United States
Environmental Protection
Agency
Environmental Sciences Research «•
Laboratory
Research Triangle Park NC 27711
Research and Development
EPA-600/S3-82-094 Mar. 1983
f/EPA Project Summary
Final Evaluation of Urban-Scale
Photochemical Air Quality
Simulation Models
Kenneth L Schere and Jack H. Shreffler
The research study discussed here is
a continuation of previous work whose
goal was to determine the accuracy of
several selected urban photochemical
air quality simulation models using
data from the Regional Air Pollution
Study in St Louis. This work reports
on the testing of three models with a
sample size of 20 days. The models
evaluated here are: The Photochemical
Box Model (PBM), The Lagrangian
Photochemical Model (LPM), and the
Urban Airshed Model (UAM). Emphasis
in this report is directed at the ability of
the models to reproduce the maximum
1 -hour ozone concentrations observed
on 20 days selected from nearly 2 years
of data. The PBM, LPM, and UAM have
been evaluated using statistical meth-
ods and graphical techniques and all
show potential as air quality manage-
ment tools. The standard deviation of
the differences between observed ozone
maxima and predicted concentrations
at the same place and time ranged from
0.04 to 0.06 ppm for maxima of 0.16
to 0.26 ppm. This measure of uncer-
tainty should be recognized by decision-
makers using these models in regula-
tory and planning processes.
This Project Summary was developed
by EPA's Environmental Sciences Re-
search Laboratory. Research Triangle
Park. NC, to announce key findings of
the research project that is fully doc-
umented in a separate report of the
same title (see Project Report ordering
information at back).
Introduction
The Regional Air Pollution Study (RAPS)
was conducted in the St Louis region over
the period 1974-1977. RAPS was de-
signed to provide a comprehensive data
set for the testing and evaluation of numer-
ical air quality simulation models on an
urban scale. While the RAPS field meas-
urements were in progress EPA surveyed
the available, state-of-the-art, photo-
chemical air quality simulation models,
and selected three for evaluation. In
addition, a simple box-model approach
was constructed by EPA and included in
the study.
The following models were investigated
in the evaluation program.
Photochemical Box Model (PBM) - a
single cell Eulerian model constructed
by EPA.
Lagrangian Photochemical Model (LPM)-
a multi-level parcel model developed by
Environmental Research and Technology,
Inc.
Livermore Regional Air Quality Model
(LIRAQ) - a single-level Eulerian grid
model developed by Lawrence Livermore
Laboratory.
Urban Airshed Model (UAM) - a multi-
level, Eulerian grid model developed by
Systems Applications, Inc.
LI RAQ was dropped from the evaluation
because of a series of technical and logis-
tics problems detailed in our first evaluation
report.* LIRAQ was unable to produce
significant ozone levels for St Louis, but
no specific correctable error could be iden-
tified.
The remaining models were tested ex-
tensively and corrections of obvious errors
'Shreffler, J.H. and K.L Schere, 1982: Evaluation of
four urban-scale photochemical air quality simula-
tion models. EPA Report, U.S. Environmental Protec-
tion Agency, Research Triangle Park, NC (in press).
-------
or deficiencies were made. No effort was
made to adjust or tune the model predic-
tions to observed concentration values.
The prevailing philosophy behind the
evaluation effort was to use the models in
an off-the-shelf mode, much as they even-
tually would be applied by a user in a
regulatory situation. However, great care
was taken in preparation of data sets and
model executions. Although model as-
sumptions vary, effort was made to use
data in similar manners in all models. Data
preparation and actual execution of the
models was accomplished solely by the
authors at EPA The goal of the evaluation
was to provide a fair and objective deter-
mination of the accuracy of a set of photo-
chemical models when tested in an opera-
tional mode against a comprehensive ur-
ban data base.
The only reasonable method of deter-
mining accuracy is to test a model against
an extensive observational data base. Pro-
viding such a data base was the purpose
behind the RAPS. Since the level of the
ozone maximum for each day is of para-
mount importance relative to the National
Ambient Air Quality Standards the com-
parison between its observed value and
the model prediction at the same time and
place would be of central interest. The
method of evaluation consists of selecting
a set of test days, executing model simula-
tions, and computing residual concentra-
tions (observed-predicted). The specific
outcome of the evaluation is a presentation
of information on residuals under the
given circumstances. Conclusions about
model acceptability require further as-
sumptions and judgements.
As an integral part of the RAPS a net-
work of 25 surface stations was estab-
lished in and around St. Louis and com-
prised the Regional Air Monitoring System
(RAMS). The RAMS stations continually
monitored various meteorological vari-
ables as well as ambient concentrations of
pollutant gases. In addition to the RAMS,
upper air balloons were released each hour
from urban and rural sites to provide wind
profiles for modeling purposes.
This report presents the results of model
simulations for 20 individual days chosen
from the RAPS data base, twice the number
from the initial evaluation detailed in the
first evaluation report These days, eleven
from 1975 and nine from 1966, account
for some of the higher 03 measurements
observed in the RAPS surface monitoring
network. Maximum hour-average single
station 03 values all exceeded 0.16 ppm
on the 20 days.
Results and Discussion
The Photochemical Box Model (PBM), a
single cell Eulerian air quality simulation
model, simulates the transport and chem-
ical transformation of air pollutants in
smog-prone urban atmospheres. The
model's domain is set in a variable volume,
well-mixed reacting cell where the physical
and chemical processes responsible for
the generation of ozone (Os) by its hydro-
carbon (HC) and oxides of nitrogen (NOX)
precursors are mathematically created. To
apply the model to the St. Louis RAPS data
base the horizontal scale of the single cell
was 20 km and the vertical scale was time-
varying, proportional to the depth of the
mixed layer. The model domain was
centered on downtown St. Louis such that
the 20x20 km area encompassed most of
the major emissions sources on either side
of the Mississippi River. A uniform distri-
bution of source emissions was assumed
across the surface of the cell. Twelve of
the RAMS surface monitoring stations
were located within the cell's boundaries
and nine others were available for deter-
mining model boundary concentrations.
The following statistics summarize the
differences between the observed hourly
maximum concentration of Os for each
day and the PBM prediction at the same
time. The concentrations are averages
over the PBM domain.
AC
AC"
s.d.(AC)
Obs - Pred (ppm)
-0.012
0.039
0.029
Evidence shows that the PBM is a useful
tool in assessing the urban air quality for
photochemically reactive pollutants, es-
pecially in stagnation conditions. The
PBM is relatively simple to use and its data
requirements are far less stringent than
most other numerical air quality simulation
models. The areas of further study that
should be pursued include: (a) the relation-
ship between the average Os concentration
observed within the model domain and the
maximum Os level observed at a single
station, (b) the continued testing and
refinement of the chemical kinetic mech-
anism within the PBM, and (c) the
sensitivity of the model to variations in
selected parameters such as initial and
boundary concentrations, initial cell depth,
emissions, wind speed, and solar radiation.
The Lagrangian Photochemical Model
(LPM) was developed by Environmental
Research and Technology, Inc. and a-
dapted under contract with EPA for use
with the RAPS data base. The LPM
considers a portion of the atmosphere as
an identifiable parcel which can be tracked
from early morning to the late afternoon.
As the parcel moves over the various
emissions sources, pollutants are assimi-
lated, vertically mixed, and subjected to
photochemical reactions in the presense
of solar radiation. The model has recently
been modified for this study to include a
Gaussian-type lateral spread of the parcel.
The following statistics summarize the
differences between the observed hourly
maximum Os concentration and the LPM
predictions at the first vertical model level
for the same time and place.
AC
AC
s.d.(AC)
JACf
Obs - Pred (ppm)
-0.023
0.058
0.052
Compared to the results presented in
the first evaluation report, the LPM has
shown considerable improvement in its
predictive capabilities resulting from changes
made to allow parcel expansion and in the
method of initialization. The model shows
promise as an effective tool to understand
Os production in an urban region. It is
relatively easy to use, inexpensive to ex-
ecute, and seems immune to various ex-
ecution errors which tend to arise unex-
pectedly in complex computations of this
sort. An area of further research may
include a study of the vertical diffusivity as
it relates to unrealistic build-up of pollu-
tants at ground level.
The Urban Airshed Model (DAM) is a 3-
D grid-type photochemical air quality sim-
ulation model developed by Systems Ap-
plications, Inc. (SAI) of San Rafael, Califor-
nia. The model structure consists of a
lattice array of cells, arranged so that the
total volume represents an urban domain
and in which the physical and chemical
processes responsible for photochemical
smog are mathematically simulated. The
horizontal dimensions of each cell are
constant but the heights of the cells vary
throughout a model simulation according
to changes in the depth of the mixed layer.
To apply the model to the St. Louis RAPS
data base, the area modeled was 60 km
wide and 80 km long. Each individual cell
was 4 km on a horizontal side. Vertically,
there were four layers of cells in total; the
bottom two layers simulated the mixed
layer and the top two represented the
region immediately above the mixed layer.
The domain of the UAM was centered just
west of downtown St. Louis and included
the entire metropolitan area. Of the 25
RAMS monitoring stations, 21 are situated
within the model domain area and the
-------
remaining 4 are available for determining
boundary concentrations.
The following statistics summarize the
differences between the observed hourly
maximum Oa concentration and the UAM
prediction at the same time and place.
AC =
AT =
s.d.(AC) =
Obs - Pred (ppm)
-0.062
0.035
0.062
The evaluation of the UAM by the Mete-
orology Division of EPA is now complete.
It is clear that the potential use of a grid-
type model such as the UAM is great,
although the complexity of the model
often makes solving the problems which
arise more difficult. This model is a
powerful tool available for use by the air
quality analyst already experienced in
working with complex simulation models.
Model results should be carefully studied
in each application and may not necessari-
ly be used in an absolute sense in any
given simulation. An area of future study
that may be pursued is model sensitivity to
variations in selected parameters such as
initial and boundary concentrations, emis-
sions, wind fields, and solar radiation.
Conclusions
The three models included in the final
analysis span a wide range in complexity
and sophistication but are all based on
numerical solutions to mass-conservative
equations. They are selected from the
general categories of box, trajectory
(Lagrangian), and grid (Eulerian) models.
Emphasis m the model performance eval-
uations is placed on ozone, although re-
sults for other pollutant species are also
discussed.
The PBM predictions for maximum Oa
for the average of the monitoring stations
within the model domain were generally
on the high side. The average Oa residual
showed a 23% overprediction over all test
days. However, for the 5 stagnation-type
days where the maximum observed 63
occurred within the PBM domain the aver-
age overprediction was 8%, considerably
better than for the entire sample. Only a
slight tendency towards overprediction
was indicated for the LPM. The biases of
the residuals were relatively small, 11 % of
the average observations at Level-1 and
only 2.5% at Level-3. The standard devi-
ations of the residuals were the highest
among the three models tested. The large
variance in the residuals might be ex-
pected since the LPM generates a predic-
tion which likely is the most specific to a
particular place and time. Model predic-
tions for maximum Oa by the UAM in a
specific sense (at the same time and
location as the observed maximum) were
consistently low for all evaluations with an
average 32% underprediction over the
sample. If the time and location of the
model predictions are not constrained to
be the same as those for the maximum
observed Oa, the average model bias for
the 20 days implied a 4% overprediction.
This excellent agreement might suggest
that the uncertainty in specifying a wind
field for a grid model like the UAM could
lead to large apparent errors in the model
results.
The choice of which particular model to
use in a specific application involves not
only the accuracy of the model but also the
resources required to operate it. The
models tested here have resource require-
ments correlated with the-:r level of com-
plexity. In terms of man-months needed
to set up a single day simulation and com-
puter time expended (minutes of CPU on a
UNIVAC 1 100/82) the approximate re-
quirements are:
PBM — 015 man-month — 1 minute CPU
LPM — 0.20 man-month — 10 minutes CPU
UAM — 0.50 man-month — 110 minutes CPU
These models are now being made
available to EPA's Office of Air Quality
Planning and Standards for further statis-
tical and sensitivity testing and ultimately
for use in their regulatory decision-making
process. Because model development is
an evolving area of research it is very likely
that subsequent "improved" versions of
the models tested here will become avail-
able. A performance test with benchmark
results, as described and tabulated in this
report, now exists for future use in urban
air quality model comparisons with any
subsequent versions of the model.
K. L Schere and J. H. Shreffler are on assignment to the U.S. Environmental
Protection Agency from the National Oceanic and Atmospheric A dministration,
U.S. Department of Commerce.
J. H. Shreffler is the EPA Project Officer (see below).
The complete report, entitled "Final Evaluation of Urban-Scale Photochemical Air
Quality Simulation Models," (Order No. PB 83-147 991; Cost: $22.00, subject
to change) will be available only from:
National Technical Information Service
5285 Port Royal Road
Springfield, VA 221'61
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
OU.S. GOVERNMENT PRINTING OFFICE 1983-659-017/7006
-------
United States
Environmental Protection
Agency
Center for Environmental Research
Information
Cincinnati OH 45268
Postage and
Fees Paid
Environmental
Protection
Agency
EPA 335
Official Business
Penalty for Private Use $300
PS 00HOTE
------- |