Development And Evaluation Of The Revised Area Source Algorithum For The Industrial Source Complex (isc 2) Long Term Model

     United States
     Environmental Protection
     Agency
Office of Air Quality
Planning and Standards
Research Triangle Park. NC 27711
EPA-454/R-92-016
October 1992
     Air
DEVELOPMENT AND EVALUATION
         OF A REVISED
    AREA SOURCE ALGORITHM
            FOR THE
 INDUSTRIAL SOURCE COMPLEX
      LONG TERM MODEL

-------
__                                   EPA-454/R-92-016

*
      DEVELOPMENT AND EVALUATION
                 OF A REVISED
          AREA SOURCE ALGORITHM
                   FOR THE
        INDUSTRIAL SOURCE COMPLEX
             LONG TERM MODEL
                U.S. Envi^r, ~ '    -:,-n Agency
                Region 5, Lu-.-r.,; (  ./ .,.;
                77 West Jackson Bc'^r^rrj -iyth r,
                Chicago, IL  60604-3590 '    °f
               Office Of Air Quality Planning And Standards
                   Office Of Air And Radiation
                U. S. Environmental Protection Agency
                 Research Triangle Park, NC 27711

                      October 1992

-------
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.   Any mention of trade
names or commercial products is not intended to constitute endorsement or recommendation for use.
                                    EPA-454/R-92-016

-------
                                PREFACE

      The ability to accurately estimate pollutant concentration due to
atmospheric releases from area sources is important to the modeling community,
and is of special concern for Superfund where emissions are typically
characterized as area sources. Limitations of the Industrial Source Complex
(ISC2) model (dated 92273) algorithms for modeling impacts from area sources,
especially for receptors located within and nearby the area, have been
documented in earlier studies. An improved algorithm for modeling dispersion
from area sources has been developed based on a numerical integration of the
point source concentration function.  Information on this algorithm is provided in
three interrelated reports.

      In the first report (EPA-454/R-92-014), an evaluation of the algorithm is
presented using wind tunnel data collected in the Fluid Modeling Facility of the
U.S. Environmental  Protection Agency. In the second report
(EPA-454/R-92-015), a sensitivity analysis is presented of the algorithm as
implemented in the short-term version of ISC2.  In the third report
(EPA-454/R-92-016), a sensitivity analysis is presented of the algorithm as
implemented in the long-term version of ISC2.

      The  Environmental Protection Agency must conduct a formal and public
review before the Agency can  recommend for routine use this new algorithm in
regulatory analyses. These reports are being released to establish a basis for
reviews of the capabilities of this methodology and of the consequences
resulting from use of this methodology in routine dispersion modeling of air
pollutant impacts. These reports are one part of a larger set of information on
the ISC2 models that must be considered before any formal changes can be
adopted.

-------
                       ACKNOWLEDGEMENTS

      This report has been prepared by Pacific Environmental Services, Inc.,
Research Triangle Park, North Carolina. This effort has been funded by the
Environmental Protection Agency under Contract No. 68D00124, with Jawad S.
Touma as Work Assignment Manager.  Special thanks go to John Irwin of EPA-
SRAB and William Petersen of EPA-AREAL, who provided helpful technical
guidance and suggestions.
                                 iv

-------
                             CONTENTS

 PREFACE	iii

 ACKNOWLEDGEMENTS	iv

 FIGURES	vi

 TABLES	xi

 1. INTRODUCTION	1

 2. THE ISC2 LONG TERM AREA SOURCE ALGORITHM	2
      2.1. The Shortcoming Of The Current ISCLT2 Area Source
      Algorithm	2
      2.2. The Implementation Of The Numerical Integration Algorithm	2
           2.2.1. Sector Average Calculation	3
           2.2.2. Smoothing the Frequency Distribution	6
           2.2.3. Convergence Criteria	6

 3. ISCLT2 AREA SOURCE ALGORITHM PERFORMANCE TEST	8
      3.1. Overview of the Performance Tests	8
      3.2. Results Of The Performance Tests		8
           3.2.1. Basic Performance Study:  Large Area Source and
           Idealized Meteorological Conditions	8
           3.2.2. Large Area Source With Idealized Hourly
           Meteorology Data Using Random  Wind Directions	28
           3.2.3. Examining The Source Geometry And Rotation
           Effects	36
           3.2.4. Large Area Source With Actual Meteorological
           Conditions	47
           3.2.5. Convergence Level Consideration	57

4. THE SENSITIVITY ANALYSIS	60
      4.1. Description Of The Study	60
      4.2. Results Of The Study	63
           4.2.1. Ground Level Sources With Downwind Receptors	53
           4.2.2. Elevated Area Source	75
           4.2.3. Ground-ievei Sources \/Vith Receptors Within and
           Nearby the Area Source	76

5. CONCLUSION	91

6. REFERENCES	93

-------
                               FIGURES

2.1.   Illustration of the Computation of the sector average impact	4
3.1 a.  Maximum Concentration Of ISCST Simulation And ISCLT
      Simulation Plotted With Down Wind Distance. 1000x1000m Area
      Source, A Stability Category	10
3.1 b.  Quartile Plot of Ratios (ISCLT/ISCST) by Convergence Levels For
      An 1000x1000m Area Source With Idealized Hourly Meteorological
      Data Set For ISCST Simulation, and the Idealized STAR Data Set
      For ISCLT Simulation. Stability Category A For All Data	11
3.1 c.  Quartile Plot of Ratios (ISCLT/ISCST) by Down Wind Distance For
      An 1000x1000m Area Source With Idealized Hourly Meteorological
      Data Set For ISCST Simulation, and the Idealized STAR Data Set
      For ISCLT Simulation. Stability Category A For All Data	12
3.2a.  Maximum Concentration Of ISCST Simulation And ISCLT
      Simulation Plotted With Down Wind Distance. 1000x1000m Area
      Source, D Stability Category	13
3.2b.  Quartile Plot of Ratios (ISCLT/ISCST) by Convergence Levels For
      An 1000x1000m Area Source With Idealized Hourly Meteorological
      Data Set For ISCST Simulation, and the Idealized STAR Data Set
      For ISCLT Simulation. Stability Category D For All Data	14
3.2c.  Quartile Plot of Ratios (ISCLT/ISCST) by Down Wind Distance For
      An 1000x1000m Area Source With Idealized Hourly Meteorological
      Data Set For ISCST Simulation, and the Idealized STAR Data Set
      For ISCLT Simulation. Stability Category D For All Data	15
3.3a.  Maximum Concentration Of ISCST Simulation And ISCLT
      Simulation Plotted With Down Wind Distance. 1000x1000m Area
      Source, F Stability Category	16
3.3b.  Quartile Plot of Ratios (ISCLT/ISCST) by Convergence Levels For
      An 1000x1000m Area Source With Idealized Hourly Meteorological
      Data Set For ISCST Simulation, and the Idealized STAR Data Set
      For ISCLT Simulation. Stability Category F For All Data	17
3.3c.  Quartile Plot of Ratios (ISCLT/ISCST) by Down Wind Distance For
      An 1000x1000m Area Source With Idealized Hourly Meteorological
      Data Set For ISCST Simulation, and the Idealized STAR Data Set
      For ISCLT Simulation. Stability Category F For All Data	18
3.4a.  Maximum Concentration Of iSCST Simuiaiion And ISCLT
      Simulation Plotted With Down Wind Distance. 1000x200m Area
      Source, A Stability Category	19
3.4b.  Quartile Plot of Ratios (ISCLT/ISCST) by Convergence Levels For
      An 1000x200m Area Source With  Idealized Hourly Meteorological
      Data Set For ISCST Simulation, and the Idealized STAR Data Set
      For ISCLT Simulation. Stability Category A For All Data	20
                                  VI

-------
3.4c.  Quartile Plot of Ratios (ISCLT/ISCST) by Down Wind Distance For
      An 1000x200m Area Source With Idealized Hourly Meteorological
      Data Set For ISCST Simulation, and the Idealized STAR Data Set
      For ISCLT Simulation. Stability Category A For All Data	21
3.5a.  Maximum Concentration Of ISCST Simulation And ISCLT
      Simulation Plotted With Down Wind  Distance. 1000x200m Area
      Source, D Stability Category	22
3.5b.  Quartile Plot of Ratios (ISCLT/ISCST) by Convergence Levels For
      An 1000x200m Area Source With Idealized Hourly Meteorological
      Data Set For ISCST Simulation, and the Idealized STAR Data Set
      For ISCLT Simulation. Stability Category D For All Data	23
3.5c.  Quartile Plot of Ratios (ISCLT/ISCST) by Down Wind Distance For
      An 1000x200m Area Source With Idealized Hourly Meteorological
      Data Set For ISCST Simulation, and the Idealized STAR Data Set
      For ISCLT Simulation. Stability Category D For All Data	24
3.6b.  Quartile Plot of Ratios (ISCLT/ISCST) by Convergence Levels For
      An 1000x200m Area Source With Idealized Hourly Meteorological
      Data Set For ISCST Simulation, and the Idealized STAR Data Set
      For ISCLT Simulation. Stability Category F For All Data	26
3.6c.  Quartile Plot of Ratios (ISCLT/ISCST) by Down Wind Distance For
      An 1000x200m Area Source With Idealized Hourly Meteorological
      Data Set For ISCST Simulation, and the Idealized STAR Data Set
      For ISCLT Simulation. Stability Category F For All Data	27
3.7a.  Maximum Concentration Of ISCST Simulation And ISCLT
      Simulation Plotted With Down Wind  Distance. 1000x1000m Area
      Source, RDU 1987 RANDOM And STAR Data	30
3.7b.  Quartile Plot of Ratios (ISCLT/ISCST) by Convergence Levels For.
      An 1000x1000m Area Source With RDU 1987 RANDOM Hourly
      Meteorological Data Set For ISCST Simulation, and the RDU 1987
      STAR Data Set For ISCLT Simulation	31
3.7c.  Quartile Plot of Ratios (ISCLT/ISCST) by Down Wind Distance For
      An 1000x1000m Area Source With RDU 1987 RANDOM Hourly
      Meteorological Data Set For ISCST Simulation, and the RDU 1987
      STAR Data Set For ISCLT Simulation	32
3.8a.  Maximum Concentration Of ISCST Simulation And ISCLT
      Simulation Plotted With Down Wind  Distance. 1000x200m Area
      Source. RDU 1987 RANDOM And STAR Data	33
3.8b.  Quartile Plot of Ratios (ISCLT/ISCST) by Convergence Levels For
      An iOOOx200m Area Source With RDU 1987 RANDOM Houny
      Meteorological Data Set For ISCST Simulation, and the RDU 1987
      STAR Data Set For ISCLT Simulation	34
3.8c.  Quartile Plot of Ratios (ISCLT/ISCST) by Down Wind Distance For
      An 1000x200m Area Source With RDU 1987 RANDOM Hour
      Meteorological Data Set For ISCST Simulation, and the RDU 1987
      STAR Data Set For ISCLT Simulation	35
                                 VII

-------
3.9.   Contour Diagram of Annual Average Rural Concentration
      for An 100x100m Area Source With Winds Come Only From 0
      Degree North	.'	38
3.10.  Contour Diagram of Annual Average Rural Concentration (jag/m^)
      for An 100x100m Area Source With Winds Come Only From 45.0
      Degree Northeast	39
3.11.  Contour Diagram of Annual Average Rural Concentration (jig/m^)
      for An 100x100m Area Source With 45.0 Degree Rotation and
      Winds Come Only From 45.0 Degree Northeast	40
3.12.  Contour Diagram of Annual Average Rural Concentration (jig/m^)
      for An 400x100m Area Source With Winds Come Only From 0
      Degree North	41
3.13.  Contour Diagram of Annual Average Rural Concentration (ng/m3)
      for An 400x100m Area Source With Winds Come Only From 45.0
      Degree Northeast	42
3.14.  Contour Diagram of Annual Average Rural Concentration (p.g/m3)
      for An 400x100m Area Source With 45.0 Degree Rotation and
      Winds Come Only From 45.0 Degree Northeast	43
3.15a. Maximum Concentration Of ISCST Simulation And ISCLT
      Simulation Plotted With Down Wind Distance. 1000x1000m Area
      Source, RDU 1987 RAMMET And STAR Data	48
3.15b. Quartile Plot of Ratios (ISCLT/ISCST) by Convergence Levels For
      An 1000x1000m Area Source With RDU 1987 RAMMET Hourly
      Meteorological Data Set For ISCST Simulation, and the RDU 1987
      STAR Data Set For ISCLT Simulation	49
3.15c. Quartile Plot of Ratios (ISCLT/ISCST) by Down Wind Distance For
      An 1000x1000m Area Source With RDU 1987 RAMMET Hourly
      Meteorological Data Set For ISCST Simulation, and the RDU 1987
      STAR Data Set For ISCLT Simulation	."	50
3.16a. Maximum Concentration Of ISCST Simulation And ISCLT
      Simulation Plotted With Down Wind Distance. 1000x200m Area
      Source, RDU 1987 RAMMET And STAR Data	51
3.16b. Quartile Plot of Ratios (ISCLT/ISCST) by Convergence Levels For
      An 1000x200m Area Source With RDU 1987 RAMMET Hourly
      Meteorological Data Set For ISCST Simulation, and the RDU 1987
      STAR Data Set For ISCLT Simulation	52
3.16c. Quartile Plot of Ratios (ISCLT/ISCST) by Down Wind Distance For
      An 1000x200m Area Source With RDU ^337 RAMMET Hcur
      Meteorological Data Set For ISCST Simulation, and the RDU 1987
      STAR Data Set For ISCLT Simulation	53
3.17.  Maximum Concentration Of ISCLT Simulation With One
      1000x1000m Area Source And ISCLT Simulation With This  Source
      Broken Down Into Four 500x500m Area Sources, RDU 1987 STAR
      Data	55
                                 VIII

-------
                               TABLES

3.1.   The Tests For Source Geometry And Rotation Effects	36
3.2a.  10 Maximum Annual Averages For Case 3.3.1.1	45
3.2b.  10 Maximum Annual Averages For Case 3.3.1.2	45
3.2c.  10 Maximum Annual Averages For Case 3.3.1.3	45
3.3a.  10 Maximum Annual Averages For Case 3.3.2.1	46
3.3b.  10 Maximum Annual Averages For Case 3.3.2.2	46
3.3c.  10 Maximum Annual Averages For Case 3.3.2.3	46
3.4a.  Maximum Annual Average Concentration Vs. Down Wind Distance
      1000x1000m Source, RDU 1987 Star Data	59
3.4b.  Maximum Annual Average Concentration Vs. Down Wind Distance
      1000x200m Source, RDU 1987 Star Data	59
4.1.   Area Source Scenarios for Sensitivity Analysis	62
4.2.   Area Source Inputs for X-Large, Close-in Scenario	62
4.3.   Comparison of Design Concentrations (i^g/m3) for the Very Small
      Source  (10m Width)	65
4.4.   Comparison of Design Concentrations (ng/m3) for the Small
      Source  (50m Width)	65
4.5.   Comparison of Design Concentrations (jig/m3) for the Medium
      Source  (100m Width)	65
4.6.   Comparison of Design Concentrations (jig/m3) for the Large
      Source  (500m Width)	66
4.7.   Comparison of Design Concentrations (ng/m3) for the Very Large
      Source  (1000m Width)	66
4.8.   Comparison of Design Concentrations (jig/m3) for the Medium
      Elevated Source (10Qm Width)	75
4.9.   Comparison of Annual Average Concentrations (jig/m3) for the
      1000m Wide Area With Receptors Located Within and Nearby the
      Area	78
                                  XI

-------
                            1.  INTRODUCTION

       Previous model evaluation studies (EPA, 1989) have pointed out the
deficiencies of the virtual point source algorithm for modeling area sources used
in the Industrial Source Complex (ISC2) Long Term (ISCLT2) model. While it is
computationally efficient, the virtual point source algorithm used in the original
ISCLT2 model gives physically unrealistic results for receptors located  near the
edges and corners of the area. Also, the algorithm cannot predict the area
source impact for receptors located inside the source itself, and it does not
adequately handle effects of complex source-receptor geometry.

       This report documents the development and evaluation of a new area
source algorithm for the ISCLT2 model, based on the numerical integration
algorithm recently developed for the ISC2 Short Term (ISCST2) model. The
evaluation  of the performance of the new ISCLT2 area source algorithm includes
performance tests, statistical analyses, and sensitivity analyses.

       For the performance tests, the new ISCLT2 numerical integration area
source algorithm is challenged in various ways. First, quality assurance tests
are conducted to examine the reasonableness of the results and the efficiency of
the algorithm.  These quality assurance tests include printing out the
intermediate calculation results to perform a line-by-line check of the computer
code of the new algorithm. Second, cases with simple area source
characteristics (square area source or rectangular area source) and idealized
meteorological conditions are used to examine the reliability and accuracy of the
algorithm.  Third, several tests are conducted to show the concentration
distribution for various area source shapes.  Fourth, tests are conducted to
examine  the effects of subdividing the area source and the effects of rotation of
the area  source on the simulated  concentration values. Finally, several cases
are examined using realistic meteorological conditions.

       In  addition to performance tests, a sensitivity analysis is presented
comparing  design concentrations using the virtual point source algorithm with
estimates using the  new numerical integration algorithm for a range of source
characteristics and meteorological data.

      The  technical description of the new 1SCLT2 area  source algorithm is
provided  in Section 2. The results of the performance tests are  presented  in
Section 3, and the results of the sensitivity analyses involving comparisons with
the virtual point source algorithm are given in Section 4.  The conclusions of this
study are presented in Section 5.

-------
           2. THE ISC2 LONG TERM AREA SOURCE ALGORITHM

 2.1. The Shortcoming Of The Current ISCLT2 Area Source Algorithm

      The algorithm used in ISCLT2 (Version 92062) for modeling area sources
 is based on the virtual point source approach. As suggested by Turner (1970),
 the virtual point source algorithm assumes that the 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 emission rate for the replacement source is set equal
 to the area source emission rate.  Therefore, the same form of calculation
 process used for point sources can be used for area sources.

      While it is computationally efficient, the virtual point source algorithm
 used in the original ISCLT2 model  has several inevitable shortcomings (EPA,
 1989). First, the algorithm does not accurately account for the impacts for
 receptors  located inside the area source itself.  The ISCLT2 model flags
 receptors  located within the area, and sets the concentration value to 0 at those
 receptors.  Second, the virtual point source algorithm is only valid for receptors
 at a sufficient distance downwind from an area source that the area source
 impact is well approximated by a point source.  Hence, for receptors close to the
 area source where the source-receptor geometry is crucial, the virtual point
 source approximation performs poorly. Third, the algorithm performs best for
 simple square shaped areas. For large area sources with complex shapes, the
 area must be subdivided into smaller square sources.

 2.2.  The Implementation Of The Numerical Integration Algorithm

      Several factors need to be considered when implementing an area source
 algorithm in the ISCLT2 model. The first and most important issue is the usage
 of the STAR (for STability ARray) meteorological data.  The ISCLT2 is a
 climatological model that uses a summary of the wind directions, wind speeds
 and stability categories encountered throughout a period (e.g., a calendar
 quarter, a  year,  or multiple years).  Therefore, the meteorological conditions are
 summarized by using a frequency distribution composed of 16 wind direction
 sectors, 6  wind speed classes and  6 stability classes.  Since all the wind
 directions  within a sector of the STAR data are assumed to oe equally iiksiy, the
 ISCLT2 model calculates sector average concentrations to determine source
 impacts. In order to account for the abrupt changes that occur in the frequency
 of occurrence of meteorological conditions at the boundaries between adjacent
 sectors, an adjustment is made to the concentration distribution. This
 adjustment is performed in the existing algorithm by a applying a smoothing
function that linearly interpolates between the concentration values calculated at
the centerline of adjacent sectors.

-------
      Since the area source is approximated by a point source in the current
ISCLT2 model, and the wind direction is equally likely to occur anywhere within
the sector, the impact at a particular distance downwind from the source does
not vary by changing the wind direction within the sector. The sector average
concentration can therefore be calculated by using a single wind direction
corresponding to the centerline of the sector.  In the new area source algorithm,
the model is treating the source as an area, and the impact at a particular
distance downwind from the source does vary by changing the wind direction
within the sector.  Therefore, the sector average calculation will need to be
based on several simulations, each corresponding to a particular wind direction
within the sector.  Instead of applying a smoothing function to the concentration
distribution with the  new algorithm, the abrupt changes in concentration at the
sector boundaries are smoothed by applying a linear interpretation to Calculate
the frequency of occurrence corresponding to each wind direction simulated.

      Another factor worth considering is that there are certain benefits to
maintaining consistency between the ISC2 short term (ISCST2) area source
algorithm and the ISCLT2 area source algorithm. These benefits include
simplifying future maintenance of the models, keeping compatibility of source
input parameters between the two models, and better consistency of results
between the two models for the same source characteristics. The new  ISCST2
area source algorithm is based on a numerical integration of the point source
concentration function over the area, and employs a Romberg integration
algorithm (Press, et al, 1986) to improve the efficiency of the computations.

      The  implementation of the new area source algorithm in the ISCLT2
model is described in more detail in the following sections.

2.2.1. Sector Average Calculation

      The  STAR meteorological  data provides the frequency of occurrence for
each of the 16 wind direction sectors. It assumes that, within each sector, all the
wind directions are equally likely. However, even for a very simple area source
shape, the  source-receptor geometry varies with the wind direction. For
example, in Figure 2.1, if the wind comes from the north, the distance it travels
over the area source is d.  If the wind comes from a direction of 5 degrees east
from north, the distance it travels over the area source is d'. In this example, d'
;s larger than d.  For 3 recsptor downwind of A.he area source, different  wind
directions within the sector result in different impacts to the receptor. Therefore,
the sector average of the concentration value cannot be calculated through the
use  of only one wind direction.

-------
                                    5 Degree
                           \  22.5 D
                            \
                           d«:::::»::::::::::::::i:::::::::::::
                                      Receptor
Figure 2.1.   Illustration of the computation of  the sector average impact.

-------
       In order to calculate the sector average concentration, several simulations
are needed for a selection of wind directions within the sector.  This selection of
wind directions can be done as follows.  First, the directions corresponding to
the boundaries of the sector, together with the direction of the centerline of the
sector are selected.  The area source impact to the receptor is computed for
these three wind directions, and a sector average calculated. Next, two more
wind directions are selected,  such that they are equally spaced between the
central azimuth of the sector and the sector boundaries. The sector average
area source impact is computed by trapezoidal integration using the impacts
computed for the five wind directions. The trapezoidal integration is used
because the directions corresponding to the sector boundaries are also used in
calculations for the adjacent sectors, and are therefore weighted by a factor of
one half.  A test is made to see if the area source impact using five wind
directions is significantly different from that obtained using three wind directions.
If the estimates differ by more than 2 percent, the sector is further subdivided,
until the 2 percent convergence criterion is satisfied.  The sector average is
calculated using a trapezoidal integration as follows:
                                                                (6)  (2-1)
                          Xjnid
                                  ,  .,	                     (2-2)
                                  Inud      o
where:       xi =   the sector average of the concentration value in ith
                   sector.
             S =   the sector width.
             f jj =   the frequency of occurrence for jth wind direction in ith
                   sector.
             z(Q) = srror term. In practice, a criterion of 5(6) < 2 percsnt is
                   used to check for convergence of the algorithm.
             x(9j) = the concentration value in ith sector.
             x(6y) = the concentration value with jth wind direction in ith
                   sector.
             0ij =  the jth wind direction in ith sector, j = 1  and N represent
                    the two boundaries of ith sector.

-------
2.2.2.  Smoothing the Frequency Distribution

       The application of a smoothing function to the concentration distribution,
as done in the current ISCLT2 algorithm, is not applicable to the numerical
integration algorithm because, as noted above, the impact at a given distance
downwind varies as a function of wind direction within the sector. In order to
avoid abrupt changes in the concentrations at the sector boundaries with the
new algorithm, a linear interpolation is used to determine the frequency of
occurrence of each wind  direction used for the individual simulations within a
sector, based on the frequencies of occurrence in the adjacent sectors.  This
"smoothing" of the frequency distribution has a similar effect as the smoothing
function used with the current ISCLT2 algorithm.  The frequency of occurrence
for the jth wind direction between i and i+1 sector can be calculated as:
             f  = Fj + (0i+1 -  GJ ) (Fi+1 - Fj) / (0i+1 -  0j)               (2-3)
where:       Fj =   the frequency of occurrence of wind directions for the ith
                   sector.
             Fj+1 =  the frequency of occurrence of wind directions for the
                   i+1th sector.
             0j =  the central wind direction for the ith sector.
             0j+i = the central wind direction for the i+1th sector.
             GJJ =  the wind direction between 0j and  0i+1
             fy =   the frequency of occurrence for the wind direction GJJ.


2.2.3.  Convergence Criteria

      This section describes the convergence criteria used to determine when
the area source calculations for a specific sector are completed. For each
combination of wind speed class, stability class and wind direction sector in the
STAR data file, at least 5 wind directions are used to approximate the sector
average area source impact. The number of wind direction simulations used, N.
can be calculated as:

             N = 2k + 1                                   (2-4)

where k is the referred  to as the level number.  If k = 1, a total of 3 wind
directions are used. This is called level 1. For level 5 (k=5), a total of 33 wind
directions are used. These wind directions are equally distributed inside the
22.5 degree sector. For example, in the case of level 5, the wind directions are
equally distributed inside the sector with a 0.68 (= 22.5/33) degree interval.

-------
      After calculations are completed for each level, the results are compared
with results for the previous level. One convergence criterion that the model
checks is whether the results for successive levels agree within 2 percent.  If the
2 percent convergence criterion is not achieved after level 2, for example, then
the model increases one more level, to level 3, which has 9 simulations.  Since 5
of these wind directions were used in the previous calculation, the model only
performs calculations for the 4 new wind directions. This procedure is
computationally efficient.

      Although this algorithm is known to converge (to within 2 percent)
eventually, the run time may be excessive for some situations. Using the
ISCST2 model, one can calculate the annual average by using the hourly
meteorological data, which requires only 8760 hourly simulations. This
corresponds roughly with the number of simulations needed for level 4
(17*576=9792).  If the ISCLT2 model were to routinely employ 10 levels (1,025
simulations for each of the 576 STAR combinations), it would be much more
efficient to run the ISCST2 model. For this reason, the algorithm is designed to
stop calculating after a certain level is reached. Several tests are needed to
determine the optimum level to ensure both reasonable  model run times and
acceptable accuracy.  The results of these tests are presented in Section 3.

      In addition to the two convergence criteria discussed above, i.e., the 2
percent comparison between results for successive levels, and the maximum
number of computational levels, a third criterion is incorporated into the
algorithm  in order to further optimize its performance. With numerical schemes
of the kind described here, it is often most difficult to achieve convergence for
very small concentration values, where truncation errors can be significant.
Since these values are also of less concern to the typical user, the model will
stop any further calculations for a given STAR  combination if the concentration
estimate is less than 1.0E-10. This avoids making excessive computations for
cases where  the algorithm is essentially trying  to converge on zero.

-------
       3.  ISCLT2 AREA SOURCE ALGORITHM PERFORMANCE TEST

 3.1. Overview of the Performance Tests

       In order to evaluate the performance of the modified ISCLT2 area source
 algorithm, several tests were designed. These tests can be classified into three
 categories.  The first category is to test the overall accuracy and performance of
 the algorithm using very idealized meteorological inputs, the second category is
 to test the reasonableness of the algorithm's performance for various source-
 receptor geometries, and the third  is to examine the algorithm's behavior in more
 detail using a realistic distribution of meteorological conditions. The latter group
 includes a series of tests to evaluate the optimum set of convergence criteria for
 the numerical integration algorithm in order to achieve an appropriate balance
 between accuracy and model run time. The performance tests include point-to-
 point comparisons, quality assurance tests, and statistical analyses.  Tables and
 graphs are used to present the analytical results in a comprehensive way.

 3.2. Results Of The Performance Tests

 3.2.1.  Basic Performance Study: Large Area Source and Idealized
            Meteorological Conditions

       The main  purpose of this test is to verify that the numerical integration
 algorithm has been correctly  implemented into the ISCLT2 model.  The test
 consists of comparisons of results from the ISCLT2 model with results from the
 ISCST2 model using the numerical integration algorithm for very idealized
 meteorological conditions.  The meteorological conditions consist of a single
 wind speed and stability category with a uniform distribution of wind directions in
 order to force the ISCST2 model to simulate sector averages for comparison
 with ISCLT2. In this study, a  1000x1000m square source and a 1000x200m
 rectangular source are used.  One  polar network of receptors is used, with the
 origin of the network located at the center of the area source.  The polar receptor
 network has seven distance rings of 250, 500, 750, 1000, 1500, 5000, and
 15000  meters, and 36 direction radials (every 10 degrees), for a total of 252
 receptors.

      To idealize the meteorological conditions, a single wind soeed and
 stabiiity category are usea for each rest. Three hourly meteorological data files
were generated for use by the ISCST2 model, one each for stability category A
 (unstable), D (Neutral), and F(Stable),  respectively.  The wind direction was
altered 0.5 degrees clockwise for each hour, and a 360 day period was used to
approximate sector average annual concentration values from ISCST2. The
 ISCLT2 model was run using  a STAR meteorological data file with frequencies
specified to select the same stability category and wind speed as used in the

-------
hourly data files for ISCST2, and a uniform distribution of frequencies for all
sectors.

      The results of the comparison between ISCST2 and ISCLT2 with the
numerical integration area source algorithm and the idealized meteorological
conditions, presented in Figures 3.1 to 3.6, are very encouraging. Three figures
are provided for each combination of source type (1000x1000m square or
1000x200m rectangle) and stability category (A, D or F). The first figure in each
group shows the maximum concentration for both the ISCST2 model and the
ISCLT2 model as a function of downwind distance. The results from the two
models are virtually indistinguishable on these plots, suggesting that the
algorithm has been correctly implemented in the ISCLT2 model.  In order to
provide a more detailed comparison of the results of the two models, the two
additional figures for each case show the "quartiles" of the ratio of
ISCLT2/ISCST2 results for all receptors, first as a function of convergence level
used in the ISCLT2 model, and second as a function of downwind distance using
no limit on the number of convergence levels (full convergence). These quartile
plots show the maximum and minimum ratios, together with the ratios that are
exceeded 25 percent of the time, 50 percent of the time, and 75 percent of the
time.

      The series of quartile plots show that the ISCST2 and ISCLT2 models
agree within ± 1 percent in nearly all cases (ratios between 0.99 and 1.01), and
that 50 percent of the ratios (between the 25 percent and 75 percent quartiles)
fall  between 0.9975 and 1.0025, corresponding to differences of less than 0.25
percent.  The quartile plots showing ratios as a function of downwind distance
show that the closest agreement occurs for the largest concentrations at
receptors located within or near the area source. The figures also show that the
ISCLT2 converge to relatively stable results by about convergence level 5,
corresponding to 33 separate wind direction simulations per 22.5 degree sector.

-------
          Maximum Cone. Vs. Down Wind Distance
                  1000x1000m Source, Idea Data, A Stab.
   1000
*
«


o<



o
•;=
(0


4)
O

O
O
   0.01 d
  0.001
                          5    6    7    8   9   10

                             Down Wind Distance (KM)
            11
                 12
                     13
                         14
                             15
                        ISCST2 Simulation
ISCLT2 Simulation
   Figure 3.1 a. Maximum Concentration Of ISCST Simulation And ISCLT
              Simulation Plotted With Downwind Distance. 1000x1000m Area
              Source, A Stability Category
                                  10

-------
1.0V
Ratio (ISCLT/ISCST) by Converg. Levels
1000X1000m Area Source, Case 2.1.1
1.0075-
« 1.005-

£.
O 1.0025-
(0
O
0.9975-I
0.995-
cc

0.9925-
0.99
-e B-
i i i i i i i i F r i i i i i i r i r i
0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21
Convergence Levels
Mac Ratio-S-76% M
-------
1.01
Ratio (ISCLT/ISCST) Vs. Distance
1000X1000m Area Source, Case 2.1.1
1.0075-
1.005-
1.0025-
0.9875-
0.895-
a9925-
0.99-
T 1 1—I I I I I
1000
100
1—I—I—I I I I I—
10000
T 1 1 1—I I I I
100000
Down Wind Distance (Meters)
• Max. Ratio -3- 75% Marie
26% Mark -jfc- Mia Ratio
60% Mark
Figure 3.1c. Quartile Plot of Ratios (ISCLT/ISCST) by Downwind Distance For
An 1000x1000m Area Source With Idealized Hourly Meteorological
Data Set For ISCST Simulation, and the Idealized STAR Data Set
For ISCLT Simulation. Stability Category A For All Data. No limit
on convergence level limit
12
-------
Maximum Cone. Vs. Down Wind Distance
1000x1000m Source, Idea Data, D Stab.
o.ooi
5 6 7 8 9 10
Down Wind Distance (KM)
11
13
14
15
ISCST2 Simulation
ISCLT2 Simulation
Figure 3.2a. Maximum Concentration Of ISCST Simulation And ISCLT
Simulation Plotted With Downwind Distance. 1000x1000m Area
Source, D Stability Category
13
-------
1.01
Ratio (ISCLT/ISCST) by Converg. Levels
1000X1000m Area Source, Case 2.1.2
-E-
-X-
-X-
-X
-Q
01234 66 7 8 9 10 11 12 13 14 16 16 17 16 19 20 21
Convergence Levels
-•-Max Ratio -X- 75% Mark -X-50%Mark
-B-25%Mak rA-Min. Ratio
Figure 3.2b. Quartiie Plot of Ratios (ISCLT/ISCST) by Convergence Levels For
An 1000x1000m Area Source With Idealized Hourly Meteorological
Data Set For ISCST Simulation, and the Idealized STAR Data Set
For ISCLT Simulation. Stability Category D For All Data. No limit
on convergence levels
14
-------
1.01
0.99
Ratio (ISCLT/ISCST) Vs. Distance
1000X1000m Area Source, Case 2.1.2
100
1000
Down Wind Distance (Meters)
100000
Max. Ratio -S-75%Mark -X-50%Mark
2S%Maric -
Figure 3.2c. Quartile Plot of Ratios (ISCLT/ISCST) by Downwind Distance For
An 1000x1000m Area Source With Idealized Hourly Meteorological
Data Set For ISCST Simulation, and the Idealized STAR Data Set
For ISCLT Simulation. Stability Category D For All Data.
15
-------
Maximum Cone. Vs. Down Wind Distance
1000x1000m Source, Idea Data, F Stab.
1000s
100=
0
~4
4J
0
h
•U

0>
O

0
O

•*
X
«
0.001
0.01 d
T
9
5 6 7 8 9 10

Down Wind Distance (KM)
11
-r
12
13
14
15
ISCST2 Simulation
ISCLT2 Simulation
Figure 3.3a. Maximum Concentration Of ISCST Simulation And ISCLT
Simulation Plotted With Downwind Distance. 1000x1000m Area
Source, F Stability Category
16
-------
1.01
Ratio (ISCLT/ISCST) by Converg. Levels
1000X1000m Area Source, Case 2.1.3
0 1 2 3 4 6 6 7 8 8 10 11 12 13 14 IB 16 17 18 18 20 21
Convergence Levels
Max. Ratio -H- 7B%Mvk -X-60%Mak
-Q-25% Ma* -afc-Min. Ratio
Figure 3.3b. Quartiie Plot of Ratios (ISCLT/ISCST) by Convergence Levels For
An 1000x1000m Area Source With Idealized Hourly Meteorological
Data Set For ISCST Simulation, and the Idealized STAR Data Set
For ISCLT Simulation. Stability Category F For All Data.
17
-------
£
O
1.01
1.0075-
1.005-
1.0025-
0.0975-I
0.995-
0.9925-
0.96-
Ratio (ISCLT/ISCST) Vs. Distance
1000X1000m Area Source, Case 2.1.3
100
T 1—I I I I I—
1000
1 r

Down Wind Distance (Meters)
i—i—i i i i i
10000
~1—I—I—I I I
100000
Max Ratio
25% Mark
75% Mark -X-50%Mark
Min. Ratio
Figure 3.3c. Quartile Plot of Ratios (ISCLT/ISCST) by Downwind Distance For
An 1000x1000m Area Source With Idealized Hourly Meteorological
Data Set For ISCST Simulation, and the Idealized STAR Data Set
For ISCLT Simulation. Stability Category F For All Data. No limit on
convergence levels.
18
-------
Maximum Cone. Vs. Down Wind Distance
1000x200m Source, Idea Data, A Stab.
1000
0.001
Down Wind Distance (KM)
ISCST2 Simulation
ISCLT2 Simulation
Figure 3.4a. Maximum Concentration Of ISCST Simulation And ISCLT
Simulation Plotted With Downwind Distance. 1000x200m Area
Source, A Stability Category
19
-------
1.01'
Ratio (ISCLT/ISCST) by Converg. Levels
1000X200m Area Source, Case 2.2.1
1.0075-
1.005-
1.0025-
1-
O 0.9075-
V)
0.995-
0.9925-
0.99-
0 1 2 3 4 5 8 7 8 9 10 11 12 13 14 15 18 17 18 19 20 21
Convergence Levels
Max. Ratio-S-75%Mark -K-50%Mark
25%M*k -^k-Min. Ratio
Figure 3.4b. Quartile Plot of Ratios (ISCLT/ISCST) by Convergence Levels For
An 1000x200m Area Source With Idealized Hourly Meteorological
Data Set For ISCST Simulation, and the Idealized STAR Data Set
For ISCLT Simulation. Stability Category A For All Data.
20
-------
1.01
1.0075-
1.005-
1.0025-
0.9075-
0.996-
0.9025-
Ratio (ISCLT/ISCST) vs. Down Wind Dist.
10OOX200m Area Source, Case 2.2.1
100
1000
Down Wind Distance (Meters)
-•-Max Ratio-S-75%Mark -X-60%Mark
-B- 25% Mark -jfc- Ma Ratio
Figure 3.4c. Quartile Plot of Ratios (ISCLT/ISCST) by Downwind Distance For
An 1000x200m Area Source With Idealized Hourly Meteorological
Data Set For ISCST Simulation, and the Idealized STAR Data Set
For ISCLT Simulation. Stability Category A For All Data. No limit
on convergence levels.
21
-------
Maximum Cone. Vs. Down Wind Distance
1000x200m Source, Idea Data, D Stab.
1000=3
0.001
6789
Down Wind Distance (KM)
10
T
11
12
13
14
15
ISCST2 Simulation
ISCLT2 Simulation
Figure 3.5a. Maximum Concentration Of ISCST Simulation And ISCLT
Simulation Plotted With Downwind Distance. 1000x200m Area
Source, D Stability Category
22
-------
Ratio (ISCLT/ISCST) by Converg. Levels
10OOX200m Area Source, Case 2.2.2
1.01
s
-§-
-B-
-e-
0 1 2 3 4 5 6 7 8 B 10 11 12 13 14 15 16 17 18 18 20 21
Convwgwic* L»v«ls
Max. Ratio -S- 75% Mark -X- 50%MarV
25% Mark -^-Win. Ratio
Figure 3.5b. Quartile Plot of Ratios (ISCLT/ISCST) by Convergence Levels For
An 1000x20bm Area Source With Idealized Hourly Meteorological
Data Set For ISCST Simulation, and the Idealized STAR Data Set
For ISCLT Simulation. Stability Category D For All Data.
23
-------
1.01
age
Ratio (ISCLT/ISCST) vs. Down Wind Dist.
10OOX200m Area Source, Case 2.2.2
100
1000
Down Wind Distance (Meters)
-•-Max. Ratio-X-75% Martc -X-50%M»k
-B-25% Marie -dk- Mn. Ratio
Figure 3.5c. Quartile Plot of Ratios (ISCLT/ISCST) by Downwind Distance For
An 1000x200m Area Source With Idealized Hourly Meteorological
Data Set For ISCST Simulation, and the Idealized STAR Data Set
For ISCLT Simulation. Stability Category D For All Data. No limit
on convergence levels.
24
-------
Maximum Cone. Vs. Down Wind Distance
1000x200m Source, Idea Data, F Stab.
1000
r 001
-r
8
T
9
6 7 8 9 10
Down Wind Distance (KM)
11
T
12
13
14
15
ISCST2 Simulation
ISCLT2 Simulation
Figure 3.6a. Maximum Concentration Of ISCST Simulation And ISCLT
Simulation Plotted With Downwind Distance. 1000x200m Area
Source, F Stability Category
25
-------
1.01
Ratio (ISCLT/ISCST) by Converg. Levels
1000X200m Area Source, Case 2.2.3
1.0075-
1.005-
1.0025-
_J
O 0.9975-
(n
0.995-
0.9625-
-s-
-X-
-B-
-S-
-X-
-X
-ED
0 1 2 3 4 6 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21
Convergence Levels
Max Ratio-X-75% Mark -X-50% Marie
25% Mark -±- Mia Ratio
Figure 3.6b. Quartile Plot of Ratios (ISCLT/ISCST) by Convergence Levels For
An 1000x200m Area Source With Idealized Hourly Meteorological
Data Set For ISCST Simulation, and the Idealized STAR Data Set
For ISCLT Simulation. Stability Category F For All Data.
26
-------
Ratio (ISCLT/ISCST) vs. Down Wind Dist.
1000X200m Area Source, Case 2.2.3
1.01
100
1000
Down Wind Distance (Meters)
Max. Ratio-S-75%Mark -X-50%Mart<
25%Mafk -Jk-Ma Ratio
Figure 3.6c. Quartile Plot of Ratios (ISCLT/ISCST) by Downwind Distance For
An 1000x200m Area Source With Idealized Hourly Meteorological
Data Set For ISCST Simulation, and the Idealized STAR Data Set
For ISCLT Simulation. Stability Category F For All Data. No limit
on convergence levels.
27
-------
3.2.2. Large Area Source With Idealized Hourly Meteorology Data Using
Random Wind Directions

In the previous section, it was assumed that the frequency of occurrence
was equal for all sectors. Using very idealized meteorological data to implement
that assumption, the ISCLT2 model successfully reproduces the sector average
concentration values calculated by the ISCST2 model. However, such a test
does not challenge the algorithm for more realistic meteorological conditions,
and especially for the smoothing function applied to the frequency distribution by
ISCLT2 for cases of non-uniform distributions. Therefore, a similar test was
performed comparing ISCLT2 and ISCST2 results using more realistic
meteorology data distributions.

The ISCLT2 model was run for the same two area sources and receptor
networks described in the previous section, with an actual STAR meteorological
data summary from Raleigh-Durham, NC, (RDU) for the 1987 annual period. In
order to simulate the sector averages with the ISCST2 model, an hourly
meteorological data file was generated that produces the same frequency
distribution as the 1987 RDU annual STAR summary. The wind speeds and
mixing heights in the hourly meteorological data file were set to match the
corresponding values used by the ISCLT2 model. To approximate the sector
average impacts, the hourly wind directions were randomly distributed within the
applicable 22.5 degree sector for use with the ISCST2 model. It is worth noting
that, although the generated hourly data set eliminates the discrepancies for the
wind speeds and mixing heights, there is no guarantee that the wind directions
are uniformly distributed within the wind sector. Therefore, some discrepancies
between the ISCLT2 and ISCST2 results are expected.

Figure 3.7a-c shows the results for the 1000x1000m square source, and
Figure 3.8a-c shows the results for the 1000x200m rectangular source. Part a of
each figure shows the maximum concentrations as a function of distance for both
the ISCLT2 and ISCST2 models. As with the very idealized case presented
earlier, the concentrations from the two models are almost indistinguishable.
Parts b and c of the figures show the quartiles of the ratio of ISCLT2/ISCST2
results for all receotors, first as a function of convergence level used for the
ISCLT2 model, and second as a function of distance with no limit on the
convergence levels. The ratios shew vary close agreement between -SCLT2
and ISCST2, less than a one percent difference, for receptors located within and
near the area source. The maximum differences increase to about 5 percent for
receptors located further downwind of the area source, but most ratios still fall
within ± 2 percent. The larger differences at downwind receptors are attributed
to the fact that the ISCST2 model results are based on randomly placed wind
directions, rather than on a uniform distribution. For sectors with relatively small
frequencies of occurrence in the STAR data file, the number of hours generated
28
-------
for the ISCST2 input file will be relatively few, and these hours may not
approximate a sector average impact very well.
29
-------
Maximum Cone. Vs. Down Wind Distance
1000x1000m Source, RDU 1987 RANDOM DATA
100
6
D>
0
U
8

i
lOz
0.1-
0.01
r
7 8 9 10
Down Wind Distance (KM)
11
12
13
T
14
15
ISCST2 Simulation
ISCLT2 Simulation
Figure 3.7a. Maximum Concentration Of ISCST Simulation And ISCLT
Simulation Plotted With Downwind Distance. 1000x1000m Area
Source, RDU 1987 Random Hourly And STAR Data
30
-------
Ratio (ISCLT/ISCST) by Converg. Levels
1000X1000m Area Source, Case 3.2.1
a B a a
•«=
~T
3
-r
5
6
T
~T
9
—i r
10 11
—i - i - 1 - 1 - 1 - r
12 13 14 15 16 17
i
18
1 - 1 —
18 20 21
Convergence Levels
-»- Max Ratio

-6~2S%Mark
75% Mark -X- 60% Mark

Mia Ratio
Figure 3.7b. Quartiie Plot of Ratios (ISCLT/ISCS7) by Convergence Levels For

An 1000x1000m Area Source With RDU 1987 Random Hourly
Meteorological Data Set For ISCST Simulation, and the RDU 1987

STAR Data Set For ISCLT Simulation.
31
-------
Ratio (ISCLT/ISCST) Vs. Distance
1000X1000m Area Source, Case 3.2.1
100
I I I I
i n
1000
I I I ITT
10000
I I I I I I
100000
Down Wind Distance (Meters)
Max Ratio-S-75% Mark -X- 50% Mark
25% Mark -^r Min. Ratio
Figure 3.7c. Quartile Plot of Ratios (ISCLT/ISCST) by Downwind Distance For
An 1000x1000m Area Source With RDU 1987 Random Hourly
Meteorological Data Set For ISCST Simulation, and the RDU 1987
STAR Data Set For ISCLT Simulation. No Limit On Convergence
Levels.
32
-------
Maximum Cone. Vs. Down Wind Distance
1000x200m Source, RDU 1987 RANDOM DATA
100
0.01
6 7 8 9 10
Down Wind Distance (KM)
11
12
13
14
15
ISCST2 Simulation
ISCLT2 Simulation
Figure 3.8a. Maximum Concentration Of ISCST Simulation And ISCLT
Simulation Plotted With Downwind Distance. 1000x200m Area
Source, RDU 1987 Random Hourly And STAR Data
33
-------
Ratio (ISCLT/ISCST) by Converg. Levels
1000X200m Area Source, Case 3.2.2
.9-
.8-
.7-
.6-
.5-
.4-
.3-
.2-

2 tt9-
o a8"
<0 tt7-
"5 0.8-
| tt5-
ffi tt4-
0.2-
ai-
o
-G H n a
=8=
«=
8 9 10 11 12 13 14 1B 16 17 18 19 20 21
Convergence Levels
Max. Ratio -S- 75%Mok -X- 50%Marie
25% Mark -A-Mia Rafio
Figure 3.8b. Quartile Plot of Ratios (ISCLT/ISCST) by Convergence Levels For
An 1000x200m Area Source With RDU 1987 Random Hourly
Meteorological Data Set For ISCST Simulation, and the RDU 1987
STAR Data Set For ISCLT Simulation.
34
-------
Ratio (ISCLT/ISCST) vs. Down Wind Dist.
1000X200m Area Source, Case 3.2.2
100
Down Wind Distance (Meters)
Max Ratio
25% Mark
75% Mark -X- 50% Mark
Mm. Ratio
Figure 3.8c. Quartile Plot of Ratios (ISCLT/ISCST) by Downwind Distance For
An 1000x200m Area Source With RDU 1987 Random Hour
Meteorological Data Set For ISCST Simulation, and the RDU 1987
STAR Data Set For ISCLT Simulation. No Limit On Convergence
Levels.
35
-------
3.2.3. Examining The Source Geometry And Rotation Effects

In the first part of the this study, a 100x100m area source and a
400x100m area source are examined using idealized STAR meteorological data.
The purpose of this test is to verify the accuracy and reasonableness of the
algorithm for simple variations of source-receptor geometry. A polar network of
receptors is used. The origin of this polar network is located at the center of the
area source. This polar network uses distance rings of 5, 10, 25, 50, 70, 150,
300, and 500 meters, with 32 receptors in each ring, these receptors are 11.25
degrees apart from each other.

Two separate STAR meteorological data files are used in these
exercises. In the first file, all of the winds are assumed to occur within the 22.5
degree sector centered on 0 degree. In the second file, winds are assumed to
occur only in the 22.5 degree sector centered on 45 degree.

In the second part of the this study, the same receptor network and
meteorological data files are used. The only difference is that the 100x100m
area source is divided into four 50x50m area sources, and the 400x100m area
source is divided into four 100x100m area sources. The results of these runs
are compared with the single source results. The purpose of this experiment is
to assure that the algorithm handles source geometry correctly.

Because the algorithm is also designed to handle rectangular sources
oriented other than north-south (i.e. rotated), several tests have been performed
to examine this capability. The source defined above is rotated for 45.0 degrees
clockwise. The wind directions in the STAR data set are changed accordingly to
show the effects of the rotation.

The various combinations used to test the source geometry and rotation
effects are summarized in Table 3.1.
Table 3.1. The Tests For Source Geometry And Rotation Effects
Case No.
3.3.1.1
3.3.1.2
3.3.1.3
3.3.2.1
3.3.2.2
3.3.2.3
Area
Source
100x1 00m
100x1 00m
1 00x1 00m
400x1 00m
400x1 00m
400x1 00m
Source
(Rotated .
From North
0
0
45
0
0
45
Wind
Speed
Category
2
2
2
2
2
2
Stability
Category
A
A
A
A
A
A
Wind
Direction
ON
45 NE
45 NE
ON
45 NE
45 NE
36
-------
Figures 3.9 - 3.11 depict contour diagrams for the concentration values
for each case with the 100x100m square shaped area source. The first feature
is that the results for north wind are symmetric across the centerline. The
second feature is that the results for the algorithm appear very reasonable for
cases involving rotation of the source and/or wind direction. Figures 3.12-3.14
depict contour diagrams for the concentration values for each case with the
400x100m rectangular shaped area source. The results for this source also
appear to be reasonable.
37
-------
100x100m Source, Rotated 0.0 Deg., North Wind

-500 -400 -300 -200 -100 0 100 200 300 400 500
500 i 1 1 1 , 1 1 1 1 1 1 500
400
300
200
100
-100
-200
-300
-400
-500
400
300
200
100
-100
-200
-300
-400
-500 -400 -300 -200 -100 0 100 200 300 400 500
-500
Figure 3.9. Contour Diagram of Annual Average Rural Concentration
for An 100x100m Area Source With Winds Come Only From 0
Degree North.
38
-------
100x100m Source, Rotated 0.0 Deg., NE Wind
-100
-200
-300
-400
100
-500 -400 -300 -200 -100 0 100 200 300 400 500
-100
-2CO
300
-400
-500
-500
-500 -400 -300 -200 -100 0 100 200 300 400 500
Figure 3.10. Contour Diagram of Annual Average Rural Concentration (jig/m3)
for An 100x100m Area Source With Winds Come Only From 45.0
Degree Northeast.
39
-------
100x100m Source, Rotated 45.0 Deg., NE Wind
-500 -400 -300 -200 -100 0 100 200 300 400 500
500
400
300
200
100
-100
-200
-300
-400
-500
100
-200
-300
-400
-500
-500 -400 -300 -200 -100 0 100 200 300 400 500
Figure 3.11. Contour Diagram of Annual Average Rural Concentration (ng/m3)
for An 100x100m Area Source With 45.0 Degree Rotation and
Winds Come Only From 45.0 Degree Northeast.
40
-------
400x100m Source, Rotated 0.0 Deg., North Wind
-500 -400 -300 -200 -100 0
500
100 200 300 400 500
400
300
200
100
-100
-200
-300
-400
-500
500
400
300
200
100
-100
-200
-300
-400
-500 -400 -300 -200 -100 0 100 200 300 400 500
-500
Figure 3.12. Contour Diagram of Annual Average Rural Concentration
for An 400x100m Area Source With Winds Come Only From 0
Degree North.
41
-------
400x100m Source, Rotated 0.0 Deg., NE Wind
-500 -400 -300 -200 -100 0 100 200 300 400 500
-100 -
-200
-300
-400
-1CO
-200
-300
-400
-500
-500 -400 -300 -200 -100 0 100 200 300 400 500
-500
Figure 3.13. Contour Diagram of Annual Average Rural Concentration
for An 400x100m Area Source With Winds Come Only From 45.0
Degree Northeast.
42
-------
400x100m Source, Rotated 45.0 Deg., NE Wind
-500 -400 -300 -200 -100 0
500
100 200 300 400 500
400
300
200
100
-100
-200
-300
-400
-500
500
400
300
200
100
-100
-200
-300
-400
-500
-500 -400 -300 -200 -100 0 100 200 300 400 500
Figure 3.14. Contour Diagram of Annual Average Rural Concentration (ng/m3)
for An 400x100m Area Source With 45.0 Degree Rotation and
Winds Come Only From 45.0 Degree Northeast.
43
-------
The results for the 10 maximum concentration values of a 100x100m
square shaped area source are listed in Table 3.2(a) -(c). Part A of the table
lists the results for winds coming from North for the whole year. Part B of the
table lists the results for winds coming from 45.0 degree Northeast. Part C of the
table lists the results for the area source rotated clockwise for 45 degrees and
the winds coming from 45.0 degree Northeast. The left hand columns of each
table contain the results generated by a 100x100m single source setup, while
the right hand columns of these tables list results from four 50x50m sources
subdivided from the 100x100m area source.

The results for the 10 maximum concentration values of a 400x100m
square shaped area source are listed in Table 3.3(a) -(c). Part A of the table
lists the results for winds coming from North for the whole year. Part B of the
table lists the results for winds coming from 45.0 degree Northeast. Part C of the
table lists the results for the area source rotated clockwise for 45 degrees and
the winds coming from 45.0 degree Northeast. The left hand columns of each
table contain the results generated by a 400x100m single source setup, while
the right hand columns of these tables list results from four 100x100m sources
subdivided from the 400x100m area source.

The differences between the single source simulation and the simulation
involving subdividing the source are very small, normally less than 0.1 percent.
For instance, Table 3.2a shows highest concentrations of 173.9104 versus
173.9131, or a difference of about 0.002 percent. These minor differences
probably result from truncation errors within the single precision calculation in
the numerical integration.
44
-------
Table 3.2a. 10 Maximum Annual Averages For Case 3.3.1.1
Rank
1
2
3
4
5
6
7
8
9
10
Results Of One
173.910400
172.949300
172.851200
170.573000
170.508100
166.425400
166.407500
162.357200
162.019400
161.908300
1 00x1 00 Area Source
{ .00, -50.00)
( -9.75, -49.04)
( 9.75, -49.04)
( 19.13, -46.19)
(-19.13, -46.19)
( 27.78, -41.57)
(-27.78, -41.57)
( .00, -25.00)
( 4.88, -24.52)
( -4.88, -24.52)
Results Of Four 50x50 Area Sources
173.913100
172.938900
172.708100
170.506600
170.492100
166.376100
166.294900
162.002600
161.688800
161.591700
( .00, -50.00)
J -9.75, -49.04)
( 9.75, -49.04)
( 19.13, -46.19)
(-19.13, -46.19)
(-27.78, -41.57)
( 27.78, -41.57)
( .00, -25.00)
( -4.88, -24.52)
( 4.88, -24.52)
Table 3.2b. 10 Maximum Annual Averages For Case 3.3.1.2
Rank
1
2
3
4
5
6
7
8
9
10
Results Of One
179.183000
173.940900
173.082100
173.013100
170.628900
170.439900
165.939900
165.885700
164.349500
163.958400
100x100 Area Source
( -50.00, -50.00)
( -35.36, -35.36)
(-27.78, -41.57)
(-41.57, -27.78)
(-46.19, -19.13)
(-19.13, -46.19)
(-49.04, -9.75)
| -9.75, -49.04)
(-17.68, -17.68)
(-13.89, -20.79)
Results Of Four
179.172000 (
173.849100 (
173.137500 (
173.011700 |
170.366600
170.257700
165.835600
165.664000
163.815600
163.423600
50x50 Area Sources
-50.00, -50.00)
-35.36, -35.36)
L -41. 57, -27.78)
L-27.78, -41.57)
' -19. 13, -46.19)
' -46. 19, -19.13)
[ -9.75, -49.04)
[-49.04, -9.75)
[-17.68, -17.68)
(-13.89, -20.79)
Table 3.2c. 10 Maximum Annual Averages For Case 3.3.1.3
Rank
1
2
3
Results Of One
173.910100
172.949200
172.851000
4 1 70.572300
5
6
7
8
9
10
170.508000
166.425300
166.407400
162.357100
162.019300
161.908200
100x100 Area Source
( -35.36, -35.36)
(-41.57. -27.78)
(-27.78, -41.57)
( -19.13, -46.19)
(-46.19, -19.13)
( -9.75, -49.04)
(-49.04, -9.75)
(-17.68, -17.68)
(-13.89, -20.79)
(-20.79, -13.89)
Results Of Four
173.912700 (
172.938800 (
172.708000 (
170.506500 {
170.492000 (
166.376000 (
1 66.294800 _1
162.002500 (
161.688700 (
161.591600 (
50x50 Area Sources
-35.36, -35.36)
-41.57, -27.78)
-27.78, -41.57)
-19.13, -46.19)
-46.19, -19.13)
-49.04, -9.75)
-9.75, -49.04)
-17.68, -17.68)
-20.79, -13.89)
-13.89, -20.79)
45
-------
Table 3.3a. 10 Maximum Annual Averages For Case 3.3.2.1
Rank
1
2
3
4
5
6
7
8
9
10
Results Of One 400x1 00 Area Source
43.721260 ( 50.00, -50.00)
43.721240 ( .00, -50.00)
43.720490 ( -50.00, -50.00)
43.614000 ( -9.75, -49.04)
43.613990 ( 9.75, -49.04)
43.290850 ( 19.13, -46.19)
43.290830 ( -19.13, -46.19)
42.799210 ( 27.78, -41.57)
42.799150 (-27.78, -41.57)
42.517910 ( 58.79, -39.28)
Results Of Four 1 00x1 00 Area Sources
43.733450 ( .00, -50.00)
43.703520 ( 50.00, -50.00)
43.703520 ( -50.00, -50.00)
43.634800 ( -9.75, -49.04)
43.510310 ( 9.75, -49.04)
43.255700 ( -19.13, -46.19)
43.210060 ( 19.13, -46.19)
42.696900 ( -27.78, -41.57)
42.687830 ( 27.78, -41.57)
42.447700 ( 58.79, -39.28)
Table 3.3b. 10 Maximum Annual Averages For Case 3.3.2.2
Rank
1
2
3
4
5
6
7
8
9
10
Results Of One 400x100 Area Source
47.265880 ( -50.00, -50.00)
47.210750 ( -9.75, -49.04)
47.099850 ( 9.75, -49.04)
47.077460 ( .00, -50.00)
46.998490 ( -19.13, -46.19)
46.793360 ( 19.13, -46.19)
46.571220 ( 50.00, -50.00)
46.541280 (-27.78, -41.57)
46.348590 ( -58.79, -39.28)
46.272030 ( 27.78, -41.57)
Results Of Four 100x100 Area Sources
47.312110 ( -9.75, -49.04)
47.264860 ( -50.00, -50.00)
47.101670 ( 9.75, -49.04)
47.060090 ( .00, -50.00)
47.007680 (-19.13, -46.19)
46.802980 ( 19.13, -46.19)
46.573250 ( 50.00, -50.00)
46.542540 ( -27.78, -41.57)
46.344420 ( -58.79, -39.28)
46.271140 ( 27.78, -41.57)
Table 3.3c. 10 Maximum Annual Averages For Case 3.3.2.3
Rank
1
2
3
4
5
6
7
8
9
10
Results Of One 400x1 00 Area Source
43.721220 ( .00, -70.71)
43.721210 ( -35.36, -35.36)
43.720460 (-70.71, .00)
43.613970 (-41.57, -27.78)
43.613960 (-27.78, -41.57)
43.290820 ( -19.13, -46.19)
43.290800 (-46.19, -19.13)
42.799180 ( -9.75, -49.04)
42.799130 (-49.04, -9.75)
42.517880 ( 13.79, -69.35)
Results Of Four
43.733390 (
43.703460 (
43.703460 (
43.634770 (
43.510270 (
43.255680 (
43.210020 (
42.696870 (
42.687810 (
42.447680 (
100x100 Area Sources
-35.36. -35.36)
.00, -70.71)
-70.71, .00)
-41.57, -27.78)
-27.78, -41.57)
-46.19, -19.13)
-19.13, -46.19)
-49.04, -9.75)
-9.75, -49.04)
13.79, -69.35)
46
-------
3.2.4. Large Area Source With Actual Meteorological Conditions

Previous tests using idealized meteorological data have verified that the
numerical integration has been correctly implemented within the ISCLT2 model,
and results from the algorithm for various source-receptor geometries appear
reasonable. This section presents results for tests of the algorithm using actual
meteorological conditions. The first tests show comparisons between the
ISCLT2 model and the ISCST2 model, both using the numerical integration
algorithm and appropriate meteorological data. The ISCLT2 results are based
on the RDU 1987 STAR meteorological data and the ISCST2 results are based
on the hourly meteorological data generated by the RAMMET meteorological
preprocessor for the same RDU 1987 data. Both models were used to calculate
annual average concentrations. Figure 3.15 shows the results of the
ISCLT2/ISCST2 comparisons for the 1000x1000m area source. Figure 3.15a
shows the maximum concentration values for both models as a function of
downwind distance. While the results show reasonable close agreement, there
is a trend for ISCLT2 to have smaller maximum concentration values than
ISCST2 as the downwind distance increases. Figure 3.15b shows the quartiles
of the ISCLT2/ISCST2 ratios as a function of ISCLT2 convergence level, and
Figure 3.15c shows the ratios of ISCLT2/ISCST2 by downwind distance. These
plots include ratios for all receptor locations. Figure 3.15b shows that the ratios
do not change significantly beyond convergence level 3, indicating that the
ISCLT2 model results are converging fairly quickly. Most of the ratios are
around 1.1, indicating that the ISCLT2 model predicts concentrations that are
about 10 percent higher than the ISCST2 model for this case. Figure 3.15c
shows that the ratio of about 1.1 is very consistent for receptors located within
and near the area, and that the range of ratios increases with increasing
downwind distance. These results using actual meteorological conditions are
very encouraging, showing that the ISCLT2 and ISCST2 models produce fairly
consistent results using the numerical integration algorithm for the receptors of
most concern located within and near the area. Figures 3.16a to 3.16c show
similar results for the 1000x200m rectangular area source.

As mentioned previously, the differences between the ISCLT2 and
ISCST2 models result from the fact that the two models use different
meteorology data sets. The STAR frequency summaries usea by the 1SCLT2
mode! lose the detail of specific combinations of wind speed, direction, stability
class and mixing height that occur within the hourly meteorological data input to
the ISCST2 model, and these specific combinations which may cause high
hourly concentrations have a significant impact on the annual averages
generated by the ISCST2 model. Comparing these results with those obtained
using the idealized meteorology data confirms this hypothesis.
47
-------
Maximum Cone. Vs. Down Wind Distance
1000x1000m Source, RDU 1987 RAMMET DATA
c
o
—i
4J
«
14
4->
C
O
o

0
u

^
X
«
£
0.1 =
0.01
56789

Down Wind Distance (KM)
11
12
13
14
15
ISCST2 Simulation —f— ISCLT2 Simulation
Figure 3.15a. Maximum Concentration Of ISCST Simulation And ISCLT
Simulation Plotted With Downwind Distance. 1000x1000m Area

Source, RDU 1987 Data.
48
-------
Ratio (ISCLT/ISCST) by Converg. Levels
1000X1000m Area Source, Case 3.1.1
1.9-
1.8-
1.7-
1.6-
1.5-
1.4-
1.3-
1.2-
1.1-
1-
0.9-
2 0.7-
"5 0.6-
5 0.5-
ff 0.4-
0.3-
ai-
o
B S S
-B-
2
3
4
6
7
8
9
10 11 12 13 14 15 16 17 18 19 20 21
Convergence Levels
Max. Ratio -S- 75% Mark -X- 60% Marfc
25% Mark -ak-Mia Ratio
Figure 3.15b.Quarti!e Plot of Ratios (ISCLT/ISCST) by Convergence Levels For
An 1000x1000m Area Source With RDU 1987 RAMMET Hourly
Meteorological Data Set For ISCST Simulation, and the RDU 1987
STAR Data Set For ISCLT Simulation.
49
-------
Ratio (ISCLT/ISCST) Vs, Distance
1000X1000m Area Source, Case 3.1.1
I
40
&
N
i
«
I
1.9-
1.8-
1.7-
1.6-
1.5-
1.4-
1.3-
1.2-
1.1-
1-
0.9-
0.8-
0.7-
ae-
as-
0.4-
as-
Q2-
0.1-
100
II
1000
10000
I till)
100000
Down Wind Distance (Meters)
Max. Ratio
• 25% Mark
75% Mark
Min. Ratio
60% Mark
Figure 3.15c. Quartile Plot of Ratios (ISCLT/ISCST) by Downwind Distance For
An 1000x1000m Area Source With RDU 1987 RAMMET Hourly
Meteorological Data Set For ISCST Simulation, and the RDU 1987
STAR Data Set For ISCLT Simulation. No Limit On Convergence
Levels.
50
-------
Maximum Cone. Vs. Down Wind Distance
1000x200m Source, RDU 1987 RAMMET DATA
100
•n

O
•H
•P
0
14
4J

«
U

e
0.01-
r
6789

Down Wind Distance (KM)
10
11
12
13
14
15
ISCST2 Simulation
ISCLT2 Simulation
Figure 3.16a. Maximum Concentration Of ISCST Simulation And ISCLT
Simulation Plotted With Downwind Distance. 1000x200m Area
Source, RDU 1987 Data
51
-------
Ratio (ISCLT/ISCST) by Converg. Levels
1000X200m Area Source, Case 3.1.2
1.8-

1.7-

| i.e-

tt 1.4-

fe 1.3-

« 1.2-

| 1.1-

2 ttfr-

o aa~

*5 ae-

1 as"

0,3-
ai-

o
x
-s-
,-£3—e=
n&
-r
2
T~
3
-r
4
-&-
S—
-r~
e
-r
7
~r
9
~l 1 1 1 1 1 1 1 1 1 1—
10 11 12 13 14 15 16 17 18 19 20 21
Convergence Levels
Max Ratio -S- 75% Mark -X-50% Mark

25%Mark -A-Min. Ratio
Figure 3.16b.Quartile Plot of Ratios (ISCLT/ISCST) by Convergence Levels For

An 1000x200m Area Source With RDU 1987 RAMMET Hourly

Meteorological Data Set For ISCST Simulation, and the RDU 1987

STAR Data Set For ISCLT Simulation.
52
-------
3.
o
W
I
100
Ratio (ISCLT/ISCST) vs. Down Wind Dist.
1000X200m Area Source, Case 3.1.2
1000
10000
Down Wind Distance (Meters)
Max. Ratio
25% Mark
75% Mart; -X- 60% Mark
Mm. Ratio
Figure 3.16c. Quartile Plot of Ratios (ISCLT/ISCST) by Downwind Distance For
An 1000x200m Area Source With RDU 1987 RAMMET Hour
Meteorological Data Set For ISCST Simulation, and the RDU 1987
STAR Data Set For ISCLT Simulation. No Limit On Convergence
Levels.
53
-------
For the large source subdivision study, the 1000x1000m large source has
been broken into four 500x500m sources. Also, the 1000x200m source has
been broken into five 200x200 sources. The 10 maximum values were
examined. Figures 3.17 and 3.18 shows the results of the comparison of the
single versus subdivided sources using actual STAR data for RDU 1987. They
show the maximum concentration at each of the 7 downwind distances of 250,
500, 750, 1000, 1500, 5000, and 15000 meters. Basically, the maximum
concentration values occur inside and nearby the area source. These results
show that subdividing the area source does not affect the design value predicted
by the model.
54
-------
Maximum Cone. Vs. Down Wind Distance
ISCLT2 Simulation, RDU 1987 RAMMET Data
100
o
••-I
JJ
a
u
41

0
O

0
o

•*
X
•)
0.01
6789

Down Wind Distance (KM)
10
11
On* 1000x1000m Src
Four 500x500m Srcs
Figure 3.17. Maximum Concentration Of ISCLT Simulation With One

1000x1000m Area Source And ISCLT Simulation With This Source

Broken Down Into Four 500x500m Area Sources, RDU 1987 STAR

Data
55
-------
Maximum Cone. Vs. Down Wind Distance
ISCLT2 Simulation, RDU 1987 RAMMET Data
100=
f)
«
«
E

o»
o
~4
4J
a
0)
u

i
10=
0.1 =
0.01
10 11 12 13 14 15
Down Wind Distance (KM)
One 1000x200m Src —•— Five 200x200m Srcs
Figure 3.18. Maximum Concentration Of ISCLT Simulation With One

1000x200m Area Source And ISCLT Simulation With This Source

Broken Down Into Five 200x200m Area Sources, RDU 1987 STAR

Data
56
-------
3.2.5. Convergence Level Consideration

Figure 3.19 shows the ratio formed by dividing the run time of the ISCLT2
model with the numerical integration algorithm by the run time of the ISCST2
model with the same algorithm. These results were obtained using the actual
RDU 1987 meteorological data for both models. The run time increases as the
ISCLT2 algorithm convergence level goes up. After level 5, the run times are
stabilized at a ratio of about 0.4, meaning that the ISCST2 model took about 2.5
times as long to run as the ISCLT2 model for this case. Table 3.4 shows how
the maximum annual concentration value at each downwind distance varies by
convergence level. In these cases, the maximum concentration value stabilizes
after convergence level 2. Overall, the algorithm appears to have converged
completely for this case by level 7.
57
-------
0.5-
RUN TIME FOR DIFFERENT CONV. LEVELS
1000x1000m Source, RDU 1987 RAMMETdata
CT3
D
C
C
o
CO
o
u_
0.4-
0.3-
£-.. 0.2-
_J
O
CO

J
O o.H
u_

Q)
—I ! 1 I I 1 1 1 1 1
10 11 1-2 13 14 15 16 17 18 19 20
Convergence Level
Figure 3.19. Computer Run Time Ratio (ISCLT/ISCST) For Real Life Simulation
58
-------
Table 3.4a.
Maximum Annual Average Concentration Vs. Downwind Distance
1000x1000m Source, RDU 1987 Star Data
DOWNWIND
DISTANCE
250 00000
500.00000
750.00000
1000.00000
1500.00000
5000.00000
15000.00000
CALCUIATION
STOPS IN
LEVEL 2
39.31793
36.33121
6.93167
3.11843
1.59176
0.24298
0.0481 1
CALCULATION
STOPS IN
LEVEL 3
39.31793
3633121
6.91585
3.11107
1 .58582
0.24080
0 04755
CALCULATION
STOPS IN
LEVEL 4
39.31793
36.33121
6.91292
3.10819
1 .58293
0.24076
0.04751
CALCULATION
STOPS IN
LEVEL 5
39.31793
36.33121
6.91339
3.10749
1 .58222
0.24076
0.04749
CALCULATION
STOPS IN
LEVEL7
39.31793
36.33121
6.91265
3.10749
1.58214
0 24076
004744
CALCULATION
STOPS IN
LEVEL 10
39.31793
36.33121
691255
3.10749
1.58214
0.24076
0.04744
UNCONDI-
TIONAL CON-
VERGENCE
39.31793
36.33121
6.91255
3.10749
1.58214
0.24076
0.04744
ISCST RUN
USING
RANDOM DATA
39.27002
36.29247
6.90980
3.10540
1 .58267
0.23798
0.04698
Table 3.4b.
Maximum Annual Average Concentration Vs. Downwind Distance
1000x200m Source, RDU 1987 Star Data
DOWNWIND
DISTANCE
250.00000
500.00000
750.00000
1000.00000
1500.00000
5000.00000
15000.00000
CALCULATION
STOPS i(4
LEVEL 2
28.54518
20.19791
1.20635
0.61886
0.29828
004810
0 0099G
CALCULATION
STOPS IN
LEVELS
28.55116
2019719
1 .20581
0.61754
0 29871
0.04794
0.00950
CALCULATION
STOPS IN
LEVEL 4
2855116
20.19719
1.20513
0.61533
0 29833
0.04793
0 00949
CALCULATION
STOPS IN
LEVEL 5
28.55116
20.19719
1.20513
0.61498
0.29823
0.04793
0 00949
CALCULATION
STOPS IN
LEVEL7
28.55116
20.19719
1.20513
0.61492
0.29820
0.04793
0.00948
CALCULATION
STOPS IN
LEVEL 10
28.55116
20.19719
1.20513
0.61492
0 29820
0 04793
000948
UNCONDI-
TIONAL CON-
VERGENCE
28.55116
2019719
1.20513
0.61492
0.29820
0 04793
000948
ISCST RUN
USING
RANDOM DATA
28.51452
20.18498
1 .20463
0.61700
0.29850
0.04807
0 00977
59
-------
4. THE SENSITIVITY ANALYSIS

4.1. Description Of The Study

The purpose of this study is to evaluate the sensitivity of design
concentrations across a range of source characteristics for the new area source
algorithm that has been incorporated into the ISC2 Long Term (ISCLT2) model
(EPA, 1992). To examine the sensitivity of the design concentrations across a
range of source characteristics, five ground-level area sources were modeled,
with sizes varying from 10 meters to 1,000 meters in width. An elevated source
scenario consisting of a 100-meter wide area with a release height of 10 meters
was modeled. An additional case involving a 1,000 meter wide ground level
area was modeled with receptors located within and nearby the area. The high
annual averages were determined for each of these source scenarios using a
full year of meteorological data. All of the sources were modeled as square
areas oriented N-S and E-W, since the original ISC algorithm was limited to
handling that source geometry. Each scenario was run for one year of STAR
data from Pittsburgh, PA (1989); one year of STAR data from Oklahoma City,
OK (1989); and one year of STAR meteorological data from Seattle, WA (1989).

Each scenario was run with both the rural and urban mode dispersion
options. The only difference between the rural mode and the urban mode that
effects the area sources modeled in this study are the lateral and vertical
dispersion coefficients, sigma-y and sigma-z. The dispersion coefficients are
somewhat larger for the urban mode to account for the increased dispersive
capacity of the atmosphere in the urban environment. The regulatory default
option was used for all scenarios. This includes a procedure for calculating
averages for periods that include calm hours. A pollutant type of "OTHER" was
specified, so that no decay was used for either the rural or the urban mode. For
the sake of efficiency, all computer runs involving the original algorithm were
performed using the ISCLT2 model, rather than the original ISCLT model. In this
way, the same input runstream file was used for both algorithms.
60
-------
A polar receptor network consisting of ground level receptors at five
distances and 36 directions (every 10 degrees) was used to determine design
concentrations. Since most area sources are ground-level or low-level releases,
the maximum impacts can be expected to occur very near the source. However,
the virtual point source algorithm does not allow receptors within the area itself,
and is known to provide unreasonable concentration estimates very close to the
source. The guidance in the ISC2 User's Guide states that if the source-
receptor distance is less than the width of the area, then the area should be
subdivided and modeled as multiple sources. Therefore, the first distance ring
in the polar network was placed at a downwind distance (measured from the
center of the area) of 1.5*XINIT meters, where XINIT is the width of the area.
This places the nearest receptors at a distance of about one source width from
the edge of the area. Additional distance rings were placed at approximately
2.0, 3.0, 5.0 and 10.0 times the initial distance, for a total of 180 receptors. For
the ground level sources, the maximum ground level concentrations are
expected to occur near the downwind edge of the area, and to decrease beyond
that distance. Therefore the maximum concentrations for these source-receptor
geometries are expected to occur at the 1.5*XINIT distance. The concentrations
at the larger receptor distances were also examined for a few cases in order to
compare the algorithms for distances downwind of the maximum concentration.

Additional receptor distances were used for the elevated source to
account for the fact that the maximum impact may occur beyond the nearest
distance ring. Additional receptor rings were included at distances of 2.0*XINIT,
2.5*XINIT, and 4.0*XINIT for the elevated release height cases to better
represent the peak concentration from the refined model.

In order to assess the sensitivity of the design values for receptors
located close to and within an area source, an additional scenario was modeled
involving a 1,000 meter wide (extra large) ground-level area source with
receptors located within the area and near the edge of the area. For the original
virtual point source algorithm, this source was subdivided into 4, 16, 64 and 100
separate areas of equal size. This was necessary because the virtual point
source algorithm cannot model impacts at receptor locations within the area
being modeled.

An emission rate equivalent to 1.0 g/s for the entire area was used for all
scenarios. The area source widths, heights of release, emission rates, and
receptor distances are shown in Table 4.1 for each scenario. Table 4.2 provides
the source inputs for the X-Large (XL), Close-in case for the 4-, 16-, 64-, and
100-source treatment used with the virtual point source algorithm. Figure 4.1
shows the location of the receptors used for the X-Large source with receptors
located within and nearby the area.
61
-------
Table 4.1.
Area Source Scenarios for Sensitivity Analysis
Source Type
X-Small, GL*
Small, GL*
Medium, GL*
Large, GL*
X-Large, GL*
Medium, EL**
X-Large, Cl***, GL
Width of
Area
(m)
10.0
50.0
100.0
500.0
1000.0
100.0
1000.0
Height of
Release
(m)
0.0
0.0
0.0
0.0
0.0
10.0
0.0
Emission
Rate
(g/(sm2))
1.0E-2
4.0E-4
1.0E-4
4.0E-6
1.0E-6
1.0E-4
1.0E-6
Receptor Distances (m)
(measured from the center of the area)
15,30,50,75, 150
75, 150,250,400,750
150, 300, 500, 750, 1500
750, 1500, 2500, 4000, 7500
1500, 3000, 5000, 7500,15000
150, 200, 250, 300, 400,500, 750,1500
250, 500,750, 1000, 1500
*GL means Ground-Level.
**EL means Elevated
*** Cl means Close-in
Table 4.2.
Area Source Inputs for X-Large, Close-in Scenario
(used for the original virtual point source algorithm only)

Scenario Description

XL, Close-in 4-sources, (2x2)
XL, Close-in 16-sources, (4x4)
XL, Close-in 64-sources, (8x8)
XL, Close-in 100-souices, (10x10)
Width of
Each Sub-
Division
(m)
500.0
250.0
125.0
100.0

Height of
Release
(m)
0.0
0.0
0.0
0.0

Emission
Rate
(g/(sm2))
1.0E-6
1.0E-6
1.0E-6
1.0E-6
Receptor Distances
(measured from the center of
the 1 000m area)
(m)
250,500,750, 1000, 1500
250,500,750, 1000, 1500
250,500,750, 1000, 1500
250,500,750, 1000, 1500
62
-------
4.2. Results Of The Study

The results of the sensitivity study are presented first for the five ground
level sources with receptors located downwind of the area, followed by the
results for the elevated source, and then for the ground level source with
receptors located within the area.

4.2.1. Ground Level Sources With Downwind Receptors

Tables 4.3 through 4.7 present comparisons of design values (10 highest
annual average values) obtained from the numerical integration algorithm in
ISCLT2 with values from the original virtual point source algorithm for the five
ground level sources of various widths. The source widths range from the very
small (10 meter wide) area source in Table 4.3 to the very large (1000 meter
wide) area source in Table 4.7.

Part A of each table presents the results using rural dispersion
coefficients, and part B for each table presents the results using urban
dispersion coefficients. The design values are generally located at the receptors
located closest to the area source.
63
-------
Example Plot Showing Location of Receptors
-1500.00
1500.00
-100000 -500.00
000
1000.00
500.00
0.00
-50000
-1000.00
-1500
1—I—1 1 1 I 1
1*1 I I 1*1 I I I I
150000
150000
1000.00
500.00
0 00
-500.00
-1000.00
-1500 00
-150000 -1000.00 -50000
0.00
500.00
1000.00
1500.00
Figure 4.1. Example Contour Plot Showing Location of Receptors Relative to
the 1000 Meter Wide Ground Level Source for the X-Large Close-
in Case
64
-------
Table 4.3.
Comparison of Design Concentrations
(10m Width)
for the Very Small Source
Site, Data & Dispersion
Options
Pittsburgh 1989, Rural
Okla. City 1989, Rural
Seattle 1989, Rural
Pittsburgh 1 989, Urban
Okla. City 1989, Urban
Seattle 1989, Urban
Numerical
Integration (New)
5469.19800
7969.95200
9681.49900
2412.35600
3668.16200
4198.99700
Virtual Point
Source (Old)
3137.21600
4032.40300
4673.74800
1305.56500
1754.99500
1948.98300
Ratio
(New/Old)
1.7433285
1.9764771
2.0714636
1.8477487
2.0901268
2.0307770
Table 4.4.
Comparison of Design Concentrations
(50m Width)
for the Small Source
Site, Data & Dispersion
Options
Pittsburgh 1989, Rural
Okla. City 1989, Rural
Seattle 1989, Rural
Pittsburgh 1989, Urban
Okla. City 1989, Urban
Seattle 1 989, Urban
Numerical
Integration (New)
280.28900
410.25400
491.23020
98.87118
150.94320
171.52150
Virtual Point
Source (Old)
160.42750
206.99070
236.76410
54.23226
73.34540
80.59711
Ratio
(New/Old)
1.7413810
1.9819924
2.0747664
1.8231064
2.0579777
2.1281346
Table 4.5.
Comparison of Design Concentrations (|j.g/m3) for the Medium Source
(100m Width)
Site, Data & Dispersion
Options
Pittsburgh 1989, Rural
Okla. City 1989, Rural
Seattle 1989, Rural
Pittsburgh 1989, Urban
Okla. City 1989, Urban
Seattle 1989, Urban
Numerical
Integration (New)
78.06039
114.53200
136.19110
25.43472
39.00346
43.95693
Virtual Point
Source (Old)
44.75477
57.90853
65.75126
14.15101
19.26075
20.92982
Ratio !
(New/Old)
1.7441803
1.9778088
2.0713078
1 .7973784
2.0250229
2.1002058
65
-------
Table 4.6.
Comparison of Design Concentrations (|ig/m3) for the Large Source
(500m Width)
Site, Data & Dispersion
Options
Pittsburgh 1989, Rural
Okla. City 1989, Rural
Seattle 1989, Rural
Pittsburgh 1989, Urban
Okla. City 1989, Urban
Seattle 1989, Urban
Numerical
Integration (New)
4.212339
6.231626
7.280537
1.217577
1.909762
2.060928
Virtual Point
Source (Old)
2.552411
3.329695
3.730223
0.721960
1.010385
1.044501
Ratio
(New/Old)
1.6503373
1.8715306
1.9517699
1 .6864882
1.8901330
1.9731221
Table 4.7
Comparison of Design Concentrations
(1000m Width)
for the Very Large Source
Site, Data & Dispersion
Options
Pittsburgh 1989, Rural
Okla. City 1989, Rural
Seattle 1989, Rural
Pittsburgh 1989, Urban
Okla. City 1989, Urban
Seattle 1989, Urban
Numerical
Integration (New)
1.290841
1.911188
2.222102
0.355397
0.566046
0.593821
Virtual Point
• Source (Old)
0.821113
1.071306
1.198315
0.218373
0.309991
0.312113
Ratio
(New/Old)
1.5720626
1.7839796
1.8543555
1.6274768
1.8260079
1.9025834
66
-------
Overall, the new numerical integration algorithm predicts higher design
concentrations than the original virtual point source algorithm. The average
ratio of the numerical integration algorithm results over the virtual point source
algorithm results (averaged over all three cities and for all averaging periods)
ranges from about 2.0 for the 10 meter wide area to about 1.7 for the 1000 meter
wide area. This trend toward smaller ratios for larger areas is illustrated in
Figure 4.2, which shows the average ratios (averaged across the three
meteorological data locations) for the five ground-level sources for downwind
receptors only, for both rural and urban dispersion. The ratios are generally
larger for the cases with urban dispersion coefficients than for the cases with
rural dispersion coefficients.

The most notable feature about these results is that the numerical
integration method produces larger concentration estimates than the original
virtual point source algorithm. One possible explanation for part of this
difference is that the numerical integration algorithm produces larger off-
centerline concentration values than the virtual point source algorithm. This is
because the numerical integration algorithm provides a more realistic
assessment of the lateral distribution of concentration by integrating over the
entire source area. The lateral distribution of concentration from the virtual point
source algorithm is essentially the distribution of a point source, although some
initial lateral spread of the plume is incorporated to account for the area source.
•
In addition to examining the design values, which all occurred at
receptors located on the nearest distance ring for the ground level sources, the
results at distances located further downwind were examined briefly to determine
whether or not the results converge with distance. Figures 4.3 to 4.8 present the
high annual average concentration values versus distance downwind for the 10
meter wide ground level area source for Pittsburgh, Oklahoma City and Seattle
data for the cases with both rural and urban dispersion coefficients. Results
show that, in all the cases, the maximum concentration values decrease as
expected with increasing downwind distance. The figures also show that the
results for the two algorithms converge for large downwind distances (beyond
about 5 to 7 source widths).
67
-------
2.2
1.2+
AVERAGE RATIO BY AREA SIZE
1989 STAR Data of Pitts., Seattle, OKC
100 200 300 400 500 600 700 800

Width Of The Area Source (Meters)
900
1000
1100
Rural Option
Urban Option
Figure 4.2. Average Ratios (New/Old) by Area Size for Ground Level Sources
For Both Rural And Urban Dispersion Coefficients
68
-------
10000-
ro
*
9000-
8000-i
7000-t
o
•*
4J
«
14
4>

I
K
«
6000-
5000-
4000-
3000-
2000H
1000H
0-t-
0
Maximum Cone. Vs. Down Wind Distance
10x10m Source, Pittsburgh, Rural
10 20 30 40 50 60 70 80 90 100
Down Wind Distance ( M)
no
120 130 140 150 160
VPS Algorithm —*— NIA Algorithm
Figure 4.3. High Annual Average Values Versus Distance for the 10 Meter
Wide Ground Level Source for Rural Dispersion and Pittsburgh
1989 STAR Data
69
-------
10000
Maximum Cone. Vs. Down Wind Distance
10x10m Source, Pittsburgh, Urban
10 20 30 40 50 60 70 80 90 100 110 120 130 140 150 160

Down Wind Distance ( M)
VPS Algorithm —+— NIA Algorithm
Figure 4.4. High Annual Average Values Versus Distance for the 10 Meter
Wide Ground Level Source for Urban Dispersion and Pittsburgh
1989 STAR Data
70
-------
10000-
Maximum Cone. Vs. Down Wind Distance
10x1 Om Source, Oklahoma City, Rural
10
20
30
4O
50
r
60 70 80 90 100 110 120 130

Down Wind Distance ( M)
14O 150 16O
VPS Algorithm —«— NIA.Algorithm
Figure 4.5. High Annual Average Values Versus Distance for the 10 Meter
Wide Ground Level Source for Rural Dispersion and Oklahoma
City 1989 STAR Data
71
-------
10000
Maximum Cone. Vs. Down Wind Distance
10x1 Om Source, Oklahoma City, Urban
10 20 30 40 50 60 70 80 90 100 110 120 130 140 150 160
Down Wind Distance ( M)
VPS Algorithm -+— NIA Algorithm
Figure 4.6. High Annual Average Values Versus Distance for the 10 Meter
Wide Ground Level Source for Urban Dispersion and Oklahoma
City 1989 STAR Data
72
-------
10000-
n
«
O
9000-
8000-1
,5 7000-
6000-(
4J 5000-

0
C 4OOO-
0
3000-
Maximum Cone. Vs. Down Wind Distance
10x1 Om Source, Seattle, Rural
1020304O 50 60 70 80
100 110 120 130 14O 1 SO 16O
Down Wind Distance ( M)
VPS Algorithm —H- NIA Algorithm
Figure 4.7. High Annual Average Values Versus Distance for the 10 Meter
Wide Ground Level Source for Rural Dispersion and Seattle 1989
STAR Data
73
-------
10000
Maximum Cone. Vs. Down Wind Distance
10x1 Om Source, Seattle, Urban
30 40
50
60 70 80 90 100 110
Down Wind Distance ( M)
120 130 14O 150 160
VPS AJgonthm
NIA Algorithm
Figure 4.8. High Annual Average Values Versus Distance for the 10 Meter
Wide Ground Level Source for Urban Dispersion and Seattle 1989
STAR Data
74
-------
4.2.2. Elevated Area Source

Table 4.8 presents comparisons of design values obtained from the
numerical integration algorithm and the virtual point source algorithm for the 100
meter wide elevated source (10 meter release height). Results are presented for
both rural and urban dispersion coefficients. The ratios for the elevated source
are smaller than the corresponding ratios for the 100-meter ground level source
(see Tables 4.5.). In fact, the ratios for the rural dispersion case are less than
1.0, indicating that the numerical integration algorithm estimates smaller
concentrations than the virtual point source algorithm for these cases. Urban
ratios are somewhat larger than rural ratios, but are still smaller than the
corresponding ratios for the ground-level source.

One possible explanation for these results for the elevated source is that
the virtual point source algorithm uses a virtual distance equal to the width of the
source for calculating the vertical dispersion parameters. This means that the
sigma-z value is growing from the upwind edge of the area source. Since the
numerical integration algorithm integrates over the entire area, the vertical
dispersion parameter for each element of the integration is representative of the
actual distance from that element of the area to the receptor location. For the
portion of the area that is closest to the receptor, and therefore having the
greatest impact on the receptor, the sigma-z value will be based on the distance
from the downwind edge of the area to the receptor location. This difference will
result in a smaller overall "effective" vertical dispersion parameter for the
numerical integration algorithm than for the virtual point source algorithm. Since
this is an elevated source with ground-level receptors, the smaller "effective"
vertical dispersion parameter for the numerical integration algorithm will tend to
cause smaller ground-level concentrations, as seen for the case of rural
dispersion. This tendency competes with the tendency for the numerical
integration algorithm to estimate higher concentrations due to differences in the
treatment of lateral dispersion.
Table 4.8.
Comparison of Design Concentrations dig/m^) for the Medium Elevated Source
(100m Width)
Site, Data & Dispersion
Options
Pittsburgh 1989, Rural
Okla. City 1989, Rural
Seattle 1989, Rural
Pittsburgh 1989, Urban
Okla. City 1989, Urban
Seattle 1989, Urban
Numerical
integration (New)
11.94122
19.64324
28.44486
16.22476
27.32439
33.55091
Virtual Point
Source (Cld)
15.29294
21.67913
29.69580
10.98232
16.86372
18.89518
Ratio
(New/Old)
0.7808322
0.9060898
0.9578748
1 .4773527
1.6203062
1.7756333
75
-------
4.2.3. Ground-level Sources With Receptors Within and Nearby the Area
Source

Table 4.9 presents comparisons of design values from the numerical
integration algorithm and from the virtual point source algorithm for the 1000
meter wide ground level source with receptors located within and nearby the
area source. Results are presented for both rural and urban dispersion
coefficients. The results for the virtual point source algorithm are presented for
each of the subdivided multiple-source scenarios examined using 4, 16, 64 and
100 areas of equal size. The ratios for the cases with receptors within and
nearby the area are generally larger than the corresponding ratios for the other
ground level cases (see Tables 4.1 through 4.7). As with the other sources
examined, the ratios are larger for the case with urban dispersion coefficients
than for the case with rural dispersion coefficients. The average ratio for the
rural cases is about 3.0, and the average ratio for the urban cases is about 4.0
The results in Tables 4.9 also show that, in general, the design values for the old
virtual point source algorithm tend to increase as the number of subdivided
areas increases. Since the impact at any receptor located within the area does
not include any contribution from the subarea in which the receptor is located, as
the number of subareas increases and the size of the subarea decreases, the
amount of contribution not accounted for will tend to decrease. In principal, as
the number of subareas approaches infinity and the individual subareas
approach point sources, the two algorithms should eventually converge.

The receptor locations for the design values are also included in Table
4.9 for the numerical integration algorithm and for the virtual point source case
based on 100 subdivided area sources. The locations are given as direction (in
degrees) and distance (in meters). Thus, a location of ( 90,500) means a
receptor located along the 90 degree direction radial, measured clockwise from
North, at a distance of 500 meters from the center of the area. The receptor
locations show good agreement. A more complete picture is provided in Figures
4.9 through 4.21, which display contour plots of high concentrations across the
receptor grid for the numerical integration algorithm (NIA) and for the virtual
point source (VPS) algorithm based on 100 sources. The rural results are
oresented first, followed by the urban results, with the numerical integration
algorithm results and virtual point source (100-source) results for the same
location and averaging period on T'acing pages to facilitate comparison. The ;cur
grid squares located at the center of the diagrams (between X = -500 to 500 and
Y = -500 to 500) define the location of the 1000 meter wide area source. The
source location and the distribution of receptor points was shown in Figure 4.1 in
Section 4.2.
76
-------
Generally, the contour plots show similar patterns between the two
algorithms, although the magnitude of the results is higher for the numerical
integration algorithm than the virtual point source. Some of the contour plots
exhibit isolated peaks and valleys, and some discontinuities (or "kinks") in the
contours. These anomalies are due to the limited number of data points (180)
on which the plots are based, and are an artifact of the interpolation and
contouring schemes used to generate the plots. Therefore, the fine-scale details
should not be given much credence in these plots, although the overall patterns
are fairly reliable. Both algorithms show generally reasonable patterns, with the
contours showing roughly the square shape of the area across the area source
itself. The numerical integration algorithm shows a steeper gradient in the
concentration distribution near the edges of the area than the virtual point
source algorithm; This trend goes along with the tendency for the numerical
integration algorithm to estimate higher concentrations within the area and to
converge toward the virtual point source estimate within several source widths
downwind of the area. The overall conclusion from these contour diagrams is
that the numerical integration algorithm provides a reasonable distribution of
concentrations for receptors located within and nearby the area source.
77
-------
Table 4.9.
Comparison of Annual Average Concentrations (ng/m3) for the 1000m Wide Area
With Receptors Located Within and Nearby the Area
Site, Data& Dispersion
Option
Pittsburgh 1989, Rural
Okla. City 1989, Rural
Seattle 1989, Rural
Pittsburgh 1989, Urban
Okla. City 1989, Urban
Seattle 1989, Urban
Numerical
Integration
(New)
34.19587(20,250)
26.40565 (330,250)
33.36413 (50,250)
14.10248(30,250)
11.19658(320,250)
13.86200(50,250)
Virtual Point
Source (Old)
4 Sources
2.52315
3.13874
3.49210
0.73693
0.96177
0.98152
Virtual Point
Source (Old)
16 Sources
6.02710
4.96530
5.99353
1.80160
1.55283
1.80063
Virtual Point
Source (Old)
64 Sources
10.01933
8.38680
*
10.09934
3.08403
2.69919
3.13583
Virtual Point Source
(Old)
100 Sources
11.18408(10,250)
9.21463 (330,250)
11.08371 (60,250)
3.47830 (10,250)
2.98885 (330,250)
3.48208 (60,250)
Ratio
New/Old-100
3.0575488
2.8656218
3.0101951
4.0544151
3.7461202
3.9809562
Note: Values in parentheses are receptor locations given as direction (degrees from North) and downwind distance (meters).
78
-------
-1500
1500
Annual Averages, Pittsburgh, Rural, NIA
-1000 -500 0 500 1000
1000
500
-500
-1000
-1500
1500
1500
1000
500
-500
-1000
-1500
-1000
-1500
-500
500
1000
1500
Figure 4.9. Contour Diagram of Annual Average Rural Concentrations
from the Numerical Integration Algorithm for the 1000 Meter Wide
Ground Level Source with Close-in Receptors Using Pittsburgh
1989 STAR Data.
79
-------
Annual Averages, Pittsburgh, Rural, VPS
-500
-1000
-1500
1500
1500
1000
- -500
-1000
-1500 -1000 -500
-1500
500 1000 1500
Figure 4.10. Contour Diagram of Annual Average Rural Concentrations
from the Virtual Point Source Algorithm for the 1000 Meter Wide
Ground Level Source with Close-in Receptors Using Pittsburgh
1989 STAR Data.
80
-------
Annual Averages, Oklahoma City, Rural, NIA
-1000 -500 0 500 1000
-1500
1500
1000
500
-500
-1000
-1500
1500
1500
1000
500
-500
-1000
-1500
-1500
-1000
-500
500
1000
1500
Figure 4.11. Contour Diagram of Annual Average Rural Concentrations (|ig/m~)
from the Numerical Integration Algorithm for the 1000 Meter Wide
Ground Level Source with Close-in Receptors Using Oklahoma
City 1989 STAR Data.
81
-------
Annual Averages, Oklahoma City, Rural, VPS
-500 0 500 1000
-1500 -1000
1500
1000
500
-500
-1000
-1500
1500
1500
1000
500
-500
-1000
-1500 -1000 -500
500 1000
-1500
1500
Figure 4.12. Contour Diagram of Annual Average Rural Concentrations
from the Virtual Point Source Algorithm for the 1000 Meter Wide
Ground Level Source with Close-in Receptors Using Oklahoma
City 1989 STAR Data.
82
-------
-1500
1500
Annual Averages, Seattle, Rural, NIA

-1000 -500 0 500 1000
1000
500
-500
-1000
-1500
-1500
-1000
1500
1500
100C
500
-500
-100C
-500
500
1000
-1500
1500
Figure 4.13. Contour Diagram of Annual Average Rural Concentrations (ng/m3)
from the Numerical Integration Algorithm for the 1000 Meter Wide
Ground Level Source with Close-in Receptors Using Seattle 1989
STAR Data.
83
-------
Annual Averages, Seattle, Rural, VPS
-1500
1500
1000
500
-500
-1000
-1500
-1000 -500
500
1000
1500
1500
1000
500
-500
-1000
-1500
-1500 -1000 -500
500
1000
1500
Figure 4.14. Contour Diagram of Annual Average Rural Concentrations (ng/m3)
from the Virtual Point Source Algorithm for the 1000 Meter Wide
Ground Level Source with Close-in Receptors Using Seattle 1989
STAR Data.
84
-------
Annual Averages, Pittsburgh, Urban, NIA
0 500 1000
-500
-1000
-1500
1500
1500
1000
500
-500
-1000
-1500 -1000 -500
-1500
500 1000 1500
Figure 4.15. Contour Diagram of Annual Average Urban Concentrations(ng/m3)
from the Numerical Integration Algorithm for the 1000 Meter Wide
Ground Level Source with Close-in Receptors Using Pittsburgh
1989 STAR Data.
85
-------
Annual Averages, Pittsburgh, Urban, VPS
-500
-1000
-1500
-500
-1000
-1500
-1500 -1000 -500
500 1000
1500
Figure 4.16. Contour Diagram of Annual Average Urban Concentrations(|ig/m3)
from the Virtual Point Source Algorithm for the 1000 Meter Wide
Ground Level Source with Close-in Receptors Using Pittsburgh
1989 STAR Data.
86
-------
Annual Averages, Oklahoma City, Urban, NIA

-1000 -500 0 500 1000
-1500
1500
1000
500
-500
-1000
-1500
-1500
-1000
1500
1500
1000
500
-50Q
-1000
-500
500
1000
-1500
1500
Figure 4.17. Contour Diagram of Annual Average Urban Concentrations(|ig/m3)
from the Numerical Integration Algorithm for the 1000 Meter Wide
Ground Level Source with Close-in Receptors Using Oklahoma
City 1989 STAR Data.
87
-------
Annual Averages, Oklahoma City, Urban, VPS
-1
1500
1000
500
0
-500
-1000
- 1 500
500 -1000

-500 0 500 1000 15

(
_»
V

-1500 -1000

/"
\
\
^^~^—-
\_

"\
}
J
• — 1

\
)

00
1500
1000
500
0
DUU
— 1 UUU
— — 1 DUU
-500 0 500 1000 1500
Figure 4.18. Contour Diagram of Annual Average Urban Concentrations(fig/m3)
from the Virtual Point Source Algorithm for the 1000 Meter Wide
Ground Level Source with Close-in Receptors Using Oklahoma
City 1989 STAR Data.
88
-------
Annua Averages, Seattle, Urban, NIA
-500 0 500 1000
-1500 -1000
1500
1000
500
-500
-1000
-1500
1500
1500
1000
500
-500
-1000
-1500 -1000 -500
-1500
500 1000
1500
Figure 4.19. Contour Diagram of Annual Average Urban Concentrations(|ig/m3)
from the Numerical Integration Algorithm for the 1000 Meter Wide
Ground Level Source with Close-in Receptors Using Seattle 1989
STAR Data.
89
-------
Annual Averages, Seattle, Urban, VPS
-1500 -1000 -500 0 500
1500
1000
500

-500

1000
1500

/
/
/

\
\
\
\
\

\~~
^
/
/ x-
l
o

\
V

^ 7 v.

^
~^\
\
-3' \
}
\ ) 1
/
/
2 —

/
^-^

1000 1500

/
/
»^
/

-1500 -1000 -500 0 500

I 3UU
1000
500

-500

— 1 OJO
— • ' — i DUU
1000 1500
Figure 4.20. Contour Diagram of Annual Average Urban Concentrations(|ig/rn3)
from the Virtual Point Source Algorithm for the 1000 Meter Wide
Ground Level Source with Close-in Receptors Using Seattle 1989
STAR Data.
90
-------
5. CONCLUSION

This report documents the development, testing and evaluation of a new
numerical integration algorithm for modeling area sources for the ISCLT2 model.
This algorithm, which is based on the numerical integration algorithm for area
sources recently implemented in the ISCST2 model, allows users to handle the
complex geometry of irregularly shaped area sources, and allows the calculation
of the area source impact for receptors located within and nearby the area
source. Detailed performance tests, statistical analyses and sensitivity analyses
have been completed to assure the reliability and reasonableness of the
modeling results. The algorithm has been compared with the currently used
ISCLT2 virtual point "source algorithm, as well as with the numerical integration
area source algorithm for the ISCST2 model.

The results show that the new numerical integration ISCLT2 area source
algorithm performs very well. Using idealized meteorological conditions, the
new algorithm achieves very good comparison results when compared with the
newly developed ISCST2 area source algorithm. For realistic meteorological
data, the discrepancies between the prediction of this new algorithm and the
prediction of its ISC2 short term counterpart are within about 10 percent for a
typical area source. The differences between the long term and short term
. algorithms using actual meteorological data are because the ISC2 long term
model uses STAR meteorological frequency distribution data, which is only a
statistical summary of the hourly data, and does not contain the precise
information on specific combinations of wind speed, wind direction, stability class
and mixing height that typically control the design values for the short term
model.

It is also concluded that the currently used ISCLT2 area source algorithm
based on the virtual point source approach underestimates the concentration
value by a factor of about 2 to 4, especially when the receptors are located
inside or near the source. This is due mainly to the fact that the virtual point
source algorithm does not properly treat the lateral distribution of the area
source plume close to the source, and also because it cannot calculate the air
quality impact inside the source unless the area is subdivided into very small
sources.
91
-------
An evaluation of convergence criteria for the algorithm shows that the
algorithm normally converges for all combinations in the STAR summary by level
7, corresponding to a maximum of about 129 simulations per wind direction
sector. It is concluded that for routine applications, computations out to level 10
(1025 simulations per sector) will provide acceptable accuracy without
significantly compromising the run time performance of the algorithm. This
convergence criterion based on a limit on the number of computations per
sector, together with the 2 percent error tolerance check and the lower threshold
cutoff of 1.0E-10, is recommended for use in the ISCLT2 model to obtain
optimum overall performance of the new area source algorithm.
92
-------
6. REFERENCES

Environmental Protection Agency, 1989. Review and Evaluation of Area Source
Dispersion Algorithms for Emission Sources at Superfund Sites. EPA-
450/4-89-020. U.S. Environmental Protection Agency, Research Triangle
Park, North Carolina.

Environmental Protection Agency, 1992. User's Guide for the Industrial Source
Complex (ISC2) Dispersion Models. EPA-450/4-92-008. U.S.
Environmental Protection Agency, Research Triangle Park, North
Carolina.

McGill, Robert, John W. Tukey, and Wayne A. Larsen, 1978. 'Variations of Box
Plots", The American Statistician 32:12-16.

Press, W.B. Flannery, S. Teukolsky, and W. Vetterling, 1986. Numerical
Recipes, Cambridge University Press, New York.

Turner, D.B., 1970. Workbook of Atmospheric Dispersion Estimates, Revised,
Sixth Printing, 1973. EPA Office of Air Programs Publication No. AP-26.
U.S. Environmental Protection Agency, U.S. Government Printing Office,
Washington, D.C.
93
-------
1. .REPORT NO. 2.
EPA-454/R-92-016
4. TITLE AND SUBTITLE
Development and Evaluation of a Revised Area Source
Algorithm for the ISC2 Long Term Model
7. AUTHOR(S)
9. PERFORMING ORGANIZATION NAME AND ADDRESS
"-' • Pacific Environmental Services *
5001 South Miami Boulevard
Post Office Box 12077
Research Triangle Park, NC 27709-2077
12. SPONSORING AGENCY NAME AND ADDRESS
U.S. Environmental Protection Agency
Office of Air Quality Planning and Standards
Technical Support Division
Research Triangle Park, NC 27711
3.
5.
6.
8.
RECIPIENT'S ACCESSION NO.
REPORT DATE
October 1992
PERFORMING ORGANIZATION CODE

PERFORMING ORGANIZATION REPORT NO.
10. PROGRAM ELEMENT NO.
11
13
14
. CONTRACT/GRANT NO. WA No
EPA Contract No. 68
. 1-131
D00124
. TYPE OF REPORT AND PERIOD COVERED
Final Report
.'SPONSORING AGENCY CODE

TECHNICAL REPORT DATA
(Please read Instructions on reverse before completing)
15. SUPPLEMENTARY NOTES
EPA Work Assignment Manager; Jawad S. Touma
16. ABSTRACT

This report includes information on an improved algorithm for modeling dispersion
from area sources, which has been developed based on a numerical integration of the
point source concentration function. A sensitivity analysis is presented of the
algorithm as implemented in the long-term version of the Industrial Source Complex
(ISC2) model. For the performance tests, the new ISCLT2 numerical integration area
source algorithm is challenged in various ways. First, quality assurance tests are
conducted to examine the reasonableness of the results and the efficiency of the
algorithm. These quality assurance tests include printing out the intermediate
calculation results to perform a line-by-line check of the computer code of the new
algorithm. Second, cases with simple area source characteristics (square arrea source
or rectangular area source) and idealized meteorological conditions are used to examine
the reliability and accuracy of the algorithm. Third, several tests are conducted to
show the concentration distribution for various area source shapes. Fourth, tests are
conducted to examine the effects of subdividing the area source and the effects of
rotation of the area source on the simulated concentration values. Finally, several
cases are examined using realistic meteorological conditions. This report is being
released to establish a basis for reviews of the capabilities of this methodology and
of the consequences resulting from use of this methodology in routine dispersion
modeling of air pollutant impacts.
17.
KEY WORDS AND DOCUMENT ANALYSIS
a. DESCRIPTORS
Air Pollution
Toxic Air Pollutants
Air Quality Dispersion Models
18. DISTRIBUTION STATEMENT
Release Unlimited
b. IDENTIFIERS/OPEN ENDED TERMS
Dispersion Modeling
Meteorology
Air Pollution Control
19. SECURITY CLASS (Report)
Unclassified
20. SECURITY CLASS (Page)
Unclassified
C. COSATI
Field/Group

21. NO. OF PAGES
22. PRICE
EPA Fora 2220-1 (Rev. 4-77)
PREVIOUS EDITION IS OBSOLETE
-------