United States
Environmental Protection
Agency
Office of Air Quality
Planning and Standards
Research Triangle Park, NC 27711
EPA-454/R-98-009
June 1998
AIR
& EPA
A COMPARISON OF CALPUFF
MODELING RESULTS TO TWO TRACER FIELD
EXPERIMENTS
-------
EPA-454/R-98-009
A COMPARISON OF CALPUFF MODELING RESULTS TO
TWO TRACER FIELD EXPERIMENTS
U.S. Environmental Protection Agency
Office of Air Quality Planning and Standards
Emissions, Monitoring, and Analysis Division
Research Triangle Park, NC 27711
-------
-------
NOTICE
This information in this document has been reviewed in its entirety by the U.S.
Environmental Protection Agency (EPA), and approved for publication as an EPA document.
Mention of trade names or services does not convey, and should not be interpreted as conveying,
official EPA approval, endorsement, or recommendation.
-------
ACKNOWLEDGMENTS
The study was performed by James O. Paumier and Roger W. Erode of Pacific
Environmental Services, Inc. (PES), Research Triangle Park, North Carolina. This effort was
funded by the U. S. Environmental Protection Agency under Contract No. 68D30032, with
Dennis Atkinson as Work Assignment Manager. The authors wish to thank John Irwin of the
Office of Air Quality Planning and Standards for helpful discussions regarding the model
simulations and analysis of the results. In addition, this document was reviewed and commented
on by John Irwin (EPA,OAQPS) and Dennis Atkinson (EPA,OAQPS).
11
-------
TABLE OF CONTENTS
Page
NOTICE i
ACKNOWLEDGEMENTS ii
TABLE OF CONTENTS iii
FIGURES iv
TABLES v
1.0 INTRODUCTION 1
2.0 MODELING DOMAIN 3
3.0 METEOROLOGICAL DATA 5
4.0 TERRAIN AND LAND USE DATA 6
5.0 SOURCE CHARACTERIZATION 7
6.0 RECEPTORS 8
7.0 MODELING OPTIONS 9
7.1 CALMET Options 9
7.2 CALPUFF Options 10
8.0 ANALYSIS OF THE CALPUFF CONCENTRATION ESTIMATES 11
9.0 SUMMARY AND CONCLUSIONS 26
10.0 REFERENCES 28
APPENDIX A A COMPARISON OF CALPUFF MODELING RESULTS
WITH 1977 INEL FIELD DATA RESULTS by John Irwin
APPENDIX B COMPACT DISCS
in
-------
FIGURES
Figure Page
1. Savannah River Laboratory field experiment site 4
2. Great Plains field experiment site 4
3. Simulated and observed 7-hour average plume for the Savannah River Laboratory tracer
study for a) actual locations and b) observed plume offset 17° to the south 20
4. Simulated and observed for the Great Plains tracer study on July 8, 1980 for a) 5-hour
average plume at the 100-kilometer arc and b) 12-hour average plume at the 600-
kilometer arc 21
5. Upper air soundings at 1200 GMT for Oklahoma City for July 8 and July 9 22
6. Location of the simulated plume using P-G dispersion coefficients for the July 8 Great
Plains tracer release. Date and time are shown next to each puff. Dots are receptor
locations at the 600-kilometer arc. Terrain elevations are in meters 23
7. Location of the simulated plume using similarity dispersion coefficients for the July 8
Great Plains tracer release. Date and time are shown next to each puff. Dots are receptor
locations at the 600-kilometer arc. Terrain elevations are in meters 24
8. Simulated and observed 6-hour average plume for the Great Plains tracer study on July
11, 1980 25
IV
-------
TABLES
Table Page
TABLE 1 SUMMARY OF THE TIME THE PLUME ARRIVED AT THE
MONITORING/SAMPLING ARCS, THE TIME FOR PLUME TO PASS THE
ARCS, AND THE LOCATION OF THE PLUME CENTERLINE 18
TABLE 2 SUMMARY OF THE OBSERVED AND ESTIMATED CENTERLINE
MAXIMUM CONCENTRATION (CMAX), LATERAL DISPERSION (oy),
AND CROSSWIND INTEGRATED CONCENTRATION (CWIC) 19
-------
1.0 INTRODUCTION
Dispersion models such as the Industrial Source Complex Short Term (EPA, 1995)
typically assume steady, horizontally homogeneous wind fields instantaneously over the entire
modeling domain and are usually limited to 50 kilometers from a source. However, applications
hundreds of kilometers from a source require other models or modeling systems. At these
distances, the transport times are sufficiently long that the mean wind fields cannot be considered
steady or homogeneous. CALPUFF is one such modeling system, consisting of several
components: CALMET, a meteorological preprocessor that utilizes surface, upper air, and on-site
meteorological data to create a three-dimensional wind field and derive boundary layer
parameters based on gridded land use data; CALPUFF, a puff dispersion model that can simulate
the effects of temporally and spatially varying meteorological conditions on pollutant transport,
remove pollutants through dry and wet deposition processes, and transform pollutant species
through chemical reactions; and CALPOST, a postprocessor that takes the hourly estimates from
CALPUFF and generates n-hr estimates as well as tables of maximum values.
Concentration estimates from the CALPUFF dispersion model were compared to
observed tracer concentrations from two short term field experiments. The first experiment was
at the Savannah River Laboratory (SRL) in South Carolina in December 1975 (DOE, 1978) and
the second was the Great Plains experiment near Norman, Oklahoma (Ferber et al., 1981) in July
1980. Both experiments examined long-range transport of inert tracer materials to demonstrate
the feasibility of using other tracers as alternatives to the more commonly used sulfur
hexafluoride (SF6). Several tracers were released for a short duration (3-4 hours) and the resulting
plume concentrations were recorded at an array of monitors downwind from the source. For the
SRL field experiment, monitors were located about 100 kilometers from the source. For the
Great Plains experiment, arcs of monitors were located 100 and 600 kilometers from the source.
Previous studies have compared the results from the Great Plains experiment to
dispersion model results. Carhart et al. (1989) intercompared the results from eight short-term,
long-range dispersion models to the Great Plains results and to a longer-term study at the
Savannah River Laboratory (not the same as used in this study). The primary method for
evaluating model performance was the use of the American Meteorological Society (AMS)
statistical measures (Fox, 1981) and graphical techniques. They concluded that model results
compared in space and in time to observations were generally poor and that predictions for a
specific location and time for averaging periods less than one day were not reliable. They also
noted that unpairing decreases the scatter. They concluded that "model improvement can be
made by better representing the wind field. The use of multiple layers seems to improve results
substantially."
The transport and diffusion of a tracer gas was simulated by Moran and Pielke (1995a,b)
using the Colorado State University mesoscale atmospheric dispersion modeling system, which
consists of a prognostic meteorological model coupled to a mesoscale Lagrangian particle
dispersion model. Results from several simulations with the model were compared to
observations from the Great Plains experiment. Their baseline simulation generally compared
favorably to observations for both arcs although directional errors were apparent by up to 20°.
The results also suggest that the nocturnal low-level jet plays an important role in transport and
1
-------
deformation of the tracer plume and that some flow regimes require better temporal resolution of
boundary layer winds than is available from the National Weather Service (NWS) twice-daily
rawinsondes.
In a study using the CALPUFF dispersion model, Irwin (1997) compared model
concentration estimates to observed concentrations at three arcs (3.2, 48, and 90 kilometers) of
monitors downwind of a three-hour tracer release in April 1977 near Idaho Falls, ID. The
primary focus in Irwin's analysis was the characterization of transport and dispersion using
different combinations of surface and upper air data. He found that the lateral dispersion was
best characterized when all the meteorological data were used but the location of the simulated
maximum relative to the observed maximum on each arc was poorly characterized regardless of
the data used.
In this report, the CALPUFF modeling domains, the meteorological data, terrain and land
use data, sources, receptors and modeling options are described in Sections 2 through 7.
Analyses and discussions are presented in Section 8. Irwin's paper is presented in its entirety in
Appendix A.
-------
2.0 MODELING DOMAIN
The CALPUFF modeling system uses a grid system consisting of an array of horizontal
grid cells and multiple vertical layers. Two grids must be defined in the CALPUFF model
meteorological and computational. The meteorological grid defines the extent over which land
use, winds, and other meteorological variables are defined. The computational grid defines the
extent of the concentration calculations, and is required to be identical to or a subset of the
meteorological grid. For the SRL and Great Plains simulations, the computational grid is defined
to be identical to the meteorological grid. A third grid, the sampling grid, is optional and is used
to define a rectangular array of receptor locations. The sampling grid must be identical to or a
subset of the computational grid. It may also be nested inside the computational grid, i.e., several
sampling grid cells per computational grid cell. For these applications, a sampling grid identical
to the computational grid was used with a nesting factor of one (sampling grid cell size equal to
the cell size of the computational grid).
To properly characterize the meteorology for the CALPUFF modeling system, a grid that
spans, at a minimum, the distance between source and receptor is required. However, to allow
for possible recirculation of puffs that may be transported beyond the receptors and to allow for
upstream influences on the wind field, the meteorological and computational domains should be
larger than this minimum.
For the Savannah River Laboratory (SRL) field experiment, a meteorological grid
extending from 32° N to 34° N latitude and from 80° W to 82° W longitude was used. Figure 1
shows the region of the SRL field experiment. The SRL facility is near the west edge of the
domain and the sampling monitors are located along Interstate 95. The distance between the
source and the receptors is approximately 100 kilometers. A 24-by-24 horizontal grid with a 10-
kilometer resolution was used for the SRL modeling.
The Great Plains site is shown in Figure 2. Two arcs of monitors were deployed during
the field experiment 100 and 600 kilometers. For this analysis, two separate grids were
defined. For the 100-kilometer arc, a grid extending approximately from 35° N to 36.5°N
latitude and from 96° W to 98.5° W longitude was defined. A 42-by-40 horizontal grid with a
10-kilometer resolution was used for this arc. For the 600-kilometer arc, the grid extended from
approximately 35° N to 42°N latitude and from 89° W to 100° W longitude. A 44-by-40
horizontal grid with a 20-kilometer resolution was used for this arc.
To adequately characterize the vertical structure of the atmosphere, six layers were
defined: surface-20, 20-50, 50-100, 100-500, 500-2000, and 2000-3300 meters. This vertical
resolution is consistent with the analysis by Irwin (1997) of the 1977 Idaho Falls field study.
-------
8AMBERG
ALLENOALE
0 5 10 IS 20
MILES
,° 2° 3,° 4,°
KILOMETERS
T I LLMAN
-95
ST. GEORGE
WALTERBORO
MP53
MP46
MP42
YEMASSEE
RIDGELAND
Figure 1. Savannah River Laboratory field experiment site.
43°W
40°
O BATS Sampler!
© BATS & LASL Samplers
R
-------
3.0 METEOROLOGICAL DATA
The CALMET preprocessor utilizes NWS meteorological data and on-site data to
produce temporally and spatially varying three dimensional wind fields for CALPUFF. Only
NWS data were used for this effort and came from two compact disc (CD) data sets (information
on where to obtain these discs is in Appendix B). The first was the Solar and Meteorological
Surface Observation Network (SAMSON) compact discs, which were used to obtain the hourly
surface observations. The following surface stations were used for each of the field experiments:
Savannah River Laboratory
Georgia:
North Carolina:
South Carolina:
Great Plains
Arkansas:
Iowa:
Illinois:
Kansas:
Missouri:
Nebraska:
Oklahoma:
Texas:
Athens, Atlanta, Augusta, Macon, Savannah
Asheville, Charlotte, Greensboro, Raleigh-Durham, Wilmington
Charleston, Columbia, Greer-Spartanburg
Fort Smith
Des Moines
Springfield
Dodge City, Topeka, Wichita
Columbia, Kansas City, Springfield, St. Louis
Grand Island, Omaha, North Platte
Oklahoma City, Tulsa
Amarillo, Dallas-Fort Worth, Lubbock, Wichita Falls
Twice daily soundings came from the second set of compact discs, the Radiosonde Data
for North America. The following stations were used for each of the field experiments:
Savannah River Laboratory
Georgia:
North Carolina:
South Carolina:
Athens, Waycross
Greensboro, Cape Hatteras
Charleston
Great Plains
Arkansas:
Illinois:
Kansas:
Missouri:
Little Rock
Peoria
Dodge City, Topeka
Monett
Nebraska: North Platte, Omaha
Oklahoma: Oklahoma City
Texas: Amarillo
-------
4.0 TERRAIN AND LAND USE DATA
CALMET requires a file of terrain elevations and geophysical parameters in order to
prepare the wind fields and other meteorological parameters. The geophysical parameters are
derived from geographical information system (GIS) land use categories. Terrain and land use
data are available on the CALMET, CALPUFF, and CALPOST Modeling System (version 1.0)
CD (hereafter referred to as the CALPUFF CD). Information on obtaining this compact disc is in
Appendix B.
The terrain and GIS land use data on the CALPUFF CD were used to define gridded land
use data for each field experiment. These data are defined with a resolution of 1/6° latitude and
1/4° longitude. The program PRELND1.EXE, also provided on the CD, was run to extract the
data from the GIS data base and map the data to the meteorological domain for each field
experiment. The program ELEVAT.EXE (also provided on the CD) was used to process the raw
terrain data into average gridded terrain data. The file of terrain and geophysical parameters
required by CALMET was constructed from the output files generated by ELEVAT and
PRELND1 with additional required records inserted manually to create the final forms of the file
for each field experiment.
-------
5.0 SOURCE CHARACTERIZATION
The primary purpose of the Great Plains (Oklahoma) experiment was to demonstrate the
efficacy of perfluorocarbons as tracers in atmospheric dispersion field studies.
Perfluoromonomethylcyclohexane (PMCH), perfluorodimethylcyclohexane (PDCH), SF6, and
two heavy methanes were released during this experiment. For this analysis, the PDCH emission
rates were used since the monitoring data appeared to have a more complete record of PDCH
concentrations.
During the Savannah River Laboratory (SRL) study, SF6 and two heavy methanes were
released. For this analysis, the SF6 tracer emission rates were used. The following source
parameters were used for this analysis:
Source
SRL
Oklahoma
(July 8)
Oklahoma
(July 11)
Release
height
(m)
62.0
10.0
10.0
Stack
diameter
(m)
1.0*
1.0*
1.0*
Exit
velocity
(ms-1)
0.001
0.001
0.001
Exit
temp.
(K)
ambient
ambient
ambient
Total tracer
released
(kg)
154
186
26
Length of
release
(hr)
4.0
3.0
3.0
Emission rate
(g s'1) and
tracer
10.69
SF6
17.22
PDCH
2.41
PDCH
* The stack diameter for each study is the same as was used for the study with the INEL data
For both experiments, the emission rate was assumed to be constant over the entire period
of the release. The emission rates were computed as follows:
total tracer released in kg lOOOg 1 hr
A- JC-
length of release in hours 1 kg 3600 sec
The exit temperature is assumed to be the same as the ambient atmospheric temperature.
CALPUFF checks the difference between the stack exit temperature and the surface station
temperature. If this difference is less than zero, the difference is set to zero. To insure this
condition, an exit temperature of 250 K was input to the model.
-------
6.0 RECEPTORS
Both discrete and gridded receptors were used in this study. The discrete receptors were
placed at approximately the same distance as the monitoring arcs. The gridded receptors were
located at the center of each sampling grid cell to provide sufficient information to plot contours
of concentration estimates.
For the SRL experiment, a single monitoring arc was used at approximately 100
kilometers. The monitors were located along 1-95 (Figure 1) from Mile Post 76 (MP76) on 1-95
near St. George, SC south to Hwy 336 west of Tillman, SC and along SC 336. The monitors
subtended an arc of about 70°. Receptors for modeling were placed along an arc every 1/4°
degree from MP76 to MP22 near Ridgeland, resulting in 261 receptor locations. The distance
between receptors was about 440 meters.
For the July 8 Great Plains experiment, there were two arcs of monitors: 100 kilometers
and 600 kilometers as shown in Figure 2. At 100 kilometers, the monitors subtend an arc of
about 75°. Receptors were placed every 1/4° degree, extending beyond the edges of the
monitoring arc, resulting in 361 receptor locations. The distance between receptors was about
440 meters. At the 600-kilometer arc, the monitors subtend an arc of about 80°. Receptors were
placed every 1/2° degree, about every 5,200 meters apart, resulting in 161 receptor locations.
-------
7.0 MODELING OPTIONS
In the CALPUFF modeling system, each of the three programs (CALMET, CALPUFF,
and CALPOST) uses a control file of user-selectable options to control the data processing.
There are numerous options in each and several that can result in significant differences. The
following model controls for CALMET and CALPUFF were employed for the analyses with the
tracer data.
7.1 CALMET Options
The following CALMET control parameters and options are chosen to be consistent with
the 1977 INEL study by Irwin (1997). The most important options relate to the development of
the wind field and were set as follows:
IWFCOD = 1 Use diagnostic wind model to develop the 3-D wind fields
IFRADJ = 1 Compute Froude number adjustment effects (thermodynamic
blocking effects of terrain)
IKTNE =1 Compute kinematic effects of terrain
IOBR = 0 Do NOT use O'Brien procedure for adjusting vertical velocity
IEXTRP = 4 Use similarity theory to extrapolate surface winds to upper layers
IPROG = 0 Do NOT use prognostic wind field model output as input to
diagnostic wind field model
Mixing heights are important in the estimating ground level concentrations. The options
that affect mixing heights were set as follows:
IAVEZI = 1 Conduct spatial averaging
MNDAV = 3 Maximum search radius (in grid cells) in averaging process
HAFANG =30. Half-angle of upwind looking cone for averaging
ILEVZI =1 Layer of winds to use in upwind averaging
DPTMIN = .001 Minimum potential temperature lapse rate (K/m) in stable layer
above convective mixing height
DZZI = 200 Depth of layer (meters) over which the lapse rate is computed
ZEVIIN = 20 Minimum mixing height (meters) over land
ZEVIAX =3300 Maximum mixing height (meters) over land, defined to be the top
of the modeling domain
-------
7.2 CALPUFF Options
The following CALPUFF control parameters, which are a subset of the control
parameters, were used. As with CALMET, these parameters and options were chosen to be
consistent with the 1977 INEL study.
Technical options (group 2):
MCTADJ
MCTSG
MSLUG
MTRANS
MTIP
MSHEAR
MCHEM
MWET
MDRY
MPARTL
MREG
= 0
= 0
= 1
= 1
= 1
= 0
= 0
= 0
= 0
= 0
= 0
No terrain adjustment
No subgrid scale complex terrain is modeled
Near field puffs modeled as elongated (i.e., slugs)
Transitional plume rise is modeled
Stack tip downwash is modeled
Vertical wind shear is NOT modeled above stack top
No chemical transformations
No wet removal processes
No dry removal processes
No partial plume penetration
No check made to see if options conform to regulatory options
Two different values were used for the option MDISP:
= 2 Dispersion coefficients from internally calculated sigmas
= 3 PG dispersion coefficients for RURAL areas
Several miscellaneous dispersion and computational parameters (group 12) were set as follows:
SYTDEP = 550. Horizontal puff size beyond which Heffter equations are used for
sigma-y and sigma-z
MHFTSZ =0 Do NOT use Heffter equation for sigma-z
XMXLEN =0.1 Maximum length of slug (in grid cells)
XSAMLEN =0.1 Maximum travel distance of puff/slug (in grid cells) during one
sampling step
MXNEW =199 Maximum number of slugs/puffs released during one time step
SL2PF =5.0 Slug-to-puff transition criterion factor (= sigma-y/slug length)
10
-------
8.0 ANALYSIS OF THE CALPUFF CONCENTRATION ESTIMATES
With field experiment results, where data are collected at arc(s) of monitors for short
periods of time, the standard statistical measures to compare observed and estimated
concentration data, such as recommended by the American Meteorological Society (Fox, 1981),
are not easily interpreted. One way to present the results is to compare "n-hour" averages, where
"n" represents the time required for the plume to pass the arc of receptors. Measures that can be
used to compare results include the fitted and absolute (centerline maximum concentration
(Cmax and Omax), the lateral dispersion, oy, and the crosswind integrated concentration (CWIC)
for the observed concentration and the model estimates. Additionally, the time required for the
plume to reach the arc, the length of time required to pass the arc, and the location of the plume
centerline can provide information on how well a model simulated the plume transport. These
measures with graphical displays of the data are used here to present the results.
Using the Colorado State University (CSU) mesoscale atmospheric dispersion (MAD)
model, Moran and Pielke (1995a,b) simulated the transport and dispersion of the tracer cloud for
the July 8 Great Plains experiment. The results from his baseline model run (referred to as run
4b in his papers) are included in this analysis. Since they were interested in mesoscale dispersion
on the 500-1000 kilometer scale, Moran and Pielke did not simulate the July 11 Great Plains
experiment.
The first part of the CSU MAD modeling system consists of the CSU Regional
Atmospheric Modeling System (RAMS), a prognostic mesoscale meteorological model that
develops three-dimensional mean wind and turbulence fields over complex terrain. A complete
description of the RAMS model can be found in Moran and Pielke (1995a). For their
simulations, Moran and Pielke (1995a) used a 41 x 46 horizontal grid with a 1/2° longitude by
1/3 ° latitude resolution and 29 vertical layers. The layer at the surface was 50 meters deep and
the domain was extended to 16.2 kilometers using a "1.15 stretch factor." Also included was an
11-level soil model extending to a depth of 0.5 meters.
Table 1 shows the time the plume reached the receptor arc, the length of time for the
plume to pass the arc completely, and the plume centerline location based on a fitted curve
through the concentration values at each monitor or receptor. Table 2 shows the statistical
measures used in the comparison. Two separate CALPUFF model runs were made: 1) using
Pasquill-Gifford (P-G) dispersion parameters, and 2) using dispersion coefficients from
internally-calculated ov and ow from the micrometeorological variables calculated in CALMET
(hereafter referred to as similarity dispersion). The central maximum concentration is estimated
from a Gaussian fit to the modeled and observed data (Cmax) and computed from the CWIC and
oy, as Cmax= CWIC/((/27i)oy). Omax is the actual maximum obtained directly from the monitored
and modeled concentrations. The CWIC was computed by trapezoidal integration. The program
that computed these measures utilized only those values that were 1% or greater of Omax.
8.1 Savannah River Laboratory
The observed concentrations are the cumulative concentration from bag samples located
along Interstate 95 from about St. George south to Ridgeland (Figure 1). Background
11
-------
concentration was estimated to be with a 0.5 part per trillion (ppt) (DOE, 1978). A continuous
tracer release started at 1025 Local Standard Time (LSI) and continued until 1425 LSI. The bag
samplers were started at different times, ranging from about 1040 to 1230 LST, and the duration
of the sampling ranged from 7.0 to 7.5 hours. Since the release started as 1025 LST, it seems
likely that sampling at the monitors would have begun prior to the arrival of the plume at all of
the monitors. The arrival time of the modeled plume was the hour ending at 1300 LST for both
P-G and similarity dispersion. The simulated plume required seven hours to pass the arc with the
P-G dispersion coefficients, but only six hours with similarity dispersion coefficients. Therefore,
seven-hour-average modeled and observed concentrations were computed for the Savannah River
Laboratory field experiment. Since the first monitors were turned on prior to 1100 LST and only
cumulative concentration is reported for the observed data, the simulated concentrations were
summed over the seven-hour period from 1100 LST through 1800 LST.
Figure 3a shows the plots of the concentration estimates at the receptors (continuous
curves) and the observed concentrations at the receptors (labeled points). The modeled peaks are
10° to 20° further to the south than the observed peak. Figure 3b shows the same plot except the
observed concentrations have been shifted by 17° to more closely align with the P-G dispersion
simulated concentrations. Clearly, there is general agreement in the shape and magnitude of the
distributions.
Note that there are two local maxima in the observations near 135° and 145°. The winds
were more northerly shortly after the release and may have resulted in the observed local peaks
(DOE, 1978) that were not captured in the modeled meteorology. The observed lateral
dispersion is 50-100% larger than the modeled dispersion due to these local peaks (Table 1). If
these two secondary peaks are omitted from the analysis, then the statistical measures of the
simulated plumes are in better agreement with the measures of the observed plume. Without
these secondary peaks, the fitted central maximum to the observations increases by 37% to 3.758
ppt (modeled: PG 7.2 ppt and Similarity 5.1 ppt), oy is reduced by 33% to 7.77 km (modeled:
PG 6.9 km and Similarity 5.0 km), and the CWIC is reduced only slightly to 0.732 ppt-m
(modeled: PG 1.29 ppt-m and Similarity 0.8 ppt-m). The meteorology, as simulated for the
Savannah River tracer release, did not characterize this initial difference in wind direction
sufficiently to transport the plume more toward the south.
8.2 Great Plains
Two study days in 1980 - July 8 and July 11 - formed the Great Plains tracer experiment.
Moran and Pielke, (1995a,b) in their analyses with the CSU MAD model, studied the July 8 case.
Results from their baseline simulation are included here for comparison, though he simulated the
PMCH release. Moran and Pielke's simulated PMCH concentrations were divided by 1.18 for
comparison to the observed and simulated PDCH concentrations. This value is the volume of
PMCH divided by the volume of PDCH, as given by Ferber et al. (1981).
12
-------
8.2.1 July 8. 1980
Beginning at 1300 LST, the PDCH and PMCH tracer gases were released at a constant
rate for a three-hour period from an open field at the National Severe Storms Laboratory in
Norman, Oklahoma. The PDCH was released at an average rate of 17.22 g s"1 as an aerosol
spray. Two arcs of air samplers recorded the passage of the tracer plume for this release - the
first at 100 kilometers and the second at 600 kilometers. The locations of the samplers at these
two arcs are shown in Figure 2. A background concentration for PDCH of 26 ppt (Ferber et al.,
1981) was removed from the observed concentrations. The results for the 100-kilometer arc are
discussed first.
100 kilometers
The samplers were set to operate for ten 45-minute periods, sufficient time for the tracer
material to pass the 100-kilometer arc. The initial sampling period began at 1500 LST, but the
concentrations in the first 45-minute period were near the background concentration. The
sampling continued through 2230 LST on July 8, with background concentrations observed for
the periods after 2100 LST (9th 45-minute period). Near-background concentrations (0-6 ppt
depending on the receptor) were observed for the periods 7 and 8 (from 1930 to 2100 LST).
Thus, the length of time the plume passed the arc was approximately five hours (1545-2100
LST). The simulated CALPUFF plume using the P-G dispersion coefficients was evident at the
100-kilometer arc for six hours, but the plume passed the arc in five hours when the similarity
dispersion coefficients were used (Table 1). Given the transit time of the observed plume and the
simulated transit time using similarity dispersion, five-hour average concentrations were used in
the analysis. The hours ending 17 through 21 were used to construct the simulated five-hour
concentration average (representing the period 1600-2100), and the seven 45-minute periods
from 1545 through 2100 LST were used to construct the observed average concentrations.
Moran and Pielke (1995b) used a transit time of 3.75 hours based on near-background
concentrations of PMCH recorded for the 7th 45-minute period and later, i.e., five 45-minute
periods were used. This difference demonstrates the variability in how to define the plume
transit time, and applies to defining the plume width as well. Moran and Pielke (1995a) discuss
the arbitrariness of defining plume width and this discussion can be extended to plume transit
time past an arc of monitors (the longitudinal dimension of the wind). There are several methods
for defining the plume width, and by extension, the transit time (with background removed):
concentrations above a percentage of the maximum concentration in the interior of the
plume, e.g., 10% or 1%,
a concentration above a threshold that is independent of the plume interior
concentrations, e.g., 0.002 or 0.003 ppt above background, and
nonzero concentrations above background.
The transit time in this analysis is defined by nonzero concentrations with background
removed (the third criterion), but the statistical measures were computed only for concentrations
greater than 1% of the maximum (the first criterion). Moran and Pielke utilized the second
13
-------
criterion, with a minimum of 0.003 ppt (without background) as an absolute minimum. The
differences in defining the transit time should not alter the conclusions of this analysis.
The five-hour average modeled and observed concentrations for the 100-kilometer arc on
July 8 of the Great Plains field experiment are shown in Figure 4a along with Moran and Pielke's
baseline simulation (experiment 4b). Two things are immediately apparent: the monitoring did
not capture the entire plume and the observed maximum concentration is very likely less than the
simulated maxima. Given the incomplete sampling of the observed plume at 100 kilometers for
this release, the statistical measures of the observed plume likely are suspect and are not
sufficient to draw conclusions regarding model performance. Nonetheless, the measures were
calculated for the observed plume and are shown in Table 2 along with the statistics for the
simulated plumes and for Moran and Pielke's July 8 simulation.
Although oy shows good agreement between simulated and observed plumes in Table 2,
from Figure 4a it appears that the lateral dispersion of the observed plume is wider than the
dispersion of the simulated plumes. From the available data, the peak may have occurred at or
near receptor 12. Aircraft samples suggest that the plume did not extend much more to the west
of receptor 12 (Ferber et al., 1981). The agreement between ground level concentrations and
aircraft observations taken between 1630 LST and 1800 LST about 1250 meters above the arc of
receptors (Figure 11 in Ferber et al., 1981) suggests a well-mixed layer through this depth. With
this information, the ground level plume was extended to the west, assuming 0.9 ppt at receptor
10 (azimuth -10°) and 0.0 ppt at receptor 8 (azimuth -15°). The statistical measures for this
'extended' plume are Cmax =1.3 ppt, oy = 10.9 km, and CWIC = 0.35 ppt-m.
Assuming a well-mixed boundary layer (which appears to be a reasonable assumption
here based on aircraft observations), the CWIC can be approximated by Q/(u z;), where u is
wind speed (meters/second) and z; is the mixing height (meters). From the afternoon sounding
for Oklahoma City on this day (not shown), the observed mixing height is about 2500 m with an
average wind speed through the boundary layer of about 10m s"1. The emission rate was
17.22 g s"1 for three hours with sampling at the monitors for five hours. These values yield a
CWIC of 0.27 x 105 ppt-m, which is slightly smaller than the observed CWIC of
0.29 x 10s ppt-m. However, the modeled boundary layer does not appear to be well-mixed for
most hours when the plume is passing the 100-kilometer arc. The modeled mixing heights are
around 2000 meters, but oz is around 1000 meters between the source and the receptor locations,
suggesting that the simulated atmosphere may be neutral, or possibly stable. These conditions
would tend to increase the surface concentration. Examining the P-G stability parameter for
these hours, the atmosphere is neutral and the CWIC for P-G dispersion is nearly a factor of two
higher than what is expected for a well-mixed boundary layer.
In comparing the CALPUFF results to Moran and Pielke's simulation, the CSU MAD
model placed the maximum about 25° west of the actual plume. The statistical measures appear
to be similar to the measures for the CALPUFF simulations. From Table 2, Moran and Pielke's
result for the 100 kilometer arc are very similar to the CALPUFF simulation using similarity
dispersion.
14
-------
600 kilometers
The sampling at the 600-kilometer arc began at 0200 LST on July 9 and continued
through 1100 LST on July 11. There were 20 3-hour sampling periods. The sampling began
after the plume arrived at the sampling arc. The reason the plume arrived sooner than expected is
discussed below. Thus, the length of time the observed plume passed the sampling monitors
cannot be determined with certainty. The average observed concentrations exclude any tracer
material that passed the monitors between the time the plume first reached the monitors and the
time the monitors were activated. After about 1400 LST on July 9, the concentrations at the
samplers were at or near background, although Ferber et al. (1981) notes that small amounts of
tracer material were detected up to two days later at several of the receptors. Concentrations
were simulated by CALPUFF at the 600-kilometer arc for 14 hours when using the P-G
dispersion coefficients and for 13 hours when the similarity dispersion coefficients were used
(Table 1). For this release and arc, 12-hour average concentrations were used for best temporal
alignment with the observed data. Moran and Pielke used 15-hour simulations in the modeling
of the Great Plains experiment.
The 12-hour average modeled and observed concentrations for the 600-kilometer arc on
July 8 are shown in Figure 4b. Two things are apparent: 1) the observed maximum concentration
is about three times higher than the simulated concentrations and 2) the maxima of the
simulations are in relatively good agreement with each other.
As noted above, the tracer arrived at the sampling arc earlier than anticipated and the
sampling likely missed some of the tracer material. Ferber et al. (1981) speculate that the plume
probably arrived just before the samplers were activated and a small amount of plume material
was not collected. The most likely reason for the earlier-than-expected arrival was the formation
of a low-level nocturnal jet, supported by a variety of meteorological measurements (Moran and
Pielke, 1995a). Hoecker (1963), in detailed studies of the low-level jet over the Midwestern
plains (from Amarillo, TX to Little Rock, AR) using a series of pibal stations, found that jet
speed maxima occur between 300 and 800 meters above local ground. In examining available
data for the 1980 Great Plains field experiment, Moran and Pielke (1994) note an approximate
doubling of the average nocturnal wind speeds from their daytime values. Examination of the
upper air wind profiles for Oklahoma City through the period indicate the presence of a jet
between 500 and 1000 meters for the 1200 Greenwich Mean Time (GMT) soundings. The wind
profiles from the 1200 GMT (0600 LST) for July 8 and July 9 are shown in Figure 5. A jet of
about 17ms"1 is very prominent between 500 and 750 meters on the mornings of July 8 and 9.
Since these soundings are taken at 0600 LST, it is possible that the low-level jet could have been
stronger prior to the sounding time.
The presence of the jet is apparent in the simulated wind fields from CALMET for
several periods during the night of July 8-9. Wind speed maxima are noted at levels 3 and 4 (at
50-100 and 100-500 meters, respectively) in the region of the simulated plume with noticeable
drop offs in speed away from the plume. Although the input meteorology and processing in
CALMET appear to have simulated the presence of a jet, the jet may not be sufficiently
characterized to simulate the transport and dispersion of the observed plume. The result is a
larger lateral dispersion and smaller central maximum of the simulated plumes.
15
-------
Another 'feature' apparent in the simulated plume using P-G dispersion is the presence of
a local maximum about 15° further east (Figure 4b). Although plume splitting was not turned on
in CALPUFF, there is a definite secondary maximum. A similar phenomenon is noted in the
modeling results for the 1977 INEL study by Irwin (1997) (see Figure 3c in Appendix A). The
secondary maximum in the INEL study appears about 20° east of the primary maximum at the
90-kilometer arc using P-G dispersion. In both the Great Plains and INEL analyses (Appendix
A), the secondary maximum is not present using similarity dispersion. A likely explanation is
that the puffs became well-mixed due to greater vertical dispersion with the P-G dispersion
coefficients, resulting in the puffs being affected more by the upper level winds. In examining
the vertical dispersion for the first several hours after the release, oz for the slug or puff farthest
from the source grows more rapidly to a larger value using the P-G dispersion coefficients (1400
m to 4200 m in 3 hours) compared to using similarity dispersion coefficients (1200 m to 3300 m
in 6 hours). The well-mixed slugs/puffs "break away" from the other slugs/puffs and are
transported by the upper level winds. In the hour before the "break away" was first noticed in the
concentration pattern, the upper level winds were more from the west and the lower level winds
were from the south-southwest. Figures 6 and 7 show a time series of the plume as it was
transported downwind from the source, across the receptor network, and beyond using P-G
dispersion coefficients and similarity dispersion coefficients, respectively. As seen in Figure 6,
for 0200 on July 9 there is plume material that is beginning to be transported away from the main
body of the plume. At 0700, plume material had been transported in a more westerly direction
across the receptor arc and was clearly south of the main body of the plume later in the period.
There was no comparable "break" when similarity dispersion was used (Figure 7).
Also of note in Figures 6 and 7 is the veering of the plume toward the east. On July 8 and
9, a large area of high pressure dominated the region. Early on July 8, a cold front associated
with weak low pressure moved through the area to the southeast. The synoptic situation was
such that the tracer, which was released in late morning, was transported on the western side of a
this high pressure system. Moran and Pielke (1995a) show the streamline fields of the observed
winds for July 8 and 9 near the surface and at about 1600 meters above ground level. The flow
in the lower level is southerly and more southwesterly and westerly in the upper level,
particularly later in the period.
8.2.2 July 11. 1980
Beginning at 1300 LST, tracer gases were released for a three-hour period using the same
system as on July 8. The PDCH was released as an aerosol spray at an average rate of 2.41 g s"1,
about 1/7 the release rate on July 8. A background concentration for PDCH of 26 ppt was
removed from the observed concentrations.
The samplers were set to operate for nine 45-minute periods, sufficient time for the tracer
material to pass the 100-kilometer arc. The initial sampling period began at 1600 LST and
continued through 2245 LST. Near-background concentrations were observed at several of the
samplers for the final period (beginning at 2200 LST) and there was no report for several others
for the final period. Thus, the transit time for the observed plume is six hours. The transit time of
the simulated plume in CALPUFF using both P-G and similarity dispersion coefficients also was
16
-------
six hours (Table 1). Therefore, six-hour average concentrations were used in this part of the
analysis.
The six-hour average modeled and observed concentrations for the 100-kilometer arc are
shown in Figure 8. As with the July 8 study, the monitoring did not capture the entire plume at
100 kilometers, although the peak appears to be a little better defined, with an observed
maximum at receptor 18. There were no aircraft flights to assist in determining the western
extent of the plume. The simulated plumes using P-G and similarity dispersion agree with each
other very well, but, as with the July 8 results for the 100-kilometer arc, the peaks are more than
twice the magnitude of the observed plume and the simulated lateral dispersion is less than the
observed plume. The statistical measures were calculated for both the observed and simulated
plumes and are shown in Table 2. Given the incomplete sampling of the observed plume at 100
kilometers for this release, the statistical measures of the observed plume likely are suspect and
are not sufficient to draw conclusions regarding model performance.
Unlike the July 8 case, no aircraft observations were available on July 11 to assist in
determining if the boundary layer was well-mixed. In examining "debug" output from
CALPUFF, several of the slugs/puffs are well-mixed within a couple hours of the release.
Comparing the simulated mixing heights to the vertical dispersion, oz, confirms that the boundary
layer appears to be well-mixed for the daytime hours. Using the approximation CWIC = Q/(u
zj for a well-mixed atmosphere, and assuming a wind speed of 8 m s"1 through a mixing depth
of 2600 m (from the Oklahoma City afternoon sounding for July 11), the CWIC is about
0.043 x 10s ppt-m, which is essentially identical to the observed CWIC. If the observed plume is
"completed" by reflecting the rightmost three points (labeled 23, 24, and 26) on the left side, the
observed CWIC becomes 0.052 ppt-m, oy = 16.9 km, and Cmax = 0.12 ppt.
As with the 100 kilometer arc for the July 8 study, the question remains - why do the
simulated plumes have higher central maxima and narrower dispersion. A more detailed
examination of the transport and dispersion algorithms in CALPUFF would be required to begin
to answer this question and is beyond the scope of this effort. With a more sophisticated
modeling system, Moran and Pielke (1995a,b) encountered similar differences in their
examination of the July 8 simulation at 100 kilometers and could not completely explain why the
dispersion model was not able to more closely represent the observed dispersion patterns at the
receptor arcs.
17
-------
TABLE 1
SUMMARY OF THE TIME THE PLUME ARRIVED AT THE MONITORING/SAMPLING
ARCS, THE TIME FOR PLUME TO PASS THE ARCS, AND THE LOCATION OF THE
PLUME CENTERLINE.
Arrival Time at Arc
(Julian day: hour)
Savannah River
Great Plains, July 8, 100 km
Great Plains, July 8, 600 km
Great Plains, July 11, 100 km
Observed
{1}
190:1545(a)
{2}
193:1645
P-G
Dispersion
344:1300
190:1600(b)
191:0300
193:1700
Similarity
Dispersion
344:1300
190:1600(b)
191:0300
193:1700
Moran
Exp4b
na
na
Length of Plume Passage (hours)
Savannah River
Great Plains, July 8, 100 km
Great Plains, July 8, 600 km
Great Plains, July 11, 100 km
Location of Plume Centerline
(degrees from north)
Savannah River
Great Plains, July 8, 100 km
Great Plains, July 8, 600 km
Great Plains, July 11, 100 km
7-7.5
5
{2}
6-7
126
1
10
15
7
6
14
6
141
357
25
10
6
5
13
6
138
360
24
10
3.75
15
17
21
{1}
{2}
(a)
(b)
na
Sample collection starting times varied from about 10:45 a.m. to 12:30 p.m. LST
Sampling started at 0200 LST on July 9 after plume arrived at the 600-km arc
Starting at the 45-minute period shown (LST)
For the period ending at the hour shown (LST)
not available
18
-------
TABLE 2
SUMMARY OF THE OBSERVED AND ESTIMATED CENTERLINE MAXIMUM
CONCENTRATION (CMAX), LATERAL DISPERSION (oy), AND CROSSWIND
INTEGRATED CONCENTRATION (CWIC).
Cmax (ppt) - fitted maximum
Savannah River
Great Plains, July 8, 100 km
Great Plains, July 8, 600 km
Great Plains, July 11, 100 km
INEL, 90 km
Observed
2.70
1.30
0.54
0.11
6.40
CALPUFF dispersion
P-G
7.20
2.70
0.11
0.26
4.00
Similarity
5.10
1.90
0.14
0.28
9.40
Moran
Exp4b
~
2.34
0.11
-
-
Omax (ppt) - actual maximum
Savannah River
Great Plains, July 8, 100 km
Great Plains, July 8, 600 km
Great Plains, July 11, 100 km
INEL, 90 km
5.10
1.05
0.38
0.11
5.00
6.90
2.60
0.13
0.25
-5.30
5.00
1.80
0.13
0.27
-9.00
~
2.27
0.10
-
-
oy(km)
Savannah River
Great Plains, July 8, 100 km
Great Plains, July 8, 600 km
Great Plains, July 11, 100 km
INEL, 90 km
11.6
9.1
19.4
14.1
13.1
7.2
9.0
64.9
9.2
18.9
6.0
6.9
42.6
8.5
12.3
~
6.9
28.6
-
-
CWIC(ppt-mxl05)
Savannah River
Great Plains, July 8, 100 km
Great Plains, July 8, 600 km
Great Plains, July 11, 100 km
INEL, 90 km
0.80
0.29
0.26
0.040
2.10
1.29
0.61
0.17
0.060
1.90
0.77
0.33
0.15
0.059
2.90
~
0.41
0.08
-
-
19
-------
(a)
7.0-1
Savannah River
Dec. 10, 1975
I.QQ.km.arc.
P-G Dispersion
Similarity Dispersion
Observations
0.0
100 110 120 130 140
Azimuth (decrees)
150
160
(b) Savannah River
Dec. 10, 1975
J.QO.kra.arc
P-G Dispersion
Similarity Dispersion
Observations
0.0
100 110 120 130 140 150 160 170
Azimuth (decrees)
Figure 3. Simulated and observed 7-hour average plume for the Savannah River Laboratory tracer
study for a) actual locations and b) observed plume offset 17° to the south.
20
-------
(a)
3.0 -I
Great Plains
JulyS, 1980
100 km arc
Q.
a.
.0
2
o
o
o
a
a.
2.0 -
P-G Dispersion
Similarity Dispersio
Observations
-20
-10
0 10 20
Azimuth (decrees)
(b)
0.40 -i
Great Plains
JulyS, 1980
600 km arc
0.30-
.0
O
O
O
Q
0.
0.20-
0.10-
0.00
P-G Dispersion
Similarity Dispersion
Observations
Moran 4b
10 20 30
Azimuth (decrees)
40
50
Figure 4. Simulated and observed for the Great Plains tracer study on July 8, 1980 for a) 5-hour
average plume at the 100-kilometer arc and b) 12-hour average plume at the 600-
kilometer arc.
21
-------
Oklahoma City Soundings
July 1980
5000 -r
4000 -
3000 -
O
2000 -
1000 -
July 9, 1200 GMT
Wind Speed (mis)
Figure 5. Upper air soundings at 1200 GMT for Oklahoma City for July 8 and July 9.
22
-------
4600-
4500-
4400-
CO
~ 4300
4200-
4100-
4000-
3900^
-100
i
100 20
200 300 400
UTM Easting (km)
500
600
700
Figure 6. Location of the simulated plume using P-G dispersion coefficients for the July 8 Great
Plains tracer release. Date and time are shown next to each puff. Dots are receptor
locations at the 600-kilometer arc. Terrain elevations are in meters.
23
-------
4600-
4500-
4400-
CO
~ 4300-
4200-
4100-
4000-
3900^
.....
\ Similarity dispersion
sp
g
3'km
receptor arc ,,,
'
,o / Julys
o g / / 02 LSI
/
JulyS
"-": July9
.17 LSI
.,-"" Julys \
12LST
15 LSI
\
-100
100 200 300 400
UTM Easting (km)
II
500 600 700
Figure 7. Location of the simulated plume using similarity dispersion coefficients for the July 8
Great Plains tracer release. Date and time are shown next to each puff. Dots are
receptor locations at the 600-kilometer arc. Terrain elevations are in meters.
24
-------
Great Plains
July 11, 1980
100 km arc
0.30 -i
0.20 -
g
"(D
0)
o
c
o
O
X
O 0.10
Q
Q_
0.00
-20
P-G dispersion
Similarity dispersion
A Observations
0 20
Azimuth (degrees)
40
Figure 8. Simulated and observed 6-hour average plume for the Great Plains tracer study on July
11, 1980.
25
-------
9.0 SUMMARY AND CONCLUSIONS
This report compares concentration estimates from the CALPUFF dispersion modeling results
to concentrations observed during two long-range transport field experiments - at Savannah River
Laboratory in 1975 and the 1980 Great Plains experiment. National Weather Service data and land use
data with a resolution of 20-30 kilometers were used in the CALMET meteorological processor to
generate the three-dimensional wind fields and other boundary layer parameters required by
CALPUFF. An earlier study by Irwin (1997) examined the sensitivity of CALPUFF simulations to
alternative combinations of meteorological data. Results of the baseline simulation performed by
Moran and Pielke (1995a,b) using the Colorado State University mesoscale atmospheric dispersion
modeling system are included for comparison to the current simulations.
For these three tracer releases, there is overall agreement between the observed times and the
modeled times for both the time required for the plumes to reach an arc and the time required for the
plume to pass completely by the arc. However, the transport direction tended to be off anywhere from
I°tol5°.
Most of the modeling results also can be considered in reasonable accord with the observations.
The statistical measures for the simulated plumes for the Savannah River and INEL field experiments
were within a factor of two of the observed plumes with no tendency for under- or overestimation. The
results at the 100-kilometer arc for both study days for the Great Plains experiment consistently
underestimated the lateral dispersion and overestimated the central maximum and the crosswind
integrated concentration. The simulated lateral dispersion and CWIC were within a factor two of the
observations, and the simulated fitted central maxima were generally 2 to 21/2 times greater than the
maxima fitted to the observed data. The results at the 600-kilometer arc for the July 8 Great Plains
experiment show a different pattern, with the simulated CWIC and central maxima two to five times
less than the observed values and the simulated lateral dispersion 2V2 -3l/2 times larger than the
observed dispersion.
Moran and Pielke (1995a,b), using the CSU RAMS model to generate the three-dimensional
wind fields and a Lagrangian particle dispersion model, encountered similar difficulties in simulating
the transport and dispersion of the tracer plumes for the July 8 Great Plains experiment. As Moran and
Pielke state, the Great Plains field experiment "was as simple an example of MAD [mesoscale
atmospheric dispersion] as one is likely to encounter," yet the simulations with a prognostic model and
additional complexities in developing the meteorological data were no better than the simulations with
the CALPUFF modeling system.
A likely reason for these differences seen in this study is that the meteorology was not
adequately characterized by the CALMET meteorological processor. This is particularly true for the
simulated concentrations at the 600-kilometer arc for the Great Plains experiment where the simulated
lateral dispersion was much greater than the observed lateral dispersion. A likely contributor to this
difference is the development of a low-level nocturnal jet on July 8 and July 9. Moran and Pielke
(1995b) suggest that the most important factor for predicting the nocturnal low level jet is a realistic
representation of the variability of boundary layer winds induced by the diurnal heating and cooling
cycle. Thus, assumptions and simplifications in the meteorological model could also contribute to the
differences seen between simulated and observed concentrations. Assumptions about atmospheric
26
-------
dispersion at these larger distances in the dispersion model itself could also contribute to the
differences.
The presence of the low level jet does not explain the differences seen at the 100-kilometer arc
where the plume passed by the arc within four to six hours of the release. At present there is no
explanation for the differences between the simulated and observed plumes.
In performing case studies such as presented here, one should keep in mind that each simulation
here is only one realization of an ensemble of realizations. Turbulence, and dispersion asssociated with
turbulence, is random and not predictable. Deterministic solutions are impossible. With additional
observation of mesoscale dispersion, an average picture of plume transport and dispersion may be
possible for the scales discussed here.
27
-------
10.0 REFERENCES
Carhart, R., A. Policastro, M. Wastag, and L. Coke, 1989: Evaluation of Eight Short-term Long-range
Transport Models Using Field Data. Atmos. Environ., 23, 85-105.
Ferber, G., Telegadas, K., Heffter, J., Dickson, C., Dietz, R., and Krey, P., 1981: Demonstration of a
Long-range Transport Atmospheric Tracer System Using Perfluorocarbons. NOAA Technical
Memorandum, ERL ARL-101.
Fox, D., 1981: Judging Air Quality Model Performance. Bull. Amer. Meteor. Soc., 62, 599-609.
Hoecker, W., 1963: Three Southerly Low-level Jet Systems Delineated by the Weather Bureau Special
Pibal Network of 1961. Mow. Wea. Rev., 91, 573-582.
Irwin, J., 1997: A Comparison of CALPUFF Modeling Results with 1977 INEL Field Data Results.
Presented at the 22nd NATO/CCMS International Technical Meeting on Air Pollution Modelling and
its Applications. 2-6 June 1997, Clermont-Ferrand, France.
Moran, M. and Pielke, R., 1995a: Evaluation of a Mesoscale Atmospheric Dispersion Modeling
System with Observations from the 1980 Great Plains Mesoscale Tracer Field Experiment. Part I:
Datasets and Meteorological Simulations. J. Applied Meteor., 35, 281-307.
Moran, M. and Pielke, R., 1995b: Evaluation of a Mesoscale Atmospheric Dispersion Modeling
System with Observations from the 1980 Great Plains Mesoscale Tracer Field Experiment. Part I:
Dispersion Simulations. J. Applied Meteor., 35, 308-329.
Moran, M. and Pielke, R., 1994: Delayed Shear Enhancement in Mesoscale Atmospheric Dispersion.
Preprint volume of the Eighth Joint Conference on Applications of Air Pollution Meteorology with
A&WMA, Nashville, TN, January 23-28, 1994.
U.S. Department of Energy, 1978: Heavy Methane-SF6 Tracer Test Conducted at the Savannah River
Laboratory, December 10, 1975. DP-1469. Prepared by E.I. du Pont de Nemours and Company,
Savannah River Laboratory, Aiken, South Carolina.
U.S. EPA, 1995: User's Guide for the Industrial Source Complex (ISC3) Dispersion Models. Volume
I - User Instructions. EPA-454/B-95-003a. U.S. Environmental Protection Agency, Research Triangle
Park, NC, 27711.
28
-------
APPENDIX A
A COMPARISON OF CALPUFF MODELING RESULTS
WITH 1977 INEL FIELD DATA RESULTS
by
John Irwin
Presentation at 22nd NATO/CCMS International Technical Meeting on Air Pollution Modelling and
Its Application. June 2-6, 1997. Clermont-Ferrand, France. To be published by Plenum Press in Air
Pollution Modelling and Its Application XU.
-------
A COMPARISON OF CALPUFF MODELING RESULTS WITH 1977 INEL FIELD
DATA RESULTS
John S. Irwin1
Atmospheric Sciences Modeling Division
Air Resources Laboratory
National Oceanic Atmospheric Administration
Research Triangle Park, NC 27711
U.S.A.
INTRODUCTION
This paper provides a summary of results from a series of analyses in which puff
dispersion modeling results were compared with data obtained following a single 3-hour late
afternoon tracer release, lasting from 1240 to 1540 Mountain Standard Time (MST),
conducted on April 19, 1977 near Idaho Falls, Idaho. The puff modeling results were obtained
using the CALPUFF dispersion modeling system (EPA, 1995a,b). This modeling system
consists of a meteoro-logical processor called CALMET, which is capable of developing time-
dependent multi-layered wind fields using a diagnostic wind model; and a puff dispersion
model called CALPUFF, which is capable of simulating the hour-by-hour variations in
transport and dispersion. The tracer release results (Clements, 1979) were obtained as a
consequence of an investigation into the feasibility of using certain perfluorocarbons and heavy
methanes as alternative tracers in place of sulfur hexafluoride (SF6). Hence, although the
results have found use for testing alternative characterizations of dispersion and transport, this
was not a primary purpose in the original design of the investigation. Draxler (1979) included
this experiment in an assessment of the effects of alternative methods of processing wind data
for characterization of the mesoscale trajectory and dispersion. He concluded that a network
of wind observations having a spacing on the order of 25 kilometers might be needed to
simulate mesoscale transport associated with variable-flow situations, and that spacing of order
100 kilometers might prove adequate for stationary and homogeneous flow situations.
METEOROLOGICAL DATA PREPARATION
The design for meteorological data collection and sampling locations relative to the
release location is shown in Figure 1. Since locations of towers and sites were extracted from
data volume figures, the relative positions are likely accurate but the absolute positions are no
1 On assignment to the Office of Air Quality Planning and Standards, U.S. Environmental Protection
Agency
-------
80.00 -
60.00
40.00 -
op
£ 20.00
-40.00
| Winds and Temperature
Winds
A PIBAL
O Receptors
+ Reiease
0o
°°
°5S
-40.00 -20.00
0.00
20.00 40.00
East-West (km)
80.00 100.00
Figure 1. The Idaho tracer experiment sampling arcs and meteorological data collection
network. The sampling arcs at 48 km and 90 km are shown. The receptor arc at 3.2 km
downwind of the release is omitted for clarity.
better than 0.5 km. The receptor arcs at 48 and 90 km downwind from the release are shown
in Figure 1. Meteorological data were available from eleven sites providing hourly-averaged
winds; four sites providing hourly-averaged winds and temperatures, three sites providing
hourly pibal observations of winds aloft (CFA, MTV, DBS). Two of the pibal sites (CFA and
DBS) also provided hourly-averaged winds and temperatures. Hourly rawindsonde
observations were taken at about 600 m northwest of the release location. The meteorological
masts ranged in height above ground with two at 6.1 m, eleven at 15.2 m, three at 22.8 m, and
two at 30 m. The pibal observations taken at Billings, Montana (well past the farthest
sampling arc downwind) were not used in this investigation. The skies were clear of clouds
and no precipitation occurred during the experiment. The National Weather Service
observations taken at Pocatello, Idaho (approximately 75 km southeast of the release location)
were included to provide station pressure (required input for CALMET).
To estimate the effects of drainage flow on the near-surface wind field, gridded values of
land use and terrain heights are needed. The land use data are used as surrogates for typical
values of surface roughness, albedo, soil heat flux, anthropogenic heat flux and leaf area index.
These surface parameters are used in estimating the surface energy balance. For this analysis,
U.S. Geological Service land use and terrain height data were extracted from data bases
included in U.S. EPA (1996). The basic grid size for these data is approximately 900 m. They
were processed into a 20 by 20 grid with a grid resolution of 10 km. Default values, as defined
in U.S. EPA (1996), for the surface parameters to be associated with the land use data were
used. The southwest corner of this grid was approximately 50 km southwest of the release.
The area depicted in Figure 1 is fairly flat, but the terrain sharply increases in height to the west
-------
and north of the area depicted. The dominant landuse was rangeland; and the surface
roughness was estimated based on landuse to be on the order of 10 centimeters.
Hourly-averaged winds and temperature were available from midnight April 18 through
midnight April 19. To mitigate the effects of not having surface data beyond midnight of April
19, the surface meteorological tower data were duplicated to form two 24-hour periods, having
identical meteorology. The assumption being made is that conditions were steady-state. The
pibal and rawindsonde data, which were available from 0700 MST to 1900 MST, were treated
in a similar manner.
CALMET assumes all upper-air observations are from rawindsondes, and thus expects
upper-air observations to provide winds, dry-bulb temperature and pressure with height.
CALMET interpolates in height for missing data values at intermediate heights in an obser-
vation; but CALMET will not extrapolate upper air data. Thus observations are rejected that
fail to reach the user-prescribed top of the modeling domain (3300 m for this analysis), or have
missing data values at the surface. To make use of the hourly pibal observed winds, temper-
ature and pressure values were added by linearly interpolating in time and height between
available rawindsonde observations, which were available every 1 to 3 hours. The pibal wind
directions were consistent with those from the one rawindsonde, but the wind speeds were
generally less in magnitude.
The CALMET wind field module is based on the Diagnostic Wind Model (DWM),
(Douglas and Kessler, 1988). A two-step procedure is involved in the computation of the
gridded wind fields. In the first step, an initial guess field is adjusted for kinematic effects of
terrain, slope flows, terrain blocking effects, and three-dimensional divergence minimization.
In this analysis, the initial guess wind field varies spatially from the available upper air obser-
vations using a 1/r2 weighting, where r is the distance from the observation to the grid point.
The second step includes four substeps: inverse distance interpolation of observations into the
Step 1 field, smoothing to reduce sharp gradients in the field, adjustments of vertical velocities
using the O'Brien procedure (O'Brien, 1970), and divergence minimization. In this analysis
the O'Brien procedure was not used, hence the vertical velocities were not constrained to be
zero at the top of the computation grid.
A purpose of this investigation was to assess the effects of having different amounts of
meteorological data for use in the development of the time varying field of meteorological
data. For this purpose four separate runs were made: Case 1 using all available upper-air and
surface mast observations, Case 2 using all surface mast observations but only the one onsite
rawind-sonde upper-air observation, Case 3 using only the CFA wind and temperature
observations with the one onsite rawindsonde upper-air observations, and Case 4 using only
the CFA wind and temperature observations with all upper-air observations. In Cases 1 and
2, all the onsite hourly wind and temperature data are employed but different amounts of
upper-air observations are used. In Cases 3 and 4, hourly winds and temperatures taken close
to the release are used with different amounts of upper-air observations. For all the CALMET
simulations, winds and temperatures were computed for six layers in the vertical, the midpoints
of which were: 10 m, 35 m, 75 m, 300 m, 1250 m, and 2650 m.
Figures 2a and 2b illustrate the major differences to be seen in the lowest-level wind speed
and mixing heights, for the grid square containing the release. The winds at CFA were higher
than those generally seen throughout the network. Hence in Cases 1 and 2 when all the onsite
winds were employed (open and closed circles in the Figure 2), the low-level winds were lower
than when only CFA data were used. In Cases 1 and 2, the afternoon stability was Pasquill
-------
category B/C (Monin Obukhov lengths of order -30 m). As a consequence of higher winds in
Cases 3 and 4, the surface friction velocities were higher, and the Monin Obukhov lengths
were larger (in magnitude), thus closer to neutral stability. The afternoon mixing heights,
shown in Figure 2b, are similar regardless of data used. This results because the "upper-air"
temperatures all have a common source, namely the rawindsonde observations taken 600 m
northwest of the release. The nighttime mixing heights are mostly a function of the magnitude
of the friction velocity. Hence, where estimated friction velocities were largest and differ most
among the various processing methods, differences were to be seen in the nighttime mixing
height values (see hours 1800 to 2300 MST in Figure 2b).
DISPERSION MODEL CALCULATIONS
Each of the four analyses of meteorology was used to produce two CALPUFF simulations
of ground-level concentrations for each of the three sampling arcs. In the first simulation, the
dispersion was described using Pasquill-Gifford dispersion parameters. In the second
simulation, the dispersion was described using dispersion parameters suggested by Draxler
(1976), which require values of the standard deviation of the vertical and lateral wind
fluctuations (referred to hereafter as "similarity dispersion"). The wind fluctuation standard
deviations estimated within CALMET are primarily dependent on the surface friction velocity.
The surface friction velocity is a strong function of stability (largest during unstable
conditions), roughness length and wind speed (increases as roughness length or wind speed
increase). The CALPUFF user's guide (EPA, 1995b) provides a complete listing of the
various equations, which is not possible to provide in this limited discussion.
CALPUFF options were set as follows: the maximum puff travel distance during one
sampling step (controls the puff generation rate) was set to 1 km, maximum puff separation
was set to 1 km, Gaussian vertical distribution was assumed, concentrations were determined
as if over flat terrain, no wet or dry deposition, and no transition to Heffter long-range
dispersion parameters was made.
CALPUFF internally computes for each sampling step, a transport wind averaged over the
depth of the puff from the multi-layer winds provided to it from CALMET. As a surface
release puff grows in the vertical, the depth through which the wind is averaged increases.
The SF6 tracer emission was reported to be steady at 25.37 gs"1 over the three hour period, and
was simulated within CALPUFF as a 3-hour point-source release at 10-m starting at 1300
MST. The release height was set at the midpoint of the lowest CALMET layer, to insure that
the internally computed standard deviations of lateral and vertical velocity fluctuations (for use
in the similarity dispersion parameter characterizations) at the specified release height, were
in accord with the wind speed used by CALPUFF for the lowest layer.
MEASUREMENT UNCERTAINTIES
The primary sampler used in this experiment was a bag sampler, consisting of a 50-liter
Saran bag enclosed in a plastic barrel. The bag was inflated with a small battery-powered
pump, which was turned on and off manually In general, 6-hour samples were available with
approximate start times of 1200 MST on the 3.2-km arc, 1300 MST on the 48-km arc, and
1500 MST on the 90-km arc. Two aliquot samples were taken from each bag, one into a 2-liter
bag and the other into a 1-liter steel cylinder. Of the 68 valid SF6 concentration values
-------
reported, 48 were from cylinder samples. For the 17 occasions where SF6 concentrations from
both the bag and cylinder samples could be compared, the bag samples were consistently 37%
lower than the cylinder samples. No discussion was present in the data volume regarding these
differences between the bag and cylinder samples, but all plots were shown using the cylinder
samples. In light of this and the differences seen between the bag and cylinder samples, only
the cylinder concentrations were used in the following analyses. To estimate the sampling
uncertainty in the SF6 cylinder samples, comparisons where made with the three
perfluorocarbons which were released simultaneously at the same location with the SF6:
PDCH (C8F16), PDCB (C6F12), and PMCH (C7F14). Comparison of SF6 concentration values
with concentrations from cylinder samples from all three perfluorocarbons was possible at 33
sites along the three sampling arcs. The perfluorocarbon concentration values were found to
be consistently 14% greater than the SF6 concentrations. For SF6 concentrations greater than
20 ppt, the standard deviation of the percentage differences was 11%; and for SF6
concentration less than 20 ppt, the standard deviation of the percentage differences was 77%,
implying greater uncertainty for the lower concentration values. A background of 0.5 ppt was
subtracted from all SF6 concentration values as suggested in the data volume.
MODEL RESULT COMPARISONS
For each 6-hour period, the second moment (lateral dispersion, Sy) of SF6 concentration
values about its centroid position along the arc was computed. The crosswind integrated
concentration, CWIC, was computed by trapezoidal integration. By assuming the concen-
tration profile along the arc is Gaussian, the central maximum, Cmax, was computed as,
Cmax = CWIC/(^ Sy).
A goal of this investigation was to assess the sensitivity of the modeling results to
different treatments of processing the meteorology, as well as to assess the performance of
CALPUFF in characterizing dispersion for transport distances beyond 50 km. Results are
summarized in Table 1 for the different wind field and dispersion treatments. Figure 3 depicts
the observed SF6 concentrations with the simulation results where all the surface and upper-air
observations were used to generate the hourly wind fields. For the observed values, there were
from 14 to 17 receptors along each arc with valid data for analysis. For analysis of the
simulation results, receptors were spaced at each arc distance at 2 degree intervals, over the 90
degree sector northeast of the release location. The second moment, Sy, represents a measure
of the puff horizontal dispersion. For these 6-hour periods, the observed lateral dispersion
ranged from roughly 22% to 15% of the travel distance downwind. The crosswind integrated
concentration values characterizes the amount of pollutant mass seen at the surface. From
Figure 2, assuming a mixed layer wind speed of 4 m/s, a mixed layer depth of 2500 m, a
sample duration of 6 hours, we would anticipate CWIC values of approximately 2xl05 ppt-m,
if the puff was well mixed. The observed CWIC values at 3.2-km and 90-km arcs are close
to 2xl05 ppt-m. The CWIC value at 48 km, shown in Table 1 as 4x105 ppt-m, would be 2x105
ppt-m, if we assumed the observed concentrations beyond receptor 425 rapidly approached
zero. There are indications of such a falloff in concentration in the bag sampling results, but
only the cylinder results shown in Figure 3 where analyzed in developing Table 1.
As shown in Figure 3 (which is typical for all of the simulations), the simulated transport
was somewhat south of the observed position along the first two arcs. It is also apparent that
the concentrations simulated for the first arc are at least a factor of 5 higher than observed.
-------
Table 1. Summary of observed and estimated centerline maximum concentration, Cmax,
lateral dispersion, Sy, and crosswind integrated concentration values, CWIC.
Cmax ( ppt)
Observed
All SFC +
All SFC +
1 SFC +
1 SFC +
A11UA
1UA
1UA
A11UA
Sy (km)
Observed
All SFC +
All SFC +
1 SFC +
1 SFC +
A11UA
1UA
1UA
A11UA
CWIC(ppt-mxl05)
Observed
All SFC +
All SFC +
1 SFC +
1 SFC +
A11UA
1UA
1UA
A11UA
Pasquill Dispersion
3.2km
103
793
808
1903
1712
3.2km
0.70
0.46
0.46
0.30
0.37
3.2km
1.82
9.17
9.21
14.22
15.89
48km
16.6
11.4
9.8
25.2
27.2
48km
9.66
5.22
5.15
4.79
4.67
48km
4.02
1.50
1.27
3.03
3.18
90km
6.4
4.0
2.5
12.5
12.0
90km
13.06
18.93
13.66
5.76
9.21
90km
2.09
1.88
0.87
1.81
2.78
Similarity Dispersion
3.2km
103
945
955
896
854
3.2km
0.70
0.66
0.66
0.55
0.60
3.2km
1.82
15.63
15.67
12.31
12.91
48km
16.6
21.0
20.0
10.0
17.1
48km
9.66
5.31
5.27
4.38
4.35
48km
4.02
2.79
2.58
1.10
1.86
90km
6.4
9.4
5.4
5.9
8.6
90km
13.06
12.27
8.43
5.38
9.31
90km
2.09
2.88
1.15
0.79
2.01
There is no obvious reason to dismiss the 3.2-km observations, as the other tracers at this arc
were within 11% of those reported for the SF6 when adjusted for differences in release rates.
The differences seen in the different treatments to develop the wind fields can largely be
explained by the increase in the lower level winds by a factor of 2 when only the winds at CFA
are used to develop the wind fields. As the transport winds increase, not only was the transport
to the arc decreased, but also the variability in the wind directions within the wind field were
greatly reduced. The transport time to the 3.2 and 48 km arcs varied from 1.5 to 3 hours,
depending on meteorology used. With these transport times and 10-km meteorology, the
lateral dispersion is relatively insensitive to the meteorology used for the 3.2 and 48 km arcs.
The CWIC values vary inversely to the product of the simulated vertical dispersion and the
-------
mean transport speed. For the Pasquill results, an increase in wind speeds results in neutral
stability, which caused the vertical dispersion to be nearly a factor of 4 less. This more than
compensated for the increased dilution, hence higher CWIC values. For similarity dispersion,
the vertical dispersion was never well-mixed and was nearly the same, regardless of
meteorology used. Beyond 75 km, an increase in transport winds tended to increase the
simulated vertical dispersion by 30% or so, which tended to reduce CWIC values for the 90-
km arc.
20.00 40.00 60.00 80.00
Arc Azimuth (degrees)
^ I ' I ' I ' I
0.00 20.00 40.00 60.00 80.00
Arc Azimuth (degrees)
~l'I^T
0.00 20.00 40.00 60.00 80.00
Arc Azimuth (degrees)
I " ' T
16.01
Hour of Day (LSI)
Figure 3. Six-hour average SF6 concentration values observed and estimated for
April 19, 1977, (A) 3.2-km arc, 1300-1900 MST; (B) 48-km arc, 1400-2000 MST; (C)
90-km arc, 1600-2200 MST. Azimuth is defined as viewed from the release position
with 0 due North and 90 due East (see Figure 1). Receptor numbers are shown just
above each observed concentration value. (D) Time history of observed PDCH and
estimated SF6 concentrations along the 48-km arc for April 19, 1977. Observed
PDCH values were multiplied by 3.16 for comparison with estimated SF6 values
(volume of SF6 divided by volume of PDCH released equals 3.16).
-------
Figure 3d provides a comparison of the time history of the puff, as it passed by the 48-
km arc. Sampling results are shown for the two-trap sampler which provided 5-minute
samples, and a cassette sampler which provided approximately 15-minute samples. These
samplers were quite close to the observed position of the 6-hour SF6 maximum along this arc.
The dispersion results are for the simulated position of the maximum, which was somewhat
displaced from that observed. The Pasquill dispersion results are in remarkable accord with
the tracer results. The similarity results arrive and depart slightly later than observed. The
slower transport for the similarity dispersion results because the vertical dispersion was less
than that simulated by Pasquill dispersion, hence the transport speed was computed over a
more shallow layer for the similarity results. These results and those discussed above suggest
that the similarity dispersion was underestimating the vertical dispersion for this case.
CONCLUSIONS
A goal of this investigation was to assess whether the CALPUFF simulations were in
reasonable accord with the observed concentrations, and the sensitivity of the simulation
results to different methods of processing the available meteorological observations. The
comparison results presented reveal as yet unexplained differences for the nearest arc, 3.2 km
downwind from the release. Possible speculations are that the puff became well-mixed sooner
than we would otherwise expect, or that the puff lifted somewhat off the surface at the 3.2-km
arc. The simulated pattern of dispersion was displaced as much as 40 degrees from that
observed, regardless of how the wind fields were characterized. For all arcs, the lateral
dispersion along the sampling arcs was best characterized by both dispersion characterizations
when all the surface tower winds were used. Except for the first sampling arc, the simulated
maxima along the arcs were typically within a factor of 2 of that observed. The Pasquill
simulations were most sensitive to how the wind fields were characterized, showing the most
variability between the various wind field results. Having only one puff release limits
conclusions to be reached. For this one realization, it would appear that simulations by both
dispersion characterizations were in best accord overall with observations when all the low-
level winds and upper-air observations were used. And for this case, the similarity dispersion
simulations may have underestimated the vertical dispersion.
DISCLAIMER
The information in this document has been funded wholly or in part by the United States
Environmental Protection Agency under an Interagency Agreement (DW13937039-01-0) to
NOAA. It has been reviewed in accordance with the Agency's peer and administrative review
policies for approval for presentation and publication. Mention of trade names or commercial
products does not constitute endorsement or recommendation for use.
REFERENCES
Clements, W.E (ed.)., 1979: Experimental Design and Data of the April 1977 Multitracer
Atmospheric Experiment at the Idaho National Engineering Laboratory, Los Alamos
Scientific Laboratory Informal Report, LA-7795-MS/UC-11, Los Alamos, NM 87545,
U.S.A., 100pp.
-------
Douglas, S., and R. Kessler, 1988: User's guide to the diagnostic wind field model
(Version 1.0). Systems Applications, Inc., San Rafael, CA, 48 pp.
Draxler, R.R., 1979: Modeling the results of two recent mesoscale dispersion experiments.
Atmos. Environ., 13:1523-1533.
Draxler, R.R., 1976: Determination of atmospheric diffusion parameters. Atmos. Environ.,
10: 99-105.
O'Brien, J.J., 1970: A note on the vertical structure of the eddy exchange coefficient in the
planetary boundary layer. J. Atmos. Sci., 27:1213-1215.
U.S. Environmental Protection Agency, 1995a: A User's Guide for the CALMET
Meteorological Model. EPA-454/B-95-002, Office of Air Quality Planning and
Standards, Research Triangle Park, NC 273 pp.
U.S. Environmental Protection Agency, 1995b: A User's Guide for the CALPUFF
Dispersion Model. EPA-454/B-95-006, Office of Air Quality Planning and Standards,
Research Triangle Park, NC 338 pp.
U.S. Environmental Protection Agency, 1996: CALMET, CALPUFF and CALPOST
Modeling System, PB 96-502-083INC (CD ROM and diskette), National Technical
Information Service, U.S. Department of Commerce, Springfield, VA, 22161.
-------
APPENDIX B
COMPACT DISCS
Solar and Meteorological Surface Observation Network (SAMSON), 1961 -1990
Version 1.0, September 1993
Available from U.S. Department of Commerce, National Climatic Data Center,
Federal Building, 151 Patton Avenue, Asheville, NC 28801
Radiosonde Data of North America, 1946 - 1992
Version 1.0, August 1993
Available from U.S. Department of Commerce, National Climatic Data Center,
Federal Building, 151 Patton Avenue, Asheville, NC 28801
CALMET, CALPUFF, and CALPOST Modeling System
Version 1.0
Available from U.S. Department of Commerce, National Technical Information
Service, 5285 Port Royal Rd. Springfield, VA 22161. NTIS PB 96-502 083.
-------
TECHNICAL REPORT DATA
(Please read Instructions on reverse before completing)
1. REPORT NO.
EPA-454/R-98-009
3. RECIPIENT'S ACCESSION NO.
4. TITLE AND SUBTITLE
5. REPORT DATE
June 1998
A Comparison of CALPUFF Modeling Results to Two Tracer
Field Experiments
6. PERFORMING ORGANIZATION CODE
7. AUTHOR(S)
8. PERFORMING ORGANIZATION REPORT NO.
9. PERFORMING ORGANIZATION NAME AND ADDRESS
Pacific Environmental Services, Inc.
5001 South Miami Boulevard
P.O. Box 12077
Research Triangle Park, NC 27709-2077
10. PROGRAM ELEMENT NO.
11. CONTRACT/GRANT NO.
68D30032
12. SPONSORING AGENCY NAME AND ADDRESS
13. TYPE OF REPORT AND PERIOD COVERED
U.S. Environmental Protection Agency
Office of Air Quality Planning and Standards
Emissions, Monitoring, and Analysis Division
Research Triangle Park, NC 27711
14. SPONSORING AGENCY CODE
15. SUPPLEMENTARY NOTES
16. ABSTRACT
The performance of the CALPUFF atmospheric dispersion model for two field tracer experiments is
summarized. The first tracer experiment was in 1975 at Savannah River Laboratory and the second was in
1980 in the central United States. Both experiments examined long-range transport of an inert tracer
material. The results generally were encouraging, with the simulated results within a factor of two of the
observed data for the statistical measures presented in the report. However, there is not a consistent pattern
of over- or under-estimation relative to the observations.
17.
KEY WORDS AND DOCUMENT ANALYSIS
DESCRIPTORS
b. IDENTIFIERS/OPEN ENDED TERMS
c. COSATI Field/Group
CALPUFF
Regional Modeling
Air Dispersion Models
Tracer Experiments
Model Performance
Long-Range Transport
18. DISTRIBUTION STATEMENT
Release Unlimited
19. SECURITY CLASS (Report)
Unclassified
21. NO. OF PAGES
51
20. SECURITY CLASS (Page)
Unclassified
22. PRICE
EPA Form 2220-1 (Rev. 4-77) PREVIOUS EDITION IS OBSOLETE
------- |