalized and unnormalized analyses show nearly identical results.
For all pairs (a), 15 of the 20 days have bias values which fall within
*Stratification in this manner ensures that the analysis is not biased.
Had the cut-off been based on observed value, the results would have tended
artificially toward underprediction.
**The bias and confidence intervals in Figure 13 are based on normalized
data obtained by dividing residuals by the term (V + IH/2.
39
-------
±15 ppm (about 20 percent o? the zero line; of these, 3 days have no
significant bias. Of the 5 days with strong bias., 2 showed negative bias
(overprediction) and 3 showed positive bias (binder-prediction). On the
whole, the number of days having negative anu positive bias is nearly equal.
There is, however, a marked shift toward underprediction
for the analyses based on the higher ozone concentrations, especially for
(c) based on the 120 ppb cut-off. Note that the degree of imprecision
also tends to grow as one restricts the analysis to data pairs with higher
cut-offs. A similar result can be seen in Fig. 14, scatter diagram of
predicted vs. observed ozone concentration for Day 195. For values lower
than 80 ppb, the points tend to cluster about the line of perfect predic-
tion. At higher concentrations, the tendency is towards underprediction
and the spread of the points tends to increase.
What explains the change in bias and confidence intervals
with the magnitude of ozone concentration? Several factors may be involved.
The all-inclusive data set contains a greater proportion of low concentra-
tions, many of which are from early in the day. The early-day ozone cal-
culations are dependent to a large extent on observed concentrations which
are used to initialize the model. In a sense, the model is forced to
come up with the right answer. Higher predictions on the other hand are
much more dependent on the mathematical simulations of physical and chemical
processes and it is possible that problems with model algorithms contribute
toward the larger bias. There is a greater chance for error and both the
size of the bias and the scatter tends to increase.
40
-------
Unpaired analysis. Table lOb presents bias values and
confidence intervals for the unpaired analysis. Results were obtained
only from the all-inclusive data set. The bias values are identical to
those of the paired (all-inclusive) analysis.* However, the confidence
limits tend to be from two to three times larger than those obtained from
the paired analysis. This ocurs because the unpaired analysis uses a
pooled standard deviation based on the standard deviations of observed
and predicted values rather than on the standard deviation of residuals.
Both the observed and predicted values show considerably more spread
than do the residuals.
* ¥ unpaired = 0" - F = R" ( Z0i _ zpi ) = N
1_
d paired = N ^(O-j - PJ) by definition.
In order to determine whether the unpaired bias varies
with concentrations, cumulative frequency distributions (CFD's) of observed
and predicted concentration were plotted for the 20 days. Two examples
(those for Days 159 and 195) are given in Figure 15. These graphs are
used by comparing the observed and predicted concentrations at the same
percentile. Figure 15a gives the CFD's for Day 195. For this day the
model -predicted values are substantially lower than observed values at
higher percentiles. At lower percentiles, predictions and observations
are simiar in magnitude. Most of the days fall into this pattern which
is consistent with the result of the paired analysis. However, there
are a number of exceptions, one of which is shown in Figure 15b (Day 159).
41
-------
For this day, predicted value's exceed observed values for percentiles
less than 40 and greater than 90. In the intermediate range, observed
and predicted values are about equal. A similar relationship is indicated
in the scatter diagram of observed and predicted pairs for Day 159 (Fig.
16). In this paired analysis, overprediction is found at lower concen-
trations and higher concentrations, while the pairs in the mid-range
exhibit both under- and overarediction.
In summary, both the paired and unpaired analyses tend
to show similar results, results which are also consistent with those of
the peak and near-peak accuracy (Section 4.2.1). The Airshed Model for
St. Louis has a systematic tendency to underestimate for areas and periods
of high observed ozone. This pattern occurred on about 75 percent of the
days. Possible causes for this underprediction are discussed in Section
4.3 which focuses on interpreting the errors.
Several additional comments on model precision are in
order. Table lOa includes the variances of observed and predicted values
for the all-inclusive data sets for each of the 20 days (S20 and S 2).
Also given is the ratio S^Q/S2 ^Calc^ ^or eacl1 day* ^or a Per^ect model
o o
application we could expect S 0 = S _ and Fca-jc = 1. However, as is shown,
S*-Q is significantly greater than S2 for all but 4 days. This indicates
that the model is not reproducing the full extent of variation exhibited by
the observed data, a result that is consistent with the finding that the
model tends to underpredict at high conce? ..-/dons and on many of the days
41'
-------
to overpredict at low concentrations. This result is evident in the
time series diagrams presented in Section 4.2.3.
4.2.3 Replication of Spatial and Temporal Patterns
The ability of the Airshed Model to replicate spatial and
temporal patterns is measured by using Pearson correlation coefficients of
paired predicted and observed values. Secondly, time series and isopleth
diagrams of predicted concentrations are compared to their observed
counterparts.
Table 11 gives spatial, temporal and overall correlation
coefficients for each day. Examination of the coefficients reveals that
the temporal correlation coefficients are very high. Similarly, time series
of predicted and observed ozone concentration shown in Figure 17 demonstrate
that the model is accurately able to simulate the diurnal pattern found in
observed data: low values in the morning, a build-up toward afternoon and
decreasing values toward evening.* (The reader should recall that the
correlation coefficients reflect only covariance between the observed and
predicted values; thus, high values can result even when substantial bias
exists). As discussed in the previous section S g/S p is significantly
greater than one for nearly all of the days; this finding indicates that
*Figure 17 gives the diurnal pattern of observed and predicted
ozone averaged over all stations and the corresponding residual 0~P
for the 20 test days.
43
-------
the model tends to underestimate the full extent of temporal variation.
The overall correlation coefficients reflect the temporal covariance and
are moderately high in value.
Figure 18 presents an overview of the daily time series
shown in Figure 17. This figure is the 20-day, all station average time
series of observed and predicted concentrations and residuals (fr-P"). This
diagram illustrates the strong covariance between observed and predicted
values, but also highlights a distinct diurnal pattern of errors. Over the
20 days the model has a tendency to overpredict in the early morning and
evening hours and to underpredict from the mid-morning to late afternoon.
This pattern is clearly discernible on 12 of the 20 days (See Fig. 17).
Several days, for example 275, were notable exceptions. On 275, the model
bias was near-zero for the early morning to mid-afternoon; however, the
predicted ozone values decreased more slowly than the observed values and
overpredicted ozone for the late afternoon and evening hours. On Day 142,
the usual pattern was reversed, underprediction occurred early and late in
the day and overprediction characterized the afternoon.
Another issue regarding the model's ability to reproduce
temporal patterns is its accuracy in predicting the time of the observed
peak. Figure 18 suggests tha.t for the overall 20-day data set, the model
does quite well; the curve for the observed maxima shows highest and equal
values at hours 1300 and 1400. The model predicted maximum value is for
the 1400 hour. To examine this issue in greater detail, the time of the
44
-------
observed and predicted maximum concentrations were compared for each day and
a histogram of the daily ^ values (time of observed max minus time of
predicted max) is shown in Figure 19. The histogram demonstrates that the
model estimate of the peak hour is early on some days and late on others.
For the 20 days as a whole, there is no consistent pattern of error in the
time of the peak and the differences cluster about zero At. (The sample
size is too small to verify the bimodal distribution suggested by the
hi stogram.)
Examination of the spatial correlation coefficients (Table
11) suggests that model's ability to reproduce geographical patterns varies
considerably from day to day. While most of the spatial r values are statis-
tically significant (0.05 level) they are generally lower than the temporal
values and vary from high positive to negative for peak hours. In inter-
preting the spatial correlation coefficients, we would emphasize that the
values are based only on those locations having monitoring sites. There is
a sparsity of sites in the outlying regions and the correlation coefficients
tend to weight the central portion of the modeling region. Thus, the analysis
is not representative of the outlying regions which often experience the
highest ozone concentrations. A similar limitation affects our interpretation
of the spatial displays of observed and predicted values discussed below.
Figure 20 presents isopleths of observed and predicted
concentrations for the hour of the observed peak for test Days 159, 195,
225, 231 and 275. Accompanying the isopleth diagrams are estimates of
45
-------
wind trajectories which pass through the hour and location of the maxi-
mum observed ozone concentration. These can be med to get a rough idea as
to how urban precursor emissions and the developing ozone cloud will move
throughout the day. In examining these figures, an attempt was made to
answer several questions for each day. Are the areas of maximum predicted
concentration coincident with tie areas of highest observed concentration
and consistent with the trajectories? Are the patterns and gradients of the
predicted values similar to those of observed values? Are the areas of
exceedance indicated by the predicted values similar to those indicated by
the observed values?*
Our ability to answer these questions is constrained by
the sparsity of sites in outlying regions. Thus, while it is possible to
draw the isopleths of predicted ozone with confidence (using all grid cells
in the modeling region), less certainty can be ascribed to the isopleths of
observed concentration particularly in the downwind regions of high ozone.
1. Highest predicted values are generally found in the
area downwind of the urban areei.
2. For those cases where sufficient monitors are avail-
able, there appears to be a general correspondence between predicted and
observed areas of high ozone (this was not the case on several days).
*Exceedance areas are tho'ie for which the ozon^ concentration is
greater than 120 ppb, i.e., the NAAQS level, 0" •.->v>>1 and predicted
exceedance zones are indicated on the i sop1 el", diagrams in Figure 20.
45
-------
3. For most of the test days, there was a general cor-
respondence in the spatial patterns indicated by the isopleths; however,
the fit varied from day to day (subjective comparisons were often difficult)
4. The model tends to underestimate the size of the
exceedance zone on most days.
These results are generally consistent with those of
previously discussed analyses. The apparent underestimation of the area
exceeding the NAAQS level is consistent with the bias analysis shown on
Figure 12c (for ozone > 120 ppb). Secondly, isopleth comparisons indi-
cate day-to-day variation in the model's ability to reproduce spatial
patterns as did the correlation coefficients. However, it is difficult
to establish a relationship between the spatial daily correlation
coefficients and subjective judgment based on the isopleth comparisons.
In certain cases (mainly stagnation days) the visual patterns suggest a
much stronger agreement than indicated by the spatial r values. Day
275 is an excellent example. Both the observed and predicted displays
show similar patterns of concentric isopleths of ozone with highest
values in and near the urban core. However, the spatial correlation
coefficient for this hour was 0.12 (barely significant at the 95 percent
level). This apparent discrepancy stems from the effects of slight
spatial displacements where gradients are extremely sharp. In such
cases, the visual comparisons probably give a better measure of the
model's ability to simulate spatial patterns than the correlation
coefficients. (For stagnation days, high concentrations are found
in the inner portion of the modeling region where the monitoring
47
-------
network is dense; thus, isopleths of observed values can be drawn with
confidence.)
An additional finding is that the Airshed Model on certain
test days creates zones in which the predicted ozone levels are well below
background (upwind) levels. These areas are located downwind of major
point sources of NOX (power plants). An example is seen in the isopleths
of predicted ozone for 1400-1500 Day 231 (Fig. 20d). Below background
ozone values extend in a plume-line pattern downwind of a cell containing a
large power plant which is the largest single source of NOX in the modeling
region (4386 kg/hr). Unfortunately, the monitoring network is not sufficiently
dense to determine whether the "deficit zones" associated with NO titration
(scavenging) are as pronounced as those simulated by the model. The model
may overestimate the size of these zones and underestimate 03 because of
the way it instantaneously mixes point source NOX emissions into a 4 x 4 km
grid cell.
Day 225 (Fig. 20c) is also worthy of detailed examination.
The bias measures indicated that the extent of model underprediction for
this day far exceeded that shown on the other test days. As Fig. 20 indi-
cates, the highest observed ozone (<160 ppb) for this day occurred during
the midafternoon in the southeast quadrant of the modeling region. Yet the
model concentrations barely reached 70 ppb, nor are any well-organized zones
of high ozone discernible. The trajectory •irforn,.-tion (shown below the
isopleths) suggest an explanation for the ^nltl em. It is probable that most
of the morning emissions were ad^p^'d out of the modeling region (i.e., to
-------
the north and east). Once mass is transported to the boundaries, it
is lost to the model. On the other hand, in "the real world," it is quite
possible for a pollutant-laden mass to be advected back into the region
later in the day. A test of this hypothesis would be to simulate this day
using a larger modeling region, one that is extended to the east. An alter-
nate and equally plausible explanation is that windfield errors resulted
in a much exaggerated transport of precursor mass out of the modeling region.
It is obvious that Day 225 is clearly an outlier and should not be used for
estimating control requirements unless substantial improvement is obtained
through refinement of the inputs or expansion of the modeling region.
Day 159 (Fig. 20a) illustrates the case where the network
of monitoring sites is insufficient to evaluate the model's ability to repli-
cate spatial patterns. The model predicts a broad region of ozone concen-
trations in excess of 200 ppb in the northwestern corner of the grid (the
predicted peak is 312 ppb). Unfortunately, there are no monitors in this
region to verify the predictions.
The pattern of errors for the area covered by monitors
is interesting. As Figure 21 shows, the afternoon hours show a persistent
tendency for overprediction (+ bias) to the west and underprediction (- bias)
to the east. One might speculate that this pattern may be associated with
errors in the windfield.
For Day 195, the model does fairly well at replicating
the overall spatial pattern. However, substantial underprediction is
evident in the size of the exceedance zone and in the concentrations
49
-------
near the urban zone; the model also shows "deficit zones" downwind of major
NOX point sources in the north central and south central portions of the
region.
4.3 Potential Sources of Error
To a certain extent., constraints of the monitoring network may
contribute to the underprediction that is evident for most test days. Evi-
dence suggests that the ozone measurements themselves may be biased. Gas-
phase titration with excess NO was used during the RAPS field studies as
the ozone calibration technique.20 Comparison studies have since shown
that gas-phase titration may an average give ozone concentrations as much
as 7.5 percent higher than does uv photometry, the latter being the EPA
designated calibration method for ozone analyzers. In controlled labora-
tory experiments, uv photometry on average yielded values 3.6 percent lower
than did gas-phase titration.21 Thus, the ozone observations used for
model evaluation may be slightly on the high side.
Secondly, the paucity of monitoring sites in outlying regions
may affect the model performance measures of bias. It is emphasized that
all of the bias measures presented are restricted to those grid cells which
lie in the immediate vicinity of monitors. Thus, to a certain extent, even
the unpaired measures are restrictive. For example, if the model peak occurs
in a region void of monitors, this peak will not be included in the F or A
residuals discussed previously. When one pints *"*"'" highest peak ozone com-
puted for any grid cell in the modeling re[ >n against the highest observed
ozone (Figure 22), the perception <"'* .r.lel performance is different and the
bias appears to disappear. T\\e problem, however with this approach is that
50
-------
the predicted values are drawn from a much larger sample than the observed
values and thus, the results are biased towards overprediction. It does
appear, however, that for several days, a displacement in the model's area
of highest ozone may have contributed to the finding of underprediction.
Days 230, 207, 231, and 184 (Fig. 23) appear to fall into this category.
For these days the model predicts a zone of maximum ozone in areas where
no monitors exist.
Constraints in the observational data base, notwithstanding, the
conclusion that the model tends to underpredict at higher ozone concentra-
tion remains. In order to explore potential sources of this problem, the
authors attempted to determine whether the model performance was related to
the type of days being modeled. To address this question, regression anal-
ysis was used to determine whether peak accuracy is related to any of the
critical atmospheric input variables used to simulate the different days.
In this analysis the F4 residual (Omax - Pmax> unpaired in location and
time for each day) is the dependent variable and mean wind speed,* tempera-
ture,* solar radiation,* maximum mixing height and ozone aloft are the
independent variables. The correlation coefficients shown in Table 12
demonstrate that only the wind speed parameter is significantly correlated
to the F4 residual. In this relationship, days with higher wind speeds tend
to exhibit more underprediction. A similar analysis for the A2 (spatially
paired average residuals) results in a significant but smaller correlation
between near-peak bias and wind speed.
*Meteorological parameters are averaged over the period 0700-1400
CST.
51
-------
Significant correlation does not necessarily imply a causal
relationship. Yet, a plausible explanation for this relationship is avail-
able. On days with low wind speed, the model results are controlled largely
by emissions. On the days with higher wind speeds, two additional factors,
advection and boundary concentrations, affect the simulations and model
performance. The additional uncertainty is heightened because the wind
fields which define advection are perhaps the most uncertain of the inputs
used in the Airshed Model. Secondly, boundary concentrations are based on
interpolations made using a limited number of measurements.
The relationship between wind field errors and advection is shown
conceptually in Figure 24. For the low wind speed case, errors in wind
field result in only a small latitudinal displacement of the model "ozone
cloud" and the cloud remains over the urban area with its dense network of
monitors. For the high wind speed case, the dislocation of the model's
urban plume is magnified as it is advected downwind. There is a much higher
probability that the model's zone of maximum ozone will be located in an area
without monitors, a factor that as stated previously, results in apparent
positive bias.
Not only is the location in error, but the emissions encountered
along the trajectory will be different from those encompassed by the actual
urban plume. Thus, the error in wind direction manifests itself as errors
in the location of and magntude of peak ozone.
To further understand the model's performance with regard to ozone,
the authors examined the model's frilly to estimate average concentrations of
NMSIC and HOX nrecursors and "0. Hgure ?.5 shews scatter diagrams of daily
52
-------
average observed and predicted concentrations of these pollutants. The
use of values averaged over all simulation hours for all sites can be
used to make a preliminary assessment as to the accurcy of the precursor
emission inventories.* Furthermore, the results give some insights into the
model's ability to simulate dispersive conditions that affect ground-level
concentration.
From the diagrams in Figure 25, it is evident that the model simu-
lates average NMHC and NOX concentrations with a high degree of accuracy
(near zero bias), correlation, and precision. Since the emissions remain
essentially constant from day to day, substantial errors in emissions should
show up as a systematic bias in the results. The decided absence of bias
suggests that in the aggregate, the NMHC and NOX emission inventories are
accurate. The day-to-day discrepancies between the observed and predicted
average precursor concentrations may be related to errors in the simulation
of advection, dispersion, and in specifying the diurnal changes in mixing
layer height.
Carbon monoxide, on the other hand, appears to be underpredicted
by the model. However, this finding may be related to the fact that a
number of CO monitors may be located near "hot spots." Model predictions,
on the other hand, are grid values averaged over 4 x 4 km squares. There
is also the possibility that the emissions inventory underrepresents the full
*Predicted values were obtained in a similar manner to predicted
ozone concentrations; estimates were derived by averaging values from
cells adjacent to the monitoring site. Both predicted and observed
concentrations were obtained from the ESRL Report.3
53
-------
extent of CO emitted in the region; however, CO unlike NMHC and NOX
has almost no effect on ozone levels.
While the precursor emission inventories appear to be accurate
in the aggregate, we can draw no conclusions about the accuracy of temporal
or spatial distributions of envssions. Similarly, our analysis says nothing
about the accuracy of the NMHC species distribution. Peak ozone is more
dependent on precursor concentrations associated with the morning urban
rush-hour emissions than it is on region-wide daily averages. Nevertheless,
it is reassuring that available evidence suggests that the overall NMHC and
NOX inventories are accurate.
Aside from the concentration of NMHC in the atmosphere, the
NMHC/NO ratio is an important determinant of ozone level. Figure 26 is
a scatter diagram of mean observed and predicted NMHC/NQX ratios (averaged
over all sites for all hours). This diagram indicates that the model has
an overall tendency to overpredict NMHC/NOX; however, there is considerable
day-to-day variation in the accuracy of the predicted ratios. This variation
is reflected in a relatively low correlation coefficient of 0.44. The
tendency toward overprediction is associated with a slight tendency towards
underprediction of MOX and the variation is associated with errors in both
NOX and NMHC.
Given the sensitivity of model peak ozone to the NMHC/NOX ratio
of emissions (demonstrated in Chapter 5), on-- <"'ch!. isk whether the model's
accuracy with regard to peak ozone is relate'! to errors in the NMHC/NOX
ratios. To address this question, :.'••* M residuals (measure of near-peak
-------
accuracy were plotted against residuals of NMHC/NOX. This diagram (Fig.
27) shows a significant relationship (correlation coefficient of .52)
between the residuals of peak ozone and the average NMHC/NOX. Note that
with increasingly large underprediction (+ bias) in NMHC/NOX, peak ozone
levels are underpredicted to a larger extent. The diagram shows that the
model is less likely to underpredict on those days for which the predicted
NMHC/NOX ratios are higher than observed.
This finding leads to the hypothesis that the model needs an
artificially high NMHC/NOX ratio to compensate for deficiencies in the
model and/or inputs, deficiencies which restrict the model's ability to
generate ozone. While this hypothesis is tentative, it is consistent with
several concerns that have been raised regarding the treatment of NMHC
emissions and boundary concentrations. First, vegetative emissions have
been ignored and secondly, it appears that the carbonyl (aldehydes and
ketones) fraction used in the NMHC emissions and boundary concentrations
may be on the low side.
Several studies 22(23 nave shown that vegetative emissions par-
ticularly in rural areas are substantial and that the reactivity of species
emitted is at least comparable to that of urban NMHC mixtures.24 The fact
that the monoterpenes and isoprenes associated with natural emissions have
not been measured in appreciable quantities possibly is related to their
rapid conversion to intermediate species. Thus, vegetative emissions may
have a substantial impact on ozone levels without contributing to the
levels of NMHC measured at urban sites. Thus, the neglect of vegetative
55
-------
factors cannot be ruled out as a contributor to the underprediction of ozone,
not only by St. Louis, but for a number of other applications including Los
Angeles,25 and Denver.26
The second factor, carbonyl fraction, is of particular concern
for St. Louis. As Table 13 indicates, the fraction of emissions for this
reactive group used for St. Louis is considerably less than that used for
other cities and is lower than the fraction that has been measured in
several cities. Whereas the fraction used for St. Louis is about 1.5 per-
cent, SAI, in a recent report, recommends an urban value ranging from 5-10
percent.27 This faction is particularly significant not only because car-
bonyl s are highly reactive species, but also because the fraction of alde-
hydes and ketones add to the total organic emissions rather than merely
redistributing it. This is because most HC emission factors do not include
these compounds.
4.4 Summary of Model i'erfomance
In summary, the assessment of model performance indicates that
the Airshed Model tends to underestimate ozone concentrations for areas and
times of high ozone. The model was found to have a systematic tendency to
underestimate peak and near-peak ozone concentrations. The assessment is,
however, limited by the sparse number of monitors in outlying areas for which
predicted values are high on many of the days.
Several factors contribute to the underprediction of peak ozone
and the aerial extent of exceedance zones. These include: (1) the current
model's inappropriate mixinq of point and possibly line) source NOX into
56
-------
entire grid cells; (2) neglect of vegetative emissions; (3) the use of a
carbonyl fraction which appears to be low; and (4) errors in advective
transfer associated with windfield errors on transport days; and (5) prob-
lems with the chemical mechanism. The evidence indicates that the overall
NMHC (sum) and the total NOX inventories are accurate and there is no cause
to search for additional anthropogenic sources that may have been neglected
from the inventory.
Correlation coefficients indicate that the model replicates
temporal variations accurately. However, while the model is generally in
phase with the observed temporal pattern, it tends to underestimate the
full degree of variation found in the observations. The "flatter" than
observed predicted diurnal curves is explained largely by an underpre-
diction of peak ozone. Paucity of sites in outlying areas restricts
assessment of the model's ability to replicate spatial patterns. How-
ever, comparison of observed and predicted fields appear to indicate that
the model is able to locate the general area of high ozone, but it tends
to underestimate the size of the NAAQS exceedance zone.
57
-------
5.0 Model Sensitivity
5.1 Introduction
Model sensitivity Defers to the resoonse of model outputs to
changes in model inputs. The purpose of sensitivity analysis is to deter-
mine the percent change (y) in a given output variable which results from
a percent change (x) in a gi^en input parameter, holding other factors
constant. In this chapter, the results of applying this simple technique
to Urban Airshed Model simulations in St. Louis are discussed.
While the t3chn1que is staightforward, the interpretation and
significance of the results are made difficult by two separate but related
factors. First, is the question of model performance; base-case simula-
tions must compare favorably with observations before much confidence can
be placed in the model as a predictive tool . A broader question is that
of model validity; changes in base-case simulations should agree with
changes in observations, given concurrent and equivalent changes in model
inputs on the one hand, and changes in physical and chemical features of
the atmosphere on the other. However, assessing model validity in this
sense is rendered nearly impossible by confounding factors which vary from
one day of observations to the next. Second, is the factor of input
interaction. Urban Airshed is a mult.i-variat.e nonlinear model. The
response of such a model to changes in a single output parameter is modu-
lated by the values taken on by other input parameter; as well. For example,
d change in mixing height mf'.y increase or "r -•.»-: the mass of pollutants
entrained into the surface layer from 9loft as well as affect pollutant
concentrations resulting from low iyvel emissions. Thus, the sensitivity
5H
-------
of a photochemical model to emission changes depends on meteorological
parameters, on the concentrations of chemical species at the boundaries
of the modeling region, and on the concentrations used to initialize the
model. Such interactions can modify both the magnitude and direction of
the model response.
Notwithstanding their limitations, model sensitivity analyses
are quite useful for a variety of objectives. These include: (1) iden-
tification of input parameters exhibiting significant influence on model
results, (2) assessment of the types of quantity and basic data from which
model inputs are derived and are needed for successful model applications,
and (3) diagnosis of possible causes contributing to unsatisfactory model
performance. In this study, primary emphasis is placed on the first objec-
tive. Furthermore, much of the study focuses on the response of the model
to changes in emissions in order to assess effects on both area-wide and
peak ozone concentrations. Inferences regarding possible outcomes associated
with regulatory actions necessarily assume a high level of model validity
over a wide range of input values. Regulatory aspects related to emission
changes are discussed in Chapter 6.
Consistent with the model performance results discussed earlier,
the sensitivity results reported here reflect the most recent version of
the Airshed Model and associated base-case preprocessor files. As discussed
earlier, changes were made to the emissions inventory, to various air quality
parameters along the borders of the modeling region and aloft, to initial
conditions, and to the photolytic rate constants input to the model. The
model itself was also revised in two respects: (1) the Carbon-Bond
59
-------
Mechanism was replaced with C.irbon-BondII27>28»29 and (2) the SHASTA
numerical technique originally employed for advection calculations was
replaced with the FCT algorithm developed by Zalesak.30 Extensive
sensitivity testing was conducted using previous versions of the Airshed
Model and St. Louis data base.31,32 These tests included uniform changes
in hydrocarbon and oxides of nitrogen emissions, in the spatial and temporal
distribution of emissions, in the mix of hydrocarbon reactivity classes,
in hydrocarbon and ozone concentrations aloft and at the borders of the
modeling region, and in the maximum afternoon mixing height. However, the
earlier test results cannot be compared to those reported here since the
revisions to the model and preprocessor files led to significant changes
in the base-case simulations.
Some 48 sensitivity tests were conducted using the most recent
version. These may be conveniently grouped and discussed as follows:
(1) uniform changes in hydrocarbon emissions, (2) uniform changes in emis-
sions of nitrogen oxides, (3) uniform changes in the mix of hydrocarbon
chemical classes (referred to hereafter as hydrocarbon reactivity), and
(4) miscellaneous changes in the spatial and temporal distribution of emis-
sions and in meteorological parameters. To date, only effects on ozone
concentrations have been analyzed.
Two concepts employed in this chapter are those of "peak" pre-
diction and "trajectory." The peak prediction is the highest one-hour
average ozone concentration simulated by t»s ~- cnodel from among all hours
and all surface layer grid cells. TV c peak prediction is therefore not
60
-------
constrained to grid cells in which ozone monitoring stations are located.
However, the peak prediction is constrained to the first (surface) layer
of grid cells. The concept of trajectory is employed in interpreting model
results although the Airshed Model is not a trajectory, or lagrangian,
model. The path of a trajectory is determined by the mean wind and may be
constructed using the three-dimensional wind field input to the model. The
actual path taken by transported pollutants is modified by wind shear and
by horizontal and vertical diffusion. Nevertheless, surface trajectories
such as the hypothetical one depicted in Figure 28, do indicate the emis-
sions burden assumed by a particular air parcel and do suggest the source
areas that contribute to high ozone concentrations predicted by the model
for a given location, particularly in the absence of extreme stagnation.
Sensitivity results vary markedly from day to day. Therefore,
before the tests themselves are discussed, the simulation days will be
described.
5.2 Test Days
All sensitivity tests were performed on one or more of 3 separate
test days. These are Monday, June 7, 1976 (Day 159); Tuesday, July 13, 1976
(Day 195); and Friday, October 1, 1976 (Day 275).
Day 159 (June 7, 1976). A large high pressure system associated
with weak pressure gradients dominated the eastern half of the nation. As
the trajectory for this day shows (Fig. 20a), flow during early morning hours
was light and northeasterly. The flow shifted to southeasterly by mid-
morning and transport toward the northwest persisted throughout the day. As
61
-------
Fig. 20a demonstrates, predicted ozone concentrations exceed ZOO ppb over a
large area of grid cells in the northwestern portion of the modeling grid.
The location of max predicted ozone is consistent with the urban trajectory
shown in the figure. The very high predicted maximum concentrations may be
plausible, given the slow passage of the trajectory over regions of dense
NMHC; however, the bias measures restricted to the monitoring sites indi-
cate a tendency toward overprediction especially at high concentrations.
Unfortunately, there were no monitors in the region of peak predicted ozone.
Day 195 (July 13, 1976). Winds were moderate and ranged from
southeasterly in the morning to SW during the afternoon. As Fig. 20b
illustrates, the area of maximum predicted and observed ozone concentra-
tions in the north central portion of the region are consistent with wind
trajectories. However, as bias statistics for this day show, there was a
tendency toward underprediction at high concentrations. Note, for example,
that the observed ozone distribution for 1500-1600 hours suggests a larger
exceedance zone than that of the predicted values.
Day 275 (Oct. 1, 1976). Of the 20 test days modeled, Day 275
recorded the highest observed ozone concentration (244 ppb). This day is
somewhat unusual in that it occurred beyond the usual ozone season; it was
characterized by extreme stagnation conditions, light winds and very low
mixing depths. (The maximum mixing depth for this day is estimated to be
about 500 m). As Fig. 20e shows, observed and predicted concentrations
were maximal near the central urban core. V' . tnere are some discrep-
ancies, the overall patterns of obse^v^-i and predicted concentrations are
62
-------
closely matched. The bias statistics for this day suggest a moderate
overprediction; however, this result is associated with the post-peak
late afternoon and evening hours. The model shows a much slower decrease
in ozone level than was observed at the monitors.
Subsequent to evaluating model performance on these 3 days, a
reexamination of the ambient nonmethane hydrocarbon (NMHC) measurements
used for specifying boundary and initial concentrations was undertaken.
Representing the difference between two relatively large numbers, total
hydrocarbon and methane, considerable uncertainty exists in the reported data.
In obtaining an hourly average NMHC value, spurious negative 1-minute read-
ings were eliminated. Also, a background value of 0.050 ppm C was assumed
so that the hourly average NMHC was never allowed to fall below this. In
addition, on Day 275, NMHC along the southern boundary was reset to a con-
stant 0.050 ppm. These changes in hydrocarbon concentrations along the
borders of the modeling region did not affect peak ozone concentrations on
Days 159 and 195. However, on Day 275 the peak was reduced from 246 ppb to
232 ppb. The revised simulation is used as the base case for all sensitivity
analyses.
5.3 Uniform Changes in Hydrocarbon Emissions
In these tests, uniform reductions of 5, 17, 42, and 75 per-
cent and increases of 1 and 6 percent were made to the hydrocarbon emissions.
All five carbon-bonds (PAR, ARO, OLE, ETH, CARB) were reduced or increased
by the same percentage even though, strictly speaking, CARB represents
aldehydes rather than hydrocarbons. This was done for each grid cell each
63
-------
hour and for all sources. In addition, concentrations of the five carbon-
bonds used for initialization were also reduced or increased by a like amount,
but only after subtracting out an assumed background value. The background
values were as follws:
PAR 33 ppbC
GARB 4 ppbC
OLE 4 ppbC
ARO 7 ppbC
ETH 2 ppbC
The adjustment to initial conditions was intended to represent the expected
effect of emission reductions on early morning ambient hydrocarbon concen-
trations. The background values could represent uncontrolled anthropogenic
sources or biogenic sources of emissions; no distinction is intended.
Hydrocarbons at the borders of the modeling region and aloft were left
unchanged.
The effect of these changes on peak ozone predicted anywhere in
the modeling region during the course of the simulation is tabulated in
Table 14 and illustrated in Figure 29. On all 3 days the change in peak
ozone concentration is in the same direction as the change in hydrocarbons
but is invariably of a smaller magnitude. However, the relative change in
ozone concentration for the same change in hydrocarbons varies widely among
the 3 days. This is readily apparent from the slopes of the curves shown
in Figure 29. Day 159 is particularly insensitive to increases in hydro-
carbon emissions. This behavior may be associated v/ith the high levels
of ozone aloft and a relatively high afternoon mixing height, the combined
64
-------
effect of which is a large influx of transported ozone into the urban plume.
The addition of more hydrocarbon to such a system may be more effective at
removing NOg (via PAN formation) than in generating additional ozone. Day
195 is the least sensitive to large reductions in hydrocarbon emissions.
This day, which experienced relatively higher wind speeds, as well as good
vertical mixing, shows the lowest predicted precursor concentrations of
the 3 days. Boundary concentrations of NMHC, therefore, may contribute
a larger fraction of the predicted NMHC concentrations, a fraction which
is not reduced by reductions in hydrocarbon emissions. Among the 3 days,
Day 275, having the poorest overall ventilation and the lowest ozone
aloft, exhibits the greatest sensitivity to both increases and decreases
in hydrocarbon emissions.
The results for these simulations may also be presented in terms
of changes in the hydrocarbon to oxides of nitrogen emissions ratio.
Figure 30 illustrates the relationship between the peak ozone prediction
and the HC/NOX ratio. The ratio is defined as the total moles of the five
carbon-bonds, expressed as carbon, divided by the total moles of NO and
N02- The totals are across all grid cells and all hours of the day.*
With NOX held constant (open symbols), peak ozone predictions invariably
increase with increasing HC/NOX ratio, but to varying degrees. A sug-
gestion of asymptotic behavior at large ratios is apparent on Day 195.
However, on Day 159, this behavior is strikingly evident. In these
*The emissions ratio so defined is much lower than ambient ratios
observed at the ground, due to the inclusion of a large mass of elevated
NOX emissions which are not instantaneously mixed to the surface.
65
-------
tests, increases in hydrocarbon emissions were limited to 67 percent.
Whether and to what extent further increases might actually suppress ozone
formation is indeterminant. Furthermore, whether the curve seen on Day
159 is due to peculiarly high HC/NOX ratio along the particular trajectory
to the peak prediction, to a high influx of ozone from aloft, mollifying
the effect of further hydrocarbon increases as suggested earlier, or to
possibly both cannot be determined without further analysis.
Uniform changes in hydrocarbon emissions affect the time and
loccition of the peak o:cne o ediction as well as the value itself. In
general, decreases in hydrocarbon emissions tend to move the ozone peak
further downwind and shift it. to later in the day. This is likely associ-
ated with a delay in the conversion of NO to M02- Conversely, increases
in hydrocarbon emissions move the ozone peak closer to the urban center
and shift it to earlier in the day. Changes in the timing of the buildup
of ozone would directly affect both the location and the time of the ozone
peak; a delay in the buildup would move the peak further downwind and shift
it to later in the day. On 'Day 159, the ozone peak moves approximately 25
kilometers to the northwest, away from the city, and is delayed 5 hours
as emissions are changed from +67 percent to -42 percent. However, a
redaction of -75 percent has little effect on the peak locations but the
time of the peak is accelerated 3 hours, relative to the -42 percent case.
A closer examination shows that the peak for the -42 percent case occurs
the last hour of the simulation (1800-1900 " "> Dearly at the boundary of
the modeling region. Along this particular trajectory, a further delay in
ozone buildup could move the peak h-;-,ond the modeling region or past the
-------
end of the simulation or, on the other hand, it might have no effect on
the timing and location of the peak if the rate of N02 photolysis drops
rapidly, effectively halting the ozone buildup. That the peak prediction
occurs 3 hours sooner when hydrocarbon emissions are reduced further to
-75 percent is strong evidence however, that an entirely different tra-
jectory, carrying a different burden of hydrocarbon and oxides of nitrogen
emissions, in fact leads to the peak ozone prediction.
A subset of this group of sensitivity tests was further analyzed
to examine the effect of hydrocarbon emission changes on area-wide ozone
predictions. The entire set of model predictions for a given sensitivity
test was contrasted with the base case using several statistics. These
included the mean absolute normalized residual and the mean normalized
residual, the latter computed as the base-case prediction minus the sensi-
tivity case prediction. Since the modeling region is 15 by 20 cells and
the simulation is run for 14 hours, these residuals are taken over 4200
data pairs. Only predictions in the surface layer are considered. In
addition, a correlation coefficient which measures the similarity of the
spatial alignment of the two sets of predictions was computed in two
steps. First, a 1-hour Pearson product-moment correlation coeffcient
was computed for all grid cells. Second, because the hourly correlation
coefficients are not normally distributed, a change of variable was
made.53 This procedure led to computation of the average population
spatial correlation coefficient.
These statistics are given in Table 15 for Days 159 and 195
for three sensitivity tests: +67 percent, -42 percent, and -75 percent.
67
-------
The mean normalized error, as used here, represents the average relative
change in predicted ozone concentrations, when the sign is reversed.
Changing the sign in Table 15 and converting to percentages, it is readily
apparent that average ozone concentrations are, in most cases, changed
much less than are peak ozone concentrations, as given in Table 14. This
is illustrated better in Figure 31 where the average percentage change, the
percentage change of the peak, and a reference 1:1 line are plotted for
Days 159 and 195. It is noteworthy, however, that on Day 159, average
ozone concentrations are increased more than is the peak ozone concentra-
tion when hydrocarbon emissions are increased by 67 percent. This behavior
would tend to suggest that the insensitivity of the peak ozone prediction
to increases in hydrocarbon emissions is associated more with the partic-
ular HC/NOX ratio along the trajectory to the peak than with ozone aloft.
Tables 14 and 15 demonstrate that day-to-day differences in model sensi-
tivity tend to be much less for average ozone concentrations than for peak
concentrations.
The effect of uniform changes in hydrocarbon emissions on com-
puted correlation coefficients between the base case and sensitivity case
is illustrated in Figure 32. The average sample spatial correlation
coefficient is reduced as hydrocarbon emissions are either decreased or
increased. Had all ozone predictions for a given sensitivity case been
reduced or increased by the same amount, the correlation coefficient
would be 1.0. The drop in the correlation coefficient, therefore, reflects
changes in the spatial patterns of predicted ozone concentrations.
58
-------
Further analysis of this subset of sensitivity runs was made to
evaluate the effect on are a-wide exceedances of the ozone National Ambient
Air Quality Standard (NAAQS) of 0.12 pm. For this purpose, the overall
normalized area! exceedance (ONAE) statistic was derived. The ONAE is
defined as the difference between the base case and sensitivity case number
of grid cells that equal or exceed 0.12 ppm during any hour of the simu-
lation, normalized by the base case number of grid cells which exceed 0.12
ppm during any hour of the simulation. The values taken on by this statis-
tic, converted to percentages, are given in Table 16. An ONAE of -100
percent indicates a doubling in the numnber of grid cells exceeding the
NAAQS while an ONAE of +100 percent indicates that no cells in the sensi-
tivity case exceed the NAAQS. A plot of these data in Figure 33 emphasizes
again the relatively large day-to-day differences. As expected, the ONAE
shows a highly nonlinear behavior. On Day 195, the ONAE approaches the
4-100 percent value ("attainment") asymptotically. On this day, a 75 per-
cent reduction in hydrocarbon emissions has the effect of reducing all
model predictions below the NAAQS. On Day 159, the data points are too
widely separated to define the model's behavior as the ONAE approaches
-i-lOO percent; two possible curves are illustrated. If the model behaves
asymptotically, as on Day 195, it appears that no amount of emissions
reduction would reduce all model predictions below the NAAQS. Given the
level of ozone aloft (0.114 ppm), such a result would not be surprising.
5.4 Uniform Changes in Oxides of Nitrogen Emissions
In these tests, reductions of 20 percent and increases of 20
percent were made to the oxides of nitrogen emissions on Days 159, 195,
69
-------
and 275. In addition, a reduction of 40 percent and an increase of 40 per-
cent were made on Day 275. Each of these reductions was combined with a
42 percent decrease in hydrocarbon emissions. In implementing these changes,
both nitric oxide and nitrogen dioxide were reduced or increased by the same
percentage for each grid cell each hour and for all sources. Besides emis-
sions, concentrations of NO and N0£ used for initialization were also
reduced or increased by a like percentage, but only after subtracting out
an assumed bckground value. The background values were as follows:
NO 3.0 ppb
NO? ?.0 ppb
The adjustment to initial conditions was intended to represent the expected
effect of emission changes on early morning ambient oxides of nitrogen con-
centrations. These background values could represent "uncontrolled" anthro-
pogenic emission sources or other sources; no distinction is intended.
Oxides of nitrogen at the borders of the modeling region and aloft were left
unchanged.
Because NOX changes were combined with HC reductions, the base
case for this set of tests was changed from the original base case listed
in Table 14. Instead the sensitivity test in which hydrocarbon emissions
wp»e reduced by 42 percent, as discussed previously, was used. Noteworthy
is the greatly reduced HC/NOX emission ratio of the new base case relative
to the original. The ratio dropped from 4.6 to 7,1 or Da,> 159, 4.2 to 2.5
on Day 195, and 5.4 to 3.1 on Day 275, M: >''• •.-. "-:-:
-------
ozone to changes in oxides of nitrogen emissions has not been examined.
Model predictions may be more sensitive to changes in NOX at higher hydro-
carbon levels when NOX is more limiting. Ideally, uniform changes in
oxides of nitrogen should be made at various hydrocarbon levels.
The effect of these tests on peak ozone predicted anywhere in
the modeling region during the course of the simulation is tabulated in
Table 17 and illustrated in Figure 34. On all 3 days, the change in peak
ozone concentration is in the opposite direction as the change in oxides
of nitrogen emissions. Decreases in oxides of nitrogen emissions lead
to higher peak ozone predictions. Furthermore, the magnitude of the change
in peak ozone is in all cases less than the magnitude of the change in
emissions. However, there is again considerable variability from day to
day. Day 195 shows the least sensitivity to oxides of nitrogen emissions
which may be related to the relatively greater importance of estimated
boundary conditions on a day characterized by advective transport. Day
159 shows the greatest sensitivity to increases in oxides of nitrogen
emissions, while Day 275 shows the greatest sensitivity to decreases.
This behavior is consistent for the 3 days with the relation-
ship established previously between the HC/NOX emissions ratio and the
peak predicted ozone concentration for constant oxides of nitrogen emis-
sions, as illustrated in Figure 30. The relationship between HC/NOX
ratio and peak ozone at constant hydrocarbon emissions is also shown in
Figure 30. (On each curve connecting the solid symbols, "B" indicates
the base case.) On Day 159, it can be seen that a decrease in the HC/NOX
71
-------
ratio (increase in NOX) gives the steepest slope, one which is nearly
identical to the HC/NOX curve established by changing HC at constant NOX.
However, a decrease of 20 percent in oxides of nitrogen emissions (increase
in HC/NOX ratio) on Day 275 produces the highest slope of the 3 days, but
a further decrease of 40 percent leads to a greatly diminished slope, and
therefore sensitivity. In general, for the limited range of hydrocarbon
and oxides of nitrogen levels examined here, changes in HC/NOX ratios at
constant HC have less effect on peak ozone concentrations than do those
at constant NOX. This is particularly evident as oxides of nitrogen are
decreased. For the case in which decreases of 40'percent in oxides of
nitrogen emissions on Day 275 are accompanied by decreases in hydrocarbon
emissions of 42 percent, the HC/NOX ratio remains essentially constant,
yet a reduction of 24.1 percent from the base case in Table 14 is seen in
the peak ozone prediction. Thus, while the behaviors at constant HC and
constant NOX are consistent, the HC/NOX emission ratio alone does not con-
vey a complete picture; the absolute level of hydrocarbon and oxides of
nitrogen emissions are primary determinants of the ]3eak ozone concentration
on any particular day.
Another way of viewing the results of these tests is to examine
how changes in oxides of nitrogen emissions modify the effect on peak ozone
of changes in hydrocarbon emissions. This is illustrated in Figure 35.
Here, decreases in oxides of nitrogen emissions are seen to diminish the
effect of a 42 percent reduction in hydrocarbon emissions on all 3 days.
Conversely, increases in oxides of nitrogen enhance the effect that
hydrocarbon emissions reductions have on reducing peak ozone predictions.
-------
Whether this same effect might occur at other hydrocarbon levels has not
been studied.
Of interest also is the effect uniform changes in oxides of nitro-
gen emissions have on the time and location of the peak ozone prediction.
In general, decreases in oxides of nitrogen emissions accelerate the time of
the peak predicted ozone concentration and/or translocate the peak closer
to the urban center. On Day 275, no well-defined advective transport
occurs and no clear pattern is exhibited in the movment of the peak ozone
prediction across the modeling grid. For the same reasons that decreases
in oxides of nitrogen accelerate the peak, increases are expected to delay
the peak. However, this does not occur. Instead, on Days 159 and 195
the peak occurs earlier in the day at about the same location as the base
case. This suggests that another trajectory carrying a different burden
of hydrocarbon and oxides of nitrogen emissions leads to the peak prediction.
Photochemical activity along the trajectory is sufficiently suppressed so
that it no longer produces the peak ozone concentration. Although the peak
prediction for the base case occurs near the downwind edge of the modeling
region, it also occurs during the late afternoon on both Days 159 and 195,
making it unlikely that extension of the modeling region would have identi-
fied a higher peak further downwind. On Day 275, increases in oxides of
nitrogen emissions have no effect on the timing of the peak and only a 40
percent increase in oxides of nitrogen has any effect on the location.
The results of these sensitivity tests suggest that controls on
oxides of nitrogen emissions are counterproductive with regard to reducing
73
-------
ozone concentrations and that allowing oxides of nitrogen to increase will
enhance th-r benefits achieved by lowering hydrocarbon emissions. However,
results with the Airshed Model in Los Angeles indicate that NOX controls are
necessary to achieve significant ozone reductions in the eastern (downwind)
part of the South Coast air basin. Whether such an effect might be seen in
St. Louis had the influence cf secondary nitrogen compounds carried over from
one day to the next or transported long distances downwind been considered,
has not been addressed. Except during episodes of extreme stagnation, such
* -i)ult: -lay simulation >,ld require qre-itly increasing the size of the
modeling region. Use of a trajectory model would be the only practical means
of evaluation in this event. Moreover, it should be strongly emphasized that
possible adverse effects on nitrogen ilioxide has not been analyzed.
5.5 Uniform Changes in Hydrocarbon Reactivity
The Carbon-Bond II kinetics mechanism incorporated in the Airshed
Model recognizes five cabon-bond categories which serve as surrogates for
the wide variety of hyrccarbon compounds emitted into the atmosphere. The
five carbon-bonds are distingushed by their markedly different reaction path-
ways and reaction rates These differences lead directly to differences in
their relative potential for contributing to the formation of ozone. The
particular mix of carbon-bonds therefore defines the photochemical potential
of the hydrocarbon emission?;. This nix is what is referred to here as
"hydrocarbon reactivty." Typically, the more reactive a particular mix,
the more ozone that is predicted by the ki,~ ; : -
Table 18 lists the mix o4" -: . .-.;ondss in tennis of the carbon
fraction of total hydrocarbon r.-iissK-'is, used in these sensitivity tests.
74
-------
The fractions were developed from different sources of information and in
part represent differences in the relative contribution of different source
types to the overall carbon-bond mix and in part different methods of
treating individual hydrocarbon species for input to the carbon-bond
mechanism.* Whether these differences are real or methodological is not
of great concern in the context of sensitivity testing; it is the effect of
these differences regardless of their origin which is.
In applying these carbon fractions to the hydrocarbon emissions,
the base case emissions of each carbon-bond must first be summed and then
redistributed. In so doing, the number of carbons associated with each
carbon-bond category must be taken into account. The total reactive
hydrocarbon emissions (in gram-moles carbon) for each grid cell each hour
was obtained as follows:
RHC = PAR + (2 * OLE) + (2 * ETH) + (6 * ARO) + CARB
The total hydrocarbon was then redistributed to carbon-bonds as follows:
PAR = C] * THC
OLE = 1/2 * C2 * THC
ETH = 1/2 * C3 * THC
ARO = 1/6 * C4 * THC
CARB = C5 * THC
where C], C2, C3, C4, and C$ are the carbon fractions corresponding to PAR,
OLE, ETH, ARO, and CARB from Table 18. This procedure resulted in a uniform
*In the Philadelphia and Tulsa inventories, for example, cycloparaffins
and internal olefins were treated as surrogate carbonyls (CARB). In St.
Louis, these were simply treated as paraffins or olefins.
75
-------
mix of carbon-bonds for each grid cell each hour. This procedure was also
used to redistribute initial concentrations. No modification was made to
hydrocarbon concentrations at the borders of the modeling region or aloft.
The first simulation listed in Table 18 uses the overall St. Louis
regional carbon-bond mix, based on the emissions from all sources for all
hours combined. This simulation, which serves as the base case for the
next three simulations, was conducted to assure a uniform distribution of
carbon-bonds and thereby eliminate temporal and spatial differences in the
actual distribution of oarbos1-bonds. The second and third simulations in
Table 18 use the overall regional carbon-bond mix derived from emission
inventories compiled for Tulsa and Philadelphia. The fourth simulation
uses a carbon-bond mix derived from the average composition of automotive
exhaust as determined from measurements of individual hydrocarbon species
taken by the Bureau of Mines (BOM).
The effect of these uniform changes in hydrocarbon activity on
peak ozone predictions on Days 195 and 275 is summarized in Table 19. The
hydrocarbon reactivity derived from the Philadelphia inventory produces
modestly higher peak ozone predictions on both days. That derived from the
Tulsa inventory produces onl.y slightly less ozone than the base case. How-
ever, a uniform reactivity corresponding to the EOF automobile exhaust data
produces markedly higher peak, ozone predictions, particularly on Day 275 where
nearly a 30 percent increase is computed. Day 275 also shows significantly
greater sensitivity than Day 195 when the Phil 9 do I f:'m'a hydrocarbon reactivity
is used. This behavior is consistent with t^e relatively greater importance
of estimated boundary concentrations on Day 195. However, Day 195 exhibits
76
-------
greater sensitivity when using the Tulsa hydrocarbon reactivity though
the difference is small. While these tests are very limited in number,
the results suggest that the reactivity of the hydrocarbon emissions,
as determined by the carbon-bond mix, does significantly influence
peak ozone predictions. Furthermore, significant differences result
from differences in reactivity representative of a variety of emissions
data bases. These changes in reactivity originate in two ways: (1) the
particular combination of source types, and (2) the method by which emis-
sions are classified for inclusion in the carbon-bond kinetics mechanism.
Either or both can have a significant impact on ozone predictions.
Another method of exploring the sensitivity of the Airshed Model
changes in hydrocarbon reactivity would be to vary the amount of each carbon-
bond, one at a time. However, in order not to alter the total carbon emitted,
one or more other carbon-bonds would have to be reduced the same amount.
While these kinds of sensitivity tests have not been performed utilizing the
present version of the Airshed Model and the revised St. Louis preprocessor
files, some indication of the relative sensitivity of ozone predictions to
the individual carbon-bonds can be gained by further examination of Tables
18 and 19. It is apparent from Table 18 that only small differences in the
ethylene (ETH) carbon fraction exist among the four simulations; the range
is 4.1 to 5.4 percent. Paraffinic carbon, represented by PAR, is recognized
as the least reactive carbon. Therefore, the differences in peak ozone seen
in Table 19 must be associated with the relative proportion of OLE, ARO, and
CARB. Broadly speaking, these represent the characteristic bonds of olefins,
aromatics, and aldehydes, respectively.
77
-------
The BOM simulation has the highest percent (30.7) of olefins
aromatic?, and aldehydes combined, and the highest peak ozone concentra-
tions. The Philadelphia simulation, having the next highest percent (21.9)
of olefins, aromatics, and aldehydes combined, has. the next highest peak
ozone concentration on both days. By contrast, the sum of the olefins,
aromatics, and aldehydes for1 Philadelphia (21.9 percent) is about the
same as for St. Louis (21.2 percent1 but the peak ozone concentrations
for nhe <'""» < • aclol phia simular.ions are higher. Aromatics and olefins are
sOTiev-hr/: lowe«~ In Phil i^lpliia (13.2 and ?.8 percent) than in St. Louis
(]4/> an.i # 8 ov'rcanij. im; (najor d" fferenco is 'n the aldehyde fraction,
which is almost four times higher in Philadelphia (5.9 percent) than in
St. Louis (1.5 percent).
Contrasting the St. Louis and Tulsa simulations, the sum of the
olefins, aromatics, and aldehydes is substantially greater in St. Louis
(21.2 percent) than in Tulsa (15.8 percent). Moreover, the sum of the
olefin and aromatic c.-D-nn fractions alone is greater for St. Louis (19.7
percent) than for Tulsa (12.2 percent). Nevertheless, on both days, the
simulations for Tulsa and St. Louis produce similar peak ozone concentra-
tions. Th1s result appears then to be associated with the fraction of
a1 'leV/d^s, which is lore tK-n twice a~. great in Tulsa (3.6 percent) as
in St. Louis (1.5 percent).
In all four simulations, the CARB f,-?- "!,,s- -is less than six
percent. Despite its relatively smaVi ^ -jltcn to the total mass of
hydrocarbon emissions, the weich-. .'^idence points strongly to the
role nlaved by aldehvo-- ,. -»•«•* other o>. -^e^.atei compounds which are
-------
included in the CARB carbon-bond category, in the formation of ozone in
urban areas. These results support the suggestion of Kill us and Whitten,^
developers of the Carbon-Bond Mechanism, that special attention be given
to the level of oxygenated compounds in emission inventories. By exten-
sion, similar attention should be given to the level of oxygenated
compounds in ambient air.
Further sensitivity tests on days 195 and 275 were conducted
to examine the effect of hydrocarbon reactivity on the change in peak pre-
dicted ozone associated with uniform reductions in hydrocarbon emissions.
For this purpose, the most reactive carbon-bond mix, the BOM, was used
in combination with a 42 percent reduction in hydrocarbon emissions.
As can be seen from Table 20, greater hydrocarbon reactivity leads to
only slightly greater sensitivity to decreases in hydrocarbon emissions
than is shown by the base case. This small difference could possibly
be explained by spatial and temporal differences in the carbon-bond
mix of the base case, which have been eliminated in the sensitivity
case, rather than the overall change in hydocarbon reactivity. Spatial
and temporal effects are discussed further in Section 5.6.
5.6 Other Sensitivity Tests
A variety of other sensitivity tests were also performed as
listed in Table 21. These tests were designed to examine the sensi-
tivity of Airshed to the spatial and temporal distribution of emissions,
to changes in mixing height, and to changes in photolysis rates.
79
-------
5.6.1 Spatial Distribution
The first of these tests replaced the grid cell by grid
cell variation in emissions implicit in a spatially disaggregated inventory
with a uniform distribution. The total! hourly ground-level emissions (from
highway mobile sources, nonhighway area sources, and minor point sources)
from all grid squares combined were divided equally among the 300 grid
squares in the modeling regioi. Major point source emissions were dis-
tributed equally among 55 fictitious point sources within each of the
four vertical layers tc be simulated. These 55 point sources were then
randomly distributed among the 300 grid cells during each hour and within
each vertical layer. This finely detailed randomization, in combination
with vertical wind shear and horizontal diffusion, should produce a
nearly uniform distribution of emissions in all vertical layers. All
emission species treated by the Airshed Model were redistributed in this way.
Table 21 indicates that a rather large reduction in peak
ozone is predicted for both Days 195 and 275. The percentage decrease is
nearly the same on each day (18.9 and 18.5 percent). A substantial reduc-
tion in the peak ozone levels is expected since the primary effect of the
redistribution is to dilute ';he high-level emissions which occur in the imme-
diate urban environs. As suggested by Figure 29, a decrease in precursor
concentrations reduces the peak ozone prediction. This occurs even when
the HC/MOX emissions ratio i> held constant, as indicated by the 24.1
percent drop on Day 275 when hydrocarbon efo ;?-;o>is were reduced by 4 per-
cent and oxides of nitrogen hy 40 p^ren*. The time and location of the
peak is also affected by the redistribution, fJn Day 195, the peak occurs
80
-------
nearly 20 km further downwind and one hour later. On Day 275, the peak
occurs an hour earlier but in the northeast corner of the region, outside
the immediate influence of the trajectory illustrated in Figure 20c for
the base case.
A second test replaced the hourly and spatially varying
mix of carbon-bonds with a uniform mix equivalent of the overall regional
mix based on the emissions from all sources for all hours combined. (This
test was discussed previously in Section 5.5 but was not contrasted with
the original base case. Table 21 shows that the peak ozone predicted is
modestly reduced on Day 275 (6.9 percent) but only slightly lessened on
Day 195 (1.1 percent). The uniform hydrocarbon reactivity has no effect
on the time or location of the peak on Day 195. However, on Day 275 the
peak is delayed two hours.
5.6.2 Temporal Distribution
A single sensitivity test was conducted on Days 195 and
275 to examine the sensitivity of peak ozone predictions to a temporal
redistribution of emissions. In this test, late morning emissions of all
pollutants were displaced to midmorning hours; namely, emissions from 9:00
to 10:00 were shifted to 6:00 to 7:00, emissions from 10:00 to 11:00 were
shifted to 7:00 to 8:00, and emissions from 11:00 to 12:00 were shifted to
8:00 to 9:00, Central Standard Time. The overall effect of this test was
to increase precursor concentrations during the morning rush hours and to
eliminate all fresh emissions for the hours immediately following. The
results are given in Table 21. As would be expected from an increase in
the morning emissions peak, maximum predicted ozone concentrations increase
81
-------
for both days (5.7 and 21.1 percent). The effect is less on Day 195
when winds were higher and precursor concentrations lower than on Day 275.
On Day 195 the p^ak ozone prediction is advanced 1 hour; however, the
peak occurs at a location some 20 kilometers fin .her downwind. Entirely
different trajectories must therefore be associated with the peak prediction
for the base case and sensitivity simulations. On Day 275, two peaks are
predicted; the early peak is seen at the same hour as the base-case simu-
lation. !'hr- ?eron;' peak, which is somev-hat greater (28*i ppb vs. 276), occurs
i boL-"% -{I*- •". *] though th-se spatial ;!nd tPmporai effects are difficult
to explain, th^ c'-vrall effect of highe* morning rush-hour emissions is an
increase in the peak ozone prediction.
5.6.3 Mixing Height
Two sensitivity tests examined the effect of uniformly
increasing or decreasing the mixing height by 25 percent throughout the day.
Because the mixing height is tie key to defining the vertical structure
of the modeling region i:1 i';e Airshed simulations, these changes can be
expected to have pronounced effects on other model inputs which in turn
influence ozone. For example, the region top is defined with respect to
the mixing height; the treatment of elevated emissions is influenced by
it; •• id ent>vj inment of pollutants }iofi. is directly oroportional to changes
in the height of the mixed layer. The sensitivity test. reou!4;s are tabu-
lated in Table 21 and illustrated in Figure 36. In al cas-iC, a change in
the Mixing he""']-*:. inrr°eses the peak o^one "'• • •.. On Day 195, peak
o/.one is insensitive to mixing height: ^<-*:* i "-creases and decreases of 25
percent produce ve?>y small incre.^ . ' ,1 arid 1.7 percent). A quite large
-------
increase in the peak ozone concentration (33.6 percent) is predicted on
Day 275 when mixing heights are reduced, but an increase though small (1.7
percent), is also seen when mixing heights are increased. In the absence
of entrainment aloft and ignoring elevated point-source plumes, an increase
or decrease in the mixing height is equivalent to an increase or decrease
in dilution. As suggested by earlier sensitivity tests, the more dilute
the mix of precursors, the less ozone is subsequently formed. However,
ozone aloft on days 195 and 275 was significant, estimated as 78 and 6 ppb
respectively. For both days it appears then that the effect of increased
ozone entrainment associated with an increase in mixing height offsets the
reduction in morning precursor concentrations. In contrast, a 25 percent
decrease shifts the balance in the opposite direction; the loss of ozone
aloft appears to be outweighed by the increase in morning precursor concen-
trations. Further analysis is required to explain why the same relative
reduction in vertical mixing produces vastly different effects on peak
ozone concentration. A possible clue may, however, lie in the morning
(7:00 to 8:00 CDT) inversion height which was estimated as approximately
420 m on Day 195 and 120 m on Day 275, and its relation to the height of
elevated point source plumes of oxides of nitrogen. It is conceivable
that a 25 percent reduction from 120 meters may have had a much larger
impact on surface HC/NOX ratios than did the same reduction from 420 meters.
Regardless, somewhat more predictable effects are seen on the time and
location of the peak ozone concentration. A 25 percent increase in mixing
height, which has little impact on the magnitude of the peak, delays the
timing of the peak 2 hours on Day 275. A 25 percent decrease on Day 195
accelerates the peak by 1 hour, although it occurs over 20 kilometers
83
-------
further downwind. Decreases on Day 275 and increases on Day 195 have no
effect on the time and location of the peak prediction.
5.6.4 Photolytic Rates
A final test explored the sensitivity of the Airshed
Model to changes in the photolytic rate constants utilized by the Carbon-
Bond II Mechanism. While the major such reaction is the photolysis of
nitrogen dioxide (NC^), which leads directly to the production of ozone,
other important photolyzable species included in the mechanism are aldehydes
(CARB) and glyoxals (G!.
-------
while a decrease of 10 percent delays it 2 hours. On Day 195, a 10 percent
increase has no effect on the time or location of the peak prediction whereas
a 10 percent decrease delays it 1 hour and moves it 5 kilometers downwind.
5.7 Discussion
In the foregoing sections, the sensitivity of the Airshed Model
to uniform changes in hydrocarbon and oxides of nitrogen emissions, to
uniform changes in hydrocarbon reactivity, and to changes in the spatial
and temporal distribution of emissions was discussed. Several additional
tests were discussed which dealt with two meteorological variables (mixing
height and solar insolation, (the latter expressed via photolysis rates).
While much information has been obtained, large gaps still exist in
understanding the sensitivity of the Airshed Model.
Numerous sensitivity tests were made using previous versions of
both the Airshed Model and the preprocessor files which were developed from
the St. Louis data base. A major sensitivity study was also conducted on
the Airshed Model using a data base for Los Angeles.2^ However, only the
study reported here utilizes the most recent version of the Airshed Model,
one incorporating both the Carbon-Bond II kinetics mechanism and the FCT
algorithm for numerical advection. As was pointed out earlier, these changes
had significant effects on ozone predictions for the St. Louis base-case
simulations. More significant was the effect these changes had on the
sensitivity of the model. The effects on sensitivity were frequently unex-
pected, sometimes large, and most often unexplained to date. Results of
previous sensitivity tests must, therefore, be viewed with a large measure
of caution.
85
-------
Despite the emphasis placed on the sensitivity of the Airshed
Model to emissions changes, this area is far from being fully explored,
Mcst notable, perhaps, is the combined effect of both hydrocarbon and
oxides of nitrogen changes. Specifically, the response of the Airshed
Model to changes in oxides of nitrogen emissions at various levels of
hydrocarbon emissions has not been examined. The sensitivity of the
Airshed Model to the spatial and temporal resolution of emissions has
not been clearly established either. The tests reported here represent
radica1 departures frc^ .-.rtual spatial and temporal emissions patterns.
These tests do not address the questions of what size grid squares
ought to be or for what time intervals emissions ought to be estimated
(hourly vs. some longer or shorter interval).
Not investigated with the current version of the Airshed Model
is the entire area of air quclity data: initial conditions, boundary
conditions along the borders of the modeling, and boundary values aloft.
Although initial conditions viere varied in the tests reported here, they
were varied along with emissions. Of particular interest for providing
guidance on modeling and establishing rational data requirements are the
sensitivity of the Airshed Model to concentrations of hydrocarbons, oxy-
gen^ ted crganics (predominantly aldoh/des), and ozone aloft. Also of con-
siderable interest are the effect of various levels of hydrocarbons, oxy-
genated organics, and organic nitrates (primarily ;'AM) present in urban
air in the early morring hours, corceotr.-f; •- '.• serve a:, initial
conditions for the model.
-------
The area of meteorological inputs is also unchanged. Although
the meteorological tests discussed in Section 5.6 do show considerable
sensitivity to photolysis rates and great sensitivity to mixing heights
under certain circumstances, such across-the-board uniform changes in
these parameters provide only very limited insight into the role which
differences in the richness of meteorological data or the methods by which
it is interpreted, analyzed, and prepared for model input have on model
results. For example, a single technique for developing a three-dimen-
sional wind field based on surface wind and temperature data and upper
air wind data was used for all Airshed Model simulations in St. Louis.
Basic to a numerical model such as Airshed is the grid cell
structure. The horizontal and vertical dimensions of each grid cell are
established early in the modeling process. (The horizontal dimension of
grid cells for modeling is to be distinguished from the size of emissions
grid squares which need not necessarily coincide). What these grid cell
dimensions should be has not been systematically investigated. Instead,
perceived computational constraints have decided the issue. The extent
to which grid cell dimensions should be tailored to the type of meteoro-
logical events being modeled has not been considered. Whereas relatively
large grid cells may be sufficient for stagnation conditions, more severe
requirements may be needed regarding the resolution of grid cells on trans-
port days, a hypothesis which is consistent with the relative performance
of the Photochemical Box Model (PBM) on different days in St. Louis.3
Wind shear may well be another factor which should be considered in estab-
lishing the vertical dimensions of grid cells. What effect changes in the
87
-------
in the grid cell structure have on both model sensitivity and model
performance could be assessed by further sensitivity analysis.
Resolution of at least two technical issues could also be
promoted by additional sensitivity testing. The first of these is biogenic
emissions of various hydrocarbon species. Biogenic emissions, components
of which have been shown in smog chamber studies to be as reactive as a
typical mix of urban anthropogenic emissions-^ and other components of
which have been shown to be relatively more reactive in terms of their
ozone formation potent:3!,^ have not been included in any of the Airshed
Model simulations in St. Louis. Of compelling interest is the extent to
which these compounds modify the sensitivity of the model to changes in
anthropogenic emissions of both hydrocarbons and oxides of nitrogen. A
second technical issue is the aromatics chemistry of the Carbon-Bond II
mechanism which has recently been updated and incorporated in what is now
Carbon-Bond III. To what extent this new understanding of the chemistry
of aromatics may influence model predictions in the base case or, more
importantly, the sensitivity of the Airshed Model to emissions changes or
other changes in inputs can only be assessed by further sensitivity tests.
88
-------
6.0 Estimating Control Requirements
This chapter demonstrates the use of the Airshed Model to assess the
types of programs that may be required in order for the St. Louis AQCR
to meet the NAAQS for ozone. It is emphasized that this study is not
intended to prescribe specific control programs relative to State Imple-
mentation Plans (SIP's) but rather to demonstrate the feasibility of the
Airshed Model approach.
Two approaches were used to estimate the type of programs that may be
required for the St. Louis Air Quality Control Region to meet the NAAQS for
ozone. First, a series of simulations were performed in which hydrocarbon
emissions were decreased uniformly. For each of the simulations, peak ozone
is compared to that of the base case. This approach enables one to estimate
the overall degree of control that is likely to be needed but ignores the
spatial, temporal and reactivity effects of specific control measures. The
second approach used is to test the effect of specific control programs.
To do this, the area source and point-source emission files were modified to
reflect reductions associated with particular source categories. The control
programs tested were: (1) Reasonably Available Control Technology (RACT) for
stationary sources; (2) the Federal Motor Vehicle Emissions Control Program
(FMVECP); and (3) Inspection and Maintenance (I & M) for motor vehicles.
Simulations were performed using 3 high ozone days from 1976 (159, 195,
and 275). These are the same days that were used for sensitivity testing
described in Chapter 5. Description of the meteorological conditions, ozone
distribution and model performance (relative to observed ozone) for the 3
days is discussed in detail in Section 5.1.
89
-------
In summary, Days 159 and 195 were days of moderate wind speed and
mixing height. Day 275, on the other hand, was characterized by extremely
low wind speeds and restricted mixing heights. For Day 275, the model was
extremely accurate in estimating observed pea. and near-peak ozone in the
base-case simulations. Day 159 showed a tendency toward overprediction
and Day 195 underprediction
6.1 Response to Uniform Changes in Emission?;
Sans iti nty tests were conducted to determine how the Airshed
!^d«i i-.redictions of c. r,- -espond to changes in hydrocarbon emissions. To
establish the response over a wide range, reductions of 5 percent, 17 per-
cent, 42 percent and 75 percent and increases of 17 percent and 63 percent
over the base case were tested. The results for the 3 test days are dis-
cussed in Chapter 5 and are summarized in Figure 38, in which percent change
in peak ozone concentration is plotted as a function of percent change in
hydrocarbon emissions. For all 3 days the response to changes in hydro-
carbon concentration is "less than 1:1; for example, on Day 195 a 40 percent
reduction in hydrocarbon emissions is required to reduce peak ozone by 20
percent. Furthermore, the model's sensitivity to HC emission changes is
substantially different for the 3 days. The model's response to reductions
(<-*• increases) in HC emissions is strongest for Day 275, the severe stag-
nation nay with the highest observed peak ozone cf the 20-ddy dsta base.
Another finding is that with greater degrees of rydrocarbor- control, the
location r*f the predicted u<:3k o/.rw is <"'• •- -low; wind and tends to
occur later in the day.35
-------
Additional sensitivity tests were conducted in order to assess how
changing NOX emissions would affect the model's ozone/hydrocarbon response
response curve. Specifically, uniform increases and decreases of NOX emis-
sions were coupled with a 42 percent reduction in HC emissions for the 3 test
days. The results of this analysis are shown in Figure 35 and indicate that
decreases in NOX emissions (by 20 percent from the base case) result in a
lessening in the effectiveness of HC emission reduction in decreasing peak
ozone concentration. The effect was small on Day 195 (ozone reduction
diminished by about 2 percent) and substantial (about 7 percent) for the
remaining 2 days. Additional tests are required to assess the effect of
N Ox emission changes with greater reduction in hydrocarbons.
6.2 Control Strategy Simulations/Methods
The sensitivity tests described above are based on uniform
reductions of total emissions. However, actual control programs have
varying effects on different sources, different parts of the region and
on different classes of hydrocarbons. For this reason, simulations
were conducted using emission inventories which were modified to reflect
changes in the actual sources to which the control programs are applied.
In the following inventories, no adjustments were made for changes in
population, production or traffic.
6.2.1 RACT
According to CFR 40, Part 51, "'Reasonable available con-
trol technology' means devices, systems, process modifications, or other
apparatus or techniques, the application of which will permit attainment of
91
-------
the emission limitations. . . ."36 Furthermore, RACT requires that atten-
tion be given to (1) the necessity of imposing emission limitations to
achieve and maintain a national standard, (2) the corresponding social and
economic impact and (3) alternative methods for achieving and maintaining
the national standard. RACT measures are further defined by available Control
Technique Guidelines (CTG's) for specific sources, e.g., EPA-450/3-78-120.37
The RACT emission inventories used in the Airshed Model
control strategy simulations described herein, were based on a number of
studies by Baverman.3&»-^,'*Q,41 These studies were performed to determine
the reactive hydrocarbon reductions that could be achieved if RACT were
applied to major stationary sources within the St. Louis modeling region.
Most of the stationary source hydrocarbon emissions are evaporative in
character and include industrial point sources (e.g., surface coating,
petroleum refining and storage) and area sources (gasoline marketing, dry
cleaning and surface coating). The application of RACT to stationary
sources results in a daily hydrocarbon emission reduction of approximately
22 percent. The hydrocarbon reduction varies from a low of 21 percent for
Day 159 to a high of 24 percent for Day 275.
6.2.2 FMVPC
The Federal Motor Vehicle Pollution Control (FMVPC) pro-
gram is broadly defined in the Clean Air Act (as amended in 1977). A more
detailed discussion is presented in CFR 40, Parts 85 and 86.42,43 in the
present study, FMVPC was considered to a K *••„;, only the hydrocarbon and oxides
of nitrogen components of the mob:1-~ sc-jree portion of the 1976 (baseline)
emissions inventory. In order to prepare the FMVPC, the St. Louis baseline
Q?
-------
mobile source inventory was modified to reflect projected 1987 hydrocarbon
and oxides of nitrogen emission levels mandated by the 1977 Clean Air Act
(CAA) amendments. The FMVPC strategy required a change in mobile source
hydrocarbon speciation as well as a reduction in mobile source hydrocarbon
and oxides of nitrogen emissions.
This study utilized estimates of total HC and NOX reduc-
tions associated with FMVPC which were prepared by Mayer.44,45 These emis-
sion reductions were determined from representative composite mobile source
emission factors generated by Mobile 1.46 Representative vehicle speeds
were chosen for each of the road classifications (i.e., freeways and arte-
rials) considered in the Mobile 1 simulations. Traffic volume and driving
patterns were held constant. The FMVPC strategy resulted in an approximate
69 percent reduction in mobile source hydrocarbon emissions. This corres-
ponds to a 41 percent reduction in area-source hydrocarbon emissions or a
33 percent reduction in total hydrocarbons. The FMVPC strategy total hydro-
carbons (carbon) reduction varies from a low of 33 percent for Julian Day 275
to a high of 34 percent for Julian Day 159. The NOX reduction represents a
43 percent reduction in NOX in the mobile source inventory, a 27 percent
reduction in area-source emissions and an 8 percent reduction in total NOX
emissions. (N02 is assumed to be 10 percent of total NOX emissions, the
remainder to be NO.)
The FMVPC strategy required a change in the distribution
of hydrocarbon species as well as a reduction in mobile source hydrocarbon
emissions. In addition to the mobile source emission factors, Mobile 1
93
-------
simulations also yielded the evaporative and exhaust-gas component frac-
tions of the mobile emissions for 1976 and 1987. The 1976 evaporative and
r-xhfust gas soecies profiles were provided by Mov^k.^ For the purpose of
this study, it was assumed that all automobiles operating in St. Louis in
1987 would be equipped with catalytic reactors for exhaust gas emission con-
trol and with canisters for evaporative emission control. The 1987 gas
emission species profile is assumed to be that given by Table 9-06-021A of
the EPA VOC .Species Hata Manual '^ The corresponding evaporative emission
no^cifE nr'-f-ne H cn'ven K" Tab" e 9-06-O^lC of the sane report. Both
p'fofi'les are based on Black n ty for inspection and
laaintenance is found in the Clean A-'- ; ...mended in 1977. Additional
requirements are r^esented in *h?; iV.:^rti Reo-; s" :v.'36'4' It should be noted
94
-------
that the I/M control strategy is considered to be a secondary control strat-
egy. As such, its effectiveness is dependent, in part, upon the effective-
ness of the primary control strategy--FMVPC. The present study assumes that
I/M affects hydrocarbon emissions but does not affect emissions of the oxides
of nitrogen. In addition, it is assumed that all automobiles are included in
the inspection and maintenance program; the inspection failure rate is 30
percent and there is mechanics training. These assumptions form the basis of
a strong inspection and maintenance program for St. Louis. According to
Mayer,45 this strong I/M program results in a hydrocarbon reduction of approxi-
mately 11,700 tons/year. This corresponds to a reduction of some 10 percent
in mobile-source hydrocarbon, 6 percent in area-source hydrocarbon and 4.5
percent in total hydrocarbon emissions.
6.3 Control Strategy Results
The 16 control strategy simulations performed in this study are
described in Table 22. The table shows that four strategies each were simu-
lated for Days 159 (transport) and 275 (stagnation). Eight control strat-
egies were simulated for Day 195 (transport). RACT and FMVPC were simulated
individually and in combination for all 3 days. The combination of RACT and
FMVPC control strategies in conjunction with boundary changes in both hydro-
carbons and ozone was also investigated for all 3 days. The effects of the
I/M strategy as well as the effects of NOX reductions were evaluated only
for Day 195. Additional boundary changes were evaluated only for Day 195.
It should be noted that boundary concentration changes in hydrocarbons, the
oxides of nitrogen and ozone were limited to changes only in the anthropogenic
portion of these pollutant concentrations.
95
-------
In using the results of uniform reduction and specific control
measure simulations, several alternative approaches are possible. First,
one can track the changes in ozone concentration that take place at the sites
of ozone monitors. The emission control requiiement is defined as that
needed so that none of the monitor sites have predicted peaks which exceed
120 ppb (the NAAQS level). This approach has the advantage of using only
model results from grid cells represented by monitoring sites, i.e., the same
portion of the modeling region for which performance measures relative to
observed concentrations have been calculated. In addition, this approach
meshes with the current practice (being used with EKMA for the 1982 SIP's)
of basing control on analyses for specific monitoring sites.
The site-specific method, however, has one drawack. The sensi-
tivity analyses discussed previously have shown that the location of the peak
ozone tends to migrate downwind. Thus, the control curve for a particular
site will be steeper (greater reduction in ozone with reduction in emissions)
than that for the region as ci whole. Given the paucity of monitoring sites
in outlying regions, there is a strong possibility that relying solely on the
behavior of peak predicted o;:one at the monitoring sites will underestimate
the full extent of control needed to bring the entire AQCR into attainment.
In tne current study, an alternative approach, tracking the peak predicted
ozone concentration regardless of position is used. This method utilizes
the entire spatial domain of the model and offers greater assurance that the
predicted control requirements will affect Ue enure region and not merely
shift the problem downwind.
-------
A second issue is that of calibration. As stated in Chapter 4, the
model has a systematic tendency to underestimate peak and near-peak ozone,
For control strategy simulations, we have selected 3 days—one for which
peak ozone was overpredicted (159), a second for which peak ozone was under-
predicted (195), and a third for which the model predicted peak ozone accu-
rately (275). The question is whether the model results should be calibrated
for the 2 days showing bias. In the current study, we have used the model in
a relative sense, such that we are concerned with percent reductions in peak
ozone.
% ozone control required = Pn,ax - 120 ppb
___ x 100%
•max
This approach contains an implicit form of calibration, one in which the
relative percent error remains constant as emissions are reduced in the
simulations. This is demonstrated in the following example. Suppose that
the predicted peak ozone is 240 ppb. However, we have shown that the model
overpredicts by 20 ppb (8.3%) on this day. Using our approach we estimate a
control requirement of
(240 _ 120) x 100% = 50%
~
However, this is exactly the same as adding a constant relative error to the
base case and the control case, i.e.,
([240-20] - [120-10]) x -,00% = 220-110 = 50%
L240-20J Z2TT
when the relative error (20 ) for the base case is equal to the relative
MO
error for the control case
120
97
-------
The results of control strategy tests for the 3 days are presented
in Table 23 and are plotted on Figure 39 along with the HC/03 response curves
for uniform HC reductions discussed previously.
Except as otherwise indicated in Figure 39, boundary concentra-
tions (i.e., transported ozone and precursors) were held constant. For a
number of strategies, the model was run with reduced boundary concentra-
tions to test the effect that parallel control programs for upwind cities
would have on St. LOUIS ozone levels.* For these cases the anthropogenic
THV *•.! ;n or boundary n - . r • corcenfations was reduced in proportion to
the emission reductions. For ozone, the anthropogenic portion (i.e., that
in excess of 0.04 ppm) was reduced less than proportionately to HC emis-
sions reductions using the HC/03 response curves (discussed in the previous
section) as a first approximation.
Examination of Figure 39 reveals that the control strategies
tested tend to follow the general patterns indicated by the HC/03 response
twveb based on uniform emission reductions. However., for the two trans-
port days (19 and 195), the combined (RACT + FMVPC) simulations tend to be
somewhat snore effective tnan the uniform reduction runs in reducing peak
ozone. For the stagnation ddy (275), the control simulations result In the
nK reductions in peak ozone as the ^nifom reductions.
A second finding is that the model appears to show a synergistic
effect betwe-:-" s^at^onary ant trnf-jPe sour'"-- •' ". for example, on Day
*Init1d"! concentrations use? . ocieling w-'.re reduced from the base
:ase in oropcrticm tc reduct^51:. T- "fissiop- *n all ;.ases.
-------
275, the RACT simulation results in an 8 percent decrese in peak ozone and
FMVPC, a 32 percent decrease when each is applied separately. However, when
the two strategies are combined, the model yields a 49 percent reduction in
peak ozone. This finding may be associated with nonlinearity in the HC/03
response curve, i.e., HC control becomes more effective as the HC/NOX ratio
decreases and ozone formation is increasingly limited by HC concentration.
Reductions in boundary concentrations representing parallel upwind
controls were found to have a substantial effect on peak ozone concentration
for the 2 transport days (159 and 195). On Day 159, reduction in boundary
concentration in conjunction with FMVPC and RACT resulted in an additional
5 percent reduction in peak ozone. For Day 275 (stagnation) the model appears
to be insensitive to changes in boundary concentrations. On this day, ozone
transported into the region was lower than on most other days and the effect
of precursor buildup associated with urban emissions tends to overwhelm the
effect of transported ozone and precursors.
Changes in NOX emissions also affect model estimates of peak ozone
reduction. For Day 195, the FMVPC simulation was run for HC reductions
alone and for the case in which both HC and NOX emissions are reduced in
accord with current statutes. Note that in the cases for which NOX control
is included, the computed reduction in peak ozone is diminished by about 5
percent This finding is similar to that reported in the section on uniform
reduction sensitivity tsts.
The changes in emissions associated with the control strategies
and uniform reductions not only reduce the magnitude of the peak ozone
99
-------
concentration, but also change the position of the peak in time and space.
As Figure 40 illustates, there is a strong tendency for the position of the
peak to migrate downwind as the level of control increases. For 2 of the
3 days, the time of the predicted peak is later in the day with increasing
control.
As in the case of the uniform emission reduction simulations, the
control strategy simulations demonstrate that the effects of control vary
strongly from day to day. For example on Day 195, the combination of RACT
ana FMVPC led to a 30 perce..*; reduction in peak ozone, whereas on Days 275
and 159, the same combination of controls resulted in a more than 45 percent
reduction of peak ozone. Moreover, the estimates oF control requirements
are quite different for the 3 days. The day-to-day variation in HC/03
response probably reflects reality. Each day has unique source-receptor
relationships, photochemical potential and boundary concentrations.
For Days 275 and 195 the analysis indicates that RACT and the
FMVPC applied in St. Louis (and in upwind cities) may be sufficient to
meet the ozone standard. On the other hand, the results for Day 159 sug-
gest that control beyond RACT, FMVPC and I/M may be required. (I/M was
tested only on one day (195); on this day the estimated effect of I/M on
peak ozone was relatively small, about 3 percent). One should note, how-
ever, that for Day 159, the model predicted peak ozone concentrations occur
in the northwest ccrne- of the region, an area without ozone monitors.
In interpreting th€> above results, we have made no formal attempt
to relate the findings to the statistical (expected exceedance) form of the
100
-------
ozone NAAQS. The use of resource-intensive photochemical grid models pre-
cludes modeling a large number of days that would be required for a strict
determination of control requirements relative to the statistical form of
the standard. However, the 3 days tested include several of the days with
highest ozone observed in 1976. (Days 275, 195 and 159 had peak observed
ozone concentrations of 244, 223 and 172 ppb respectively). The 3 days
represent a good cross section of the types of days associated with high
ozone. The most stringent control requirements were found on Day 159; how-
ever, the predicted peak for this day was found in the northwestern portion
of the modeling region where there were no monitors to evaluate the maximum
predicted concentration of 312 ppb. Monitoring coverage for Days 275 and
195 was more adequate. Thus, one might be tempted to use the estimates of
control requirements for Days 195 and 275, i.e., 50-60 percent NMHC emis-
sion reduction achieved largely through a combination of RACT and FMVPC.
While the study indicates that Inspection and Maintenance results in a small
(several percent) reduction in peak ozone, I/M may be necessary to assure
that the full benefits of FMVPC are realized.
The reader is cautioned, however, that the results are based on
1976 emissions and air quality data. Thus, the design concentration and
the estimated control requirements may not be in exact agreement with those
based on more recent assessments.
6.4 Uncertainties
While it is presently difficult to quantify the uncertainties
associated with the analysis, it is possible to identify major areas of
uncertainty. As we have shown, the estimates of control requirements are
101
-------
very sensitive to the slope of the HC/Os curve. The curves differ from
•Jay to day and are complex functions of precursor arid ozone transported
into the region, dilution, the particular mix of emissions that occurs as
air is advected between grid colls and the react vity of the precursor mix.
Any errors (for example, inadequacy of the chemical mechanism, errors in
the speciation of NMHC emissions or in estimates of boundary concentrations)
will result in errors in the HC/03 response curves, and in our estimates
of the cont"oi »'eqji'ed to mee; the ambient ozone standard.
^oeci ticatior ," t • ''Jary concentrations for future years is
particularly difficult. While the assumption that parallel efforts are
being made in upwind cities, one must assume the portion of incoming pol-
lutants are anthropogenic. Moreover, we can only make approximations as
to how control efforts in distant cities affect ozone and precursor con-
centrations in air that has been transported over hundreds of kilometers.
The uncertainty associated with boundary concentrations is likely to
increase as emissions are "turned-down" and peak ozone concentrations grow
increasingly sensitive to boundary concentrations.
Unfortunately, there is no readily available way to test the
veracity of the HC/Ch response- curves. However, it may be possible (for
3t. Luuis or for other cities! to trad; changes in 07.016 and in emissions
over a multi-year period and ";o compara the observer! relationship with
that shown by the ^odel ,
The principal difficulty in svc'r an approach is that it is often
difficult to separate the effect-. ./ -,-;, eoreloy.v f"um that of emissions.
-------
as possible with regard to meteorology. In assessing emissions-related
reductions in ozone, model results from meteorologically similar days should
be used.
103
-------
7.0 Conclusions
The purpose of this chapter is to highlight the primary findings of
the study with regard to model performance, model senstivity and the
model's estimates of control requirements for the St. Louis AQCR. The
chapter also integrates the information presented in earlier chapters in
order to address the issues of uncertainty and the model's tendency to
underpredict ozone at high concentrations. Finally the chapter discusses
the feasibility of the Airshed ^lodel approach for regulatory analyses and
R3ke?: recommendations reoaHing the use of the model.
7.1 Conclusions on Model Performance
An evaluation of the Airshed Model for St. Louis was conducted
using the latest EPA version of the model and the RAPS data base. The
resulting ozone concentration estimates were compared using statistical
procedures (performance measures) recommended by the AMS Workshop on Model
Eva!uation.
In general, the results indicate that the model is sufficiently
accurate for use in control strategy simulations. Specific findings are
discussed below:
a. Model Bias (0—F) was used to assess the model's abilty to
estimate observed ozone concentrations over the full range of ozone
values. The analysis indicates that there is no overall pattern of under-
or overestimation for the 20 deys when all of the data pairs are used;
most of the bias values are within ±20 percent of the zero value.
There is, however, a marked shift toward underprediction for the analyses
which use higher ozone concentrations.
b. Consistent with the finding in ;ti;, the Airshed Model was
found to have a systematic tendency to underestimate peak and near-peak
ozone concentrations. This conclusion <~ based on a number of performance
measures recommended by the AMS and D> lex.
1C'
-------
c. The selection of performance measure has a pronounced effect
on perception of model bias. The measures of peak accuracy based on a
single residual (Omax - Pmax) tenc' to overstate the degree of underpre-
diction. Because these measures are based on the location and/or time
of the observed maximum, temporal and/or spatial displacements result in
a positive bias (underprediction) even if the model correctly predicts
the magnitude of the observed peak. On the other hand, the measures are
obtained by averaging (Omax-Pmax) residuals over all hours for a temporally
paired comparison and over all sites for a spatially paired comparison.
These are more robust and free of this inherent bias. These "A" measures
indicate that the average bias for spatially paired residuals is 14 ppb
(about 11 percent). The predicted spatially averaged maxima for nearly
all of the days fall within ±30 percent of the corresponding observed
maxima.
d. The model accurately represents temporal variation of
ozone; however, the model's ability to replicate spatial patterns varies
from day to day. While the model is able to locate the general area of
highest concentrations for most days, it tends to underestimate the size
of the NAAQS exceedance zone. A lack of monitoring sites in the areas
downwind of St. Louis limit the assessment of the model's spatial accuracy.
e. The model was found to estimate 24-hour average NMHC and
NOX concentrations accurately for the 20-day data set. The lack of bias
indicates that the overall emission inventories are fairly accurate;
however, no conclusions may be drawn with regard to the accuracy of the
spatial, temporal or species distributions in the inventories.
7.2 Model Sensitivity
7.2.1 Findings
Conclusions resulting from a large number of sensitivity
tests (described in Chapter 5) are summarized as follows:
a. The sensitivity of the Airshed Model to changes in model
inputs shows large day-to-day variation, variation which demonstrates a high
degree of interaction among different input parameters.
b. Uniform changes in hydrocarbon emissions result in
nonlinear changes in peak ozone predictions which are in the same direction.
However, the relative magnitude of the change in peak ozone is invariably
less than that for emissions. Peak ozone concentrations are more sensitive
to uniform reductions in hydrocarbon emissions than are area-wide ozone
predictions.
105
-------
Compared to a reduction in NMHC emissions alone,
simultaneous reductions in both NMHC and NOX emissions appear to lower
the reduction of oeak model ozone. Conversely a reduction in hydrocarbon
emissions accompanied by an i icrease in NOX etn. sions, tends to enhance
the reduction of peak ozone. These conclusion^, however, were obtained
at only a singie !42%) reduction in HC emissions and for a limited set of
NOX changes. No conclusions :an be drawn regarding the sensitivity of
nitrogen dioxide predictions to changes in NOX or the effect on carryover
or long-r^nge transport of ozone precursors.
d. Peak ozone predictions are a function of the absolute
amount of NMHC and NOX emissions and not merely of the NMHC/NOX ratio.
Equal re,',; '.'•:,'"•-^ <-• both foMHC and NOX emissions result in lowered peak
ozone.
r. - . .res i" ';nr reart','«i ty of hydrocarbon emis-
cionb tfhicr; c'-e "tjpreS'vf.'Cdt'VEJ of enn's":on inventories in different cities,
produce modest differences in peak ozone predictions. It follows that the
photochemical potential in each city will vary according to the species
distribution of its emissions.
f. Despite the small fraction that carbonyls represent
in the total hydrocarbon emissions, these oxygenated organics play a very
significant role in contributing to peak ozone predictions.
g. Radical shifts in the spatial and temporal distribu-
tion of emissions produce significant changes in peak ozone predictions.
Spreading out, or diluting, emissions irt time and space reduces peak ozone
predictions.
h. "me importance of nonuniform spatial and temporal
emissions patterns are illustrated by the effect which large across-the-
board changes in emissions have in shifting the time and location of the
peak ozone prediction. These shifts are associated with differences in
the trajectory and emissions burden of the air parcel arriving at the time
and location of the peak. The burden of emissions in turn reflects
variations in the temporal and spatial distribution of emissions.
i. Changes in the height of the mixed layer can result
in unexpected ana very significant changes in peak ozone predictions.
This behavior is associated vn'th a complex interaction among ozone and
precursors aloft, elevated emissions sources, and Vie beigh-, of the mixed
1 aver
j. Changes in peak OZCK.V ,. t Dictions are nearly directly
proportional to changes in photolysis rates [which in turn are nearly
ji -eotly r.fOjjO^riona1 to total TO"- ' - <'"0,
-------
7.2.2 Recommendations for Further Sensitivity Analyses
In Chapter 5 gaps in our knowledge of Airshed Model
sensitivity were identified. The following analyses are recommended to
fill these gaps:
a. The sensitivity of Airshed predictions of both peak
and area-wide ozone and nitrogen dioxide levels from combined changes in
both hydrocarbon oxides of nitrogen emissions should be assessed. Spe-
cifically, both increases and decreases in oxides of nitrogen at various
levels of hydrocarbon reductions should be examined.
b. The sensitivity of peak ozone predictions to the
spatial and temporal resolution of emissions should be evaluated. Emis-
sions grid squares should be enlarged such that adjacent grid squares in
the base case are averaged, thereby diluting precursor concentrations in
areas of high emissions density. Similarly, the time interval at which
emissions are updated (presently hourly) should be lengthened, thereby
diluting precursor concentrations associated with peaks in emitting
actvities.
c. Initial conditions, boundary conditions along the
borders of the modeling region, and boundary values aloft should be
scrutinized with respect to their effect on both peak and area-wide
ozone predictions. Attention should be given initial conditions for
hydrocarbons, oxygenated organics, nitric acid, and organic nitrates,
which act as reservoirs of photochemical activity, particularly on
stagnation type days when pollutant carryover is significant. The effect
of ozone and hydrocarbons being advected into the modeling region along
the borders should be examined, particularly on transport type days when
advection is most pronounced. The significance of aloft concentrations
of ozone, hydrocarbons, and oxygenated organics, which all serve as
sources of reactive materials, should be assessed on a day when aloft
concentrations are relatively high and a large diurnal rise in the depth
of the mixed layer occurs and entrainment is maximized.
d. The sensitivity of the Airshed Model to changes in
the wind field needs to be explored. The response of the model to both
changes in the richness of the basic data (radiosonde, pibal, surface
wind speed, direction, and temperature sensors) and to the method (inter-
polation and divergence reduction or diagnostic wind modeling) by which
the wind field is generated from the basic data should be analyzed.
e. The effect of different choices of grid cell structure
on model predictions should be systematically investigated. Changes in
the horizontal dimensions of grid cells on both transport and stagnation
type days and in the vertical dimension on days of prominent wind shear
should be examined.
107
-------
f. The role of biogenic emissions in modifying the effect
of changes in angthropogenic emissions of both hydrocarbons and oxides of
nitrogen needs to be elucidated with regard to its effect on peak and area-
wide ozone concentrations. Specifically, combined hydrocarbon and oxides
of emissions changes should be- made at various estimated levels of biogenic
organic compounds.
7.3 Conclusions with Regard to Control Requirements
In order to estimate control requirements, a series of simula-
tions were conducted in which NMHC emissions were reduced uniformly for all
sources. Specific control strategies (i.e., RACT for stationary sources
and rMVi'C and 1 & M for >
-------
b. The sensitivity and control strategy analyses suggest that
the most effective strategies are those which rely on substantial across-
the-board reductions in NMHC emissions. The model shows a synergistic
effect between stationary and mobile source emissions reductions such that
the combined effect of RACT and FMVPC is greater than the sum of the
separate reductions. The sensitivity analyses do indicate that modest
reductions in peak ozone may be accomplished through significant changes
in species distribution (reactivity of NMHC) or through the spreading out
of emissions; however, the most effective strategy appears to be associ-
ated with substantial reductions in total NMHC emissions, a finding which
supports current EPA policy.
c. Reductions in boundary concentrations representing parallel
control efforts in upwind cities were found to have a substantial effect
on peak ozone concentration for transport days, but a negligible effect on
stagnation days in St. Louis.
d. Simulations in this study suggest that reduction of NOX
emissions may increase HC control requirements; however, only a small
number of tests for 3 days at one level of HC reduction (HC/NOX ratio)
were conducted. Secondly, as discussed previously, the model's handling
of NOX emissions from point sources is inadequate. A complicating factor
is that reductions in NOX emissions may result in several benefits:
reduced urban N02, reduced levels of nitrate and a potential reduction
in ozone concentrations in regions downwind of the area modeled.
e. The emissions and air quality data used are based on 1975-
1976 information. Thus, the current analysis cannot be used to formulate
a specific control program. Nevertheless, the simulations suggest that
both the Federal Motor Vehicle Pollution Control and RACT for stationary
HC sources are minimally required for attainment of the ozone NAAQS for
sizeable urban areas.
7.4 Errors and Uncertainties
In this section, we review findings related to model error and
uncertainty. Chapter 4 clearly demonstrates that the Airshed Model has a
systematic tendency to underestimate ozone concentrations, particularly
at higher concentrations, for St. Louis. The reasons for this tendency
have not been established to date. Nevertheless, the evidence presented
in this report points to several possibilities. Two lines of evidence
suggest that the carbonyl fraction used for the St. Louis NMHC emissions
and background concentrations were erroneously low: (1) the sensitivity
109
-------
analyses indicate that the magnitude of the carbonyl fraction is an
important determinant of peak ozone and (2) the value used for St. Louis
is substantially lower than the size of the faction measured in a number
of urban areas. Whereas SAI now recommends that a value of 5-10 percent
carbonyl fraction be used, the fraction used in this study is about 1.5
percent. The 'je-isitivity tejts suggest that an increase in carbonyl
fraction from this level to about 5 percent (the lower end of the recom-
mtndea ra^i^'t •„-i" increase p-?ak ozone concentration by 10 percent cr more.
Iris e^P5"""1" i" r: MTO5raHT° to the 11 oercent average bias for spatially
pa'ro': i evirii'c >•; call, >'-.\. MO for the 23-clay tesi period.
Several other factors may also contribute toward underprediction
of ozone. These include the current model's inappropriate mixing of point
source NOX into entire grid cells, the neglect of vegetative emissions, and
errors in advective transfer associated with windfield errors on transport
days.
In ihis stuuj k no attempt lias been made to assess the uncertainty
associated with the estimates of control requirements. Airshed Modeling
is a complex process based on a large number of assumptions and input vari-
ables. Uncertainties associated with these oarameters and assumptions con-
tru-ute tc the- overall uncertainty. While many important varifbU''- have
not been tested systematically, the sensitivity tssts, at least in a pre-
limirujrv way. point tc those variables which t\:-v? •• vrry strong impact on
peak ozone. Among the variables found tc c ;\.iufi are: the distribution
:* 'iyd.-oc{t -hc-n -.pecies (reactivit-' " ' .-U'.'ulorly carbonyl fraction,
C1 ra'-io, >r->? r>hotol,>s'is ';.te- and n1'•'i'-'-i height Furthermore,
110
-------
testing with OZIPP/EKMA indicates that predictions are extremely sensitiive
to concentrations of precursors transported into the modeling region.53
Errors in these parameters may lead not only to errors in predicting observed
concentrations, but also to errors in the HC/03 response curves used to esti-
mate control requirements. Additional factors which may contribute to problems
in the HC/03 response curves are the designation of future year boundary (back-
ground) concentrations and deficiencies in the model treatment of chemistry,
point-source diffusion, and vertical mixing. A number of specific tests are
suggested in Section 7.2.2.
7.5 Feasibility of the Airshed Approach
The St. Louis ozone modeling project and other studies utilizing
photochemical grid models have demonstrated that such programs require
large and costly data bases, lengthy lead times and considerable resources
and expertise. A typical modeling study for an urban region can take 3
years or more and require resources equivalent to $1-2 million (including
data-base collection and preparation).^ Modeling requires the availability
of an air pollution engineer, a systems analyst or advanced programmer, an
air pollution meteorologist and an atmospheric chemist. A large computer
facility and from 30-70 hours of computer time are also required. In
addition, potential users can anticipate that initial modeling may identify
anomalous model output, a result which will necessitate a detailed examina-
tion of the model set-up and inputs. Several modeling iterations may be
required to sort out the problems and delays may be anticipated.
There is no doubt that the three-dimensional grid model approach
encompassed in the Airshed Model is superior to simple approaches (e.g.,
Ill
-------
OZIPP/EKMA) in its spatial, temporal and species resolution and in its
treatment of physical and chemical processes. However, many uncertainties
remain and comparative evaluations of Airshed c.-
-------
8.0 REFERENCES
1. Reynolds, S. D., "The Systems Applications, Inc., Urban Airshed
Model: An Overview of Recent Developmental Work," EPA-600/3-77-001b,
U.S. Environmental Protection Agency, Research Triangle Park, North
Carolina, 27711, 1977.
2. MacCracken, M. D., J. D. Wuebbles, J. J. Walter, W. H. Duewer,
and K. E. Grant, "The Livermore Regional Air Quality Model,"
J. App. Met., Vol. 17, pp. 254-272, 1978.
3. Schere, K. !_., and J. H. Shreffler, "Final Evaluation of Urban-Scale
Photochemical Air Quality Simulation Models," (Draft), Environmental
Sci ences Research Laboratory, U.S. Environmental Protection Agency,
Research Triangle Park, North Carolina 27711, 1982.
4. Fox, D., "Judging Air Quality Model Performance," (A Summary of the
AMS Workshop on Dispersion Model Performance, Woods Hole, Mass., Sept.
1980). Bulletin of the Amer. Met. Soc., Vol. 62 No. 5, pp. 599-609,
1981.
5. Layland, D., "Guideline for Applying the Airshed Model to Urban Areas,"
Publication No. EPA-450/4-80-020, U.S. Environmental Protection Agency,
Research Triangle Park, North Carolina 27711, October 1980.
6. Briggs, G. A., "Plume Rise: A Recent Critical Review," Nucl ear Safety,
Vol. 12, pp. 15-24, 1971.
7. Killus, J. P., et al., "Continued Research in Mesoscale Air Pollution
Simulation Modeling--Vol. V: Refinements in Numerical Analysis, Trans-
port, Chemistry, and Pollutant Removal," draft report for Environmental
Sciences Research Laboratory, Office of Research and Development, U.S.
Environmental Protection Agency, Contract No. 68-02-2216, Systems
Applications, Incorporated, San Rafael, California, 1977.
8. Littman, F., "Regional Air Pollution Study: Emission Inventory Summari-
zation," Publication No. EPA 600/4-79-004, U.S. Environmental Protection
Agency, Research Triangle Park, North Carolina 27711, 1979.
9. Strothmann, J. A., and F. A. Schi ermei er., "Documentation of the Regional
Air Pollution Study (RAPS) and Related Investigations in the St. Louis
Air Control Region," Publication No. EPA 6004-79-076, U.S. Environmental
Protection Agency, Research Triangle Park, North Carolina 27711, 1979.
10. Shreffler, J. H. andR. B. Evans., "The Surface Ozone Record from
the Regional Air Pollution Study, 1975-1976." Atm. Environment,
Vol. 16, p. 6, 1982.
113
-------
11. Karl, T. R., "Ozone Transport in the St. Louis Area," Atmospheri c
2 > H21-1431, 1978.
12. Karl, T. R., "Day of the Week Variations of Photochemical Pollutants
in the St. Louis Area, Atmospheric Environment, Vol. 12, pp. 165-1667,
1978.
13. Karl, T. R., "A Study of the Spatial Variabilit y of Ozone and Other
Pollutants at St. Louis, Missouri," Atmospheric Environment, Vol.
14, pp. 681-694, 1980,
14. Schiermei er, F. A., "RAPS, Field Measurements Are In," Envir. Sci .
Techno! . Vol. 12, pp. 644-651 1978.
15. Arider-.;n, G., "Objective Windfield Analysis for Airshed Model in
Killus, J. P., et al . , "Continued Research in MesoscaleAir Pollution
Simulation ModeTipq-'-Vol V: Refinements in Numerical Analysis, Trans-
pert; Chemistry, c • 'J '•' itar-.t Removal , " d*-aft report for Environmental
Scirnf_s>: H • "•>rd-"e1' ,aDDrd";ory , Office oc Research and Development, U.S.
Environmental Protection Agency, Contract No. 68-02-2216, Systems
Applications, Incorporat3d, San Rafael , California, 1977.
16. Schere, K. L., "Air Quality Model Response to Objectively Analyzed
Wind Fields," Fifth Sympasium on Turbulence Diffusion, and Air Pol-
lution of the America?! Meteorological Society, Atlanta, Georgia,
March 1981.
17. Shir, C. C. and L. J. Shi eh, "Development of the Urban Air Quality
Simulation Model with Compatible RAPS Data," EPA-600/4-75-005-a,b,
U.S. Environmental Protection Agency, Research Triangle Park,
North Carolina 27711, 1975.
13. Cox, W. M., letter to Mr. Richard Londergan, TRC Environmental
Consultants, October 20, 1981.
19. Cole, H. S., Newberry, C. F., Cox, W., Moss, G. K., and D. Layland,
"Application of the Airshed Model for Ozone Control in St. Louis."
U.S. Environmental Protection Agency, Research Triangle Park, North
Carolina 27711. Presentated at 75th Annual Meeting of Air Pollution
Control Assoc,, New Orleans, Louisiana, June 1982.
20. Rehme, K. A., J. C. Puzak, M. E, Beard, F. Smith, and R . J. Paur,
"Evaluation of Ozone Ca'ibration Procedures," Publication Ho.
EPA 600-54-80-050, U.S. Environmental Protection Agency, Research
Triangle Park, North Carolina 27711, 1981.
21. Fried, A., and 6. Hodge'ion. Laser F*'";';^ j<^'<:\c Detection of
Nitrogen Dioxide in the Gas-phase TiJ.r-dfi on of Nitric Oxide with
Ozorc," A-al . Chew. , ¥o>. 5*- : '••".. 1982.
-------
22. Zimmerman, P. R., "Testing of Hydrocarbon Emissions from Vegetation,
Leaf, Litter, and Aquatic Surfaces, and Development of a Methodology for
Compiling Biogenic Emission Inventories," EPA-450/4-79-004, U.S.
Environmental Protection Agency, Research Triangle Park, North Carolina
27711, 1979.
23. Flyckt, D. L., H. H. Vlestbery, and M. W. Holdren, "Natural Oranic
Emissions and Their Impact on Air Quality,": Paper No. 80-69.2, 73rd
Annual Meeting of the Air Pollution Control Association, Montreal,
Quebec, June 1980.
24. Kamens, R. M., H. E. Jeffries, M. W. Gery, R. W. Weiner, K. G. Sexton,
G. B. Howe, "The Impact of a-Pinene on Urban Smog Formation: An
Outdoor Smog Chamber Study," Atmos. Environ., Vol. 15, p. 969, 1981.
25. Systems Applications, Inc. "The Sensitivity of Complex Photochemical
Model Estimates to Detail in Input Information," EPA-450/4-81-031a,
U.S. Environmental Protection Agency, Research Triangle Park, North
Carolina 27711, 1981.
26. Dennis, R. L., M. W. Downton and R. S. Press, "Evaluation of Perform-
mance Measures for an Urban Photochemical Model," Draft, June 1982.
27. J. P. Killus, G. Z. Whitten, "A New Carbon-Bond Mechanism for Air
Quality Simulation Modeling," EPA-600/3-82-041, U.S. Environmental
Protection Agency, Research Triangle Park, North Carolina 27711,
1982.
28. G. Z. Whitten, H. Hogo, M. J. Meldgin, J. P. Killus, P. J. Bekowies,
"Modeling of Simulated Photochemical Smog with Kinetic Mechanisms,
Volume 1. Interim Report, EPA-600/3-79-001a, U.S. Environmental
Protection Agency, Research Triangle Park, North Carolina 27711,
1979.
29. G. Z. Whitten, J. P. Killus, H. Hogo, "Modeling of Simulated Photo-
chemical Smog with Kinetic Mechanisms, Volume 1. Final Report,"
EPA-600/3-80-028a, U.S. Environmental Protection Agency, Research
Research Triangle Park, North Carolina 27711 1980.
30. S. T. Zalesak. "Fully Multi-dimensional Flux-Corrected Transport
Algorithms for Fluids," J. Comput. Phys., Vol.31, p. 335, 1979.
31. C. F. Newberry, H. S. Cole, G. K. Moss, "The St. Louis Oxidant Modeling
Project: Additional Sensitivity Testing," draft report, Office of Air
Quality Planning and Standards, Research Triangle Park, North Carolina
27711, May 1981.
32. H. S. Cole, J. Summerhays, K. MacKay, C. Newberry, "St. Louis Oxidant
Modeling Project: Progress Report," Draft Report, U.S. Environmental
Protection Agency, Research Triangle Park, North Carolina 27711,
September 1980.
115
-------
33. R. M. Kamens, H. E. Jeffries, M. W. Gery, R. W. Weiner, K. G. Sexton,
6. B. Howe, "The Impact of o-Pinene on Urban Smog Formation: An
Outdoor Smog Chamber Study," Atmos. Environ., Vol.No. 6, p. 969,
1981.
34. R. R. Arnts, B. W. Gay, J~., "Photochemistry of Some Naturally
Emitted Hydrocarbons, " EPA-600/3-79-081, U. S. Environmental Pro-
tection Agency, Research Triangle Park, North Carolina 27711, 1979.
35. Gipson, G., "Comparison of Three Ozone Models: Urban Airshed,
City-Specific EKMA, and Proportional Rollback." Office of Air
Quality Planning and Standards, U.S. Environmental Protection
Agency, Research Triangle Park, North Carolina 27711, 1982.
36. Code of Federal Regulations (CRF) 40, Protection of Environment,
Part 51, Revised as of July 1, 1981, pp. 534-585.
37. Peterson, °, R. arr.i R, ?. Sakaida, "Summary of Group I Control
Technique Guideline Documents for Control of Volatile Organic
Emissions from Existing Stationary Sources," EPA-450/3-78-120,
U.S. Environmental Protection Agency, Research Triangle Park,
North Carolina 27711, June 1979.
38. Braverman, T., "Estimated RACT Emissions Reductions for Point-
Source Evaporative HC Emissions in St. Louis," Office of Air Quality
Planning and Standards, U. S. Environmental Protection Agency,
Research Triangle Park, North Carolina 27711, June 1979.
39. Braverman, T., "Estimated RACT Emissions Reductions for Point-Source
HC Emissions in St. Louis from Petroleum Refining, Primary Metals
Production, Chemical Manufacturing, and Recently Added Point-Sources,"
Office of Air Quality Planning and Standards, U.S. Environmental
Protection Agency, Research Triangle Park, North Carolina 27711,
September 1979.
40. Braverman, T., "Estimated RACT Emissions Reductions for Area-Source
Evaporative HC Emissions in St. Louis from Nonindustrial Surface
Coating, Gasoline Marketing at Service Stations, and Dry Cleaning
Operations," Office of A:: r Quality Planning and Standards, U.S.
Environmental Protection Agency, Research Triangle Park, North
Carolina 27711, 1979.
41. Braverman, T., "Revision of RACT Control for RAPS," Memorandum to
David H. Barrett, Model Application Section, Office of Air Quality
Planning and Standards, U. S. Environmental Protection Agency,
Research Triangle Park, North Caroline 271n, October 18, 1979.
42. Code of Federal Regulations (CFR) 40, Protection of Environment,
Part 85, Revised as of July 1, 1«H1, pp. 186-242.
43. Code of Federal Regulations (C?\F) 40, Protection of Environment,
Part 86, Revised as of July 1, 1981, pp. ^2-870.
116
-------
44. Mayer, N., "Approximate Effects of Mobile Vehicle Emissions Reduction
Strategies on Emission Factors," Memorandum to David H. Barrett, Model
Application Section, Office of Air Quality Planning and Standards, U.S.
Environmental Protection Agency, Research Triangle Park, North Carolina
27711, March 1979.
45. Mayer, N., "Approximate Effects of Motor Vehicle Emission Reduction
Schemes on Hydrocarbon and Nitrogen Oxide Emissions in RAPS,"
Memorandum to David H. Barrett, Model Application Section, Office
of Air Quality Planning and Standards, U.S. Environmental Protection
Agency, Research Triangle Park, North Carolina 27711, March 1979.
46. "Mobile Source Emission Factors (for low-altitude areas only),"
Publication No. EPA-400/9-78-006, Office of Transportation and Land
Use Policy, U.S. Environmental Protection Agency, Washington, D.C.
20460 March 1978.
47. Novak, J., Personal commnication. Data Management and Systems Analysis
Section, Meteorology and Assessment Division, Environmental Science
Research Laboratory, Office of Research and Development, U.S. Environ-
mental Protection Agency, Research Triangle Park, North Carolina 27711,
September 1980.
48. "Volatile Organic Compound (VOC) Species Data Manual," Second Edition,
Publication No. EPA-450/4-80-015, U.S. Environmental Protection
Agency, Research Triangle Park, North Carolina 27711, July 1980.
49. Black, Frank and Larry High, "Procedures to Determine Nonmethane Hydro-
carbon Emission Rates from Automobiles," Paper No. 770144, Annual Meeting
of the Society of Automotive Engineers, Detroit, Michigan, February 28-
March 4, 1977.
50. Black, Frank and Larry High, "Passenger Car Hydrocarbon Emissions
Speciation," Publication No. EPA-600/2-80-085, U.S. Environmental
Protection Agency, Research Triangle Park, North Carolina 27711,
May 1980.
51. Johnson, W. I., "Hydrocarbon Composition, Sulfur, and Manganese Content of
U.S. Motor Fuels Sampled for the DuPont 1978 Summer Road Octane Survey,"
Publication No. PLR--78-68, E.I.duPont de Nemours & Company, Wilmington,
Delaware 19898, November 1978.
52. P. G. Hoel, Introduction to Mathematical Statistics, 3rd edition,
John Wiley & Sons, Inc., New York, 1962, p. 167.
53. Gipson, Gerald L., "Evaluation of the Carbon-Bond Mechanism for EKMA,"
Internal Draft Reports, Air Management Technlogy Branch, Office of Air
Quality Planning and Standards, U.S. Environmental Protection Agency,
Research Triangle Park, North Carolina 27711. Part 1: June 1982;
Part 2: July 1982; Part 3: September 1982.
117
-------
U.< SPECIFICATION OF THE GRID
. Transport
&
V
Transport
Tranjport
3 Transport
o
o
Transport
•*-
*
~^-
T
Chemistry
Elevat'd Emissions
A Transpo
1
Cnemistry
Elevated Emissions
iTranspo
if
Chemistry
Elevated Emissions
j Transpo
! Cnemistry
j Surface Emissiont
1
rt
rt
rt
Tranjport
Transport
Transport
Transoort
Top of Modeling Region
Top of Mixed Layer
Ground Surface
Surfiice Removal
t to ;0 kilometer*™
(b> ATMOSPHERIC PROCESSES 1REATFD '»-,' •-• :.,.,.:> Vis OF GRID CELLS.
Figure -» Schematic illustrate • of ih° g^ id used ^n ' tn
processes feated m the ainhtx .'-odti Id.'.apVd •' "if. n 'jvnclas.
Tesche. and Reid, 1978K
TIB
-------
N02
SUNLIGHT
(a) THE NO2-NO-O3 CYCLE
«I02
PHOTOCHEMICAL
BY-PRODUCTS
ORGANIC COMPOUNDS
FREE
RADICALS
(b) ORGANIC OXIDATION OF NO TO N02 WITH OZONE BUILDUP
Figure 2. Photochemical production of oxidants.
119
-------
\ (EDWARDSVILLE
0115 ^
Figure 3. The St. Louis ,-rQ« with locations of the
RAPS surface stations.
120
-------
?x / iQ i"v «*""«"> i
LLLr^ ^ W
1 ' ™ i\™T-™^^^^™'"' •^^^^*'T«fc*iki«»^Tr'wv i — — •
Figure 4. The St. Louis area with locations of the RAPS surface
stations and 4 x 4 km modeling grid superimposed.
121
-------
DRY8195-1976
HR>8
THC EMISSIONS
D
0
0
0
9
35
53
41
If
21
lit
4
59
71
95
51
4
«
16
19
0
C
a
0
21
94
SS
31
ZS
31
ni
7>
JSS
128
S!
22
29
42
26
2
a
0
21
21
21
30
103
MB
12
73
95
rfo
96
30
9S
29
35
33
9
9
7
4
3
53
ZM
«
ts
94
It*
313
MB,
6S
46
45
27
44
8
0
0
41
42
34
49
41
f«
ttt
m.
2ft
M*
278
m
71
14
17
24
10
0
0
79
ft
26
93
93 ;
4U"^
345
228
431
fetj
«»
«*
JSS
37
Its
it*
0
0
96
»M
71
40
491
4tt
348
*«
*«*
££
SOS
(N
1*3
m
»«
so
9
0
2
44
MSt
75 j:
98;
JSS
4(12
««*';
^••i
BJ
**£
308
V?:
«"
33
31
6
8
0
7
¥
««;'
34';
w.
31
n
M
^
It
iic
72
49
12
S
6
9
0
3;
no
ZS9
?»3
»0
87
$«
>49
*4i.
^
^
^a]
69
33
77
16
1
9
0
11
20
W4
«fl
»«
132
77
5tJ
+1*
^
^
i}<
89
12
8
36
16
17
34
0
21
a
ttt
S
w
zoo
67
221
63
ISO
310
til
til
10
u
14
«
10
14
0
8
12
m
J*7
ic*;
17
69
70
33
UJ
in
tu
IW
27
14
14
8
•
f
0
f
12
48
3
33
11
41
44
14
146
SO
17
!!«
ltd)
25
20
If
10
.5
0
1
If
16
14
72
M»
4.
71
114
112
21
?«.
tu
*»
32
23
22
15
9
0
S
9
10
12
20
60
42
3t
19
10
12
57
5«
5?
24
If
If
9
4
0
2
1
to
11
12
33
33
33
20
2
7
77
61
24
3
IS
15
35
J2
Figure 5. Total source emissions of THC, in kilograms, for the
RAPS area on Day 195 of 1976 at 0800 CST. Grid cells
are 4 km squares; the SW corner corresponds to UTM
(x. y) coordinates of Zone 15: (706, 4236).
<100
100-500 500-1000 >IOO'
122
-------
DRY: 195-1376
NOX EMISSIONS
t
0
44
1C
s
s
It
3
it
it
• 0
IS
1
1
i
4
\
1
0
0
2S
21
11
12.
322
57
3:
:s
2S
17
s
s
3
1
I
4
0
s
zs
9S
u
37
1CU
•16»
t7
41
23
13
e
t
i
3
4
2
e
7
142
f£T
13!
104
S3
rr»
iif
us
13
IS
11
10
s
4
3
2
a
7
k
T«
127
;3s
m
trc
127
us
72
It
20
3
3
3
10
12
*•
S
35
:t*
I/'
200
S3
•ii
!JS
HI
03
tu
il
II
(4
(I
114
)4
37
C
7 ]
78
t?S
it!
1S>
443
•J33
!*«
107
304
(3
"
P
19
S
3
7
pM
iw
78
S*
SI
4
34
ti
16
Si
17
1
t
1
4
3
3
3
3
0
26
14
3J
37
U
$s
s:
Sf
4.4
IS
t
f
1
4
3
2
4
0
11
1
22
24
40
4S
IS
IS
S3
4;
12
t
7
e
s
2
3
0
S
22
21
IS
32
17
II
»
17
fl
IS
1
3
3
S
3
3
0
11
17
:t
is'
IS
7
7
23
IS
21
11
IS
10
*
7 ,
3
3
0
11
S
u
15
11
S
s
16
29
II
6
10
10
13
13
S
S
Figure 6. Total source emissions of N0y, in kilograms, for the
RAPS area on -Day 195 of 1976xat 0800 CST. Grid cells
are 4 fan squares; the SW corner corresponds to UTM
(x, y) coordinates of Zone 15: (706, 4236).
mm
wrn
|SO*-IOOO > 1000
123
-------
•o a;
c u
re
4->
c: to
ro
S-
re
S- VI
a-
O J- ^
•r- tt) OJ
+J C O
C >i
•r- O) +->
ro
r- CD XJ
ro to Q)
C
cr o; +->
4-> re
I/I .C
«4- ro
O CJ C/)
i- ro
>, ro QJ
ro i-
t— i— ro
O. ro
to -i— C
•i- S- ro
(/) S-
T3 3 3
•i— T3 -Q
S- C 3
C3 -r- I/I
ai
3
D-,
124
-------
Totally paired
Temporally paired
Spatially paired
Unpaired
1 1_
R
» • 1
F2-
I • I
Ai
A2
I & 1
1 4
-ZO O
40 60
Fin. 8. 20-day averaqe residuals with 95^
confidence limits for peak/near-peak
measures.
125
-------
/ 82 f-
(63
107
2.09
•» Z 1 l-
Z.S 1 *~
•J <" i I—
Ifoo
21 I ( •• 1
137
175" 1 ^.... ... •
-« ( S
- .., i M
1 , *,.. j
i ja . , * . ..., .__ j
1
t 1
-
V
i. ,- . *
j • f
»_A _^
i i i
-40 — Za & 2.O 4° fc° (
F3 • A2 H— »— f
m
f
Finure 9. Spatially paired measures for each day,
See text for details.
126
-------
Hl-
178
162.
(83
2.31
25-1
XII
111
2-J7
"-H
_L
X
X
~2O
2.0
40 60 HO (,-rr)
Figure 10. Unpaired measures (F4 and A4) for each day.
See text for details.
127
-------
\-
^ «•»
o
£
; O
_ O
X
v
X
"J8
X
\
in
OJ
i-
x la. •
10 >-.
+J r— •
C <0 ^->
OJ T- C
0 4-> -r-
C (O O
U O C. Q
4-)
~- C^ U~ *D
°- > -D
S-
rt i f~
UJ * -t_
CO .O (J
O .^— fO
-j o cu
\ « .
\ h i*- -1^0
\ -f o ! on oi +J
\o T'
\ ' \
\ \
X
x VN- S-
s^ \«
S «\? £
*x X « -
\ e
<4 \ • ^ 10
2« \ ° "
\ « •
/*" * \
^ * • \ v £ *. .
— - \ i-
3 •!->
o -c <* x
5? o>
S- J- .r-
CL-I- cr
ID
<4- O- in
O S-
>> Oi
• t/1 ^— fi
2 E-E
?, res to 3
o s- •,- c
O -(->
*c *o >^
o -i- O- (C
o "C CO 73
_ S_
'^
re i —
O • 3
I/O ro O
IO
I —
0)
i.
Q
Lj
128
-------
\
T
u.
1
*^1 m
.£
\
x \ "
x \\ '-
x , \ "•
\ \ 10"
\ \
V \ *
\ \
I
X N !
X\- 1
x X \ !
\ V-
^X
1 1 \
s I i
'0
""
T
m
•
|i
J_
la.
10"
X
O en
O O)
M S-
tn
10
Ol
E
T -o
n O)
ID t-
o ••-
s.
c
Z3
I ^
i/l
c
0
•M
C O
O 0.
O
10
-o •»->
o) re
o oi
• o
N t iQ
! o
—i o
I rr
Ol O)
4-> C
-------
a.
Based on all
(paired) observed
and predicted values.
lo ya to
'511—«
1 2'f
Based on pairs for which
obs. or pred. values 80 ppb.
llf
O 10 <»o
c.
151
Based on pairs for which obs.
or pred. values 120 ppb.
183
4 18-1
HSI
MS «-»—I
Figure 12. Model Bias (ppb) using all pairs (a), and stratified
data sets (b and c).
130
-------
Ot
I CM
Based on ell (paired1 observed
and predicted ozone values
underpredlctlon
i-CH
KH
Based on pairs for which
observed or predicted
values > 80 ppb
Based on pa^rs for wflicn
observed or predicted
values > 120 ppb
-0.80 -0.60 -0.40 -0.20
0.20 0.40 0.60 0.80
1.00
Fig. 13. Normalized Model Bias uslnq all oalrs («), a«<1 stratified
data sets (b and c). TO obtain the normalized bias, mean
residuals (ff-F) were divided by the term (f »
131
-------
i
!
i
i
A '
tt
CL
fib
I
*o
CD
O
M!
O
(D
O
tCROSS)
**
A
0)
(L
&
V
Kt
O
1 -U
H-
O
at
O.
fs^
o
Cc
Q.
z
a
0
o
Q
C
«
QC
O
QC
z
w
i/1
C
C
c
c
*
* '
1 !
3 !
3 1
• *
iA
« 1
rw
3
D
•
irt '
iA
0
a '
• *
S
f*
D
0 1
in
O
3
•
3
3
n
3
>
•
9
3
1
a
o
o
fib
IM
O
C
c
OJ
>
>
•
9
4
O
O
O
3
g
*
,
O
«n
«
! o
1 c
c
tr,
^
*
•
«
O
O
o
«
.
~
: .
«
• R
•
« •
M
k
* • *
v «
rsj
• *n
« «
04 M «
I 4
^ *
1 ft
•
:
o
c
o
»
1
•
c
«
•
«
IM
k M
«
«
*
•
"
«
fM
SI
3
•
3
9
!
^
IM
IM
•1 *
*
J
T *>J
^1
*
*
•
0
o
•
0
It
1
1
1
L
p*
C ^
1 ^
L *
r *
*
i
i
i «
i ru
*
4
C
c
III
)
3
t
•P
» 1
i
*
C
=
.X
9
9
1
i
>
9
y
l •
LO
C^
3
; >,
3 ro
•. f-~.
M '-'
S-
o
l»-
s. ^
; c
; o
•f^
^->
(O
3 l-
3 •!->
£
c
o
o
(L!
C
o
M
O
•o
a>
>
n Q}
" to
X3
O
(/I
3 ^
•o
O)
4->
3 (J
•r—
• -o
2 O)
K S-
Q.
*+—
3 O
£=
03
s_
cn
(t3
3 T3
; s-
3 CD
r> 4J
-l->
ID
O
3 ^
3
*
•3-
O)
s_
33
cn
•r—
U_
132
-------
-S
f-M
•-
.*="
Fiqure 15. Cumulative frequency distributions of predicted and
observed concentration.
a. Day 195 b. Day 159
133
-------
en
in
ro
O
S-
o
10
c
o
•r—
4->
to
QJ
O
E
O
O
O)
c
O
Nl
O
Ol
C/l
Jd
O
-c
QJ
+J
O
• r-
t:
o>
o.
en
re
•r*-
-o
O)
1C
C_)
OO
OJ
CD
134
-------
0
z
0
N
E
P
P
B
0
Z
0
N
E
P
P
B
see
lee —
e
-lee
OBS
PRED
RES
LIN
6 8 ie ia 14 i£ is ae
HOUR
DAY 142
aee
lee —
-lee
1 I • I • I • I ' I ' I ' I '
4 6 8 16 18 14 16 18 2«
HOUR
DAY 178
Figure 17. Time series of observed and predicted ozone concentrations and
residuals averaged over all stations for each test day.
135
-------
0
z
0
N
E
P
P
B
0
Z
0
N
E
P
P
B
see
DBS
PRED
RES
LIN
iee —
-lee
4 6 8 le
IS
HOUR
DAY 182
14 16 118 20
see
tee —
| I | I [—1 | r—p
6 8 16 18 14
HOUR
DAY 183
16 is ae
Figure 17 (continued)
136
-------
0
z
0
N
E
P
P
B
298
iee
-lee
OBS
PRED
RES
LIN
' I '
ie
I ' I '
14 16
18 26
HOUR
DAY 184
200
0
z
0
N
E
P
P
B
100 —
-iee
1 I ' I
468
I ' I M
ie 12
HOUR
DAY 207
14 16 18 20
Figure 17 (continued)
137
-------
0
z
0
N
E
P
P
B
see
leeH
-lee
— OBS
PRED
RES
LIN
1
6
8 10
( i | i |— i | r
ia 14 16 is se
HOUR
DAY
see
o
2
0
N
E
P
P
B
iee
8 10 IP. 14 16 18 80
Figure 17 (continued)
138
-------
0
z
0
N
E
P
P
B
200
DBS
PRED
RES
LIN
100 —
-100
I
8
I
10
i2
HOUR
DAY 830
14 16 18
200
0
Z
0
N
E
P
P
B
100
-100
1 ^ 1 ^ I
10 12 14
HOUR
DAV 231
16 18 20
Figure 17 (continued)
139
-------
OBS
0
z
0
N
E
P
P
B
see
lee
-lee
o
z
0
N
E
P
P
B
see
lee H
-lee
T
6
T
8
ie
14
16
I
18
HOUR
DAY 1S9
Figure 17 (continued)
140
-------
0
z
0
N
E
P
P
B
see
lee —
-lee
OBS
PRED
RES
LIN
\
1 I ' I ' I ' I ' T
4 6 8 10 12 14
HOUR
DAY 160
16 is ae
aee
o
z
0
N
E
P
P
B
iee
-lee ,
468
' I F
12
HOUR
DAV 195
14 16 18 29
Figure 17 (continued)
141
-------
0
z
0
N
E
P
P
see
lee —
-lee
— OBS
PRED
•-• RES
LIN
8 Id
IS
HOUR
DAY 211
\
14
16
I
is
ae
260
0
z
0
H
P
P
B
tee —
-lee
I
8
ie
I '
ia
HOUR
9AV S12
-r-| i—y—r
16 18 s«
Figure 17 (continued)
142
-------
0
z
0
N
E
P
P
B
200
OBS
PRED
RES
LIN
100 —
-iee
I ' I ' \ r I
8 10 12 14
HOUR
DAY 225
I
16
18 20
300
0
Z
0
N
E
P
P
B
i0e
•100
8 10
IS
HOUR
DAV 226
16 18 20
Figure 17 (continued)
143
-------
0
z
0
N
E
P
P
B
800
100 —
-lee
OBS
PRED
RES
LIN
I
6
I
8
16 18
HOUR
DAY 837
16 18 80
0
Z
0
N
E
P
P
B
800
100 —i
-100
\
4 6
I
8
10 IE
HOUR
DAV 875
!<< 16
I
18
80
Figure 17 (continued)
144
-------
OJ
4->
O
•r—
-C
c
CO
CU
0) .
vi .c
O
a< c
•*~ c.
O) in
C
'O ro
4->
W C
O
• — 4-J
ro ro
ro QJ
XJ (_)
i c
cr o
C\J (J
a;
3
145
-------
Number
of Days
-5 -1-3-2 -J o | z 34 5
Separation in hours
Figure 19. Histoqram showino the frequency of laa tines between
observed and predicted oeaks (t . - t .}.
(Day 183 excluded because it has multiple^peaks.
146
-------
ST. LOUIS JUNE 7, 1976 (DAY 159)
ESTIMATED CONCENTRATIONS OF 0 (IN Pril)
HOUR 1400-1500 3
ST. LOUIS JUNE 7, 1976N(DAY 159)
OBSERVED CONCENTRATIONS OF 0 (IN PPB)
HOUR 1400-1500 3
OOT.II5-II1I
nil
NOX tr.JSSIONS
Fici. 2Qa. Soatial fields o-f observed and predicted ozone con-
centrations for hour of peak observed 0-^ for Day 159.
Rough trajectory estimates are superimposed on THC and
NO emission density diagrams in the lower insets.
Concentrations in ppb.
147
-------
\
ST. LOUIS JULY 13, 1976 (DAY 195)
ESTIMATED CONCEN'IHATIONS OF 03 (IN PPH)
IIOUK 1500-1600
ST. LOUIS JULY 13, 1976 (DAY 195)
OBSERVED CONCENTRATIONS OF 03 (IN Pl'IJ)
HOUR 1500-1600
out. H5-1I16
THC CESSIONS
Figure 20b. Spatial fields of observed and predicted ozone concentra-
tions for hour of peak observed 03 for Day 195. Rough
trajectory estimates are superimposed on THC and NOX
emission density diaarams in the lower insets.
Concentrations in ppb.
148
-------
ST. I B'JtS SU
FSUMSTEO
HOOK 1300-140
tCPY 225)
P3 'IN FFBI
SI . LOO,5 flu: 12. 1911; ICBY ?js)
CBSFKVfD CCNClNlRSTiesS PF P3 MN rr
HOOK 1300-1400
Fiqure 20c. Spatial fields of observed and predicted ozone concentra-
tions for hour of peak observed 03 for Day 225. Rouqh
trajectory estimates are superimposed on THC and NOX
emission density diagrams in the lower insets.
Concentrations in ppb.
149
-------
ST. LOUIS AUG. 19 1975 (DAY 231)
ESTIMATED CONCENTRATIONS OF 0 (IN PPB)
HOUR 1400-1500
P'f
Si^S'^SL.1?-.".". o*r M»
THC rnjSSIONS
"• rw"
-1
E£j"i
:'i'" i'
&r,
Figure 20d. Spatial fields of observed and predicted ozone
concentrations for hour cf peak observed 0^ for
Day 231. Rough trajectory estimates are suDerimoosed
on THC and NOX emission density diagrams in the
lower insets. Concentrations in ppb.
150
-------
ST. LOUIS OCT. 1, 1976 (DAV 275)
ESTIMATED CONCENTRATIONS"OF 0, (IN PPB)
HOUR 1400-1500 J
ST. LOUIS OCT. 1, 1976 (DAY 275)
OBSERVED CONCENTRATIONS OF 0 (IN PPB)
HOUR UOO-1500
MTMK-llll
Mil
THC tnlSSIONS
OATIIII-II7I
Nflll
M0» CESSIONS
Figure 20e. Spatial fields of observed and predicted ozone concen-
trations for hour of peak observed CL for Hay 275.
Rouqh trajectory estimates are superimposed on THC and
NO emission density diagrams in the lower insets.
Concentrations in ppb.
151
-------
in iMii*
"i i7 i-j'e i it
K f 1C2
in i r 3
112 n i_ _. _ i H9
1C
-------
153
-------
CO
CL,
O
CSI
re fi
Q O
VM
O
in
r-. tc
o*. c
rH O
•H
» J-l
00 «
• c
6C CU O
3CO
< c -a-
O rH
u
w
o
•o o
3 0) m
C > rH
*" QJ Id
• 1C 3
4J ,c o
W O B
cs
p-
0 C
ro -H
CNJ ^ —
to O
O
U-, M
r^ C
ON C
bE O C
3 E C
< o CM
S-
QJ >^
S-
O
t/1 QJ
"O c:
r— O
QJ Nl
•r- O
T3
^- QJ
fO >
-I-' QJ
TO l/l
O--Q
IX) O
fO
ro
C\J
QJ
•3
CTl
4-J CC C •«-
154
-------
(O
OJ
Q.
14-
O
s_
o
S-
O
CU
O
c
O
O
Ol
c.
O
O
•r~
"C
OJ
c.
c
CO
•c r--.
0) C
> CSJ
0) >>
-------
(X
CU
Q C
fc
m
r-- co
o> Z
tH O
owe
POO
co o
M O C
P W «*
O > rH
• CO B
H » O
cr. O s:
-------
02
p-
PH
CO
< o
i/l to
r-. Z
O> O
H
>* Z
iJ W O
C3 O O
•-> Z sr
O T-l
O I
w o
MOO
:= W rn
O > r-i
>J Oi
K K
• tr. Z3
HBO
U) C S
1C
OJ
O-
EC
CM
P-i
•a-
00
f)
o
t~ o
ON M
IS
fi H
>H W
tJ U O
5 s c
•-3 O -a-
U iH
I
to c o
W Id C
H W O
S-
o
li-
t/1
c
o
re
s_
•M
c
Ol
o
c
o
(J
01
o
N
o
•c
OJ
4->
o
•r—
-o
Ol
S-
Q-
re
•c *i-
oj co
> r-
S-
Q> >>
I/) (O
^D C
O
<4- O
O M-
in O)
TD c
i— O
Q> N
•r- O
H-
•c
r— QJ
re >
-t-> QJ
re i/>
Q. -Q
1/1 O
-o
CO
CM
QJ
=3
a
-------
OJ
to
O
c
CT
(O
4->
I/O
(L)
(C
(J
O
O.
O)
to c
c o
O tsl
E O
O
CL ro
<1) r~
O Q-
C 00
O -r-
OJ
3
158
-------
5.
«
sr
*-x
v.^
O
c
f<
cL
to
C.
o
o
o
" N
'i QQ
1^''
v° <
— IT,
•O
01
tfi
S-
3
O
.? X) "d >,
ei OJ C «0
O •!-> (O C3
— U
•i- C
•o o ,
(O *»-
«O i- •(->
S- O) C
cr> > QJ
E
o > a
1/5 ID Z
ro
in
CM
a>
3
U)
159
-------
XI
O)
cr
f<3
t-
O>
-
X *•'<
\ fs
\."
N. ~
\ 'i £
\ '""^ *
\ ^ "
\
\
\
\
\^
V &
X^ %
X **
\ a-
V "*"
\o 2 •
7i -;.^N
• . M N ~
t N-%: ^
*-^(5 .^ *
• i Oi
f-^ ^2."
5 •" \ -
H *,
N
x
\
\
i * i i i i i 11
f>
c
o
•r~
4-)
to
x^y J-
to +i
Cl c
3 g
0
0
0 SX
^fj *Z~-
c
1
O -c
r> a>
i_
ai
J5
s -
^ •
ii to' .
v- -D S- >,
o. ai 3 i
E tO -r-
tO OJ 4->
S- 4-> C
O to a>
•4-> V X!
0 to 0) E
rTt U > 3
rv oo 0 z:
o
in
O <^
"~ O)
s_
3
Cr
u_
/~\ • T '
S^S1 o o o° o o «^
a.«J r^ « in <& co N
a
Ul
160
-------
01
Ol
•
m /v.
*. 5
o
" £
p«
I' II
oo
c
o
o
c c
<0 «J
g,-
01 10
-------
It
•ill
10 f
.27?
130
.20?
• 32'
12.
14
Figure 26. Scatterdiagram of mean predicted vs. observed MMHC/NO
ced over all sites for all hours.
ratios averaced over al
162
-------
•H 01
•o e
u o
o -»• co \
o »•
1 3
• •
, " !
o -» °
H ~ ~
X53*
^*
\
2 S
' i
o
«o
l
«n
ii
\
•o *
->
o u
VI C
i — 3
ro
=
0) -r-
x \n
\
163
-------
SNOISSI1AI3
o
O
O)
cu
.c
•(->
o
Q.
JC
fO
en
c
o
r—
ra
c:
cu
+j
ro
CL
to
c:
O
CO
CM
cu
S-
164
-------
c
o
u
'(£
c
o
JO
C
o
01
c
o
N
O
1C
(U
o.
S r?
OD o
en
CJ
165
-------
300 I-
O
153
O.
Q.
C
O
H3
O)
O
O
-------
a. Day
Q —
— Average Change
— Peak Change
to
40
20
/
10 4O
Percent Change In HC
tO
--20
r -40
SO
b. Day 195
fcf Average change
^- ' Peak change
-60
Figure 31. Average Percent Change and Percent Change 1n Peak 03 1n Response to Uniform Changes In Hydrocarbon Emissions.
167
-------
0/57
D
c
o
Ol
i-
o
o
0.7 -
Percent Change in HC
Figure 32. Spatial Correlation Coefficients of Base Case vs. Sensitivity Case
Changes in Hydrocarbon Emissions.
168
-------
100-
5D-
-60 -40
Percent Change in HC
O '59
-2O ^
_4O _
-60 V
Figure 33. Overall Normalized Area! Exceedance (ONAE) in Response to Uniform
Changes in Hydrocarbon.
169
-------
O 4.
NO
§T
l/»
r>.
rt
o
cT
(O X
CO
i« O)
U O)
CO
ro Ol
CO 10
o
(/>
O
^N
1
<->
r<
J
• 7
/
/
f
I
|
i
/ /
i / /'
1 - r -
• ~ i \a /
g4 °^V
i / //
1 / /
• i
'!'<•'
J^S
l*^t 1 i
Jf& '
//' ^ °
. // / CX •O^*
/// / 1 ,
//•/
///
// 1
// 1 0
<$$ 2> ^
/ 1
' , *
)
•
X
0
-$ =
1 '"~
O)
! O>
! c
1 n
1 SI
0
i 4J
10 =
7 g ^^o ^
1 o5 'i Ov N
Q- \ \ C4
r, n , ,
14- 43 1
C TJ
3 C^
O
O 3
4-> t3
OJ
CO i-
C
O O
•r- J_
•M eo
O
•i- CJ
T3 Z
2i H
a.
(U O
c -r-
O il
N 3
O
a
*"* *r-
rtJ
QJ C
Q. 0
«»- 4J
O fO
O 3
co E
C •!-
O in
CL
a)
QJ
00
O)
3
170
-------
o DO
cr> m LD
in cy> f-
•— r— CM
>,>>>,
H3 *O *O
O O O
r § x
0
z
_c
CD
CD
C
(O
1 O O
T ^
•4-J
f~.
- — | _ — - - . . .
-
!
k_
a;
0
fc
o.
>•- o
o
("\. 1
W>4
1
CD
C
TO
E
o
ac
c
O
3
CVJ
•3-
o
(J
QJ
t O
U3
I
O)
^i
4->
«»-
O
c
O
•r-
•»->
-------
c
o
ra
S-
4->
c
O)
(J
(LI
O
N
O
(O
0)
D-
O)
cr
-30
— 10
— 10
-20
-30
2.0
Percent Change in Mixing Height
D I3S
O 275
Figure 36. Response of peak ozone predictions to uniform changes in mixing hei
172
-------
to
•)->
C
cu
o
c
o
o
o
N
O
-20
2.0 *•
fO
Percent Change in Photolysis Rate
-jo
D
O 275
-2O
Figure 37. Response of peak ozone predictions to uniform
changes in photolysis rate.
173
-------
Day
195
c«y ft i i
159 / / /
Day
275
1:1 line
ej
&
.5 30
I
I 20
u
«j
o
10
_L
a.
20 40 60
Percent change in BC
10
-20
-30
-40
-50
-60
Day 159
Day 195
Day 275
1::1 Line
Figure 38. Response of Airshed peak ozone to uniform changes in
HC emissions fcr 3 days.
174
-------
60
50
I
:.
10
*• /
«'/
/
/
/
'
/ Day 139
0 20 40 60 80
Z Reduction in HC
<»»>
cTSO
-s
•*
40
130
10
20 40 60 80
I Reduction In EC
(O
50
cT *°
S. 30
J 20
*J
C*
I 10
"
•
•
_____ 3
/
/
/
/
/
/
2V
/
/
/
/
',
D«y 275
CONTROL STRATEGY CODE
1. IACI
2. WVECP (HC)*
3. THVECT (HC I HO )
4. RACT •«• IKVTCP-de)-
5. RACT + WVZCP (HC) + BOCKDART
6. RACT + fHVECP (HC) + I/M
7. RACT + FHVECP (HC) + 1/H + BOUNDART
8. RACT •»• WVZCP (HC « HO^) + I/M + BOUKDAK
* P»r«nth«$ti iodicate pollutant* controlled.
A level of control required to
meet ozone NAAQS.
20 40 60 80
Z Induction in HC
Figure 39. Control strategy simulations for 3 days.
discussion.
See text for further
175
-------
no v i ss
UHT Ifaa MET
HBCT FHVECP FMfECP
,
D
i?12
,
D
288
n-ie
e
D
zss
*
D
£?•
i
D
216
0
167
'
0
PS
16-18
• HCrCX
»HCs-42I «HCi-BBX
rnvECf
0
174
16-17
,
D
1SS
17-1B
D
a
1SS
16-17
«
D
141
IB-IB
•
D
199
17-18
W
D
121
16-17
*
D
103
17-18
»MCr-27Z
»HCr-6BI *HCc-7SX
•HCrOZ
RRCT
fftncr
D '
232
14-IE
0 .
209
16-17
0 .
214
1E-16
o
0
160
18-17
It
0
149
18-17
•
D
120
15-16
D
D
as
16-17
»MCx-J4I *HC:-42Z
= Urban area.
= Location of predicted peak ozone.
(Concentration in ppb and time of
peak shown on each diagram.)
Figure 40. Summary of Airshed Control Strategy Simulations.
176
-------
2
r
M
2
O
H
U)
VI
p
b
0
0
o
0
o
o
-
1 *
in
sr
w
o
s
H «
Bo
p w
I >
[> o.
H X
..
3 0
» O
i O
3 0
3 O
3
3
D O
3 O
> 0 O 0
D
O
D
5 O O O O
4 vO
O N
"»
g ui r- o» jr
3 a- co
D 0* W
m
'
W
/I
0 P,
•e H
0 H
5 Z X ^ W
C O O W «
* H H H
J M H U *
3 0 U H
< tf D t 0
i W PS U W
H U >
H Q W OB «<
1 Z Z O Pu
3 H o u as
:
i
•
o *
O 0
CO CO
r
in <
r- i
•J
n
•v
n
00* O*
Is 5
M
£
~
M
-
•»
T
. . | .
mo Qt
«O N N
vo o a-
*- o
o »> o
O ff* \O
or* a*
r r
I
J f* CT.
JO r-
op j:
o* ir to
4O
in «
U M
u u
p: pt
3 3
0 0
in in
in
T * W O
1 W Z <
k PC H O
t-« =3 K 3 O *i » a •< 1-3 fti
H r s ^
C3
P
1
!
9 H
•1
s
• • i .
.
m
~
s
I
- 0
T «
M m Id- «o
BO y r» --
o
m in ea «- er>
ll »- => •< a. o u
i U, X •-* U =
J O O Q O
H H H H E
»> W W W W E
4 U M W W O
C CL PC « p! U
*
,
SS
«
r
0 0
o
p- 0
•n
*
n *-
3 0
~*
r o Io in
" SS S
-
r o ff o o o
4 CO
N
o in
•i
1
i
- *0
n in
3 <*f
-0
•"» M
i m
- M
Q 0
0 0
O
n
M
•^ ro CO
D O
i
n
o
ootnooc o o o
r* o f- o o o r o o p
" P
- 0
n
-0
0 CO
o
4 UJ
3
4
—
o
* '
9 O
0
H M H
H £ X
< rt <
OHM
U ^1
• U
U O
< u, H X
C . (A £D «
3 J W * Q
H 0 W 1 1 1
< a. o o o o
e HJ u x 3: n
: c E Q. P- a.
: n n < S «:
3 0 0 > > >
J U O W M W
2
0 -
c^
rf
0
0 0.
t/1
H
0
UKAL
UASTE
3 0
3
p
U
M
^
0
o tu
3= se
b H •<
Z ui t-1
' H •< U
> Q H U U
M 2 « O
J H U W
•* O fS W H
1 i/l P. Q C
j ,
' i
J
0
O tO
eo r>
BO 0>
« ^ ' ' in
j s- •- in o
r- <*» p* *o
s
») (*>
. . .
o eo •-
N to m
N
ro o
N iO
o% m
0- o> eo o
-- «- k- CO
N£sss
M
» r-
u
« p a.
W & H
r H -*
3 U O 3 U H
: >^ , 01 & P:
u « w p; H
< PS ~ H «
< 0 Z C M 3
X H 3 06 Z Q
•1 0 < < 0 Z
C E >J fc. O H
!>
f
i— t
tO C
1 a* '
i
•j
o
o
i
c
n
CO
A In
^ o
s. p
s-
O 0
n er. •
t JT
^«
1 (*•
0 D
r co i
r *n <
3
*>
W
U
* pi
< p
3 O
i E
> C '
tt rt
< H
Q D
SI
(
3 H <
[
« f
t t
•
1
177
-------
Table 2
01976A8CR76
»aCR(070) 'METROPOLITAN
S REPORT U/-14-BZ *
ST. LOUIS (ILL-MO)
EMISSIONS IH TONS/YEAR
FART
SOX
FOR 1976.
NOX
HC
CO
FUEL COMBUSTION
EXTCOttB BOILER
BITUMINOUS COAL
KEaj-UuAi, U!L
DISTILLATE OIL
TOTALCELECTRIC GENER)
INDUSTRIAL
RESIDUAL OIL
NATURAL GAS
OTHER/NOT CLASIFD
BITUMINOUS COAL
DISTILLATE OIL
TOTAL(COMMERCl-IHSTU)
INTLKKLUOHBUVI'ION
ELECTRIC GENERATN
NATURAL GAS
TUTALtLLECTRIC lilM.R)
TOTAL (INTERNLCOMBUST)
INTERNLCOMBUSTION
ELECTRIC bEKEKATN
DIESEL
UlLbLL FUEL
TOTALCINDUSTRIAL )
TUTAUlNTtKKLUUniJUl.!)
TOTAL(FUEL COMBUSTION)
6357
it
to
»>t07
51
72
63
1512
2*1
1572
1
1
1
1
11791
.20
.70
. ~l i
.57
.50
.33
.00
.77
. 11
.02
.«»8
.77
.77
.61
.76
.?B
.69
£53131
517
858252
1181
171
2251
1592
6
1947
i
i
i
i
8«8601
.95
.68
.31
. 11
.31
.00
.01
.91
.91
.07
.87
.67
.59
.17
. 17
.20
2855i!3
liiio
2')7
2878:>9
126
3911
1856
199
2
151
228
2 JZ
232
«
21
*k
298
-------
c
i
Ul
VI
1
o
CM
£
o
X
£
•o c
•p
J
u
I1
£
o
o
t
,2g
•a M
moi L.
t-
X 4J J) Q.
<0 <« E Q.
§
2 2
179
-------
Table 4
DESCRIPTION OF THE INPUT FILES TO THE AIRSHED MODEL
DIFFBREAK This file contains the mixing height for each column
of cells at the beginning and end of each hour of the
simulation.
RE6IONTOP This file contains the height of each column of cells at
the beginning and end of each hour of the simulation. If
this height is greater than the mixing height, the cell or
cells above the mixing height are assumed to be within an
inversion.
WIKD This file contains the x and y components of the wind
velocity for every grid cell for each hour of the simu-
lation. Also the maximum wind speed for the entire grid
and average wind speeds at each boundary for each hour
are included on this file.
METSCALARS This file contains the hourly values of the meteorological
parameters that do not vary spatially. These scalars are
the N02 photolysis rate constant, the concentration of
water vapor, the temperature gradient above and below the
Inversion base, the atmospheric pressure, and the exposure
class.
AIRQUALITY This file contains the initial concentrations of each
species for each grid cell at the start of the simulation.
BOUNDARY This file contains the location of the modeling region
boundaries. This file also contains the concentration of
each species that 1s used as the boundary condition along
each boundary segment at each vertical level.
TOPCONC This file contains the concentration of each species for
the area above the modeling region. These concentrations
are the boundary conditions for vertical integration.
TEMPERATUR This file contains the hourly temperature for each surface
layer grid cell.
EMISSIONS This file contains the ground level! emissions of NO, N0?,
five carbon bond categories, and CO for each grid square
for each hour of the simulation.
PTSOURCE
TERRAIN
This file contains the point source information,
including the stack height, temperature and flow
rate, the plume rise, the grid cell into which the
emissions an; emitted, and the emissions rates for
NO, NOp, five carbon bond categories, and CO for
each point source for each hour.
This file contains the value of the surface roughness
and deposition factor for each grid square.
180
-------
Table 5
Example of an array of predicted and observed concentrations
from which performance measures are calculated*
OAT 226
HR/STH
5
6
7
8
»
10
11
12
13
14
15
It
17
18
If
20
IUX
OB
PR
OB
PR
OB
PR
OB
PR
OB
PR
OB
PR
OB
PR
OB
PR
OB
PR
OB
PR
OB
PR
OB
PR
OB
PR
OB
PR
OB
PR
OB
PR
OB
PR
101 in 103 104 105 106 107 108 109 11« 111 112 113
10 11 10 16 20 19 16 17 19 33 32 24 17
9 21 '-8 20 4 -8 17 21 19 15 -8 -8 21
22 25 22 29 30 28 29 31 30 43 38 29 31
41 48 -8 25 17 -8 42 74 38 23 -8 -8 58
28 33 30 35 34 31 33 40 40 46 38 33 34
89 83 -8 59 65 -8 73 107 89 56 -8 -8 85
34 43 38 38 38 39 42 48 44 46 44 40 43
-8 106 117 89 94 -8 98 116 141 125 -8 -8 106
43 51 50 44 43 42 48 60 53 53 47 44 48
-8 122 134 117 121 -8 102 121 180 146 -8 -8 126
47 55 52 46 46 45 54 58 63 50 53 53 53
-8 122 122 125 135 -8 103 128 194 -8 -8 -8 142
54 70 56 51 52 56 68 65 69 58 72 74 72
-8 129-178 131 136 -8 99 129 (liT) 178 -8 -8 142
61 90 68 57 57 66 84 76 ( 77j 71 90 86 90
-8 133 -8 148 159 -8 116 124 -8 -8 -8 -8 139
75 104 84 70 65 72 103 83 97 80 95 96 99
-8 149 -8 152 180 -8 128 132 -8 -8 -8 -8 124
81 111 102 90 77 79 111 89 114 95 92 108 102
-8 446 155 114 143 -8 132 155 146 137 -8 -8 101
89 104 112 108 94 79 91 100 109 112 101 105 101
-8 121 116 60 100 -8 104 124 118 107 -a -a 134
105 95 111 111 110 97 86 103 64 111 104 84 88
66 86 80 57 • 78 -B 72 92 83 86 -8 -8 87
79 83 69 se 80 86 10 56 21 55 70 87 81
56 50 61 38 56 -8 40 58 68 67 -8 -8 59
27 41 19 23 32 35 43 12 12 21 25 34 53
31 27 41 32 28 -8 17 37 52 45 -8 -8 25
141 115671 111
89 149 178 152 180 132 155 I 225) 178 142
79 111 112 111 110 111 103 l_l£»J 112 102
114
-a
20
-8
36
-8
43
-a
48
-8
55
-8
59
-a
60
-a
71
-8
74
-a
64
-a
70
-8
95
-a
84
-8
48
-8
17
115
-8
34
-8
47
-8
54
-a
66
-8
68
-a
71
-8
67
-a
65
-8
67
-8
71
-a
73
-a
48
-8
18
-8
12
-8
10
116
4
31
14
43
51
55
90
66
134
84
147
86
158
92
142
101
180
98
170
87
139
60
111
28
83
12
57
3
55
9
180
101
117
12
3
36
-8
46
-8
54
-8
62
-a
67
-8
79
-8
94
-8
104
-8
112
-8
112
-8
85
-8
34
-8
16
-a
13
-8
3
3
36
118 119
is -a
19 15
11 -8
43 32
15 -8
50 31
34 -a
48 35
64 -8
49 40
105 -8
51 48
129 -8
SI 71
-8 123
62 92
139 125
as 100
-a 120
99 109
-8 125
104 117
121 153
97 119
114 -8
82 102
aa -a
58 65
72 -8
10 10
43 -8
1
139 153
97 119
120
-a
7
-a
20
-a
22
-8
33
-a
37
-a
59
-a
76
-a
85
-8
89
-a
95
-a
107
-8
117
-8
111
-8
es
-a
45
-8
28
1".
21
15
23
30
46
36
59
39
82
46
90
50
87
66
-a
82
91
90
-8
94
-8
90
80
83
69
72
83
83
71
69
45
38
91
90
MAX
21
19
23
43
46
50
74
55
107
66
141
84
iao
86
194
92
225
[101
180
109
180
117
155
119
134
111
92
90
72
69
55
38
225
119
~" — v
ft
F4
F3
*The F and A measures are defined 1n Table 6. Missing values are
denoted by the value -8.
181
-------
Table 6a
Measures Calculated for Each Day Separately
\Measure
PairingX
Method \
Fl (Totally Paired)
F2 (Paired in Time)
F3 (Paired in Space
F4 (Unpaired)
Al (Paired in Time)
A2 (Paired in Space)
A4 (Unpaired,**
highest 25)
Residual* Number of Bias Standard 95% Conf-
(d) Residuals /-n Deviation of Interva'
Per Day va; S Bias
\ \
/ 1
1 > NA > NA ) NA
/ 1
' 1 J J >
/ N hrs d" = sd/N Srf/ /
/ M sites 3" = id/M Sd^ /
25 obs. d~ = U-P~ S /
25 pred. S / /
All residuals (d) are based on 0-P (Observed minus predicted concentrations]
thus positive residual or bias is underprediction.
A4 is an unpaired bias calculated as follows
(B)25 - m
25
where (
-------
Fl (Totally paired)
F2 (Paired in Time
F3 (Paired in Space)
F4 (Unpaired)
Table 6b.
Measures for 20-Day Data Set
\ Measure
Pairing\
Method \
20 Day
Bias
Estimate
Standard
Deviation
of Bias,
95% Confidence
Intervals
Scatter Diagrams and
Correlation Coefficients
of Predicted vs. Observed
Concentrations
See (a)
See (a)
(0 vs. P)
Diagrams and r values
obtained with the 4
different pairings.
Each diagram contains
20 points, 1 per
day
Al (Paired in Time)
A2 (Paired in Space)
A4 (Unpaired,
highest 25)
See (b)
Sd
S-.S- see (c) '
(ff vs. V)
Diagram and r values
obtained with the 3
different pairings
of average observed
and predicted
concentrations
a. For the F measures, the 20-day bias is the average of single residuals
(d, Table 6a). S2d = z(d.-?)2/(20-l) 1 = day
b. For the A measures, the 20-day bias is the average of average residuals
(3" , Table 6a).
The average standard deviations for the 20 days are obtained as follows:
2
7) /(20-1) t « day
2 7 Q
S = z
2 20 2
S - * z (cTj-'o) /(20-1) t « day
2 20,
5 a *
/(20-1) 1 = day
c. 95$ confidence intervals are constructed as follows:
B" + t n,K S/vfT where N = number of days
•" • U^D
S = appropriate standard deviation (Sd for paired analysis and
+ S for F4 and A4, unpaired measures).
o p
183
-------
Table 7
Procedures to Assess Overall Accuracy and Precision*
Attribute
Accuracy
Precision
Measure
Pairing/
Stratification
Paired Analysis
1. All pairs
2. > 80 ppb**
3. > 120 ppb***
Unpaired Analysis
Frequency
Distribution
(0 vs. P)
'
Bias
;
'
Avge.
Absolute
Error
1 71
/
Standard
Deviation.
S
V
;:;
0
sp
95%
Confidence
Intervals
J
'
Other
For all
Scatter
diagram
observe
predict
concent
Analysi
spatial
tempora
tributi
errors.
Formulae
Paired
(
( 0 . p )
—IT— T —TT~™ '
*Statistics were calculated separately for each day.
':*"Pairs with observed or_ predicted values > 80 ppb.
Pairs with observed or_ predicted values > 12o ppb.
184
-------
Table 8
Procedures to Assess the Model's Ability to
Replicate Spatial and Temporal Patterns
I. Spatial Patterns
A. Visual comparison of spatial plots of observed and predicted ozone
concentrations for the hour of maximum observed ozone concentration.
Comparison of exceedance zones.
B. Spatial Correlation Coefficients (for each day)
r is the spatial correlation coefficient for the hour of the
m maximum observed concentration.
F is the average spatial correlation coefficient (averaged
over all hours).
where
m'
m'
sm
where
and
Pim
m
m
is the observed ozone for station i at the hour of
the maximum observed ozone
is the predicted ozone for station i at the hour of
the maximum observed ozone.
is the mean observed ozone at the time of the observed
maximum averaged over all sites.
is the mean predicted ozone at the time of the observed
maximum averaged over all sites.
I is the number of sites;
- 1
rs ' T
where T is the number of hours and r . is the spatial correlation
coefficient for hour t.
185
-------
Table 8 continued
II. Temporal Patterns
A. Visual comparison of time series for observed and predicted
ozone for each day.
1. for the station of the maximum observed ozone.
2. averaged over all stations.
B. Temporal Correlation Coefficients (for each day)
r.
where
where
and
m
is the temporal correlation coefficient for the station
of the observed maximum concentration.
is the average temporal correlation coefficient (averaged
over all stations).
(0
tn
(P
tm
m
2 T
( E (P
tm
- V V
'tm
tm
m
m
is the observed ozone for hour t at the site of the
maximum observed ozone;
is the predicted ozone for hour t at the site of the
maximum observed ozone.
is the mean observed concentration at the site of the
maximum observed ozone, averaged over all hours;
is the mean predicted ozone concentration at the site
of the maximum observed ozone averaged over all hours;
T is the number of hours.
rt = T
.
where I is the number of sites and r. is ;he temporal correlation
'i
coefficient for site i.
186
-------
Table 8 continued
III. Overall Correlation
r - overall correlation coefficient for each day
TI
IE (Qt1 - *D (Pt1 - P)
r 8 ^—^
[EI (Ot1 - ff) zz (Pt1 - F) ]
where
T = number of hours
I = number of sites
Q+i - observation at hour t, site i
IT = daily mean concentration averaged over all sites and hours
?t^ = prediction at hour t, site i
F = daily mean concentration averaged over all sites and hours.
187
-------
CO
to
*
in
ro
O
O
CM
u
(O
OJ
s_
o
u
(O
3
O
u
ct
ro
0)
c u
oe
O. _l
0. U
=
•"•3-1
1 OU
«_1
oS
«UJ
cc
in
CO
i
^.
OK
CV
1
09
CD
CM
'
^
O
00
5
m
i
03
P)
1
in
' 0
CVI
1
».
CM
1
K
f
CVI CM « M
•*• -o •»• in
•y ••• CM CM
o n •« •*
fs ts O »•
(v/ •«•
1
•«• K T MJ
OK "•• so K
1
in OK «v «•
•r o so »
OK •>• (VI (VI
<• -O N *
U) CM (VI *•
Is (VI < *
<»-•«- (VI •«-
CM •»• CM CM
rs co •«- o
"• 1 «•
1 1
tf OK CM sO
CM co f in
•«• in co CM
•«• *• CM P)
-o in f -r
p> i i
i
•«• -• OK 0
N in •>- *
i
•C P> -O O
•c CM o! V
«• in ^ CM
». is. so rs.
J fc 8 T
i
CO O -O (V
o ^ •£
-o o tvi in
So •»• o
000
< O OD Is.
rs" o CM *
in •* <• *^
«O 00 O I.'O
•c co in •«•
00 OK OK O
«• f CM
1
o rs. in ^
CM co -r oo
P) CM CO t
O OK OK K
CM * CM IVI
•»•*•»• 00
1
•»• -O •«• 09
in -o o •«•
•^ CM CM
OK
PJ
O
»•
(VI
o
0
CM
00
CM
CM
CVI
1
rs.
CM
CO
P)
00
CM
CM
CVI
OK
•C
sO
P)
m
o-
O
o
"»•
fs
CM
1
OK
(VI
CM
(M
0
CO
t-
10
K
1
„
CVI
1
0>
OK
(«
j
<
0
03
IO
o
in
i
•. CM o n
•o
00
in
00
I
CM
0
to
CM
O
•*•
o
•c
•o
so
o
CM
OK
CM
"O
•o
PJ
<
OK
PJ
tn
OK
CO
»
(VI
10
PJ
K
OK
0
CM
»
in
T
CM
sO
£
CM
*
^
*
CM
O
OK
*
z
CM
1
*
in
o
CM
CM
t
i
CM
P)
PJ
1
PJ
•C
I
^.
rt
i
O-
1
« ^r * ts CM
PJ I CM •«• UJ
CVI
oo
Ifl
OK
3
P) CVI CM
•O »• •«•
00 »
OK CVI
li.
M
UJ«C9
0§J«
XOJ
«ffi
CO
>.
OK sfl OK fs (VI
fs •» in ^ uj
(VI 00 » CM OK
<• so o PJ o
in , CM CM » *
O O » ts 00
•«• CM •»• f *•
CM CO CM CO *
•r rs oo co oo
•f CM
•C 00
•«• K
in o
«- (VI
rs o-
o o
00
CM
CM
-0
•o
CM
00
OK
OK
CO
co
OK
o
1
(VI
o
P)
PJ
CM
PJ
P>
PJ
OK
OK
OK
K
U)
«
o.
CM
CM
rs
OK
tn
^
o
o
09
•O
CM
CM
O
«O
"
PJ
<0
UJ
PI
CM
CM
in
o
CM
CM
OK
a>
tn
^.
CM S>
•»• -o
CM «•
O sO
K sO
CM UJ
•«• CM
(M (VI
00 K
PJ O
» CM
in so
CM IS
CM •«•
•O Is
(M CO
CM CM
CM
CO
Is
CM
in
(Vi
188
-------
Table 9b
'•/\<« and "p" Residuals of Peak Accuracy: 20-Day Average
Residuals and 95% Confidence Intervals*
Measure
Fl
F2
F3
F4**
A1
A2
A4**
Residual
63
48
48
31
12
14
16
Low CL
47
31
34
11
5
4
-4
UP CL
79
65
62
51
19
24
35
Std. Dev.
36
38
31
45
16
22
44
Source
AMS
AMS
AMS
AMS
Cox
Cox
Cox
Residuals, deviations in ppb.
**
Confidence intervals for unpaired measures based on S = /^ . <.2
o 5 p
189
-------
•o
01
VI
+J
C5 f-
r— _J
at «
J3 C
in c
*"^
z
•
c
O VI
g s 4J
Jn-r-
en _j
£
z
Q.
[/) C
"*»^ p-
O n
c.
1/1 U.
ex
in
o
CO
d£
c
o tta
M •"
in c
en •-*
n"
to
z
>^
s
ro o> vO co vn o O cr^ »— ^- r— co vO
o co Q m *~ invocr»cou^r—tr>ro
+ 1
,
•— «nr-.«»-inooenincOr-oo
CM CMr— »— enioiomwenr— «n
tooinr— ,— t^f-wcnoinroo
in en r— in w «• i— en r— r— en
cocMCMcncMCOiocMOVr— cor»
toiocMentrcnmLnoDCOtoenen
^ ^
coocnoo3r*»*rentootinom
inr«.encMcnenenioinrN.«^cnr—
r— CM I— CO I— CM CM
1 I 1
inr*«roeninr^CMio»wcMro
oencotocncMCMOininoocn
• co t*» co o CM in o oo f*- o co oo CM
t r-'Cn^-ocoi'^cOf— enr»»cDOi—
p-cor»CM«»i— snor»ototom
tocMincncocn^rN-oor<<^tOr-^
CM CO CM ^~
cotooocncoo*— CMCMO\^>^>^>
en en f*~ oo r~ i^* en CM i& en f^ **** CM
Or— CMCncnOror^^mcMcorw
CMCO CM»— r— r— r— CM PO
vcncncMiocnr-. cnootoinr— en
enenCMcncncMcncMCMCMCMCoen
4-1
toin^-tochtoininenocMmin
Ornininoomr».cM<»*CMr—
T ii *~ *~ T i T "~
en in *^ »"~ to o^ co ^™ to en !•." c^ c*4
rocoC3^r^vtoehrocn^f^.
^.
in
r~
CM
r—
r™
r™
CO
in
^.
00
1
ffl
s
vn
o
03
CM
*™
in
CM
_
O
1
CO
in
CM
gBV
CO
O
co
»•»
to
CM
^.
CM
r-
en
CO
?
S
en
en
SI
en
eo
*~
in
CM
r-
V—
en
CM
tM
CM
Ol
«>
CJ)
^p
r"
1"
HI
03
L-5
en
t.a
W5
n
C4
CO
«
'"
INI
im
'£>
10
CM
CM
in
CM
CM
to
rs.
en
o
to
en
in
CM
to
to
r«.
o
CO
in
CM
0
co
o>
r-
o
in
10
CO
CM
en
to
CM
CM
in
r-»
CO
r^
in
in
in
f-
«9-
in
3
0
to
t
CM
^
0
3
CM
CM
o
^.
CO
er>
CM
r—
CM
!».
en
CM
en
_
^"
en
in
en
en
in
en
,,.
in
en
i
CO
r^
en
o
^
§
in
^^
en
in
en
en
CM
i
n
in
CM
•
0
CO
Al
Vt
3
"i
>»
•%
U
1
CL
I.
O
•a
S
Q)
VI
•§
JT
x
L.
•i?
(O
CX
*
1
^
^
1
g
a.
in
®
w
c
(O
x:
M
Irt
at
i/i
•»~
jE
*
en
•
o
u.
a
o w
•o Jj o
ttJ ij
0 £ „
^5 ** 3
cL -M w
*" 4J OJ
0 « N
"S S"5
> r— I/I
J- Id T-
O W r~
s- tn 3
01 U
I/I JC. r—
i *• 3
« c
O- •-" «
« ""
190
-------
Table lOb
Bias and Confidence Limits (Unpaired)
Day N TMP 95% Confidence Limits
142
178
182
183
184
207
209
221
230
231
251
159
160
195
211
212
225
226
237
275
233
235
304
241
276
249
263
291
236
299
281
279
282
224
258
243
247
191
277
300
-10.6
3.5
- 5.4
•-5.6
0.9
10.6
5.5
17.5
-12.3
4.0
4.2
-12.3
11.5
5.9
-10.1
1.1
26.6
28.6
29.8
-23.9
+ 9.0
10.4
4.8
7.0
6.3
6.9
5.7
5.9
5.8
5.3
5.3
8.0
8.1
8.1
6.1
7.01
5.0
9.0
6.8
11
191
-------
Table n
Correlation Coefficients Between Predicted and Observed 0.
Day
142-75
178-75
182-75
183-75
184-75
207-75
209-75
221-75
230-75
231-75
251-75
159-76
160-76
195-76
211-76
212-76
225-76
226-76
237-76
275-76
r
.86
.85
.83
.84
.73
.88
.72
.80
.77
.79
.81
.85
.89
.90
.84
.91
.69
.76
.69
.70
r.
s_
P
.26
.58
.56
.47
.83
.70
.46
.80
.15
.90
.74
.52
.54
.73
.81
.78
-.38
-.29
.46
.12
r.
s
.27
.47
.46
.38
.38
.36
.29
.36
.27
.32
.37
.35
.56
.39
.42
.59
.44
.10
.32
.23
r.
t_
P
.87
.93
.65
.83
.88
.96
.89
.91
.86
.94
.97
.93
.78
.98
.94
.96
.75
.78
.90
.83
r.
t
.90
.90
.86
.86
.70
.92
.75
.82
.84
.83
.86
.90
.90
.92
.74
.92
.74
.76
.76
.78
Avg. .81 .49 .37 .88 .83
r ^ the overall correlation coefficient for each day, obtained
by using all paired data sets.
r * the spatial correlation coefficient for the hour of the
p maximum observed concentration.
7 * the average spatial correlation coefficient (averaged over
all hours).
rt * the temporal correlation coefficient for the station of
p the observed maximum concentration.
Ft ^ the average temporal correlation coefficient (averaged over
all stations).
192
-------
Table 12
Correlation of F4 Residuals vs. Input Parameters
for 20-Day Test Set
Variable
Wind speed
Average temp.
Solar rad.
Max. mix ht.
Ozone aloft
Correl .
coeff, r
.53
.37
-.21
-.29
-.28
Sig. at
95% lev.
Yes
Marginal
No
No
No
Table 13
CARBON-BOND FRACTIONS (AS CARBON) OF TOTAL REACTIVE HYDRO-
CARBON EMISSIONS IN ST. LOUIS, TULSA, PHILADELPHIA, AND
LOS ANGELES.
AVG. CARBON FRACTION OF RHC
PAR
OLE
ETH
ARO
CARB
ST.
LOUIS
74.5
4.8
4.3
14.9
1.5
TULSA
80.0
4.2
4.2
8.0
3.6
PHIL
74.0
2.8
4.1
13.2
5.9
LA
70S
4.9
9.0
15.4
4.6
193
-------
cu
Cn
C
in
fi»j
I ^
CM
>,
10
Q
.*
03
CU
O-
XT
O
CO XI
O 0-
0.
CM
CO
CM
r-.
CM
ro
V£>
CM
O
en
If!'
CM
C\il
CO
in
co
i
O
CM
CM
vo
i
CO
oo
O C
4-> O
CO £
(/) T-
c -o
O C
a. o
in o
cu
a: r-
co
c ••-
in
&
ai
c
fO
cu co xa
a. o o.
CL
in
CM
CO
0>
•
CM
I
ai
us
vc
CO
I
CTi
in
o
CM
01
CO
CO
o
«d-
I
CO
O
c c
O I-H
•r—
•M T3
ro c
5- (O
c
cu
o
C
O
in
c
o
-r-
V)
in
o
isl C
o o
XI
-D t-
u
o o
•r- i.
T3 T3
CU >,
i. a:
Q.
C
JlJ •!-
0}
CU v
a. cu
en
c c
•i- to
x:
cu o
en
c
IO
x:
o
in
3
C3
ai
en
c:
cu co xs
tx o a.
o.
Ovl
CM
^1-
in
CM
CO
CO
I—
I
CTl
co
CM
oo
•
o
CO
I
IO
CM
in
m
i
cr>
co
c o
o ••-
•r- +->
4-) Q.
ro -.-
r— S_
3 O
e ui
•^- CO
00 O
tu
t/1
(O
cu
in
1C
ua
in
c
o
en
in
CU
'*. VD
O
c >.
c
o
•I—
in
in
UJ
cu
in
ro
cu
in
O
m
tn
cu
in
(O
ai s«
v. in
o
OJ >>
c
o
in
E
CU
in
>
Q XJ
in
O
cu
V)
(O VJ
CU CM
S- «*
CO >>
O XI
in
c
o
•r-
en
1/1
•f
UJ
cu
in
to *«
cu m
i- r-~
u
cu >>
O XJ
194
-------
U)
cu
en
c
fO
s-
(C
o
o
s_
-o
O)
ro
E
O
o c
to +-> •<-
•r- i-
•
O
cr.
ID
CT)
00
ID
O1
CJ
CTl
o
E
O
00
CO
LlJ
•o
N ~>
t-~
03
(1) i—
ISI O.
in
01
CO -
O>
O)
t^)
-------
Table 16
Effect of Uniform Hydrocarbon Emission Changes on
Overall Normalized Areal Exceedance (ONAE)
i
Sensitivity Test ONAE
Description %
+67% HC, D159 -31.0
+67% HC, D195 -81.1
-42% HC, D159 +34.3
-42% HC, D195 +80.6
-75% HC, D159 +89.7
-75% HC, D195 f100.0
, NGE. - NGEC
'ONAE = —2 !_ x 100%
NGEb
where NGE = number of grid cells exceeding the NAAQS
for ozone during simulation
and b = base case
s = sensitivity case
196
-------
Table 17
Change in peak predicted ozone concentration in response to uniform
changes in oxides of nitrogen emissions and initial conditions
Simulation Description
Decrease HC Emissions
by 42%
Decrease HC Emissions
by 42%, Increase NO
Emissions by 40$ x
Decrease HC Emissions
by 42%, Increase NO
Emissions by 20% x
Decrease HC Emissions
by 42%, Decrease NO
Emissions by 20% x
Decrease HC Emissions
by 42%, Decrease NO
Emissions by 40% x
Day 159 Day 195 Day 275
Peak% Peak% Peak%
0- Change 0. Change 0, Change
(PPb) (ppb) (ppb)
216
139
187 -13.4 131 -5.8
242 +12.0 142 +2.2
149
123 -17.5
134 -10.1
170 +14.1
176 +18.1
197
-------
Table 18
Fraction of total reactive carbon in each carbon-bond category
used for uniform reactivity sensitivity tests
Simulation Carbon Fraction
Description PAR OLE ETH ARO CARB
St. Louis Uniform
Reactivity 0.745 0.048 0.043 0.149 0.015
Philadelphia Uniform
Reactivity 0.740 0.028 0.041 0.132 0.059
Tulsa Uniform
Reactivity 0.800 0.042 0.042 0.080 0.036
BOM Uniform
Reactivity 0.639 0.054 0.054 0.203 0.050
198
-------
Table 19
Change in peak predicted ozone concentration in response
to uniform changes in hydrocarbon reactivity
Simulation Description
Day 195
Day 275
Peak% Peak
0- Change 0- Change
(ppb) (ppb)
St. Louis Uniform Reactivity
(Base Case)
172
216
Philadelphia Uniform
Reactivity
180 +4.7
241
+11.6
Tulsa Uniform
Reactivity
170 -1.2
215
- 0.5
BOM Uniform
Reactivity
187
+8.7
280
+29.6
199
-------
Table 20
Effect of hydrocarbon reactivity on the change in peak predicted ozone
concentration in response to uniform changes in hydrocarbon
emissions and initial conditions
Simulation Description Day 195 Day 275
Peak % Peal< %
0, Change 0, Change
(ppb) (pjib)
Base Case 174 - 232
Decrease HC Emissions
by 42% 139 -20.1 149 -35.8
BOM Uniform Reactivity 187 - 280
BOM Uniform Reactivity,
Decrease HC Emissions
by 42% 149 -20.3 174 -37.9
200
-------
Table 21
Change in peak predicted ozone concentration in response
to other sensitivity tests
Simulation Description
Base Case
Day 195
Peal<~~~ %
GO Change
(ppb)
174
Day 275
PeaF %
0~ Change
(ppb)
232
Uniform Spatial Distribution
HI -18.!
189 -18.5
St. Louis Uniform Reactivity
172 - 1.1
216 - 6.9
Temporal Redistribution
184 + 5.7
281 +21.1
Increase Mixing Height
by 25%
176 +1.1
236 + 1.7
Decrease Mixing Height
by 25%
177 + 1.7
310 +33.6
Increase Photolysis Rates
by 10%
187 + 7.5
254 + 9.5
Decrease Photolysis Rates
by 10%
160 - 8.0
214 - 7.8
201
-------
fD
CM
CM
CU
rei
Ul
C
o
s.
o
VI
o
c:
O
re
cn
CU
4J
CO
O
4J
O
*»->
••— ^
l_
t— 5
«=c
Q£
V_-x
*»-
O
c
o
•r-
4-J
re
u
•r—
r—
O.
Q.
to
C. >,
CO i~^
•r-*
4-> Ul
Q. C
•i- O
i. •<-
O i/t
U) V)
CU T-
Q E
CU
>
S-
re
c:
o
•r~
4-*
re
4-)
Ul
CU
O
^3
-o
CU
Oi
c
o
•p-
4->
•r—
"O
c
o
o
p—
re
•p-
4->
•r—
C
•r~
O
31
CU
CJ
3
-0
CU .
ex. in
t^*s
CM
-
Ul "O
cu c:
o re
s_
3 LO
O cn
V) i —
cu cn
i. in
t3 **"""
T5 V)
C >,
to to
0
4->
c
•r™ •
O >>
Q.I—
p—
cu re
4J c:
re o
•r— «p~
&_ 4-^
Q- S-
0 0
S- Q.
Q- O
Q. S-
re ex
cu
T3
cu
u.
cu
jz:
4->
re
u
•p~
p-~
CX
CX
^J
cu
-C
4-J
>>
XI
U)
c:
o
•r—
t/1
V)
•r™
cu
o
3Z
CU
4-> T-
0 TD
3 C
"O O
cu o
S-
*—
cu re
Ul •!-
O) 4J
XT *r—
4-J CI
»f—
>^
r— CU
CX4J
<£
j2
•p™
•
•*""»• ~C
Q- C
O C
z: o
u_ .^.
• — - cn
— a)
e s.-
re
s- cr
Cn £;
O -r-
0- *CJ
T3 *
r— O IT)
o f= r^.
S- CM
4->
0 4J C
c_> re
;to
cu ID in
r- > CT1
O 3 r—
*r~
x: >> «
cu r— cn
=> E LT)
W r—
S- 0
0 «t- Ul
4-> -i- >,
o c: re
2: 3 Q
CU i—
4= re
4-J -r-
4~>
H- CO •>•-
O Ul C
^J •^"
c x:
O 4-> <0
•f- J=
4-* ^) 4^
re •—
u CL c
•r- Q_-f—
i— ^^
a- -a
0. C
re • re
^-^.
CU CV. C
x: o o
4-J LU *r-
> O)
>,S CO
XI (J> S-
^*^
U) CT1
c E c
O t3 *r~
•r~ ^ <™-
VI Cn CU
Ul O T5
•r- i- O
E CL E
CO
r— co
xo x:
O i- 4J
z: 4J
C 1-
"O O OJ
C 0 >
re o
cu
0 r- >^
31 U^-
*X3 -*rr t»
c cu o
re i> M-
«^-
CU S- C
r- 0 3
•r- 4->
xi o u)
§sr c
o
r— 'r—
cu re 4->
u s- o
3 cu 3
T3 "C1 TD
Q) rt) QJ
o: u. s-
•
LO
cn
>>
re
o
•
Ul
c
o
•r-
+J
•r—
-o
c
o
0
-o
cu
c
•f—
JO
E
0
o
a>
c~
4-^
>^
X2
VI
C
0
•^
(/)
t/)
"r-
E
O
a:
cu
o
S-
^3
O
VI
>5
S-
re
c
o
4*>
re
4->
VI
TD
C
re
cu
»—
•r-
f~i
0
E
ai
o
3
T3
CD
IX
1
O
V-
cx
V)
c
o
•p—
4J
•r—
•j^
C
0
o
r»
re
•r~
4-^
•r—
C
•r-
cu
o
3
"O •
CU LO
a: r^
CM
• TD
D- C
o re
LU
5> Lf)
E2
T3 -
c cn
re LO
p —
l—
O Ul
(
fV fQ
Q
t-
O
C- **^
o •—
4-> re
re c
o o
•1— •!—
r— 4-J
cx s.
cx o
re a.
•o *— ~
CO CU
c o
•r- C
xi re
E c
0 CO
CJ 4->
c
O) -r-
x: re
j > E
^^ "O •
X) C 10
re cn
Ul i —
c C
o o >>
•r- ••- re
Ul 4-> 0
V) U
•r- CO
E a. •
CO V) >i
c • —
O -i- r—
:r-— re
d
cu 2: o
i- *•— < 4-J
3 S_
O T3 O
V) C CX
re o
>> s-
S- CL. CX
re o
C LU U)
o s» c
•r-s: o
4-> U, •!-
re 4->
4-> «T-
VI >— -O
o c
T3 >SC 0
c D; o
re
cu o "re
•r- C 4->
Xi O -P-
O v- C
E 4J -r-
re
cu o cu
O-r- «J
3 r— 3
"O Q-"CJ
CU Q- CU
a: re a;
"CD C
«.
E J-
o re
u -o
c
CU 3
x: o
4-> JO
^}~O
x> c
re
Ul
C r— •
o re LO
•r™ »r~ r*^.
Ul 4-> CM
VI -r-
•i- C T>
E-p- c
cu re
cu
o o LO
:r 3 cn
T3 P—
CU CO
O C£ «
s- cn
3 LO
O • •—
V) D-
o vi
>>LU >)
i- > re
re s: o
c u.
o
•r* "O *
4-J C ^>
re re i —
u) (— re
O c
-o <: o
c cr v-
re 4J
<*- s-
cu o o
«— cx
•r- C O
/**) c? i-
O T- CX
re u)
CO 0 C
O T- O
T3 CX4->
CU CX-p-
Oi to TO
-0 4-
cu re
•p- c:
XI 3
E 0
O XI
o
T3
cu c
.c: re
p~
>> re
xi ••-
Ul •!-
c c
O T-
•p-
Ul CU
Ul O
•p- 3
E-0
QJ CU
a:
a:
cu s: LO
S- •— i i —
3
0 TD >>
vi c re
re o
i- CV.
re O •
C LJ >>
•i- 3E J—
4-> u. re
re c
4J •> O
Ul | — -p-
^ ^ 4J)
T3
• pM ^-^ f(
XI O C
E 3:
o — -
o
s: ;
CU "^-P-
x: i— i p-
4>^ r
"O £
>, C C
X) re *P
4
Ul. — !
C XJ
0 O !
V) !
VI X> 1
•p- C
E re i
cu
co a: •
s_
3 CX '
0 <_>
U) LU
^^ ^y
s- ti-
re
C "
•p- O
4"^ 31
re — -
U) 1—
o
"X2 "cC
c a;
re
14-
co o
P»«
•p~ a
X3 O
E 4->
re
CU U
o •<-
3 i —
-0 Q
-------
in
^.
OJ
ro
O
in
ro
o
c
o
•r-
4->
ro
>
LU
"D
C
(O
>}
CT) O*t
ro
CvJ •*•? ' —
rjj S- >,
•— -4-> ro
.£-) CO O
1— •—
O
i-
4-3
C
O
( )
in
•r-
^3
O
_J
4->
CO
en
c: i
ro
<5
&a
ro J3 CM
QJ CTQ-CO
D. O Q-OJ
OJ
CD
C 1
fx^
rt ^> Q_i—
OJ
en
ro
c™ I
O
5^
ro JQ OJ
Ol COO-i—
Q. O Q-cO
CT
QJ
+J
ro
&_
4_>
OO QJ
r— CO
0
+-> OJ
c. to
O ro
CO m
^S ^9 <5-°
co o co
r^ i — oo
co »^-
i i i
•* 0 O
r— \£> OJ
CM i— r—
S^ ^^ ^^ ^^
CTi O 00 in
O CO O
1— 1— r— CO
1 1 1 1
in > — o • —
in «* to CM
00
r-» oj in
• • •
00 r- 10
OJ «3"
i i i
in r^ r^
00 OJ 10
CM OJ r-
^~***
X ^^"^
O 0
ciS
•o
C D.
ro (_>
— * LU
r_j r_> >
5- 5- L?
O. O. +
0 O
r— LU LU H-
0 > > O
< 2: 2: - O
o; oo
<:
CD +
•z.
= s: 2:
o «^. -^
OQ 1— i I— 1
+ + +
**—-» *— ^» «^-s
000
S- 5- •55-
Q- D. Q.
0 O O
LU LU LU
S S ^
U- U_ Lu
+ + +
r— 1— (—
0 0 O
<: «c ^
en
0 £
e— CO
>- 05-
ce: c en
c
s: co
^^. O r ^
K— i CO
O>
+ QJ r—
I— O
X fO J^
O i — QJ
^^ -r— ^»
CO
•o > $-
c
>i O
0 i— £•
*-^ ro i—
C ro
D. O J-
O «/) OJ
LU CO -O
> QJ
«c «a: s
ce: o: u-
0)
o
c
CO
c:
O)
+->
c
•r—
CO
-o
C
ra
c
o
TJ
o
&
10
c
1— 1
3E
^.
I— 4
203
-------
TECHNICAL REPORT DATA
(Please read Instructions on the reverse before completing)
1. REPORT NO
EPA 450/4-83-019
4. TITLE AND SUBTITLE
2.
3 RECIPIENT'S ACCESSION NO.
The St. Louis Ozone Modeling Project
5. REPORT DATE
August 1983
6. PERFORMING ORGANIZATION CODE
7 AUTHOR(S)
8. PERFORMING ORGANIZATION REPORT NO.
Henry S. Cole, David E. Layland, Gerald K. Moss,
Conrad F. Newberry
9. PERFORMING ORGANIZATION NAME AND ADDRESS
U.S. Environmental Protection Agency
Office of Air Quality Planning and Standards
Monitoring and Data Analysis Division (MD-14)
Research Triangle Park. North Carolina 27711
10. PROGRAM ELEMENT NO.
A13A2A
11. CONTRACT/GRANT NO.
12. SPONSORING AGENCY NAME AND ADDRESS
13. TYPE OF REPORT AND PERIOD COVERED
14. SPONSORING AGENCY CODE
15. SUPPLEMENTARY NOTES
16. ABSTRACT
The results of applying a refined photochemical model to the St. Louis area
are described. The model is a three-dimensional grid model which incorporates a
generalized chemical kinetics mechanism. The report describes the performance of
the model using a variety of statistical and graphical techniques. Model simula-
tions for a set of 20 days during 1975 and 1976 are utilized in the analysis. The
report also examines the effect of changes in emissions on predicted ozone con-
centrations for a smaller subset of days.
17.
KEY WORDS AND DOCUMENT ANALYSIS
DESCRIPTORS
b.IDENTIFIERS/OPEN ENDED TERMS C. COSATI I'leld/Group
Air pollution
Atmospheric models
Photochemical reactions
Smog
Ozone
Nitrogen Oxides
Hydrocarbons
Urban Airshed Model
SAI Airshed Model
Carbon-Bond Mechanism
St. Louis
18. DISTRIBUTION STATEMENT
19. SECURITY CLASS (Tills Report!
Unclassified
21, NO. OF PAGES
207
Unlimited
20. SECURITY CLASS (Thispage)
Unclassified
22. PRICE
EPA Form 2220-1 (R«v. 4-77) PREVIOUS EDITION is OBSOLETE
-------
Brwsnmental
ion V, Library
South
'. Illinois
------- |