United States
Environmental Protection
Agency
Environmental Sciences
Research Laboratory
Research Triangle Park NC 27711
Research and Development
EPA-600/S3-84-079 Aug. 1984
Project Summary
An Evaluation of Alternative
Gaussian Plume Dispersion
Modeling Techniques in
Estimating Short-Term Sulfur
Dioxide Concentrations
Thomas E. Pierce
A routinely applied atmospheric
dispersion model was modified to
evaluate alternative modeling tech-
niques that allow for more detailed
source data, onsite meteorological
data, and several dispersion methodol-
ogies. These techniques were evaluated
using hourly measured SO2 concentra-
tions at fixed receptors around coal-
fired power plants near Paradise,
Kentucky, during 1976, and near
Johnsonville, Tennessee, during 1977.
A significant finding of the evaluation
was that the more sophisticated models
did not appreciably outperform the
routinely applied models. The models
using airport meteorological data
performed as well as the models using
onsite wind data. With the Pasquill-
Gifford and Briggs dispersion schemes,
small differences in model performance
were observed. More substantial differ-
ences occurred with models using
onsite turbulence measurements. The
model using Pasquill's recommenda-
tions tended to overpredict peak
concentrations. The models based on
Draxler's and Cramer's approaches
using onsite turbulence yielded mixed
results, perhaps in part because the
lateral standard deviation of wind
direction available was the 1 -h average
of 5-min values (rather than a 1 -h value),
thus eliminating the longer period
fluctuations that are important in
estimating 1-h concentrations in
addition to the shorter period
fluctuations. Additional research is
recommended to improve the
application of onsite turbulence
measurements and to provide more
accurate estimates of plume trajector-
ies for input to atmospheric dispersion
models.
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 docu-
mented in a separate report of the same
title (see Project Report ordering in-
formation at back).
Introduction
The EPA routinely uses Gaussian
plume models to make decisions having
important environmental and economic
impacts on air quality. Despite the
economic and environmental importance
placed on air quality modeling, the predic-
tions from dispersion models are often no
better than within a factor of two of the
observations. Recognizing the limitations
of these models, the EPA and the Ameri-
can Meteorological Society (AMS) have
initiated steps to improve current
modeling practices. The general
consensus is that more validations of air
quality simulation models are needed to
improve existing models and to guide the
development of future models.
-------
In the current study, a Gaussian plume
dispersion model incorporating alterna-
tive techniques was evaluated using a
comprehensive data base developed for
two Tennessee Valley Authority coal-
fired power plants: Johnsonville and
Paradise. Alternative modeling tech-
niques were included in a routinely
applied Gaussian dispersion model, in-
cluding different approaches for emis-
sions, meteorology, and dispersion.
The benchmark model from which the
alternatives were developed was the
MPTER algorithm. Multiple Point
Gaussian Dispersion Algorithm with
Optional Terrain Adjustment, an air
quality simulation model that is used by
EPA in regulatory applications. The
MPTER was especially suited for this
study because actual coordinates for
each monitoring site could be specified,
along with the physical separation
between stacks. Furthermore, the
author's in-depth experience with
MPTER allowed for careful and expedient
inclusion of the alternative models.
Modeling Alternatives Used
In modeling for regulatory applications,
the design capacity or the maximum
operating capacity of resulting SO2
emissions and source parameters are
often required as model inputs. The daily
and seasonal operating rates vary
significantly at most coal-fired power
plants. Furthermore, maximum surface
concentrations at a power plant can occu r
under less than full-load conditions.
Thus, for the Johnsonville and Paradise
locations, hourly S02 emissions, stack
gas temperatures, and gas exit velocities
were computed for comparison to full-
load conditions.
The meteorological data required for
EPA regulatory applications are often
selected from the National Weather
Service (NWS) stations that are most
representative of the site being modeled.
In this study, the airport data from
Nashville, Tennessee (BNA) were used as
model inputs for the Paradise and
Johnsonville sites. Evansville, Indiana
(EW) surface data were also included for
comparison at the Paradise site. Because
the use of onsite meteorological data is
preferred, the hourly average 110-m
onsite wind data were used at both power
plants.
An accurate estimate of dispersion is
critical in air quality analysis. In this
regard, much effort was made in
developing the alternative models in this
study. The MPTER uses Pasquill-Gifford
(PG) dispersion coefficients. As one
alternative, Pasquill's approach for
estimating buoyancy-induced dispersion
(BID) was evaluated. Briggs' rural disper-
sion coefficients combined with Pas-
quill's1, Draxler's2, and Cramer's3 that
use onsite turbulence measurements
were evaluated for the Johnsonville site.
A matrix of model alternatives was
developed for use in Paradise in 1976 and
for use in Johnsonville in 1977. These
matrices are shown in Tables 1 and 2
respectively. In the development of the
alternative models, an effort was made to
keep each change in a technique inde-
pendent of the other techniques.
Accomplishments
Short-term S02 concentrations were
evaluated at Paradise with seven model-
ing alternatives and at Johnsonville with
eleven modeling alternatives. The
accomplishments of this study included
the following:
1) The development of two years of
data (one year for Paradise and one
year for Johnsonville) consisting of
hourly S02 emission rates and stack
characteristics, hourly S02 concen-
trations monitored at ten to twelve
receptor sites, hourly average onsite
meteorological parameters, and
hourly NWS meteorological data
collected at BNA, and EW;
2) The incorporation of alternative
modeling techniques into the EPA
Gaussian plume dispersion model,
MPTER, including modifications of
MPTER to allow for the input of
onsite meteorological data, the
development of an algorithm for
calculating dispersion using Briggs'
rural interpolation formulae, and
adaptations of onsite turbulence
schemes using methods described
by Pasquill, Draxler, and Cramer;
3) The implementation of these
models using the available data on
EPA's UNIVAC computer; and
'Pasquill, F. 1976. Atmospheric Dispersion Parame-
ters in Gaussian Plume Modeling Part II Possible
Requirements for Change in Turner Workbook
Values. EPA-600/4-76-030b. Environmental Pro-
tection Agency Research Triangle Park, North Caro-
lina. 53 p.
2Draxler, R. 1976. "Determination of Atmospheric
Diffusion Parameters."Atmos. Environ. 10: 99-105.
3Cramer, H. E. 1976 "Improved Techniques for Mod-
eling the Dispersion of Tall Stack Plumes." In: Pro-
ceedings of the Seventh International Technical
Meeting on Air Pollution Modeling and Its Applica-
tion. N. 51, NATO/CCMS Airlie House, Virginia, pp.
731-780.
4) The analysis of the model's results
and observed S02 concentrations
using relatively straightforward sta-
tistical comparisons.
Results
Paradise
At the Paradise plant, greater differ-
ences in predicted SO2 concentrations
resulted from using various emissions
data than from using alternative disper-
sion schemes and meteorological data.
Furthermore, the "conservative"full-load
emissions models compared more favor-
ably to peak observed concentrations
than the "more realistic" models using
hourly emissions data, which underpre-
dicted peak concentrations by about a
factor of two. This unexpected result was
attributed to inaccuracies associated
with estimating stability, mixing heights,
wind vectors, and/or S02 emission rates.
Another unexpected feature of the
Paradise analysis was that the models
incorporating onsite wind data did not
show significant improvements over the
models using airport meteorological data.
Although the concentrations from the
two modeling alternatives differed hour
to hour, the overall results, especially for
the peak concentrations, were similar.
Although emissions input and
meteorological data have been
recognized as having important effects in
dispersion modeling, dispersion schemes
are usually discussed as a primary basis
for model differences. The concentra-
tions predicted with the three dispersion
schemes at Paradise, however, resulted
in average differences of less than 10
percent. Buoyancy-induced dispersion
did seem to increase concentrations with
the models using the Briggs and PG
dispersion coefficients, although the
Briggs coefficients potentially can yield
lower concentrations for downwind
distances up to two kilometers.
Furthermore, measurements taken from
receptors closer to the source than those
located at Paradise should be examined
for comparison with the Briggs and PG
curves.
Johnsonville
With the 11 Johnsonville alternative
models, the principal emphasis was on
the prediction of peak concentrations,
especially the concentrations related to
the National Ambient Air Quality Stand-
ards for SO2. Some additional analyses
were used to diagnostically evaluate the
models.
The significant differences between
the emissions data alternatives that were
-------
Table 1. Modeling Alternative Matrix Used for Paradise
Model Number
Technique
Emissions input
Full-load conditions
Hourly averages
Meteorology input
BNA airport data
EVV airport data
Onsite wind data
Dispersion scheme
PG
PC with BID
Briggs formulae with BID
Table 2. Mode/ing Alternative Matrix for Johnsonville
Model Number
Technique
2 3
10 11
Emissions input
Full-load conditions
Daily averages
Hourly averages
Meteorology input
BNA airport data
BNA, exclude calms
Onsite wind data
Dispersion scheme
PG
PG with BID
Briggs with BID
Pasquill's 1976 scheme
Draxler's scheme
Cramer's scheme
XXXXXXXX
X X X X X X
noted in the Paradise analysis were not
evident in the Johnsonville analysis. In
fact, the models using full-load emissions
data compared similarly to the models
using hourly and daily emission data for
the peak concentrations. The differences
between the models using daily and
hourly emissions data seemed insignifi-
cant; in retrospect, it probably would have
been more cost effective to use the daily
emission rates rather than the hourly
emission rates. The similar performance
noted for the models using full-load,
daily, and hourly emission rates may have
been due to the increased plume rise
associated with full-load stack
parameters, which would cause a reduc-
tion in maximum surface concentrations.
It appeared that the procedure to elim-
inate calm winds had little effect on peak
concentrations. At other power plant
sites where the receptors are located
closer to the source and where A- and B-
stability has a greater effect on peak
concentrations, the elimination of calm
winds might be an important
consideration.
As with Paradise, little additional bene-
fit was gained by using onsite wind data
in preference to airport data. This is
probably because the airport site, BNA,
and the Johnsonville site are similar in
topography. In the diagnostic evaluation,
the model using onsite wind data did
show better performance over the airport
model, resulting in higher correlations
and mean fractional errors closer to zero.
The differences between the models
using different dispersion schemes were
again small in contrast to the differences
noted with the emission and meteorolog-
ical alternatives. In predicting the peak
concentrations, the three basic disper-
sion models (PG, PG with buoyancy-
induced dispersion, and Briggs)
performed about equally. The models
using on-site turbulence data, however,
did not perform as well as expected. The
Pasquill scheme overpredicted most peak
concentrations. The modeling
alternatives using the Cramer and
Draxler dispersion schemes gave results
that were comparable to the models
using the PG curves but differed
substantially, on occasion, from the
observations. Because the Cramer and
Draxler models limited dispersion during
nighttime hours, further work is needed
to develop a scheme to practically
estimate dispersion using routine
meteorological measurements during all
meteorological regimes.' In this study, the
PG model appeared to perform as well as
the other dispersion schemes.
Conclusions and
Recommendations
In studies of this kind, the results often
suggest a need for improving modeling
techniques and for obtaining more so-
phisticated input data. This study did not
deviate from this tradition. The alterna-
tive modeling approaches used with the
Johnsonville and Paradise plants yielded
results that were within an order of
magnitude of the peak SO2 concentra-
tions. Yet, further improvements are
possible with more accurate source term
data, more site-specific meteorological
data, and more refined dispersion
schemes.
The major differences in model per-
formance resulted from differences in
emission rates and to a lesser extent from
the use of different meteorological data.
The use of alternative dispersion
schemes was not as significant.
However, efforts to improve the measure-
ments of the standard deviations of
horizontal and vertical wind direction (as
well as other turbulence parameters)
could provide a more reliable means for
estimating dispersion.
The task of model evaluation in this
study was especially onerous, because
slight differences in predicted plume
trajectories overwhelmed the statistical
comparison used to examine the
alternative modeling approaches. This
further substantiates the observation
that the Gaussian plume dispersion
model is especially sensitive to the input
of wind direction. Model evaluations
could perhaps be simplified by following
the advice of Calder and Johnson, i.e.,
that each model component be separately
evaluated. Otherwise, it is difficult to
compare the strengths and weaknesses
-------
of different modeling techniques. The
difficulty in performing model evaluation
is further compounded by lack of field
data. The availability of field data from
studies such as Cinder Cone Butte, Idaho,
and Kincaid, Illinois, will hopefully make
the tasks of model evaluation and model
improvement less formidable.
The EPA author Thomas £. Pierce was on assignment to the Environmental
Sciences Research Laboratory, Research Triangle Park, NC 27711 from NOAA,
U. S. Department of Commerce {presently with NUS Corporation, Gaithersburg,
MD).
D. Bruce Turner is the EPA Project Officer (see beiow).
The complete report, entitled "An Evaluation of Alternative Gaussian Plume
Dispersion Modeling Techniques in Estimating Short-Term Sulfur Dioxide
Concentrations," (Order No. PB 84 -223 62 7; Cost: $16.00, 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:
Environmental Sciences Research Laboratory
U.S. Environmental Protection Agency
Research Triangle Park, NC 27711
US GOVERNMENT PRINTING OFFICE, 1984—759-015/7775
United States
Environmental Protection
Agency
Center for Environmental Research
Information
Cincinnati OH 45268
Official Business
Penalty for Private Use $300
T i' T t- I
-, X A n Y
' ' i n 'I S T N
i f i
t I
------- |