jO\
-------
EPA-450/4-83-020
Evaluation of Urban
Air Quality
Simulation Models
by
Richard Londergan, David Minott,
David Wackter, Roderick Fizz
TRC Environmental Consultants, Inc.
800 Connecticut Boulevard
East Hartford, CT 06108
Contract No. 68-02-3514
Prepared for
U.S. ENVIRONMENTAL PROTECTION AGENCY
Office of Air Quality Planning and Standards
Monitoring and Data Analysis Division
Research Triangle Park, NC 27711
July 1983
-------
PREFACE
TRC Environmental Cnsultants, Inc. has produced a set of model
performance statistics for urban Gaussian dispersion models. This work
has been perormed for the U.S. Environmental Protection Agency (EPA),
Office of Air Quality Planning and Standards (OAQPS), under EPA Con-
tract 68-02-3614, Work Assignment 13, "Evaluation of Urban Air Quality
Simulation Models."
DISCLAIMER
This report has been reviewed by the Office of Air Quality Planning
and Standards, U. S. Environmental Protection Agency, and approved for
publication as received from TRC, Environmental Consultants, Inc.
Approval does not signify that the contents necessarily reflect the views
and policies of the U. S. Environmental Protection Agency, nor does men-
tion of trade names or commercial products constitute endorsement or
recommendation for use. Copies of this report are available from the
National Technical Information Service, 5285 Port Royal Road, Springfield,
Virginia 22161.
-------
CONTENTS
Preface ii
Figures v
Tables vi
1. Introduction 1
2. Urban-Model Evaluation Data Base 4
Emissions and Source Data 4
Meteorological Data 10
Ambient S02 Data 11
3. Statistics Approach 14
Data Sets For Comparison of Observed and Predicted
Concentrations 14
Peak Concentrations 16
Comparisons of All Concentrations 18
Statistical Analysis of Model Performance 19
Statistical Evaluation Procedures 25
4. Description and Adaptation of the Urban Models 27
Distinguishing Features of the Urban Models . . 27
Plume Rise 30
Dispersion Coefficients 30
Stability Classification « 31
Meteorological Joint Frequency Function (STAR) for
Annual Models 31
Mixing Height 32
Wind Profile 32
Area Source Treatment 33
Pollutant Half-Life 34
Model Modifications and Options 34
CDM: Modifications and Options ............ 35
AQDM (Briggs Plume Rise Version): Modifications and
Options 36
ERTAQ: Modifications and Options 37
TCM: Modifications and Options 38
TEM-8A: Modifications and Options 39
RAM: Modifications and Options 40
5. Model Performance Results 42
Annual Average Models 42
Short Term Models 47
Unpaired Data Sets for 25 Highest Values 47
Paired Data Sets for Peak Values 54
Paired Data Sets for All Values 61
6. Conclusions 68
7. References 70
-111-
-------
APPENDICES
A Annual Average SOX Emissions Inventory for Point and Area Sources
B Annual Average Meteorological Joint Frequency Function from the 1976
RAPS/RAMS Data Base
C Highest and Second-Highest S02 Concentrations Observed and Predicted
(RAM and TEM-8A) in 1976 for the RAPS/RAMS Stations
D Hourly Meteorological and Observed Concentration Data for Selected
Days with High Modeled Concentrations
-IV-
-------
FIGURES
Number Page
2-1 Location of the Regional Air Monitoring System (RAMS) Stations
with SC>2 Monitors Indicated by Underlines 5
2-2 Locations of S02 point sources in the 1976 RAPS Inventory . . 6
2-3 Geographic Distribution of All RAPS Area Sources Including Those
in the Study Subregion ". . . . 8
2-4 Detail of the Distribution of RAPS Area Sources (1 km square) in
the Region of High Source Density 9
-v-
-------
TABLES
Number page
2-1 RAMS S02 Monitoring Stations and Corresponding Modeling
Receptor Number Used in the Study 12
2-2 Percentage of 1976 RAPS/RAMS Hourly S02 Monitoring Data
Accepted for Urban Model Evaluations 13
3-1 Summary of Data Sets for Urban Model Evaluation with RAPS Data
Base 15
3-2 Statistical Estimators and Basis for Confidence Limits on
Performance Measures ... 20
3-3 Performance Measures and Statistics for Unpaired (25 Highest)
Data Sets 23
3-4 Performance Measures and Statistics for Data Sets Paired in Time
and Location 24
4-1 Distinguishing Features of the Urban Models As Run for the
Current Evaluation ...... 28
4-2 Wind Profile Exponent by Stability 33
5-1 Urban Annual Averge Measured and Predicted SO, Concentrations
for St. Louis 1976 43
5-2 Comparison of Annual Average Observed and Predicted Concentration
Values Paired by Station 44
5-3 Comparison of 25 Highest Observed and Predicted S02
Concentration Values (Unpaired in Time or Location) for the
1, 3, and 24 Hour Averaging Periods 48
5-4 Comparison of 25 Highest Observed and Predicted S02
Concentration Values (Unpaired in Time or Location) for the
1 Hour Averaging Period 50
5-5 Comparison of 25 Highest Observed and Predicted S02
Concentration Values (Unpaired in Time or Location) for the
3 Hour Averaging Period 52
5-6 Comparison of 25 Highest Observed and Predicted SO2
Concentration Values (Unpaired in Time or Location) for the
24 Hour Averaging Period 53
-VI-
-------
5-7 Comparison of Maximum (Highest1 and Second Highest) Observed and
Predicted Concentration Values, Paired by Station for the
1 Hour Averaging Period 55
5-8 Comparison of Maximum (Highest and Second Highest) Observed and
Predicted Concentration Values, Paired by Station for the
3 Hour Averaging Period 57
5-9 Comparison of Maximum (Highest and Second Highest) Observed and
Predicted Concentration Values, Paired by Station for the
24 Hour Averaging Period 58
5-10 Observed and Predicted Highest Second-Highest Values 59
5-11 Comparison of Highest Observed and Predicted SO2 Concentration
Values Event-by-Event (Paired in Time) for the 1, 3, and 24
Hour Averaging Periods 60
5-12 Comparison of All Observed and Predicted Concentration Values,
Paired in Time and Location 62
5-13 Comparison of All Observed and Predicted 1 Hour Average
Concentration Values Paired in Time and Space for Specific
Data Subsets 63
5-14 Comparison of All Observed and Predicted 3 Hour Average
Concentration Values Paired in Time and Space for Specific
Data Subsets 65
5-15 Comparison of All Observed and Predicted 24 Hour Average
Concentration Values Paired in Time and Space for Specific
Data Subsets 66
5-16 Comparison of Daily Maximum Observed and Predicted 1 Hour
Concentrations (Highest by Day) 67
-vii-
-------
SECTION 1
INTRODUCTION
In March 1980 EPA published a notice in the Federal Register which
provided an opportunity for organizations to submit dispersion models for
possible inclusion in the next revision of EPA's "Guideline on Air Quality
Models." A large number of models were submitted in response to this
notice, including six in the "urban model" category (four annual average and
two short-term models). To decide in an objective manner which models should
be included in the guideline and what recommendations should be made
concerning the use of these dispersion models for regulatory applications, EPA
has undertaken a systematic evaluation of urban models. TRC, working under
contract to EPA, has assembled an air quality data base, set up and run the
dispersion models, and produced statistical comparisons of observed and
predicted air quality. These comparisons have been summarized in tabular form
and have been forwarded to the reviewers.
In September 1980 the American Meteorological Society (AMS) organized a
workshop to consider the issue of model performance evaluation. The 1980
workshop held at Woods Hole, Massachusetts, produced a report entitled
"Judging Air Quality Model Performance". This report contains recommended
statistical procedures for comparing observed air quality with model
predictions. The procedures recommended by the Woods Hole workshop provided
the basis for the statistical comparisons presented in this report. In 1982,
TRC performed a similar study for EPA to evaluate eight rural models. On
the basis of that study and subsequent comments by the AMS peer reviewers, TRC
-1-
-------
recommended a series of changes to simplify and streamline the statistical
calculations. These changes have been adopted for the urban model evaluations.
The air quality data base to which model predictions were compared was
acquired with a 13-station network of continuous S0_ monitors/ operated in
metropolitan St. Louis. The data were obtained from the EPA RAMS/RAPS
archive. Coincidental air quality and emissions data for calendar year 1976
were used in this study. Specific features of the data base are described in
Section 2.
In Section 3 the statistical approach is described. For the short-term
models, the set of observed and predicted concentration values has been sorted
in a variety of ways to provide statistical model performance comparisons that
reflect either high concentration values or all concentration values, with and
without pairing according to time and space. For the annual average models,
only one observation and prediction are available for each monitor. These
data sets are defined, and the specific statistical tests applied to each are
outlined.
In Section 4 the distinguishing features of the urban models are
summarized. Particular attention is devoted to describing the technical
differences among the models (as run for this study), how model options were
selected, and what modifications were required to obtain model predictions
appropriate for this evaluation.
Prior to running the urban models for evaluation with the RAPS data base,
it was desireable to confirm that the models would be run in accordance with
the expectations of the model developers. To accomplish this, a test-run
package was prepared by TRC and supplied to the model developers for their
formal review and concurrence. The package supplied to each model developer
contained the following information:
o Description of the urban-model evaluation data base;
—2 —
-------
o Summary of model-code modifications;
o Summary of input options;
o Test-run data (listings of all input and output data) for the
model developer's particular model;
o Complete listing of the model code "as run," to enable the model
developer to confirm the code line-by-line.
Comments on these documents from the model developers were addressed by TRC
prior to executing the final model runs for the statistical evaluations.
The results of the study are described in Section 5. The tables of
statistical comparisons for all six models, based on the performance measures
recommended by the AMS workshop, are presented in this section. Appendix D
provides tables of hour-by-hour model input and observed SO air quality for
each of 11 selected days when high SO,, concentrations were measured.
Finally, in Section 6, conclusions from the work assignment are
presented. These conclusions concern primarily the evaluation methods used in
the study and how these methods may influence the results.
Four appendices contain tabulated data. Appendix A contains the annual
average SO emissions inventory for all point,and area sources as modeled.
The annual average meteorological joint frequency function used as input to
the annual average models is listed in Appendix B. Tables of highest and
second highest observed and predicted concentrations for 1-, 3-, and 24-hour
periods at each station are presented in Appendix C. Appendix D provides
hourly meteorological and observed concentration data for selected days with
high predicted or observed concentrations.
-3-
-------
SECTION 2
URBAN MODEL EVALUATION DATA BASE
The data base used for urban model evaluations is a subset of the
extensive data archive established by the EPA for the Regional Air Pollution
Study (RAPS), ' a series of monitoring programs conducted between 1973 and
1978 for the St. Louis area. The RAPS data base was previously reviewed and
recommended as an appropriate data base for evaluating the urban models.
Data for calendar year 1976 were selected for the model evaluation data base
because the quality of the emissions data is better than for other years,
and because the 1976 data year provided all requisite model input. The
criteria pollutant sulfur dioxide (SO-), was selected for the model
evaluations. The SO emissions inventory of the RAPS data base represents
A
all area sources and point sources in both the Missouri and Illinois portions
of the St. Louis metropolitan area. The RAPS data base includes measurements
of meteorology and total sulfur or S0_ concentrations made at the 25
Regional Air Monitoring System (RAMS) stations operated in association with
the RAPS program. The map shown as Figure 2-1 depicts the RAMS station
locations and geographic extent of the RAPS study area.
EMISSIONS AND SOURCE DATA
The RAPS emissions inventory contains 480 point sources and 1989 area
sources in the St. Louis metropolitan area, including 235 point sources with
non-zero SO emissions. The general locations of these point sources are
X
shown in Figure 2-2. For ease of graphical presentation, multiple sources at
-4-
-------
Figure 2-1. Location of the regional air monitoring system (RAMS) stations with
SOp monitors indicated by underlines, (from "Documentation of the
Regional Air Pollution Study," December, 1979, EPA-600/4-79-076)4
-5-
-------
4330
' *13(122) '
4320 r
4310 -
4300 -
4290 -
4280 -
4270 -
4260 -
4250 -
4240 -
4230 -
4220
X 9(115)
XUK116)
LEGEND
S02 POINT SOURCES
• <50 (xlO3 kg/yr)
O 50-250
a >250
S02 MONITORS
X RECEPTORS (RAMS ID)
120 km.
710
720
730
740
750
760
770
780
790
300
810
Figure 2-2.
Locations of SC>2 point sources in the 1976 RAPS inventory,
(Multiple sources at the same facility, while modeled
separately, are shown combined for ease of graphical
presentation).
-6-
-------
the same facility were combined in this figure. A table of annual average
point source emissions and source characteristics is presented in Appendix A.
A map showing the distribution of RAPS area sources is presented in Figure
2-3. The RAPS study employed a fine-mesh source grid in the portion of
metropolitan St. Louis with highest emissions density. The grid for the
high-density region is detailed in Figure 2-4. For purposes of this study,
TRC reduced the number of area sources from 1989 to 1536 by excluding 453 RAPS
sources located more than 30 kilometers from the nearest S0_ monitoring
station. The rationale for this reduction is that the excluded RAPS SO
x
area sources are too distant to have any significant impact at the RAMS
monitors. (This assumption was confirmed by comparing results of CDM annual
average model runs based on 1536 area sources versus 1989 area sources.) A
table of annual average area source emissions is presented in Appendix A.
Hourly source data consist of SO emission rate, stack temperature and
A
volume flow rate for each of the point sources, and SO emission rate for
A
each of the area sources. For the long-term models, annual averages of these
variables were also available. The requisite fixed source data, including
geographic coordinates, stack diameter, stack height and area source width,
were also included in the model evaluation data base.
EPA compiled and made available to TRC a set of area source heights for
use with the RAPS emissions inventory. The area source heights range from 10
meters in rural areas to as high as 23 meters in the St. Louis downtown
areas. The heights were assigned based on land use patterns, starting from
3
values determined by Turner and Edmisten for an earlier St. Louis area
source inventory. This earlier inventory, however, did not encompass the full
region of the RAPS inventory. Area sources not in the original inventory were
later classified by EPA either as rural, suburban, or urban, with a height of
10 meters, 14 meters, or 18 meters, respectively.
-7-
-------
QJ
to
O
3
c
CO
0)
CJ
O
to
"3
cu
"3
1/1
Q.
fO
c •
O C
•i- O
•U T-
3 CD
-Q OJ
•r- i_
S- X2
4J 3
LO tO
•^
-o >,
TJ
O 3
•r- -M
-C to
Q.
(O 0)
S_ .C
ai+J
o
O) E
0)
3
cn
-8-
-------
4320-
km.
4310-
740
750
760
770 km.
Figure 2-4.
Detail of the distribution of RAPS area sources (1 km square)
in the region of high source density.
-9-
-------
METEOROLOGICAL DATA
A composite meteorological data set was made available by EPA for input to
the models. This data set contains hourly values of temperature, pressure,
wind speed and wind direction spatially averaged from the 25 RAMS stations.
Temperature and pressure in this data set were calculated as hourly arithmetic
means over the RAMS network. Hourly vector mean wind speeds (WS) and wind
directions (WD) were calculated from horizontal components of the wind vector
(u. and v.) at each station (i = 1 to N) as follows:
N N 1/2
WS= !_ u.2+ 1_ v,
N . , l N . ,
WD = Arc tan
1
1
N
U.
N
v.
i-1 ^
2
X
2
180
IT
If WS or WD differed by more than 4 m/sec or 75 degrees, respectively, from
the observed value at any given station, the data from the outlier station was
excluded and the vector wind components were reaveraged.
%
Wind measurements were taken from the 30 meter tower level at 17 of the 25
meteorological monitoring locations and from 10 meters at the other eight
locations. Most of the wind measurements made in the St. Louis urban area,
where the majority of the SO sources are located, were at 30 meters. A
height of 30 meters was therefore used for models requiring measurement height
as input.
9
Hourly values of stability were available based on the "Turner method"
using the composite wind speeds as well as cloud-cover observations from the
nearby National Weather Service station at Lambert Field.
Hourly mixing height values, calculated in the RAPS study, were also
available. The hourly values had been determined by interpolation from
-10-
-------
measured morning and maximum afternoon mixing heights using the CRSTER
preprocessor program. Measurements of the morning mixing height were based on
acoustic-sounder data. The afternoon mixing height was measured in the
RAMS/RAPS study with radiosondes. Where acoustic-sounder or radiosonde data
were missing/ the monthly mean mixing height value had been substituted.
Meteorological data for the annual models was available iri the form of a
6-category day/night STAR deck (Stability Classes A, B, c, D-day, D-night, and
E-F). A tabulation of this data is provided in Appendix B. Annual average
temperature and mixing height values were obtained from climatic records and
the standard Holzworth tables, respectively.
AMBIENT SO DATA
Hourly average SO concentrations were available for the 13 S02
monitoring stations in the RAPS/RAMS network. in order to allow a direct
comparison between the standard model predictions and the observations, TRC
converted these concentrations from parts per million to micrograms per cubic
meter using hourly pressure and temperature. Annual average SO
concentrations were calculated for each of the 13 stations. Figure 2-1 shows
the RAMS network, with the 13 S0_ monitor locations underlined. Table 2-1
gives the modeling receptor number corresponding to each SO,, monitoring
location.
For the urban model evaluation study, background levels of S02 a-re
assumed to be zero. The comprehensive regional emissions inventory minimizes
the likelihood of a significant background level. For many transport wind
directions, none of the RAPS stations is located upwind of the source region.
It is therefore extremely difficult to quantify whatever background there may
be with any confidence.
-11-
-------
TABLE 2-1
RAMS SO2 MONITORING STATIONS AND CORRESPONDING
MODELING RECEPTOR NUMBER USED IN THE STUDY
RAMS Station
101
103
104
105
106
108
113
114
115
116
120
121
122
Receptor Number
1
2
3
4
5
6
7
8
9
10
11
12
13
In reviewing the results of the performance evaluation, the reader should
oe aware of the criteria used for selection of data for the analysis. After
discussions with EPA personnel, acceptance criteria were established for the
hourly data that are based on the size of the instrument span drift and the
completeness of sampled data. Specifically, hourly SO- concentrations were
deemed acceptable if both of the following conditions were met:
(1) Span drift did not exceed 15 percent.
(2) The number of one-minute concentration samples making up the
hourly average value is 30 or greater.
Data recovery figures for the 13-station network are summarized in Table
2-2. The hours of SO concentrations, categorized as either missing,
excluded or accepted for analysis, are shown for the 13-station total and as
ranges across the individual stations. Approximately half of the missing
hours are attributed to the month of July. The reader should also be aware
that the operating range of the SO- monitors was such that no hourly
measurements exceeding 1 ppm were reported, i.e., values above this level are
"missing".
-12-
-------
TABLE 2-2
PERCENTAGE OF 1976 RAPS/RAMS HOURLY S02 MONITORING DATA
ACCEPTED FOR URBAN MODEL EVALUATIONS
13 Station Station-by-Station
Average (%) Range (%)
Missing Data 16 11-21
Data Excluded by acceptance criteria 13 10-17
Total Data Loss 29 23-34
Accepted Data 71 66-77
-13-
-------
SECTION 3
STATISTICS APPROACH
The 1980 AMS Woods Hole workshop on model performance evaluation
recommended a comprehensive list of performance measures and statistics for
evaluating air quality models. In addition, the workshop recommended
comparisons of the full set of observed-predicted data pairs, of the highest
observed and predicted concentration per event and of the highest N values
(unpaired in time or space), plus comparisons for subsets representing
individual monitoring stations or selected meteorological conditions.
TRC and EPA reviewed the workshop report and formulated a statistical
approach for the rural model evaluations based on workshop
recommendations. The approach was modified, following the rural model
evaluations, primarily to reduce the volume of information by eliminating
redundant performance measures and statistics. The statistical approach
followed for the urban evaluation is described below.
DATA SETS FOR COMPARISON OF OBSERVED AND PREDICTED CONCENTRATIONS
The data sets listed in Table 3-1 represent the different types of
comparisons recommended by the AMS workshop. In each instance, comparisons
were recommended for the basic 1-hour unit for model predictions and also for
3- and 24-hour averaging times. The numbering scheme in the table is derived
from a summary prepared by William Cox of EPA of the data sets and
statistics recommended by the AMS workshop. For annual average comparisons,
-14-
-------
rH
1
U
J
o
^
*
111
01
a
10
0.
•<
X
s-
s
z
o
I
u
J
i
z
a
at
D
o:
£
CO
g
u
W
<
s
a
o
as
tn
or "•*
3
-H
JJ
T3
0> C
0 ""*
*O 4)
0) u
QJ HJ •—
Q* 10
«-*
0) *-« O
(0 rH
?«s
833
^
rH
f
ca
g
ID
JJ 01
C C
4) 0
o a
Is
i a
rH O
t
CO
•a
0)
u
3 Q
«1 c
> 5 >
S •= *
.a I e
« c o
J= * O
J3 — 1
CJ I4H Qj
_ .^
^
1
<
C
o
jj
ID
u
JJ
C CO
a> e
o o
P "^
ID
•* a
ID g
o) o
4
41 in of
y jj 3 *u 4> IP
rH C OJ rH C S C
(D-rllfl ID ID ~H **H Q >l
> a > JJ O -H JJ
VulO 0)>HjQ JJ'uC
O ID • O ^ 01 O
*O Ool 'U'^HiD 3 TJ
4>«l*H4i 4) UlrHO'
u *Q JJ IH 13 '^ n] 4.'
O*04^"H Qi4)JJ W(,}34
IDQ«WC IDCLlD JJrHC
0* '" fl S ° "3
4ISWO OJOITJ 0
kJ Jj £i --H W Q (U ^~* JJ d
we o WjJ£. IE >^
0>O OJJ "^>^-iJ'o
•t-i c: > o] c jo •<-•
(fl (TJ 0) Qr^jjQ JJ .Q
t-U O ^ Qj(0(0'rj(Ucn'(l>
3kjiojj s u EOI^E
^- ^-* «
(N m •»
1 t 1
CD 33 C3
- (W
c *o •c *c o
CJ^ O C^0(0f fl(fcWC>030>rH JJtJ
3-^JSJJ 3Cn^«-l r-t WOIQ ij4J>rt) V C
^MjjiO ,_) c w n AI OJ 3 — i 0) D — « jJ « -^
>jJjJO > jJ <0 C 3 2l "J (0 -O tJ JJ 3fl •
•H fl rH JJclA-H^ T3-WQ 0>O O>"3 OlU'd>t
rrj C JJ TjOJflj 01 4) W ™4 0 •£ "^ClO
— * > 6 OJ flj -^ >JZ0 3Or-*4)
Q 0 ^ JZ**4 Q JS u g Z0)£<#t aO It -^ E
U Q g ^-, g u ^ g 3 a>!Ztu 0) 2 U4 0 ' (Q W >
0 ^ fl jj 0 x *^ > ^ "J -i C x ^ rtj fl pi -w a* -^ JJ tJO
c .^flJ-^C"*^ flflJ QJ rt -^tnoj -^OTOTO>Q) *S ui'o
_e>«»o-4J'— s>»'D';Jc»- S>» * jcotoi -cojincca *-wcc
h -w *0 fl> 4) C 3 C 41 "U W JCQ) JM(U-H -^ O tJ
01 1U(UCJJJ-)E 4)VjjV O O *"-<'3 uJ-:>c w-^'u W-^TJO*'* « i» jj — .
5 8£^S.Sg S2S.-HI.S IIII Hi SI 2 §3 loll
^. *^ -^ — » --* — ^
r* nj jo «j ^ in
i n m TT -w I
< I i I f
-------
the data set consists of one pair of observed and predicted values for each
monitor, directly analogous to set (A-2) for the peak concentration
comparisons.
For some hours during a year, none of the monitoring stations experienced
significant observed or predicted SO impact. These hours of effectively
zero observed and zero predicted impact are relatively uninteresting for the
evaluation of air quality models for regulatory purposes. Including those
hours in statistical analyses adds to the computational burden and tends to
dilute the model performance results from hours with significant impact.
Consequently, threshold values were imposed to screen the data base for
statistical analyses. If, for a given time period, both the observed
concentration and the predicted concentration at a station were below the
threshold, that data pair was excluded from further analysis. A threshold
value of 25 pg/m3 was used for 1- and 3-hour averages, and a value of
5 yg/m3 was used for 24-hour averages.
Peak Concentrations
For peak concentrations, comparisons are made to determine model
performance both on an unpaired basis and for various pairings in time and
space. The first two items in Table 3-1 represent a comparison of the highest
observed and highest predicted concentrations, paired in time (A-l) and paired
in location (A-2). For the RAPS data set, these two comparisons provide quite
different measures of performance since the number of events is large (1 year
represents 366 days or 8,784 hours) while there are only 13 stations. An
additional (A-2) data set was added for the urban evaluation, representing the
second-highest values observed and predicted at each station.
-16-
-------
Item A-3a represents a comparison of the highest observed concentration
values, regardless of time or space, and predicted values representing
different time and space pairing. Item A-3b is directly analogous to A-3a,
but starts from the highest predicted value. Results for data sets (A-3a) and
(A-3b) for the rural evaluation were relatively uninforrnative. These sets
were therefore dropped from the urban evaluations.
Items A-4 and A-5 involve comparisons of the "N" highest observed and
predicted values, unpaired in time or space. The AMS workshop recommended
that such comparisons be based on the upper 2 to 5 percent of concentrations,
rather than on one or two extreme values. As an alternative to the percentile
approach, TRC recommended using a small number (N=25) which would more
appropriately represent the set of highest observed and predicted values,
while still providing a statistical basis for establishing confidence limits.
On a percentage basis, 25 values represent roughly 7 percent of the 365
24-hour values in a year, about 1 percent of the 3-hour values, and about 0.3
percent of the 1-hour values. The statistical methods recommended by the AMS
workshop for these data sets assume that each data point is independent. This
assumption is not strictly valid, however, since the ranking process
introduces a dependence among the data values. The confidence intervals
calculated assuming independence will, therefore, tend to be too narrow.
Air quality data often exhibit spatial and temporal correlation,
particularly over time periods of a few hours. For 1- and 3-hour periods, the
highest 25 values were screened to eliminate cases with two or more high
values from the same period, or with two consecutive high values at the same
location. This screening is intended to reduce the effect of auto-correlation
and to avoid double-counting a single event. For non-overlapping 24-hour
averaging periods (midnight to midnight), less correlation is expected, and
this screening was not included.
-17-
-------
Comparisons of the highest 25 observed and predicted values were performed
for all stations combined (A-4a), for each station individually (A-4b) and for
subsets of events corresponding to time of day and to selected meteorological
conditions (A-5). For 1-hour periods, data subsets were established by
dividing the total data set into groups according to time of day or to the
model input wind speed or atmospheric stability class for each period. The
time of day subsets were not used for the rural evaluation but were added for
the urban evaluation. Hours of the day were divided into four groups: 0000
to 0600 hours; 0600 to 1200 hours; 1200 to 1800 hours; and 1800 to 2400
hours. Three wind speed groups were defined: low wind speed (less than 2.5
m/sec); moderate (2.5 to 5 m/sec); and high (greater than 5). Four
atmospheric stability groups were defined: unstable (class A or B); slightly
unstable (class C); neutral (class D); and stable (class E, F, or G).
Comparisons of All Concentrations
In addition to peak concentration analyses, the AMS workshop recommended
that comparisons be made based upon all observed and predicted concentration
values. Table 3-1 lists four items of this type. Item B-l is the comparison
of observed and predicted values at a given monitoring station (for all data
pairs above the threshold values). Item B-2, comparison of observed and
predicted values for a given time period, was recommended by the AMS workshop
but was not implemented for this study. With relatively few monitors and many
time periods, separate statistics for each time period are not practical.
Item B-3 represents comparisons based on the set of values from all 13
stations combined. Item B-4 represents subsets of B-3 to reflect time of day
and specific meteorological conditions. The same wind speed, atmospheric
stability, and time of day criteria described for item A-5 above were used to
define subsets for 1-hour averaging periods here.
-18-
-------
STATISTICAL ANALYSIS OF MODEL PERFORMANCE
The statistical measures employed in the rural model evaluation were based
on the 1980 AMS Woods Hole Workshop recommendations, as summarized in W. Cox's
letter of September 1981. In preparing for the urban evaluation, TRC
proposed a modified list of statistical measures and analyses. The basic set
of estimators used for comparisons of observed and predicted concentration
values are summarized in Table 3-2, together with the statistical methods
recommended for establishing confidence intervals.
For paired comparisons, the performance measures are based on an analysis
of residuals. Model bias is indicated by the average and/or the median
residual, with a value of zero representing no bias. The characteristic
magnitude of the residuals is an indicator of the scatter between observed and
predicted values on an event-by-event basis. Three measures of noise or
scatter were computed:
o Variance 1 (d. - d)
~FT~ . X
o Gross variability 1 d.
i
o Average absolute residual 1 Id I
i
where d. is the residual (observed minus predicted) for data pair i, d is
the average residual, and N is the number of data pairs. The correlation of
paired observed and predicted values is measured by the Pearson correlation
coefficient.
For unpaired comparisons, the list of performance measures is somewhat
shorter. Model bias is indicated by the difference between the average (or
median) observed value and the average (median) predicted value. A ratio of
-19-
-------
TABLE 3-2
STATISTICAL ESTIMATORS AND BASIS FOR CONFIDENCE LIMITS ON PERFORMANCE MEASURES
Performance
Measure
Basis for Confidence Interval
Estimator Paired Comparison
Unpaired Comparison
Bias
Average One sample "t," with
adjustment for serial
correlation
Median Wilcoxon matched pair
Two sample "t"
Mann-Whitney
Noise/Scatter
Variance
Gross
variability
Average
absolute
residual
Chi-squared test
on variance of
residuals
None
F test on variance
ratio
Not applicable
None
Not applicable
Correlation
Pearson
correlation
coefficient
Fisher "z1
Not applicable
Frequency
distribution
comparison
Maximum
difference
between
two
cumulative
distribution
functions
Not Applicable
Kolmogorov-Smirnov
(K-S) test on f(obs.)
vs. f(pred.)
-20-
-------
tne variances of the observed and predicted values is provided to indicate
whether the distribution of values in the two data sets is comparable.
Similarly, the frequency distribution of observed values is compared with that
for predicted values.
Standard statistical methods have been used to estimate confidence limits
for each of the performance measures. Discussion of the statistical
procedures may be found in most statistics textbooks. For parametric
procedures, the reader is referred to Snedecor and Cochran (1967), while
for nonparametric procedures Hollander and Wolfe (1973) provide an
appropriate description.
For paired comparisons, the confidence interval on the average residual
can be estimated using the one-sample t test. This parametric test
incorporates the assumption that the residuals follow a normal distribution,
but for large N, departures from normality are not critical when the number of
events is large. Serial correlation can affect results significantly,
however, since the number of "independent events* will be overestimated and
the calculated variance may understate the magnitude of the actual random
error component. The AMS workshop recommended the adjustment of confidence
limits for serial correlation. A method described by Hirtzel and Quon
14
(1981) has been used to adjust the confidence interval from the one-sample
t test. The interval given by the standard one-sample t test is multiplied by
the factor [(1+r)/(l-r)] , where r is the lag-one autocorrelation
coefficient of the residuals.
An analogous nonparametric indicator of model bias is the median
residual. The statistical method for estimating a confidence interval on the
median residual is provided by the Wilcoxon matched-pairs test. No
straightforward method of adjusting the confidence intervals from the Wilcoxon
test for serial correlation has been identified.
-21-
-------
A confidence interval for the variance of the residuals is calculated
using a chi-squared test. No adjustment was made for serial correlation. No
standard method is available for estimating confidence intervals for the gross
variaoility or average absolute deviation measures. For the Pearson
correlation coefficient, the Fisher z test provides a method of estimating the
confidence interval. «.
Comparison of two cumulative distribution functions is accomplished using
the Kolmogorov-Smirnov (K-S) test. For this test, the two distribution
functions are compared across the full range of concentration values, and the
maximum frequency difference between the two functions is identified.
For unpaired comparisons, two bias measures are computed. The average of
the observed values is compared with the average of the predicted values. The
confidence interval on the difference of the averages is estimated with a
two-sample t test. The difference of the medians is also computed, and the
confidence interval is estimated using the Mann-Whitney nonparametric test.
As noted previously, assumptions regarding data independence are not strictly
valid for the "highest 25 value" data sets.
The variance of observed values is compared with the variance of predicted
values for unpaired data sets. The performance measure is the ratio of the
variances; the F test provides confidence limits on the ratio. The frequency
distribution comparison for unpaired data sets provides a measure of the
difference between the observed and predicted distribution functions. The K-S
test is again used to assess the statistical significance of the maximum
frequency difference.
The specific performance measures and statistics calculated for each data
set are summarized in Tables 3-3 and 3-4. The notation for identifying
-22-
-------
CO
I
m
9
E-
w
E-
H
Cn
K£
^
Q
tn
u
EC
O
^_4
re
ID
CN
Q
Prl
M-l
PS
^
ff
a
c
a;
O
ft.
w
CJ
H
E*
en
M
§
E-
en
Q
z
<
MEASURES
H
u
§
o
&4
tf
w
rt.
l-M
•
•U C
W o
•S.2
u
* rt
cr a
<" e
u O
&. a
Q)
o
s °
(TJ -l-t
•H 4J
u m
to K
<4-l <
M-l
•H M-l
Q o
'O
0) 0)
O1 jj
03 o
^ -H
01 >o
> 0)
< s
Cu
Average
Observed
M
U
^ X X
-^
•
M
•
0
C N. -s.
*
H
cj
^. X X
«
,»—•,
•
H
•
u
N. % N.
X. *St X.
» » ^
-^ N. s.
•
4J
\ 01
M to to € to
C JJ \ -U C
0 C. C C >i 0
•H Ol 0 01 J3 -H
jj > ^. .H > ^ Jj
(OOJio JJOIJD tn-i-t^.
jj TJ< (0 n> JJ T3 in
W H | JJ r-l | 01 C |
Hrf** f/N J L« f« /*\ L(
fl, wJ *^ <; w u itf
^ to — . a — ^u--.
•H >i 3
rt ffl W
erval
jj
c
•rH
0>
O
c
01
T3
>r4
M-l
C
0
u
II
•
M
o
•K
-23-
-------
«q»
1
ro
W
J
§
E-
Z
O
l-l
E-
<;
g
a
ai
O
w
2
M
E-"
Z
i-(
Q
W
tf
M
*C
ft
cn
g
cn
<
E-1
c
U (0
0)
e
cn -H
JJ EH
C JC C
1) O O C —
> m •
0) 10 -H f*l
> TJ JJ 1
63 ci) 0) <0 CO
U 6 O ~~
.H -H -H c5
rH (0 E-t i-S
< ft
C
O >i +-
JJ -H JO C
cn jj o •^
0) (0 TJ -H «N
x: jj cu jj i
d cn u 10 "<
•H •'"I O ^*
DM U CO Q
0) ft J
ft
JJ C
JJ C "H
cn ID ---
0) > T3 01 i-l
jc: u
^_i
CU «3
ja LI
e
Z <
C
H
•
u
M
o
o
u
cu
g
cu
M
0)
M-l
u
O
O
10
u
(0
ft
0)
u
o
cn
jj
cn
u
cu
(0
M
(0
J3
g
73 cn 05
X X
•
X X
^ ^>
^ ^
X X
tn
jj
c
CU
ion Coeffici
son R
\
rman p
jj Wj (0
« 10 a;
rH CU ft
cu ft w
u
u
8
X
X
^-«
•
IH
CJ
^>
—
M
•
CJ
c
X
comparison
Variance
X
X
^•^
•
M
g
^v
^
l-l
•
g
s.
^^
•
•
g
*— *
%
Frequency
Maximum
jj
w
cu
jS
C7^
•H
jG
1
T3
C
0
o
cu
cn
cu
£
jj
U
O
M-l
•a
c
(0
cn
(0
•-I
(0
D
C
C
(0
u
0
M-l
T3
cu
•a
-H
0
W
ft
o
cn
a
<8
CU
u
cu
»
cn
o
•i-i
"2 -^
(0 cn
> -H
kj JJ
0) 10
JJ JJ
C M
•H
confidence
performance
11 cu
. CO
M _£
g ^
* 4-
er station.
values p
-24-
-------
data sets corresponds to that employed in Table 3-1. Table 3-3 indicates that
the full set of estimators and confidence interval calculations will be
provided for the 25 highest values over all stations and events (A-4a), but
only a partial set of measures is provided by station (A-4b) or for subsets by
time of day or meteorology (A-5).
For the paired data sets (Table 3-4), the highest priority is placed on
comparisons of the highest value per station (A-2) and all events paired in
time and location (B-3). The remaining data sets received a more limited
analysis. For the annual average data set, the estimators and confidence
intervals indicated for the (A-2) data set are provided.
STATISTICAL EVALUATION SYSTEM
The statistical evaluation system adapted for the model evaluations
consists of two components: a preprocessor to sort the "work files" of
observed and predicted hourly concentrations into data sets for statistical
analysis; and a statistical package to compute values and confidence intervals
for the performance measures. The work files, plus associated hourly
meteorological parameters, are sorted by the statistics preprocessor into a
number of data sets. The preprocessor computes block-average values,
beginning each day at midnight, for 3- and 24-hour periods, screens each pair
of measured and predicted concentrations according to threshold values, and
then constructs the individual files required to perform each type of
comparison listed in Table 3-1.
The statistical package then calculates the specific performance measures
listed in Table 3-2. The statistical computations were performed on the EPA
Univac 1110 computer, using the Statistical package for the Social Sciences
(SPSS). ' TRC constructed two basic SPSS runstreams, one to implement
-25-
-------
the paired comparisons, the other for unpaired comparisons, and applied each
as appropriate to the various data sets.
The SPSS output from the Wilcoxon matched-pairs test could not be used in
the form provided by SPSS, and this comparison has therefore been dropped from
the result tables. (The Wilcoxon results are generally redundant with the t
test, and were therefore judged to be dispensable in light of the considerable
effort required to recompute them separately.)
-26-
-------
SECTION 4
DESCRIPTION AND ADAPTATION OF THE URBAN MODELS
TRC has evaluated the performance of six Gaussian urban air quality models
using performance measures recommended by the American Meteorological
17 18
Society. Four are annual-average climatological models (AQDM, COM,
19 20 21
ERTAQ and TCM ) and two are hour-by-hour models (RAM and
22
TEM-8A ). AQDM, COM and RAM are EPA models; TCM and TEM-8A were developed
by the Texas Air Control Board; and ERTAQ was developed by Environmental
Research and Technology, Inc. (ERT). (ERTAQ, while primarily a long-term
model, does have a short-term mode. ERT, however, recommended that only the
long-term mode be evaluated.) The distinguishing features of the urban models
are summarized below, then the model input options and code modifications
required to run the models are documented.
DISTINGUISHING FEATURES OF THE URBAN MODELS
Distinguishing features of the urban models as run for the current
evaluation are listed in Table 4-1, and described briefly below. Particular
model options and run modes were specified by the model developers. It is not
the intent here to fully describe each of the urban models. In-depth
technical discussions of each model can be obtained from the individual
appropriate model-user guides. (Documentation for the current version of the
RAM model is contained in comment statements embedded in the computer code.)
The reader is encouraged to refer to the user's manuals for technical details
and references.
-27-
-------
TABLE 4-1
DISTINGUISHING FEATURES OF THE URBAN MODELS
AS RUN FOR THE CURRENT EVALUATION
AQDM
• Final plume rise
Five-category regular (A-E) STAR deck
Stability E is changed to stability D for az calculations
Mixing height varies with stability
Area sources modeled as virtual point sources
Rural vertical dispersion coefficients
Linear interpolation between 22.5° sectors for horizontal distribution
Total plume reflection at surface and mixing lid
No increase in wind speed with height
Uniform vertical mixing when downwind distance >_ 2X£,
where XL = distance where az = 0.47* mixing layer depth
• Input area source heights
CDM
Transitional plume rise
Six-category day/night STAR deck
Mixing height varies with stability
E stability class changed to class D for point sources
All stabilities (except A) are made one class less stable for area
sources
Rural vertical dispersion coefficients
Linear interpolation between 22.5° sectors for horizontal distribution
Wind speed increase with height is stability dependent
Four-hour pollutant half-life input for half-life option
Impact of area sources computed using sector integration
Total plume reflection at ground and mixing lid
Initial az of 30 meters for area sources
Initial az for point sources is 30 m for stacks <20 m; 0 for
stacks >50 m; and linearly interpolated in between
• Uniform vertical mixing when downwind distance _>_ 2XL,
where Xr = distance where CT_ = 0.47* mixinq layer depth
Ju £i
• Input area source heights
ERTAQ
Final plume rise
Five-category regular (A-E) STAR Deck
Mixing height varies with stability
45° triangular crosswind distribution
Stability classes reduced by one class for urban modeling
Initial az = 30 meters for area sources
Initial az for point sources is a function of stack height
Rural vertical dispersion coefficients
Perfect reflection at ground and mixing lid
Wind speed increase with height is stability dependent
Infinite half-life used for pollutant decay
Minimum allowable downwind distance = 10 meters
Area source impacts computed using rectangular increments in the upwind
direction
Input area source heights
-28-
-------
TABLE 4-1 (Continued)
DISTINGUISHING FEATURES OF THE URBAN MODELS
AS RUN FOR THE CURRENT EVALUATION
RAM
Final plume rise
Stack tip downwash
Buoyancy induced dispersion
Hourly meteorological and source data
Four-hour pollutant half-life input for half-life option
Wind speed increase with height is stability dependent
Input area source heights
Urban horizontal and vertical dispersion coefficients
Plume reflection at ground and mixing layer
Plume penetration when effective plume height > mixing depth
Minimum wind speed limited to 1.0 m/s for calculations
TCM
• Transitional plume rise
• Six-category day/night STAR deck
• Rural vertical dispersion coefficients
• E stability class changed to class D for point sources
• Stability class reduced by one for area sources
• Pollutant decay not used
• 22.5° sector averaging
• Area source contributions calculated only for a maximum of five basic
area source grids upwind of receptor
• Wind speed increase with height is stability dependent
• Perfect reflection at ground
• No treatment of mixing lid
• Area source emissions assumed from ground level
• 10 meters £ effective stack height <^ 2000 meters
TEM-8A
• Transitional plume rise
• Hourly meteorological and source data
• Stack tip downwash option not used
• Rural horizontal and vertical dispersion coefficients
• ofy corrected for averaging times other than 10 minutes
• Pollutant decay option not used
• Area source contributions calculated only for a maximum of five area
sources upwind of receptor
• Uniform vertical mixing when downwind distance >^ 2XL,
where XL = distance where oz = 0.47* mixing layer depth
• Plume penetration of mixing lid (L) for effective stack height >_ 2 *
L, when physical stack height < L
• Perfect plume reflection from ground but not from the mixing lid
• 10 meters £ effective stack height <_ 2000 meters
• Area source emissions assumed from ground level
• Wind speed increase with height is stability dependent
• No minimum wind speed for calculations
-29-
-------
Plume Rise
All of the models calculate an effective stack height for point source
emissions based on various Briggs plume rise formulations (see appropriate
user manuals for references). The transitional plume rise concept employed by
CDM, TCM, and TEM-8A uses the distance-dependent plume rise formulations.
AQDM, ERTAQ and RAM use final plume rise for calculating effective stack
height at all distances from the source.
TCM does not include a limit to vertical mixing. TEM--8A allows plume rise
through the top of the mixing layer only when the effective stack height is at
least twice the mixing layer depth (see sub-section on mixing height). RAM,
AQDM, ERTAQ, and CDM allow plume rise through the top of the mixing layer.
RAM also computes the effect of stack tip downwash on plume rise.
Dispersion Coefficients
The rural, Pasquill-Gifford vertical dispersion coefCicents are used by
all of the urban models except RAM which uses the urban, McElroy-Pooler
vertical dispersion coefficients. CDM and ERTAQ assume an initial a = 30
z
«
meters for area sources, and an initial a dependent on stack height for
z
point sources.
TCM, AQDM and CDM employ 22.5 crosswind sector averaging, with AQDM and
o
CDM using a linear interpolation between adjacent sectors. ERTAQ uses a 45
triangular crosswind distribution. RAM and TEM-8A use the urban,
McElroy-Pooler and the rural, Pasquill-Gifford horizontal dispersion
coefficients, respectively. The TEM-8A model enhances the horizontal
dispersion coefficients as a function of stability to account for the
dispersive effect resulting from atmospheric motions on time scales greater
than 10 minutes. RAM contains an algorithm to account for enhanced horizontal
and vertical dispersion resulting from buoyant plume rise.
-30-
-------
Stability Classification
All of the urban models classify atmospheric stability as follows: A is
extremely unstable, B is moderately unstable, C is slightly unstable, D is
neutral, E is slightly stable, and F is moderately stable. ERTAQ and AQDM use
5 stability categories (A, B, C, D, E-F) ; CDM and TCM expect the neutral
category, D, divided into day and night components, yielding six stability
categories (A, B, C, DD, DN, E-F). Both RAM and TEM-8A accept 7 stability
categories as input (A, B, C, D, E, F, G), where G represents extremely
stability; RAM then treats stability G as equivalent to F. Additionally,
TEM-8A internally splits the neutral category into daytime and nighttime
components.
To simulate the effects of enhanced turbulence in urban environments,
several models adjust input stability class to a less stable category. ERTAQ
reduces each input stability category by one, except A stability. In AQDM,
stability E-F is changed to stability D for a calculations. TCM and CDM
convert E and F stability to D for point source computations, and shift all
input stabilities except A to the next less stable category for area source
computations. TEM-8A (in the urban mode) treats stability classes E, P, and G
as class D.
Meteorological Joint Frequency Function (STAR) for Annual Models
Wind speed, wind direction, and stability class data are input to the
annual models AQDM, CDM, ERTAQ and TCM with the use of a joint frequency
function (also known as a stability array or STAR deck). The STAR data, based
on meteorological observations at Lambert field and at the 25 RAMS stations,
consists of the fractional frequencies of occurrence for each possible
combination of stability (5 or 6 categories), wind direction (16 categories),
and wind speed (6 categories). Stability categories for the annual models are
described above. TRC created a 5-stability category STAR deck for AQDM, and
-31-
-------
ERTAQ by combining the D-day and D-night categories from the original (6
category, day-night) STAR deck. Wind speed and direction categories were
identical for all of the long-term models, and are documented in Appendix B.
Mixing Height
All of the urban models, except TCM and TEM-8A, assume that a plume having
an effective release height less than the mixing height will be reflected by
the elevated stable layer. When the effective plume height exceeds the mixing
height, however, these same models assume full plume penetration of the
elevated stable layer, resulting in zero ground level concentrations.
Slightly different assumptions are made by the models TCM and TEM-8A. TCM
does not include any treatment of mixing height for either plume reflection or
plume penetration. TEM-8A uses an inversion penetration factor (I, set to I =
2). When the effective stack height exceeds twice the mixing height (L),
TEM-8A assumes that the plume escapes the mixed layer (i.e., ground level
concentrations are set to zero). Otherwise the effective stack height (with
an upper limit of L) is used in the dispersion calculations.
In the models TEM-8A, AQDM, COM and ERTAQ uniform vertical mixing is
assumed to result beyond twice the distance where a exceeds 0.47 times
z
the mixing height. With RAM, uniform mixing is assumed beyond the distance
where a exceeds 1.6 times the mixing height. Uniform vertical mixing is
z
not simulated in TCM.
Wind Profile
All the models except AQDM use a power law formulation to adjust wind
speed from measurement height to stack height. The wind profile exponents, as
used in this study, are shown below in Table 4-2.
-32-
-------
TABLE 4-2
WIND PROFILE EXPONENT BY STABILITY
Model
RAM
COM
TEM-8A
TCM
ERTAQ
Area Source
A
.15
.10
.10
.10
.10
Treatment
B
.15
.15
.15
.15
.15
C
.20
.20
.20
.20
.20
D
.25
.25
.25
.25
.25
E
.40
.30
.30
.30
.30
F
.60
.30
.30
.30
.30
TCM and TEM-8A use a method developed by Gifford and Hanna to calculate
area source contributions. These models require a rectangular grid of square
area sources, with the grid size equal to the side length of the smallest area
source. With these models, the simulation is limited to a maximum of five
area sources for calculation of impact on a given receptor for a given wind
direction. The five sources include the area source containing the receptor
and up to 4 upwind area sources. If the area sources in question are larger
than the basic grid size, fewer than 5 area sources may impact a receptor for
a given wind direction. Also, area sources are assumed to emit at ground
level in TCM and TEM-8A.
In ERTAQ, the contribution of each area source to each receptor is
calculated by integrating over the total area of the area source. All area
sources upwind of a receptor may have an impact on that receptor. Area source
heights can be input separately for each source.
AQDM simulates area sources through the use of virtual point sources. The
virtual emission point is located upwind from the area source such that the
width of a 22.5 angle originating at the virtual point and extending to the
-33-
-------
midpoint of the area source equals the width of the area source. Area sources
which do not fall entirely within a 22.5 sector upwind of the receptor are
reduced by a factor equal to the fraction of the area source contained within
the 22.5 sector. Area source heights are input by the user.
o
COM performs an angular integration over the 22.5 sector upwind of the
receptor in question to compute area source impact at the receptor. The
o
number of angular sections into which each 22.5 sector is divided for
integration, and the radial distance increment of integration are user
inputs. Values of 4 angular sections and an initial radial increment of 250
meters were chosen for the integrations. Area source heights are input by the
user.
RAM uses a narrow plume approximation to compute the concentration at a
receptor due to area sources. As run, the RAM model places each area source
into one of three area source height categories before performing the
integrations.
Pollutant Half-Life
RAM and COM were run using an exponential pollutant decay half-life of 4
hours. ERTAQ, TCM and TEM-8A used an infinite pollutant half-life. AQDM does
not allow for pollutant decay.
MODEL MODIFICATIONS AND OPTIONS
Certain modifications to the model codes were needed to carry out the
evaluations. Modifications were required specifically:
• To adapt some models to the EPA UNIVAC computer.
• To enable particular models to accept the large source inventory.
-34-
-------
• To format calculated concentrations for input to the statistics
system.
Prior to running the urban models for evaluation with the RAPS data base,
it was desireable to confirm that the models would be run in accordance with
the expectations of the model developers. As described in Section 1, a
model-specific test-run package was prepared and supplied to each model
developer for formal review and concurrence. Comments received from the model
developers were addressed by TRC prior to performing the final model runs for
the statistical evaluations. The modifications required for each model are
described below, and in addition, the user-supplied technical options selected
for each model by its developer are listed.
COM; Modifications and Options
a. Technical Modifications ..to...COM
EPA provided TRC with a version of COM modified to increase the number of
point and area sources which can be input to the model. This permitted
modeling of the 235 point sources and 1536 area sources in the RAPS data base
in a single run. TRC made several additional modifications to CDM. Code was
added to facilitate writing calculated concentrations to a work file for
subsequent statistical analysis. TRC modified the CDM program to replace
dimensioned variables with simple variables wnen used as exponents. This
change was necessary because the current EPA UNIVAC operating system does not
correctly compute an arrayed exponential when more than 65 K words of core are
required by the program. TRC also added statements to ensure that the model
would run when input stack temperature is less than ambient temperature.
b. CDM Input Options and Variables Description
• DELR = 250 meters Initial area source integration
increment.
-35-
-------
• DINT = 4
• SA = 0
• HT = 1400 meters
• HMIN = 400 meters
• TOA = 13.3°C
• SZA (1-6) = 30 meters
• YD = 1.05, YN = 0.97
• GB (1) = 4 hours
Number of intervals used to
integrate over 22.5° sector.
Briggs plume rise used.
Holzworth afternoon mixing height
for St. Louis.
Holzworth morning mixing height
for St. Louis.
Climatological mean ambient
temperature for St. Louis.
Initial az for each stability
class for area sources.
Ratios of average daytime and
nighttime emission rates to the
24-hour emission rate.
Pollutant decay half-life.
AQDM (Briggs Plume Rise Version); Modifications and Options
a> Technical ModificationstoAQDM
The version of AQDM was utilized that provides for use of Briggs plume
rise. TRC modified AQDM to write calculated concentrations to an annual work
file, to allow 13 non-grid receptors instead of 12, and to input data as
formatted READ, rather than NAMELIST format.
b. AQDM Input Options and Variables
• DPTHMX =1400 meters
TA = 286.5 K
PA = 1000 mb
Description
Holzworth average afternoon
mixing depth for St. Louis.
Climatological mean ambient
temperature for St. Louis.
Ambient pressure
default value.
model
-36-
-------
ERTAQ; Modifications and Options
a. Technical Modifications to ERTAQ
TRC altered ERTAQ in three areas. The model was adjusted to accept input
data from the model input file on Unit 18 rather than Unit 5. Statements were
added to facilitate writing calculated concentrations to an annual work file
for subsequent statistical analysis. Finally, the model input read statements
for source data were modified to accept more than 99 sources.
b. ERTAQ Input Options and Values
• A, B, C = default values
• XI, X2, X3 = default values
• EX = default
• ZQ = 30 meters
* XMIN = 10 meters (default)
• NCOMP = 5 (default)
• REGION = URBAN
• METHOD = 2
• WS = default
• DEPTH = 1400 meters
• TAMB = 286.5 K
• PAMB = 1000 mb
Description
Vertical dispersion coefficients.
Crossover distances for vertical
dispersion.
Exponents for wind profile.
Reference height for wind
profile.
Minimum allowable downwind
distance.
Maximum number of area source
subdivisions.
Dispersion option.
Triangular horizontal dispersion.
Wind speed for each class.
Mean mixing height.
Climatological mean ambient
temperature.
Ambient pressure for printout.
-37-
-------
TCM; Modifications and Options
a. Technical Modifications to TCM
The TCM model was modified to accept input data from a disk file, and to
write calculated concentrations to an annual work file for subsequent
statistical analysis. The TCM model assumes a fixed anemometer height of 10
meters, so code changes were made for the TCM model to assume an anemometer
height of 30 meters (consistent with Texas Air Control Board recommendations).
b. TCM Input Options and Variables Description
• LX = LY = 1 Number of rows and columns in
the receptor grid.
• NPRISE = 0 Transitional plume rise used.
• IURBAN = 1 Urban dispersion used.
• TA = 13.3°C Climatological mean ambient
temperature for St. Louis.
• ASCALE =1.0 Area source emission scaling
factor.
c. Othe_r TCM Technical Considerations
The specification of receptor locations in TCM is complicated by the
linkage between the area source grid and receptor grid. In order to specify
receptor locations exactly in TCM, only one receptor can be input for any
given model run. Therefore, TCM was run 13 separate times, once for each of
the 13 RAMS SO monitor locations. TCM calculates impacts only for area
sources located within four emission grid squares from the grid square in
which a receptor resides. Since the emissions grid width for the RAPS area
sources is one kilometer, only area sources within four to five kilometers of
each receptor were considered by TCM. Potentially important impacts from area
sources that exist beyond that distance would not be simulated.
-38-
-------
The TCM and TEM-8A user's guides also recommend that the area source
emissions grid be designed with the same spacing as the receptor grid, but
displaced such that receptors (at the receptor grid intersections) are located
at the center of the area-source grid squares. Restructuring of the 1,536
RAPS area sources with respect to each of the 13 monitoring stations in order
to accomodate this recommendation would be prohibitively costly. Following
discussions with EPA, the Texas Air Control Board agreed that use of the RAPS
inventory, as originally structured, will result in a useful performance
evaluation of the TCM and TEM-8A models.
TEM-8A| Modifications and Options
a. Technical Modifications to TEM-8A
TRC added code to TEM-8A to write calculated concentrations to hourly and
annual work files for subsequent statistical analysis. The code was altered
to allow input from disk file rather than cards. TRC also inserted logic to
read in hourly values of point and area source emissions, stack temperature,
and volume flow rate, and to convert these values into TEM-8A compatible
units. As with TCM, the TEM-8A model assumes an anemometer height of 10
meters. Code changes were made to the TEM-8A model to assume an anemometer
height of 30 meters.
b. TEM-8A Input Options and Variables Description
• NTOPT = 9 Hourly meteorological data on
tape; plume penetration
factor = 2.0.
• NSTDWN = 1 Stack-tip downwash algorithm
not used.
• LX = LY = 1 Numbers of rows and columns
in receptor grid.
-39-
-------
DTDZ = default Potential temperature gradient
for stable conditions.
ASCALE = 1.0 Area source emission scaling
factor.
IWIND = 1 Measured wind direction
entered to the nearest degree.
c. Other TEM-8A Technical Considerations^
Considerations regarding the specification of receptor locations and area
sources in TEM-8A are identical to those described previously for TCM.
RAM; Modifications and Options
a. Technical Modifications to RAM
TRC made modifications to RAM in five areas. The number of area sources
allowed in the model was increased so that the 1536 RAPS area sources could be
input in a single run. Statements were added to read in hourly source data,
and to convert source data into units compatible with RAM. Code was inserted
to compute and write calculated concentrations to an hourly work file for
subsequent statistical analysis. As with COM, the RAM model requires more
than 65 K words of computer core. Therefore/ use of arrayed variables as
exponents (a problem with the current EpA-Univac 1100 system) can lead to
computational errors. RAM was modified to circumvent this problem. Finally,
TRC changed RAM so that actual, rather than interpolated, hourly stack data
could be used in the calculation of plume rise.
b. RAM Input Options and Variables Description
• MUOR = 1 Urban mode.
• 2=0. meters Receptor height.
• IOPT (1) = 0 Include stack downwash.
-40-
-------
• IOPT (2) = 1
» IOPT (3) = 1
• HANE =30 meters
• HALF = 14,400 seconds
(4 hours)
• PL (1-6) = .15, .15, .20,
.25, .40, .60
« FH
0.75
• XLIM = 115 kilometers
• NHTS = 3
• HINT = 10, 15, or 20 meters
• PPH = 12 or 17 meters
No gradual plume rise.
Include buoyancy-induced
dispersion.
Anemometer height.
Pollutant half-life.
Wind profile exponents for
stabilities A-F.
Fraction of area source height
which is physical height.
Distance limit on integration
for area sources.
Number of heights to be used for
area sources.
Area source heights.
Breakpoint heights between area
source heights.
c. Other RAM Technical Considerations
The area source algorithm in the RAM model allows the definition of up to
three area source height categories to be used in the integrations. Following
discussions with Bruce Turner (EPA), the use of 10 meter, 15 meter, and 20
meter area source heights with breakpoints at 12 meters and 17 meters were
recommended as input variables to the RAM model.
-41-
-------
SECTION 5
MODEL PERFORMANCE RESULTS
Comparisons between observed and predicted concentrations have been
produced for four annual average models and two short-term models. The
performance measures and statistics calculated for each model are described in
Section 3. The model performance results are organized into a series of
tables. In this section, the results are presented and discussed.
ANNUAL AVERAGE MODELS
For the annual average models, the entire data set consists of one
observed and one predicted concentration value for each of the 13 RAPS
monitoring stations. These values are listed in Table 5-1. When observed and
predicted annual concentrations are compared, several differences are
evident. The highest measured annual value occurred at Station 104 and is
more than twice as large as the second-highest value. The highest predicted
values occurred at Station 101 for all four models. The highest predicted
value for each model is lower than the highest observed value. The lowest
predicted values occurred at the same two stations (120 and 122) for all of
the models, while the lowest observed value occurred at Station 116.
Performance measures and statistics for the annual average models are
presented in Table 5-2. The average difference between the observed and
predicted values for all stations is a measure of model bias, i.e., whether
the model systematically over- or underpredicts. ERTAQ gave the largest
overprediction (a negative difference means predicted is greater than
-42-
-------
TABLE 5-1
URBAN ANNUAL AVERAGE MEASURED AND PREDICTED S02 CONCENTRATIONS
FOR ST. LOUIS 1976 Ug/m3)
Station
1 (101)a
2 (103)
3 (104)
4 (105)
5 (106)
6 (108)
7 (113)
8 (114)
9 (115)
10 (116)
11 (120)
12 (121)
13 (122)
Average
Measured
55
36
116
43
52
37
36
35
28
24
27
30
27
42
AQDM
82.4
50.1
62.2
56.6
60.7
41.9
40.2
33.1
29.1
25.6
22.7
24.6
19.9
42.2
COM
83.4
51.2
62.4
45.5
45.7
42.2
32.9
31.6
43.9
23.0
15.3
18.1
14.3
39.2
TCM
102.0
43.8
52.6
34T4
39.6
41.7
27.1
29.4
46.6
24.0
12.4
17.1
15.0
37.4
ERTAQ
100.9
61.3
79.5
57.1
58.1
52.6
42.8
44.1
49.5
31.3
24.8
30.1
26.9
50.7
aRAPS/RAMS monitoring ID codes in parentheses.
-43-
-------
CN
I
a
CQ
o
LU
u
o
at o
ce 2
UJ W
co at
cc —
o <
0.
UJ
U CO
< UJ
CE 3
o. Z
C UJ
u
z z
C O
CO (_}
Q.
£
O
U
*: 3 i-i a.
UJ QD X U.
3 •— - 2. 01
x en o a
u. »-> u u,
o —
LU o a
u m uj
i; —i z
« at a.
ct a. in
< £ a:
>• o o
o —
a: z u.
« CE aj
UJ o O
a. u u
m
O » •
05 z U.
UC X UJ
< a o
Ui U U
a.
o: _i o
LU O -I
i» tn U. 0)
< •-« C
o
LU UJ
« a: 3
a: LU _i
uj :n *t
• in »in »LP »t
O • O • C • O
O)
ry
i in 03 ^ CT- 1*1 ro •
i 9 o ru
0000
0 a o
-< rvi nj
o IT!
-------
observed) and TCM the largest underprediction. The average difference,
however, is not significantly different from zero for any of the models, at a
95 percent confidence level. The fraction of positive residuals (stations
with observed value larger than predicted) ranged from 0.23 to 0.69. (When
this fraction is greatly different from 0.5, model bias is indicated.)
•
The magnitude of differences between observed and predicted annual-average
values at each station is characterized by three measures: the standard
deviation of residuals, root mean square error, and average absolute
residual. AQDM has the smallest values for all three measures, and TCM has
the largest. The confidence intervals on the standard deviation values
indicate that differences between the models are not significant at a 95
percent confidence level.
The Pearson and Spearman correlation coefficient values indicate that
correlation between observed and predicted values at the same station is
comparable for AQDM, CDM, and ERTAQ, but somewhat lower for TCM. Conversely,
the variance of the concentration values predicted by TCM is closer to the
observed variance (the ratio is closer to one) than the variances predicted by
the other three models. The confidence intervals indicate that none of the
variance ratios are significantly different from one. The frequency
difference comparisons indicate that the observed and predicted cumulative
distributions (of 13 values) differ by at most 20 to 40 percent (0.2 to 0.4).
In summary, the performance statistics for the annual average models
indicate some differences in performance among the models, but for this small
data set none of those differences are significant at a 95 percent confidence
level. All of the models underpredicted the highest annual average value, and
none predicted the highest value where it was was observed.
-45-
-------
Prom Table 5-1, the annual-average concentration observed at Station 104
is much larger than that observed at any other station. Also, the
concentrations predicted by the models at Station 101 are much larger than at
the other stations. In addition, from information presented in the next
section it is apparent that Stations 104 and 101 dominate the 25 highest
observed and predicted short-term concentrations, respectively. Model
performance in relation to Stations 101 and 104 has recently been investigated
23
by Ruff. While the conclusions from this investigation are not
definitive, Ruff does suggest that the aggregation of several small but
distinct sources as one area source could be responsible for the
overprediction at Station 101. For Station 104, Ruff believes that certain
emission sources may have been inadequately quantified or were neglected in
the RAPS inventory, leading to model underprediction. To the degree that the
RAPS emissions inventory is subject to such shortcomings, the model evaluation
results involving Stations 101 and 104 would be affected.
If Stations 101 and 104 were excluded from the analysis, many of the model
performance statistics presented here for both the annual and short-term
models would change significantly. For example, if Station 104 were excluded
from the data set, the annual-average observed value (Table 5-2) would
decrease from 42 to 36 yg/m , and the observed variance would decrease by
a factor of five. Only the non-parametric measures (fraction of positive
residuals, Spearman correlation, and frequency distribution comparison) would
not change substantially if Station 104 is removed.
The present study is concerned with the operational evaluation of urban
models, that is, as typically applied in the regulatory setting. From this
standpoint, the limitations of the comprehensive RAPS emissions inventory are
no greater (and likely fewer) than what one would encounter using any other
urban emissions inventory in model applications. The model evaluation results
presented here, therefore, are certainly representative in an operational
sense.
-46-
-------
SHORT TERM MODELS
For the two short term models, RAM and TEM-8A, a large number of
performance measures have been calculated for data sets representing selected
peak values and various data pairings for 1-, 3-, 24-hour averaging periods.
Results are presented first for the unpaired (25 highest values) data sets and
then for the paired data sets.
Unpaired Data Sets for 25 Highest Values
Table 5-3 summarizes results for the 25 highest observed and predicted
values, over all events and locations, for all three averaging times. For
TEM-8A, the average of the 25 highest predicted values is roughly twice as
large as the average of the 25 highest observed values for each averaging
time. For RAM, the average of the 25 highest predicted 1-hour values is lower
than the observed average by about 15 percent; for 3-hour values, the average
predicted by RAM is 40 percent lower than observed; and for 24-hour values RAM
underpredicted by a factor of 2. Statistics indicate that the difference
between the observed and predicted averages is non-zero at a 95 percent
confidence level. Results for the difference of medians are very similar to
those for the difference of averages.
The variance comparison results for TEM-8A indicate that the range of the
25 highest 1-hour and 3-hour values predicted by TEM-8A is much larger than
observed. For 24-hour values, however, the variance ratio for TEM-8A is not
significantly different from unity. Conversely, for RAM, the variance ratio
is not significantly different from unity for 1-hour and 3-hour values, but
the variance of predicted 24-hour values is much smaller than observed.
The frequency distribution comparison results indicate large differences
(0.76 to 1.0) between the observed and predicted cumulative distributions for
both models for all three averaging times. In general, there is little or no
-47-
-------
a
j
a
<
H
£ O
H U
x a
£2
>i O
u c
Eca>
3 0) u
g 3 01
~* en aj
x ai >w
(0 W -H
z u. a
11 O 'O
O W 0}
C -H U
fl Ij ft
re e a
=•
Difference
of Medians
(Obs-Pced)
(pg/m3)
a a ~
o ye
c a a) —
m Vj u f
u i \
LM < to u*
"u ^ =•
-^ >w o —'
Q 0 ~
•0
J) 41 —
m -u 4t^
« 0 = S
ij -^ M \
§"§ 5 2
cn >
us "
U
vfi
«-t
O
r* r*
o » ^1
• m • ,
O o o •
O
o
m
o »
• M
o o
o
o -
• r*
00 O
CM » CO
Mm m <
r-i cn i
I
Cft
yj »
0 0
(N V0
1 T
fV
1
1
-48-
-------
overlap between the distributions of the 25 highest observed and 25 highest
predicted values.
As discussed previously, the upper range limit of the instrumentation used
in the RAPS network to observe SO was 1 ppm (approximately 2600
. 3.
ug/m ). In computing hourly average SO concentrations for the RAPS
archive, therefore, EPA deleted any 1-minute SO concentrations in excess of
instrument range.
Comparisons of the 25 highest observed and predicted values by monitoring
station and for various data subsets reveal more detailed aspects of model
performance. Results of such comparisons for 1-hour average values are
presented in Table 5-4. For TEM-8A, comparisons by station indicate that the
model overpredicted the average of the 25 highest values by more than a factor
of two everywhere except at Station 104. At most stations, the variance of
the 25 highest TEM-8A-predicted values was also larger than the observed
variance. RAM overpredicted the average of the 25 highest values at 10 of the
13 stations; at 7 stations the predicted average was within 20 percent of the
observed average, while three stations showed disagreement by more than a
factor of two. Variance ratios for RAM showed large differences between the
observed and predicted range of the 25 highest values, even at several
stations where the average values were similar.
The results by station in Table 5-4 once again reveal the influence of
Station 104 on peak observed concentrations. The average of the 25 highest
observed values at Station 104 (1886 pg/m } is more than twice as large as
that at any other station, and obviously dominates the average of the 25
highest values over all stations (1929 ug/m , from Table 5-3). For
TEM-8A, the peak predicted values at Station 101 dominate the "all stations"
results in a similar fashion.
-49-
-------
a
E-> a.
So
E- Z
Z M
M U
U <
z a
o w
u >
4
04
O «
Cfl 3
ai
a
s:
Q X
B^ 5
£ g 5
Q 2 "
i en
z 2
00 §
a« **
> E-
3 y
en o
m ,J
ga
£°
S H
CJ h- 1
w 5-i
E
Z
IT) 1-4
o o
U —
a —
O O1 41 *^
C (0 um
41 w ft* fi
U 41 1 X.
41 > ffl O1
« < a a.
M O —
a o ~
ftj 4) «••>
S" O 3 S
U -rt ^H \
4) T3 flj CH
> 4> > a
<£ -
"U
fl)
o- > J3 > =-
«C O —
C —
01 0 03
U ffl 0)
C -^ Li
(QUO,
•H fl \
W Qj U)
fl S J3
>S£
to
O QJ -—>
4) w w S
u oi a \
41 > ) O-
u-i < m 3.
S w O ^
Q O —
«a
w JJ 4)f*T*
it O 3 S
> 4) > 3.
•o
0) 4) —
01 > a»f*>
«) u 3 S
4) (fl (0 CT1
CO
AJ
4>
W
(0
tt
a
OvoiN^o -trDin^tN >o ^ - 4
O^>-HO^ i^inOO^ ^*O3N
-H -1 -H .H
r-co^roco chinncocn eo^-^
-Hcomvo^f o>f-oOin fN-HOfH
W — . « -» ^. -* «^«-^\0 O^ffN
U O M CN f">
•• 4J
cm ccccc ccccc ccc
O4) OOOQO QOOOo OcO
£= wwSSw SSSww Sww
> CN
ffl —
O -4 Cn in
O —t -H CN
<*> o
O O
o o
*>• o
IN 33
IN |
a
3
O
o
0 4)
o o o o
O o <= o
vfl r*l CO rf
O -I -H cs
I I I I
O o O O
O o O O
O V0 fN CO
00-1 —
< U Q
ifl m to
41
•8
u
-50-
-------
Results for subsets by time of day show a striking difference between
observed and predicted behavior. Both models predicted higher peak values at
night and lower peak values during the day/ while peak observed values were
very similar for all four time intervals.
Results for subsets by meteorological conditions show distinct differences
in performance for the two models. TEM-8A overpredicted the 25 highest values
for low wind speed conditions by more than a factor of two, but predicted the
average of the 25 highest values for intermediate and high wind speeds within
10 percent. By contrast, RAM predicted the 25 highest values for low wind
speeds within 10 percent, but underpredicted for highest wind speeds by about
40 percent. TEM-8A predicted the average of the 25 highest values for Classes
A & B and Class C within 20 percent, but overpredicted for Class D by 60
percent and for Class E & F by more than a factor of two. RAM underpredicted
the 25 highest values for Classes A & B (by 30 percent), Class C (by 60
percent) and Class D (by 55 percent), but predicted the average of the 25
highest values for Class E & F within 10 percent.
Station-by-station results for the 25 highest observed and predicted
values for 3-hour and 24-hour averaging periods are presented in Tables 5-5
and 5-6. These tables reveal a pattern very similar to the 1-hour results.
TEM-8A overpredicted the 25 highest values at all stations except Station 104,
generally by more than a factor of two. By contrast, RAM predicted the
average of the 25 highest values within 10 to 20 precent at roughly half of
the stations for each averaging time. RAM underpredicted the peak 3- and
24-hour values at Station 104 by more than a factor of two and overpredicted
consistently at Stations 101, 105, and 115.
-51-
-------
IT!
TABLE 5-
RATION VALUES
G PERIOD
DICTED SO2 CONCENT
HE 3 HOUR AVERAGIN
§H 2
BSERVED AND P
LOCATION) FOR
RAPS 19"
°«
6» 0
S iii
gs
M El
X
Z
in M
fN
a
a. u
0 «
COMPARISON
(UN PA
IX
g
i
c —
41 O -D
o a 41
« u a,
-| 0 0
O ^~
ffi ~*
41 41 13
U Cfl 41 ^
C "3 w t*l
41 u & &
u 41 1 X
41 > to W
« < a a
« ^ £. ~
a o
•o
11 01 -.
p> *J 4) r*l
« o a S
u •-< -H X
41 -o ia 01
> 41 > =l
< £ -
•a
41 41 _
CT > 01 ^
u 41 .H X
41 CO fl O^
> a > a
<: o —
c ~
oi o •a
0 M 41
« w a,
s 11
>62
01
0) 0) ~
U O1 ^ -^.
a; u i- E
0) > I O>
«u < to a
•w J3 —
-H "W O
0 0 —
•o
at oj «
^i jj ai m
U -H r-4 X
(U fl (0 CT1
2 £ > -
ft-
•O
(U 41 -v
CT- > 0)f^
* !S a s
u w iH N,
tt) 3 m 01
55=*^
CO
JJ
41
W
a
^j
S
r--^^ONO «-Hom-^ ff\iN«
GO^OOiN O~HOOCN ON3-*
-H
r-ro-Hoco o% IH n «H i ^D^-tN
•H 1 P" (N | | -H ^- l
II II
^fS'nmeo or-'Oanco -nmtN
(Nr^crsr»<*i n^oo^'V r-m^
u")<*io^fN nmp^cN^1 mo^
^mrOcN'v oir«J(Nr*j-*mco oovooOi-*r*> irt^otn
(N *-) pH
mmOvofN fiinr-.fj^' r^om
^•nf^ra^T O4 (N
03 —
parentheses.
c
00
0)
•§
u
c
M
tr
c
•«4
u
0
C
0
ra
X
0*
<
a:
n
-52-
-------
to
in
jj
j
a
M
u
<
I
0)
01
•H
a
o)
Ol
a
0)
>
<
u
g
(0
u
10
a
01
&
^
01
>
<
0
1
u
•rt
TJ
01
>
41
O
01
U
a
u
fl
a
1
M
41
>
O
•o
01
o
1
t3
0)
>
1
0
c
o
01
8
~
41
&
U
a
0
3
— t
1
>
3
>
C
O
a
10
o
I
04
1
to
a
0
3
3
a
>
"U
4)
w
\
01
43
O
—
g
CT
-*
S
ex
•n
E
S
TJ
01
£
:£
'•'
CO
4J
0>
01
o
Q
o
£££SS SSS^m S * S
o * r^ o a\ O -H o o '-o r^^-N
r»m^rcofN (N-HI-H (N.H-N
-( 1 W | | _<
l i
«,.-,-.-, ^0,0,0 »»r.
OCftWiH^1 i-Nr^fSOiTI 'H.H'W
0 -1 T 0 *H OOOOO OOO
^•ino*cof*i o vo tA N o cof^^*
r*-cftri*^r* f^w^jir^fi coeooo
—( -^ ,H r«g ^H •Hi^ro.,.^ l|l
-(I It 1 III
1
ertr-ifNcn^1 -o-o^-oo CNP*-^
S"1*"""*' M^^"^ ^-"^
fn^H*c^nr* o^ocoen cocri^
"c
O
^
4J
*
01 . G 0^01
U o ^
coi ccccc cccce s = g
OO) OOOOO OOoOo OoO
4 (QiQlQiQIO Ot (M
o —
01
01
a
9
c
41
a
a
e
a
4>
U
a
H*
C
W
0
c
0
E
tn
>
Ot
X
-53-
-------
Paired Data Sets for Peak Values
Comparisons of the highest observed and predicted concentration values,
paired in location or time, utilize the same measures and statistics described
earlier for annual average model results. Table 5-7 presents the results for
the highest observed and predicted 1-hour concentration values at each
station, plus similar comparisons of the second-highest values. For TEM-8A,
the average difference between observed and predicted highest values is
negative, indicating overprediction, and the magnitude of this difference is
larger than the average observed value. Results for second-highest values for
TEM-8A also show large overprediction. For RAM, the average difference
between highest observed and predicted values is small, relative to the
average observed value; the average difference for second-highest values is
even smaller. Confidence intervals indicate that the overprediction by TEM-8A
is significant at a 95 percent confidence level, while the average differences
for RAM do not represent statistically-significant bias. The results for the
fraction of positive residuals are consistent with the average differences.
TEM-8A overpredicted the highest values at 11 of the 13 stations, and
overpredicted the second-highest values at all 13 stations. RAM
underpredicted the highest and second-highest values at more than half of the
stations.
The standard deviation of residuals, root mean square error, and average
absolute residual for TEM-8A are all larger than the average observed value.
All of these measures are smaller for RAM, indicating less scatter between
observed and predicted values. Correlation coefficient values for both
models, however, are lower than 0.3 in every instance. For RAM, the
correlation between highest observed and predicted values station-by-station
is negative. The ratio of observed to predicted variance of concentration
values for TEM-8A was significantly less than unity, reflecting the larger
-54-
-------
r-
i
m
M
3
S
03
a
j
z
o
M
|
S
M
3C
U
Q
Ps
M Sd
a a
a a.
2
Q M
Z t3
Q aj
Ed >
U t£
CO 3
o O —
OS -o
— ,-{ S
Ed S
I H U3
gs S
= 2 2
Q
Z Z
0 O
01 S
O 01
%»
to
ui g
is
X 0<
~
M
X
2
o
g
£
B.
g
U
t
c
at 0 -*
u uj -o
C H O -0
o o
c
B * *
w u *S
S 8 8
W
C
o • •
W \4 W
o<
0) rH
01 JJ «J >^
O» 3 3<*>
(0 -H T) £
U O -H V.
O
0« £ O* Ed 3-
Cfl -
*0)
u O n] -^
*8 '*^ 3^
(Q -H to Q*
i> > a> 3-
to o> os —
Q
o
*o
01 OJ
c a> ^ u.
O > ITJ a-
-H -*4 9 A
y o -^ -4 M
3 6
j_J U J5 'X.
4J H m ^
* o ^
0) U
J3 « u
E ** JJ -<
3 0 (0 fl
Z O 04
-o
£
Cft ^ iTV "4
in co r- p-
COOCOCN nO^-^
— i co (N *r
O1* OP*- or*- -H^
O (N O T
00 0 O
co ** r< r-t
O (N O O
0 O 00
1
ID O V O
o m 04 -*
o o o o
1
o\ i*n ^o vo
•-t « ^ -4
r- co ^ in
rH -H
mm vo c^
eft a\ *-i in
rg cri co \o
CN ,-1
0 O O ff»
f*l fl ff\ fH
\O (*- CO .H
(N -H ^H ,H
M 00 ^ P»
c^*^*. ^ » r* •»
m CM O i-( rHr^VOXO
iH ^ *-4 IT —< PJ CO
-H r- co ^i
<«.
in ^ o M
(-H VO O *fi
o o a o
co ^^ in --»
^- r- in *•>
r* r^ r* r-
1 CO t 1*1
rH m \£» -*
,H*CO* ^ *. ro »
r- ^ f-t ,-* -v r* I .H
i— ( ^H (H »H
JJ
(fl
0)
CT
.r^
X
4J
tn M <
x e- a: co H a
— * -C7
"•i —
to
o
d
V
Je
ij
c
u
O
c
0)
*o
«w
C
0
0
Jj
c
(U
u
u
a>
a
m
en
-55-
-------
magnitude and range of predicted values; variance ratios for RAM were not
significantly different from unity.
Results for the highest and second-highest values paired by station for 3-
and 24-hour averaging periods are presented in Tables 5-8 and 5-9. In
general, the results for average difference show model bias similar to that
for 1-hour values. TEM-8A overpredicts for both 3- and 24-hour averages; the
average differences are as large or larger than the average observed values.
RAM underpredicts by less than 15 percent for 3-hour average values and by
about 25 percent for 24-hour averages. Measures of scatter between observed
and predicted values are larger than the average observed value for TEM-8A,
but are much smaller for RAM.
Correlation coefficients for 3-hour values are noticeably higher than for
1.0-hour values, indicating somewhat better success at predicting how peak
values vary from station to station; but correlation coeffficents for 24-hour
values are lower than for 3-hour values.
The variance comparison results for TEM-8A continue to show ratios less
than 1, but for 24-hour values the confidence intervals include unity. For
RAM, the variance ratios increase with averaging time; for 24-hour values, the
predicted variances are significantly smaller than observed.
The highest and second-highest values observed and predicted at each
station for each averaging time are listed in Appendix C. These values
represent the basis for the measures and statistics presented in Tables 5-7,
5-8, and 5-9. Appendix D provides tables of hourly meteorological and
observed concentration data for selected days with high modeled concentrations.
Current air quality standards are based on the highest, second-highest
SO concentration value at any station for 3-hour and 24-hour periods. For
the reader's information, Table 5-10 lists the observed and predicted values
corresponding to the air quality standards. These single-value comparisons
-56-
-------
CO
1
in
H
i
H
u
2
•C
Z
0
E-i
£
&]
g
O
u
Is
as
0 H
& ^
0* 3
O *-*
2
s >
3 «
S «
w =
a o —
/*i '"*
f* "••«
us ~
s fc» in
E O (S
U-
a
o o
o *•*
£
O (Q
** S
" i
M •«
S *
£
3
£
S
&•
0
§
tfl
1
U
c
<]> O ••"•
O 01 T3
C -H OJ
It) M W
•^ rtj fl*
u CX \
ffl E w
> O Q
0 E
C
8 w «
S 8 8
U3
S* *
U U
w w (U
S8 S
V <-*
w i> j3 Q)
a) uf*r
*J C w O S
0 4) 3 u CT1
K 2 CT W =L
*
"2 o^-
HJ -^1 3 f*l
fl ^ 0) 01
OJ > If 3-
M o a —
o
•a
C 0) rH U
O > us 04
««-4 -^* 3 A
O O -« -H 0)
u 0 91 O
t a. BJ —
4) —
Si C 'S m~
2 2 £ -^
> iw m 3-
O —
S* > 3 E
S «> ID 0>
* 5 > —
o ~
"
g iy jj .^
3 O fl *0
Z Q 04
•-H
I
i
O 3
r-4
0
o
1
o
o
in
o
00
en
at
S
f-(
CO
in
CM
cn ao
I
0
o
o
cs
eo •.
ct m
in
I
in
r-
m
r-(
til
OH GO
•S l
d s
=i &
o r*
-------
in
ta
j
m
§
D
J
>
Z
0
s
i
M
g
O
o
s s
1s
Sg
Q g
g>
1*
si ~
O "" "-o
- N «
E"1 »-i
§9* w
^ o.
•M Q OS
o ***
0 §
r/J ^*
i »
* X
i§
U
£
X
i.
O
Z
O
en
i
&
e
o> o -»
o w -a
c ->-t a;
•H (Q a*
is g en
> O -0
o 2,
c
10 . .
u y *!>
35 8
en
c
0 • •
tfl w w
M w 0)
S88
CU
0 rH
O =•
u
0
^
tJ
en a)
2 .2 3 S"
•W »M -U -O
0 0 "•* -* 01
(0 10 0] A
,n P 5 O
b, ucC'—
*
0) ^*
5) O fQ *-,
n 11 u e
U W 04 \
< |M 5 —
ni fll o>M
Js > 3 e
U u rH \
^H IT) (Nl vA i— (
ro r- ^*
o ^ >* >^ o «•
p-» O M
o f o
(N 03 r-
^a- m n
o o o
VO ^ (N
O f*l —4
O O O
O O o •»
n ot f*i
*~* *••* *••*
00
LI
«
jj
e
01
o
c
01
T3
•r4
<4-l
C
o
o
JJ
c
1>
o
M
V
a.
in
(T»
«
-58-
-------
provide no basis for statistical measures, however, and the AMS workshop
recommendations do not include evaluations based on a data set comprised of
highest second-high values.
TABLE 5-10
OBSERVED AND PREDICTED HIGHEST SECOND-HIGHEST VALUES
_ 3-Hour Average _ 24-Hour Average _
Observed 1609 1170
RAM 1127 424
TEM-8A 4108 1852
Another paired data set consists of the highest observed and predicted
values over the monitoring network from each sampling period, paired in time.
Results for all three averaging periods are presented in Table 5-11. While
the data sets discussed up to this point contained relatively few points,
event-by-event comparison for a year of data involve much larger volumes of
data (i.e., a large "N").
Results for 1-hour average values show negative average differences for
both models, indicating overprediction. Both values are significantly
different from zero (the large N leads to a narrow confidence interval), but
the average difference for TEM-8A is more than twice the average observed
value, while for RAM it is only about 20 percent of the average observed
value. (Because pairs with both observed and predicted values below 25
were excluded from analysis, the number of events is slightly
different for the two models.) The standard deviation of residuals for both
models is larger than the average observed value; the standard deviation for
TEM-8A is much larger than for RAM. The maximum difference between the
observed and predicted cumulative frequency distributions is also much larger
for TEM-8A.
-59-
-------
TABLE 5-11
COMPARISON OF HIGHEST OBSERVED AND PREDICTED S02 CONCENTRATON VALUES
EVENT-BY-EVENT (PAIRED IN TIME)
FOR THE 1, 3, AND 24 HOUR AVERAGING PERIODS
RAPS (1976)
Numbe r
of
Model Events
Averaging Time:
1 Hour
TEM-8 7891
RAM 7769
Averaging Time:
3 Hours
TEM-8 2475
RAM 2431
Averaging Time:
24 Hours
TEM-8 339
RAM 339
Average Average
Observed Difference*
Value (Obs-Pred)
( ug/m3 ) ( yg/m3 )
165 -354
(-382, -326)
167 -35
(-49, -21)
142 -320
(-350, -290)
145 -20
(-36, -4)
130 -317
(-366, -268)
130 -5
(-34, 24)
Standard
Deviation
of Residuals*
(pg/m3)
518
(510, 527)
290
(285, 295)
455
(443, 468)
225
(218, 231)
290
(270, 314)
165
(153, 178)
Maximum
Frequency
Difference
0.52
(0.022)
0.16
(0.022)
0.55
(0.039)
0.17
(0.039)
0.74
(0.104)
0.35
(0.104)
*95 percent confidence interval in parentheses.
-60-
-------
Results for 3- and 24-hour average highest values paired in time are
generally very similar. Both models overpredict on average, TEM-8A by a much
larger degree than RAM. The standard deviation of residuals and the maximum
frequency difference are also larger for TEM-8A. For 24-hour values, the
average difference for RAM is not significantly different from zero at a 95
percent confidence level. «.
Paired Data Sets for All Values
The largest data sets considered in this evaluation represent all
concentration values, paired in time and location. Results for these data
sets, for all three averaging periods, are presented in Table 5-12. The size
of the data sets for 1-hour values were so large that non-parametric
statistics could not be calculated, due to computer work-space limitations.
Results for the average difference between observed and predicted values
indicate overprediction by TEM-8A (by a factor of 3) and by RAM (by 20 to 25
percent). The standard deviation, root mean square error, and average
absolute residual values are generally larger than the average observed value
for both models, but are substantially larger for TEM-8A than for RAM.
Correlation between observed and predicted values increased with averaging
time, but with little difference between the models. Variance ratios for
TEM-8A were consistently less than 0.2, indicating a predicted variance five
times larger than observed, while ratios for RAM were significantly greater
than one for both 3-hour and 24-hour values. Maximum frequency differences
are also larger for TEM-8A than for RAM.
Comparisons of all observed and predicted values were also made by station
and for subsets of events based on time of day and meteorological conditions.
Results for 1-hour values are presented in Table 5-13. For TEM-8A, the
average difference is negative at every station and for every data subset,
-61-
-------
(M
•-4
1
to
TABLE
.
z
O
&*
J
a
5
u
~
z
M
a
S
j
04
1
z
O
£* ^»
1 5
u ._.
a «
8 S
a
U
s
a
3
a.
a
z
o
u
CO
to
(O
o
J
o
z
o
tn
S
s
>| 0)
3 C C
E 01 0)
••H 3 ij
x a1 O 13
O 01 01
e -p4 w
(5 WJ Ckl
•«* (0 "^
u 0« 01
* - S
~
C
IB • •
u u Ct)
BOO
Q*
CO
C
S* *
U U4
U M CJ
288
a.
01 H
41 4J (0 ^^
o> 3 sen
10 PH TS S
u 0 -^ • — ,
*5"5
_
0) u m
•U C u 0 E
,§ S 5-U ^
01
•O c 10
£ O 3^
•a ij •* s
a £ * ^»
** > a 5
Q (U
0
01 T3
C 01 r-l CU
O > <0 w
•In -i 3 &
a ° ~m~a "i
u o ^ 5
^•*
O 13
" « 1 >»
a -'
, •o
fl> at ^
o> > a>m
ra u 3 E
— 4 -H H f*i j*i in r*i
*o*o fi o — ' o ^ro-^o
«•*• .... ....
o o oooo oooo
'— — * *-• — ' »-.-«.
vO CN 'tO *H CO in
mo CN^4 ^OON^H ^OOQO^H
i-l O —( fN M vO
o«r-Hm OMF-tr* OIH^HCN
p-t o\ fH i-i — I m
O O O -M O ^H
f*l ^H O rH
•H * -o * n • vp»\.0k
f>\o*Hin cNtN^Hco »HO r^
m fH (N pH ,-( —
r-) m fH * \o
* * CM ^ (NT
* * • • • •
O O O O
rH -H rH rH P- ^
II II II
m co CD vo r- QO
m .. rH - CN«rH» CO * 1 %
rHmlO i-HVfilO 1O CN
1 ^ (N 1 m CN O ch CO T T V T
in T -H rH
" 1* >•
Q) (1) QJ
.e s a
e- H S
!T 01 tT
•^ f -H I "^ 1
ra S < J u 2 J a 1
i^fri 0^ wE^fX* wt*0£
Ot Of 0)
< < 3
t
JJ
0)
• in
01
11 a
ta .U
41 n
-C >Q
^J
C IM
2 °
2, a
-S '"
01
-> OJ
4V J3
c 2
u
o eu
o
§ °
•a
»w 3
0
o *a
tu
S '
2 §
01 o
a
in o
tn Z
« «
-62-
-------
•H
1
H
I
a
a
3
z
0
M
L
§&
ys
0 =
u en
03 <
P
> O
E
OS t™*
•D u
o 2
aj o. ^
en ,0
"'a r-
00 2
Ed 64
EH """
" P, en
52 1
gg; S
"•a
a z
z <
< M
a £
ft] M
ft] Z
en M
S a
0 §
•J 2
< <
C*
O
^
o
en
M
a.
u
i
4
03
1
1
C en
•a o ~< —
w — « toft
tO t^ 3 S
TJ 10 « "3 X
10 > tn a
AJ 01 0> ~"
en Q i
01 —
O T3 -»
01 01 u =
2 Of | tjl
0) w .Q —
«as
13
01 01 —
Cfl > Oltt
n ^ a e
0) tQ 10 (71
> ^ > a
< O —
u
J3 «J u
S W 4J -H
3 0 10 to
Z Q tti
C 10
tj 0 iH -»
10 -U 3 S
10 > 01 a
JJ 01 4> *"*
en Q K
0) _
O "O — >
en o> !b"*e
2 S t w
O d)^
u a! ,H v.
fi ;> 3-
< O ~
w
(U Cfl
^ « u
g »W iJ ^
3 O fO tO
z a a,
Ul
Jj
0)
-HOco CNOP*
II 1 1 II
-t^o5o^ mr-\ooui -4t-m
M
ss^ss ;ssss sss
\& r*. \o \o if) oo>a3Cheo acn^1
•*ocft-H^ f*ir*o4*OM r-cocn
CN m m vfl ON inincn^fH MCO^
m^cnoo or^cocftepi ^Omi^>
V^-ltr-lrH iHl|H
mmcoco enr-p* -Hvc^fn
•^^•r*O '^rC'*1 2y5.r>l*^
(Ninficn vncN^H vo co (— < (N I~H fn
CN 1 | H M -H | 1 1 -H CN
11 II 1 1
u-» in r^ t.n \o
inp*mm *^^om ^or^m^
r-r*vOm fso,— t ^OLir^ut
^"inu^in incoco i—^rimo
mcocOr-i o^co fNuT-Hm
nrnfN^-j vs^o -vofNco
M^^-trH -HCN^ fN^H
C
O
JJ
5 a
C 3
tj CO U
>-H 13 £
(0 tp >i *3 ^
.. o aj w *" •*•*
>. i-tOt'x •*H
g O 0 O O **
•^5OvflO4cO ^ * *
E-00-^.H Z< ffl
>t >•
flj Q
t
w
0)
5)
£
jj
C
c
u
O
jj
C
0
E
cn
\
03
a.
a:
fl
-63-
-------
indicating further the systematic overprediction by this model. For RAM,
average differences show a mixture of over- and underprediction; most average
differences were less than the corresponding average observed values.
Overprediction is indicated for RAM, and is especially large for TEM-8A, at
Stations 101 and 115, during night-time hours (0000-0600 and 1800-2400), for
low wind speeds, and for Class E & F stability. The largest standard
deviation of residuals for RAM occurred at Station 104, where RAM
underpredicted on average.
Results for all concentration values at each station for 3- and 24-hour
periods are presented in Tables 5-14 and 5-15, respectively. The results for
average differences are very similar to 1-hour results in Table 5-13. The
number of data pairs is greatly reduced, however, and the standard deviation
values decrease as the averaging time increases.
One additional paired data set was analyzed in this study. During the
earlier evaluations of rural models, low correlation and large scatter between
predicted and observed highest hourly values were noted. In order to explore
whether significant improvements in model performance could be achieved by
relaxing the time-pairing constraint, a data set was constructed for the urban
study consisting of the highest observed and predicted 1-hour values for each
day. Performance statistics for this "highest by day" data set are summarized
in Table 5-16. Comparing these results with those for 1-hour values in Table
5-11, no reduction in data scatter is apparent. In fact, since the average
values are much larger for the "highest-by-day" data set, both the average
difference and the standard deviation of residuals are much larger in Table
5-16 than with the stricter time pairing of Table 5-11.
-64-
-------
u
i
U]
ta
s
J
Z
o
§s
5'2
85
u 10
01 ~~
c '«"
CT> 41 2 S
a n cu x
u
JJ O u E
« ^ IX \
u 0) i en
ni <4-t W 3.
> U-l ^ *-
* a E
•a
ai a> -~
i ° S2
in in in in -«r
f*4 CO
t/1 ^> t
1 1
csOmoo
• vo o -* !N
O o ° O ° O
Ifl
IT
c: c c
0 0 0
en w w w w co en co
0!
»
1
en
a:
-65-
-------
tn
1
TABLE
tn
1
z
O
3
£
*g
» S
y »
0 g
« 3
li
2y
OS *_J
§3
"" 0" _
•V "> «
M a; r-
g£ £
*-> rj w
li i
a- Q
Q Z
* U
si
i-
1°
o a,
•j 2
J *?
*• &
Eb
O
2
O
CO
a
<
&
!
i
GO
i
c a
•GO -< —
IS JJ 3 £
•a is " is ^v
c -H o -H en
IS > 0] =>
4J fl) qj ^^
tn a K
tu ^^
O T3 -~
tj> a) u s
a u a, >.
u tu i en
0) »w n a
S S 0 ~
a —
•o
aj a> ~-
5n > on
is i- 3 e
M at M \
o w ra tr>
5 0 —
14
.42 IS u
3 0 IS 10
z a a.
c m
IS JJ 3 S
•0 0 « -0 "X
C •*•* O -H Cn
is > 01 a
jj at a; ~-
tn o K
a —
0 TJ —
D C 4JM
1 t7>
(y — .
Q» > djn
RJ >- 3 S
U
< O ^^ —
*-***->
i-Hd^in^O 03f*l^'in^_| CNfNCN
O i-t CN rn
0 OoOflO oOQOo OoO
w oatnwaaw tntncnww cncntn
>i
a
,
m
u
0)
c
ka
fl
a
c
m
u
a
D»
C
•"
o
c
o
w
^
04
a:
fl
-66-
-------
TABLE 5-16
COMPARISON OF DAILY MAXIMUM OBSERVED AND PREDICTED 1 HOUR CONCENTRATIONS
(Highest by Day)
Model
TEM-8A
RAM
Numbe r
of
Data
Pairs
347
347
Average
Observed
Value
(ug/m3)
442
442
Average
Difference*
(Obs-Pred)
(ug/m3)
-952
(-1077, -827)
-202
(-299, -105)
Standard
Deviation
of
Residuals*
(ug/m3)
934
(870, 1010)
599
(557, 647)
*95 percent confidence interval in parentheses.
-67-
-------
SECTION 6
CONCLUSIONS
The performance evaluation of the urban models has produced a great
variety of measures which compare observed and predicted concentration
values. In this report, the results have been discussed and explained, but no
attempt has been made to compare the performance of one model versus another.
The conclusions and recommendations presented below are concerned with model
evaluation methods and with the performance of the models as a group.
The evaluation of annual average urban models was based on a very limited
number of data points. The following conclusions can be drawn from the
results:
o In many respects, all four annual average models performed
similarly. With few data points, it is difficult to discriminate
effectively between models. Statistical confidence intervals for
different models frequently overlapped.
o The observed concentration value at one station was large,
relative to the remaining 12. Several measures were strongly
influenced by this one value.
The evaluation of the short-term urban models involved the calculation of
performance measures for a variety of data sets representing selected peak
values, pairing in time or location, and subsets of events based on
meteorology and time of day. Three general conclusions can be drawn from
these results:
o Comparisons between observed and predicted extreme values, such as
highest second-highest 3- and 24-hour concentrations (Table 5-10),
provided distinctly different indications of model bias than
-68-
-------
statistical measures for peak values (Tables 5-7 through 5-9).
Single-value comparisons, while pertinent to a specific regulatory
application, provide an unreliable basis for inferring general
performance characteristics. Conversely, good statistical
performance is no guarantee of a model's success in a specific
regulatory situation.
o Performance statistics for data sets representing all stations
combined can be strongly influenced by high concentration values
unique to a single station. The 25 highest observed values, for
example, are dominated by Station 104, while the 25 highest
predicted values for TEM-8A are dominated by Station 101 (see
Tables 5-3 and 5-4). The results illustrate the importance of
examining model performance at individual stations, as well as
collectively.
o Both short-term models predicted substantial variation of peak
1-hour concentration values with time of day, but very little
variation was observed (see Tables 5-4 and 5-13). The models also
predicted more variation of peak values with wind speed and
stability than was observed. Such discrepancies suggest serious
problems with either model inputs or model formulation.
Comparison of results from the rural and urban model evaluation studies
led to the following additional observations:
o The highest 1-hour values observed and predicted for the urban
case are associated with light winds and stable (Class E & F)
conditions. By contrast, peak values for the rural case were
associated with Class A and B stability.
o Statistics for 1-hour values, paired in space and time, indicate
little or no correlation between observed and predicted values for
either urban or rural models. For the urban case, however,
correlation for 24-hour values was significantly better than for
the rural case.
o The reduction in the volume of statistics from the rural to the
uroan case was achieved without the apparent loss of essential
information.
-69-
-------
SECTION 7
REFERENCES
1. United States Environmental Protection Agency, 1978. Guideline On Air
Quality Models. EPA-450/2-78-027, OAQPS, Research Triangle Park, NC.
2. Fox, D.G., 1981. Judging Air Quality Model Performance (A Summary of the
AMS Workshop on Dispersion Model Performance, Woods Hole, MA, 8-11
September 1980). Bull. Am. Meteorol. Soc., 62, 599-609.
3. Londergan, R.J., D.H., Minott, D.J. Wackter, T.M. Kincaid and D.M.
Bonitata, 1982. Evaluation of Rural Air Quality Simulation Models.
Prepared for EPA by TRC Environmental Consultants, EPA-450/4-83-003,
OAQPS, Research Triangle Park, NC.
4. Strothmann, J.A. and F.A. Schiermeier, 1979. Documentation of the
Regional Air Pollution Study. EPA 68-02-2093, U.S. Environmental
Protection Agency, Research Triangle Park, NC.
5. Schiermeier, F.A., 1978. Air Monitoring Milestones: RAPS Field
Measurements Are In. Env. Sci. & Tech., Vol. 12, No. 6.
6. Minott, D.H., 1982. Development of Test Data Sets and Work Plan for the
Evaluation of Air Quality Simulation Models. EPA-68-02-3514 (W.A.4),
TRC-1671-R80, TRC Environmental Consultants, Inc., E. Hartford, CT.
7. Novak, J., 1982. personal communication with EPA/ORD,
8. Turner, D.B. and N.G. Edmisten, 1968. St. Louis S02 Dispersion Model
Study - Basic Data. Unpublished Draft Manuscript, U.S. Department of
Health, Education, and Welfare, National Air Pollution Control
Administration, Durham, NC.
9. Turner, D.B., 1970. Workbook of Atmospheric Dispersion Estimates.
AP-26, Office of Air Programs, Environmental Protection Agency, Research
Triangle Park, NC.
10. Holzworth, G.C., 1972. Mixing Heights, Wind Speeds, and Potential for
Urban Air Pollution Throughout the Contiguous United States. AP-101,
Office of Air Programs, U.S. Environmental Protection Agency, Research
Triangle Park, NC.
11. Cox, w., 1981. Letter to R.J. Londergan, 20 October 1981. U.S.
Environmental Protection Agency, Research Triangle Park, NC.
-70-
-------
12. Snedecor, G.W. and W.G. Cochran, 1967. Statistical Methods, 6th
Edition. Iowa State University Press, Ames, Iowa.
13. Hollander, M. and R.A. Wolfe, 1973. Nonparametric Statistical Methods.
John Wiley and Sons, New York, NY.
14. Hirtzel, C.S. and J.E. Quon, 1981. Estimating Precision of
Autocorrelated Air Quality Measurements. Summary of Proceedings
Envirometrics 81, 200-201.
15. SPSS, 1975. Statistical Package for the Social Sciences, Second Edition,
N.H. Nie, ed. McGraw-Hill Book Company, New York, NY.
16. SPSS, 1981. Statistical Package for the Social Sciences Update 7-9: New
Procedures and Facilities for Releases 7-9. C.H. Hull and N.H. Nie,
series editors. McGraw-Hill Book Company, New York, NY.
17. TRW Systems Group, 1969. Air Quality Display Model. United States
Department of Health, Education, and Welfare, 1969. PH-22-68-60, NTIS
NO. PB-189-194, National Air Pollution Control Administration,
Washington, DC.
18. Busse, A.D. and J.R. Zimmerman, 1973. User's Guide for the
Climatological Dispersion Model. EPA-R4-73-024, U.S. Environmental
Protection Agency, Research Triangle Park, NC.
19. Weisenstein, O.K. and J.H. Wallace, 1980. ERTAQ User's Guide.
M-0186-001E, Environmental Research and Technology, Inc., concord, MA.
20. Texas Air Control Board, 1980. User's Guide to the Texas Climatological
Model. Permits Section, Texas Air Control Board, Austin, TX.
21. United States Environmental Protection Agency, 1978. User's Guide for
RAM. EPA-600/8-78-016a, Environmental Sciencies Research Laboratory,
Research Triangle Park, NC.
22. Texas Air Control Board, 1979. User's Guide to the Texas Episodic
Model. Permits Section, Texas Air Control Board, Austin, TX.
23. Ruff, R.E., 1983. Application of Statistical Methods to Diagnose Causes
of Poor Air-Quality Model Performance. Atmospheric Environment, Vol. 17,
No. 2, 291-297.
-71-
-------
APPENDIX A
ANNUAL AVERAGE SOX EMISSIONS INVENTORY
FOR POINT AND AREA SOURCES
Data Description A-l
Point Source Inventory A-2
Area Source Inventory A-7
-------
The point sources described in this section represent the 235 sources in
the RAPS study which had non-zero SO emissions. The RAPS ID consists of a
x
two digit state code, a four digit county code, a two digit plant code, and a
two digit stack identifier. X and Y coordinates are kilometers in the
Universal Transverse Mercator system. All stack parameters are annual average
values with SO emissions in units of metric tons per year.
A
The 1536 area sources contained in this listing are those area sources in
the RAPS study which lie partially or totally within 30 km of one or more of
the 13 SO monitors. Each area source is a square with a minimum side
length of 1 km. A unique RAPS grid ID number was assigned to each area source
by the EPA in their development of the inventory. X and Y coordinates are
kilometers in the Universal Transverse Mercator system.
A-l
-------
E C\J
UJ LJ
»- a
U. PO C\J r* ^*
E
in z:— ......I................".
o H- v
K a
o
• >* • •••••€••••••••••••••••••••••••• !»•••••••••••••••••
o
\L
a:
z ..............;.... ~^..
>^ ^ c\jc\icccccx.o<£^^i/'inu*jU~. ininifio ^J c\. ce
o o
a. o
vi
A-2
-------
x^^
3 1-1
fo^
GO E **•
o
to
o
CL
«* OOCOCCCCCCCOCCXCOCCCiajCCCCCCCCCCCCC
r^r^h-ojc^cxevr\.c\je\jojCvr^
COOOOoOOOO^OOOOoOOOOCDOOOOOOOOOO
roK}i^rOfO
to
^o^o^K^<^
o =>
a. o
A-3
-------
oi u~ic\j(r*- c\j CM i-icofOfOrocM^occ
O>s ^H »-* PJOCC
Kl CNJ
3:0
UJ bJ
i- a
u^
^xc'^cr^^
OS: CNJ
K: *jjrHc-cvjC\.c\jc^a.c\J'-|'-<'Or^crx— ac ex K: ro fO «- K- .
*-<4-HO CNJ •-* CC CO CM — • ^ ^ ^, ^, ^-,
ru ro CM (\. o a; r- u~> tf>
< xD — • O
C^
O
0.
a.
ce.
^^(r^ovosrj^HCNj^
•••••»••.»•••••*•
'trtr)^
^f»-\i^ixOw\cr-^r-\4;x^'X^-r^
^c^
O 3
a. o
00
^w
A-4
-------
cr^ Co (r -x o ir>«-t^- tnin
•O"OvD'-«c\jf**-c>a^'a-
o in ir> IT in to \£ to
10 -x^" c\j
UJ LJ
t— C
O
a.
inu^u^'^O'^irjs£ii^'^
co
Cu
^a;ccccc\icNjcjo^oo^^c-^
ccvONiJ^acccaf^^^r^^xc^
J
c\irj^ojrjcx^^a^
cr cr c cr o c- c c- cr cr IT c^ ~
c\/ ex. r^ CM c\» cv c\j c^
O 3
a. o
CO
i o »-HC\iK^-tf',n^or"-cccriJ3^H(\)>o-3-iin\£}f''"'COff* ^
• cccorcccccc^^ccxo^^Xff-a^c^^^o^c^ o
A-5
-------
cv*
O
[fl r— sD s£»
t-l t-. CM PO
Ui UJ
>- a
K. O
O
o
a.
•-« ^-i c\j cv
Jc^
O 3
a. o
A-6
-------
LTJ VL. Kj C\J O O C" Q C'
0 cc c\i c\: r— o,' a
ec •-« »H «-• «-» c\i
CO — i-«
oe\j^^
a
h-i
ai
O
tn
a.
m in in to r- cc CD a. ac cr a^ o* o o o a o o o c= CD L^ if> \r> in m o o o o o o o L~. IT LT ^ IT- tn ir>
^. ac.ro^C"ao"ac--a-c>;raa^xa^a'C'T>crfr'CDC'ti>c;ri. oooczc.oc-cici^'*-i'-<^-«i-t^— <•-'•-' ^H^-,^H^-.«
-------
o >» •••••••••••••••••••••••••••••••••••«•.••..».••••••
STrtM i ~ . .3- „ j _ „ „
UJ ^
oz: , ,
** c\j(\jojcNj^c\jc\jc\jcv*csic\j<\jC\ic\fcsjcvjc\jc\jC\j^cNjcvjc^c\ir\j CO
<
u
a:
«z
jj ~
^ O^ *****' **^" *^*"*o"* *"**"(0»o""*rtr^"^» "*"*"«** "**«^"**'!*
c. —
^x
a:
ui —
OE ••••«..•
VI -*
^ CJCVCSl^CsiCsiriiCSJOJO^rviCNJCJfvjC^CNJCNJCVJOjrofNJC^CNJWOJOJCVJOJPdK^rO^^'^^^^^C^^^CSJOdCVCJC^CJCVJCNJCNJ
Q
o __ _
*~' <-7 I
A-8
-------
£ ,-c —I r-l —1
UJ -,
os- • ••»•••«•••••«••••
Q
O
0.
IT
VI
CX Q: a? C\JC\JC\jCVJf^'^CNJC\Jf^CMOJCsiCvriJC^rv.C\j(XOJC\JCMC^C\jr\j^C\JcvrjCVOJC\jCyC\JCSjt\jCVOjrljC^CXJ^C\.C\JC^O-C\jClaC\jO\JfO
O <£ ^
o o
VI fO fO r*"j "I K, KJ f~ "•" ^"i *~j
Q
o -sr ^^^i^i^)iOir>t^'iOiPi^i^i/^if''^^i^Lnu~)U*'^s'L'-iipinLn'j->i/"'in ii~> in i^ u"' u^ i^". 10 i^ LO u^ tn to u". ^* ,0 LP IP LO j~i in LT">
a.
oca; ^cjc\ic\jcsjf\icNicNjr^c\jc\ic\)Cjr\jcsjrsjojc\jcNjryc>jcNJCMC\(c\jcvc\ic\jc\jcsjc\jc\jrvJc\JCsjc\JC\lcyc\irjc\iCMC\jrj
s
in
A-9
-------
o v ....i. L. .....^'u.i".
sr tn 1-1 ro «-<.3-r\jcMr~u°5^jro*-H ^^
LJ —
** CMOdCMCM^CMCMPtJCSJ CM £>J Pd CM CM CSJ CM CM CM CMCMCMCMCMCMCMCMCMCMCMCMCNi CM CJ CM CM CM C\J CNJ <\J CM CM C\J CM CM CM C\J CM C\j CM CM
X ^ 'O'Of")rO'Of*"3''^fOfOfO to IO K) IO K5 K? fO K"' ^ to fO fO *O fO to K} *~> ^O lO fO fO f*7 **^ f"> f"^ K} Ki fl ^ to f*~, K) fO i*^ fO fO r»^ h'j fO fO
O
Q
0-
«S
<£
00
QC ^ ££ m fO ^O f.-) *^> >"") ^ fO fO HO fO fO f*i »O N^ fO K^ f*- K, K} K} **"• *O *0 ^ K/ K} fO fC I0 10 O K5 ro fO >••• KJ *"> *"j K~, fO rO *O ro fO fO K; IO fO -3°
^ « 13
O O
C/> M
<
Ul
c:
<
vD —
CC E "-t Cd «-H ^H CM
a —
o ("^ ^ K> ^ ^ r-T N* "O *") f") h") r^ *~) K"
O
a
00
0.
cr
ai CC K5fO'OfO'*OfOK)rO(*OfO'OfOfO'O fO rO'OtO'O'OfOfOtO"OfO'O'»'^'OfO'1Of^fOf'^lfOfOrOfOK)fOfOrOK^fOtOfO'O fO ro *O fO
O
-------
in
ac
o
w
LJ
<£
-1
CM £ ? 5 £ £ «£ £ S £ £ £ £ ° a
** •
CO —
5-E iri^r-cc^o-iojroa-^vDi>-a
- CM S CM CM OU OJ CM CM ft) CXJ C(-iiff-c:
JCNJCM^CVJC\JfOC\JC\JCyC^C\Jr\JCSJC\JC\JC\JCyC\JCVJC\JCNJC\JC\JC\tf\ICSJCNJ(\JrjC\Jf\JOdC\JO_'C\jC\i
;ONO-. CT-(^ONl>I^CP^CrOOOOOOOOOOrt-H«^rt^«^l(M(MC--.c\JA,vMCxJ^
'^-wrl-tfnnnfOuiJ^^rl^J^J^Jc-V t ^* M to c] ^ « LT* -! - -' " '
a
a
i*- x
tr• o*-
A-ll
-------
r~\ •—1
CO tr* C\J r-«^r-*^K}K3OQ'v^£>in^fO^H i-^CJ cr fVj c\j «— « ^* cvj c\j i-t ^- u" c\j »-
CC
cv
c. s:
^oi^coc3C\jr^,^cs.K^
'O>oc'C3^^^^^r^h-r*r^f^r^^^^c^
3T OOOOOOOOOOOOOOOOOOCSOOOOOOOOOOOOGOCr—»^^-Hr-*^-—'.-H"-<~H*-I--I-H— ~1^
o
o
(X
CO
o o
« 00
in^inu'jLniniO^^vxJvT'^v^^^O'aj^r^^^r^r^r^f^f^r^f^^ttcccrxzcoa.c^
in IT. ui in in in m u"i u~> tr IT in IT, ic ^ IT. m IT tr, vr m m in IT in in ir> IT m in ID IT LP in iri IT ^- IT tr> IP in m LO in IT t^ in ir> c
'Or-uTcc.ox 3 CL in »o OL sc O
«^a^
Q.
iO'O'*"i'*")'OfOfOc" ^•?'C"*r'ci^
-------
a£ cccuocccr''3'a vOinv£o^cjrjoa.^«^in^i^o^K- r^cc^^LTi^tr^r^a?K^^a3Oou^r*ro^xc\jooo^Lnc\]cCK}r^in^^oro^r^«^
O"v •«» •*••• *..••.•«......*... • *,,•*•
oo t— ccfyjoactoroctOvcovCKjcro^i^ CT^CU e\j—*fiin^c«-ip— fOr^oa-i-H in.3-cC'3'K}r-Ktc^ior^v£ CM cv
31 iP .-. CM •-« «-« CM vO o r*- r- csj if) »-• •-« CM CM CM 1-1 •-• C\JK}^HOK>OC\J E r^tr «-*fOm\i r^x^o^cyf^^iTsijccc3«fvjKic^^xa-0'-c^c^O"'C3coooi^^^^^LOirLr'r^r^^r^r^^a:c^aa
*•* cNjo«c\jruc\jCsJOjcvjr^CNjc\jryc\jcsjc\jc\jc\jCMCUcNjcNjC\;^hOKiK)K>rt
a
a
*-• iT)^^^ONO^c\i<)^u^^f^^o^o^cNjKi^in^r^a2a^o^ojK^^in^co^o^CNjroK)fOfOK)fO^^^^or— cvj*o>rif)^fH-o2aNC.»-»rjK3'*'ifJvfi^-a;frc*-ic«f*"j^L~OP*CLC' o
U uic_ ^inu^inu")U^u^inir>v0^^j3O^%£!^\ivj3r^^r^^r^^^r^r-^GCa.cc=Ca^
Q; O^ttT vC^\£sCO\£^^^CisOvCvfi^£^D<4j^v£1X^D\£^. vi'Jv^^^Ovf'X^ >^ti sCvi)slj'Jv£,'^, ^T\Os£»£'^^£sC\ri\i'\i;^r'\^O'^~
3
o c
CO - ^ry0^^f^«^LPooiOcNjc\j^\DtT'iri^^^»^^cciricr^~^^i^o«^rjacf^^^«^^^^LOoc"c\j^
-H O^ •••••*••• ••*•
ton- c^^TC\J''Oi-ic\ifOry-a- «• (TvsmtncNjKi.H.c-ioa^cocDr-r-.-. <-i fO cv ro ID -r ^r^a-^o-^coKi^inso^cca^^cNjK^^irJ^^asmo^rgfO^ririor^cccro^f^roc-xa^o
^ ajx^£^^o^cr-o^toir)u^LPir'r*r^r^^r^r^r^r*r-coa;coa:cca:QC(Xixa:o^o>c^^o^oc>^^
— c\jc\jrjc\jcuc\!c\Jc\iroooc\;cuOjcsjc\.c\ic\iC\.rgc\Jc\jryr\jr^ojc\jc^
^ -^^^^^^^i^^fursjrjCNjCNjcv'rvjcvjC^roruc^rycNjcjcvjcMf^rj^cvjcyc^c^
a
o
p— X^O^C\JK^(r^O'C'Cr'^ccro^cr^c'^ai^(^o^^
CO
a.
<
-------
ox ••••••••••••••• • •*...••••*.....•«.» •••»*
*- .^ _ ^
Uj *-
c ar •*• .••* .......*. ...........
to ^
** f\JC\JC\JOJC*iJfO''^ K) K> fO fO CO *O C\l Oj C\J (\J CMCNjCXiCNJCXJCNJCMCxJCSJ CM CNjrjrO'O'O "OfOfOfO *O (\JPJC\JC\jC\JOjC\JOJC\jC\JCyr^C\J
Q
a
0.
ce
M
13 < 3 ~
O O
V5 t/)
<
Lu
cr
«t
vO «
rt OX .......•...>!•... •..••..!...«•....•..... . . ~ .
Q. ~
c:
LLJ —
QZ • .•..j.«»jj«».«...»......^.^.....j.....«
oo ^
a
3 __
CO
a.
-------
O "x »»»«•••••••*«•••••••**••••••••*•••••••••*••*••*••»
3£ C\J -f «• in *O *-> CM ^ CU CM vfl ^••3'«-H CNj.-H^.-iCVCMCM'a-CMCU «—i rO CM ^
UJ *•
ca: •••••••••••*•••••••••••••••••»
00 —'
•*• C\lOjrjCMC\jCM(MOJO^CMOs|C\JCMC\ICyCUCMC\JCMK5KiK^K^K^fOrOK^KlC\JCM^
x ic tr/ u"i m m in m m 1.0 in in m in in in tr in in UT uo uo tn m IT in m u~ u~ in tn IT in IT- & IT. in in tn m in in m tr in IP in tn in IT in LT
a
Q
O CMPvJruCMPjC\jCM
-------
C\, CG ^ O •
cC f1"1 c£ ^2
*-» »•-» cc cc CM ir> PO o tr
OLOCC aj c£ f*» ^ <\
K) r— u") o if> f*- P>O -arced.
t*« oc «"* c; Q> «r ^ u5 ^n c\,
1 H- f-i CM C\J •
Q r
CO **
x^ inini^tnu^tT'LOinu^u^inin^inininuou^bT^i^inuTinL^in
cr
CD
CO
c_
a;o^w^'^c^
ir>in^D^^uJvc^ \a vfi *£ ^ ^- r*-
u u o
a: cr ex
O < Z>
o o
CO tO
xcccs;ac-
C >v
CO h-
E
^ tr •
a; o
^G •sOr*«-'-3'XaCX
itp^CCiXQ1 in^OCJ
a y
•-. ^
00 **•
CypJCNICNJAj(NdCNJC^
uv in *r- if> u~j u'f IT \r. ir, if> J^ xj IT Ji m in urJ u~ LT & \r> u~ IT m in in in to L~>
CO
Q_
a;
z>
o
*o K*. •a- -a- •
A-16
-------
.
c\i>- r^^r^^^o^crv^^or^cco^o^cvju^K^^c^xc^ovJ^cc^K^cvK^or^^^
O *v *.*•••••»•••*.•••*• •• ••••••••«••••••••••
3C •-< ro ro .-i ou »-t •-( foes- CM r,j cu f« .-i miOcr^o
Lu •"•>
az •••• *..»
oo <~
>- si ojK}^cy*oi^c\ifO*in^r^coCT^Oi0^0''r^^o^^OvCT>c^ c-1
x *z & ID IT. LO y> in in lO L^; ID in m ^ in LI uO u"> u^ in in co in in IT tf> tn ui iT; IT- u^ ^ ID ir> in in u"> in in u~> in LTI u^ in u^ IT, ^ ^ c o ^c
a
c
(—. xoNO'^cN^^^\or^co0^a»^fsjfO^uivOCTvOrHC\ifO^in^r^aia^o^^c\jrO^u^viir^co^*^'^c^ -Q
en c\ii^KjK^fo^^ro^rOK^KOK)c'^^fl-c-^^^u^uiinu^tnu^u^inLnin^^vc^vfl^^^^c^
CO
0.
<
IX
w
uj o^ ^o
c: 2: tr oooooaooooocioOOOococ3Qooc>cooooooc3c:oOcrc5O£jOOc_-oooc>c?oo—«
o o
CO CO
<
UJ
X.
«t
x.0 «
r^. Q£ ^s^U?^CCKirvjO^O^v^^-^TrOP^CV^^CCOCCt^^^Hff^CN;^h-a'^TCCCJ^^'^O^rO{TlNCl^
cr^ r«>- o>o^ccoNu^^^^^^^c^*^vflu^r^tnu"''^'^v£*Ki^^^CNjf^f^ccc\jcxjc^inrvjoc>s£cy
^ O^v •....
f)>- o^o a-hOc^JCOo^^^-^£otnCyK)'O^or-\£all^DcC'CCaOfVt^'O'O (-ia;r*r-r»r^«a;(r'^f*- -S-K?K-,
(/) S r- tfJ p-«^H^v£ifO ,-t »-, ^( r-, -ff- P-I^H
a. —
<
C£
UJ *-*
ox: ..•.......••••.*.. ••.».•*....•.......
»«^ LOLnC"'-^'^'^'^^-**—I^HfHi—"CNJlPinin •~H, I-H ^^ «—* «-^ ^H »-H i—i ,-^ -—( i^«-^i—t^H.-^tO'^'^'-H'-^'^^ •—li—«.-H ^H •C-'-'*.-^^^r-iT--s: u^oiroso-^ojK^^Ln^r^a:inoinor^oj^^lpi^r^oicro«c\j^^inc^O'^fvjK^^ui^p^
i£ ^-i^)xj3sOh"f^-r*r**l**r*- f^1 ^~r^« aoij^o^ocsoooo oooo«~*•-* *-**-**-^^-tixr^r^r^r^-r^-r^f^^-oo o o ^ o o «—» «-^
— C\iCSiCVrUC\iCy(\iCNJC\iCyC\JC\jCgOOC\iC\JPOfO^OK^K7K^fOKOK^^^K)K^^K^^
s: in in in IT, *n »TJ m tfi ir» in if> \n tfi tT> \rvnmiT'ir^nu^ir. u^ir. ^inirifjiTJifjinin\C'O\coi^oo\i;vD^^a^)sD^jv^^D\o>jDO
x it: [r> uo i-r' u~ in ir, u"5 ui u*< in in in IT- IT. in in tf> ir in -^ tn LT in IT. j~. u~ in IT L"J IT- in ^n u*; tn in LT u~ u~- IT x. in in tn in ir. in ur> u~ in L*
o
c
*-> vDr-ccjNO^cyfO^-in j:r-co^cv"Oo-HC\j fO-3-^^jr*coaso*-*fNjfO-c-j^^r-.^crcD-HC\jfO j-ui^f*-
oi r-r^^f^o^aca^xcCcoar)cC'ccitriaNiTvONcr!Ti!T^tT'fTNOoc3OC3C3oooo.—i *-»^-«^-tr-(r-4^^^-«<—' —tour\jojc\j<\j(\ic\jC\j
CO
a
<
£f^xc^o»^c\jK^^Ln^f^c20^Oi^ojrofl'in\r^xo^o^-«c\;'^^rLno^ — o^Or-iOj'O^-'j^vcr'—a;^0
UjU oooooooaOi^^^^^^^rt^^cvJC*jv\jf\iC\*r\jcvrj^OjK;»ctOfo«'OtO(OfOto* •a-*a-*'S'^-
-------
o^ • •«.*••••»»••••»•«•»••••»».».•••«.•••»••••..••.••.
UJ —
c £ • ••....•.••..••••••••••• • . ....... .
00 — -« — .H _< ,-1 tH ' ' *" ' *H
*•* CNJCNJC\J{\iC\JC\JCWCWC\Jc'vJC\JCMCN(C*JCUC\jC\JK3rOCVCMC\JC\JCMCM(\JC\J ^O fO C\JC\JC\JCyO\jc\J'OfOC\JC\jrMtocUC\Jf\JC\IC\JC\J\\JCMCV
x ^ ^*im\i3v^s£o^v^^^*£^^^'OiJJ^or~f^'tN-^-r~^~r^f*-r^F~-r^r^f^»f^-i>»is-r^r~ccctcccccCiXi^c ..<...".
0. ~ ~
-------
CD O C O
C" -3* C- O
I/) h- ^C «-H C\j ~H
O E
it^fOocM-3-vOTC3^'S-'-2Qrt-j'-* KJ in r*- cr »—« K^OC. oj-c-o^ -x
crc>cr'in,nLr ir'^vCvc^vrccarcccca: a:crcrcrL'~;uiu'"J^ou~
;
o: cr^croooooooo
O C3Of->^t»-<^-
o
cccccccc^coxaicC' CCCT. tr c c1- o-- cr a^ CT> o^ cr (.3
X
f—
cr
Q.
aNa%crc^(r
>^ONCa;L'^iP^xa:iriin a cox SCLDUI^^J-X^
o
Q
CO
a.
) CM CM CsJ CM CNJ CNJ CN. C1, C1^
uJ O
CC K
< O
O
CO
A-19
-------
ar o as cc to CM cc o-
«-« to cr- i— i
c- r- ec o — •
c- CM in en ov IT r-- r-- cs r- ro cr o -3- r- o
CM
c-\ccco<>j^\3^u^a,^^
-tr)Lriir'^^'«Os£'C)';r^-ipLOir)^'j^^3\r^^oo\£sC^^w;
C:
o
oo
a.
^ui'jDr^cOCT'O'^OjrO^uT^r^
C-^-
OX
QE
»— i ic
CO —
X Jd OJ CJ CM CX C\* C\J Aj C\l CVJ Cg C\; CNJ C\i OJ £M C-- C\; CV C>. fy CM C\j
' f\j C\J C\, tX' C^ CN* C\J Ou C\, Ts, C\i P-j iN CN,
O
Q
O
j ro c\t c\j eg c\j cv:
A-20
-------
3C »—i CNJ i^ «-« «—I C\J CSj <-< C\fi-»fO «—i CO CM
Lu —
CO ^
*•* c>jrorOroc\i fO fO fOf'OfOK^flOfOf<"jK5 KlK^fO f'OfOfOrOhOfOl'O fO K)K/fOfOfOK"J'O^"J(iO K~i fO K"; fO K) K^ rO fO fO K) *O
a.
a:
o o
00 CO
LJ
CL •*• OJ
JC\jrOK^
O
cr ^c\jOjf\,. cv c\j
a.
•<
or
O ^ ^ ^ '-«•-' ^ ^-t ^,^^^,^^^^H^^H^
GO
A-21
-------
o
a:
c
PC >o OD o f*- s* o- cc
o u~) C"> cr> D"* a
on
oo ~
0" T^CT^oo
l
at a:
in in in in LO if> in ^n in ir>
in LO ir^ IT* in in -i -.o i
A-22
-------
APPENDIX B
ANNUAL AVERAGE METEOROLOGICAL JOINT FREQUENCY FUNCTION
FROM THE 1976 RAPS/RAMS DATA BASE
Data Description B-l
Joint Frequency Function B-2
-------
The meteorological joint frequency function contained in the following
pages is in the Standard National Climatic Center format for a 6 stability
category day/night STAR deck (1=A, 2=B, 3=C, 4=D-day, 5=D-night, and 6=E-F) .
o
Sixteen 22.5 wind direction sectors are included, beginning with the
northernmost wind direction sector and proceeding clockwise (i.e., N, NNE,
NE, ...). The six wind speed classes, designated as ul through U6, represent
wind speeds in the range of 0-3, 4-6, 7-10, 11-16, 17-21, and greater than 21
knots. Central wind speeds for each class did vary by model, however.
B-l
-------
OOOOOGOOOOOOOOOO
OOOOOOOOOOOO OOO O
oooooooooooo o o o c
oooooooooooo o o o c:
OOOOOGOOCfOOOOOOO
OOOOOOOOGOOCTOOOO
OOOOOOO OOOOOOOOCj
oooooooooooo oooo
OOOOOOOOOOOOOOOO
OOOOOOOOOOOO O O O o
ooooocoooooooc. o o
oooooooooooooooo
O O O O C OOOOOOOOOOO
oooooooooooooooo
oooooooooooooooo
oooooooooooooooo
oooooooooooooooo
OOOoOOCDOOO GOO OOO
OOOOOOOOOOOO CP OOO
OOOOOOOOOOOO O O O O
OOOOOOOOOOOOOOOO
oooooooooooo er> o oo
OOOOOOOOOOCDOOOOO
C\J
00
—I
u
-J
u
--r 000
oooooooooooo
oooooococ ooooooo
OOOOOOOOO C3OOOOC3O
oooooooooooooooo
ooooooooooc^ooooo
OOOOOOOOOOOOOOOO
OOO OOOOOOOO^ '3 OOO
OC3OOOOOOOOOQOOCO
oooooooooooooooo
OOO OOOOOOC»OOOO OO
C3OOOOCDOO&OOQOOC5O
oooooooooooo oo oo
o
h-
U
OOOOOOOOO0OOOOOO
COOOCJOOOOOCOOOOO
OOOOOOOOOOOOOOOO
OOOOOOOOO -J1 O O O O O O
oooo
O O O O
I I I I I I I I I I
CDOOOOOOOOOO
oooooo CD ooo
OOOOO
UJ OOOOOOOOO»-<
(£ •••.••••*••
u.
Ul
o
a
c
-3
oooooooooooooo
C3OOOOC3OOOOOOOO
I I I I I I I I I I I I I 1
OOO OOOOOOOOO O OOO
oooooaooooooooo o
I I I I I I I I I I I I t I I I
OOO O^OOOOOOJOOOCSO
oooooooooooooooo
K) -
Kl >O K~
GOO
OOO
)OO OOOO^OOOOO
I I I t I I I I I I I I I I I
»ooooooooooooooo
1000000)0000000-00
IOOOOOOOOOQOOOOO
ry »o 10 c\j cy c\j oj o. c\j fo Cj >o CM c\i »o
-------
ooooooooooooooo
oooooooooaooooo
OOOCJOOOOOOOOOOO
ooooooooooooooo
OOOOGOOOCJOOOOOO
o o
t I
o o o o o o oooooooooo
OOOOOOOOOOOOOOOO
oaoooooooooooooo
O O O O O C: O O O O O a LO O O O
oooooooooo o o a o
oooooooooooooo
OQOOOOQOQOOOOO
o o
oa
OO
fO PO c\j cv ro K) K-, oj ro fO
O O OOC3OOOOCJ
O O COCDOOOOO
I I I I I I I I I 1
ooooc»ooooooooo o
ooooocooooooooo
ooooaooo»Hooo\£oo
-3- o o o o a; CD vOffvLT>r*-c\jiX<** .3-
oooooooooi-ioooooo
OO OOOOOOOOOO
oo oooooooooo
II I I I t I I f I I I
cooooooooooooooo
OOOOOOOOOOOOC3OOO
OC3OOOOC3OCMO OOOOOO
OfOCCoOOCOO'iU">CCC\JLOO^'-*CUCC<
CD tr cviCjOocovocua- ^-in\D«-ia-csj
OOOOOC3OC3OOO
OOOOQOOOOOO
I I I I I I I I t I t t
ooooooooooo>ooooo
ooooo^ooroo^aoo oo
>-i 000
CO
<
(/} ** *4 ^H I
o
u
o
LJ
o:
u.
OC/OOOCDCJOOC3 OOOOOO
CDOOOOOOOOOOOC.OOO
I I I I I I I I I I I I I I I I
oooooooooooooooo
OOOOOO C- OOO OOOOOO
OOOOOOOOOOOOOOOO
o^oooocaoococ^ooooo
I I I I I I 1 I I I * I I I I I
oooooooooooooooo
OOOOO OOOOOOOOOOO
z
o
OOOOOO OOOOOOOOOO
oooooooooooo o CD a o
I 1 I I I t I I I I I I 1 ( I I
OOOOOCiOOOOOOoOOO
C3CDCOOOOOOOOOOCDCDO
LJ
X
CMOJCMOdCdCJCNJCMOJ rvJfSJCVOjCM
OOOOOOOC5OOOOC.O
OOC3OOOOOCDOOOOO
I I I I I I I I I I I I I I
OOC5OC3OCDOCDC3OOOO
oooooooooooooo
oooooooooooooooo
I I I 1 I I t f I I 1 1 I I 1 I
(ZSOOOOOOOOOOOOOOO
oo 00000= ooooo
C3^ OOOOOO,OOOOO
I I I I I I t t I I I I I
OO C1OOOOOOOOOO
OO OOO OOOOOOOO
r- r— -x •-»•-«
0" CM u"' -r «-«
B-3
-------
OOOOG^OOOOOOOOOO GOOGOOOOOOOOOOOQ
OOOOOGOOOOOOOOOO OOOOOOOOCTOOOOOOO
oooooQoooooooooo oooooooooooooooo
OOOOOOCGQOOOOOOG OOOOCJOOOCrOOC-oOOC^
OOCJOOOOOGOOOGOOO OOOOOGOQOOOOCJOOO
fOCNJfOfO K) C\J CV K")
O O O C3 OOOO
oooo o o o o
till I I I I
OOOoGOOOOOOOOCtOO OOOOOOOOOOOOC. OOO
OOOOOOOCOOOOOOOO OOOCTOOOOOOOC'OOOO
ooooooooccooooo^-o oooooooooooooooo
OOOOOOQ-tf-O-y-TOCVJCCJ''?'- OOOOOOOOOOOOOOOO
ooooooof-Hp^.— i-td.a-.a-tncr' ooooaoooooraoooc:. o
to
- K> Kl KlfOfO OfOCSJfO
OOO ^OOOOOOOOO *- O O OOO OOOO
OOO OOOOOOOOOO t-* O O OOO OOOO
I I I I I I I I t I I I I -J I I 111 till
OOOOOOOOOOOOOOOO «-r OOOOOOOC3CSOOOOOOO
CDciOOOOCSoP*-OO<^OCSOO CO OOOaoOSGOC-OGOOOO
oc
c
jcjycsjf^-«jj(ij \jfj<\ji-if-ii\jcivj*-*c\;
U. GOOOOOOOOOOOOOOC3 U. OCC5OOOOGOOOCJGOOO
oooooooooooooooo ooooooooooooc-ssoo
>- I I I I I I I I I I I I I t I I >- I I I I I 1 I t 1 I I I I I I I
U oooaoooooooooooo O CTOOOOOOOOOOOOOOO
Z oooooci/O'ooooooooo ~z. ooonfaootrr^c^oococ"*^
UJ r-csj'd-ooc3f*-'S*«-^«^cCcco\j3^-i— t LJ *H*^ovDc»Or^-a-c\jo^-cC'^i/i^o^
3 lOff^KiGOCT'a.inocruaxcC'a-or-t'O o ccojsr\i>^'-«c\j^-to^-infii-Lrr-fO(M
c cv^)o^-^D^rr**-Hr»-{\it-(»-i«-cciPX O r-cc-3-fO^i-icCLnincNJLOU">FOfooff»
"*
C
-5
L>J
ojc\i(\jc\jc\jf\JOjrd(\Jc\j^fNiCM^oojc\j X ojc\jr-.c\if\jc\)«-«>-«'-'^'-«'-t'-tcacvc\j
OCJOOOOOOOOOOOOOO h- OOOCiOOOOOOoOOOOO
oooooooooooooooo oooooooooaoooooa
I I I I I I I I f I I I I I I 1 I I I I ! I I I I I I I I I I I
OOOOOOOO>OOC5OCDC3OO OOOOOOOOOCS^OO^OO
OOOOOOOOOOOOGOOO oof-")Oo^o-a-ioc\je\1''3'crc3Oo
cj m rg ro f\- c\i c\, ro c\i Cw K» -o F~) CNJ fO o- Ou ru cy cu c\, r\i oj c\i »-« rj r« cy ry cu oj CNJ
o— cjoc^aoooooocscr-oG o^oooooc'joc.oooescjci
oooooooooooooooo oooooooaoooooooe
i I i i i i i i I t t i i i t I t i i i i i f I t i i t t i i i
OOOOOC3OOCTOOOOOOO OOOOOCJOOdOOOOOOO
OOOOOOOOOOOOOOOO OCCTOOOOC2OOOOOOCiO
o
*—
o
B-4
-------
APPENDIX C
HIGHEST AND SECOND-HIGHEST SO2 CONCENTRATIONS OBSERVED
AND PREDICTED (RAM AND TEM-8A) IN 1976 FOR THE RAPS/RAMS STATIONS
-------
M
1
u
a
j
n
*
10
CT>
.H
M
-x
en
0.
2
K
co
z
O
s
$
g
u
o
a
Ed
E.
u
rH
a
Ed
«
a.
a
z
<
a
Ed
>
tt
Ed
M
O
o
a
3
O
S
H
3
§
a
d
i
1
_
a
«
>
01
ta
XI
0
u
s
ra
a
i
o
i
g
\
&
—
u
X
I
£
^
1
U
S
>i
T)
1
*—»
•0
g
3.
1
5
CO
in in
o o
CO CO
CN fs
1 t
0 0
-t -H
GO p-
CM CTl
m m
00 0
0 CM
CO \O
1 t
rH rH
O rH
^J- CO
CN en
CN r-
rH
(N •<*
CN rH
M ^O
CN o
t 1
fH rH
O rH
m ^
•«T ^<
in vo
a)
o S
rH CN
inin^oinfNinnrnmfNr-"
OOOOOQO oooo
coco'*^co'*orn^ «9* 'ninrn
CMCNfJfNfMcNrHrH CNrHfM
t | 1 1 1 1 1 1 1 1 I
OO^O^CQ mm cOrHfn
rH-HOrHOOOO O rH O
CN m co m * co ^ m mcncn
^•oco^tnoovD^ enco1^
e^^«p*i^)Chinin mr* mr»
-—lOOOfHO OO OoO
i i i i t i ii iii
mrHcnco-^co^ mfNop*1
O O O O O O O O O rH O
vD^rHen^min co r^rHQO
OrH*ncnf^>oM'Hco tn «HJ
*OvoF""Cor""*o en p«.\och'"fc
-H rH (N CN
^Icocnv0'-Hin CM CM cnmcN
rHOOOrHrH rHCM Oo^
^Op*\DP*-<^f*l CO rH rH^TO
OrHC5O-HCS--liH rHrHM
1 1 1 1 1 I II III
CN^oCMO*!*'iH rHrH ^-(NCN
H O *H ^ O O OO O rH O
r-^comvOfN m encomo
co^iceo'HvoO^ fnh^* coo
^cn'>''^*l'*o in o o ^«cn
fM rHtN-H^ ,-i-H rH
~* —• . .-. — .
^"in^oco'^^'in^ c^fsiCM
O rH CN <*>
m v LH 10 r^ co enrH HrH^H
.
en
0)
c
f4
U
0
jj
C
i
i
\
in
Q<
<
X.
a
C-l
-------
t
u
TABLE
ON
W
o*
2
g
0
1
§
CJ
a
u
EH
CJ
PREDII
a
z
a
H
a:
u
OJ
m
o
OS
o
X
•H
E*
en
1
X
a
o
a
u
a>
1
s
E-
i
"O
>
1-1
o
in
€
I
in
a
4
_
•o
e
o>
s
1
i
*"•
\
5
u
S
t
£
•n
£
CT
~
0
(0
U3
OOOOH Of-iojaoooM
coin^m ino>^*o\chpi*in(*ifri
^NrsjrJH OiNOOjOJ-^OJcNOJ
1 t t I 1 i 1 1 1 1 1 t 1
•HOO-H Ooooooooo
^cofnxr«)mo4p-f*iof-i^r-
co i— i o o Cf»o3P*r*r*>^ii'>inco
co^ininP-oj^'ch*s*OP*coo
OoOfN OOJ OOJOO—i-H O
Oco^^f ojco r*^-p-m^HO
r^OjOi— trHO'HHOJ^|P-frH'-H
i l I t i i i i i i t i i
^OOHrH r»ojQO fsi^r\ao>p*.Op»\o
— 1 -H r-t f-( M
f*i m f*i co f*i CN oj v0r-trHO~4crt
O4,-tO^MOJ<-(i-4OJr4<-404O
OJQ^fH OH^H OJ^r-t'HO'H
1 1 1 1 t t 1 t 1 1 1 1 1
•-IfH-HfNOJ CN^'H'-^'-t^OOJ^
ojp^oaa^Ovoo p^poFH^rHP*
in^f*iipo^tP"'CO"siPi*
rH »-4
rHf*i^Ln^o eo'*^^'^^^s'ojoj
o -^ (N m
^Hojo^in VOP* CD cri <— i —* .— i •-<
0)
01
f-
c
0)
rtj
a
c
01
0)
0
Q
t-i
O*
c
u
o
C
g
05
g
fl*
s
fl
C-2
-------
r*>
1
u
J
o
EH
to
r*
W
W
0.
2
z
o
M
a
1
1
8
a
I
a
u
Q.
a
z
«S
a
a
en
o
o
a:
§
•*•
m
H
u
EC
M
X
<
CO
1
i
-
1
4)
>
U
0)
u
.Q
O
U
S
1
i
—
\
O>
a.
X
g
„
n
S
3.
_
^4
ffl
S
1
s
^H.
E
1
0
•iH
JJ
U3
o
CO
CN
1
O
r-t
r-
o
in
o
0
i-H
0
•*
O
rH
CN
CS
1
r-l
SO
i-H
rH
«J
O
1-t
OOOOoOoOoOoO
incoino>coorn'j3p*cj^^op*'
,_i CNH I I I l i i i I
-H O-H^COtnooMp*. CNi/i<^
rHrHHOOOoOoOoO
p^cOiH'^'fl'coincoin chmo>
p-t-HfNiHfH r-4r-t^H -H
O O OOOO*OOO'HHO>HOOrH OrH>H
I 1 1 1 1 1 1 1 t 1 1 1
fM'H^H'Hfi
oo O-HO^OOOO oo
mcr\'«s**firH^lp*i'1>rM'^ r*^H
mco.Hso^'mco^cMCNricM
H tHeNOCNO,HCM,H ^HQ^H
O
(Nfn^'invor-COO^rHrHrH'H
,
m
0)
a>
JZ
Jj
% c
u
s*
c
o
o
a
rH
C
u
O
-H
C
o
€
03
i
en
OH
$
a
C-3
-------
1
U
U
•J
1
IO
r-
ON
W
VI
cu
3
o
fc
U)
z
o
M
1
6^
M
5
8
Q
H
M
O.
O
a
^
a
in
o
o
as
ss
ro
i-
tn
u
3
M
X
a
z
<
CO
I
I
1
u
o
t
i
—
"xl
3.
.2
w
cn
Oo^OOoOoOoOoO
fN-sj H-t-H-4 OtN (M-HfM-^CN
t I 1 1 1 1 1 1 1 1 1 1 1
HH-H-IOHOOOOOOO
00 ^^^0*^"^ C*1CO "TrHinON
>H f)f"-infMr*J r"^\o fnr*mm
HrH O—tOrjCMofNl«-HO^fM
1 1 I 1 t 1 I 1 1 1 1 1 1
•HtHOfHCMo'"* ON —^ OtNH^n
O-l-t-HO^IOoO-HO-lO
ON ONf^OCNcOONONCMoOaoOO
^'CS'tOCMinHONONH ON ON CO CO
fM_| OrHrHcM O O CvJ O O H <-!
(Nosin^aOfH o mH in «— i ao ^
fMo^Hi-HpHi-H O>-< O *-H O i— 1
1 1 1 1 1 1 1 1 I 1 1 1 t
OH-HO'-fH'H O^ O O O *-l
maxO^cof^coON mos moNrs1^
ONinOeou^r*1^ PH^HCMOC^I
r-4 H
to _» .-» _ *^
•Hm'*tn^Oco'*i^ru^ HCMrgtM
~ — ^w — ^^^— o ^ CM «
iHc^^^mvor* COON HH^^
U
Ul
OJ
^
<1>
U
CLi
C
01
5
y
a
M
c
o
•*^
c
o
e
cn
|
OJ
i
-------
tn
1
O
u
a
en
z
O
5
u
O
u
a
u
£*
u "*
a
z \
i!
S a
12
O
0!
3
O
S
fM
f-
2
O
s
^J
CO
1
S
S
H
-
i
•a
>
aj
tn
8
33
Q
i
s
6
O>
S
It
>t
(0
O
i
£
n
e
IL
Ul
X
2
t
£
^
A
O
fl
JJ
m*ovoininr~innu3vo»r--t
1 1 1 1 l l 1 i 1 l 1 1 1
mw-ioo^orjnovoomj;
o^r-r-^mm^o^m^^
SSo^SSSSSSSSS
i i iT*i i i7i iT1?!
aOr»»ocMi»ioo>r~t~momoo>o'n
f-i
O ^ CM ^
.
m
w
a
j?
JJ
C
Q)
u
a
c
01
3
U
Q
M
C
•H
w
S
i
cu
(0
C-5
-------
W
z
o
CENTRAT
8
Q
a
S
Q
a, m
Q 'H
z w
<€ 2
0 2
> w
K ^
U <
ca as
0)
og
a u.
s
o
X
CM
C
(fl
Cu
X
SECOND
4
00
c
I
1
Observed
u
~*
fl1
4
S
S
S
X
id
a
i
o
£
—
"^
a>
3.
*^
X
I
1
•x
I
o
'H
a
4J
Ul
'Tinino>oo^oovniovijino\in
1 1* I ? I 1 I ? I V 1 1* 1
^M^O-HrHOOOOOOO
SSoS°5SKSSSSS
i i i i i i i i i i i i i
OO^O
-------
APPENDIX D
HOURLY METEOROLOGICAL AND OBSERVED CONCENTRATION DATA
FOR SELECTED DAYS WITH HIGH MODELED CONCENTRATIONS
Data Description D-l
Meteorological Data D-2
Observed Concentration Data D-13
-------
The tables in this appendix provide sets of daily hour-by-hour
meteorological and air quality data (negative values indicate missing data)
used in the urban model evaluations. The sample data are provided to support
case study analyses of days when high concentrations were observed in the RAPS
network or predicted by the short-term air quality models (TEM-8A and RAM) .
The days selected, and the basis for the selection are given in Table D-l.
TABLE D-l
SELECTED DAYS OF DATA AND SELECTION CRITERIA
Date
01/15/76
01/26/76
01/26/76
01/27/76
08/23/76
08/23/76
10/28/76
10/28/76
11/15/76
11/16/76
12/06/76
12/11/76
12/15/76
12/31/76
Criterion
RAM-H3
RAM-2H24
TEM-2H24
RAM- Hi
RAM-2H1
TEM-2H1
TEM-H1
TEM-H3
TEM-H24
*
OBS-H1
OBS-H3
OBS-H24 '
RAM-H24
Hour Ending
3
24
24
5
5
5
5
6
24
*
14
18
24
24
Receptor
3
9
9
9
8
8
1
1
1
*
3
3
3
1
* Several high or second high values were observed and predicted for 1-, 3-
and 24-hour averaging periods on 11/16/76.
RAM - RAM model predicted concentration
TEM - TEM-8A model predicted concentration
OBS - observed concentration
Hi/ H3, H24 * Highest 1-, 3-, and 24-hour average concentration for the year.
2H1, 2H3, 2H24 = Second highest 1-, 3-, and 24-hour average concentration for
the year
D-l
-------
O CO
O CL
_J<
Oc:
or in
oco«-i
IT LT tr tr u"i a"1 iT. c- o in « IT c u~' u* u^ u^ F-") (M ^i CT- a. r- c
O
a.
c/5
o
a;
C QO *•••••••»•• ».•*
Q. U.' f^ "T CC c: i-t \C ~2 IT C~ r- CX' -^ O K: rfj r*- ?« a. CM n-, o o rO
00
D-2
-------
X
ii> GO
•— i Q;
UJLJ
Si
<
a:
05 CO
O Q.
C.
s:
o
Q.
o
o:
CO
d
n.
o o
z: 2
D-3
-------
01
z
o
•—' Q^ OOOOOOdf-igsu-j,-*!
LJ wJ *«*••••••••
L_ ITJ U"' IT. ^~ LT. U". IT. •
X E »-* *-H r^ ^H ^^ t~4 i~1 '
a: o CN. «• ^C
O 00
O Q.
o a:
ir
O 1/3
CL X
OJ UJ O
^H O
c; oo
3
O ul
a
a.
o
a.
Q O
u:
c. c
a 2
o o
D-4
-------
Uj UJ
x i—
X £ M rt _, -I
M
s:
«s
cc
_
C ci
o; fO
O 00 CM
-< in o a-
O IP
X CM
O
Q.
Q '
2 :
3
3-
5
Q O
LlJ
C C:
CC OC C\J ^ f
o
tr: 2
Z! —
O O
D-5
-------
CO
00 W
x
c< 01 ooooooK-3-tT5'jDr^-
t-i o£ oooooof-roc-'ir — •
O 00
c a.
_j--3-ccco<7*fi*--j*-^cNJr*ec-3-o-GC
oa: o.ii ^-tu^Lno^T--a-L'*i'^"'C3fi-coiO4-ooa"io-HC\:c\i*-i
H-OO o (\jrycyc\jojcyf\jOdCuOjajc\Jcvf^cxjcsi
UJ C «-i
a: _j N.
o
>- • r-
0.
oo
t
a *
o
Q.
o c
X 2"
D-6
-------
on
CO CO
H- _J
w o
CO *—
O O 0? OOOCJOOi—< ,-( *-( i—'OOoOOCiOC'>OJ\£ONf'"'.'s£Q
*-» •-•a: ooooooFT«'*'tf)var*-cso*oooinir>a:o(Mif)r*o
h- UJUJ •••»••••••••**••••»•••••
< XH- ^Xv£^iX>^f^-iJLO-a'lOC\ii-«*H«-^'^'J'CSJc-r»»ON^^pOO
v- LW crc*0vc~crai^-^'U^ri*(r'-'fOir:irir-irJoixv*:fl-"^f-fCs
CO XE ^CMrO-«-inf*aCO>trff^ffvCCvI3lO-^-'OCNJ
oo z:
£
<
01
>- X.
O CO
O CL
_j< LOiT)incr-TpcccC'-»fr'C1''^i^(Mri-r-cuiri'3'fOc\jo^f-Jr-«-»
a: in n ••••••••«*•*•••••••
ooo--t LJCS ro»o^c\jc%j»-ii-«fOLn^cc(r'CrooONQC^infl-«r-tf-^OK)
>— Z3 —" O c\jf\jcucuojc\jrucvj^<^csjc\jco(\jojwojc\jojfuojc\Jc\jC\j
UJ O T-I
E _J x.
^
>- • r-
_j i-
K oo a
^ Q.
OiD oo u^cc*£rio«-H:ciPvD^ou?«-»(rCT*K">ri'-cuc\is£ccir)^cc''0
ZCM co >xu~'CCfO-a'O^r-^-'~JOccc\;fOC"-3-r— sfifO
a^ •••••«*.•••.**,*•••.•••*
o —•
ae ^
u.
i—
— cc
00 i—
o oo ••••••••••••••
E oo «-'ccn«-(^-(^Hr'jc\Jocrr-\£fO*-<\rcc'i^r-
O Z <-< *-( r-*
u «
s
o
c:s «-tc\ifO^-j-sir-a:c'Or-iCgfoc-irOr^cc
3 i-^ rH--«i-^^^«—(r^^^^^^H
o o
D-7
-------
. _l
>
00
m co
t/j (-)
oo t-
O o oo ooc-ooor-ro
- "v
O 00
O C.
_j
-------
00
ffi 00
OOOOCifOP^— (L
i—i a;
LJUJ
X >—
Ooi
Q£
oco
>- •
_J t—
a; oo
3
OiD
Xcv
E
O
u
O Q
x :r
D-9
-------
CD 00
<:<[ •a-ir»mintnu"'iO'3''C-ro«-fOfOK>'3''C'.3-
i— -i
co o
W *-
Z X
O OH LnOOOOC1t*C;.-t^i/-)r«.CCCJOSc3OC5aOOOO
M >—t Q£ •tf-ooooooC'NCNO>'tr(^(yNCC)CToooooooo
I— UJLJ • ••••*••*•«•••••«•••••••
jeNJ'-|T-'--<'H'-t—*>-i'-i'-<^-if-1
t/) x£ -^)^-('-'^Hr-4^-t^-iojK)fl-invaf*-ccecti;occocococccc«co
oo E
=:
<
a:
>- v,
O V)
O CL
-J< aOfO^^K)rOfOKicoa;^'ir)^oj(\if^.r--'iOfOcococCtf'«'
Oir z. ^ •a-cr'0Nc^crc'tro'c-'C'Cj«-«f\;ccG;*-. r''>C'cri-=r-c-'—• x. t— LJ \&^a\Q\Q\£\£\£<*D<^\£\£\ct\Q'*c\Qr~*r*'r+-r*'r**f^r^t*-t*-
t— — CM C CSjCUC\JCsJC^CVJC\jOJCNjC\JC\JOJC\;OJ(\JC\JC\JC\jCNJC\JCVjC\JC\(CNJ
LJ O •-!
S -J x.
O
>- • f»-
_J t—
(K (^ Q
— a.
om o^ «-occ-tT>oso*r»r*aJc\js£^-vOfVt-i'-"-*'c\j«
z c\, oo o^csiC^U"Cr-ir-0sC*Cvf1^C*lcS'—»
O *Z CMC\J-H«-»<-H^-tr-''-<«-tryCM
o —.
3
^3
a: ^ f-HrjrO'S'u-'s^r-ccchOf-'C.fOfl-ifi^or^xo'orHrji.-;^-
^••^ »^i^i-^i-H«^*-»i-^^H^^i-^C\jC\JCSjCNjC\J
O O
I 3
D-10
-------
»- a:
LJUJ
2:
«r
a:
o oo
o a.
as in
2:
OlO
""
IT
o
01
o
E
O
QO — <•— <
c
r
D-ll
-------
1/1
CO V>
l/> O
co *-
2 I
o o co
< X »—
^- t-j
oo xs
to s:
>• X.
cr s/3
o a
Oa: a. i
oc —< s:
o -«
_i -x
<£
a: oo a
=> a.
Oin (/}
O
01
ce
t—
O
(_)
C Q
Z 2
D-12
-------
UJ 2
CO O
I I •-• I I
CO*-* •••••••••••••••••*
— CM ^H —t —i OJ ^
•— -^ C\i CM rvj rO -«
o: 2
OO —H^^"™^'"^«~t^^'™'—^"~*»^c\ic\ic\jr\jc\j
D-13
-------
c\<-. •••*•••••*••••••••
«-H •—t
G *•* *••••••••••••••••*
i"* O CO tO tO lO CU fO v£ C" T** fO ^ ^H fO CM C3 ff* h~ O i
01 O
*t ^
CO H»
- -
JD
1/3
CL <**
<; »-;
Q;* ?O^* »***•••••*•*•*••••••••••
cr s: r- it
o v. *•
LJ_ O
o oo
s:-
-J ^
UO fO^ ..**••*..*•**
~ ^. II
,-» CMCNJCNJ^H—i CM •-« ^-* i I i i i if* •*• to I i •«••«• -a- ro i i
o
a: 2 -. *. —
O O ^H i-^ *^ I—1 i-« i—4 »•! r-4 ^» ^H CNJ CNJ CM f\J CM
D-14
-------
C\J CSJ CD !
« o
C2 H-
*— c-:
«a
O >-
o_
00
CL ~
a: ^
O N»
u. o
C\l CM rH -H ,-H
t I t I I I t I I I CM
_J UJ
LJ ce
Q D
LJ c:
_1 3
U_ C
o a
D-15
-------
Ui 2
CO C
a. —
«l K)
o >,
U. CJ
O
hM K^
CvJ
a. < «
= ce.
O
O
o m
co 0-
Q 3
O «
E< -r— .
I . I OK;
Qs:-^
LJ **
I— >-
O _l
u- c:
_i o
UJO KJ-^
a: s
O —«
CO
< « ~ i i i i i
Q >• —
D-16
-------
Ox*. ••••••••••••••••••••••••
r-t <-t fO C\J I
UJ ^ *-
CO C
X t-
<£H- -3- -^ • •*•.•••••••••»•»•••••••
G >• —
C
(/5
C_ ~
< fj
ce * KI ~
* *-» r-
tz: E ^
Ox. v
^—
-
UO fO« •••••••••••
t-f i «—
-------
c\i i i i t i i i CM csj
^* m ro i i
— ro
UJ "Z.
C-0 O
csr* 10 -*
OX.
u. o
fOCMOdt-"
,-t ,-t CM —.
o
»-. in
h-O
< o
D-18
-------
" if"1 ^ ^ O
•I 1 I I I I C\J O \O C\J .-i t
ro CM i-i rj rv
o <••
CM -H
co o
«r h-
t— to
<
o >-
OJ
a. "*
< fO
a: *
sfl o >—• •—<
x cc i i
CO
2
O
, LJ ^O O ^n
_J c\
CO
i tr>
-------
o —
C\J 1-1
CD H-
X PO -C- 0s -3- >-» CT 03 if) 00 ^i
i v2 4* CVJ <-) I 1 I I I
c: t\jrofNji—oirod—"
Z LJ
C <_) r
^ z
t- O
< u
o
_J CV
<£ O
> V)
LU
O
—I LJ
LJ o:
0 13
O CO
s: <
LJ
OS.
om oa*rofO^-f-»roa5aor*o-Hor*»O(j»cvjoa\
-------
r-r— .—* c^ *-H r- •
fOt-'io cc J3 in ^ cc cc \£
o *-.
C\J r-t
2
O
i i error- i
CC
CO
a —
«l I/) cc *^
c z o o
O -H
''—t ««H O* rO *• fO (M
2 ce c\j
;r v. o —
o i_> t~ ^
1— O
< u
o m ^
LJ
O
-J UJ
uj a:
a D
UJ O
CO X
D-21
-------
f\J «
!/) O
«s <—
CD t-
C\J OJ 1-1
i i cc co ^H ^H i i i
inr-cc--
-------
~ -H I I I I I I I I I I I I 1 I I I I I I I I I I
^ O ^» •••••••t*»»t««»**l»«««*«
—> ^ I I 1 I I I
ri r-i -3-1-ro m ^ ro cc r* K} 1-1 KJ i— ro II
V) O
ca i-
c ^ —
h-
d z ojs Q^CC^~*^^^CCCCOCQ^^"^— ^•f^cccr. ccccccccf^r^c*cot^
a >-i i-i «-i i c\j i i
D I— K5
2 Ct CM
I- O
_i ru
UJ «-i ^Cjmf^infl-«-^H CMC\JC\JC\JC\JC^r^-3-(\Jlt
c —
_ UJ
U 0^
O 3
o c/o
h— >•
U-O K>-» •••••••••••••«••••«•••••
X Z
UJ
D-23
-------
TECHNICAL REPORT DATA
(Please read instructions on the reverse before completing)
1. REPORT NO.
3. RECIPIENT'S ACCESSION NO.
EPA 450/4-83-020
J_
4. TITLE AND SUBTITLE
Evaluation of Urban Air Quality Simulation Models
5. REPORT DATE
July 1983
6. PERFORMING ORGANIZATION CODE
7. AUTHOR(S)
8. PERFORMING ORGANIZATION REPORT NO.
Richard Londergan, David Minott, David Wackter,
Roderick Fizz
9. PERFORMING ORGANIZATION NAME AND ADDRESS
TRC Environmental Consultants, Inc.
800 Connecticut Boulevard
East Hartford, CT 06108
10. PROGRAM ELEMENT NO.
11. CONTRACT/GRANT NO.
68-02-3514
12. SPONSORING AGENCY NAME AND ADDRESS
U.S. EPA
OAQPS, MDAD, SRAB (MD-14)
Research Triangle Park, NC 27711
13. TYPE OF REPORT AND PERIOD COVERED
. SPONSORING AGENCY CODE
15. SUPPLEMENTARY NOTES
William M. Cox, Project Officer
16. ABSTRACT
This report summarizes the results of a comprehensive evaluation of "urban"
air quality simulation models using SOp and meteorological data collected as part
of the St. Louis RAPS study. The report contains numerous tabulations of each
model's performance in terms of statistical measures of performance recommended by
the American Meteorological Society.
The purpose of the report is two-fold. First, it serves to document for the
models considered, and similar models, their relative performance. Second, it
provides the basis for a peer scientific review of the models. To stay within the
spirit of this latter purpose, the report is limited to a factual presentation of
information and performance statistics. No attempt is made to interpret the sta-
tistics or to provide direction to the reader, lest reviewers might be biased.
17.
KEY WORDS AND DOCUMENT ANALYSIS
DESCRIPTORS
b.lDENTIFIERS/OPEN ENDED TERMS
c. COSATI Field/Group
Air Pollution
Mathematical modeling
Meteorology
Sulfur Dioxide
Statistical Measures
Performance Evaluation
St. Louis RAPS Study
Air Quality Impact
Assessment
18. DISTRIBUTION STATEMENT
Release to public
19. SECURITY CLASS (This Report)
Unclassified
21. NO. OF PAGES
300
20. SECURITY CLASS (Thispage)
Unclassified
22. PRICE
EPA Form 2220—1 (R*v. 4-77) PREVIOUS EDITION is OBSOLETE
------- |