United States
Environmental Protection
Agency
Environmental Sciences
Research Laboratory
Research Triangle Park NC 27711
Research and Development
EPA-600/S3-84-087 Sept. 1984
Project Summary
Evaluation of the Pollution
Episodic Model Using the
RAPS Data
William R. Pendergrass and K. Shankar Rao
The Pollution Episodic Model (PEM)
is an urban-scale model capable of pre-
dicting short-term average ground-
level concentrations and deposition
fluxes of one or two gaseous or particu-
late pollutants at multiple receptors.
The two pollutants may be nonreac-
tive, or chemically coupled through a
first-order chemical transformation. Up
to 300 isolated point sources and 50
distributed area sources may be con-
sidered in the calculations. Concentra-
tion and deposition flux estimates are
made using the hourly mean me-
teorological data. Up to a maximum of
24 hourly scenarios of meteorology
may be included in an averaging
period.
This report describes an evaluation of
the PEM that used data from the St.
Louis Regional Air Pollution Study
(RAPS). This evaluation was designed
to test the performance of the model
by comparing its concentration esti-
mates to the measured air quality data
by using appropriate statistical meas-
ures. Twenty days, ten summer and ten
winter, were selected from the RAPS
data base for the PEM evaluation. The
model's performance was judged by
comparing the calculated 12-h average
concentrations with the corresponding
observed values for five pollutant
species: SO2, fine and course sulfates,
and fine and coarse total mass. A first-
order chemical transformation of SO2
to fine sulfate was considered in the
calculations in addition to the direct
emission and dry deposition of all five
pollutants.
For the twenty PEM evaluation days,
PEM predicted average concentrations
of SO2, and fine and coarse sulfates to
within a factor of two. The model over-
predicted the average concentrations
of fine and coarse total mass by a factor
of three to four over the evaluation
period. This was attributed primarily to
overestimation of emission rates and
incorrect location of area sources,
which dominate the fine and coarse
total mass emission. Other possible
sources of errors in the calculations are
listed and discussed.
This Project Summary was de-
veloped by EPA's Environmental Sci-
ences Research Laboratory, Research
Triangle Park, NC, to announce key
findings of the research project that is
fully documented in a separate report
of the same title (see Project Report or-
dering information at back).
Introduction
The Pollution Episodic Model (PEM)
is an urban-scale model capable of pre-
dicting short-term ground-level concen-
trations and deposition fluxes of one or
two gaseous or particulate reactive pol-
lutants in an urban environment with
multiple point and area sources. It is in-
tended for studies of the atmospheric
transport, transformation, and deposi-
tion of acidic, toxic, and other pollutants
in urban areas to assess the impact of
existing or new sources or source mod-
ifications on air quality, and for urban
planning. The effects of dry deposition,
sedimentation, and a first-order chemi-
cal transformation are explicitly ac-
counted for in PEM.
This report describes an evaluation of
the PEM that uses the St. Louis Reg-
-------
ional Air Pollution Study (RAPS) data.
This evaluation was designed to test the
model's performance by comparing its
concentration estimates to the mea-
sured air quality data by using several
standard statistical measures of perfor-
mance.
Twenty days, ten summer and ten
winter, were selected from the RAPS
data base for the PEM evaluation. The
model's performance was judged by
comparing the calculated average con-
centrations with the corresponding ob-
served values for the following five pol-
lutant species: SO2, fine sulfate, coarse
sulfate, fine total mass, and coarse total
mass. In these species, the cut-off size
between fine and coarse particle frac-
tions was 2.5 |j.m. A first-order chemical
transformation of SO2 to fine sulfate
was considered in the calculations in
addition to the direct emission and dry
deposition of all five pollutants.
Data Base
Twenty days, ten summer and ten
winter, were selected from the RAPS
data base by the U.S. Environmental
Protection Agency (EPA) for the PEM
evaluation. Detailed emission inven-
tories of the RAPS region and meteorol-
ogy and concentration measurements
corresponding to these evaluation days
were supplied by the EPA from the
RAPS data base.
Hourly area and point source emis-
sion inventories for a typical winter day
and a typical summer day for the St.
Louis metropolitan area were supplied
by the EPA on two magnetic tapes. The
emission inventories were supplied on
a numerical grid with a fixed origin at
XUTM = 710 km and YUTM = 4250 km,
which extended to 60 km in both x and
y directions. The size of each emission
grid cell for area sources was 5x5 km,
thus giving 144 emission squares in the
grid. Data tapes contained information
on SO2, sulfate, total paniculate mass
emissions, and the particle size derived
from the 1976 RAPS Emission Inven-
tory. An average conversion rate of
1.85% of S02 emissions was used to es-
timate the sulfate emission rates for
both area and point sources from the
known information on SO2 emission,
provided particulate emissions existed.
in the case of point sources with no par-
ticulate emissions, but relatively large
S02 emissions, the sulfate emissions
were calculated on the assumption that
in a short period of time, the conversion
of SO2 to SO3 occurs and contributes to
2
the total mass in the region of interest.
However, in the case of area sources,
the SO2 emissions were relatively small
(3% of total SO2 emissions); therefore,
sulfate emissions could be neglected if
there were no associated particulate
emissions.
The size distributions of sulfate parti-
cle emissions from area and point
sources were more difficult to estimate.
Approximate base size distributions of
sulfate for typical winter and summer
days were determined. Based on other
studies, about 50% of the sulfate was
assigned to the size range less than 0.25
|o,m for the summer aerosol, and ap-
proximately 25% of the sulfate was as-
signed to this size range for winter
aerosol. The base size distributions ap-
proximated were then used to estimate
the sulfate size distributions in the area
and point sources by relating the total
particulate emissions (with associated
size spectrum) from these sources to
sulfate emissions.
Hourly measurements of wind
speeds and directions, and tempera-
tures from the 25 station Regional Air
Monitoring System (RAMS) network
were supplied by the EPA. Input files
contained urban mixing heights, wind
speeds and directions, atmospheric sta-
bility class, and temperatures: the input
winds were RAMS network resultant
winds. The stability classifications were
supplied in the format required by PEM
(i.e., stability classes 1-7).
Data tapes containing the observed
gas concentration values from the
RAMS network, corresponding to the
twenty evaluation were provided. Sepa-
rate files of high-volume and dichoto-
mous sampler data were also provided.
The data files were scanned for hourly
average SO2 concentrations, and 12-h
average concentrations of total mass
and total sulfur. The observed gaseous
total sulfur concentrations were used to
approximate SO2 concentrations at the
RAMS stations where the latter were
not measured. The high-volume and
dichotomous sampler data contained
total sulfur and total particulate mass
concentrations in |j,g/m3. The particu-
late data were further divided into fine
and coarse categories based on a cut-
off size of 2.5 jj,m. The total sulfur mea-
surements were multiplied by a factor
of three (the ratio of the molecular
weight of SO4 to the molecular weight
of sulfur) to obtain the equivalent total
sulfate concentrations.
Concentration measurements were
not made at all of the 25 RAMS stations.
The observed S02 concentrations are 1-
h average values. The total sulfur and
total mass concentrations measured by
eight out of the ten reporting RAMS sta-
tions were 12-h average values; only
stations 103 and 105 recorded 6-h aver-
ages. To facilitate comparison with the
model calculations, the observed con-
centrations of SO2, fine and coarse sul-
fates, and fine and coarse total mass
were converted into 12-h averages. This
procedure gave two (12-h average) ob-
served concentrations per day for each
of the five pollutants.
Model Evaluation Results
PEM concentration predictions were
evaluated against the measured con-
centrations for five pollutants: SO2, fine
particulate sulfate, coarse particulate
sulfate, fine particulate total mass, and
coarse particulate total mass. These five
quantities were calculated in three
model runs that used different sets of
input data. It was assumed that SO2
chemically transforms into fine sulfate
at a constant rate, and there is no con-
tribution to coarse sulfate concentra-
tions from this transformation.
A chemical transformation rate of
SO2 to fine particulate sulfate of 5% per
hour was used. This value was held
constant throughout the model runs re-
gardless of meteorological and other
conditions.
Deposition (Vd) and gravitational
settling (W) velocities were varied de-
pending on the pollutant in each model
run. For example, Vd was 2.0 cm/s and
W was assumed to be zero for SO2. No
attempt was made to vary them by the
atmospheric stability class or other
meteorological conditions.
PEM uses a fixed calculation and re-
ceptor grid system. A grid system was
designed such that PEM receptors
either matched or formed a grid around
the actual RAMS network stations. For
point comparisons with the RAMS net-
work stations, the four receptors in the
grid squares around the RAMS station
were summed and their average was
assigned to the RAMS station location.
The numbers of point sources in this
evaluation were 286 in winter and 275
in summer, thus nearly utilizing the
maximum capacity of the model of 300
point sources. For point source calcula-
tions in this evaluation, a modification
was made to the PEM program such
that concentrations were calculated
only for the receptors surrounding each
RAMS station and not at the rest of the
receptors. This required calculation o'
-------
only 84 out of a total of 2500 receptors.
The model performance was evaluated
by using several statistical measures.
Two general measures of performance
were used here: a) measures of differ-
ence, which included bias, variance,
gross variability or root mean squared
error (RMSE), and average absolute
gross error; b) measures of correlation
paired in space and time.
The ratio of the calculated (P) and ob-
served (0) means for SO2, P/O, was
1.24, and the ratio of the corresponding
standard deviations was 1.12. However,
the correlation coefficient was only 0.23
over the compared range (6.5 - 250 jj,/
m3) of concentrations. This suggests a
large degree of randomness in the indi-
vidual case-by-case comparisons of
SO2 concentrations. The differences D,
between observed and calculated SO2
concentrations showed a clear bias for
PEM to overpredict observed concen-
trations less than 75 |xg/m3 and under-
predict observed concentrations
greater than about 125 (xg/m3. The bias
D over the entire evaluated range of
SO2 concentrations was -12.8 |o,g/m3.
Thus, PEM was conservative, with a
tendency to slightly overpredict the av-
erage SO2 concentrations. The average
absolute gross error IDI was 48.5 |xg/
m3, which was less than the mean of
observed concentrations. Therefore, on
the average, PEM predictions are within
a factor of two of the observed SO2 con-
centrations.
For fine sulfate concentrations, P/O
was 1.1 and the ratio of the correspond-
ing standard deviations was 1.2. The
correlation coefficient was 0.41 over the
compared range (1 to 30 (xg/m3) of the
concentrations. The model tended to
overpredict O, < 18 ^g/m3 and under-
predict O, > 20 |j,g/m3. The bias D over
the entire range of concentrations was
-1.0 jxg/m3, (i.e., the model is slightly
conservative). The average absolute
gross error IDI was 4.8 jxg/m3, which
was much less than the mean of ob-
served concentrations. Therefore, aver-
aged over the entire data base, PEM cal-
culations of fine sulfate concentrations
were within a factor of two of the cor-
responding observed values.
Coarse sulfate (particle size ^ 3 ^m)
concentrations from the model evalua-
tion results for direct emissions of
sources only were low (generally less
than 3 |j,g/m3). The ratio of the means
of calculated and observed values of
concentrations, P/O, was 0.52, and the
ratio of the corresponding standard de-
viations was 0.9. The correlation coeffi-
cient was 0.38 over the compared range
of concentrations. The model slightly
underpredicted the concentrations,
with a bias of D = 0.5 n-g/m3 and an
average absolute gross error of 0.66 |xg/
m3. The latter was 59% of the mean of
observed concentrations. Thus, on av-
erage, the calculated coarse sulfate con-
centrations were within about a factor
of two of the corresponding observed
values. The model performed some-
what better in winter than in summer.
The results clearly showed that PEM
overpredicted fine total mass concen-
trations. The observed concentrations
were less than 80 |xg/m3, but the corres-
ponding calculated values ranged up to
300 |xg/m3. The larger calculated con-
centrations were generally associated
with weak diffusion conditions charac-
terized by strong stabilities, low wind
speeds, and shallow mixing depths that
were typical of several of the winter
evaluation days. The ratio of the means,
P/O, was 3.1 and the ratio of the corres-
ponding standard deviations was 6.0.
The model significantly overpredicted
the concentration, with a bias of D =
-70.8 jjig/m3 and an average absolute
gross error of 72.1 |o,g/m3, which was
2.1 times the mean of observed concen-
trations. The model overpredicted con-
centrations at stations within the city by
a factor of three or less, and accurately
modeled the two outlying stations.
The model evaluation results for
coarse total mass (particle size ^ 3 |^m)
concentrations were qualitatively simi-
lar to those obtained for fine total mass
evaluation.
Conclusions
This report described an evaluation of
PEM that used twenty days of data from
the St. Louis RAPS. This evaluation was
designed to test the model performance
by comparing its concentration esti-
mates for five pollutants to the mea-
sured air quality data by using appro-
priate statistical measures of perfor-
mance.
For the twenty evaluation days, PEM
predicted average concentrations of
S02 and fine and coarse sulfates to
within a factor of two. The model over-
predicted the average concentrations of
fine and coarse total mass by a factor of
three to four over the evaluation period.
The significant differences between the
calculated and observed total mass
concentrations may be attributed to a
number of reasons:
1. Hourly point and area source emis-
sion inventories were available for
only one winter day and one sum-
mer day. These inventories were
further averaged over two 12-h
periods per day for use as input to
PEM. Analysis of the emission inven-
tories indicated a core of steady
emission sources with various other
area and point sources coming on or
off line throughout the modeling
period. Running PEM on an hour-by-
hour basis might account for this
variability of emissions but the mod-
eling costs would be prohibitive. De-
spite this variability, both fine and
coarse sulfates were predicted to
within a factor of two for the total
means as well as across the 12-h av-
eraging period. However, the vari-
ability in emissions appears to be
very important for fine and coarse
total mass, because these emission
rates were significantly larger and
dominated by ground-level sources.
2. Point sources dominated the emis-
sions of SO2 and fine and coarse sul-
fates. Area sources dominated the
emissions of fine and coarse total
mass, and the sulfate components of
fine and coarse total mass emissions
from area sources were negligible
compared to the nonsulfate compo-
nents. The nonsulfate total mass
consisted of fugitive dust, highway,
residential, commercial, industrial,
and other particulate emissions of
different sizes that were difficult to
estimate accurately. No information
is available on the variability of these
emissions. Any errors involved m
the estimation and location of these
sources would significantly affect
the calculated concentrations due to
the relatively large emissions from
area sources.
3. Because of the 12-h averaging for
periods 00-12 and 13-24 h, little can
be said about the diurnal variation
of model performance in this evalua-
tion. There were also significant dif-
ferences between the first and sec-
ond averaging periods in the mean
residuals of fine and coarse total
mass. This may have been as-
sociated with the diurnal variability
of area source emissions of these
species and with errors in stability
classification. The first 12-h averag-
ing period was generally charac-
terized by stable conditions, with
weak diffusion conditions. Hence,
the calculated concentrations and re-
siduals were larger for this period.
-------
4. Constant deposition and settling ve-
locities and transformation rates
were used throughout the 12-h av-
eraging period. This ignores the de-
pendence of these variables on
meteorological conditions such as
wind, humidity, and thermal stratifi-
cation. Also, using one constant set
of values for deposition and settling
velocities to describe the broad parti-
cle size spectrum ^ 3 i^m may not
accurately represent the behavior of
particles of different sizes.
5. The wind speed and direction input
to PEM were the RAMS network re-
sultant values These are approxima-
tions to real conditions. Errors in
wind direction may cause the model
to affect particular receptors that
may be completely ignored in real-
ity. An underestimation of the actual
wind speed leads to overprediction
of the calculated concentrations.
Additional effort should be directed
toward an examination of the model's
response to emission variability, stabil-
ity classification, and area source emis-
sions and location. Experience has
shown that area sources are the pri-
mary determinant in modeling urban
ground-level concentrations of nonsul-
fate particulate matter.
William R. PendergrassandK. ShankarRaoare with Atmospheric Turbulence and
Diffusion Division, National Oceanic and Atmospheric Administration, Oak
Ridge, JN 37830
Jack H. Shreffler is the EPA Project Officer (see below)
The complete report, entitled "Evaluation of the Pollution Episodic Model Using
the RAPS Data." (Order No PB 84-232 537; Cost. $10.00, subject to change)
will be available only from.
National Technical Information Service
5285 Port Royal Road
Springfield, VA22161
Telephone. 703-487-4650
The EPA Project Officer can be contacted at.
Environmental Sciences Research Laboratory
U.S. Environmental Protection Agency
Research Triangle Park, NC 27711
- U S GOVERNMENT PRINTING OFFICE, 1984 -- 759-015 '7813
United States
Environmental Protection
Agency
Center for Environmental Research
Information
Cincinnati OH 45268
Official Business
Penalty for Private Use $300
- i
•-! 'v - o I ; < t- (• I
"5 i) O il iJ
------- |