United States
Environmental Protection
Agency
Atmospheric Sciences
Research Laboratory
Research Triangle Park NC 27711
Research and Development
EPA/600/S3-87/022 Aug. 1987
&ERA Project Summary
Analyses of PEM-2 Model
Evaluation Results for Short-Term
Urban Paniculate Matter
James M. Godowitch
The Pollution Episodic Model Version
2 (PEM-2), an urban dispersion model,
has been evaluated with measurements
from the 1982 Philadelphia Aerosol
Field Study data base in order to in-
vestigate its ability to model 12-hour
average concentrations of participate
matter less than 10 micrometers (PM10).
Modeled fine and coarse paniculate
total masses were combined and then
statistically evaluated against cor-
responding PM10 measurements at six
monitoring sites for a 29-day experi-
mental period.
Model performance was determined
from statistical measures of difference
and correlation between observed and
modeled concentrations paired in time
and location. The regional background
dominated many of the evaluation
statistics since it contributed about 70%
to the measured urban PM10 con-
centrations.
Concentration estimates from PEM-
2 and the RAM model are compared
from independent evaluations with this
same data base. Mean and high-five
PM10 concentrations from the PEM-2
model were about 25% lower than RAM
predictions at four sites within the city
limits. Differences in model results are
attributed to particulate removal by dry
deposition and settling processes in
PEM-2 and the different treatments of
area source emissions by these models.
Results of statistical measures were still
quite similiar for both models. Due to the
dominant role of regional background
concentrations in this evaluation study,
it was not possible to conclude which
model performed more accurately.
This Project Summary was developed
by EPA's Atmospheric Sciences Re-
search Laboratory, Research Triangle
Park, NC, to announce key findings of
the research protect that Is fully docu-
mented In a separate report of the same
title (see Project Report ordering In-
formation at back).
Introduction
The proposed National Ambient Air
Quality Standards (NAAQS) regulations
will establish a 24-hour standard for par-
ticulate matter in the size range less than
10 micrometers (PM10). Once this short-
term standard is promulgated, state and
local regulatory agencies will be required
to develop implementation plans to attain
and maintain the new standards. Air
quality dispersion models are expected to
be relied upon for urban regulatory ap-
plications and emission control strategies
contained in the state implementation
plans. In support of the Agency's policy to
regulate urban particulates, the Atmo-
spheric Sciences Research Laboratory has
sponsored the development and evalua-
tion effort of the Pollution Episodic Model
Version 2 (PEM-2). The PEM-2 is an
urban-scale Gaussian plume diffusion-
deposition model which has been
designed to compute short-term (up to
24-hours) ground-level concentrations of
one or two species of particulate or
gaseous pollutants. The model accounts
for the transport, dispersion, and the
deposition and settling processes of
particulates from multiple point and/or
area emission sources.
This report contains the results of
various analyses to evaluate the ability of
-------
PEM-2 to model PM,0 concentrations.
The model results for fine and coarse
participates were aggregated and statis-
tically evaluated against corresponding
PM10 measurements derived from ob-
served fine and coarse particulate con-
centrations from the Philadelphia data
base. Statistical measures of ^difference
and correlation for modeled and measured
concentrations paired in time and space
were used to examine model perform-
ance. In addition, comparative results from
PEM-2 and the RAM are presented.
Model Evaluation Data Base
All the measurements necessary to
perform a model evaluation for urban
particulates were obtained during the
Philadelphia Aerosol Field Study (PAFS).
The PAFS field program was conducted
during an intensive 31-day period from
14 July to 14 August 1982 in the
metropolitan region of Philadelphia,
Pennsylvania.
A comprehensive inventory of fine (FP)
and coarse (CP) particulate total mass
emissions was developed specifically for
the experimental period. Some real-time
sampling was performed and an effort
was made to determine whether im-
portant sources were operating con-
tinuously or off-line at times during the
study period Nevertheless, it is acknow-
ledged that the hourly emissions were
primarily derived from long-term values
and should not be construed to represent
actual emissions measurements. The
components of the inventory included
300 major point sources, 289 area
sources, gridded mobile sources, gridded
minor point sources, and 25 sources of
Industrial Process Fugitive Particulate
Emissions (IPFPE) The area, gridded
mobile, and minor point source emissions
were combined into a single file for input
to the model because they were con-
structed on the same 17x17 grid with
individual cell sizes of 2.5 km on a side.
The IPFPE sources were also incorporated
into this emissions grid where each
existed, although their grid cell sizes were
either 0.2 km or 0.5 km.
There were six PAFS monitoring sites
equipped with dichotomous filter sam-
plers which provided continuous FP and
CP measurements during the 31 -day ex-
perimental study. All 6 sites were located
within the urban emissions region and 4
sites were situated inside the city limits
The FP and CP total mass measurements
consisted of 12-hour average concentra-
tions. The two averaging periods of
0600-1800 EOT and 1800-0600 EOT a re
essentially representative of daytime and
nighttime conditions, respectively.
The model requires hourly values of
wind speed and direction, temperature,
mixing height, and stability class. Hourly
surface observations made by the
National Weather Service at the Philadel-
phia International Airport (PHL) located in
the southwest section of the city were
obtained for the model evaluation. High
resolution upper air temperature and
relative humidity measurements up to
2000 m were obtained by an airsonde
system which was launched three times
daily at 0400, 1000, and 1600 EOT from
16 July through 14 August. The height of
the lowest elevated temperature inver-
sion base defined the mixing height. The
observed morning and afternoon mixing
heights, and hourly surface observations
were input to the RAMMET meteorological
processor in order to derive hourly values
of mixing height and stability class re-
quired by PEM-2.
Model Evaluation Results
The technical features and options
chosen for the model runs included: urban
wind profile exponents, stack-tip down-
wash, new plume rise/penetration
methods, and a constant height of 10 m
was specified for all area sources. The
deposition (Vd) and gravitational settling
(W) velocities for the particulate species
are given below.
Dry Deposition and Settling Velocities
for the PEM-2 Evaluation Runs
Paniculate Vd (cm/s) W (cm/s)
Flange Day Night Day Night
Fine (FP)
Coarse (CP)
O.2
0.5
0.1
0.5
O.O
0.25
0.0
0.25
There is uncertainty associated with
these values since there is a lack of
experimental deposition measurements
in urban areas. Nevertheless, these
estimates provide for differences between
the time periods and for different size
ranges.
PEM-2 was executed to compute hourly
concentrations of FP and CP due to hourly
emissions and hourly meteorological pa-
rameters for the 29-day period from 16
July through 14 August 1982.
The determination of the regional
particulate component is an important
factor of urban PM10 modeling because it
may be a relatively large fraction of the
total measured concentration at urban
receptor sites. The regional background
measurement must be representative of
the incoming concentration into an urban
domain. Hence, it should be measured at
an upwind site or suitable 'remote' loca-
tion that is not impacted by nearby
sources or influenced by emissions from
the urban area being modeled. Unfortu-
nately the PAFS sites, were all located
inside the Philadelphia emissions area.
The hourly surface wind observations
were averaged over 12-hour intervals
corresponding to the time period of the
measured concentrations in order to
determine which site was upwind of the
city. In an attempt to account for these
contributions of urban emissions at the
upwind site, the regional background
(PMb) was determined by subtracting the
predicted concentration (P.) from the
observed concentration at the upwind
site (PMup) for each period.
Two separate sets of statistical results
were determined in this model evaluation.
For Set 1, observations (O,) were com-
pared to the corresponding sum of model
predictions and background (i.e. P, + PMb).
For Set 2, results were obtained for model
predictions (P,) and corresponding re-
sultant observed values derived by sub-
tracting the background for each period
from the measured concentration (i.e. 0,
- PMb).
In the analysis where PMb was included
with predictions, paired values from the
upwind site were omitted in the sample
data set. Although this criterion reduced
the sample size, the statistical measures
would be artifically improved if upwind
values were included due to the method
of determining PMb. PEM-2 slightly
underestimated PM10 concentrations with
an overall positive bias (d) of 5.3 /ug/m3.
The mean modeled concentration was
43.9 Mg/m3. A value of 0.96±0.3 for the
ratio of P/O was favorable. However,
measures of correlation from linear
regression analysis departed from desired
values. A relatively large intercept (A) of
23.7, a slope (b) of only 0.41, and a
correlation coefficient (R) of 0.56 were
determined from all available paired con-
centrations. A few cases of high observed
concentrations being greatly underpre-
dicted had a definitive influence on the
linear correlation measures and also ef-
fected the difference measures, such as
bias.
Due to the large influence of PMb on
the previous evaluation statistics, the Set
2 of statistical results were obtained with
concentration pairs composed of model
predictions and resultant observed values
-------
minus the regional backgrounds. Several
of the statistical measures revealed poor
model performance as results differ
largely from the other set of values. Al-
though bias and absolute error were
nearly equivalent in both sets of results,
their magnitudes represented a much
greater fraction of the observed and
predicted means for these concentration
pairs. The relatively large absolute error,
near zero slope, and small R also indicate
considerable scatter and little correlation
between these paired concentrations. In
fact, the standard deviations were com-
parable to the mean resultant observed
values. These results also demonstrated
the powerful role of the regional par-
ticulate background in this evaluation
process.
Both PEM-2 and the RAM model are
Gaussian plume models with many of
the same technical features and options.
However, there are a few important dif-
ferences between these models, which
may yield variations in estimated con-
centrations. A relevant factor is that
PEM-2 accounts for dry deposition and
gravitational settling, while these pro-
cesses cannot be considered by the RAM
model. Another difference between these
models is in the treatment of area emis-
sions and their source heights. PEM-2
computes contributions from up to eight
upwind area cells to the concentration at
a given receptor, while RAM can consider
the impact at a particular receptor from
all upwind area sources. Additionally, a
single area source height can be specified
in the PEM-2 runs. In contrast, RAM
allows for input of different area source
heights. Both models were executed with
the same emissions data base, and hourly
meteorological parameters, although
there may be small differences in the
mixing height since values from a nearby
rawinsonde site were used instead of the
PAFS airsonde site data.
A single area source height of 10 m
was input in the PEM-2 model runs. In
contrast, one of three possible source
heights was specified for area sources in
RAM; namely, 13.7 m, 9.1 m, or 4.6 m,
were assigned to area sources based on
emission rate. The grid size for both
models was 2.5 km. This means the
contributions from area sources beyond
20 km from a site were not considered by
PEM-2 in the concentration calculations
for a given site. All area sources upwind
of a site were included in RAM.
The RAM evaluation results were taken
from another report and its background
values were used with the PEM-2 model
results in this phase of the analysis to
make the direct comparison with RAM
possible. The common feature found from
both sets of statistics was that PEM-2
predictions were consistently lower than
RAM's results in the mean and peak
concentrations. In the set where modeled
predictions were considered alone, the
PEM-2 mean value of 10.1 /*g/m3 is 3.1
jig/m3 less, or about 75% of RAM's mean
concentration of 13.2 jig/m3. The model
differences described earlier were be-
lieved responsible for the different pre-
dictions. Modeled concentrations by
PEM-2 were reduced due to losses by
deposition processes. RAM's considera-
tion of more upwind areas sources and
lower source heights for some area
sources, particularly those grid cells out-
side the city, compared to a uniform 10m
height for PEM-2 also contributed to
higher concentrations since PEM-2 com-
putations are limited to eight upwind
area grids.
The relative similarity in results for
both models made it difficult to state
which model performed better or more
accurately. While the correlation mea-
sures for PEM-2 are slightly better than
the RAM results, the large positive biases
revealed both models greatly underpre-
dicted observed values.
An interesting feature explored was
the difference in model predictions for
the various sites. A revealing result when
comparing these model predictions was
the ratio of this difference to the RAM
prediction (i.e. PRAM. PPEM / PRAM) at each
site. Interestingly, the percentage of this
ratio was about 20% for the central urban
site and at the 3 sites closet to it, but
results jumped to 35% and 40% at two
outlying sites. These latter two sites were
most distant from the central downtown
area. It appeared that differences between
the models were accentuated when ap-
proaching the boundary of the model
domain, which in this case was farthest
from the city and larger emissions.
Of particular relevance in regulatory
applications is how a model simulates
the highest concentrations since these
are the values which may exceed a pol-
lutant standard and provide the basis for
the design concentration upon which a
control strategy is implemented. Results
for high-five concentrations were similar
to those obtained from the mean con-
centration results. PEM-2 predictions
were almost always lower than RAM's
results. The values of the concentration
ratio PEM-2/RAM were lowest at the 2
outermost sites. Both models significantly
underpredicted the peak observed con-
centrations even when the regional
background values were used.
Large positive biases in the comparison
of PEM-2 and RAM results may be ex-
plained by underestimated regional back-
grounds used in the RAM evaluation. The
strong similarity in the statistical results
revealed little evidence to distinguish
between the performance of these
models. Model accuracy was also difficult
to assess from the evaluation statistics
due to the dominant role of the regional
background component.
The EPA author J. M. Godowitch (also the EPA Project Officer, see below)
is with the Atmospheric Sciences Research Laboratory, Research Triangle
Park, NC 27711, and is on assignment from the National Oceanic and
Atmospheric Administration. U.S. Department of Commerce
The complete report entitled "Analyses of PEM-2 Model Evaluation Results
for Short-Term Urban Paniculate Matter," (Order No. PB 87-199 667/AS;
Cost: $13.95, subject to change) will be available only from:
National Technical Information Service
5285 Port Royal Road
Springfield, VA 22161
Telephone: 703-487-4650
The EPA Project Officer can be contacted at:
Atmospheric Sciences Research Laboratory
U.S. Environmental Protection Agency
Research Triangle Park, NC 27711
-------
United States Center for Environmental Research
Environmental Protection Information
Agency Cincinnati OH 45268
Official Business
Penalty for Private Use $300
EPA/600/S3-87/022
0000329 PS
U S EHVIR P«2I1STION *GENCY
------- |