United States
Environmental Protection
Agency
Office of Air Quality
Planning and Standards
Research Triangle Park NC 27711
EPA-450/4-89-020
NOVEMBER 1989
Air
REVIEW AND EVALUATION
OF
AREA SOURCE DISPERSION
ALGORITHMS
FOR EMISSION SOURCES
ATSUPERFUND SITES
-------
EPA-450/4-89-020
REVIEW AND EVALUATION OF
AREA SOURCE DISPERSION
ALGORITHMS
FOR EMISSION SOURCES
AT SUPERFUND SITES
By
TRC Environmental Consultants, Inc.
East Hartford, CT 06108
EPA Contract No. 68-02-4399
EPA Project Officer Jawad S. Touma
Office Of Air Quality Planning And Standards
Office Of Air And Radiation
U. S. Environmental Protection Agency
Research Triangle Park, NC 27711
November 1989
-------
This report has been reviewed by the Office Of Air Quality Planning And Standards, U. S.
Environmental Protection Agency, and has been approved for publication as received from the
contractor. Approval does not signify that the contents necessarily reflect the views and policies of the
Agency, neither does mention of trade names or commercial products constitute endorsement or
recommendation for use.
EPA-450/4-89-020
-------
ACKNOWLEDGEMENTS
The authors wish to acknowledge the contributions of numerous individuals,
without whose assistance this study could not have been successfully
completed. The Project Officer, Mr. Jawad Touma, provided guidance and review
comments throughout the course of the project, including numerous constructive
suggestions regarding the draft report. Mr. Joseph Tikvart, Chief of the
Source Receptor Analysis Branch, also provided a critical review of the
report. A large number of investigators were contacted during the review of
technical literature and recent modeling developments. Dan Reible of
Louisiana State University, Jim Bowers of U.S. Anny-Dugway, and Richard
Schultz of Trinity Consultants were particularly helpful in providing
information to aid this effort. Dr. Brian Lamb of Washington State University
provided valuable information concerning the forest canopy area source
experiments. The sensitivity analysis plots presented in Section 3 were
produced through the diligent efforts of Ms. Maureen Hess.
-111-
-------
TABLE OF CONTENTS
SECTION PAGE
ACKNOWLEDGEMENTS ii
1.0 INTRODUCTION 1-1
1.1 Approach 1-1
1.2 Identify Superfund Site Area Source
Characteristics 1-2
1.2.1 Source Site and Shape Considerations 1-3
1.2.2 Environmental Influences 1-7
1.2.3 Hunan Activity 1-7
1.2.4 Summary of Findings 1-8
2.0 REVIEW OF AREA SOURCE DISPERSION ALGORITHMS .... 2-1
2.1 Introduction 2-1
2.2 Overview of Area Source Algorithms 2-2
2.2.1 Virtual Point Source 2-2
2.2.2 Point Source Array 2-3
2.2.3 Line Source Segment(s) 2-3
2.2.4 Line Source Integration 2-4
2.3 Existing Area Source Models 2-6
2.3.1 Industrial Source Complex Short-Term Model . . 2-7
2.3.2 Fugitive Dust Model 2-7
2.3.3 Point, Area, Line-Source Model 2-10
2.3.4 Gaussian-Plume, Multiple Source Air Quality
Algorithm 2-10
2.3.5 SHORTZ Model 2-12
2.3.6 TEM/PEM 2-13
2.3.7 ISCLT 2-13
2.3.8 Climatological Dispersion Model 2-14
2.3.9 LONGZ Model 2-16
2.3.10 VALLEY Model 2-16
2.3.11 Air Quality Dispersion Model 2-18
2.3.12 Area Source Screening Techniques 2-18
2.3.13 Toxic Release Models 2-19
2.3.14 Other Existing Models 2-21
2.4 Overview of Area Source Dispersion Literature . . 2-21
2.5 Limitations of Existing Models 2-24
3.0 ANALYSIS OF MODEL PREDICTIONS FOR EXAMPLE
APPLICATIONS 3-1
3.1 Approach 3-1
3.2 Tests of Mathematical and Physical Principles . . 3-3
3.3 Predicted Concentrations for Base Case 3-8
3.3.1 Short-Term Models 3-8
3.3.2 Long-Terra Models 3-10
3.4 Tests of Mathematical and Physical Properties . . 3-11
3.4.1 ISCST 3-11
3.4.2 FDM 3-13
3.4.3 PAL 3-14
3.4.4 RAM 3-14
3.4.5 SHORTZ 3-15
3.4.6 ISCLT 3-17
-v-
-------
TABLE OF CONTENTS
(CONTINUED)
SECTION PAGE
3.4.7 CDM 3-17
3.4.8 VALLEY 3-18
3.5 Discussion of Results 3-19
3.5.1 Short-Tern Results 3-20
3.5.2 Sector Average Models 3-22
4.0 COMPARISON OF MODEL PREDICTIONS WITH EXPERIMENTAL
DATA 4-1
4.1 Database Description 4-1
4.2 Results 4-5
4.2.1 Unpaired Comparisons 4-6
4.2.2 Paired Comparisons 4-6
5.0 CONCLUSIONS AND RECOMMENDATIONS 5-1
5.1 Conclusions 5-1
REFERENCES R-l
-VI-
-------
LIST OF FIGURES
FIGURE PAGE
2-1 SUBDIVISION OF A SQUARE AREA SOURCE INTO FIVE LINE
SOURCE SEGMENTS NORMAL TO THE WIND DIRECTION .... 2-5
2-2 SUBDIVISION OF RECTANGULAR AREA SOURCE INTO LINE
SEGMENTS BY PAL 2-11
2-3 ILLUSTRATION OF SECTOR INTEGRATION USED IN CDM .... 2-15
2-4 SCHEMATIC OF THE VIRTUAL POINT SOURCE AS PROJECTED FROM
AN AREA SOURCE 2-17
3-1 AREA SOURCE AND RECEPTOR CONFIGURATION FOR BASE CASE
SCENARIO 3-2
3-2(a) ILLUSTRATION OF INFLUENCE ZONE FOR NEAR-FIELD RECEPTORS 3-5
3-2(b) SUBDIVISION OF BASE CASE AREA SOURCE 3-6
3-2(c) SOURCE-RECEPTOR CONFIGURATION FOR DIAGONAL WIND
DIRECTION 3-7
3-3 to 3-47 GRAPHS OF CENTERLINE CONCENTRATION VS. DOWNWIND
DISTANCE FOR SELECTED MODELS, METEOROLOGICAL
CONDITIONS AND SOURCE CONFIGURATIONS 3-23 to 3-67
4-1 ISOPRENE FLUX EXPERIMENT SAMPLING GRID, RELEASE POINTS
AND WOODLOT 4-3
4-2 RANGE OF MAXIMUM PREDICTED AND OBSERVED CONCENTRATIONS
FOR TRACER RELEASE EXPERIMENTS 4-7
-vi i-
-------
LIST OF TABLES
TABLE PAGE
1-1 AREA SOURCE CHARACTERISTICS 1-4
2-1 CHARACTERIZATION OF AREA SOURCE ALGORITHMS IN EXISTING
MODELS 2-8
3-1 SUMMARY OP SENSITIVITY TEST RESULTS FOR SHORT-TERM
MODELS 3-21
4-1 METEOROLOGICAL AND EMISSION CHARACTERISTICS FOR TRACER
RELEASE EXPERIMENTS IN A FOREST CANOPY 4-4
4-2 MAXIMUM OBSERVED AND PREDICTED NORMALIZED TRACER
CONCENTRATION FOR EXPERIMENT 4-8
4-3 SUMMARY OF RELATIVE DIFFERENCES BETWEEN OBSERVED AND
PREDICTED MAXIMUM VALUES 4-10
-ix-
-------
1.0 INTRODUCTION
An evaluation of dispersion modeling techniques available for estimating
ambient concentrations produced by emissions from sites contaminated with
toxic pollutants has been conducted. The focus of this study has been the
application of area source dispersion algorithms to these emission sources.
These sources can be characterized as low level releases with little buoyancy
due to either momentum or temperature. Potential applications include both
the estimation of short-term concentrations associated with site remediation
activities and the calculation of long-term concentrations for estimating
population exposures in the vicinity of a landfill, lagoon or waste disposal
facility.
1.1 Approach
The study consisted of four tasks. The report has been organized to
present the results of each task in sequence.
Task 1 - Identify Source Characteristics. The emission characteristics
representative of superfund/landfill sources were examined to identify
modeling requirements and related technical issues associated with estimating
ambient concentrations near these sites. Task 1 findings are described in
Section 1.2.
Task 2 - Review Available Area Source Models. Existing models and the
technical literature were reviewed to identify available modeling techniques
for estimating short-term and long-term concentrations due to area sources.
Specific models were selected for further evaluation, based on their potential
suitability for landfill/superfund applications. Task 2 findings are
described in Section 2.
Task 3 - Analysis of Model Predictions for Example Applications. Five
short-term area source models and three long-term (sector average) models were
1-1
-------
applied to a number of example applications in order to compare the magnitude
of concentration predictions and to test whether near-field and far-field
concentration estimates were consistent with mathematical and physical
principles. Task 3 results are presented in Section 3.
Task 4 - Comparison of Model Predictions with Observed Concentrations.
The five short-term models were applied to estimate ambient concentrations for
a series of tracer dispersion experiments involving low-level releases within
an isolated grove of trees. These experiments simulate an area source release
over an area of 15,000 square meters. Predicted and observed tracer
concentrations were compared for thirteen one-hour experiments at sampler
locations approximately 100 m downwind of the source region. Task 4 results
are summarized in Section 4.
Conclusions and recommendations from this study are discussed in Section 5.
1.2 Identify Superfund Site Area Source Characteristics
Landfills and other large area sources have characteristics which have a
substantial bearing on the air quality modeling method used to simulate
pollutant dispersion from such sources. Many of the currently used pollutant
dispersion models are based on a Gaussian formulation originally developed for
point sources. Over time, the model formulations have been adapted for
application to distributed sources (line sources, area sources, volume
sources) but many of the original point source assumptions are retained in
these models. The purpose of this discussion is to describe the types of
emissions from superfund sites that may be modeled as area sources.
Subsequent sections of this report will provide detailed descriptions of the
current model formulations used to predict dispersion from these area sources.
Various types of toxic waste sources fall into the area source category.
These include landfills, waste lagoons, evaporation and settling ponds,
1-2
-------
agricultural fields which have been treated with chemicals, and regions where
long-tern exposure to toxic chemicals (due to manufacturing, mining, power
generation, etc.) has contaminated the soil. For all of these sources,
pollutants are emitted at or near ground level. The sizes of these sources
can range from a few square meters in the case of settling ponds to a few
square kilometers in the case of contaminated soils. The effect of these
sources on the population also varies based on source size, strength and
type. Many sources are located in areas remote from the general population.
Others, by their nature, occur in centrally located areas where the potential
for exposure is large. Emissions from many of these sources are a function of
human activity. Addition of material, maintenance or disruption of the
surface layer will have a bearing on emissions. Emissions may also be
affected by environmental conditions such as wind speed, air temperature,
ground surface temperature, and surface moisture.
1.2.1 Source Site and Shape Considerations
Estimating concentrations from surface area sources presents a number of
challenges for the modeler. Available area source algorithms contain a
variety of assumptions and offer different strengths and weaknesses. More
reliable concentration estimates can be obtained by careful problem definition
and by selection of the most appropriate model/algorithm. A summary of
superfund site area source characteristics is included as Table 1-1. The
location, geometry and relative elevation of a typical waste storage or
landfill site are important factors in the dispersion characteristics of the
site. Gaussian dispersion models generally assume an elevated source which
injects pollutants into a moving air stream. Since many landfills are
low-lying or in pits, the mechanisms by which pollutants enter the atmosphere
are more complicated. The rate of entrainment of pollutants which exit the
1-3
-------
TABLE 1-1
AREA SOURCE CHARACTERISTICS
- spatially continuous emissions (non-point)
unique site geometry including depressions, piles,
lagoons, etc.
temporally variable source strength dependent on ambient
atmospheric conditions (primarily temperature, wind speed,
precipitation) and maintenance activity levels
- surface sources characterized by low wind speed at surface
- density flows may be important over surface area sources
chemical type of emission may vary spatially across
landfills
1-4
-------
surface will be controlled by diffusion, mechanical (turbulent) flushing and
suction effects. In the case of a surface source, the horizontal velocity at
the surface is zero ("no slip" boundary condition) and the rate of emission of
the pollutant is determined by the diffusive velocity, which is based on vapor
pressure gradients and boundary layer resistance. The near surface layer is
characterized by low mean wind speeds (often less than 1 m/s). The Gaussian
models generally avoid extrapolating wind speeds down to the surface, however,
because low wind speed values can cause unrealistically high concentration
predictions. Most models specify a minimum 1.0 m/s wind speed for calculating
concentrations.
Many modern landfills have substantial vertical extent (>10 m) and
therefore may be modeled as elevated area or volume sources. Since the
landfill is projecting into the flow, the ambient wind will ventilate the
surface layer of the landfill. The actual height is physically limited by
base area of the landfill and the angle of repose of the material. In
practice, the maximum height may be dictated by regulation and/or constrained
by the need for access by heavy machinery. In the case of a large elevated
landfill, the source itself may be an obstacle to the flow and establish the
flow field for some distance to the leeward side of the landfill. The modeler
may account for this effect by using existing building wake or downwash
algorithms. However, the elevated landfill is unlike a building because its
sides are not vertical and its entire surface area (top and sides) may be
emitting pollutants. The sloping sides of a landfill represent a more
"aerodynamic" shape and are likely to produce less turbulence than a building
with vertical walls. Since the landfill flanks are angled, the source
strength per horizontal area of the flanks may be greater than that of the
top. Elevated landfills may have a region of high concentration in the lee of
1-5
-------
the landfill pile with dilution primarily due to wake eddies entraining
unconcaminated air from windward of the landfill.
The estimation of emissions and dispersion is also complicated if the
source is below the local land surface elevation. Situations in which trash,
waste sludge, etc. are placed in a pit in the earth are common, and such
situations may require emission and/or dispersion algorithms to treat a sunken
source and a cavity flow problem. With existing algorithms, the sunken source
may be approximated by a surface source of lesser strength. The rate of
pollutant emissions to the ambient air will depend not only on the rate of
emissions from the ground surface into the air within the pit, but also on the
rate of exchange (ventilation) between the pit and the air above it.
In all of these scenarios, it may be useful to characterize mixing and
surface conditions and use these characterizations in the calculation of
source strength. Unlike the combustion effluents created by manufacturing and
power production which are proportional to the production rate, the source
strength of a landfill or other area source is more directly related to
environmental conditions. Under conditions of limited mixing, the partial
pressure gradient over the area source is reduced and volatilization of a
given chemical decreases. When mixing is enhanced, partial pressure gradients
are steep and the source strength increases. Snow cover may effectively cap
an area source and rainfall greatly reduces emissions.
In the situation of a low lying or sub-surface source with emissions at
ambient temperature (neutral thermal buoyancy) the worst case concentration
scenario will be a surface-based inversion where a relatively shallow layer of
dense, cold air sits on the surface of the source providing a thin mixing
layer and high concentrations. It could be expected that this effect would be
local because wind speed and transport are limited in most strong surface
inversions. Conversely, the fact that the air in these inversions is denser
1-6
-------
near the surface allows density flows to develop. In regions of complex
terrain this effect could lead to anomalously high concentrations down the
hill from a source. Air "draining" from a source of toxics and "pooling" over
a low-lying area could lead to high exposures.
1.2.2 Environmental Influences
The source strength of an area source may be strongly correlated to source
surface temperature. Chemical reactions and decay rates of the emitted
pollutant may also depend on temperature. In the case of an area source such
as a landfill, many gaseous emissions occur by evaporation. Evaporation rates
are a function of the ground surface temperature. For many organic compounds,
evaporation rate is strongly dependent on temperature across ambient ranges.
After emission, some evaporated compounds undergo chemical reactions which are
strongly temperature dependent. In general, algorithms to estimate
evaporation rates and decay rates as a function of environmental conditions
are not yet available.
1.2.3 Human Activity
The degree to which area source strength is coupled to the environment may
depend on the surface characteristics of the source. In the case of an
abandoned site, the source surface may be capped and/or vegetated which would
lower source strength sensitivity to ambient temperature and wind. In the
case of a cap, source strength would be largely determined by sub-surface
diffusion. In an active landfill, material may be periodically "turned"
causing source strength to be dependent on both the ambient environmental
conditions and on landfill operation or clean-up schedules.
The source emissions from many large area sources will be spatially
inhomogenous with respect to both strength and type. For example, at a
1-7
-------
typical landfill, different types of waste (household trash, construction
waste, industrial waste, etc.) will be dumped in different areas, and dumping
will occur in one area while grading occurs in another. These differences are
often important for estimating peak short-term concentrations, but are less
important on an annual basis.
1.2.4 Summary of Findings
Air pollutant emission sources at sites contaminated with toxic pollutants
are typically low-level, distributed sources, ranging in size from about 10
square meters to several square kilometers. Considerable spatial variation in
emission rates across the source area is common, particularly for larger
areas. Both short-term (one-hour) and long-term (annual) average
concentration estimates are needed, depending upon the specific application.
Typical source-receptor distances for short-term concentration estimates range
from 10 m to 1000 m and may include "on-source" receptors (to estimate worker
hazard) and "fenceline" receptors immediately adjacent to the contaminated
area. For long-term average estimates, population exposure is most often the
primary concern. Typical source-receptor distances for long-term estimates
range from several hundred meters to about 10 km.
Source geometries encountered at some contaminated sites pose special
challenges for air quality modeling. Elevated sources (e.g., elevated
landfills), below-grade sources (e.g., excavated areas), and small heavily
contaminated areas "nested" within larger source regions all represent
difficult scenarios for many existing area-source models.
Source emission rates at contaminated sites often vary with environmental
conditions and as a result of human activities. Many of the same factors
which affect emissions will also influence dispersion behavior. It is
therefore important to account for these variations when performing air
quality modeling analyses.
1-8
-------
2.0 REVIEW OF AREA SOURCE DISPERSION ALGORITHMS
2.1 Introduction
Accurate prediction of ambient concentrations resulting from pollutant
emissions from area sources is a problem of considerable importance.
Spatially distributed sources of pollution include landfills, settling ponds,
lagoons, agricultural fields and storage piles. Residential and commercial
urban areas may also be considered as area sources, although emissions
actually originate from many small point sources.
The most widely used air quality models are those incorporating the
Gaussian plume equation to describe the transport and dispersion of emissions
from a point source. These models provide an efficient and easily understood
calculation procedure for estimating concentrations. Area source algorithms
based on the Gaussian plume equation are of primary interest for this study,
since these algorithms can be easily incorporated into another existing
model. Area source algorithms from non-Gaussian models are not transferable
and are not as well suited for routine applications. Non-Gaussian area source
algorithms were reviewed, however, in an effort to identify any approaches
which might represent significant improvements in calculation accuracy or
efficiency.
For a Gaussian model, the concentration from a point source at a receptor
can be calculated readily, using a simple set of equations. If the emissions
are spread uniformly over an area, however, the resulting concentration can be
determined only by integrating over the source area. The area source
contribution cannot be computed exactly using simple equations. Area source
algorithms represent calculation procedures which have been designed to
estimate the area source contribution in an efficient manner.
2-1
-------
Pollutant dispersion models which are currently available to predict
pollutant concentrations at user-specified receptor locations have been
examined. One important consideration for this study is the availability of
FORTRAN source code. Eleven of the models included in the UNAMAP6 package
(EPA, 1986) contain area source algorithms. Additional models were identified
by contacting researchers active in developing models for regulatory
applications and by searching the recent technical literature.
2.2 Overview of Area Source Algorithms
Four basic techniques for estimating area source impacts were identified.
Two of these approaches are primarily source-oriented, while two (upwind
integration and point source array) are receptor-oriented. Different models
employ variations on a given technique or a hybrid of two methods. The four
basic techniques are described below.
2.2.1 Virtual Point Source
The virtual point source method as suggested by Turner (1970) to estimate
area source impacts assumes that the pollutant plume downwind of an area
source can be simulated as a point source. The initial source dimensions are
accounted for by placing the point source upwind of the actual area source
location, so that the lateral spread of the plume at the area source is
comparable to the source width. The emissions for the replacement source are
set equal to the area source emissions. Thus, the equations used to compute
concentrations for an area source are the same as those used for a point
source. One limitation inherent in this approach is that the point source
plume does not accurately reproduce the crosswind concentration distribution
on the source. The point source plume distribution is Gaussian, with higher
concentrations in the center and lower concentrations at the edges, while the
2-2
-------
actual emissions are uniform across the source. Another limitation of this
method concerns the choice of a single source location for oz, which
eliminates the effect of the along-wind source distribution. Concentration
predictions are therefore distorted, particularly in the near-source region.
Virtual point source algorithms are generally not utilized in short-term
models. For long-term models, which incorporate a sector-averaged plume
width, this technique is more commonly used. Examples include ISCLT, LONGZ
and VALLEY.
2.2.2 Point Source Array
The receptor-oriented point source array provides a second method of
estimating area source concentrations using the Gaussian plume equation for a
point source. Using this approach, the model generates an array of point
sources centered on the receptor. The total area of the region covered by the
source array is apportioned among the point sources, so that each point is
representative of its surrounding area. The emissions due to area sources are
then apportioned to the point source array, based on the location of each
point source and the size of the area it represents. Two of the models
reviewed for this study calculate area source contributions using a point
source array. CDM, a sector-average model, generates a radial source array
around each receptor based on the 22.5° sectors which it employs for
calculating concentrations. The Texas Episodic Model (TEM) and its
derivative, PEM, generate a rectangular source array around each receptor.
2.2.3 Line Source Segment(s)
For a Gaussian model, the concentration calculation for an area source can
be simplified if the source area is simulated as one or more line segments
oriented normal to the wind direction. The computation effort required for a
2-3
-------
line source with uniform emission density is roughly equivalent to that for
two or three point sources. A schematic diagram is provided in Figure 2-1
illustrating the simulation of a square area using five line source segments.
If the endpoints of each segment correspond to the boundaries of the source
region, the simulated source will provide a reasonable representation of
source geometry. The number of segments used by a model represents a choice
between accuracy and efficiency. Depending upon the number of lines employed
and the positioning of the lines within the source area, the collapsing of a
two-dimensional area into a line may significantly distort the source-receptor
geometry. This problem is most significant in the near-field and most
pronounced with algorithms that use single line segments to represent an
area. The majority of the short-term models reviewed for this study use the
line source segment approach. Examples include ISCST, FDM, PAL, and SHORTZ.
2.2.4 Line Source Integration
The use of line source segments normal to the wind direction provides a
convenient method for simulating area source contributions, but the
computational effort can be substantial if each area source is subdivided into
many line source segments. Certain simplifying assumptions can greatly
enhance computational efficiency by accounting for the contributions of many
line segments within a single calculation. Two techniques were identified
which represent integration over many line segments.
Crosswind Integration. In urban regions, the emissions density generally
does not change abruptly between adjacent area sources. If the emissions
density is assumed to be constant in the crosswind direction, area source
contributions can be calculated based only on the emissions from sources along
a line directly upwind from the source. The RAM model incorporates this
"narrow plume hypothesis" to compute area source contributions. This
2-4
-------
WIND
DIRECTION
LINE SOURCE
FIGURE 2-1.
SUBDIVISION OF A SQUARE AREA SOURCE INTO
FIVE LINE SOURCE SEGMENTS NORMAL TO THE
WIND DIRECTION
2-5
-------
technique will give accurate results only if the emissions density is
relatively uniform over a crosswind distance comparable to the upwind
source-receptor distance. Concentration calculations do not account
accurately for source-receptor geometry for a single (isolated) area source
square.
Upwind Integration. For receptors located within an area source or at
near-field distances, the region upwind of a receptor is often occupied by a
single area source. In this situation, the crosswind term in the Gaussian
equation integrates to a constant, independent of downwind distance. For a
ground-level release height, it then becomes possible to integrate over the
source area in the upwind direction. This technique is only appropriate at
near-field distances and only for receptors downwind of the center of the
source area. Both SHORTZ and LONGZ employ upwind integration at near-field
distances to calculate the vertical dispersion term in the Gaussian equation.
2.3 Existing Area Source Models
Model algorithms for area sources are contained in dispersion models
designed for a wide variety of applications. The primary focus of this study
is area source algorithms suitable for estimating concentrations due to
routine (non-accidental) air emissions from Superfund sites and other areas
contaminated with toxic materials. Models designed for regulatory
applications to estimate concentrations for time periods ranging from one hour
to one year over distances between 10 m and 10,000 m are most relevant. Area
source algorithms in five existing short-term models designed to predict
concentrations based on hourly meteorological conditions were reviewed (ISCST,
FDM, PAL, RAM, SHORTZ). Six long-term (sector-average) models designed to
predict concentrations based on the frequency distribution of meteorological
2-6
-------
conditions were also reviewed (ISCLT, FDM, CDM, VALLEY, LCNGZ, AQDM). A
number of other models designed for air toxics or other applications were also
included in the review of area source algorithms. Table 2-1 characterizes the
area source algorithms in some widely used models.
2.3.1 Industrial Source Complex Short-Term Model
I SCSI uses the finite line source approach to model area sources. Each
square area source is modeled as a single line segment oriented normal to the
wind direction. The line segment length is the diameter of a circle
containing the same area as the square. For estimating lateral dispersion,
the line source is located at the downwind edge of the area source. For the
vertical dispersion coefficient oz, ISCST uses a "virtual distance" Xy = X +
XQ, where X is the along-wind distance from the downwind edge of the area
source to the receptor, and XQ is the side length of the area source square.
The ISCST algorithm predicts zero concentration for a receptor located
within an area source. The ISC User's Guide recommends that receptors not be
placed within a distance of one side length from any area source. If
receptors are placed at closer distances, the source(s) should be subdivided.
The ISCST area source algorithm does not accurately account for
source-receptor geometry. The use of a single line source segment to simulate
a square area eliminates the along-wind distribution of source emissions. The
length of the line segment used by ISCST is independent of wind direction, but
the crosswind extent of an area source square changes with wind direction.
The effect of these simplifications is greatest at near-field receptors.
2.3.2 Fugitive Dust Model
The Fugitive Dust Model (FDM) was developed to model both short-term and
long-term average particulate emissions from surface mining and similar
2-7
-------
TABLE 2-1
CHARACTERIZATION OF AREA SOURCE ALGORITHMS
IN EXISTING MODELS
Model
Area Source Algorithms
(a) Short-Term Models
ISCST
FDM
PAL
RAM
SHORTZ*
single line segment with virtual
distance for vertical dispersion
multiple line segment
multiple line segment
upwind integration (infinite line
source)
line segment
- near-field: upwind integration
- far-field: single line segment
(b) Long-Term (Sector-Average) Models
CDM
ISCLT
FDM
LONGZ*
VALLEY
AQDM
upwind integration (point source
array)
virtual point source
virtual point source(s)
- near-field: upwind integration
- far-field: virtual point source
virtual point source
virtual point source
* ground-level source height only
2-8
-------
sources. FDM accounts for deposition losses as well as pollutant dispersion.
The area source algorithm is designed for application to rectangular area
sources with principal axes oriented north-south and east-west. FDM
subdivides each area source into five line source segments oriented normal to
the wind direction (see Figure 2-1). For short-term averages, each line
source is then modeled using the CALINE line source algorithm, which has been
incorporated into FDM.
CALINE is a line source model developed by the California DOT to determine
the dispersion of pollutants from highways. For each line segment, FDM/CALINE
predicts the concentration using the Gaussian model equations for a finite
line source. Each line segment contributes based on its upwind and crosswind
distance from the receptor and on ov and oz. On-source receptors are allowed,
but only the line segments upwind of the receptor influence the receptor.
FDM/CALINE requires a computational effort for each area source which is
roughly 10 times the computational effort for one point source.
The FDM/CALINE model uses the Pasquill-Gifford dispersion coefficients,
with adjustments to az to account for the initial dispersion associated with
vehicle-induced turbulence and the buoyancy of vehicle exhaust emissions, plus
averaging-time adjustments to ay. Consequently, the ay and oz values used by
FDM are larger than the unadjusted Pasquill-Gifford coefficients.
For long-term averages, FDM employs a sector average treatment of lateral
dispersion. Each area source is again divided into five line segments normal
to the wind direction. The emissions from the area source are apportioned
among the five line segments in proportion to their lengths. Each segment is
then modeled as a virtual point source, located at the center of the line
segment. The Pasquill-Gifford oz coefficients are used without adjustment for
long-term average calculations. For long-term averages, FDM was judged to be
similar to ISCLT and was not evaluated separately.
2-9
-------
2.3.3 Point, Area, Line-Source Modal
The Point, Area, Line-Source Model (PAL) Version 2.0 is a steady-state
Gaussian plume model. This model is recommended for source dimensions of tens
to hundreds of meters. The area source algorithm uses multiple line sources
arranged perpendicular to the wind to simulate a rectangular area source.
The program initially calculates concentrations from 9 line sources spaced
equally across the source area. The Gaussian model equations for a finite
line segment are used. Next, PAL computes the concentrations from 10 line
segments located midway between each pair of the original nine lines and the
two corners of the source area (see Figure 2-2).
PAL estimates the accuracy of the computed concentrations, based on the
difference between the two concentration estimates (Cg - CIQ) divided by the
average % (Cg + CIQ). If this relative difference exceeds a user-specified
accuracy limit, PAL continues with further calculations. (The User's Guide
recommends an accuracy limit of .02.) For the third iteration, 20 lines are
placed midway between the 19 lines (and two corners) used previously.
Iterations continue until the accuracy limit is satisfied.
PAL requires a computational effort for each area source which is at least
20 times the effort for a point source. PAL Version 2.0 uses the
Pasquill-Gifford rural dispersion coefficients when the rural option is
selected.
2.3.4 Gaussian-Plume, Multiple Source Air Quality Algorithm
The Gaussian-Plume Multiple Source Air Quality Algorithm (RAM) is a
steady-state Gaussian plume dispersion model. RAM was developed for use in
urban and rural areas with low relief.
The area source algorithm in RAM calculates concentrations for a grid of
square area sources based upon contributions from sources located directly
2-10
-------
FIGURE 2-2
Subdivision of Rectangular Area Source
Into Line Segments by PAL
WIND
AREA
SOURCE
10 LINE SOURCES LOCATED MIDWAY
BETWEEN ORIGINAL 9 LINE SOURCES AND
CORNERS
©
RECEPTOR
IF *10 DIFFERS FROM THE AVERAGE
OF x9 AND *10 BY LESS THAN THE
TEST VALUE THEN RESULTS ARE
AVERAGED AND NEXT AREA SOURCE IS
ANALYZED
2-11
-------
upwind of the receptor. This algorithm uses the "narrow plume
simplification," which assumes that the emissions density directly upwind of
the receptor is representative of emissions at all crosswind locations. This
assumption allows the concentration at a receptor to be calculated using the
Gaussian model equation for an infinite line source. RAM computes
concentrations due to area sources along the path upwind of the receptor using
a distance spacing which starts at 10 m for near-field calculations and
increases in four steps to a maximum of 1,000 m at distances beyond 15 km.
The narrow plume hypothesis is most suitable in situations where there is
relatively little lateral variation in source strength, such as an extensive
urban area. In the case of an isolated area source, this simplification is
not valid, except for near-field receptors downwind of the center of the
source.
2.3.5 SHORTZ Model
SHORTZ is a short-term dispersion model suitable for use in flat or
complex terrain. SHORTZ calculates concentrations for ground-level area
source squares utilizing a finite line source approach. Beyond a distance
equal to 3 times the source width, the area source is simulated as a single
line source segment oriented normal to the wind direction. The length of the
segment is the crosswind projection of the source area. For near-field
receptors, the vertical term is calculated as an integral from the downwind
edge to the upwind edge of the source. The lateral term is calculated with
the line segment located at the center of the source area. (This separation
of the lateral and vertical terms in the Gaussian equation is only correct for
receptors downwind of the source center.) SHORTZ accept a user-specified
"source height" to produce an initial az for area sources, but a ground-level
release height is always assumed. The dispersion coefficients used in SHORTZ
2-12
-------
are based on the Pasguill-Gifford dispersion curves, but they do not
correspond to the Pasquill-Gifford ay and az values used in EPA models.
2.3.6 TEM/PEM
The Pollution Episodic Model (PEM) is a Gaussian plume model which
includes deposition/settling and first order chemical transformations. PEM is
an urban scale model and is designed to incorporate multiple-point and area
sources. It is designed to calculate short-term, ground level concentrations
and deposition velocities. PEM incorporates the framework of the Texas
Episodic Model (TEM) and the TEM area source treatment. PEM and TEM use a
point source array to estimate area source concentrations. The emissions from
area sources are apportioned among a rectangular array of points. The
distance spacing of the point array is specified by the user. Area source
contributions are calculated only for sources located within 8 distance
increments of a receptor. This area source technique has serious limitations,
and the TEM model developer does not recommend the use of this model for area
sources. Based upon the initial review, neither PEM nor TEM was chosen for
further evaluation in this study.
2.3.7 ISCLT
The algorithm used in ISC long-term {ISCLT) to model square area sources
is a virtual point source approach. The concentration calculations in ISCLT
are based on a 22.5° sector-average plume width. The distribution of
concentration is smoothed between radial sectors using a smoothing function to
produce a continuous distribution of concentration vs. direction. For
estimating lateral dispersion, the location of..the virtual point source is
displaced upwind of the area source center so that the sector width at the
source is equal to the diameter of a circle with the same area as the square.
2-13
-------
For estimating vertical dispersion, the point source is located at the center
of the source area.
2.3.8 Climatoloqical Dispersion Model
The Climatological Dispersion Model (CDM) is used to model long-term
average pollutant concentrations (seasonal or annual) using average emission
rates and a joint frequency distribution of wind direction, wind speed and
stability.
The COM area source algorithm is designed to calculate concentrations for
a grid of area source squares. CDM generates a radial array of points around
each receptor based on angular and radial spacing (DINT and DELR) specified by
the model user. CDM overlays the polar coordinate point array over the
square area source grid and assigns emissions to each point based upon its
location and the area (in the polar coordinate array) that each point
represents. The process is illustrated in Figure 2-3. The point source array
is then used to calculate concentrations. CDM allows a minimum angular
spacing of 1.25° and a minimum radial spacing of 10 m for the point array.
The radial spacing increases with distance according to the following table:
Radial Distance Radial Spacing
< 2500 m DELR
2500-5000 m 2 * DELR
>5000 m 4 * DELR
CDM computes contributions for a maximum of 100 radial increments. The
size of an area source and its distance from the receptor can only be resolved
in multiples of DELR. A small radial spacing provides improved spatial
resolution but reduced distance range... If DELR is 25 m, the distance range is
2.5 km. Conversely, as DELR increases the spatial resolution of the point
array decreases.
2-14
-------
la.
„ EMISSION
GRID
Figure 2- 3
Illustration of sector integration used in COM. In this figure, R is
receptor location, Pm is the maximum distance to the edge of the grid
from the receptor, DELR is the upwind step width and n is the step
number, (reproduced from Irwin et al., 1985)
2-15
-------
2.3.9 LONGZ Model
The LONGZ model is the long-term average version of SHORTZ. The area
source algorithm in LONGZ is designed for application to square, ground-level
area sources. A virtual point source approach is used for distances beyond 3
source widths. The source location is displaced upwind for simulating lateral
dispersion in a manner identical to ISCLT. For vertical dispersion, the point
source is located at the area source center.
For near-field concentration estimates, LONGZ calculates vertical
dispersion based on the integral of the vertical term in the Gaussian equation
between the downwind edge and the upwind edge of the source area. The
procedure for lateral dispersion is not affected. Aside from differences
related to dispersion coefficients, LONGZ closely resembles ISCLT.
Consequently, LONGZ was not evaluated separately in this study.
2.3.10 VALLEY Model
The VALLEY model was developed primarily to simulate dispersion from point
sources in the deep valleys typical of mountainous terrain. The model employs
a sector-average treatment of lateral dispersion to simulate worst-case
dispersion conditions. The area source algorithm used in this model
incorporates the virtual point source approach for a square area source. For
vertical dispersion, the area source is modeled as if the emissions were due
to a point source located at the center of the source area.
For lateral dispersion, VALLEY treats three types of source-receptor
relationships: (1) far-field, where the entire source area contributes to a
given receptor; (2) near-field, where only a portion of the source area
affects a given receptor; and (3) the receptor is inside the source area. The
first two cases are illustrated in Figure 2-4.
2-16
-------
VIRTUAL SOURCE
LOCATION FOR A
A' • AREA "SEEN" BY
RECEPTOR 2
I
22.SO SECTOR
WIND
DIRECTION
RECEPTOR 1
SOURCE AREA A
(TOTAL SQUARE)
Figure 2- 4
Schematic of the virtual point source as projected from an
area source, (reproduced from Burt, 1977)
2-17
-------
In the first case. Receptor 1 is sufficiently distant from the source that
the entire area is encompassed by a 22.5° wedge directed from the receptor
toward the source. In the far-field case, the virtual point source is
displaced upwind of the source center by a distance such that the 22.5° sector
width at the area source center matches the width of the area source.
In the second case. Receptor 2 is close enough to the source that a 22.5°
wedge directed from the receptor toward the source does not encompass the
entire source. In this case, the source emissions are reduced based on the
portion of the square outside of the 22.5° sector. The upwind displacement of
the virtual point source in this case is based on the area source width,
reduced to account for the area outside of the sector.
In the third case, a receptor is within the boundaries of the area source
and is only affected by the upwind portion of the source area. The same
procedures used for case 2 are applied for this situation.
2.3.11 Air Quality Dispersion Model
The Air Quality Dispersion Model (AQDM) was designed primarily for
estimating long-term averages of S(>2 and particulates in urban areas. AQDM
uses a virtual point source approach to modeling area sources. The area
source treatment in AQDM closely resembles that in ISCLT. AQDM was not
evaluated separately.
2.3.12 Area Source Screening Techniques
A basic technique to estimate short-term pollutant concentrations due to
area sources is given in the EPA document "A Workbook of Screening Techniques
for Assessing Impacts of Toxic Air Pollutants" (McNaughton and Bodner, 1988)
which is based on the work of Turner (1970). This approach uses the virtual
point source method. Area source width is used to calculate an approximate av
2-18
-------
at the area source using properties of the Gaussian distribution. A set of
curves is provided which relates ay to downwind distance for each
Pasquill-Gifford stability class. The distance thus obtained is termed the
virtual distance (xv). The virtual distance is then added to the actual
distance between the receptor and the area source. This combined distance is
then used to obtain av and oz to be used in the calculation of concentration.
The concentration is calculated in a straightforward manner using emission
rate, wind speed, the plume dispersion parameters and source height. The
formulation avoids integration (or the numerical equivalent) making it a
"calculator friendly" method.
Another screening method for estimating concentrations of pollutants from
an area source is given by New York DEC, Division of Air Resources. In this
treatment, an average per area emission rate and coefficient based on area
size are used to calculate concentrations within the area source. A table of
concentration reduction factors versus distance is presented. As a first
approximation, this technique is valid for areas with sides greater than 350
feet in length and for -distances closer than 3S of the source, where S is the
source side length.
These techniques are intended to provide preliminary, worst-case
concentration estimates to help the modeler decide whether more detailed
analysis is necessary. Screening techniques are not designed to account
accurately for source-receptor geometry and were not evaluated in this study.
2.3.13 Toxic Release Models
Chemical spill or accident models are of interest primarily because of the
chemical libraries included in the codes and the provisions for treating
evaporation, heavier-tnan-air gases, or other specialized problems. The spill
models generally contain area source algorithms because when a liquid is
2-19
-------
released, it forms a pool which evaporates and acts as an area source. Many
of these models have relatively sophisticated algorithms to deal with chemical
considerations, but most use very simple area source algorithms.
The simplest (very conservative) approach to calculating dispersion from
an area source is to use the area of a pool to calculate total evaporation
from the pool and then to associate the amount evaporated with a point source
at the center of the pool. This approach obviously is not adequate in the
near-field in the case of extensive pools. The Areal Locations of Hazardous
Atmospheres Model (ALOHA) is a spill model developed by NOAA which provides
graphical output of dispersion plumes due to chemical spills. ALOHA uses this
point approximation of the pool source as described.
The model entitled "A Portable Computing System for Use in Toxic Gas
Emergencies by the Ontario Ministry of the Environment" (known as the OME
Toxic Release Model) has an extensive chemical library but also treats the
pool of spilled chemicals as a point source.
The Air Force Toxic Chemical Dispersion Model (AFTOX) is a Gaussian
puff/plume dispersion model which was developed to model emissions from
chemical spills. AFTOX is a coupled emissions/dispersion model and uses an
inventory of chemicals and corresponding chemical properties to model
dispersion from instantaneous or continuous leaks or spills of any chemical in
the inventory. The emission algorithm predicts whether a given spilling
chemical will be in gaseous or liquid form based on an ambient temperature
input. The area of the source is then calculated if the chemical is in liquid
form, based on the spill rate. A source strength is then calculated based on
an evaporation algorithm. The dispersion algorithm itself is non-reactive and
no decay or deposition occurs. It is a Gaussian plume model under steady
state, non-stable conditions and it is a Gaussian puff model otherwise. AFTOX
uses a standard virtual point source algorithm to simulate an area source.
2-20
-------
The Shell Development Company Evaporation/Air Dispersion Model for
Chemical Spills on Land (SPILLS) is another model to simulate the properties
of an accidental chemical release. This model also uses the virtual point
source method to model area sources. An initial ov is assumed based on area
source width and characteristics of the Gaussian distribution, a virtual
distance is then calculated.
None of these models contains an area source algorithm which warrants
evaluation for the present study.
2.3.14 Other Existing Models
A number of other existing models were identified which contain algorithms
for treating distributed (area and volume) emission sources. These models
either were designed for applications distinctly different from the air
quality issues pertaining to Superfund/contaminated sites, or contain area
source algorithms which were judged to be redundant with algorithms from
models already chosen for evaluation. Listed below are three other models
which were reviewed:
• Mesopuff - Designed for modeling long-range transport. Not
appropriate for the time and distance scales of interest.
• Photochemical Box Model - Designed for modeling reactive
photochemical pollutants. Employs a gridded box model equivalent
to volume source. Not appropriate for near-ground emission
sources or near-field concentration estimates.
• APRAC1A - Designed for modeling gridded traffic emissions based on
line-source algorithm. Not appropriate for a single, isolated
area source.
2.4 Overview of Area Source Dispersion Literature
A search of recent technical literature identified few published articles
representing unique methods for estimating concentrations downwind of area
2-21
-------
sources. Two published articles were identified which discussed alternative
approaches to modeling area sources. Neither approach represents a computer
algorithm suitable for evaluation in the present study.
The models suggested by Hwang (1987) and Chitgopekar et al. (1988) attempt
to resolve some of the known shortcomings of the currently utilized area
source dispersion algorithms. The discussion by Hwang is theoretical but
examines both a Gaussian approach and one based on transport equations in the
atmospheric boundary layer. Chitgopekar et al. (1988) attempts to resolve
problems of near-field prediction through the use of a "top-hat" formulation.
Chitgopekar et al. (1988) presents an area source model developed in
response to problems with virtual point models in the near-field for area
sources. These authors state that "the most rigorous" treatment of Gaussian
dispersion from area sources would be to model them as a dense matrix of
multiple point sources. This idea can be conceptualized as increasing matrix
density until, in the limit, inter-point spacing goes to zero and every point
in the area is emitting. This approach is computationally intensive and is
rarely used in the standard models. (The finite line segment approach in FDM
or PAL is mathematically equivalent but far more efficient, if Gaussian
dispersion is assumed.) Virtual point source methods do not require extensive
computations and can be simplified to allow manual calculation. However, the
virtual point source method should not be used if the source width is greater
than 40* of the distance between the source centerpoint and the receptor
(Hwang, 1986 as cited in Chitgopekar et al., 1988). This is a serious
limitation due to the fact that the region of interest in many area source
pollutant dispersion and exposure situations is in the near-field.
The treatment by Chitgopekar et al. is to divide the dispersion of the
plume into high frequency and low frequency components. A low pass filter is
used so plume meander is included but small turbulent scales are not. The
2-22
-------
model uses a multiple line source approach. By eliminating high frequency
fluctuations, the distribution immediately downwind of the area source
approximates a top hat, with a region of uniform concentration in the across
wind direction, downwind of the source. This treatment is equivalent to
modeling the near-field region downwind of the source as if the source length
was infinite. Edge effects are eliminated by filtering of the smaller (high
frequency) fluctuations. These authors have developed this model so that it
can be used with Pasquill-Gifford stability classes to indicate plume meander
or more sophisticated boundary layer parameters obtained through eddy
diffusion (diffusion analogues or similarity) equations. The authors believe
that the "top hat" distribution of concentration in the near-field is more
realistic that the Gaussian plume shape predicted by the virtual point source
method.
Hwang (1987) gives a method for estimating on-site concentrations at toxic
waste disposal sites. Gaussian dispersion models using the virtual point
method perform poorly in this type of near-field application. Hwang states
that when using a virtual point approximation for an area source of
pollutants, the receptor should be at least the distance from source center to
the property line or 100 m whichever is greater, downwind of the source center.
Hwang suggests two models for near-field (on-site) dispersion modeling.
The first is a mathematical formulation of the multiple point source idea. By
treating the area as being composed of differential point sources, the
concentration contributions to an on-site receptor from each differential
source area can be summed. This approach utilizes the standard point source
Gaussian concentration equations.
The second method utilizes a partial differential equation describing
atmospheric dispersion. This formulation utilizes wind speed and transfer
coefficients to yield concentration. The transfer coefficients are
2-23
-------
parameterized using the standard deviations of pollutant spreading. The
equation can then be solved with the appropriate boundary conditions.
Hwang compares the results of these formulations with results obtained
using a simple box method. The two formulations given in this paper agree
quite closely. The box model yields values of concentration greater by a
factor of two than either of the methods proposed by Hwang.
Hwang and Chitgopekar have demonstrated the advantages of alternative area
source methods compared to either the virtual point source approach or a box
model. Neither author, however, discussed the finite line segment approach
which is widely used in short-term models. The proposed methods appear to
offer few advantages when compared to the multiple line segment approach of
FDM or PAL.
2.5 Limitations of Existing Models
Gaussian dispersion models have inherent limitations which are accepted
for reasons of convenience and simplicity. The errors induced by simplifying
Gaussian assumptions may be compounded by the treatment of area sources.
General problems with many Gaussian plume models are:
1) The effects of atmospheric turbulence are not well represented by
a Gaussian plume distribution. To represent dispersion of
pollutants by atmospheric turbulence by a Gaussian distribution
introduces error.
2) Characterization of the mixing layer as a closed surface layer may
introduce error. Escape of pollutants at the top of the mixing
layer and settling/deposition at the surface should both be loss
terms in a correct formulation. Some of the models include
simplified treatments of settling/deposition.
3) Many of the dispersion models have optional decay formulations
which allow a constant chemical half-life to be entered. Many
VOCs have chemical reaction rates which vary substantially as a
function of solar radiation, ambient temperature, humidity, and
the presence of other chemical pollutants.
4) Most of the dispersion models do not treat low wind speed
scenarios adequately. Zero wind speed values cause computational
2-24
-------
failure and very small values may cause unrealistically high
concentration predictions. This problem is typically avoided by
arbitrarily setting all wind speeds less than 1 m/s to be equal to
1 m/s.
The limitations noted above and uncertainties relating to emissions
estimates as described in Section 1.2 contribute to a number of specific
problems which might be anticipated based on the source characteristics of
contaminated sites and similar area sources.
1) Since many area sources are true surface sources, low wind
velocities will occur frequently at the source (from a
computational standpoint, a "no-slip" or zero velocity boundary
condition may force velocity to be zero at the surface).
2) Diffusion may be the dominant process in the initial stages of
chemical releases from many surface area sources. A more
sophisticated treatment of this process may be needed.
3) Emissions from many area sources may be dominated by VOCs which
are chemically reactive and have half-lives which are strongly
dependent on environmental conditions.
The dispersion models discussed here do not contain emission models. In
the case of manufacturing processes, it may be reasonable to model with
constant emissions. For the types of pollutant sources discussed here,
steady-state emissions will not be a reasonable assumption. Ambient
temperature, wind speed, moisture, and site geometry are important factors in
determining source strength in many of these situations.
2-25
-------
3.0 ANALYSIS OF MODEL PREDICTIONS FOR EXAMPLE APPLICATIONS
The analysis of model predictions for a set of hypothetical area source
test cases is designed to accomplish two objectives:
• to characterize the range of concentration predictions expected
from each model, and the differences between models, as a function
of input meteorology, source characteristics, and receptor
locations.
• to examine whether model predictions are consistent with a number
of basic mathematical and physical principles.
3.1 Approach
To achieve these goals, each model was applied to a standard set of
prediction scenarios which represented variations on a "base case"
source/receptor configuration. The "base case" source and near-field
receptors are illustrated in Figure 3-1. A single square area source with
dimensions 150 m x 150 m was chosen. The wind direction is parallel to one
side of the source. Six receptors were replaced at selected distances
downwind of the area source center ("centerline" receptors). The receptor
distances are 100 m, 125 m, 175 m, 325 m, 875 m, and 1575 m downwind of the
area source center. (The first four centerline receptors, numbered R1-R4, are
shown in Figure 3-1.) A second group of receptors was placed downwind of the
source edge. These "edge" receptors are numbered R7-R10. In addition, model
predictions were obtained at two "on-source" receptors, Rll and R12, as
shown. Flat terrain was assumed.
Concentration predictions were obtained from each model for these 12
receptors, using an emission rate of 1,000 g/s. Three different stability
conditions (Class B, D, and F) were modeled, with a constant wind speed value
of 2 m/s and mixing height of 5000 m. For Class D stability, both a ground
3-1
-------
i
U
o:
u
»-
z
u
o
8
Cl
8
•ON
I
•
CO
•00
•fc
t*
is
£
^x
U
(/>
i—i
Q
Q
n
o
o
O
in
OJ
(J
U
E
o
in
3-2
-------
level source and a 10 m source height were modeled. For Class B and Class F,
only the ground level source was modeled. Rural Pasquill-Gifford (P-G)
dispersion coefficients were selected for each model.
3.2 Tests of Mathematical and Physical Principles
A series of test cases representing variations on the "base case" source
geometry were devised to examine whether concentration predictions from each
model behave in a manner consistent with mathematical and physical
principles. These tests are designed primarily to examine concentration
predictions for low-level emission sources (height 10 m or less) and for area
sources ranging from 50 x 50 m to 500 x 500 m. The primary focus is D
stability, which represents the most common (but not the worst-case) stability
condition. The following tests were applied:
Stability Comparison. For a ground level source, predicted centerline
concentrations should increase with atmospheric stability (Class B = lowest.
Class F = highest). The influence of stability class should be more
pronounced at larger distances, since horizontal dispersion near the source is
dominated by initial source dimensions.
Center Versus Edge (Near-Field). Immediately downwind of the source,
predicted concentrations at edge receptors should be roughly one-half of the
concentrations at corresponding centerline receptors when the wind direction
is parallel to the source side. For a Gaussian model, the sources which
produce significant predicted impact at a given receptor are those located
upwind of the source. As the crosswind distance between source and receptor
increases, the relative impact decreases according to the Gaussian equation.
A source at a crosswind distance of 3.03 ay produces one percent of the impact
of an identical source located directly upwind of the receptor. The source
region which lies within ±3
-------
the predicted impact. At near-field distances, the "zone of influence" within
±3 ay upwind of the centerline receptor is entirely filled by the area source,
as shown in Figure 3-2(a), while the corresponding zone upwind of the edge
receptor is half empty. The empty and filled portions of the influence zone
for the edge receptor are mirror images. The predicted concentration should
therefore be one-half of the concentration predicted when the entire zone is
filled.
Subdivision. The base case scenario is divided into four equal parts with
the same total emissions, as shown in Figure 3-2(b). No significant change in
predicted concentrations should result, since the overall configuration of
source(s) and receptors is identical.
Far-Field Convergence. At downwind distances which are large relative to
source dimensions, predicted concentrations should depend on total emissions
but not on source size. (At downwind distances where Oy is as large as the
area source width, the plume from an area source will be equivalent to the
plume from a point source with the same emission rate.) Centerline
concentrations at distances out to 8 km were compared for three area sources
with identical emissions: 150 m x 150 m (base case), 75 m x 75 m, and 450 m x
450 m. At 8 km, oy for 0 stability is roughly 450 m, so model predictions
should be similar for all three sources.
Source Orientation. The orientation of the wind direction relative to the
area source will influence predicted concentrations near the source.
Predictions for the base case were compared to predictions for the
configuration shown in Figure 3-2(c), with the wind direction at a 45° angle
to the source axes. Centerline receptors were placed at the same distances
downwind of the source's center. Two results are expected:
1. Predicted centerline concentrations near the source should be
higher for the "diagonal" configuration, because a larger source
area falls within the "zone of influence".
3-4
-------
I* .
LJ
£ ™
u ^ w
Z^L^
i—i ^* C-0
-------
«co
o:
u
-p
Qj
OJ
E
ID
E
CD
LO
0)
u
L
3
O
(X)
Ci
0)
ui
<3
u
0)
(f\
C5
CQ
U-
O
c
O
ui
cu
I
n
cu
i.
3
O)
CI
3-6
-------
LJ
\
\
\
E
CD
LO
\
E
o
LD
-------
2. Farther from the source, the two configurations should produce
similar concentrations. (See the discussion of "far-field
convergence".)
Source Height. The use of non-zero source height should produce lower
predicted concentrations in the near-source region. Differences should
decrease as downwind distance increases. Predictions for a 150 m x 150 m
source with 10 m height were compared against the base case. At 875 m
downwind distance, oz for 0 stability is roughly 28 m, and centerline
concentrations predicted for a 10 m source height should be about 94 percent
of predictions for a ground-level source.
All of these tests are well-suited for testing short-term (hourly)
models. Some of the tests are more difficult to interpret for long-term
models, which are designed to predict sector-average concentration values
based on wind direction frequency. The center/edge comparison and the source
orientation test were only applied to the short-term models.
3.3 Predicted Concentrations for Base Case
3.3.1 Short-Term Models
Predicted concentrations for the base case source-receptor configuration
were obtained for three stability conditions (Class D, Class B, and Class F)
at a constant wind speed of 2 m/s. The centerline concentrations predicted by
the five short-term area source models are illustrated in Figures 3-3, 3-4,
and 3-5. In these figures, the downwind distance is measured from the center
of the area source. For all three stability conditions, a similar pattern is
evident. FDM consistently predicts the lowest concentration values at all
distances. SHORTZ generally predicts the highest concentrations within 150 m
of the source, while RAM predicts the highest concentrations beyond 500 m.
PAL and ISCST predict intermediate values at all distances.
3-8
-------
At small downwind distances, the concentrations predicted by RWi ara lower
than these predicted by PAL and SHORTZ. (The closest distance, 37.5 m,
represents receptor Rll, located within the area source. See Figure 3-1. The
ISCST model will not predict concentrations inside of an area source.) The
gap between predictions by RAM and the other four models increases with
downwind distance for all three stability conditions.
At a distance of 100 m, centerline concentration predictions for D
stability (Figure 3-3) range from a low value of 2.6 x 10s ug/ra3 predicted by
FDM to a maximum value of 7.5 x 10s ug/m3 predicted by SHORTZ. At 1575 m
downwind, centerline predictions range from 1.3 x 10* ug/m3 (FDM) to 6.2 x 104
ug/m3 (RAM). Near the source, at distances between 100 m and 175 m, PAL and
SHORTZ give nearly identical centerline predictions. As distance increases,
SHORTZ predictions decrease faster than PAL's. The ISCST prediction at 100 m
is roughly a factor of 2 lower than PAL or SHORTZ, but ISCST and PAL
predictions converge as distance increases.
For B stability (Figure 3-4), SHORTZ predicts the highest centerline
concentration at 37.5 m (on the source). At distances beyond 200 m, RAM
predictions are the highest and FDM predictions are lowest. Predictions by
PAL and SHORTZ are very similar at all distances for B stability. Predictions
by ISCST are lower in the near-field but converge with PAL and SHORTZ at
375 m. At 1575 m, the RAM concentration is a factor of 4 higher than all
other models.
For F stability (Figure 3-5), SHORTZ predicts the highest centerline
concentrations at 38 m and 100 m. FDM predicts the lowest concentrations at
all distances. SHORTZ predictions span the largest range for F stability,
since SHORTZ predicts the highest concentration at 37.5 m and matches FDM for
the lowest concentration at 1575 m. PAL predictions agree with SHORTZ near
3-9
-------
the source, while ISCSI predictions are lower by about a factor of 2. At
greater distances, however, PAL, RAM and ISCSI predict similar values.
These base case results illustrate the effects of differences between area
source algorithms and in the models' treatment of dispersion. The RAM "narrow
plume simplification" apparently causes RAM predictions to diverge from the
other models at large distances for B and D stability. SHORTZ and PAL
generally produce similar near-field predictions, which indicates that their
area source algorithms are similar (from an operational perspective). ISCST
and PAL generally agree at 875 m and 1575 m, but disagree closer to the
source, indicating differences in the area source algorithms, but similar
dispersion treatment. FDM's treatment of both the area source (near-field)
and dispersion coefficients is unique among this group of models.
The magnitude of centerline concentrations predicted by the short-term
models changes significantly with source height. Predicted concentrations for
a 10 m source height for D stability are illustrated in Figure 3-6. The
near-field concentrations predicted by RAM and PAL decrease by more than a
factor of 10 when the source height changes from zero (Figure 3-3) to 10 m
(Figure 3-6). The maximum predicted concentration in Figure 3-6 is 1.9 x 10s
ug/m3, for SHORTZ, versus 1.8 x 10* ug/m3 for SHORTZ in Figure 3-3. The
relative rank of predictions by different models changes with source height at
distances out to 175 m. For example, FDM predicts higher concentrations than
PAL or RAM for the 10 m source height.
3.3.2 Long-Tenn Models
The concentrations predicted by the three long-term (sector average)
models for the base case are illustrated in Figures 3-7 through 3-10. For a
ground-level 150 x 150 m source and D stability (Figure 3-7) ISCLT predicts
the highest centerline concentrations near the source, and COM predicts the
3-10
-------
highest values beyond 200 m. VALLEY predicts the lowest near-field
concentrations. Beyond 500 m, ISCLT and VALLEY predict similar values.
For B stability (Figure 3-8), CDM predicts the highest near-field
concentrations. Beyond 500 m, CDM and VALLEY predict similar values. ISCLT
predicts higher near-field values than VALLEY, but predicts the lowest
concentrations beyond 300 m.
For F stability (Figure 3-9), VALLEY predicts the highest concentrations,
CDM predicts the lowest near-field concentrations, and ISCLT predicts the
lowest concentrations beyond 800 m.
For a 10 m source height and D stability (Figure 3-10), ISCLT predicts the
highest concentrations at 100 m and 125 m, while CDM predicts the highest
values beyond 200 m. Predictions for these long-term models were less
sensitive to the change in source height than predictions from several of the
short-term models.
3.4 Tests of Mathematical and Physical Principles
The results obtained for the tests of physical principles defined in
Section 3.2.1 are described below for each area source model.
3.4.1 ISCST
Stability Comparison. The centerline concentrations predicted by ISCST
for the base case configuration for Class B, D, and F stability are compared
in Figure 3-11. As expected, predictions increase as stability increases.
The relative difference in predicted concentrations also increases with
distance from the source. Concentrations for B and F stability differ by a
factor of 5 at 100 m and a factor of 30 at 1575 m.
Near-Field (Center Versus Edge). Predicted concentrations at receptors Rl
and R7 are identical. Similar results were obtained at other near-field
3-11
-------
receptors. (If ISCSI correctly accounted for source geometry, the
concentration at R7 would be one-half the value predicted at Rl.)
Subdivision. Predicted centerline concentrations are shown in Figure 3-12
for the single 150 m x 150 m source and the subdivided source (four 75 m x
75 m sources, as shown in Figure 3-2a). At 100 m distance (25 m from the
downwind edge of the source area), ISCSI predicts concentrations different by
a factor of 2 for these equivalent source configurations. At 325 m,
concentrations differ by 25 percent. Higher concentrations are predicted for
the subdivided source. These results are not physically reasonable.
Source Orientation (Figure 3-13). Centerline concentrations predicted by
ISCST are nearly identical for normal and diagonal wind directions. I SCSI
predictions do not reflect the difference in source-receptor geometry between
these two cases.
Source Height. Centerline concentrations predicted by ISCST for a 10 m
source height are lower than those predicted for a ground level source, as
shown in Figure 3-14. Differences are largest near the source, exceeding a
factor of 2 at 100 m, and diminish as distance increases. These results are
physically reasonable and consistent with P-G dispersion coefficients.
Far-Field Convergence. Centerline concentrations predicted by ISCST are
shown in Figure 3-15 for three area sources with equivalent total emissions
but different areas (75 x 75 m; 150 x 150 m; and 450 x 450 m). Differences
are largest near the source and diminish as distance increases. The smallest
area source produces the highest predicted concentrations. At 100 m distance,
the centerline value for the 75 x 75 m source is more than 5 times the
corresponding value for the 450 x 450 m source. These results are consistent
with source-receptor geometry and Gaussian dispersion.
Summary - ISCST. Results obtained for ISCST are reasonable for the tests
based on sensitivity to stability class and source height. ISCST did not
3-12
-------
correctly account for source-receptor geometry in the center/edge,
subdivision, and source orientation tests.
3.4.2 FDM
Stability Comparison (Figure 3-16). Predicted centerline concentrations
increase with stability, as expected. At 100 m distance, concentrations for B
and F stability differ by only a factor of 2. At 1575 m, the difference is a
factor of 20.
Center Versus Edge. Predicted concentrations at receptors downwind of the
source edge (R7, R8) are one-half of the corresponding values at centerline
receptors. This result is consistent with source-receptor geometry.
Subdivision (Figure 3-17). Predicted centerline concentrations are
approximately 20 percent lower at 100 m distance for the subdivided source,
and 10 percent lower beyond 500 m. These differences represent an inaccurate
treatment of source geometry.
Source Orientation (Figure 3-18). Predicted centerline concentrations are
higher near the source for the diagonal orientation, consistent with the
source-receptor geometry. Results converge at greater distances.
Source Height (Figure 3-19). Predicted centerline concentrations at 100 m
from the source are a factor of 2 lower for the 10 m source height than a
ground level release. At 325 m and 875 m, the differences are smaller than
values estimated using P-G dispersion coefficients.
Far-Field Convergence (Figure 3-20). In the near-field region, predicted
centerline concentrations decrease as source area increases. As distance
increases, results for all three source sizes converge.
Summary - FDM. FDM provides physically reasonable results for all but one
of the tests, indicating that the treatment of source-receptor geometry is
generally accurate. The subdivision test results suggest some room for
improvement.
3-13
-------
3.4.3 PAL
Stability Comparison (Figure 3-21). Canterline concentrations increase
with increasing stability. Differences become larger as distance increases,
growing from a factor of 5 at 100 m to a factor of 50 at 1575 m.
Center Versus Edge (Near-Field). Predicted concentrations at edge
receptors near the source (R7, R8) are one-half the corresponding centerline
concentrations. This result is consistent with source-receptor geometry.
Subdivision (Figure 3-22). Predicted centerline concentrations for the
subdivided source are almost exactly equal to the corresponding values
predicted for the single source at all distances. These results are
physically reasonable.
Source Orientation (Figure 3-23). Predicted centerline concentrations for
the diagonal source orientation are higher than the values for a normal wind
direction. Differences become insignificant as distance increases.
Source Height (Figure 3-24). Predictions in the near-source region are
very sensitive to source height. With a 10 m release height, the
concentration at 100 m is lower by a factor of 10. Predictions converge as
distance increases.
Far-Field Convergence (Figure 3-25). Predictions for three area source
sizes converge at 8000 m from the source. In the near-field, concentrations
decrease as source size increases. These are the expected results for a
properly functioning model.
Summary - PAL. The results for PAL are physically reasonable for all
tests. Predictions are very sensitive to source height.
3.4.4 RAM
Stability Comparison (Figure 3-26). Centerline predictions by RAM
increase with stability. The difference between B and F stability increases
from a factor of 4 at 100 m to a factor of 10 at 1575 m.
3-14
-------
Center Versus Edge (Near-Field). RAM predicts identical concentrations
for centerline receptors (Rl, R2) and for edge receptors (R7, R8). For this
test, RAM did not correctly account for source-receptor geometry.
Subdivision (Figure 3-27). RAM correctly predicts identical centerline
concentrations for the single source 150 x 150 m base case and the subdivided
(identical) source.
Source Orientation (Figure 3-28). RAM predicts higher centerline
concentrations for the diagonal wind direction than for the normal direction.
The results are reasonable near the source. At larger distances, predictions
are 25 percent higher for the diagonal orientation. These results are not
consistent with conservation of mass.•
Source Height (Figure 3-29). RAM predictions are extremely sensitive to
source height. Centerline predictions for 10 m source height are lower than
predictions for a ground level source by a factor of 10 at 100 m. Predictions
for these two cases converge as distance increases.
Far-Field Convergence (Figure 3-30). RAM predictions for three different
area source sizes do not show convergence even at distances beyond 2000 m. In
fact, the relative differences remain constant between 800 m and 8000 m.
Summary - RAM. RAM predictions are not reasonable for a single, isolated
area source. (The "narrow plume hypothesis" employed by RAM is designed for
application to a large area source grid, and assumes that emissions vary
slowly between adjacent grid squares.) RAM does not produce physically
reasonable results for the center-edge, source orientation, and far-field
convergence tests.
3.4.5 SHORTZ
Stability Comparison (Figure 3-31). Predicted centerline concentrations
increase with stability at all distances. The difference between Class B and
3-15
-------
Class F predictions is a factor of 5 at 100 m but increases to a factor of 20
at 1575 m. These results are physically reasonable.
Center/Edge Comparison. Near-field concentrations predicted by SHORTZ at
"edge" receptors (R7.R8) are one-half of the corresponding centerline
predictions, consistent with the source-receptor geometry.
Subdivision (Figure 3-32). Predicted centerline concentrations for the
150 x 150 m source and the equivalent four 75 x 75 m sources are identical at
all off-source receptors. For the on-source receptor Rll, a higher
concentration is predicted for the subdivided source. The results indicate
that source geometry is correctly accounted for.
Source Orientation (Figure 3-33). SHORTZ predicts identical centerline
concentrations for diagonal and normal wind directions. SHORTZ does not
account for the different source-receptor geometry and its effect on
near-field concentrations.
Source Height (Figure 3-34). Predicted centerline concentrations near the
source are lower for the 10 m source height than for the ground level source,
as expected. Differences of 20 and 10 percent remain at 875 m and 1575 m,
respectively. These differences, while small, are larger than expected based
upon the Pasquill-Gifford vertical dispersion coefficients for D stability.
Far-Field Convergence (Figure 3-35). Predicted centerline concentrations
within 1000 m of the source are significantly different for the three source
sizes (75 x 75, 150 x 150, 450 x 450 m). The larger source areas produce
lower predictions. Centerline predictions converge at greater distances.
These results are physically reasonable.
Summary - SHORTZ. Results indicate that the SHORTZ area source algorithm
predicts reasonable concentration patterns for most scenarios, but does not
always respond to details of source-receptor geometry.
3-16
-------
3.4.6 ISCLT
Stability Comparison (Figure 3-36). Predicted centerline concentrations
for B, 0, and F stability increase with increasing stability. Differences
between B and F stability are a factor of 5 at 100 m and a factor of 10 at
1500 m.
Subdivision (Figure 3-37). ISCLT predicted higher concentrations for the
subdivided source area (four 75 x 75 m sources), compared to the single (150 x
150 m) source. The difference is approximately a factor of 1.5 at 100 m, and
a small difference (10 percent) remains at 1575 m.
Source Height Comparison (Figure 3-38). Centerline predictions by ISCLT
are not strongly affected by source height. Differences between predictions
for a 10 m source height versus a ground level source are less than a factor
of 2 at 100 m and decrease as distance increases.
Far-Field Convergence (Figure 3-39). Predictions for the 75 x 75 m and
150 x 150 m area sources converge gradually with increasing distance. At
8000 m, the centerline concentration for these sources is 20 percent higher
than the corresponding value for the 450 x 450 m source.
Summary - ISCLT. The area source predictions by ISCLT show differences
for the subdivision and far-field convergence test which are not consistent
with source geometry and Gaussian plume dispersion behavior.
3.4.7 COM
Stability Comparison (Figure 3-40). Centerline concentration predictions
by COM increase with increasing stability. The difference between B and F
stability is less than a factor of 2 at 100 m and increases to a factor of 6
at 1500 nr.
Subdivision (Figure 3-41). CDM's predictions for the subdivided source
are indistinguishable from the base case concentrations at all distances.
3-17
-------
Source Height (Figure 3-42). CDM predictions are not strongly sensitive
to source height. The difference between predictions for 150 x 150 m
ground-level and 10 m sources is only 25 percent at 100 m. Using P-G vertical
dispersion coefficients (CTZ) for D stability, a larger difference was
estimated at 100 m, but a much smaller difference was estimated at 1500 m.
Far-Field Convergence (Figure 3-43). The differences between
concentrations predicted for three area source sizes are relatively small at
near-field distances. The predictions appear to converge at 3km downwind, but
large differences are predicted by CDM at 8 km. For the 75 x 75m source, the
predicted concentration at 8km is higher than at 3km.
Summary - CDM. CDM predictions for the stability comparison and
subdivision tests are physically reasonable. For the source height and
far-field convergence tests, CDM predictions were not consistent with
source-receptor geometry and Gaussian plume dispersion behavior. Inaccuracies
may be related to the spatial resolution of the array used by CDM for upwind
integration.
3.4.8 VALLEY
Stability Comparison (Figure 3-44). Predicted centerline concentrations
for F stability are distinctly different from, and much higher than, those
predicted for B and D stability. Near the source, concentrations for B and D
stability are very similar, and differences increase with distance. The
near-field concentrations for F stability are greater by more than a factor of
10. No physical basis for these differences is apparent.
Subdivision (Figure 3-45). Predicted centerline concentrations increase
when the base case source is subdivided. VALLEY predicts concentrations 50
percent higher at 100 m and 5 percent higher at 1500 m.
3-18
-------
Source Height (Figure 3-46). Predicted concentrations for an elevated
area source are only 5 percent lower than corresponding predictions for a
ground level source. Much larger differences were estimated at distances of
100-200 m using the P-G dispersion coefficients. Larger differences were also
predicted by the other models.
Far-Field Convergence (Figure 3-47). Concentrations predicted by VALLEY
for three area source sizes converge gradually at distances beyond 800 m from
the source. At a distance of 8000 m, the predicted concentration for the
450 x 450 m source is 20 percent lower than values for the 75 x 75 m and
150 x 150 m sources.
Summary - VALLEY. The differences in concentrations predicted by VALLEY
as a function of stability, source geometry and source height are not
consistent with predictions by other models and do not correspond to Gaussian
plume dispersion behavior. Inaccuracy of 20 to 50 percent was found in
several cases.
3.5 Discussion of Results
Predicted concentrations were compared and analyzed for eight air quality
models applicable to area source emissions. The model scenarios used for this
analysis represent variations on a base case with a single 150 x 150 m area
source and receptors extending out to 1575 m from the source. Model
predictions were analyzed to characterize the range of concentrations
predicted as a function of stability class and source height. Test cases were
designed to identify changes in predicted concentrations associated with
specific changes in the source-receptor configuration. These tests were
analyzed to identify strengths and weaknesses in each area source algorithm,
independent of each model's treatment of atmospheric dispersion.
3-19
-------
3.5.1 Short-Term Models
Five short-term models (ISCSI, FDM, PAL, RAM, SHORTZ) were analyzed in the
greatest depth. In general, FDM predicted the lowest centerline
concentrations. Within 200 m of the source, SHORTZ predicted the highest
concentrations. Beyond 400 m, RAM predicted the highest values. (The low
concentrations predicted by FDM result primarily from differences in
dispersion rates, not from the area source algorithm.)
The tests of physical principles provide a basis for judging each area
source algorithm's strengths and weaknesses based on "absolute" performance
criteria, independent of any measured concentrations. The results are
summarized in Table 3-1 and are briefly discussed below:
Stability Comparison. All of the models' predictions agreed with the
expected changes in concentrations with stability class and downwind distance.
Source Height. All of the models predicted lower near-source
concentrations when the source height was increased from zero to 10 m. For
SHORTZ, however, predictions did not converge at a distance of 1575 m.
Far-Field Convergence. Predictions were obtained for three area sources
with different dimensions but the same total emissions. For all of the models
except RAM, predicted concentrations were independent of source size at a
distance of 8000 m.
Center/Edge. Near the source, predicted concentrations downwind of the
area source edge should be one-half of centerline concentrations. FDM, PAL,
and SHORTZ correctly predicted this behavior. RAM and ISCST predicted no
difference in concentration between centerline and edge receptors.
Subdivision. The base case area source was subdivided into four equal
parts, representing (in total) the identical source. PAL, RAM and SHORTZ
correctly predicted identical concentrations. FDM predictions were different
by 10 to 20 percent. ISCST predictions for the subdivided source were higher
by a factor of 2.
3-20
-------
rH
1
cn
w
EH
1
l
i
Q
«E
jg
Oft
W
1
1
i
OT
i
W
K
H
3
H
K
M
W
H
W
VI
§
on
fa
O
>
M
§
•H
01 4J
U «
Ll 4J
3 e
5.3
OT H
o
§
•H
01
•H
>
•H
;j
10
0)
*
ta
i*<
u
0)
4J
1
0)
•a o
rH C
O 01
•H OI
ELI U
1 0)
U >
(0 C
fa 0
u
jj
Oi
•H
0)
S
01
o
14
o
OT
e
>i O
jj a
•H »H
iH U
!a &
to e
-u 5
co O
4J
09
0)
H
"J "«
S c
S a>
£ «-•
s£ 8 8 s£ i
^^ "
•H '^
•o *o
§ §
^ ^^
? 0)
H 8
0) 0)
Ll •-! U
0) 10 0)
<4J C Ifc4
i^j -a a 8 8 8
u tj
8S -8 8 8 iS
•-s «
S I
8 8 8 § i
r* -. -^
? >i- >i
*J 4J 4J
* -H -H
'> > >
Tj -u J **
•tn
•H
. o
e£
«i!
Q]
iJ-S
most oi
erious
a
n
4J n
•H 3
3 O
n LI
01 01
{_! C
9
0) C
1—4
.£» -O
«} 0)
C S
0 0
m J3
10 01
2 %
>J
rH -.
Oi 2
0)
££
0
a 01
01
at jj
£ 01
4J a)
4J
^
S S
^^ ^
m 4J
u
0) -O
i «
O •-!
•H
(0
-------
Source Orientation. Centerline concentrations were compared for wind
directions along a source side and along the diagonal. Near-source impacts
should be higher for the diagonal direction, but differences should be
negligible at greater distances. ISCST and PAL predicted this behavior
correctly. FDM predicted near-source impacts correctly, but predicted
differences at large distances. ISCST and SHORTZ predicted no differences
near the source. RAM predicted substantial differences at large distances.
Overall, the FDM and PAL algorithms consistently gave physically
reasonable results. SHORTZ gave reasonable results except for the source
orientation test. The ISCST algorithm failed three tests sensitive to source
geometry. The RAM algorithm showed numerous serious deficiencies.
3.5.2 Sector Average Models
Three sector average models were examined: ISCLT, CDM, and VALLEY.
Results for all three models indicate that source integration methods are
subject to inaccuracies of 10 or 20 percent. Aside from these inaccuracies,
the ISCLT algorithm gave reasonable results based on the stability comparison,
source height, and far-field convergence tests. ISCLT failed to account
properly for source geometry, however, based on the subdivision test. CDM
gave reasonable results for the subdivision, stability, and source height
comparisons, but the far-field predictions at 8km were not reasonable. The
VALLEY algorithm gave physically unreasonable results based upon the stability
comparison, the subdivision test, and its lack of sensitivity to source height.
3-22
-------
Rgure 3—3
10
7_
10 4-
Model: SHORT TERM GROUP
Test: INTER-MODEL COMPARISON
Stability. 0
Source Size: 150m X 150m
Source Height- Orn
Wind Speed: 2 m/s
Wind Direction: 360
Total Source Strength: 1 kg/s
1 I
500 1000 1500 2000
Downwind Distance (m)
3-23
-------
10 ••a
Figure 3—4
c
o
10 5-
o E
O D
£010<
TI.U
CD c~
-^ C
CD
O
103
Model: SHORT TERM GROUP
Teat INTER-MODEL COMPARISON
Stability: B
Source Size: 150m X 150m
Source Height: Om
Wind Speed: 2 m/s
Wind Direction: 3oO
Total Source Strength: 1 kg/s
I I I I I I I I I I I I I ! I I I
RAM
0
500 1000 1500 2000
Downwind Distance (m)
3-24
-------
10 7q
o E
O D
-------
Rgure 3—6
10 f-i
oE
O O
-------
Rgure 3—7
Model: LONG TERM GROUP
Test- INTER-MODEL COMPARISON
Stability. D
Source Size: 150m X 150m
Source Height: Om
Wind Speed: 2 m/s
Wind Direction: 360
Total Source Strength: 1 kg/s
COM
500 1000 1500 2000
Downwind Distance (m)
3-27
-------
Figure 3—8
10
Modok LONG TERM GROUP
T««t: INTER-MODEL COMPARISON
Stability: B
Source She: 150m X 150m
Source Height: Om
Wind Speed! 2 cn/s
Wind Direction: 3GO
Total Source Strength: 1 fcg/e
COM (x)
VALLET (o)
500 1000 1500 2000
Downwind Distance (m)
3-28
-------
10
p *
-M *
E
oE
o o
CD
-------
10 *n
o
U-*
o
oE
00
CD
(D
O
10
Figure 3—10
Ikxtek LONG TERM GROUP
Twfc INTER-MODEL COMPARISON
Stability: 0
Source Size: 150m X 150m
Source Height- 10rn
Wind Speed: 2 m/s
Wind Direction: 360
Total Source Strength: 1 ka/e
COM
VALLET
ISCLT
500 1000 1500 2000
Downwind Distance (m)
3-30
-------
10 ^
c
o
5_
*
C £
o E
O D
£ 6io4-
CD
CD
O
10
Rgure 3—11
Model: I5CST
Test STA8IUTY COMPARISON
Stability: D. B, F
Source Size: 150m X 150m
Source Height Om
Wind Speed: 2 m/s
Wind Direction: 360
Total Source Strength: 1 kg/s
STABILITY F
STABILITY D
STABILITY B
500 1000 1500
Downwind Distance (m)
2000
3-31
-------
10 en
o E
O 0
en
O
CD
C _
T- O
CD
O
10
Figure 3-12
Model: ISCST
Test: SUBDIVISION COMPARISON
Stability: 0
Source Size: 150m X 150m. 4<75m X 75m)
Source Height Om
Wind Speed: 2 m/s
Wind Direction: 360
Total Source Strength: 1
4<75m X 75m) (»)
150m X 150m (o)
0
500 1000 1500 2000
Downwind Distance (m)
3-32
-------
10 *n
o
C £
o E
00
••!§
i VJ
C'
CD
O
10
Figure 3—13
Modefc SCST
Test SOURCE ORIENTATION
Stability: 0
Source Size: 150m X 150m
Source Height- Om
Wind Speedt 2 IH/S
Wind Direction: 360, 315
Total Source Strength: 1 kg/»
360 (o)
315 (*)
500 1000 1500 2000
Downwind Distance (m)
3-33
-------
10 fn
*
*
E
oE
o o
o> S1
r- O
CD
-f-»
C'
-------
10
e_
10s-
Figure 3—15
Mod«b BCST
Teet FAR-HELD CONVERGENCE
Stability: D
Source Size: 75m X 75m. 150m X 150m. 450m x 450m
Source Height: Om
Wind Speed: 2 m/s
Wind Direction: 360
Total Source Strength: 1 kg/e
75m X 75m (*)
150m X 150m (o)
450m X 450m(x)
I ' I • I • T • l
23456
1 2345678
Downwind Distance (km)
3-35
-------
10 *q
c
o E
o o
c _
T: 5
CD*
c
O
0 10 4-
y
E
10 3-
Figure 3—16
Model: FDM
Test: STABILITY COMPARISON
Stability: D. B. F
Source Size: 150m X 150m
Source Height Om
Wind Speed: 2 m/s
Wind Direction: 360
Total Source Strength: 1 kg/s
STABILITY F
STABILITY D
STABILITY B
500 1000 1500
Downwind Distance (m)
2000
3-36
-------
Figure 3—17
10 'n
oE
O o
|g
o3*F
0
O
10
Modal: FDM
T«t: SUBDIVISION COMPARISON
Stability: 0
Source Size: 150m X 150m. 4(75m X 75m)
Source Height Om
Wind Speed: 2 m/s
Wind Direction: 360
Total Source Strength: 1 kg/e
150m X 150m (o)
4(75nT\ 75m) (*)
0
500 1000 1500
Downwind Distance (m)
2000
3-37
-------
10 *-i
Rgure 3—18
Modal: FDM
Test SOURCE ORIENTATION
Stability: D
Source Size: 150m X 150m
Source Height: Om
Wind Speed: 2 m/s
Wind Direction: 360, 315
Total Source Strength: 1 kg/3
315 (o)
360 («)
500 1000 1500 2000
Downwind Distance (m)
3-38
-------
10'I
Rgure 3—19
" JMo ••
§E
o o
0)
CD
O
104-
Model:
Twt SOURCE HDGHT COMPARISON
Stability: D
Source Sin: 150m X 150m
Source Height: Om, 10m
Wind Speed: 2 m/8
Wind Direction: 360
Total Source Strength: 1 ka/e
Om (*)
10m (o)
500 1000 1500 2000
Downwind Distance (m)
3-39
-------
10 'q
Rgure 3-20
10 »•:
oE
O 0
CD
C
TI
(D
-i-»
C
Q>
O
cn
10 3-
102-
Model: FDM
Teefc FAR-FIELD CONVERGENCE
Stability: D
Source Size: 75m X 75m. 150m X 150m. 450m X 450m
Source Height Om
Wind Speed: 2 m/s
Wind Direction: 360
Total Source Strength: 1 kg/a
75m X 75m (o)
150m X 150m (x)
450m X 450m (*)
• ••••tt
1 234567
Downwind Distance (km)
8
3-40
-------
10 7q
10
6_
-------
Figure 3-22
10 7q
c
— X^""S1 n '—
-4-* T 1U _
c"
oE
o o
0)
c
en
o 10
s_
(D
o
104-
Model: PAL
Tttt: SUBDIVISION COMPARISON
Stability: 0
Sourc* Sfe«: 150m X 150m. 4<75rn X 75m)
Source Height: Om
Wind Speed: 2 m/9
Wind Direction: 360
Total Source Strength: 1 kg/e
150m X 150m (*)
4<75m X 75m) (o)
500 1000 1500
Downwind Distance (m)
2000
3-42
-------
107T3
2x-x § j
*-4^ KTi 1 0 "
o£
O D
-------
10 7
o E
OD
(D
O
e_
10
Figure 3-24
Modefc PAL
Test- SOURCE HQGHT COMPARISON
Stability: 0
Source Sire: 150m X 150m
Source Height: Om, 10m
Wind Speed: 2 m/s
Wind Direction: 360
Totd Source Strength: 1 kg/e
Om (*)
10m (o)
500 1000 1500 2000
Downwind Distance (m)
3-44
-------
C
o
C'
CD
O
10 eq
Rgure 3-25
•¥•
oE
O o
| 010
'-c.a
CD C
4_
10
Mod«l: PAL
T«rt: FAR-FIELD CONVERGENCE
Stability: D
Source Siz«: 75m X 75n>. 150m X
Source Height* Om
Wind Speed! 2 rn/8
Wind DirBCuoni 960
Total Source Strength: 1 kg/s
150m. 450m X 450m
75m X 75m (*)
150m X 150m (o)
450m X 450m (x)
I f • • I I I
1 234567
Downwind Distance (km)
8
3-45
-------
Rgure 3—26
10 ^
oE
o o
gpio
0)
O
10
Mod*fc RAM
T«st STA8UJTY COMPARISON
Stability: D, B, F
Source StiaK 150m X 150m
Sourco H
-------
10'-]
*
*
E
00
-S
C
CD
O
10 4-
Figure 3—27
Model: RAM
•tot- SUBOIMSION COMPARISON
Stability: 0
Soun* Ste«: 150m X 150m. 4(75rn X 75m)
Source Heiahb Om
Wind Speed: 2 m/s
Wind Direction: 360
Total Source Strength: 1 ko/i
150m X 150m (*)
4<75m X 75m)(o)
500
1000
1500
2000
Downwind Distance (m)
3-47
-------
CD
CD
C'
-------
10
f_
oE
o o
-------
Rgure 3—30
10 *q
c
1 n *—
1U _
oE
00
0)
-»-»
Q
CD
O
10 '•
\
Model: RAM
Test: FAR-FIELD CONVERGENCE
Stability: O
Source Size: 75m X 75m. 150m X 150m. 450m X 450m
Source Height- Om
Wind Speed: 2 m/s
Wind Direction: 360
Total Source Strength: 1 kg/8
75m X 75m (*)
150m X 150m (o)
450m X 450m (x)
• I i I i I ' I i I i I i I i i
012345678
Downwind Distance (km)
3-50
-------
10 7
10 f-
•*
*
oE
O D
E8
*— O
-------
10 7^
Figure 3—32
Modal: SHORTZ
Teefc SUBDIVISION COMPARISON
Stability: D
Source Size: 150m X 150m. 4(75m X 75m)
Source Height: Om
Wind Speed: 2 m/s
Wind Direction: 360
Total Source Strength: 1 kg/e
150m X 150m (*)
4(75m X 75m) (o)
10
0
500 1000 1500 2000
Downwind Distance (m)
3-52
-------
10 7q
Rgure 3—33
Ifcxtek SHORTZ
T«at SOURCE ORIENTATION
Stability: 0
Source Sa«: 150m X 150m
Source Height Orn
Wind 5p60d! 2 m/8
Wind Direction: 360, 315
Total Source Strength: 1 tog/»
5_
360 (*)
(o)
104-
500 1000 1500 2000
Downwind Distance (m)
3-53
-------
10 7i
*
*
oE
O D
01
O 10
•C.O
-------
10 *-
c
o
^
10 5^
o E
O D
(D
O
10 «•
Rgure 3—35
Mod*: SHOR1Z
T«8t FAR-HELD CONVERGENCE
Stability: D
Source Siz«: 75m X 75m. 150m X 150m. 450m X 450m
Source Height: Om
Wind Speed: 2 m/s
Wind Direction: 360
Total Source Strength: 1 kg/e
75m X 75m (*)
150m X 150m (o)
450m X 450m (x)
I ' I • I ' I
1234
i • r
6 7
8
Downwind Distance (km)
3-55
-------
10 *q
CD
CD
-4—'
C1
CD
O
0 10 4
o
10
Rgure 3—36
Model: SCLT
Test STABJUTY COMPARISON
Stability: 0. B, F
Source Size: 150m X 150m
Source Height Om
Wind Speed: 2 rn/a
Wind Direction: 360
Total Source Strength: 1 kg/s
STABILITY F
STABILITY D
STABILITY B
0
500 1000 1500 2000
Downwind Distance (m)
3-56
-------
10
f_
(D
O
oE
o P
Is
-.2
0>E
10 4-
Rgure 3-37
Model: SCLT
Twt- SUBDIVISION COMPARSON
StobHRy: 0
Souro Sfe«: 150m X 150m. 4(75m X 75m)
Source Height Om
Wind Speed: 2 m/»
Wind Wriction: 360
Total Source Strength: 1 kg/e
4(75m X 75m)(o)
1SOm X 150m (*)
500 1000 1500 2000
Downwind Distance (m)
3-57
-------
10 f-i
o
4-»
p *
oE
O 0
£p
(D
CJ
10
0
Figure 3—38
Mod»k
T«»t SOURCE HBGHT COMPARISON
StabJBty: D
Source Size: 150m X 150m
Source Height Om, 10m
Wind Speed! 2 m/s
Wind Direction: 360
Total Source Strength: 1 kg/»
Om (*)
10m (o)
III
I I I I
I I I ^
500 1000 1500 2000
Downwind Distance (m)
3-58
-------
10 §q
Rgure 3—39
C £
O E
00
V ?
r- O
0
C'
(D
O
4_
10
Model: BCLT
Tmfc FAR-FJELO CONVERGENCE
StobUtty: 0
Source Size: 75m X 75m. 150m X
Source Height Om
Wind Speed: 2 tn/9
Wind Direction: 360
Total Source Strength: 1 kg/e
150m, 450m X 450m
75n» X 75m (x)
150m X 150m (o)
450m X 450m (*)
i
1 23456
Downwind Distance (km)
8
3-59
-------
10 "q
Figure 3—40
c
o
oE
o o
£pio-
(D
CD
O
10
Model: COM
Test STABIUTY COMPARISON
Stability: D, B, F
Source Size: 150m X 150m
Source Height: Om
Wind Speed: 2 m/s
Wind Direction: 360
Total Source Strength: 1 kq/s
STABIUTY F
STABIUTY D
STABIUTY B
500 1000 1500 2000
Downwind Distance (m)
3-60
-------
Figure 3—41
10
C
o
+^
oE
o o
CD
O
10
Model: COM
Test: SUBDIVISION COMPARISON
Stability: D
Source Size: 150m X 150m, <75m X 75m)
Source Heiphfc Om
wind Speed: 2 m/s
Wind Direction: 360
Total Source Strength: 1 kg/s
4<75m X 75m) (o)
150m X 150m (»)
0
500
1000
1500
2000
Downwind Distance (m)
3-61
-------
10 6-i
o *
*
oE
o o
TI.2
0) C
-»-» C
Cx^-x
-------
10'-q
Figure 3—43
Model: COM
Test: FAR-FIELD CONVERGENCE
Stability: D
Source Size: 75m X 75m, 150m X 150m, 450m X 450m
Source Height Om
Wind Speed: 2 m/s
Wind Direction: 360
Total Source Strength: 1 kg/3
c
Oro
•-P V 10
D *
•|E
d> w
£E
0 0
u^
^§10
• ^™" * .^_
"SE
5_
-
-
-
-
_
4_
—
-
c
CD
o
10*
75m X 75m (*)
150m X 150m (o)
450m X 450m (x)
I ' I
5 6
1 234567
Downwind Distance (km)
8
3-63
-------
10*i
Rgure 3—44
10'
Modek VALLEY
Test- STABILITY COMPARISON
Stability: D. B. F
Source Size: 150m X 150m
Source Height: Orn
Wind Speed: 2 m/s
Wind Direction: 360
Totd Source Strength: 1 kg/s
STABIUTY F
STABILnY D
STABILITY B
500 1000 1500 2000
Downwind Distance (m)
3-64
-------
10 *H
*
*
*^
M«
§E
O 0
m C7>
Q)
-»-•
C
CD
O
10 4-
Figure 3—45
Model: VALLEY
fob SUBDIVISION COMPARISON
Stability: 0
Sourt* Size: 150m X 150m. 4<75m X 75m)
Source Heiaht Om
Wind Speed: 2 m/s
Wind Direction: 360
Total Source Strength: 1 kg/e
4<75m X 75m) (o)
150m X 150m (»)
500 1000 1500 2000
Downwind Distance (m)
3-65
-------
105n
Figure 3—46
Model: VALLEY
Teat SOURCE HEIGHT COMPARISON
Stability: 0
Source Size: 150m X 150m
Source Height: Om, 10m
Wind Speed: 2 m/s
Wind Direction: 360
Total Source Strength: 1 kg/*
CD
Om (*)
10m (o)
10 4-
0
500 1000 1500 2000
Downwind Distance (m)
3-66
-------
10 ••=
^C^10 4>
oE
o o
0
CD
c 010
"— o
&
CD
O
10
Rgure 3-47
Mod«l: VALLEY
Test: FAR-FIELD CONVERGENCE
Stability. D
Source Size: 75m X 75m. 150m X
Source Heiahb Om
Wind Speed: 2 m/3
Wind Direction: 360
Total Source Strength: 1 kg/e
150m. 450m X 4SOm
75m X 75m (»)
150m X 150m (o)
450m X 450m(x)
If • I I I • I
01 234567
Downwind Distance (km)
8
3-67
-------
4.0 COMPARISON OF MODEL PREDICTIONS WITH EXPERIMENTAL DATA
Predictions for the short-term area source models have been compared to
measured concentrations for an experimental data set involving releases of an
inert gaseous tracer within an isolated stand of trees/ surrounded by open
grassland. These experiments produced a distributed emission source which can
be simulated as an area source. Comparisons of observed and predicted
concentrations for this small set of experiments provide an informative test
of model performance.
4.1 Database Description
A review of available information from field measurement programs
characterizing observed concentrations in the vicinity of area source releases
was recently carried out as part of a larger effort to identify and acquire
databases for the evaluation of air toxics models (Zapert, et al., 1989). The
criteria which were used to select databases for model evaluation included the
following:
• Reliable emissions data
• Site-specific meteorological measurements
• Adequate spatial resolution of ambient concentrations
No suitable databases representing actual sources of toxic pollutant
emissions were identified. In general, measurement programs for actual
sources are characterized by uncertain emissions estimates and poor spatial
resolution (concentration measurements at only two or three locations).
The five databases chosen for air toxics model evaluation all represent
measurement programs with controlled, artificial releases designed to simulate
actual pollutant releases. Four of these databases involve dense gas
(heavier-than-air) releases and are not suitable for the present study.
4-1
-------
The database selected for testing area source models represents a series
of tracer experiments conducted in south-central Washington in 1982-83
(Allwine, et al., 1985). The inert gaseous tracer sulfur hexafluoride (SFg)
was released from an array of points within an isolated stand of oak trees.
The stand covered an area of 1.5 hectares (15,000 m2) to an average height of
8 m. The experiments were designed primarily to estimate isoprene emissions
from the forest canopy. By measuring SFg and isoprene concentrations
simultaneously, the isoprene emission rate could be inferred, assuming that
SFg and isoprene were both emitted uniformly from the canopy. Integrated
one-hour concentrations were measured at sampling points approximately 100 m
downwind of the release area. Samplers were deployed at different locations
depending upon the wind direction. The tracer release was initiated 10
minutes before ambient sampling began, and continued until sampling was
completed. Wind speed, wind direction and temperature measurements were also
collected during the experiments. A schematic diagram of the experimental
configuration is provided in Figure 4-1.
Thirteen tracer experiments from this program were chosen for model
evaluation. Three tests were excluded because the wind direction shifted
during these experiments and the peak measured concentrations occurred at the
end of a line of samplers. Emission characteristics and meteorological
conditions for the thirteen selected tests are summarized in Table 4-1. All
of the experiments were conducted during the daytime, when isoprene emissions
were expected to be highest. Consequently, the tests only represent unstable
and neutral conditions. This is a serious limitation for the present
application, since worst-case short-term impacts from near-ground emission
sources are expected during stable conditions.
4-2
-------
Sampling points
N
Sampling points
Met.
station
Release*
points
Met.
station
lOOm
FIGURE 4-1
ISOPRENE FLUX EXPERIMENT SAMPLING GRID, RELEASE POINTS AND WOODLOT
4-3
-------
TABLE 4-1
METEOROLOGICAL AND EMISSION CHARACTERISTICS FOR
TRACER RELEASE EXPERIMENTS IN A FOREST CANOPY
Test Number
Tracer
Emission Rate
(g/s)
Wind Speed
(m/s)
Atmospheric
Stability
Class
1
2
3
4
5
6
7
8
9
10
11
12
13
.102
.074
.104
.084
.081
.089
.094
.097
.095
.093
.092
.113
.112
1.0
3.6
5.8
6.2
2.2
8.5
1.5
1.1
1.3
1.6
1.8
2.2
3.2
D
D
D
D
B
D
D
C
C
C
C
D
D
4-4
-------
While these forest-canopy experiments represent a useful database for
testing area source models, the experimental configuration introduces a number
of complicating factors. The simple modeling scenario assumes that tracer
emissions will become thoroughly mixed within the canopy and evolve out of the
top of the source region. The following items contribute to the uncertainties
involved in modeling these experiments:
• Dispersion within the canopy is probably incomplete, and emissions
will not occur uniformly across the source region.
• Thermal stratification may "trap" some of the tracer within the
canopy. Emissions to ambient air may persist for several hours
after the tracer source is shut off.
• In the trunk space below the canopy, winds may transport some of
the tracer material out of the source region.
• Enhanced turbulence is likely in the lee of the source region.
The forest grove may produce the equivalent of a "building wake"
as ambient air moves over and around this obstacle.
The effects of these complicating factors are difficult to quantify. Some
will tend to increase observed tracer concentrations, others to lower
concentrations. The modeling approach does not account for any of these
factors. The source is modeled as four adjacent area source squares, each
61 x 61 m, comprising a total area of 1.5 hectares. The source emission
height was chosen as 8 m, the average canopy height.
4.2 Results
Predicted and observed tracer concentrations were compared for each tracer
experiment. Statistics were computed both for event-by-event comparisons
(paired in time) and for measures based on unpaired comparisons. Observed and
predicted concentrations were divided by the tracer emission rate before
computing statistics.
4-5
-------
4.2.1 Unpaired Comparisons
The maximum observed and predicted normalized concentration values are
illustrated in Figure 4-2. Each vertical bar spans the range of the thirteen
maximum values. All five of the short-term models overpredict the observed
range of maximum values. The highest observed maximum value (over all tests)
is 1.5 x 10™* s/m^, while maximum predicted values range from 2.4 x 10~4 s/m3
by FDM up to 7.3 x 10~* s/m3 by RAM.
The range of maximum values predicted by FDM and SHORTZ overlaps
considerably with the range of observed values. The lowest 8 or 9 maximum
values predicted by these models fall in the same range as the top 10 observed
maxima. By contrast, most of the maximum values predicted by ISC, PAL and RAM
exceed the highest observed value. Overall, the maximum values predicted by
FDM are higher than the maximum observed values by roughly a factor of 1.6,
SHORTZ over-predicts by a factor of 2, and the remaining models by a factor of
4 to 5.
4.2.2 Paired Comparisons
The maximum observed and predicted normalized concentration values for
each experiment are summarized in Table 4-2. The bias toward over-prediction
illustrated in Figure 4-2 is again evident in Table 4-2. Lack of correlation
is also apparent, as events with peak predicted values do not coincide with
high observed values. For example, Test 1 has the highest predicted value for
all five models, but has the second-lowest observed maximum value.
Correlation coefficients for observed and predicted maximum values are
negative for all five models. (By contrast, the correlation between maximum
values predicted by any two models exceeds 0.95. Inter-model correlation
results from the use of the same meteorological inputs for all of the
models.) The relative differences between maximum observed and predicted
4-6
-------
z
uj Z
Z
U -D
£•-
c^
1 1
1 1
1 1
1 1
I— 1
Q
UJ
1
1
1
1
X
z
u
z
4-1
o
l
l
1
l
1
1
t-
{/>
UJ
I
1
1
1
1 N
|c£
O
co
H
I—C
U
Q
UJ
>
UJ
CD
o
I '
n
' O
10
to
t-
"Z.
UJ _
> &.
Q£ UJ
LU 0_
CO X
CQ UJ
O UJ
10
D <
5"
< UJ
I- uj
U U
^ <
o
-z.
UJ UJ
u u
< o
c£ U
«N
I
UJ
C£
D
NOIlVaiN3DNOD
4-7
-------
TABLE 4-2
MAXIMUM OBSERVED AND PREDICTED NORMALIZED
TRACER CONCENTRATION FOR EXPERIMENT
Test
Number
Observed
Concentration
(l
-------
values for each event are summarized in Table 4-3. For FCM, predicted maximum
values agreed within a factor of 2 with maximum observed values for eight of
the 13 experiments. All five models overpredicted by more than a factor of
1.25 for the majority of events.
While maximum predicted values are closely tied to meteorological
conditions, observed maximum values do not show a similar pattern. The
highest observed values occurred for tests 10 and 11 (C stability, low wind
speed), but tests 8 and 9 with similar meteorology have low observed maximum
values. Three experiments with D stability and low wind speeds (tests 1, 7
and 12) have high predicted concentrations but relatively low observed
values. The Turner stability method (based on wind speed and cloud cover)
does not provide an effective method of classifying the observed dispersion
behavior for these experiments.
Summary - Comparisons of observed and predicted maximum concentrations for
the 13 one-hour experiments demonstrate systematic overprediction by all five
short-term models. FDM showed the least bias, over-predicting by roughly a
factor of 1.6. Maximum values predicted by FDM were within a factor of 2 of
the observed maximum for 8 of 13 events. The range of maximum predicted by
SHORTZ was about a factor of 2 higher than observed; predicted and observed
maxima agreed within a factor of 2 for 6 events for SHORTZ. ISC, PAL and RAM
all overpredicted the range of maximum values by more than a factor of 4. For
these three models, predicted and observed values agreed within a factor of 2
for 4 or 5 events. RAM overpredicted by more than a factor of 1.25 for all 13
events. Paired comparisons showed negative correlation between observed and
predicted maximum values for all models.
4-9
-------
TABLE 4-3
SUMMARY OF RELATIVE DIFFERENCES BETWEEN
OBSERVED AND PREDICTED MAXIMUM VALUES
Ratio (R) of Predicted
to Observed Maximum
Concentration
ISC
FDM
PAL
RAM
SHORTZ
R > 2
2 > R > 1.25
1.25 > R > 0.8
0.8 > R > 0.5
0.5 > R
9
2
2
-
*»
5
2
2
4
_
8 9
2 4
3
- -
_ M
6
2
3
1
1
4-10
-------
5.0 CONCLUSIONS AND RECOMMENDATIONS
5.1 Conclusions
A number of existing dispersion models are available for estimating
ambient concentrations due to area source emissions from landfills, waste
disposal areas and contaminated sites. Three basic methods of calculating
concentrations due to area sources are employed by existing models: virtual
point source, line source, and upwind (numerical) integration. A group of
five short-term models (FDM, ISCST, PAL, RAM, SHORTZ) and three long-term
(sector average) models (ISCLT, CDM, VALLEY) incorporating these methods were
selected for evaluation in the present study.
None of these dispersion models incorporates a source algorithm capable of
estimating evaporative emission rates from a landfill or waste disposal area.
For many applications, the variation of emissions as a function of
environmental factors (ambient or ground surface temperature, wind speed,
moisture) is a major source of uncertainty for determining ambient
concentrations.
Example Applications. Each of the eight models was applied to predict
ambient concentrations for a base case hypothetical scenario and for several
variations designed to examine whether model predictions are consistent with
mathematical and physical principles.
The base case represented a single 150 x 150 m ground level area source,
with receptors located downwind at distances ranging from 100 m to 1,575 m
from the source center. Among the short-term models, FDM generally predicted
the lowest concentrations. The other four models' predictions varied in rank,
depending upon distance and stability class. At large distances, RAM
consistently predicted the highest concentrations.
All of the models except RAM showed reasonable far-field behavior. PAL
predictions were physically reasonable and consistent with source-receptor
5-1
-------
geometry for ail of the configurations analyzed. FDM produced small
inconsistencies (generally 10 percent or less) for two tests sensitive to
source geometry. Predictions by SHORTZ and ISCST in the near-field did not
accurately reflect source-receptor geometry for one or more comparisons. RAM
did not give physically reasonable results for either near-field or far-field
predictions.
None of the three long-term average models consistently gave reasonable
results. Comparisons of predictions for different scenarios indicated
inaccuracies of 10 percent or more in calculated concentrations. Serious
deficiencies in VALLEY were indicated by several tests. ISCLT and COM gave
reasonable results for most comparisons. ISCLT, however, predicted large
differences for the subdivision comparison, while CDM gave erratic results for
the far-field convergence test.
Comparison with Observed Concentrations. Observed and predicted
concentrations were compared for a data base of thirteen one-hour experiments
involving tracer releases within an isolated grove of trees. The release was
modeled as a 1.5 ha area source. All five of the short-term models predicted
higher maximum tracer concentrations than were measured roughly 100 m downwind
of the source region. FDM and SHORTZ overpredicted by a factor of 2 or less,
while ISCST, PAL and RAM overpredicted by roughly a factor of 4. All of the
models showed poor correlation between observed and predicted maximum values,
paired by event.
Recommendations. Overall, the short-term models which employ algorithms
based on the line-source method (FDM, PAL) appear to provide an adequate
treatment of near-source geometry and reasonable far-field behavior. Subject
to further performance testing, either FDM or PAL is recommended for use with
near-ground area sources. (The area source algorithm in FDM is not a separate
modular component of the model. This algorithm is coupled to line source and
5-2
-------
dispersion algorithms taken from CALINE. which uses a modified form of the
Pasquill-Gifford rural dispersion coefficients.) It may also prove feasible
to modify either SHORTZ or ISCSI to provide more reasonable near-field
predictions. The RAM model is not recommended for application to an isolated
area source.
In light of the widespread use of ISCST for regulatory applications, the
feasibility of modifying or replacing the ISC area source algorithm deserves
serious consideration. The inaccuracy of near-field area source predictions
by ISCST is sufficiently large that the users guide recommends subdividing the
source if receptors are placed near that source. In the present study,
results have clearly demonstrated that ISCST does not account properly for
source-receptor geometry at near-field receptors.
The long-term average results indicated that ISCLT and CDM generally
provide adequate treatment of area source dispersion, but each model has
specific shortcomings. Subject to further testing, either ISCLT or CDM is
recommended. The FDM, LONGZ and AQDM models contain area source algorithms
similar to ISCLT and are expected to produce similar predictions. Potential
modifications to correct specific model shortcomings should be investigated.
The VALLEY model is not recommended for application to area sources.
5-3
-------
REFERENCES
Allwine, G., Lamb, B. and Westberg, H., Application of Atmospheric Tracer
Techniques for Determining Biogenic Hydrocarbon Fluxes from an Oak Forest,
The Forest - Atmosphere Interaction, Ed. Hutchison, B.A. and Hicks, B.B.,
D. Reidel Publishing Company, Boston, 1985.
Benson, P.E., CALINE3 - A Versatile Dispersion Model for Predicting Air
Pollutant Levels Near Highways and Arterial Streets, PB80-220841, November
1979.
Bjorklund, J.R. and Bowers, J.F., User's Instructions for the SHORTZ and LONGZ
Computer Programs, Volumes I and II, EPA-903/9-82-004a and 004b, March
1982.
Hurt, E.W., Valley Model User's Guide, EPA-450/2-77-018, September 1977.
Gschwandtner, G., Eldridge, R. and Zerbonia, R., "Sensitivity Analysis of
Dispersion Models for Point and Area Sources," JAPCA, Vol. 32, #10,
October 1982.
Guideline on Air Quality Models (Revised), EPA-450/2-78-027R, July 1986.
Hodanbosi, R.F. and Peters, L.K., "Evaluation of RAM Model for Cleveland,
Ohio," JAPCA, Vol. 31, #3, March 1981.
Hwang, S.T., "Methods for Estimating On-Site Ambient Air Concentrations at
Disposal Sites," Nuclear and Chemical Waste Management, Vol. 7, 1987.
Irwin, J.S. and Brown, T.M., "A Sensitivity Analysis of the Treatment of Area
Sources by the Climatological Dispersion Model," JAPCA, Vol. 35, #4, April
1985.
Irwin, J.S., Chico, T. and Catalano, J., CDM 2.0 — Climatological Dispersion
Model, EPA/600/3-85/029, PB86-136546, November 1985.
Rao, K.S., User's Guide for PEM-2: Pollution Episodic Model (Version 2),
EPA/600/8-86/040, PB 87-132 098, December 1986.
Schere, K.L. and Demerjian, K.L., User's Guide for the Photochemical Box Model
(PBM), PB85-137164, November 1984.
Schulze, R., Practical Guide to Atmospheric Dispersion Modeling, Trinity
Consultants, Inc., Dallas, 1989.
Schulze, R.H., Ed., "All Area Source Models Are Not Created Equal,"
Atmospheric Diffusion Notes, Trinity Consultants, Inc., Issue #7, July
1982.
Scire, J.S., Lunnann, F.W., Bass, A. and Hanna, S.R., User's Guide to the
Mesopuff II Model and Related Processor Programs, EPA-600/8-84-013, April
1984.
R-l
-------
Simnon, P.B., Patterson, R.M., Ludwig, F.L. and Jones, L.B., User's Manual for
the APRAC-3/Mobilel Emissions and Diffusion Modeling Package,
EPA/909-9-81-002, July 1981.
Turner, D.B., Workbook of Atmospheric Dispersion Estimates, EPA Office of Air
Programs, Pub. # AP-26, 1970.
Turner, D.B. and Novak, J.H., User's Guide for RAM, Vol. 1, Algorithm
Description and Use, EPA-600/8-78-016A, November 1978.
Wackter, D.J. and Foster, J.A., Industrial Source Complex (ISC) Dispersion
Model User's Guide - Second Edition (Revised), Vol. 1, EPA-450/4-88-002a,
December 1987.
R-2
-------
TECHNICAL REPORT DATA
(Please read fnstmciians on tHe reverse before completing)
REPORT NO.
EPA-450/4-89-020
2.
3. RECIPIENT'S ACCESSION NO.
4. TITLE AND SUBTITLE
Review and Evaluation of Area Source Dispersion
Algorithms for Emission Sources at Superfund Sites
5. REPORT DATE
November 1989
6. PERFORMING ORGANIZATION CODE
7. AUTHOR(S)
I. PERFORMING ORGANIZATION REPORT NO.
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.
68-02-4399
12. SPONSORING AGENCY NAME AND ADDRESS
13. T
U.S. Environmental Protection Agency
Office of Air Quality Planning and Standards
Research Triangle Park, N. C. 27711
•*ee of REPORT A
Final Report
NO PERIOD COVERED
14. SPONSORING AGENCY CODE
15. SUPPLEMENTARY NOTES
EPA Project Officer:
Jawad S. Touma
16. ABSTRACT
This report examines air quality dispersion modeling algorithms and related
technical issues associated with estimating ambient concentrations from area
sources at Superfund sites. The report describes the area source emission
characteristics associated with Superfund sites and provides a review of
existing, available techniques for modeling area sources. It also describes the
results of applying five short-term and three long-term area source models to
a number of example applications and one field data base in order to compare the
magnitude of concentration predictions and test whether concentration estimates
are consistent with mathematical and physical principles. The report provides
conclusions and recommendations from this study.
17.
KEY WORDS AND DOCUMENT ANALYSIS
DESCRIPTORS
b.lOENTIFIERS/OPEN ENDED TERMS C. COSATI Field/Group
Air Pollution
Hazardous Waste Assessment
Toxic Air Pollutants
Air Quality Dispersion Models
Dispersion Modeling
Meteorology
Air Pollution Control
13 B
18. DISTRIBUTION STATEMENT
Release Unlimited
19. SECURITY CLASS (This Report!
Unclassified
21 NO OP PAGES
120
20 SECURITY CLASS (Tins page}
Unclassified
22. PRICE
EPA Form 2220-1 (Rev. 4-77)
PREVIOUS EDITION IS OBSOLETE
-------
July 1991
ERRATA FOR
REVIEW AND EVALUATION OF AREA SOURCE DISPERSION ALGORITHMS
FOR EMISSION SOURCES AT SUPERFUND SITES
Office of Air Quality Planning and Standards
Technical Support Division
Research Triangle Park, NC 27711
-------
LIST OF ERRATA
REVIEW AND EVALUATION OF AREA SOURCE DISPERSION ALGORITHMS FOR
EMISSION SOURCES AT SUPERFUND SITES
EPA-450/4-89-020
NOVEMBER, 1989
NTIS ACCESSION NO. PB90-142753
The following pages are replaced:
1. Page ix
2. Page 2-22
3. Page 4-1
4. Page 4-4
5. Pages 4-6 through 4-10
6. Page 5-2
7. Page R-l
-------
LIST OF TABLES
TABLE PAGE
1-1 AREA SOURCE CHARACTERISTICS 1-4
2-1 CHARACTERIZATION OF AREA SOURCE ALGORITHMS IN EXISTING
MODELS 2-8
3-1 SUMMARY OF SENSITIVITY TEST RESULTS FOR SHORT-TERM
MODELS 3-21
4-1 METEOROLOGICAL AND EMISSION CHARACTERISTICS FOR TRACER
RELEASE EXPERIMENTS IN A FOREST CANOPY 4-4
4-2 MAXIMUM OBSERVED AND PREDICTED NORMALIZED TRACER
CONCENTRATION FOR EXPERIMENT 4-8
4-3 SUMMARY OF RELATIVE DIFFERENCES BETWEEN OBSERVED AND
PREDICTED MAXIMUM VALUES 4-9
—ix-
-------
sources. Two published articles were identified which discussed alternative
approaches to modeling area sources. Neither approach represents a computer
algorithm suitable for evaluation in the present study.
The models suggested by Hwang (1987) and Chitgopekar et al. (1990) attempt
to resolve some of the known shortcomings of the currently utilized area
source dispersion algorithms. The discussion by Hwang is theoretical but
examines both a Gaussian approach and one based on transport equations in the
atmospheric boundary layer. Chitgopekar et al. (1990) attempts to resolve
problems of near-field prediction through the use of a "top-hat" formulation.
Chitgopekar et al. (1990) presents an area source model developed in
response to problems with virtual point models in the near-field for area
sources. These authors state that "the most rigorous" treatment of Gaussian
dispersion from area sources would be to model them as a dense matrix of
multiple point sources. This idea can be conceptualized as increasing matrix
density until, in the limit, inter-point spacing goes to zero and every point
in the area is emitting. This approach is computationally intensive and is
rarely used in the standard models. (The finite line segment approach in FDM
or PAL is mathematically equivalent but far more efficient, if Gaussian
dispersion is assumed.) Virtual point source methods do not require extensive
computations and can be simplified to allow manual calculation. However, the
virtual point source method should not be used if the source width is greater
than 40% of the distance between the source centerpoint and the receptor
(Hwang, 1986 as cited in Chitgopekar et al., 1990). This is a serious
limitation due to the fact that the region of interest in many area source
pollutant dispersion and exposure situations is in the near-field.
The treatment by Chitgopekar et al. is to divide the dispersion of the
plume into high frequency and low frequency components. A low pass filter is
used so plume meander is included but small turbulent scales are not. The
2-22
-------
4.0 COMPARISON OF MODEL PREDICTIONS WITH EXPERIMENTAL DATA
Predictions for the short-term area source models have been compared to
measured concentrations for an experimental data set involving releases of an
inert gaseous tracer within an isolated stand of trees, surrounded by open
grassland. These experiments produced a distributed emission source which can
be simulated as an area source. Comparisons of observed and predicted
concentrations for this small set of experiments provide an informative test
of model performance.
4.1 Database Description
A review of available information from field measurement programs
characterizing observed concentrations in the vicinity of area source releases
was recently carried out as part of a larger effort to identify and acquire
databases for the evaluation of air toxics models. The criteria which were
used to select databases for model evaluation included the following:
• Reliable emissions data
• Site-specific meteorological measurements
• Adequate spatial resolution of ambient concentrations
No suitable databases representing actual sources of toxic pollutant
emissions were identified. In general, measurement programs for actual
sources are characterized by uncertain emissions estimates and poor spatial
resolution (concentration measurements at only two or three locations).
The five databases chosen for air toxics model evaluation all represent
measurement programs with controlled, artificial releases designed to simulate
actual pollutant releases. Four of these databases involve dense gas
(heavier-than-air) releases and are not suitable for the present study.
4-1
-------
TABLE 4-1
METEOROLOGICAL AND EMISSION CHARACTERISTICS FOR
TRACER RELEASE EXPERIMENTS IN A FOREST CANOPY
Test Number
1
2
3
4
5
6
7
8
9
10
11
12
13
Tracer
Emission Rate
-------
4.2.1 Unpaired Comparisons
The maximum observed and predicted normalized concentration values are
illustrated in Figure 4-2. Each vertical bar spans the range of the thirteen
maximum values. Four of the five models overpredict the maximum observed
value and all five models overpredict the minimum observed value. The highest
observed maximum value (over all tests) is 1.5 x 10~^ s/rn-^, while maximum
predicted values range from 1.2 x 10~4 s/m^ by FDM up to 3.2 x 10~~4 s/m^ by
RAM.
The range of maximum values predicted by FDM is within the range of
observed values. The lowest 12 maximum values predicted by SHORTZ fall in the
same range as the top 10 observed maxima. By contrast, most of the maximum
values predicted by ISC, PAL and RAM exceed the highest observed value. The
median value predicted by FDM is higher than the maximum observed median by
roughly a factor of 1.3, SHORTZ overpredicts by a factor of 1.7, and the
remaining models by factors of between 2.7 and 3.2.
4.2.2 Paired Comparisons
The maximum observed and predicted normalized concentration values for
each experiment are summarized in Table 4-2. The bias toward overprediction
illustrated in Figure 4-2 is again evident in Table 4-2. Lack of correlation
is also apparent, as events with peak predicted values do not coincide with
high observed values. For example, Test 1 has the highest predicted value for
all five models, but has the second-lowest observed maximum value.
Correlation coefficients for observed and predicted maximum values are
negative for all five models. The relative differences between maximum
observed and predicted values for each event are summarized in Table 4-3. For
FDM, predicted maximum values agreed within a factor of 2 with maximum
4-6
-------
fc
UJ
I
(J
X
i
i
i
i
1
T.
Z
1
1
1
1
I*
1
Z
<
Q
2
i
i
i
j
/
i
jj=
-C
4-1
O
1
1
1
1
1
1
to
UJ
$
o
_l
1
1
1
1 N
. 1-
1 o
tS)
o
o
o
to
Qi
] li
co
CD
o
o oo
z i-
<: z
UJ
Q S
UJ t—<
I— cc.
<_> uj
l—i O.
O X
UJ UJ
OL
Q- UJ
H
I—E
I—[
]—I
CO
o
X 0£
s a:
UJ
u_ o
o <
OL
UJ I—
CJ
z a:
-------
TABLE 4-2
MAXIMUM OBSERVED AND PREDICTED NORMALIZED
TRACER CONCENTRATION FOR EXPERIMENT
Test
Number
1
2
3
4
5
6
7
8
9
10
11
12
13
# Closest
# Closest
Observed
Concentration
(10~6 s/m3)
24.5
87.8
33.7
74.6
88.5
58.5
67.1
21.7
37.0
99.8
151.9
45.9
42.8
to Observed
to Observed
ISC
292.
174.
101.
101.
109.
71.
172.
227.
190.
162.
143.
180.
181.
0
1
FDM
123.
88.
55.
53.
50.
36.
76.
94.
82.
70.
63.
81.
81.
8
8
PAL
295.
149.
89.
88.
102.
57.
169.
222.
186.
159.
140.
191.
157.
3
3
RAM
322.
232.
139.
135.
120.
91.
206.
304.
253.
182.
160.
187.
209.
1
1
SHORTZ
163.
101.
64.
60.
38.
42.
98.
125.
103.
97.
86.
104.
102.
1
0
and Greater
4-8
-------
TABLE 4-3
SUMMARY OF RELATIVE DIFFERENCES BETWEEN
OBSERVED AND PREDICTED MAXIMUM VALUES
Ratio (R) of Predicted
to Observed Maximum
Concentration
ISC
FDM
PAL
RAM
SHORTZ
R > 2
2 > R > 1.25
1.25 > R > 0.8
0.8 > R > 0.5
0.5 > R
7 3
3 4
3 1
4
1
7
2
4
8
4
1
5
2
2
3
1
4-9
-------
observed values for 9 of the 13 experiments. Three models overpredicted by
more than a factor of 2.0 for the majority of events.
While maximum predicted values are closely tied to meteorological
conditions, observed maximum values do not show a similar pattern. The
highest observed values occurred for tests 10 and 11 (B stability, low wind
speed), but tests 8 and 9 with similar meteorology have low observed maximum
values. The Turner stability method (based on wind speed and cloud cover)
does not provide an effective method of classifying the observed dispersion
behavior for these experiments.
Summary - Comparisons of observed and predicted maximum concentrations for
the 13 one-hour experiments demonstrate a general tendency towards systematic
overprediction by all five short-term models. FDM showed the least bias,
overpredicting the median maximum value by roughly a factor of 1.3. Maximum
values predicted by FDM were within a factor of 2 of the observed maximum for
9 of 13 events. The median maximum predicted by SHORTZ was a factor of 1.7
higher than observed; predicted and observed maxima agreed within a factor of
2 for 7 events for SHORTZ. ISC, PAL and RAM all overpredicted the median
maximum value by approximately a factor of 3. For these three models,
predicted and observed values agreed within a factor of 2 for 5 or 6 events.
RAM overpredicted for all 13 events.
4-10
-------
geometry for all of the configurations analyzed. FDM produced small
inconsistencies (generally 10 percent or less) for two tests sensitive to
source geometry. Predictions by SHORTZ and ISCST in the near-field did not
accurately reflect source-receptor geometry for one or more comparisons. RAM
did not give physically reasonable results for either near-field or far-field
predictions.
None of the three long-term average models consistently gave reasonable
results. Comparisons of predictions for different scenarios indicated
inaccuracies of 10 percent or more in calculated concentrations. Serious
deficiencies in VALLEY were indicated by several tests. ISCLT and CDM gave
reasonable results for most comparisons. ISCLT, however, predicted large
differences for the subdivision comparison, while CDM gave erratic results for
the far-field convergence test.
Comparison with Observed Concentrations. Observed and predicted
concentrations were compared for a data base of thirteen one-hour experiments
involving tracer releases within an isolated grove of trees. The release was
modeled as a 1.5 ha area source. All five of the short-term models tend to
predict higher maximum tracer concentrations than were measured roughly 100 m
downwind of the source region (Table 4-2). FDM and SHORTZ overpredicted by a
factor of 2 less than half of the test cases, while ISCST, PAL and RAM
overpredicted more than half of the cases by a factor of 2 or more. All of
the models showed poor correlation between observed and predicted maximum
values, paired by event.
Recommendations. Overall, the short-term models which employ algorithms
based on the line-source method (FDM, PAL) appear to provide an adequate
treatment of near-source geometry and reasonable far-field behavior. Subject
to further performance testing, either FDM or PAL is recommended for use with
near-ground area sources. (The area source algorithm in FDM is not a separate
modular component of the model. This algorithm is coupled to line source and
j-2
-------
REFERENCES
Allwine, G. , Lamb, B. and Westberg, H. , Application of Atmospheric Tracer
Techniques for Determining Biogenic Hydrocarbon Fluxes from an Oak Forest,
The Forest - Atmosphere Interaction, Ed. Hutchison, B.A. and Hicks, B.B.,
D. Reidel Publishing Company, Boston, 1985.
Benson, P.E., CALINE3 - A Versatile Dispersion Model for Predicting Air
Pollutant Levels Near Highways and Arterial Streets, PB80-220841, November
1979.
Bjorklund, J.R. and Bowers, J.F., User's Instructions for the SHORTZ and LONGZ
Computer Programs, Volumes I and II, EPA-903/9-82-004a and 004b, March
1982.
Burt, E.W., Valley Model User's Guide, EPA-450/2-77-018, September 1977.
Chitgopekar, N.P., D.D. Reible and L.J. Thibodeaux,''"Modeling Short Range Air
Dispersion from Area Sources of Non-Buoyant Toxics',1 AWMA, August 1990,
Vol. 40, #8.
Gschwandtner, G., Eldridge, K. and Zerbonia, R. , "Sensitivity Analysis of
Dispersion Models for Point and Area Sources," JAPCA, Vol. 32, #10,
October 1982.
Guideline on Air Quality Models (Revised), EPA-450/2-78-027R, July 1986.
Hodanbosi, R.F. and Peters, L.K., "Evaluation of RAM Model for Cleveland,
Ohio," JAPCA, Vol. 31, #3, March 1981.
Hwang, S.T., "Methods for Estimating On-Site Ambient Air Concentrations at
Disposal Sites," Nuclear and Chemical Waste Management, Vol. 7, 1987.
Irwin, J.S. and Brown, T.M., "A Sensitivity Analysis of the Treatment of Area
Sources by the Climatological Dispersion Model," JAPCA, Vol. 35, #4, April
1985.
Irwin, J.S., Chico, T. and Catalano, J., COM 2.0 — Climatological Dispersion
Model, EPA/600/8-85/029, PB86-136546, November 1985.
Rao, K.S., User's Guide for PEM-2: Pollution Episodic Model (Version 2),
EPA/600/8-86/040, PB 87-132 098, December 1986.
Schere, K.L. and Demerjian, K.L., User's Guide for the Photochemical Box Model
(PBM), PB85-137164, November 1984.
Schulze, R., Practical Guide to Atmospheric Dispersion Modeling, Trinity
Consultants, Inc., Dallas, 1989.
Schulze, R.H., Ed., "All Area Source Models Are Not Created Equal,"
Atmospheric Diffusion Notes, Trinity Consultants, Inc., Issue #7, July
1982.
Scire, J.S., Lurmann, F.W., Bass, A. and Hanna, S.R., User's Guide to the
Mesopuff II Model and Related Processor Programs, EPA-600/8-84-013, April
1984.
R-l
------- |