EVALUATION AND SENSITIVITY ANALYSES RESULTS
OF THE MESOPUFF II MODEL WITH CAPTEX MEASUREMENTS
by
James M. Godowitch
Atmospheric Sciences Modeling Division
Atmospheric Research and Exposure Assessment Laboratory
Research Triangle Park, North Carolina 27711
ATMOSPHERIC RESEARCH AND EXPOSURE ASSESSMENT LABORATORY
OFFICE OF RESEARCH AND DEVELOPMENT
U.S. ENVIRONMENTAL PROTECTION AGENCY
RESEARCH TRIANGLE PARK, NORTH CAROLINA 27711
-------
EVALUATION AND SENSITIVITY ANALYSES RESULTS
OF THE MESOPUFF II MODEL WITH CAPTEX MEASUREMENTS
by
James M. Godowitch
Atmospheric Sciences Modeling Division
Atmospheric Research and Exposure Assessment Laboratory
Research Triangle Park, North Carolina 27711
ATMOSPHERIC RESEARCH AND EXPOSURE ASSESSMENT LABORATORY
OFFICE OF RESEARCH AND DEVELOPMENT
U.S. ENVIRONMENTAL PROTECTION AGENCY
RESEARCH TRIANGLE PARK, NORTH CAROLINA 27711
-------
DISCLAIMER
The information in this document has been funded wholly or in part
by the United States Environmental Protection Agency. It has been
subject to the Agency's peer and administrative reviews, and has been
approved for publication as an EPA document.
AFFILIATION
The author is on assignment to the Atmospheric Sciences Modeling
Division, Atmospheric Research and Exposure Assessment Laboratory, from
the National Oceanic and Atmospheric Administration, U.S. Department
of Commerce.
11
-------
ABSTRACT
The MESOPUFF II regional Lagrangian puff model has been evaluated
and tested with the Cross-Appalachian Tracer Experiment (CAPTEX) data
base. The model was applied to the six full-scale CAPTEX episodes in
order to simulate the tracer plume formed from the 3-hour release of an
inert, non-depositing perfluorocarbon tracer gas from either one of two
selected sites. Surface tracer measurements were made at 3 hour or 6 hour
intervals over an extensive sampling network of arcs downwind to 1100 km
in the northeastern U. S. and southeastern Canada.
An operational evaluation was conducted by exercising the model
transport and dispersion components with all default features set according
to the user's guide. Transport in the model is simulated by a mixed-
layer averaged wind field and an upper level wind field averaged from
the mixing height to 700 mb level. Puff growth within 100 km downwind
is governed by horizontal and vertical dispersion parameters derived
from expressions fitted to the standard PGT curves, while time dependent
formulas are applied at greater distances.
Model performance was quantitatively determined from standard stat-
istical measures of difference and correlation between modeled and obs-
erved tracer concentrations paired in time and location. The model over-
predicted peak and mean concentrations, whether paired in time and/or
location. Modeled mean concentrations were within a factor of two of
observed mean values in four of the experiments. The overpredictions
are primarily attributed to an underestimation of vertical dispersion
during neutral stability conditions which were specified during the after-
noon hours after releases. Graphical maps and comparative tests between
observed and modeled concentration pairs revealed that spatial displacements
between the observed and modeled plumes contributed to the large scatter
and low correlations found in the evaluation statistics. Analysis of plume
centroid positions revealed that the greatest changes in the separation
of the observed and modeled plumes occurred during the nocturnal periods.
Revisions are suggested to better treat this source type in the model.
Diagnostic model test results with optional wind fields in the model
(e.g. surface, 850 mb single level wind fields) and various dispersion
options are compared to those from the operational model and the observed
data. A surface wind field transported the modeled plume considerably
slower and generally to the left of the observed plume while the single
level 850 mb wind field consistently shifted the modeled plume to the right
(clockwise) of the actual plume location. Model results with the mixed-
layer averaged wind field displayed the best overall performance in comp-
arisons of the position and time of impact of the peak concentration with
the observed plume values at the 300 km arc.
The 24-h peak and plume average concentration results from model test
runs, which focused on different options and variations in key parameters
of the the dry deposition and chemical transformation modules, are compared
to base case values. All model runs were performed with SOX emissions
from a realistic elevated point source. Peak SO? concentrations showed
negligible variations when deposition or chemical transformation were omitted
however, mean SC>2 concentrations in the plume were more sensitive. Mean
and peak sulfate concentrations were more sensitive than SC^ by variations
in the surface resistance and by changes in parameters controlling the
chemical transformation rate.
iii
-------
CONTENTS
Abstract iii
Figures vi
Tables viii
Acknowledgements x
1. INTRODUCTION 1
2. DESCRIPTION OF MODEL COMPONENTS AND TECHNICAL FEATURES 4
Wind Fields and Related Meteorological Parameters 7
Dispersion Parameters 10
Dry Deposition and Chemical Transformation Methods 10
Model Grid System 12
3 . MODEL EVALUATION DATA BASE 14
Description of the Field Measurement Program 14
Data Preparation for the Meteorological Processor 18
4. MODEL EVALUATION PROCEDURES AND RESULTS 30
Operational Model Evaluation Analyses and Results 33
Statistical Results for Mean and Peak Concentrations ... 33
Comparative Analyses of Plume Patterns and Positions ... 54
Diagnostic Model Test Results 64
5. MODEL SENSITIVITY RESULTS 82
6. SUMMARY AND CONCLUSIONS 93
REFERENCES 96
APPENDIX A Model test results for different
puff release/sampling rates 99
-------
LIST OF FIGURES
Number £^&
1 Flow diagram of the MESOPUFF II modeling system - 5
2 Surface sampling network for the CAPTEX field program 16
3 Meteorological grid in MESOPAC II for the Dayton cases
of the CAPTEX field program 19
4 Closest rawinsonde site (1 6) to each grid point in
the MESOPAC II meteorological grid for Dayton cases 22
5 Surface meteorological stations considered in the
MESOPAC II runs for the CAPTEX evaluation. Boundaries
for meteorological grids are denoted (heavy solid and
dashed lines) 23
6 Closest surface meteorological station (1 25) to
each grid point in the MESOPAC II meteorological grid
for Dayton cases during CAPTEX 25
7 Land use categories in the MESOPAC II meteorological
grid for the Dayton cases during CAPTEX 29
8 Average concentrations obtained from sites with observed
and modeled nonzero concentrations over the sampling
periods -during a) CAPTEX #1 and b) CAPTEX #2 41
Same as lla-b, except for c) CAPTEX #3 and d) CAPTEX #4 42
Same as lla-b, except for e) CAPTEX #5 and f) CAPTEX #7 43
9 Cumulative frequency distributions of the model and
observed concentrations from the entire data set 44
10 Cumulative frequency distributions of model and observed
concentrations for a) CAPTEX #1 and b) CAPTEX #2 46
Same as 13a-b. except for c) CAPTEX #3 and d) CAPTEX #4 47
Same as 13a-b, except for e) CAPTEX #5 and f) CAPTEX #7 48
11 Observed (0) and modeled (*) plume patterns depicted
from sites with nonzero concentrations during;
a) CAPTEX #1 and b) CAPTEX #2 56
Same as 8a-b, except for c) CAPTEX #3 and d) CAPTEX #4 57
Same as 8a-b, except for e) CAPTEX #5 and f) CAPTEX #7 58
VI
-------
Number Figures (continued) Page
12 Ratios of downwind distances of the model and observed
plume centroid locations (Dp/DQ) and separation
distance between centroid positions to the observed
downwind distance (DS/DQ) versus time of day starting
from the day of release. Negative and positive values
for DS/D0 denote the model plume position is to the
right and left of the observed plume, respectively.
a) CAPTEX #1 and b) CAPTEX #2 results 61
Same as 9a-b, except for c) CAPTEX #3 and d) CAPTEX #4 62
Same as 9a-b, except for e) CAPTEX #5 and f) CAPTEX #7 63
13 a) Difference between the downwind distances of the
centroid locations of the model (D ) and observed (DQ)
plume patterns versus time after release.
b) Separation distance (D£) between model and observed
plume centroid locations versus time after release.
Symbols in both figures denote events for each experiment;
CAPTEX #1 (0), #2 (X), #3 (triangles), #4 (squares),
#5 (stars) , and #7 (diamonds) 65
14 Upper air wind speed (0) and wind direction (X) profiles
at a) 18 GMT (i.e. 13 LST) and b) 00 GMT for CAPTEX #2 72
15 a) Wind speed (0) and wind direction (X) profiles and
b) temperature profile at Dayton, Ohio from CAPTEX #3 74
16 Tracer concentration pattern from aircraft flight legs
at different levels during the afternoon of October 3, 1983
during CAPTEX #3 75
17 Hourly variation of mixing height (Z-j) from the
MESOPAC II meteorological processor for Dayton, Ohio
during September 18-19, 1983 of CAPTEX #1 81
18 Average 24-h S02 concentration fields from the model
base case run for a) Day 1 and b) Day 2. All values
in units of pg m"J 85-86
19 Average 24-h sulfate concentration fields from the model
base case for a) Day 1 and b) Day 2. All values
in units of /ig m 87-88
VII
-------
Number
LIST OF TABLES
Page
1 Calculation sequence of meteorological parameters
in MESOPAC II
2 Computation sequence of the MESOPUFF II model code ...............
3 Tracer release information for CAPTEX ..................... ..... 15
4 Tracer sampling periods during each CAPTEX field experiment ...... 15
5 Upper air measurements and specifications ........................ 21
6 Surface meteorological station information ....................... 24
7 Hourly surface observations and specifications .................. 26
8 Land use categories and surface roughness lengths
for MESOPAC II ............................................... 28
9 Input control parameters and default features of
MESOPAC II runs for CAPTEX episodes ........................... 31
10 Input parameters and specifications for MESOPUFF II runs ......... 32
11 MESOPUFF II model evaluation results: observed and
modeled concentrations paired in time and location ............ 35
12 Test results of modeled and observed concentrations
paired in time and location ................................... 37
13 Selected statistical evaluation results for each event
of the CAPTEX experiments ..................................... 39
14 Distribution of observed and modeled concentrations
over the entire CAPTEX data set .................... . .......... 49
15 Peak observed and model concentrations
paired in location only ....................................... 50
16 Peak observed and modeled concentrations paired in time only
for the events of each CAPTEX experiments ..................... 52
17 High-25 observed and modeled concentrations unpaired
in time and location for each CAPTEX experiment ............... 53
18 Statistical results of observed and modeled concentrations
paired in time and location from sampling arcs
for CAPTEX #1-4 ............................................... 55
19 Peak observed and modeled concentrations unpaired in time
and location for each arc from CAPTEX #1-4 ................ ..... 55
20 Comparative results for different wind fields on
peak concentration/plume transport
at the 300 km arc for CAPTEX #1 ................................ 67
viii
-------
Tables (continued)
Number Page
21 Comparative results for different wind fields on
peak concentration/plume transport
at the 300 km arc for CAPTEX #2 68
22 Comparative results for different wind fields on
at the 300 km arc for CAPTEX #3 69
23 Comparative results for different wind fields on
at the 300 km arc for CAPTEX #4 70
24 Selected meteorological observations and model results
at the release site for each CAPTEX experiment 71
25 Results of analysis of observed upper air wind profiles
for the CAPTEX release days 77
26 Peak concentrations for CAPTEX #1 using
various dispersion options 79
27 Comparison of observed and modeled mixing heights 79
28 Source characteristics and emission rates for all model
sensitivity test runs 83
29 Results of model sensitivity runs: Comparisons with
24-h base case peak and mean concentrations 89
30 Results of sensitivity runs from variations in
select key parameters in the model 91
Al Differences in peak values at the initial arc
for different puff release/sampling rates 99
IX
-------
ACKNOWLEDGEMENTS
The author wishes to extend his appreciation to Terry Clark, Norman Possiel,
and Jerome Heffter for their thorough reviews and helpful suggestions which were
incorporated into this final report. Thanks are also expressed to Terry Clark
for providing a copy of the CAPTEX data base and for the useful discussions
concerning regional model evaluation methods. Appreciation is also expressed
to James Reagan for his assistance with the land use inventory.
-------
SECTION 1
INTRODUCTION
The ability of a regional-scale air quality model to reproduce
spatial pollutant concentration fields or deposition patterns on a short-
term basis is strongly dependent upon its formulations for simulating the
transport and dispersion processes. Models need to treat the important
diurnal variations occurring in the mean and turbulent structure of the
three-dimensional wind flow which directs the path of and impacts the
spread of pollutant plumes. Thus, an evaluation of these crucial model
components with field measurements is a critical element in establishing
the credibity of any model. Although a comprehensive evaluation of all
components in a model, including the chemical mechanism and removal methods,
is highly desireable, the data necessary to undertake such an effort have
been difficult to compile.
Experimental field studies with certain tracer gases have provided
valuable meteorological and concentration measurements on regional scales,
which allow evaluations to focus on the transport and dispersion components
of models. One of the most recent tracer data sets available for this
purpose is the Cross-Appalachian Tracer Experiment (CAPTEX). This field
program was was specifically designed and conducted with the intent of
acquiring an accurate tracer data set from an extensive surface sampling
network with concurrent upper air meteorological measurements for use in
transport and dispersion model evaluation efforts. During the CAPTEX
episodes, a chemically inert, non-depositing perfluorocarbon tracer gas
was released at ground-level over a 3-h period from either one of two
selected sites. Uncertainties often existing in emissions were minimized
as the tracer emission rate was highly controlled and concentrations were
accurately measured above a relatively low background level. There was also
no known interference from other sources to complicate the interpretation
of the tracer measurements. The CAPTEX field program was conducted in the
northeastern U.S. and southeastern Canada and the data base consists of a
challenging set of seven experiments with measurements spanning up to two
diurnal cycles in each episode.
The results from a few models which have been applied to particular cases
or to the entire CAPTEX data set have already been reported. Lee (1987)
applied a particle-in-cell numerical model to all CAPTEX cases and relied
heavily on various graphical analyses to investigate model performance.
Even though plots of modeled and observed plume patterns showed relatively
good qualitative agreement in most cases, a scatter diagram of modeled and
observed concentration pairs showed considerable variability and no corre-
lation. In fact, the large number of zero concentrations for either the
observed or predicted values indicated that notable spatial plume displace-
ments must have existed during the course of each experiement.
The ARL (Air Resources Laboratory) Lagrangian puff model, which distributes
mass vertically into different layers and transports the mass by different
trajectories at each level, was evaluated in several ways with the CAPTEX
data base by Draxler (1987). Several model runs were performed with various
-------
combinations of the meteorological surface and upper air data; some simulations
applied finer spatial resolution and/or a more frequent temporal interval
than the standard twice-daily profiles of the routine measurement network.
Results were not conclusive as to whether improved spatial or temporal reso-
lution in the input data produced more accurate model plume patterns. However,
the best results for the plume trajectory were indeed found when both spatial
and temporal resolutions were enhanced. Nevertheless, the predicted concen-
trations displayed order-of-magnitude differences from observed values when
pairs were compared even when using the most highly-resolved input data. Thus,
spatial plume displacements still were sufficient to produce large scatter and
low correlations in concentration pairs in time and location.
Davis et al. (1986) demonstrated the ability of a Lagrangian puff model
(MLAM), which allowed puffs to be vertically distributed and transported
within seven vertical layers and also simulated decoupling from the surface,
to explain the plume patterns in two CAPTEX episodes. While no statistical
results were provided, the graphical displays showed the ability of this
model to treat the considerable vertical direction shear present in one
of the CAPTEX cases simulated.
Most recently, a four-dimensional data assimilation technique incorporated
into a three dimensional hydrodynamic model was also applied to two CAPTEX
cases by Kao and Yamada (1988). who utilized it with a random-particle dif-
fusion model. Their graphical results of concentration pairs revealed notable
scatter. Differences between observed and simulated wind components, which
developed with time over the modeled plume history were mainly attributed
for the discrepancies in plume centerline positions. Spatial displacements
were found to be larger in the latter stages of each episode.
Clearly, the statistical and graphical analyses from these efforts revealed
the difficulty that current models have in successfully repli'cating the three-
dimensional wind flows and dispersion processes that characterize the observed
data, as exemplified by concentration differences and plume pattern displace-
ments between the models and observations.
In this report, the results of an evaluation and testing effort of the
MESOPUFF II model (Scire et al., 1984a,b) with the CAPTEX data base are
presented. MESOPUFF II is a second generation Lagrangian puff model which
was designed to treat transport, dispersion, chemical transformation, and
the removal (dry and wet) processes impacting pollutant emissions from
elevated point and/or area sources on a regional scale for short-term
applications. Although the chemical and deposition modules were specifically
developed for sulfur oxides and nitrogen oxides, any non-reacting and/or
non-depositing gaseous species may be modeled by electing not to simulate
these processes during model execution. Thus, no significant changes in
the mode code were needed for this evaluation effort.
The current operational version of MESOPUFF II (version 4.3) was applied
and evaluated with the measurements obtained from six CAPTEX episodes The
model can take advantage of two different layer-averaged wind fields derived
from surface and upper air data to transport a pollutant plume, while puff
dispersion is treated with traditional dispersion parameters. Section 2
describes the model system and provides the pertinent details about the
methods employed in the model to simulate these processes, while Sectio 3
-------
discusses the CAPTEX data base and preparation of the various input data
for the model runs.
Model performance was examined with various analyses techniques in
order to investigate the model's ability to predict concentrations and
plume patterns out to 1000 km from the tracer release sites. Thus, the
analysis tools included, but were not limited to traditional statistical
measures of difference and correlation between modeled and observed concen-
trations paired in time and/or location. While these statistical results
provide useful quantitative information, the results of previous regional
model evaluations with the CAPTEX data base have also shown the utility of
various graphical analysis and more innovative quantitative measures, which
helped clarify 'observed and modeled plume positions. Consequently, graphical
displays are used to demonstrate qualitative spatial agreement between
observed and simulated plumes, while analysis of plume centroid locations is
the method applied to obtain quantitative measures of differences in the down-
wind position and separation of the respective plumes. In Section 4, all of
these analysis techniques are employed in the evaluation process in order
to obtain the best overall picture of the capabilities and limitations of
the model for this particular application.
Uncertainty should be considered in the analysis of model predictions.
The uncertainty is caused by errors in the measurements, model formulation
and model input errors, as well as the inherent uncertainty associated with
the stochastic nature of turbulence and diffusion of a plume. The uncertainty
is not addressed directly within the evaluation analyses. However, any
inadequacy in the input data is pointed out where it may contribute to model
error. Particular attention in the analyses will focus on highlighting short-
comings in the model's simulation of the transport and dispersion processes.
Since the model evaluation with the CAPTEX data is limited to an assess-
ment of the transport and dispersion components of the model system, a select
group of model test runs was also undertaken to investigate the impact on
pollutant concentrations from optional methods in the model or to changes
to key technical parameters of the chemical transformation and dry deposition
mechanisms. The model test runs simulated the SOX emissions from a single,
elevated point source and apply the same meteorological fields from one of
the CAPTEX episodes. Section 5 discusses the results of model test runs
from comparisons of 24-h peak and average plume concentrations from each
test simulation to those of a base case run.
-------
SECTION 2
DESCRIPTION OF MODEL COMPONENTS AND TECHNICAL FEATURES
The principal program components of the model system and their
important input/output features are depicted in Figure 1. In particular,
separate computer codes perform the specific functions for processing
various meteorological data (READ56, MESOPAC II), for computing pollutant
concentrations (MESOPUFF II), and for postprocessing operations with
the calculated concentrations from receptor sites and/or grid points
(MESOFILE II). All of these model components were exercised during this
evaluation and testing effort. An overview of the attributes of the
model components is given herein and the essential details of the methods
used to compute the parameters relevant to the transport and dispersion
processes, and treatments of dry removal and chemical transformation are
also presented. This information provides some background to assist in
the interpretation of the evaluation results and sensitivity test runs.
Complete details about the various methods in the model programs can be
found in Scire et al. (1984a) and complete instructions for applying the
model system and optional features are contained in Scire et al. (1984b).
The READ56 processor program examines and reformats the twice-daily
(i.e. 00 and 12 GMT) rawinsonde sounding measurements, and creates a
formatted file of temperature and wind speed/direction profile data for
each site. This program screens the profile measurements to determine
if values at mandatory levels are missing. However, data quality assurance
tests and inserting missing values must be performed by the user.
MESOPAC II is the primary meteorological processor program. It
generates hourly gridded fields of meteorological parameters from hourly
surface meteorological observations, the upper air data files produced by
READ56, and land use category information. Precipitation measurements are
necessary when modeling wet episodes, otherwise these data are optional.
Precipitation measurements were not used in this model evaluation effort
as dry conditions existed during the CAPTEX episodes. Table 1 shows the
execution sequence of the MESOPAC II program code and the order of calcula-
tion of meteorological parameters. The processor program is highly modular
in that each parameter is computed within a separate subroutine. A single
output file of gridded meteorological parameters is produced by MESOPAC II,
which serves as the primary input file to the model program.
Both processor programs were specifically designed to accept as input
the various formatted meteorological data archived by the National Climatic
Data Center in Ashville, North Carolina. The National Weather Service (NWS)
upper air data (TDF5600 format) and hourly surface observations (CD1444
format) are the expected data sets for these programs, although comparable
observed data can be substituted if all the necessary variables have been
measured. Hourly measurements of ceiling height, cloud cover, and near-
surface observations of temperature, pressure, wind speed and direction
and relative humidity are required.
-------
/ Rawinsond* \
REA056 Control
Parameter Inputs
REAOSS Upper Air
Preprocessor Progra
MESOPAC II
Control Parameter
Inputs
r-ormaned Twice
Daily Rawmsonde
Data files
MESOPAC II Meteorological
Preprocessor Program
Hourly
Meteorological
Variables
MESOPUFFII
Control Parameter
Inputs
MESOPUFF II DISPERSION MODEL
Concentration
Tables
MESOF1LEII
Control Parameter
Inputs
MESOFILS II
Postprocessor Progra
Concentration
Tables
Figure 1. Flow diagram of the MESOPUFF II modeling system. (Scire et al., 1984a)
5
-------
TABLE 1
CALCULATION SEQUENCE OF METEOROLOGICAL PARAMETERS IN MESOPAC II
> READ USER INPUT FOR CONTROL AND TECHNICAL FEATURES
> READ UPPER AIR PROFILE DATA (00 AND 12 GMT FOR FIRST SIMULATION DAY)
< ITERATE OVER DAY(S)
* COMPUTE SUNRISE/SUNSET TIMES, HOURLY SOLAR ELEVATION ANGLES
> READ PRECIPITATION DATA (OPTIONAL)
< ITERATE OVER HOURS
> READ HOURLY SURFACE METEOROLOGICAL STATION OBSERVATIONS
> IF HOUR IS 00 GMT, READ NEXT 12 GMT UPPER AIR PROFILE
> IF HOUR IS 12 GMT, READ NEXT 00 GMT UPPER AIR PROFILE
* CALCULATE SURFACE WIND FIELD AT GRID POINTS
* CALCULATE PGT STABILITY CLASS AT GRID POINTS
* CALCULATE SENSIBLE HEAT FLUX (H) AT GRID POINTS
* CALCULATE FRICTION VELOCITY (U*) AT GRID POINTS
* CALCULATE MIXING HEIGHT (Z±) AT GRID POINTS
* CALCULATE OBUKHOV LENGTH (L) AT GRID POINTS
* CALCULATE CONVECTIVE VELOCITY (W*) AT GRID POINTS
* CALCULATE LOWER WIND FIELD AT EACH GRID POINT
* CALCULATE UPPER WIND FIELD AT EACH GRID POINT
* OUTPUT/STORE ALL GRIDDED METEOROLOGICAL PARAMETERS
<- END OF HOUR ITERATION
< END OF DAY ITERATION
* CLOSE FILES
* END PROGRAM EXECUTION
-------
MESOPUFF II computes hourly surface pollutant concentrations on a
Cartesian grid and/or at user-specified, nongridded receptor sites. The
model applies a Gaussian puff superposition technique to simulate a
continuous plume by a series of individual puffs. The puff release and
sampling rates, and source and emission information must be supplied by
the user. Each puff is individually transported and its mass is subjected
to dispersion, deposition (dry and/or wet) and chemical transformation.
Table 2 reveals the overall execution sequence of the MESOPUFF II code
and the order of simulation of the physical processes. Options exist
that permit one or more of the removal mechanisms to be omitted or for
input values to override default values of key technical parameters.
The MESOFILE II postprocessor package is capable of performing
various analyses of individual or multiple concentration files produced
by the model. Time-averaging of concentrations may be performed in an
evaluation effort or statistical results can be computed to investigate
variations in concentration fields produced by different model runs.
The latter function is particularly useful when conducting model
sensitivity studies in a comparison of different results to a base case.
WIND FIELDS AND RELATED METEOROLOGICAL PARAMETERS
Two different layer-averaged wind fields are generated by MESOPAC II
in the default mode. A mixed-layer averaged wind field is determined from
hourly surface observations and twice-daily upper air profile data in order
to represent the mean boundary layer flow pattern between the surface and
mixing height (Z^). An upper wind field is determined over the layer from
Z^ to the 700 mb height. The user may override these default fields by
selecting alternate wind fields according to particular input specifications
(see Scire et al., 1984b) . Optional wind fields include; a single level
wind field for surface, 850 mb, 700 mb, or 500 mb levels, or vertically
averaged winds from Z^ to 850 mb or to 500 mb. A single wind field model
is also possible by specifying the same wind field selection as the lower
and upper level wind fields.
The methodology to determine the layer-averaged wind fields consists
of several steps. Surface wind components (us, vg) are determined at
each grid point from the hourly surface station observations with the inverse
distance-squared (1/d ) technique. An alignment weighting factor is also
applied which gives greater consideration to measurements at sites more
directly upwind or downwind of a grid point. The 00 GMT and 12 GMT rawinsonde
profiles are assumed to be representative of neutral/unstable conditions and
stable conditions, respectively. With the appropriate rawinsonde profile,
layer-averaged wind speed and direction are computed from all measurements
between the surface and the current Z^. Then the ratio of the layer-averaged
and surface wind speeds (R^), and the angular difference between the layer-
averaged and surface wind directions (d6) at the rawinsonde site are determined
Finally, the mixed-layer averaged horizontal wind components (Ura^,Vm-j_) at a
grid point for a particular hour are computed by applying R^ and dG from the
closest rawinsonde site to the spatially-interpolated surface wind speed and
direction, respectively, at that grid point.
The layer-averaged upper wind field is determined by linear interpolation
in time between soundings and averaging all data over the layer from Z^
-------
TABLE 2
COMPUTATION SEQUENCE OF THE MESOPUFF II MODEL CODE
> READ INPUTS FOR TECHNICAL OPTIONS, GRID SIZE, SOURCES/EMISSIONS
> READ THE GRIDDED WIND FIELDS FOR INITIAL HOUR OF SIMULATION
< ITERATE OVER THE HOURS FOR THE SIMULATION
> READ OTHER GRIDDED METEOROLOGICAL PARAMETERS FOR CURRENT HOUR
> READ GRIDDED WIND FIELDS FOR NEXT HOUR
* INITIALIZE CONCENTRATION ARRAYS
< PERFORM PUFF ITERATIONS
* FOR NEW EMITTED PUFF - COMPUTE PLUME RISE
< PERFORM PUFF SAMPLING ITERATIONS
* TRANSPORT THE PUFF
* COMPUTE AND APPLY DISPERSION PARAMETERS TO PUFF, AND
TRANSFER MASS IF IN 3-LAYER MODE
* CALCULATE AND APPLY CHEMICAL TRANSFORMATION RATE(S)
* CALCULATE AND APPLY WET REMOVAL
* CALCULATE AND APPLY DRY DEPOSITION
* SAMPLE PUFF TO COMPUTE CONCENTRATION CONTRIBUTION
AT GRID POINTS/NONGRIDDED RECEPTOR LOCATIONS
< END PUFF SAMPLING ITERATIONS
< END PUFF ITERATIONS
* CALCULATE PUFF CONCENTRATIONS FOR NEXT HOUR'S CHEMISTRY
* CALCULATE MULTI-HOUR AVERAGE CONCENTRATIONS, IF SPECIFIED
* PRINT/STORE MODELED CONCENTRATIONS
< END HOUR ITERATIONS
* END PROGRAM EXECUTION
-------
to the 700 mb height. The upper level wind components (Uu]_, Vu-|_) are
derived at each grid point by the 1/d2 method with the summation of contri
butions over the rawinsonde sites.
The appropriate wind field for puff transport is automatically
selected by a comparison of the puff center height (Zp) and the current
Z^ on an hourly basis The source height and any plume rise are summed
to obtain Zp. For ZpZi, the upper wind field is used to transport the
puff and its growth is governed by stability class 5 (i.e. PGT class E by
default). Puff splitting is not performed by this model. However, when
the multi-layer model feature is exercised, puff mass can be transferred
between the mixed-layer and upper layer. For example, the amount of mass
removed from the mixed-layer depends on the fractional change of Z^ since
its peak value. This model feature is usually selected with the surface
depletion method for dry deposition in an attempt to account for the large
difference of boundary layer resistance to mixing at night compared to day-
light hours.
The mixing height has particularly important roles in this model
because it is involved in both transport, as just discussed, and in
dispersion processes. Since Z^ is the demarcation level defining the
vertical extent of the mixed-layer averaged wind field and the starting
height of averaging for the upper level wind field, variations in Z^
can lead to possibly notable differences in puff transport, especially
when strong vertical shears in speed and/or direction exist. As in many
models, Z^ is also the critical height scale in vertical dispersion
calculations. Thus, it is essential to model performance that the temporal
behavior and spatial variations of Z^ be simulated realistically.
During the daylight hours, convective and mechanical mixing heights
are computed with the greater of the two values being selected as Z^
at each grid point. The convective mixing height (Z^c) is strongly
controlled by the surface sensible heat flux (H = w'T') and the temperature
structure aloft. The behavior of H is highly dependent on the incoming
solar radiation and the observed cloud cover. Eqs. (1) and (2) are used
to compute Z^c and the temperature discontinuity (AT) at the mixed-layer
top, respectively, at the next hour (t+1) over an hourly time step (dt).
Zic = (Zic2 + 2H(l+E)dt/V 2ATtZic/^ + AT /V> (1)
t+1 t t t+1
AT = (2 E tf H dt)H (2)
t+1
The potential temperature lapse rate (V>) is determined over a 200 m layer
(default) above Z^c. An entrainment coefficient (E) is included in an
effort to parameterize an additional contribution to Z^c growth from
downward transport of warmer air with a default value for E of 0.15.
A daytime mechanical mixing height (Z^m) is computed from the
surface friction velocity (U*) and the Brunt-Vaisala frequency
according to;
9
-------
Zim - 2U/(fNb) (3)
where f is the Coriolis parameter. During nocturnal hours, the
mixing height is defined solely by Zim, which has an even stronger
dependency on U* according to the empirical expression in Eq. 4.
Zim - C U* (4)
The current default value for C is 2400.
The friction velocity is a function of the surface wind speed,
surface roughness length (ZQ) , and a stability adjustment factor
(Scire et al.,1984a). The land use type for each grid point should
be accurately specified because ZQ is assigned internally based on
representative values for the broad land use types of Sheih et al. (1979)
DISPERSION PARAMETERS
Each puff in the model is horizontally symmetric with a Gaussian
distribution. Puff growth is governed by the horizontal and vertical
dispersion parameters, a and CTZ . Values of these parameters are
computed by power law functions which were derived from fits of the
Turner (1970) curves. The form of these functions is given by;
a = A Xb (5)
where A and b vary with stability class and are different for a and CTZ
(Scire et al., 1984b). Eq. (5) is applied to puff growth out to downwind
distances (X) of 100 km (default). Thereafter, puff growth is determined
with the time-dependent formulas of Heffter (1965). The crossover distance
where the change in methods occurs may be overidden. An optional feature
in the model uniformly disperses puffs throughout the mixed-layer depth
immediately after being released.
The stability class for each grid point is determined each hour
according to the Turner (1964) method. Since these PGT stability
classes are defined according to cloud cover, ceiling height, solar'
elevation angle, and the surface wind speed, values can vary across the
model grid each hour. Stability classes range from 1 (A very unstable)
to 6 (F - moderately stable).
DRY DEPOSITION AND CHEMICAL TRANSFORMATION METHODS
Although the tracer gas which is the subject of this evaluation
effort is nondepositing and chemically inert, the methodologies that
simulate dry removal processes and determine the transformation rates in
MESOPUFF II are described because the results of the sensitivity analyses
in Section 5 are obtained from variations to the default methods.
10
-------
Both of these physical processes are highly complex in the atmosphere
and revisions to model formulations are often performed as new experimental
information becomes available. Methods to simulate these processes must be
species dependent. While MESOPUFF II is capable of simulating SOX and
NOX simultaneously, the information contained in this overview is limited
to this model's treatment of SC^ and a secondary product, fine sulfate (SO^) .
Dry deposition has received greater recognition over the past decade
as an important removal process of pollutants over regional and larger scales
Consequently, more resources have been devoted to obtaining experimental
dry deposition measurements of certain pollutants (Hales et al . , 1987),
which has provided more information to improve our understanding of depos-
ition processes and has advanced the development of deposition parameter-
ization methods. There has also been a notable evolution in the treatment
of dry deposition in models from the simple technique of specifying a
constant deposition velocity of a specific pollutant, as in the first
generation of MESOPUFF (Benkley and Bass, 1979) to a physically realistic
method which attempts to account for the spatial and temporal variations
in meteorological parameters and surface features that influence dry
removal processes. Currently, atmospheric and surface resistances are
commonly employed to compute spatially-averaged deposition velocities for
grid cells by consideration of the proportion of various land use types
within each cell (Walcek, et al. , 1986).
The dry deposition method in MESOPUFF II is also parameterized by a
deposition velocity (V^) . In this scheme the removal processes consist
of the vertical turbulent diffusion of pollutants through a turbulent
atmospheric surface layer and a laminar sublayer, which is followed by
uptake into the vegetative or surface material. Individual resistances
are used to represent the ability of a species to transfer through the
atmosphere and surface cover. The deposition velocity is computed as the
inverse of the sum of individual resistances according as follows;
vd - '
where r_ is the aerodynamic resistance, r^ is the quasi laminar
boundary layer resistance in the proximity of the canopy or surface
interface and r is the surface resistance. Values of ra and r^ are
temporally and spatially variable since both are dependent upon the
meteorological conditions and land use. These aerodymamic resistances
are computed from;
ra = (k U^)-1(ln(z/Z0) - Sh <6a)
rb -
(6b)
where Sh is a stability correction factor and B is a surface transfer
function. The value of kB"^ for S02 is 2.6. However, r^ is assumed
to have a constant value of 1000 s/m for SO^. Specifications for other
modeled pollutants are given in the user guide (Scire et al . , 1984b) .
11
-------
Surface resistances for SC>2 are a function of land use and stability
class. The default values are specified internally in the module for
summer conditions only and are the same as those in Sheih et al. (1979).
For SO^, the total resistance is determined by the sum of ra and rb
because rs is currently assigned to be zero in the module.
In the multi-layer mode of the model, only puff mass within the mixed
layer is subjected to dry deposition. In contrast, the source depletion
method is exercised when the single layer mode of the model is selected.
The treatment of chemical transformation in the atmosphere has also
shown great advances. There is a hierarchy of methods to simulate chemical
reactions and their rates that vary from the linear treatments with fixed
transformation rates to sophisticated chemical mechanisms containing
numerous reactions which require considerable computer resources.
Chemical conversion of S02 to SO^ in MESOPUFF II is controlled
by a transformation rate which is computed from a parameterized expression.
Scire et al. (1984a) described the photochemical model mechanism and
analyses of the model simulation results which lead to defining the
primary variables that dominated the transformation rates. Expressions
for transformation rates were developed from statistical regression
analyses of many photochemical model run results for a range of conditions.
In particular, the hourly transformation rate (k) for SC^ to sulfate
for the daytime period is computed from;
k = 36R0.55030.71S-1.29 + 3*10-8^ (7)
o
where R is total solar radiation (kwatt/m ), 63 is the background
ozone concentration (ppm), S is the stability class, and RH is relative
humidity (%). These parameters were found to be the dominant variables
from regression analysis of the photochemical model results. Thus,
chemical conversion of SC^ is enhanced in the model under conditions of
high solar radiation, strong dispersion, high background OT, and high
humidity. The default background 63 concentration in the model is
currently 80 ppb, although the model is capable of accepting measured
03 concentrations at sites within the model domain. Model run results
in Scire et al. (1984a) indicated values for k were 2-3% per hour during
the daytime, which were comparable to existing experimental results from
field studies (Wilson, 1981). During the nocturnal period, the SOo
oxidation rate is set to 0.2% h . Equations comparable to (7) have
also been developed for NOX transformation rates.
GRID SYSTEM
The grids for MESOPAC II and MESOPUFF II are based on a Cartesian
coordinate reference frame. Since it is not constructed on a latitude-
longitude framework, alignment differences between N-S grid lines and
the true N-S direction lines defined by longitude can become serious for
grid domains larger than a few hundred kilometers, especially at middle
and high latitude regions. An approach used to minimize the impact of
12
-------
this effect on the grid system for the CAPTEX region will be noted later.
A meteorological grid defines the set of grid points where the
meteorological parameters are computed by MESOPAC II. Currently, the
computer code limits the size to 40 x 40 grid indices. The computational
grid in MESOPUFF II can be defined to be the same size as the meteorological
grid or a subset of it. When a puff travels beyond the computational grid,
it is eliminated from further consideration in concentration calculations.
A sampling grid defines the set of gridded receptor points where concen-
trations are computed and this grid may be identical to or a subset of the
computational grid. The model is also capable of computing concentrations
at so-called nongridded receptor locations. This model feature can be
particularly useful in a model evaluation, such as in this application to
the CAPTEX tracer sampling sites, since measurement networks are not designed
according to a uniformly spaced grid. Model execution time is also reduced
significantly by restricting concentration calculations to the nongridded
receptor locations with an option to omit calculations for the sampling
grid. The grid size and grid point spacing must be defined by the user in
each application. Additional details and an example configuration of the
model grid systems are presented in Scire et al. (1984b).
13
-------
SECTION 3
MODEL EVALUATION DATA BASE
DESCRIPTION OF THE FIELD MEASUREMENT PROGRAM
The field study phase of CAPTEX consisted of seven individual tracer
releases from either Dayton, Ohio or Sudbury, Ontario over the period
from mid-September to late October 1983. Information about each tracer
experiment is given in Table 3. With the exception of experiment #6, which
was very short to permit the start up of the next release from Sudbury,
the amount of tracer gas emitted for these experiments was comparable. All
releases were from ground level and a constant emission rate is assumed
over a 3-hour period during each full-scale experiment.
The first four tracer releases were undertaken from Dayton and the
release periods were during afternoon hours (Table 3). In contrast, both
Sudbury releases were conducted at night. Nevertheless, the meteorological
conditions at both sites permitted strong vertical mixing and the prevailing
wind pattern in each case provided the desired transport of the tracer
plume across the study region. Dayton releases were performed when the
region was under the influence of large high pressure systems. At the
release site, wind flows were from the southwest or west and mostly fair
or partly cloudy sky conditions existed. The Sudbury releases occurred
after cold front passages with northwesterly winds and mostly cloudy skies.
Brown et al. (1985) provides detailed synoptic maps and discusses the
weather conditions for each CAPTEX episode.
The field study region during CAPTEX encompassed the northeastern
U. S. and southeastern Canada. The surface sampling network shown in
Figure 2 was designed to encompass this area. It consisted of an array of
individual arcs deployed at approximately 100 km intervals extending from
300 km to about 1100 km from of Dayton. Each sampling site was designated
by a 3-digit number (e.g. 302); the leading digit (3) signifying the arc
number in hundreds of kilometers and the next two digits (02) representing
a site's relative position along each arc with numbers increasing from
south to north.
The experimental measurement plan was to commence sampling concurrently
at all sites along a given arc prior to the expected arrival of the tracer
plume and to obtain measurements over the entire duration of plume passage.
Each site was equipped with an automatic sequential sampler that was preset
to collect tracer concentrations over a prescribed period. The tracer
sampling system is described by Ferber et al. (1986). Tracer sampling
began at the nearest arc soon after the start of a release, and at 3-h or
6-h intervals later at subsequent arcs. The sampling interval was set to
3-h at sites on the 300-km arc for Dayton releases and at some Canadian
sites after the Sudbury releases to provide better temporal resolution near
the release point. All other measurements were generally made over 6-h
intervals. A maximum of six consecutive samples were obtained at each site,
14
-------
TABLE 3
TRACER* RELEASE INFORMATION FOR CAPTEX
EXPT
(#)
1
2
3
4
5
6
7
SITE
DAYTON*
DAYTON
DAYTON
DAYTON
SUDBURY
DAYTON
SUDBURY
DATE
(1983)
9/18
9/25
10/2
10/14
10/25-6
10/28
10/29
TIME AMOUNT RELEASED RATE
(LST) (kg) (g/s)
1200-1500
1205-1505
1400-1700
1100-1400
2245-0145
1030-1100
0100-0400
208
201
201
199
180
32
183
19
18
18
18
16
2
16
.26
.61
.61
.44
.67
.96
.94
* Released from Wright-Patterson Air Force Base, otherwise at
Dayton International Airport (DAY).
# Tracer used was perfluoromonomethylcyclohexane (CyH-^)
TABLE 4
TRACER SAMPLING PERIOD FOR EACH CAPTEX FIELD
EXPT START SAMPLING
(#) (DATE-TIME (LST)
1 9/18-1400
2 9/25-1300
3 10/02-1600
4 10/14-1300
5 10/26-0300
6* 10/28-1100
7** 10/28-0100
STOP SAMPLING$
(DATE/TIME (LST)
9/20-0800
9/28-0400
10/04-1900
10/16-1000
10/28-0300
10/29-0800
10/30-0800
EXPERIMENT
DURATION
(hours)
42
63
51
45
48
21
33
$ Sampling end time for most sites up through the 1000 km arc.
* Canadian sites were not put in operation in anticipation of #7.
** One site continued sampling until 11-31/1500
15
-------
O O O O3I-1
320 3IQ 3I6 312
Figure 2. Surface sampling network for the CAPTEX field program.
-------
so measurements spanned either 18 or 36 hours. In most cases this sampling
strategy was successful since the arrival and departure times of the tracer
plume, and peak concentrations could be determined at each arc to within
the averaging interval of the measurements. Table 4 lists the start time
and most common ending time of sampling to reveal the overall duration of
each experiment. Measurements at a few sites on the most distant arcs
continued even later than those posted in Table 4. The sampling period for
all experiments, with the exception of #6, extended over at least one diurnal
cycle. Measurement periods for the four Dayton experiments covered episodes
of approximately two full days before the actual tracer plume departed the
sampling network.
The acquisition of upper air meteorological data on more frequent
temporal and finer spatial scales was also an important ingredient of the
CAPTEX field program. During the experimental periods the National Weather
Service rawinsonde sites in the CAPTEX region launched four rawinsondes
daily at 6-hour intervals, instead of the routine twice-daily soundings.
These supplemental soundings were made at 06 and 18 GMT. Even though the
meteorological processor was not designed to accept these profiles as
input, Z^ observations from these soundings may be compared to
modeled values and the intermediate wind profiles are also useful in
diagnosing and interpreting the causes of model variations. There were
10 additional upper air sites deployed in the region during this
experimental program to provide greater spatial density of the rather
sparse NWS rawinsonde network. Rawinsonde soundings were also performed
during the tracer release periods at both release sites.
Several research aircraft were specially-equipped with modified
tracer samplers for the CAPTEX studies. Sampling flights were conducted
to obtain the horizontal dimensions of the tracer plume at various levels
at different downwind distances and aircraft soundings were conducted to
investigate the vertical distribution of tracer. The aircraft flight paths
for each experiment and listings of the tracer concentrations measured
along each flight path are provided by Ferber et al. (1986). Since the
aircraft tracer measurements represented spatially-averaged concentrations
sampled over several minutes along a flight path at various heights above
ground, these measurements will not be directly used in the quantitative
statistical analysis of the model results. However, the aircraft
measurements provided valuable information about the tracer plume
position, and its spatial dimensions at various times and distances from
the release point. The aircraft data can be helpful in an examination
of modeled and observed plume locations and in qualitative assessments
of modeled and observed concentrations when possible. Results of analyses
of the horizontal and vertical distribution of the aircraft tracer data
have been presented by Raynor et al. (1984) and Stunder et al. (1986),
respectively.
The tracer measurements collected at each sampling site were subjected
to quality assurance audits and tests (Lagomarsino et al., 1987). Tracer
concentrations were provided in units of femtoliters per liter (fl I"1)
which is equivalent to parts per 10 . According to Ferber et al. (1986),
an ambient background concentration of 3.4 fl 1" was subtracted from all
tracer measurements. The final data base contained concentrations to the
nearest 1 fl 1 at each site with accompanying date/time (GMT) information,
17
-------
and sampling interval (3 or 6 hours). The data base contained no entries
for sampling periods when tracer concentrations were not collected at a site
or when measurements were rejected by quality assurance procedures. A
conversion factor of 1 fl 1 - 1.56x10 g m was applied to modeled
concentrations for their evaluation against observations.
DATA PREPARATION FOR THE METEOROLOGICAL PROCESSOR PROGRAM
Prior to the selection of surface stations and data preparation for
the model evaluation runs, a meteorological grid for MESOPAC II was
designed with sufficient size and adequate resolution to cover the CAPTEX
region. It was also decided to use different grid sizes for the Dayton
and Sudbury cases. The meteorological grid was set to 30 x 19 for the
Dayton cases and 30 x 24 for both Sudbury releases. However, the grid
point spacing of 37 km was the same for both grids. Therefore, the grid
domain covered an 1100 km x 700 km area for the Dayton cases and an
1100 km X 888 km region for the Sudbury cases. The selection of the grid
point spacing was partly based on the resolution of the land use inventory
which will be discussed later. This combination of grid size and the grid
point spacing provided sufficient coverage and spatial resolution of the
study region. Figure 3 shows the meteorological grid for the Dayton cases.
The release site is situated a sufficient distance from grid boundaries and
the domain extends eastward to include the 1000 km arc. The northern bound-
ary of the grid extended to 47°N for the Sudbury cases. These grid sizes
appeared to stretch the applicability of this model grid system to a maximum
limit due to the convergence of longitude lines with latitude. Differences
in alignment of grid and longitude lines, taking the overall domain into
consideration, were minimized by defining the N-S grid lines in the middle
of the model domain to be parallel with longitude lines. Thus, alignment
differences increased slowly toward the E-W boundaries.
A magnetic tape of the CAPTEX data base contained the rawinsonde
data, surface tracer concentrations, aircraft tracer data, ancillary
meterological measurements from several towers, and terrain heights
over the CAPTEX region according to the format specifications documented
in Ferber et al. (1986).
The rawinsonde sounding measurements at the release sites and the
regular NWS upper air network sites were processed for input to MESOPAC II.
Since a survey of the rawinsonde measurements from the 10 additional
sounding sites revealed that winds were consistently missing for many
levels within the boundary layer, these sites were not included in the
meteorological processor runs. The wind and temperature records for each
sounding were combined from separate blocks on the tape. An additional
complication was that the significant wind and temperature reporting
levels did not necessarily correspond, as temperatures were often detected
at fewer levels. Consequently, an upper air profile with temperatures and
winds at common levels up to 500 mb was constructed by deriving values by
linear interpolation at heights where the other parameter was measured.
The formatted upper air data files were reviewed to insure that soundings
existed at 00 and 12 GMT for all days in the simulation period required to
model each experiment. These files were also examined to make sure values
existed at all mandatory levels (i.e. 1000, 850, 500 mb) because the MESOPAC II
code ceases execution when an upper air profile or data are missing. Table 5
18
-------
N47.
SI.OO*
47.00*
* * K * *
Figure 3. Meteorological grid in MESOPAC II for the Dayton cases
of the CAPTEX field program.
-------
contains information about the upper air measurements required as input to
MESOPAC II. These upper air data were subjected to various quality control/
assurance tests during processing and archival procedures (Ferber et al., 1986)
Figure 4 displays the closest rawinsonde station to each grid point by a
number from 1 to 6. There were six rawinsonde sites available for the Dayton
cases. It is evident from Figure 4 that each site is assumed to represent
a rather large section of the model grid, ranging from 300 km to over 400 km
in most directions.
The hourly surface meteorological observations at all regularly
reporting stations in the northeastern United States and for several
stations in Ontario were also obtained, as the processor cannot compute
fields of meteorological parameters without hourly surface measurements.
Since the number of stations available was much greater than MESOPAC II's
limit of 25 sites, a single group of stations was selected and their
observations were prepared for model input. Figure 5 shows the full set
of stations selected for the model evaluation. These sites were chosen in
order to cover the entire CAPTEX region and to also provide a relatively
uniform intersite spacing across the region. The hourly observations at
each release site were included as input for the experiments with a release
from that site. Table 6 lists the entire group of stations whose observations
were use as input for the model runs. A set of 25 stations was used in all
cases, however, a few sites were only used in the Dayton experiments and
other stations were substituted for the Sudbury cases. In particular,
stations closer to a release location were selected in an attempt to provide
better definition of the flow pattern around the time of release. For
example, two stations in Indiana and one in Michigan were included in the
Dayton cases, but these stations were excluded from the model runs for the
Sudbury releases in -lieu of sites closer to Sudbury. Although the dimensions
of arrays in the program code could be easily increased to accommodate data
at additional stations, the purpose of the operational evaluation was to
exercise the code as specified in the user guide. Figure 6 reveals that
the greater number of surface sites is translated into finer spatial resolu-
tion than that for the upper air stations.
There are notable elevation differences among the stations in Table 6.
The relatively level terrain across Ohio and Ontario gives way to higher
elevations and complex terrain variations among the stations across
Pennsylvania and New York. Nevertheless, elevation data are not required
by MESOPAC II since flat terrain is assumed by this model.
The MESOPAC II code contains no provisions for missing data and execution
stops if it detects missing data or an error occurs in the date/time inform-
ation. Therefore, it is also imperative that the hourly surface observation
files be carefully scrutinized for missing data and that editting corrections
be performed before model execution. Fortunately, only a small fraction of
observations were missing from the 25 selected sites. In most cases a single
hour of observations was missing, although 2 or 3 consecutive hours of missing
data occurred occasionally. In all instances, values were estimated by linear
interpolation from observations taken immediately before and after the missing
hour(s). Table 7 presents the surface meteorological variables required as
input to MESOPAC II. The information about each variable is taken from CD144
(Card Deck 144) format specifications. The files of hourly observations were
also reviewed and values were derived when a single variable was missing.
20
-------
TABLE 5
UPPER AIR MEASUREMENTS AND SPECIFICATIONS*
VARIABLE UNITS RESOLUTION
Pressure mb 0.1 mb
Height m, MSL* 1 m
Temperature degrees (°K) 0.1°
Wind speed m/s 0.1 m/s
Wind direction degrees (°) 1 °
# Units and resolution of the format requirements in MESOPAC II,
* altitude above mean sea level
21
-------
STATION HUMEE3 OF CLOSEST RAWIMSOUOe STATION TO EACH GRID POINT
MULTIPLY ALL VALUES BY 10 ** 0
19
17
16
14
1 3
12
11
10
'
c
7
"
5
A
i
2
1
\
\
\
\
\
\
\
\
\
\
\
\
V
\
\
\
\
\
\
\
^
\
\
\
\
\
\
\
\
\
\
A
'
6
6
6
A
A
A
4
1
1
1
1
I
1
1
A
4
i
A
A
A
4
A
1
1
1
1
1
1
1
A
A
6
i
A
6
A
6
1
1
1
1
1
A
6
6
A
6
A
4
A
1
1
''
1
1
6
A
A
4
A
A
6
A
1
1
1
1
A
A
A
A
A
A
A
A
1
1
1
1
A
A
A
A
A
A
A
A
1
1
1
1
1
4
A
A
A
6
A
A
A
1
1
1
1
A
A
A
4
A
A
4
4
6
1
t
1
A
A
A
A
4
A
A
A
A
1
1
1
4
A
A
4
A
A
A
i
A
A
A
1
A
A
A
A
6
A
4
A
A
A
A
A
5
5
5
5
5
5
5
A
A
4
A
A
">
5
5
5
5
5
5
A
A
4
A
A
^>
5
5
5
5
5
5
5
A
A
A
A
5
5
5
5
5
5
5
5
A
A
A
A
5
5
5
5
5
5
5
5
A
A
A
2
*>
5
5
5
5
5
5
5
A
*
.,
*
2
2
^
5
5
5
5
5
5
5
A
?.
2
2
5
5
5
5
5
5
5
5
5
2
2
2
^
5
5
5
5
5
5
5
5
2
2
2
5
5
5
5
5
5
5
5
5
2
2
2
b
5
5
5
5
5
5
5
5
2
2
2
3
3
3
3
3
3
3
5
3
2
2
2
3
3
3
3
3
3
3
3
3
2
2
2
3
3
3
3
3
3
3
3
2
2
2
3
3
3
3
3
3
3
3
2
2
2
3
3
3
J
3
3
3
3
2
2
2
3
3
3
3
3
J
3
3
3
2
2
3
3
3
3
3
3
3
3
3
3
2
2
9 10 11 12 13 1A 15 1A 17 U 19 20 21 22 23 2A 25 2i 27 23 29 30
Figure 4. Closest rawinsonde site (1 - 6) to each grid point in
the MESOPAC II meteorological grid for Dayton cases.
-------
47.00*
U)
Figure 5. Surface meteorological stations considered in the MESOPAC II
runs for the CAPTEX evaluation. Boundaries for meteorological
grids are denoted (heavy solid and dashed lines).
-------
TABLE 6
SURFACE METEOROLOGICAL STATION INFORMATION
LOCATION
Albany, NY
Altoona, PA
Avoca, PA
Buffalo, NY
Cleveland, OH
Dayton, OH
Dubois, PA
Elmira, NY
Erie, PA
Flint, MI
Ft. Wayne, IN
Grand Rapids ,MI
Harrisburg, PA
Indianapolis , IN
Messina, NY
Newark , NJ
Philadelphia, PA
Pittsburgh, PA
Parkerburg, WV
SITE ID LATITUDE
ALB
AOO
AVP
BUF
CLE
DAY
DUJ
ELM
ERI
FNT
FWA
GRR
CXY
IND
MSS
EWR
PHL
PIT
PKB
*Sault Ste Marie, MI SSM
Syracuse, NY
Toledo, OH
Williamsport , PA
Youngtown, OH
Canadian Sites:
London, ONT .
*0ttawa, ONT.
*Sudbury, ONT.
Trenton, ONT.
Alternate Site:
#Columbus , OH
* applied in the
# Substituted for
SYR
TOL
IPT
YNG
YXU
YOW
YSB
YTR
CMH
Sudbury
DAY in
(DEC
42.
40.
41.
42.
41.
39.
41.
42.
42.
42.
41.
42.
40.
39.
44.
40.
39.
40.
39.
46.
43.
41.
41.
41.
43.
45.
46.
44.
40.
cases only
a test case
. N)
75
30
33
93
42
90
18
17
08
97
00
90
22
75
93
70
88
50
27
47
12
60
25
27
05
32
62
12
00
in place
only
LONGITUDE
(DEG
73
78
75
78
81
84
78
76
80
83
85
85
76
86
74
74
75
80
81
84
76
83
76
80
81
75
80
77
82
of FWA,
. W)
.80
.32
.53
.73
.87
.20
.90
.90
.18
.75
.20
.50
.85
.25
.85
.17
.25
.22
.57
.37
.12
.80
.92
.67
.09
.67
.80
.53
.88
IND,
ELEVATION
(m ASL)
89.
448.
289.
215.
245.
306.
556.
291.
225.
234.
252.
245.
107.
246.
65.
9.
8.
374.
194.
221.
124.
211.
160.
362.
278.
126
348.
81.
254.
YNG
24
-------
ME50PUFF VERSION 4.0 LEVEL 870421
STATION IIUtlBER OF CLOSEST SURFACE MET. STATION TO EACH GRID POINT
MULTIPLY ALL VALUES DY 10 *» 0
19 S
IB \
17 N
16 \
15 N
14 \
11 \
10 \
9 \
a \
7 \
6 N
5 \
4 \
3 N
2 \
1 \
\
14
14
14
14
14
14
i &
14
14
14
13
13
13
13
15
15
15
15
15
14
14
14
14
14
14
1 A
14
14
13
13
13
13
15
15
15
15
15
14
14
14
14
14
14
1 ft.
I A
14
14
13
13
13
13
13
15
15
15
15
14
14
14
14
14
14
1 A
14
14
13
13
13
13
13
7
7
7
7
12
12
12
12
12
12
14
22
22
13
13
13
7
7
7
7
7
12
12
12
12
12
12
12
22
22
22
22
7
7
7
7
7
7
12
12
12
12
12
12
12
22
22
22
22
7
7
7
7
7
7
12
12
12
12
12
12
12
22
22
22
22
22
7
7
7
7
7
12
12
12
12
12
12
12
22
5
5
5
5
7
7
20
20
20
25
25
25
25
25
25
5
5
5
5
5
5
5
20
20
20
20
25
25
25
25
25
25
25
5
5
5
5
5
5
20
20
20
20
25
25
25
25
25
25
25
5
5
5
5
5
19
20
20
20
20
25
25
25
25
25
25
10
10
23
23
23
19
19
19
20
20
20
25
25
25
25
25
25
10
10
23
23
23
19
19
19
19
20
20
24
25
25
25
25
25
10
10
10
23
23
19
19
19
19
20
20
24
24
24
4
4
4
10
10
10
a
a
19
19
19
19
19
20
24
24
24
4
4
4
4
10
a
8
8
a
2
2
2
2
2
24
24
24
24
4
4
4
a
8
8
a
a
2
2
2
2
2
24
24
24
24
24
4
4
a
8
8
a
2
2
2
2
2
2
24
24
24
24
24
4
9
9
16
16
2
2
2
2
2
2
2
24
24
24
24
24
24
9
9
16
16
16
6
6
6
6
6
6
24
24
24
24
24
21
9
9
16
16
16
6
6
6
6
6
6
17
24
24
24
21
21
9
9
16
16
16
6
6
6
6
6
6
17
17
17
21
21
21
9
9
3
3
3
6
6
6
6
IB
IB
17
17
17
21
21
21
21
3
3
3
3
3
18
18
18
IB
18
17
17
17
17
21
21
3
3
3
3
3
3
ia
18
18
18
18
17
17
17
17
1
1
1
3
3
3
11
11
18
18
18
18
18
17
17
17
17
1
1
1
1
3
11
11
11
11
18
ia
18
18
17 17
17 17
17 17
1 1
1 1
1 1
1 1
1 1
11 11
11 11
11 11
11 11
11 11
11 11
18 11
18 18
18 18
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30
Figure 6. Closest surface meteorological station (1-25) to each grid point
of the MESOPAC II meteorological grid for Dayton cases during CAPTEX.
-------
TABLE 7
HOURLY SURFACE OBSERVATIONS AND SPECIFICATIONS
VARIABLE
Wind speed
Wind direction
Temperature
Pressure (station)
Relative humidity
Cloud cover
Ceiling height
(Precip. type)*
UNITS
knots
degrees (°)
F°
Inches Hg
%
Tenths (xlO)
Ft, AGL*
RESOLUTION
1
10°
1
0.01
1
1
100
* Card Deck 1444 specifications
# height above ground level
26
-------
However, no quality assurance tests were performed on the surface observations
since various quality control procedures are routinely conducted during the
data archival process by the National Climatic Data Center.
It should also be stated that these surface observations do not represent
actual time-averaged measurements over 1-hour periods. The hourly observations
are actually taken over a relatively short interval prior to each hour. It
is particularly notable that wind direction is recorded only to the nearest
10°. There is no random number technique in MESOPAC II to generate a value
for the units place and observed directions are multiplied by 10 in the model.
While wind speed is reported to the nearest 1 knot, observed wind speeds less
than 3 knots are rare due to instrument response limitations. These light
wind conditions are usually reported as calm (i.e. zero wind speed).
The land use type must also be provided at each grid point in the
meteorological domain. A list of the land use categories and their
associated surface roughness lengths built into the preprocessor code
are given in Table 8. Since land use information was not part of the
CAPTEX data base, land use types were obtained from an existing in-house
land use inventory (Page,1980). Briefly, this inventory contained the
percentage contributions of 12 different land use types for cells
covering 1/6° latitude by 1/4° longitude areas (i.e. approximately
18.5 km square grids). The land use cell whose coordinates were closest
to each grid point was determined and the land use type exhibiting the
largest percentage was selected to represent the land use at each grid
point. At most grid points, a single land use type was dominant with
percentages greater than 70%. There were a few cases when the agriculture
(Category 1) and forest (Category 4 or 5) categories were the major land
use types and exhibited nearly equivalent percentages. Category 2 (see
Table 8) was selected as a representative compromise in these situations
because only one land use category could be specified for each grid
point. It is acknowledged that this is a simplification to the real-
world where the landscape usually exhibits various land uses on subgrid
scale of this model.
Figure 7 displays the grid for Dayton experiments with the land use
categories. The principal water bodies (12) include the Great Lakes and
Atlantic Ocean. Agricultural cropland (1) dominates Ohio while forested (5)
regions cover central Pennsylvania, and parts of New York state and Ontario.
There are a few grid points defined as urban land use (11). The spatial
pattern in Figure 7 was also found to closely correspond with the coarser
grid scale land use map presented by Sheih et al. (1979).
27
-------
TABLE 8
LAND USE CATEGORIES AND SURFACE ROUGHNESS LENGTHS FOR MESOPAC II
CATEGORY
1*
2*
3
4
5*
6
7
8
9
10
11*
12*
NOTE:
Contents
* land use
LAND USE DESCRIPTION
crops and pasture
crops, woodland, and
grazing land
irrigated crops
grazed forest and woodland
ungrazed forest and woodland
ROUGHNESS LENGTH ZQ (m)
0.20
0.30
0.05
0.90
1.00
subhumid grassland and semiarid 0.10
grazing land
open woodland grazed
desert shrub land
swamp
marshland
urban
water
of table taken from Sheih et
types input at grid points in
0.20
0.30
0.20
0.50
1.00
0.0001
al. (1979)
the model domain
28
-------
MESOPAC VERSION 1.1 LEVEL 860328
LAUD USE CATEGORIES FOR EACH GRID POItfT
MULTIPLY ALL VALUES BY 10 »»»
19 \ \Z 12 12
18 \ 12 2 2
14 V 2 2 2
1 \ 1 1 1
22255
2 2 5 5 12
22212
11125
12
12
1
5
5
12
12
12
2
2
5
5
12 12
12 12
12 1
1 0
1 2
1 1
2 2
2 2
12
2
1
2
2
5
5
12
12
1
2
2
5
5
12
12
1
.
j
i
i
5
5
1Z
0
2
2
5
5
5
5
12
2
5
5
5
5
5
12
2
5
5
2
5
5
12
2
5
5
2
5
5
12
5
2
5
1
5
5
12
2
j
1
1
5
5
12
..
j
1
1
5
5
2
j
j
2
5
1
1
5
.
2
2
1
1
1
5
2
2
2
1
1
2
5
-
1
2
12
1
2
5
2
5
5
5
5
5
5
2
5
12
12
5
12
5
1 ?
1 ?
12
12
1
2
5
1 ?
1 >
12
12
9 10 11 12 13 l'« 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30
Figure 7. Land use categories in the MESOPAC II meteorological grid
for the Dayton cases during CAPTEX.
-------
SECTION 4
MODEL EVALUATION PROCEDURES AND RESULTS
The MESOPUFF II modeling system was applied to the six full-scale
CAPTEX experimental cases. Experiment #6 was omitted in this model
evaluation effort because it was limited to a half-hour release with few
nonzero measured concentrations over the sampling network. All model runs
and data analyses were performed on a Sperry Univac 1182 computer system.
The important input variables and standard features specified in the
MESOPAC II processor runs are given in Table 9. All MESOPAC II runs began
at hour 00 EST (midnight, Eastern Standard Time) of a release day and the
number of simulation hours was sufficient to cover the entire sampling
period of each experiment. All technical default features in the oper-
ational model runs were applied according to the user guide (Scire et al. ,
1984b). The data files of hourly surface observations from 25 selected
stations, the maximum number permitted, were input in each case and data
files of upper air measurements for 6 and 7 rawinsonde sites were used for
Dayton and Sudbury cases, respectively. The gridded meteorological fields
generated by the processor were stored on a single disk file in each case
and an output file was also produced with results printed at two hour
intervals for use in analysis and interpretation of model results. The
MESOPAC II program required about 400,000 bytes of computer core during
execution and computational times varied from 400 to 600 seconds for these
episodes spanning two to three simulation days.
The input values and specifications for the MESOPUFF II model runs are
presented in Table 10. Due to the conservative nature of the tracer, the
wet/dry deposition and transformation processes were not simulated. The
computational grid in the model was defined to be identical to the
meteorological grid. The emission rate (g s ) was slightly different
in each case as individual rates were computed from the actual amount of
tracer emitted over the release period (see Table 3). The tracer gas
existed under ambient temperature and exhibited neutral bouyancy. Thus,
no plume rise occurred. The height of the release nozzle was specified
to be 1 m above the surface in all model runs.
The selection of rates for puff release and puff sampling is an
important element for any puff model because sufficient puff overlap must
occur under all wind speeds encountered in an application for the model
to properly simulate plumes. The results of several model test runs
conducted with various combinations of rates are listed in Appendix A.
Results showed that individual peak concentrations at 300 km arc sites
differed by 3% or less from the average peak value computed over all test
runs. Since these results indicated that modeled concentrations were
not sensitive to combinations of rates between 2 and 12 puffs h ^ at the
nearest arc, the puff release rate and puff sampling rate were both set to
4 h in all model runs. However, an optional feature was selected for
these model runs which allowed the puff sampling rate to increase, if
necessary, in order to insure sufficient puff overlap. The start time for
30
-------
TABLE 9
INPUT CONTROL PARAMETERS AND DEFAULT FEATURES
OF MESOPAC II RUNS FOR CAPTEX EPISODES
FEATURE
SPECIFICATION
METEOROLOGICAL GRID
GRID POINT SPACING
TIME STEP
SIMULATION PERIOD
REFERENCE TIME ZONE
METEOROLOGICAL STATIONS
UPPER AIR STATIONS
WIND FIELDS (default)
OUTPUT FIELDS
COMPUTER REQUIREMENTS
30 x 19 (Dayton), 30 x 24 (Sudbury)
37 km
1 hour
VARIABLE start at 00 EST on the day
of release and go to end of sampling
period (e.g. range from 56 to 87 h)
Eastern Standard Time (EST)
25 hourly surface sites
6 - Dayton , 7 - Sudbury cases
Mixed-layer average (Sfc. to Z^),
upper layer average (Z^ to 700 mb)
Storage on disk file, printer output
at 2-hour intervals
400,000 bytes; 10 minutes CPU time
on a Sperry Univac 1182 computer
31
-------
TABLE 10
INPUT PARAMETERS AND SPECIFICATIONS FOR MESOPUFF II RUNS
TRACER EMISSIONS
COMPUTATIONAL GRID
TIME STEP
PUFF RELEASE RATE
PUFF SAMPLING RATE
WET/DRY DEPOSITION/CHEMICAL
TRANSFORMATION
PUFF DISPERSION
SAMPLING GRID CALCULATIONS
COMPUTATION REQUIREMENTS
3-hour release at constant rate
Source Height 1 m; neutral buoyancy
IDENTICAL TO METEOROLOGICAL GRID
1 hour
4 per hour
4 per hour
PROCESSES NOT SIMULATED
a and az according to PGT curves
to 100 km;time dependent expressions
of Heffter (1965) at greater distances
deactivated; calculations at
nongridded site coordinates only
400,000 bytes of core
32
-------
model execution corresponded with the beginning hour for each release.
The model code was modified slightly to stop new puffs from being
emitted after 3 hours. Thereafter, only existing puffs were tracked and
sampled during the remainder of the simulation period. Hourly tracer
concentrations (g/m3) were computed at the coordinates of all operational
tracer measurement sites in the model domain, instead of at sampling
grid points. This feature reduced the execution time of individual model
runs to less than two minutes for this application.
In order to compare the modeled results with the surface tracer
measurements, the MESOFILE II postprocessor program was applied in order
to compute 3-h or 6-h average concentrations from the model output file
over the same time periods as the measurements and modeled concentrations
were also converted into the same units as the measurements (fl 1 ).
OPERATIONAL MODEL EVALUATION ANALYSES AND RESULTS
Sets of statistical measures of differences (residuals) and correlation
between observed and modeled concentration pairs are presented from time
and/or space pairings in order to provide quantitative evidence of model
performance. Concentration pairs with both observed and modeled values
equal to zero were excluded in the calculation of the statistical measures
because neither the observed nor the modeled plumes had an impact in these
cases; 1088 out of 1895 pairs were in this category. In addition, analyses
of observed plume concentration patterns at individual time periods revealed
a large number of concentrations of 1-2 fl 1 ; some at sites separated
from the main plume distribution by adjacent sites which exhibited zero
concentrations. Specifically, measurements at these isolated locations
were believed to be anomolous since there was considerable uncertainty as
to whether a measured concentration so close to the natural background
really indicated an actual impact by the tracer plume. Potential error
in the analytical procedures used to determine concentrations might be
attributed for some of these values above background (Ferber et al., 1986).
However, instead of eliminating all such extremely low values, a screening
procedure redefined 73 values to be zero where an observed concentration
of < 2 fl 1 occurred at any outlying site separated from the tracer
plume pattern. Other observed concentrations < 2 fl 1 at sites
next to locations with nonzero tracer concentrations were not changed in
the data set since it was deemed reasonable to assume that the tracer plume
had indeed impacted such locations. Therefore, the final evaluation data
set contained all paired values that formed the union of both the perceived
actual and predicted plumes. As a consequence of applying these screening
criteria, the full data set was composed of 734 concentration pairs from
the six modeled experiments.
Statistical Results For Mean and Peak Concentrations
Statistical results were computed from observed and modeled concentra-
tion pairs matched in time and location for each experiment and over the
entire data set. These criteria provide a stringent test of model per-
formance and a critical test of model skill as each predicted concentra-
tion was compared to the measured value at the same site and for the same
time period.
33
-------
The results in Table 11 reveal that MESOPUFF II overpredicted mean
concentrations in each case. Nevertheless, the predicted mean concen-
tration (P) was within a factor of two of the observed mean value (0)
in experiments #2,#4, and #7 and for the entire data set. The overall
mean residual (d) was -33.8 fl 1"^, while individual values ranged from
-58.3 fl I"1 for CAPTEX #1 to a low of 12.8 fl I"1 in CAPTEX #2. The
most significant model overprediction occurred in CAPTEX #3 as P was five
times greater than 0; however, the observed mean concentration in CAPTEX #3
was considerably lower than the mean values of the other experiments.
Results of additional analyses will be presented later in an effort to
explain this large discrepancy.
Since no absolute standards have been established for defining acceptable
model performance, statistical results like those in Table 11 are primarily
useful in an intercomparison of different models. However, comparable
statistics for other models have yet to appear. Nevertheless, results of
several statistical measures in Table 11 did indicate that considerable
scatter existed among the paired concentrations. The standard deviations
of the residuals (SD^) are much larger than the observed mean concentra-
tions. Furthermore, there is greater variability in the modeled results
as exemplified by the standard deviations of the modeled concentrations
(SD ), which are greater than corresponding standard deviations for
observed values (SDQ). The results for the average absolute residual (AAR)
and root mean square error (RSME), when compared to observed concentrations,
also indicate that notable scatter exists between modeled and observed
concentrations. In addition, the Pearson correlation coefficient (R) is
also generally near zero which reveals no linear relationship among the
concentration pairs for these individual data sets and over the entire
data set.
Similar results for the statistical measures were also obtained by
Policastro et al. (1986) when the predicted results of eight regional
models, including MESOPUFF II, were evaluated in time and space against
measurements from two different tracer field studies. Large scatter
and low correlations were generally found between model and observed
results, and most models also displayed a tendency to overpredict mean
concentrations. Consequently, scatter plots of observed versus modeled
concentrations showed many pairs where either the observed or the predicted
value was zero (Policastro et al., 1986; Draxler, 1987; Lee, 1987). Thus,
it was evident that the modeled plume was either displaced away from
particular sites where the actual tracer was measured or modeled concen-
trations impacted the particular sites at different times that the observed
plume. Some statistical measures were strongly sensitive to these circum-
stances and the ensuing results reflected situations when plume displacements
occurred. In particular, significant scatter (e.g. large values for d,
AAR, and RMSE) and low correlations were obtained when observed and modeled
plumes exhibit little or no overlap, even though modeled mean concentrations
may have compared favorably with observed mean values (Policastro, et al.,
1986). However, the statistical measures could only provide partial inform-
ation concerning regional model performance and more innovative analyses,
to be shown, are certainly required to provide better quantitative informa-
tion to assist in assessing and understanding the causes of model/measured
differences.
34
-------
TABLE 11
MESOPUFF II MODEL EVALUATION RESULTS:
CONCENTRATIONS PAIRED IN TIME
OBSERVED AND MODELED
AND LOCATION
STATISTIC^ CAPTEX EXPERIMENT
#1 #2 #3 #4
N 143 247 124 96
0 22.6 45.7 10.0 54.9
SDQ ±138.1 ±172.0 ±31.4 ±172.2
P 80.9 58.5 49.9 90.5
SDp ±252.8 ±231.1 ±163.2 ±226.5
d -58.3 12.8 39.9 -35.7
SDd ±165.5 ±295.5 ±168.2 ±294.6
AAR 71.8 95.7 57.1 123.6
RMSE 174.9 295.2 172.2 295.2
R 0.80 -0.05 -0.07 -0.08
$ - units are f 1/1 , except for N and R
N - number of concentration pairs
#5 #7
68 56
41.7 79.7
±112.7 ±135.4
86.0 113.5
±166.8 ±214.8
-44.4 -33.8
±178.4 ±283.0
103.2 185.5
182.6 282.5
0.23 -0.27
ALL
734
38.6
±143.3
72.4
±218.6
-33.8
±244.6
95.7
246.7
0.14
0 mean observed concentration ; P = mean modeled concentration
SD standard deviation about the mean
d average residual =0 P
AAR - average absolute residual
RMSE = root mean square error
R = Pearson correlation coefficient
35
-------
Objective tests between the concentration pairs were performed in order
to provide additional quantitative comparative results which could provide
a better overall picture of the model's capabilities to simulate the
spatial plume pattern over the time frame of these experiments. In
particular, comparative test results that demonstrate the relative amount
of correspondence between modeled and observed plumes could provide relevant
information about the abilities of the transport and dispersion components
of the model to simulate the movement and growth experienced by the tracer
plume. Table 12 contains results in the form of percentages for certain
comparative tests on the concentration pairs. One of the interesting
tests is the percentage of pairs where both observed and modeled values
are nonzero (Tl). High percentages for Tl demonstrates that a substantial
overlap must have existed between the modeled and observed plumes for the
sampling periods of an experiment, which also suggests the overall speed
and path, and growth of the plume was accurately simulated. Conversely,
low percentages would reveal that the area of intersection between the
respective plumes was rather small which implies inadequate treatment of
plume transport/dispersion by the model. The percentages for Tl in Table 12
vary among these cases with the highest being 46.9% for CAPTEX #4. Tl
was less than 20% for experiments #1, #3, #5, and #7 which indicated
rather small areas of intersection for the observed and modeled plumes
during these cases.
Additional evidence of spatial displacements between the respective
plumes is provided by test results showing the percentages when either the
observed (T2) or the predicted (T3) concentration was zero while its
counterpart was nonzero. Low percentages for T2 and T3 would suggest
a strong spatial correspondence of the respective tracer plumes. However,
these results show a notable number of pairs where either the observed
or predicted concentration was zero. In fact, results indicate that either
T2 or T3 was greater than 50% in some experiments with most values
for T2 and T3 above 30%. The best combination of T2 and T3 is found for
CAPTEX #4. which also displayed the highest value for Tl. Finally,
the results for T4, the percentage of cases where observed concentrations
were greater than modeled values, vary with values ranging from about 35%
to 54%. Results from the T4 test are sometimes used as indicators of the
performance of dispersion in a model. For example, when T4 is significantly
greater than 50%, it might be suggested that the modeled plume exhibited
less horizontal spread than the observed plume. Nevertheless, T4 results
are difficult to interpret in this manner and are also complicated by the
potential effects from both transport and dispersion. Even though the model
certainly overpredicted the mean concentrations, as shown by the earlier
statistical results, spatial displacements due to differences in the speed
and/or the directional components of transport between the model and
observed plumes are believed to be the principal factors responsible for
the variations in T4 results among these cases.
Mean concentrations and selected statistical measures for each experiment
are presented in Table 13 to illustrate model behavior on an event by
event basis. An examination of Table 13 indicates that observed mean
concentrations generally exhibited an increase with time and then gradually
decreased toward the last sampling period. This temporal trend reflects
the arrival of the tracer plume accompanied by generally high concentrations
36
-------
TABLE 12
TEST RESULTS OF MODELED AND OBSERVED CONCENTRATIONS
PAIRED IN TIME AND LOCATION
CAPTEX EXPERIMENT
TEST #1 #2 #3 #4 #5 #7
N 143 247 124 96 68 56
Tl (%) 16.8 33.6 16.1 46.9 19.1 16.1
(P&O >0)
T2 (%) 52.4 32.0 33.9 27.1 48.5 37.5
(0=0, P>0)
T3 (%) 30.8 34.4 50.0 26.0 32.4 46.4
(P=0, 0>0)
T4 (%) 35.7 53.0 54.0 44.8 35.3 48.2
(0>P)
ALL
734
26.4
37.6
36.0
46.7
37
-------
TABLE 13
SELECTED STATISTICAL EVALUATION RESULTS FOR EACH EVENT
OF THE CAPTEX EXPERIMENTS
EXPT
1
1
1
1
1
1
1
1
1
1
2
2
2
2
2
2
2
2
2
2
2
2
3
3
3
3
3
3
3
3
3
3
3
MM/DD-HR
(LSI)
9/18-14
9/18-17
9/18-20
9/18-23
9/19-02
9/19-05
9/19-08
9/19-14
9/19-20
9/20-02
9/25-13
9/25-16
9/25-19
9/25-22
9/26-01
9/26-04
9/26-10
9/26-16
9/26-22
9/27-04
9/27-10
9/27-16
10/02-16
10/02-19
10/02-22
10/03-01
10/03-04
10/03-07
10/03-13
10/03-19
10/04-01
10/04-07
10/04-13
N
4
5
15
4
21
3
26
26
26
14
4
5
6
14
8
24
33
36
37
32
25
10
1
4
6
16
4
29
27
15
12
5
5
0
(fl/1)
5.8
P
(fl/D
59.0
552.2 1012.8
23.7
0.8
16.6
0.
4.5
3.2
2.0
2.4
0.2
0.2
45.7
130.4
311.6
125.6
32.6
21.0
17.8
17.7
11.7
5.1
1.0
18.0
50.8
3.6
5.0
6.9
6.7
12.3
12.6
12.0
1.8
155.9
29.0
61.8
118.7
49.2
38.1
36.6
13.0
51.0
827.6
419.0
155.8
118.8
33.3
32.7
33.6
17.2
8.6
8.7
9.7
0.
1.5
257.5
118.8
5.5
39.5
28.9
49.1
4.7
0.
0.
d
(fl/1)
-53.
-460.
-131.
-28.
-45.
-118.
-44.
-34.
-34.
10.
-50.
-827.
-373.
-25.
192.
92.
-0.
12.
0.
9.
3.
-4.
1.
16.
-206,
-115.
-0,
32.
-22
-36.
7
12
1
2
5
3
.2
1
7
7
8
6
6
8
4
.3
.4
,9
3
.1
,6
,6
.1
,0
,6
,0
,5
.7
,1
.5
.6
.3
.9
.9
.0
.8
AAR
(fl/D
53.2
615.5
140.7
29.8
75.5
118.7
52.5
41.3
38.1
15.0
51.2
827.8
464.0
285.9
421.4
150.1
50.0
42.3
23.1
21.9
16.7
13.8
1.0
17.5
308.0
122.4
10.5
45.2
32.8
49.1
12.6
12.0
1.8
NNQ
2
0
3
2
8
3
16
17
17
7
3
4
3
5
4
6
9
10
9
8
6
6
0
2
4
7
1
11
8
5
4
0
0
NN
r
0
1
4
2
10
0
6
9
7
5
1
1
2
8
2
13
13
10
11
13
8
1
1
2
1
9
3
14
10
5
7
5
5
Tl
(%)
25.
75.
53.
0.
14.
0.
8.
0.
8.
7.
0.
0.
17.
7.
25
21.
30.
42.
43.
31.
36.
30.
0.
0.
0.
0.
0.
10.
30.
27.
8.
0.
0.
T4
(%)
0.
50.
40.
50.
52.
0
31.
35.
31.
36.
25.
20.
33.
57.
38.
71.
70.
47.
54.
59.
52.
40.
100.
50.
17.
56.
75.
48.
48.
40.
67.
100.
100.
38
-------
TABLE 13
EXPT
4
4
4
4
4
4
4
4
4
4
4
5
5
5
5
5
5
5
5
7
7
7
7
7
7
7
7
DATE/TIME
10/14-13
10/14-16
10/14-19
10/14-22
10/15-01
10/15-04
10/15-07
10/15-10
10/15-16
10/15-22
10/16-04
10/26-03
10/26-09
10/26-15
10/26-21
10/27-03
10/27-09
10/27-15
10/27-21
10/28-23
10/29-02
10/29-05
10/29-08
10/29-11
10/29-14
10/29-20
10/30-02
N
1
7
5
16
9
24
6
16
8
9
10
4
9
14
12
13
10
3
1
0
3
3
14
4
16
11
3
0
0.
133.9
276.6
47.8
30.1
58.8
27.0
9.6
13.2
8.8
0.
86.2
181.1
57.5
2.8
0.6
0.6
1.3
1.
0.
1.
35.7
25.6
179.5
75.9
196.0
32.7
P
67.0
167.6
521.0
230.0
77.0
51.8
62.3
33.2
11.2
7.3
0.
0.
176.8
118.5
130.5
74.5
2.0
0.
0.
0.
0.
190.7
220.8
8.0
107.3
63.2
83.3
(CONTINUED)
d
-67
-33
-244
-72
-46
7
-35
-23
2
1
86
4
-61
127
-73
1
1
1
1
-155
-195
171
-31
132
-50
.0
.7
.4
.2
.9
.1
.3
.6
.0
.5
.2
.3
.0
.7
.8
.4
.3
.
.0
.2
.5
.4
.8
.7
AAR
67.0
301.4
784.4
138.7
102.0
70.1
69.0
26.3
14.0
7.2
86.2
181.1
175.1
128.7
74.2
2.4
1.3
1.
1.
155.0
234.6
187.5
179.7
259.2
116.0
NNQ
1
5
3
3
3
4
3
5
2
1
10
0
2
6
6
10
7
0
0
0
2
8
1
6
2
2
NNp
0
2
1
7
3
5
0
0
3
4
10
4
4
5
2
1
2
3
1
3
0
2
3
6
9
1
Tl
0.
0.
20.
38.
33.
58.
33.
69.
38.
44.
0.
33.
14.
33.
8.
0.
0.
0.
0.
33.
21.
0.
25.
0.
0.
T4
0.
29.
20.
62.
33.
50.
17.
19.
62.
67.
100.
56.
43.
17.
8.
20.
100.
100.
100.
0.
14.
75.
44.
82.
33.
NN , NN - number of observed
Tl P Snd 0 both > 0
T4 - 0 > P
predicted zero concentrations.
39
-------
at the closest arc or sites a few hours after release, and this peak was
followed by an general decrease in the mean concentration due to the plume
being dispersed as it traveled across the sampling region. Modeled concen-
trations also exhibited a similar temporal pattern, although the time
variation of mean model values was often out of phase with the time series
of observed mean values. The residuals were usually negative as the model
appears to overpredict in most events. The largest biases were usually
found during an early time period when concentrations also happen to be
highest. Differences in the time and location of impact of the peak
concentration are attributed for these significant departures between the
observed and modeled mean concentrations. The results for the number of
observed (N ) and predicted (N ) concentrations being zero coupled
with the low Tl results indicate important spatial plume displacements.
In fact, during some events there was no intersection of the respective
plumes (i.e. Tl = 0). Low values of Tl were more common during the early
events and also near the end of an experiment. Differences in the time of
initial impact of the respective plumes at the first arc (or closest sites
for the Subdury cases) can produce low Tl values soon after release. At
the later stages of an experiment a more rapid departure of the observed
plume from the sampling region or large spatial displacements between the
plume patterns existing on the sampling domain was also believed to lead
to small fractions in Tl.
Averaged concentrations for each event were also computed separately from
observed and modeled concentrations (excluding any zero values) to assess
differences in magnitude between the plumes with time. The results are
presented as a function of time after release in Figure 8a-f. For CAPTEX
#1-4 (Figures 8a-d), the highest observed and modeled average plume concen-
tration occurred within 12 h after release as a result of plume impact at
the 300 km arc sites. The modeled values (X) replicated the time history
and magnitudes of observed (0) concentrations best in #2 and #4. For a few
events there were very few or no observed nonzero concentrations, which
was responsible for the values near or on the abscissa. These results also
show the overall decrease in average plume concentrations with time.
Concentration standard deviations were comparable to mean values for most
events. The results for CAPTEX #5 and #7 (Fig. 8e-f) did not exhibit concen-
trations at early periods as high as the Dayton cases; probably because the
plumes did not intersect sites or an arc as close to release point as in the
Dayton cases. Despite the log scale for concentration, it appeared that
average modeled concentrations from these calculations were more often higher
than average observed values, which suggests that peak concentrations were
overpredicted, although more definitive information about model performance
at peak concentration levels will be presented later.
Cumulative frequency distributions were determined to establish the
relative agreement between the observations and predicted values across
the range of concentrations. Although frequency distributions do not
demonstrate information about model skill, since values are not paired in
time or location, the results can provide evidence about how well the modeled
concentrations replicate the magnitude and overall distribution of observed
values. The results from the entire data set in Figure 9 concur with the
earlier tests that large numbers of either observed or modeled concentrations
in the evaluation data set were zero. The cumulative frequency curve for
observed concentrations lies above that for the modeled values. The
40
-------
a CAPTEX 1
10 c| 1 1 1 1 1 ! 1 1 1 i 1 1 i 1 1 1 T
CT 103
en
LJ
O
o
1°
X
o
X
X X
X
o
x
X
o
x - MODEL
o - OBSERVED'
x
O
x
X -
! 1 - 1 - 1
10 20 30
TIME AFTER RELEASE (h)
40
b io2
LJ
O
-z.
o
0
10' -
b CAPTEX 2
|iiii|iii:iiiiir
X
X O
X
o
X
o
2<
X
X
o
x - MODEL
o - OBSERVED"
10 20 30 40 50
TIME AFTER RELEASE (h)
60
Figure 8. Average concentrations obtained from sites with observed
and modeled nonzero concentrations over the sampling periods
during a) CAPTEX #1 and b) CAPTEX #2.
41
-------
10*
F 10J
c CAPTEX 3
o
c;
UJ
O
o
10
10
10
x - MODEL
o - OBSERVED"
X
o
X X
x
o
X
o
o
o
o
o
X
o -
I
I I I I I I.
10 20 30 4.0
TIME AFTER RELEASE (h)
50
-------
10
LJ
o
CJ
10
10
o
X
o
I
e CAPTEX 5
x
o
o
I I I I
X
o
x - MODEL
o - OBSERVED"
X
o
o
I I vJ-v I
10 20 30 40
TIME AFTER RELEASE (h)
50
f CAPTEX 7
- I i F 1 i i i i I i i i i i T
1
x - MODEL ]
vT 103
2
O
a;
1
z:
LJ
O
§ '«'
o
10°
o - OBSERVED
^ H
x o o ]
X v
-3
E o 3.
~" ~i
I X
- -
vi, 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1
5 10 15 20
TIME AFTER RELEASE (h)
25
Figure 8. Same as 8a-b, except for e) CAPTEX #5 and f) CAPTEX #
43
-------
00
80
o
l_d
ID
a
Ld
60
ID
^20
=D
O
0
CAPTEX
i i i i 1111
MESOPUFF II (clashed)
OBSERVED (solid)
11
o
10 1 10 2 10
CONCENTRATION (fl/l)
10
Figure 9. Cumulative frequency distributions of the model and observed
concentrations from the entire data set.
-------
separation between the curves increased from 1 to 10 fl 1 , which
indicated that a greater number of observed values exist at these low
concentrations compared to modeled results. The separation remains about
the same between 10 and 100 fl 1 . However, the gap between the curves
closed rapidly above 100 fl 1 as a relatively greater number of modeled
values occur at higher concentrations.
Individual cumulative concentration distributions are shown in
Figures lOa-f to illustrate the variability in the distributions among these
experiments. The best overall agreement between the cumulative distribution
curves of observed and modeled concentrations appeared in experiments #2,4,
and #7. CAPTEX #2 and #4 also happened to exhibit the best results for
plume overlap (i.e. highest Tl values in Table 13) among these experiments.
The frequency distribution results of observed and modeled values
in Table 14 quantitatively reinforce the previous cumulative distribution
findings. These results that reveal a relatively large number of observed
concentrations still existed at 1 fl 1 , however, there was no basis to
eliminate or modify these values since all were found at sites on the edge
of the observed plume. These results also verify that larger numbers of
observed values occurred at low concentrations. A crossover point appeared
at about 25 fl 1 , as the frequencies of modeled values were greater at
higher concentrations. In fact, these results provided strong evidence
that the model overpredicted the highest concentrations because twice as
many modeled values are found in the intervals above 100 fl 1 observed
values.
The ability of a model to predict the highest concentration should
certainly be an important aspect of an evaluation effort, since models
are often applied in order to assess peak concentrations in regulatory
situations. In the evaluation for peak concentrations, results were
determined from observed and modeled values paired in time only or location
only. In fact, there were very few cases where the modeled peak concentra-
tion occurred at the same site and time as the observed peak concentration,
which is not unexpected in view of the spatial plume displacements
reported earlier.
Table 15 contains the results for peak observed and modeled concentra-
tions paired in location only. This evaluation framework allows peak
observed and modeled concentrations to occur at different time periods
at a given site. These results show that the model overpredicted peak
concentrations for all experiments. Except for CAPTEX #3, the peak
modeled concentrations were within a factor of two of peak observed values.
The smallest negative residuals are found in CAPTEX #2 and #4. In these
experiments, the results for Tl in Table 12 indicated that plume overlap
was higher and the other test results were more favorable than in the other
cases. The largest overpredictions of peak concentrations occurred in
CAPTEX #1 and #3. As reported earlier, the mean observed concentration
for CAPTEX #3 was unusually low in relation to the other cases. Results of
analysis of peak concentrations at the 300 km sites, where the highest
observed concentrations occurred for the Dayton cases, revealed that the
main portion of the observed plume did indeed pass to the west of the
300 km arc in this case. This explains the low observed peak and mean
concentrations for this experiment and the reason for the significant
45
-------
100
80
O
z.
UJ
8"
UJ
cr
u.
UJ
>*1Q
220
-D
O
a CAPTEX
i M
10
MESOPUFF II (dashed)
OBSERVED (solid)
Mill I I I I I I I ll I I 1 I I I I
10 ' 10 10
CONCENTRATION (fl/l)
10"
100
b CAPTEX i
MESOPUFF II (dashed)
OBSERVED (solid)
10 ' 10 * 10
CONCENTRATION (fl/l)
Figure 10. Cumulative frequency distributions of model and observed
concentrations for a) CAPTEX #1 and b) CAPTEX #2.
46
-------
100
c CAPTEX #3
80
o
60
a
LJ
LJ.
ID
220
r>
o
I I 1 T FIT I 1 I I I I 1 yrr ' - r i1 I J. .L Uf- ] [ Tllll
MESOPUFF II (dashed)
OBSERVED (solid)
10
10' 102 10
CONCENTRATION (fl/l)
100
80
o
z
LJ
ID
o
LoJ
o:
Lu
60
^20
Z)
o
I 1 I 1 I I 1 1
d CAPTEX
I I I Mill
MESOPUFF II (dashed)
OBSERVED (solid)
1 L 1 I I
10
10 10 2 10
CONCENTRATION (fl/l)
104
Figure 10. Same as lOa-b, except for c) CAPTEX #3 and d) CAPTEX
47
-------
too
e CAPTEX #5
80
60
O
z:
UJ
ID
O
ui
Lu
Ld ,
220
Z>
O
MESOPUFF II (dashed)
OBSERVED (solid)
10
10 ' 10 2 10
CONCENTRATION (fl/l)
10
100
f CAPTEX ^
MESOPUFF II (dashed)
OBSERVED (solid)
CONCENTRATION (f|/|)
Figure 10. Sa.e as lOa-b, except for e) CAPTEX *5 and f) CAPTEX *7
48
-------
DISTRIBUTION OF
FOR
CONCENTRATION OR
INTERVAL (fl/1)
0
1
2
3-5
5-10
11-15
16-25
26-50
50-100
100-150
151-200
201-250
251-300
301-400
401-500
501-600
601-701
701-800
801-900
900-1000
1001-2001
>2001
TABLE 14
OBSERVED AND MODELED CONCENTRATIONS
ENTIRE CAPTEX DATA
NUMBER OF
OBSERVED
1437
68
53 50.4%
57
53
29
51 36.9%
47
42
14
11
8 10.7%
3
7
6
2
0
1 0.9%
0
1
5 1.1%
0
SET
VALUES
MODELED
1425
36
25 34
49
50
30
49 40
67
46
24
25
16 20
12
12
6
4
4
2 2.
3
0
8 2.
2
.8%
.9%
.2%
8%
1%
49
-------
PEAK
EXPT N
#
1 42
2 58
3 43
4 26
5 25
7 28
ALL 69
OBSERVED
0
(fl/D
66
±251
86
±229
17
±48
111
±262
95
±169
111
±171
197
±315
1
2
3
3
7
6
0
6
2
0
3
5
4
7
TABLE 15
AND MODELED CONCENTRATIONS PAIRED IN LOCATION ONLY
P
(fl/D
176
±427
133
±406
109
±231
151
±345
178
±225
173
±263
391
±526
.4
.0
.2
.2
.0
.7
.7
.8
.8
.4
.9
.2
.1
.3
d
(fl/D
-110
±257
-46
±84
-91
±242
-40
±458
83
±181
-62
±344
193
±550
3
6
9
5
4
2
7
8
6
4
6
7
6
6
AAR RMSE R T4 (0>P)
(fl/1) (fl/1) (%)
141.3 277.4 0.83 31.0
180.6 482.6 -0.09 60.3
115.4 256.2 -0.11 34.9
208.7 451.8 -0.12 57.7
126.0 196.4 0.61 20.0
236.5 344.2 -0.22 42.9
326.0 579.9 0.22 29.0
50
-------
overpredictions was because the modeled plume directly intercepted this arc.
Other results in Table 15 also reveal large scatter and low correlations
between peak pairs matched in space. The absolute value of residuals and
RMSE are large relative to observed peak values and correlation coefficients
are generally small and negative.
A comparative analysis of peak concentrations was also performed
for observed and modeled values paired in time only. With this framework,
the highest observed concentration is compared with the peak modeled value
from any site for the same time period. Table 16 contains the results for
the events of each experiment. Travel time is also given as the number
of hours from the start time of release to the beginning hour of each
sampling interval. Clearly, in most of the events the modeled peak concen-
trations were greater than observed values, as exemplified by the large
number of negative residuals. An encouraging result, however, is that the
highest modeled concentration within an experiment occurred at the same
time as the corresponding observed peak value in #1,#4, and #5. This
result suggests that there was apparently good agreement in the transport
speed and direction between the modeled and actual plumes during the initial
hours after release. On the otherhand, highest modeled peak concentrations
for experiments #2 and #7 occurred much earlier than the observed highest
values in these cases. Since these results are unpaired in space, it is
also interesting to note that a survey of -all 60 events revealed that in
only 3 cases the observed and modeled peak concentrations during a partic-
ular event were found at the same site. However, there were 21 events
where the peak values occurred at different sites along the same arc.
Larger displacements obviously existed in another 25 events where the peak
observed and modeled values were located at sites on different arcs.
Finally, the relative positions of peak concentration pairs in the remaining
11 cases were indeterminant since either the observed or modeled value was
zero. These results also reveal that the model generally overpredicted
peak values and notable time and/or spatial displacements also existed
between observed and modeled peak concentrations.
Statistical results in Table 17 were computed over the high-25
observed and modeled concentrations fully unpaired in time and location
for each experiment and for the overall data set. These results are also
not unexpected based on the previous analyses of the peak concentrations
and frequency distributions. The model consistently overpredicted for this
group of concentrations even when values are completely unpaired. The
absolute values of the bias were less than the average peak observed
concentrations for experiments #2, #4, and #7. Since time and location
constaints are irrelevant in this comparison, the peak concentration dif-
ferences given in Table 17 were not likely attributed to transport speed
and direction variations of the observed and modeled plumes. These
results and those in previous tables for peak concentrations suggest that
dispersion of the modeled plumes may not be as great as that experienced
by the actual tracer plumes. Results of analysis of additional model runs
to be discussed later will attempt to confirm this hypothesis and identify
the underlying cause(s).
The configuration of the sampling network presented an opportunity
to investigate the four Dayton experimental results separately from the
Sudbury cases. Consequently, a statistical analysis was performed on this
51
-------
TABLE 16
PEAK OBSERVED AND MODELED CONCENTRATIONS PAIRED IN TIME FOR EACH EVENT OF THE CAPTEX EXPERIMENTS
1
2
EVENT
5678
10
11
12
CAPTEX ffl
DATE7TIME (DD/IIRl
TIME FROM RELEASE (h)
Peak Obs. (fl/1 )
Peak Modeled (fl/1 )
RESIDUAL (Op - Pp)
CAPTEX #2
TIME FROM RELEASE (h)
Peak Obs. (fl/1 )
Peak MODELED (fl/1 )
RESIDUAL (Op - Pp)
CAPTEX #3
AVE +_ SD
18/14 18/17 18/20 18/23 19/02 19/05 19/08 19/14 19/20 20/02
2 5 8 11 11 17 20 26 32 38
22 1586 121 2 89 0 47 16 14 12 191 +_ 492
153 2367 862 67 283 221 221 212 197 60 168 +_ 668
-131 -781 -711 -65 -194 -224 -177 -226 -183 -48 -277 +_ 262
"25/13 25/16 25/19 25/22 26/01 26/04 26/10 26/16 26/22 27/01 27/10 27/16
1 4 7 10 13 16 22 28 34 40 46 52
1 1 228 1348 1575 1399 200 96 73 96 43 44 425 +_ 618
156 2767 1430 1020 381 413 245 214 144 50 27 26 573 +_ 779
-155 -2766 -1202 328 1194 986 -45 -118 -71 46 16 18 -148 + 1017
DATE/TIME (DD/IIR)
TIME FROM RELEASE (hrs';
Peak Ohs. (fl/1)
Peak MODELED (fl/1 )
RESIDUAL (Op - Pp)
CAPTEX #4
02/16 02/19 02/22 03/01 03/04 03/07 03/13 03/19 04/01 04/07 04/13
2 5 8 11 11 17 23 29 35 41 47
1 71 304 21 15 63 68 162 79 28 3
0 4 874 1416 22 264 201 294 28 0 0
1 67 -570 -1395 -7 -201 -133 -132 51 28 3
T47T3 14/16 14/19 14/22 15/01 15/04 15/07 15/10 15/16 15/22 16/04
2 5 8 11 14 17 20 23 29 35 41
0 935 1350 248 216 210 156 40 37 28 0
67 830 1448 687 491 392 221 117 51 35 0
-67 105 -98 -439 -275 -182 -65 -107 -14 -7 0
74 + 89
282 + 426
-208 + 433
TIME FROM RELEASE (h)
Peak Obs. (fl/1 )
Peak MODELED (fl/1 )
RESIDUAL (Op - Pp)
CAPTEX #5
DATE/TIME (DD/IIRl
TIME FROM RELEASE (h)
Peak Ohs. (fl/1 )
Peak MODELED (fl/1 )
RESIDUAL (Om - Pm)
26/03 26/09 26/15 26/21 27/03~27/09 27/15 27/21
4 10 16 22 28 34 40 46
167 160 381 33 125 10 2 1
0 794 478 382 272 5 0 0
167 -334 -94 -319 -117 5 2 1
293 ^ 439
397 +_ 426
-104 + 142
118 +_ 181
241 ^ 277
- 93 + 178
CAPTEX 117
DATE/TIME (DTT/HR)
TIME FROM RELEASE (h)
Peak Ohs. (fl/1 )
Peak MODELED (fl/1 )
RESIDUAL (0 - P )
28/23 29/02 29/05 29/08 29/11 29/14 29/20 30/02
-2 1 4 7 10 13 19 25
1 1 107 309 277 546 553 98
0 0 299 1011 32 578 652 230
1 1 -192 -732 215 -32 - 99 -132
236 +_ 224
354 +_ 351
-118 + 280
-------
TABLE 17
HIGH- 25 OBSERVED AND MODELED CONCENTRATIONS UNPAIRED
IN TIME AND LOCATION FOR EACH CAPTEX EXPERIMENT
EXPT
1
2
3
4
5
7
ALL
± one
(f?7
121 ±
341 ±
43 ±
196 ±
113 ±
178 ±
637 ±
standard
a,
-------
subset of concentration pairs to compare observed and modeled concentrations
for each arc from Dayton cases. The results for mean concentrations in
Table 18 were derived from paired values at all sites along each arc from
CAPTEX #1-4. These results confirm that the largest overpredictions occurred
at the closest arc with the largest negative residuals at the 300 and 400 km
arcs. The mean observed concentration at the 300 km arc was significantly
greater than values at other arcs by as much as a factor of 10. Another
notable result is that observed mean values appear to display no strong
trend with downwind distance between the 400 km and 1000 km arc over these
cases. The modeled mean concentrations, on the otherhand, definitely exhibit
a gradual decrease with distance. The most significant differences between
the respective mean values existed from the 400 km to 600 km arcs.
The results for the peak concentrations in Table 19 are similar to those
for mean values over these arcs. The peak observed and modeled values were
computed from the highest concentration at each of the sites along the arcs
for these cases. These results show that model overpredictions of peak
concentrations were also certainly largest at the 300 km arc. While the
peak observed concentration at the 300 km arc is considerably larger than
peak values at any of the other arcs, there is also an indication of a
secondary peak in the results for the 700 km arc. This feature was believed
to be attributable to fumigation of elevated tracer material to the ground
during the radid growth of the mixing height to higher levels over the
morning period of the day after release.
Comparative Analyses of Plume Patterns and Positions
The graphical results in Figure lla-f are shown to give qualitative
displays of the overall observed and simulated plume patterns for each of
the modeled CAPTEX experiments. The overall spatial patterns of the
observed and modeled plumes are depicted by symbol(s) at the sites that
experienced a nonzero observed (0) and/or modeled (*) concentration(s)
during any sampling event of an experiment. The sites where both nonzero
observed and predicted concentrations occurred represent an overlap region
of the actual and modeled tracer plumes over the entire course of a
particular experiment.
During CAPTEX #1-4 (Figures lla-d) with releases from Dayton, both
modeled and observed plumes crossed the 300 km arc in northern Ohio,
although some spatial displacement in the positions of plume impact are
evident even at this arc. The modeled plume appeared to overlap the
observed tracer plume very well at the 300 km arc during CAPTEX #1
(Figure lla), however, the observed plume traveled north of the modeled
plume as it passed across Lake Ontario and upper New York state before
exiting the sampling region. The modeled plume during CAPTEX #2 (Figure lib)
intercepted the 300 km arc to the left (i.e. counterclockwise) of the
observed plume, however, there appears to be better overall spatial agree-
ment in the plume patterns farther downwind in this case.
In contrast, the modeled plumes were shifted to the right (clockwise)
of the observed plumes at the 300 km arc during CAPTEX #3 and #4 in
Figures lie and lid, respectively. The spatial discrepancies between the
respective plumes are particularly noticeable from the 300 km to 500 km arcs
during CAPTEX #3. In fact, the observed plume nearly missed the 300 km arc
54
-------
TABLE 18
STATISTICAL RESULTS OF OBSERVED AND MODELED CONCENTRATIONS
PAIRED IN TIME AND LOCATION FROM SAMPLING ARCS FOR CAPTEX #1-4
ARC
300
400
500
600
700
800
900
1000
N NN 0
(SITES) (PAIRS) (fl/1)
16
9
12
10
10
13
8
3
91
71
75
78
93
112
63
27
140.8
12.0
18.7
15.2
18.8
16.8
12.2
13.7
P
(fl/D
230
110
55
34
27
14
12
13
.3
.7
.4
.3
.0
.1
.5
.4
d
(fl/D
-89.5
-98.7
-36.7
-19.2
-8.2
2.7
-0.3
0.3
AAR
(fl/D
321.
109.
53.
41.
39.
27.
22.
17.
8
5
7
5
3
0
1
6
T4
41.
23.
30.
41.
55.
67.
60.
63.
(0>P)
8
9
7
0
9
5
3
0
PEAK OBSERVED
UNPAIRED IN TIME
ARC N
300 12
400 8
500 9
600 9
700 9
800 12
900 6
1000 2
TABLE 19
AND MODELED CONCENTRATIONS
AND LOCATION FROM CAPTEX #1-4
NN 0
(fl/D
12 439
±658
8 61.
±69.
9 71.
±59.
9 75.
±63.
9 84.
±74.
12 71.
±41.
6 62.
±34.
2 55.
±37.
.0
.2
1
8
7
1
2
6
8
2
0
4
5
3
5
5
P
1041
±924
386
±328
170
±95
113
±78
135
±82
59
±53
70
±111
123
±73
d
/D (fl/D
.7
.8
.1
.6
.4
.7
.3
.9
.2
.6
.3
.5
.7
.5
.0
.5
-602
±1136
-307
±354
-98
±94
38
±65
-50
±132
11
±75
-8
±125
-67
±36
.7
.8
.0
.8
.8
.2
.1
.6
.4
.6
.7
.2
.2
.5
.5
.1
AAR T4
(fl/D (%)
983.5 16.7
307.0 0.0
107.7 22.2
64.6 22.2
123.6 33.3
53.5 58.3
80.8 50.0
67.5 0.0
55
-------
a CAPTEX #1 RESULTS
X1S.CO-
\
vf
x tr.ecf
b CAPTEX #2 RESULTS
Figure 11. Observed (0) and modeled (*) plume patterns depicted from sites
with nonzero concentrations during a) CAPTEX #1 and b) CAPTEX #
56
-------
I
5
c CAPTEX #3 RESULTS
d CAPTEX #4 RESULTS
Figure 11. Same as lla-b, except for c) CAPTEX #3 and d) CAPTEX
57
-------
I
5
*
"li.CO-
e CAPTEX #5 RESULTS
n X 41.»f
f CAPTEX #7 RESULTS
R
*
Figure 11. Same as lla-b, except for e) CAPTEX #5 and f) CAPTEX #7
58
-------
during #3, as the bulk of it apparently traveled over extreme northwestern
Ohio. Clearly, the best overall spatial agreement between the plumes
existed during CAPTEX #4. Results of analysis of winds to be shown later
will indicate that vertical wind direction shear was comparatively smaller
on this case than in the other Dayton episodes.
Figures lle-f show the plume patterns during CAPTEX #5 and #1 for the
Sudbury releases. There was an interesting difference between these cases
as the modeled plume traveled to the right (clockwise looking downwind) of
the observed plume during CAPTEX #5 and to the left (counterclockwise) of
the observed plume during #7. There is also evidence that the spatial
separation between the respective plumes increased with distance downwind
for these two cases. However, no firm conclusions can be reached about
the relative spatial agreement between the plumes at any particular
time period within an experiment from Figure lla-f since each showed only
the overall spatial 'footprint' of the modeled and observed plumes over
a particular experimental episode. Individual graphical maps of the
concentration distributions, too numerous to display herein, are needed
to demonstrate the patterns of the observed and modeled plumes at each time
period.
Another analysis technique that has emerged to assess spatial displace-
ments between modeled and observed plume patterns relies on plume centroid
positions. With this technique, the difference in relative plume positions
and plume separation can be explored. A centroid location of a plume is
defined herein as the density-weighted maximum concentration coordinate
position (XC,YC). The plume centroid positions for the modeled and
observed plumes were determined at each time period from concentrations
at sites with Eq. (8).
Xc(t) - 2 Xj.Cj / Z Cj , (8a)
Y(t) = 2 y.C / 2 C (8b)
In Eq. 8, x.= -y^ are the latitude- longitude coordinates of each site ( j )
with a nonzero concentration (C) , where C represents either an observed or
modeled concentration over a particular time period (t). The summation is
performed over all sites separately with the observed and modeled plumes.
Using the results of Eq. 8, the downwind distances of the observed (DQ)
and predicted (D ) plume centroid locations with respect to the particular
release site were determined. In addition, the results of Eq. (8) were also
applied in order to compute the separation distance (Ds) between the
observed and modeled plume centroid positions.
These distance parameters have been advocated by Clark et al. (1988)
in an assessment of long range model performance. For example, the stat-
istical and test results presented earlier may not be able to distinguish
between the performance of two different models in situations where both
calculate zero concentrations at the same sites. That is, one model's
plume pattern may slightly miss sampling sites, while the plume from another
model may be displaced by a much greater distance from the same measurement
sites. With these plume distance parameters, the downwind distance ratio
59
-------
(D /DQ) and a normalized separation distance ratio (DS/DQ) can
be applied to more rigorously assess and distinguish between the performance
of different models. For example, values of D /DQ that are greater
than unity indicate the modeled plume traveled faster than the observed
plume, while ratios less than unity signify the modeled plume was transported
slower than the observed plume. On the otherhand, the values of DS could
implicate the speed and/or direction components of transport. However, if
both plumes were the same distance downwind (i.e. D /DQ=1), a separation
between the respective plumes would be entirely attributable to directional
errors by the model.
The results of these ratios for each CAPTEX episode are presented in
Figure 12a-f. Values have been plotted at the mid-point in time of
each sampling interval. Focusing on the CAPTEX #1-4 (Fig. 12a-d), most of
the values of D /DQ are less than 1, which reveals the modeled plumes
traveled slower than the observed plumes. In particular, the evidence
in some of these cases showed that the modeled plume tended to travel
more slowly than the actual plume during the first overnight period. The
sign convention adopted for the ratio of DS/DQ in these figures
specifies whether the modeled plume centroid was to the left (minus) or to
the right (positive) of the observed tracer plume. The most dramatic
shifts in DS/DQ also occurred during the nocturnal hours of the first
night. During the stable night hours, large vertical shears in both speed
and direction are expected to develop in association with the nocturnal
low level jet. In addition, graphical maps of plume concentration patterns
during the morning periods of the second day in these cases also verified
that the leading edge of the observed plume was impacting sites along arcs
considerably farther downwind than the modeled plume.
An examination of the MESOPUFF II code revealed that the modeled plume
was always transported by the lower wind field (mixed layer averaged) even
during the nocturnal hours. Transport of the modeled plume was not switched
to the upper layer wind field because with a surface-based nonbouyant
release as in this application, Z_ always remained less than Z^. Thus,
even though the mixed layer depth dropped to relatively low values at night,
Z^ continued to be greater than Zp. Plume transport by the mixed layer
averaged field, which at night is representative of the wind flow over a
relatively shallow layer, also explains why the observed tracer probably
outpaced the modeled plume during the nocturnal period. At night when
significant vertical shears in speed and direction are prevalent, the
mixed layer averaged winds were certainly slower than upper level wind
speeds which more likely advected the observed tracer plume. In addition,
plume separation was also magnified because larger differences in wind
direction between these layers often exist during stable conditions at
night. Examinations of nocturnal wind profiles during the CAPTEX
episodes revealed notable vertical shears in both speed and direction,
although there was variability from among the cases.
The results for the two Sudbury cases where the releases were at night
under moderate winds behind cold fronts differed somewhat from the previous
cases and each other. The results for CAPTEX #5 in Figure 12e were closer
to ideal as higher winds and smaller vertical shears were responsible for
transporting the rather narrow plumes rapidly off the U. S. east coast
(see Figure 8e for plume patterns). Results for CAPTEX #7 in Figure 12f are
60
-------
2.0
1.5
1.0
.5
0.0
-.5
a CAPTEX 1
-' i' i' i' i' i i' i' r i' i' r r r r r r i' i' i' r p r i'
x - DP/DO
o - DS/D0J
-1.0
X X
x
o
o o
o
X
o
X
12 16 20 00 04. 08 12 16 20 00 04 03 12
14 18 22 02 06 10 14 18 22 02 06 10
TIME (LSI)
2.0
1.5
1.0
.5
0.0
-.5
-1.0
b CAPTEX 2
' | ' | ' | ' | ' | ' | | ' |
X - Dp/D0
O - DS/D0
X
X X
xxx
X
X
o
O O O
o
o
o
o
o
"1
12 16 20 00 04 08 12 16 20 00 04 OS 12
14 18 22 02 06 10 14 18 22 02 06 10
TIME (LST)
Figure 12. Ratios of downwind distances of the model and observed
plume centroid locations (D /DQ) and separation
distance between centroid positions to the observed
downwind distance (DS/DQ) versus time of day starting
from the day of release. Negative and positive values
for DS/D0 denote the model plume position is to the
right and left of the observed plume, respectively.
a) CAPTEX #1 and b) CAPTEX #2 results.
61
-------
2.0
1.5
1.0
c CAPTEX 3
o
.5
0.0
-.5
-1.0
x
x - Dp/D0
o - DS/D0
X
"X"
X
X
X
o x
X
o o
o
o
o o
o
12 16 20 00 0-1 08 12 16 20 00 04. OS 12
U 18 22 02 06 10 14 18 22 02 06 10
2.0
TIME (LSI)
d CAPTEX 4
1.5
1.0
.5
0.0
-.5
-1.0
.' i ' i ' i ' i ' i ' i ' i ' i ' r r i ' i ' i ' i ' i ' i ' i ' i ' i ' i ' i ' i ' i
: x - Dp/D0
x o - DS/D0
X
X
X
X
X
X
X
o
o
o
o
o
o o
o
o
12 16 20 00 04. 08 12 16 20 00 04 08 12
14 18 22 02 06 10 14 18 22 02 05 10
TIME (1ST)
Figure 12. Same as 12a-b, except for c) CAPTEX #3 and d) CAPTEX #4
62
-------
2.0
1.5
1.0
e CAPTEX 5
<
.5
0.0
-.5
X
X
o
o
' I ' I ' I ' I ' I '.
- Dp/Do :
- DS/D0 :
X
- n I I 1,1 . I, I. I .1 , I, I. I ,1 , I , I , I ,1 , I. I, I . I , I , I, I ,| ,-
00 04 08 12 16 20 00 04. 08 12 16 20 00
02 06 10 14 18 22 02 06 10 14 18 22
TIME (LSI)
2.0
f CAPTEX 7
1.5 -
1.0
o
^.5
0.0
-.5
-1.0
i' i' ii'i'*' i' i' i' i' i'i'i' i' r i i' i i' rT i' i
x - Dp/D0
x o - DS/DO.
x
X
X
-x--
o o
00 04 08 12 16 20 00 04 08 12 16 20 00
02 06 10 14 18 22 02 06 10 14 18 22
TIME (1ST)
Figure 12. Same as 12a-b, except for e) CAPTEX #5 and f) CAPTEX #7
63
-------
in agreement with the overall plume patterns shown earlier in Figure llf
as the modeled plume traveled to the left (looking downwind) of the observed
plume and at a faster pace, which was a departure from the other cases.
The non-normalized values of these parameters are presented in
Figure 13 to show the actual difference between D and DQ and the
magnitude of DS. The results have been plotted against time after
release for all episodes. A notable feature is that the largest diff-
erences in downwind distance between the respective plume centroid loc-
ations appeared from 12-24 hours after releases with values around 200 km.
This time frame also coincided with the nocturnal period for the Dayton
cases. The largest values of DS were also found during this time period
as values near 300 km occurred among these cases. Some caution must
be exercised when interpreting the results for the later time periods
because the bulk of the observed plume probably already moved beyond the
sampling domain. Consequently, greater uncertainty exists with these
points as some values may not accurately reflect the true differences
between the respective plumes as the observed plume centroids were
derived from plume remnants still being sampled at the farthest
arcs from the release locations. Nevertheless, these results have
provided valuable quantitive information about the magnitude of spatial
plume displacements and helped focus on the probable causes for deviations
in plume transport by the model.
DIAGNOSTIC TEST RUN RESULTS
Additional processor and model runs were conducted in order to
exercise some of the optional selections for the wind field and dispersion
methods and to compare these results with those already obtained from the
operational default model runs. It was decided to conduct the diagnostic
runs for the Dayton cases only with the emphasis in the analysis on the
peak concentration results at the 300 km arc. This test strategy was
selected in order to uncover the extent of the differences in the plume
impact location and timing for the various wind fields and to identify
the cause of the model overpredictions of peak concentrations which were
pronounced at the 300 km arc sites. A single optional feature was exercised
in each of the diagnostic runs while all other default parameters and
methods remained unchanged. The input surface and upper air data files
were identical to those in the default runs.
Although MESOPAC II contains options for several different wind fields,
it was decided to limit the diagnostic runs to two alternative wind fields;
single-level wind fields for the surface and the 850 mb levels. Either of
these wind flows might be selected in certain model applications. Winds at
the 850 mb level are generally representative of the flow at the top of the
daytime mixing layer, and it is expected to provide a strong contrast from
the transport provided by a surface wind flow. MESOPAC II processor runs
were performed to produce meteorological files for each case with these
wind field selections.
Of particular interest are differences in the location and travel time
for the peak concentration and its magnitudeat the 300 km arc for the
different wind fields and how the results compare with those of the observed
plume. Tables 20-23 present the comparative results for CAPTEX #1-4,
64
-------
400
DOWNWIND DIFFERENCE
200
o 0
Q
I
CL
-200
-400
1 I i I I I I
i I i I i l i r
o
o
X X
o o
A
XOX° X
o a
x x
x x
X
I I I I
o
i i i i i i i
10 20 30 40
HOURS AFTER RELEASE
50
400
300
£
.^r
w
Q
100
I X
X
SEPARATION DISTANCE
l i i i i i i l I I i i i i i i i i i
O O
o
o
o
X
X
X
X
00, I
I I 1 I I I I 1 I I I I I
ire.
X
t I tilt
10 20 30 40
HOURS AFTER RELEASE
X|
J
J
50
Figure 13. a) Difference in the downwind distances of the centroid
positions of the modeled (D ) and observed (DQ) plume
patterns versus time after release, b) Separation distance (Ds)
between modeled and observed plume centroid locations versus
time after release. Symbols in both figures denote events for
each experiment; CAPTEX #1 (0), #2 (X), #3 (triangles),
#4 (squares), #5 (stars), and #7 (diamonds).
65
-------
respectively. The individual results will be discussed separately in
this section. The azimuth angle of each site with respect to true north
are also given to help quantify the angular difference in the relative
positions of plume impacts with the different wind fields. The position
of each site in these tables is consistent with the configuration of this
arc, as shown earlier in Figure 2, with site numbers increasing from right
(east) to the left (west).
Table 24 presents the hourly NWS surface wind observations and results
of certain parameters from the operational model runs for the release period.
This table is provided with the intention of giving supplemental information
for the interpretation of the model results. Surface winds in Table 24 do
indeed show that the Dayton releases all occurred under primarily southwesterly
wind flows. Consequently, for this wind flow and with the configuration
of the sampling arc, a modeled plume impacting a site to the right (left)
of the observed plume location on the 300 km arc would indicate a greater
influence by a more westerly (southerly) component in the wind field.
The results in Table 20 for CAPTEX #1 indicate that the location and
timing of impact of the peak concentration with the mixed layer (default)
wind field compares favorably with the observed results. Site 316 was
the location of impact for both the observed and model default peak concen-
tration and the transport times were also the same at 5 h after the start
of release. In contrast, the surface wind field caused the modeled peak
concentration to occur much later and at a site slightly to the left of
the observed results. The 850 mb winds transported the modeled plume
at about the same speed as the observed plume but directed it to the right
of the observed peak concentration location. These results are to be
expected when wind direction becomes more westerly with height which was
evident from the rawinsonde profiles; the wind flow at 850 mb had a greater
westerly component than the surface winds. It appears that winds averaged
over the mixed layer simulated the observed transport very well during the
afternoon period in this case.
The results in Table 21 for CAPTEX #2 are certainly somewhat different
from the CAPTEX #1 in the relative positions of plume impact. During this
experiment the model default peak concentration intercepted the arc much
faster and to the left of the location of the observed peak value. In fact,
several 'model-specific' sampler sites were included in the model run in
an effort to better define the modeled plume position in this case. The
default peak occurred at a site about 25° to the west (left) of the
observed peak. Mixed layer averaged wind directions at Dayton in Table 24
were indeed more southerly than the surface winds, which indicated why the
modeled plume moved to the left of the observed plume position, but this
did not explain why the mixed layer averaged wind possessed a greater
southerly component. However, the afternoon and the regular evening
(00 GMT) wind profiles at Dayton in Figure 14 appeared to show why the
modeled plume behaved in this manner. The 00 GMT profile .which was used
to determine the modeled winds, revealed that wind direction backed with
height from 230° at the surface to 250° near 1000, while the sounding
at 18 GMT indicated that wind direction displayed the more typical veering
with height during the release period. Consequently, d6 derived from the
surface and layer-averaged directions was positive from the 00 GMT
profile, which produced the stronger southerly component found in the mixed
66
-------
TABLE 20
COMPARATIVE RESULTS FOR DIFFERENT WIND FIELDS ON
PEAK CONCENTRATION/PLUME TRANSPORT AT THE 300 KM ARC FOR CAPTEX #1
RELEASE - 9/18 (1200 - 1500 LST)
320
318
SITE NUMBER
316 314
312
310
AZIMUTH (°) (24.5)
(36.2) (42.5) (45.3) (55.4)
(60.5)
0 1 425 1586 197
DAY/TIME? 18/1700 18/1700 18/1700 18/1700
TRAVEL TIME (h) 5 555
MODEL WIND FIELD
DEFAULT
SURFACE
850 mb
176
18/2300
11
0
116 2367 1568
18/1700 18/1700 18/1700
555
4456 752 163
18/2000 18/2300 19/0200
8 11 14
0
0
0
1961 642
18/1700 18/1700
5 5
$ TIME IS THE START HOUR OF THE SAMPLING INTERVAL
* TIME INTERVAL FROM THE START OF RELEASE AND TO BEGINNING OF
OF SAMPLING INTERVAL
67
-------
TABLE 21
COMPARATIVE RESULTS FOR DIFFERENT WIND FIELDS ON
PEAK CONCENTRATIONS/PLUME TRANSPORT AT THE 300 KM ARC FOR CAPTEX #2
RELEASE 9/25 (1205-1505 LSI)
SITE NUMBER
324* 323* 322* 321* 320 318 316 314 312 310
AZI.(°) (-1.5) (4.7) (10.8)(16.7)(24.5)(36.2)(42.5)(45.3)(55.4)(60.5)
0 - 64 1575 768 407 12 10
DAY/TIME 26/04 26/02 26/04 26/01 26/04 26/04
TRAVEL TIME (h) 16 14 16 13 16 16
MODEL WIND FIELD
DEFAULT 9 320 2767 1430 390 20 0 0 0 0
26/01 25/16 25/16 25/19 25/22 26/04
13 4 4 7 10 16
SURFACE 12 79 286 459 1106 1324 873 656 92 74
26/07 26/07 26/07 26/10 26/10 26/13 26/13 26/16 26/19 26/19
19 19 19 22 22 25 25 28 31 31
850 mb 0 0 0000 533 1002 303 0
25/19 25/22 2-5/22
7 10 10
* 'Model-specific' sites only
68
-------
TABLE 22
COMPARATIVE RESULTS FOR DIFFERENT WIND FIELDS ON
PEAX CONCENTRATIONS/PLUME TRANSPORT AT THE 300 KM ARC FOR CAPTEX #3
RELEASE - 10/02 (1400-1700 LST)
SITE NUMBER
320 318 316 314 312
310
308
306
AZI. (°) (24.5) (36.2) (42.5) (45.3) (55.4) (60.5) (68.8) (74.4)
Omax 304 0
DAY/HR 02/2200
TIME (h) 8
MODEL WIND FIELD
03/0100
11
DEFAULT
0 102 38 1416
02/2200 02/2200 03/01
8 8 11
SURFACE 258 10 3 11
03/0400 03/0400 03/0700 03/0700
14 14 17 17
850 mb
0
0
0
0
0
524 8
02/2200 03/0100
8 11
0
0
14 754 1357
03/0100 03/0100 03/0100
11 11 11
69
-------
TABLE 23
COMPARATIVE RESULTS FOR DIFFERENT WIND FIELDS ON
PEAK CONCENTRATIONS/PLUME TRANSPORT FOR THE 300 KM ARC FOR CAP
RELEASE - 10/14 (1100-1400 LSI)
SITE NUMBER
320 318 316 314 312 310 308 306
AZI.(°) 24.5 36.2 42.5 45.3 55.4 60.4 68.8 77.4
Omax 000 09 1350 33 0
DAY/HR 15/19 14/19 14/19
TRAVEL TIME (h) 888
MODEL WIND FIELD
DEFAULT 000 00 0 1448 145
14/19 14/16
8 5
SURFACE 65 799 1859 1492 231 119 0 0
15/07 15/01 15/01 15/04 15/04 15/04
20 14 14 17 17 17
850 mbOOO 0 0 00 0
M missing measurements
TEX #4
304 302
83.9 93.1
M M
0 0
0 0
626 81
14/16 14/19
5 8
70
-------
TABLE 24
SELECTED METEOROLOGICAL OBSERVATIONS AND MODEL RESULTS
AT THE RELEASE SITE OF EACH CAPTEX EXPERIMENT
SFC OBS. MODEL ML WINDS$
SITE- HR WS WD SPEED DIR PGT TEMP
DATE (LST) (m/s) (DegxlO) (m/s) (Deg) CLASS (°F)
(DAY-9/18)
#1 12
13
14
15
(DAY- 9/2 5)
#2 12
13
15
(DAY- 10/2)
#3 14
15
16
17
(DAY- 10/14)
#4 11
12
13
14
(SUD-10/25-26)
#5 23
00
01
02
(SUD-10/29)
#7 01
02
03
04
7.
9.
8.
8.
4.
5.
6.
3.
7.
5.
4.
5.
7.
7.
7.
8.
4.
3.
5.
5.
8.
6.
3.
6.
7
3
8
2
1
2
2
6
2
7
1
2
7
2
2
2
6
6
2
7
2
2
1
2
$ Mixed layer (ML)
* Ceiling height in
22
24
21
22
24
20
22
22
24
22
23
23
25
26
24
25
32
32
33
31
33
33
33
33
averaged
hundreds
13
16
15
14
9
12
15
10
17
13
10
12
14
14
14
16
4
3
5
5
17
13
8
13
.4
.0
.3
.3
.4
.8
.6
.1
.3
.7
.1
.4
.8
.5
.9
.3
.6
.6
.2
.6
.9
.0
.8
.0
winds at
of feet
224
245
216
224
230
189
206
205
266
248
258
257
269
280
264
271
319
319
329
310
326
326
339
326
.4
.0
.7
.8
.1
.5
.6
.2
.0
.7
.5
.6
.4
.9
.2
.9
.9
.9
.8
.0
.8
.9
.7
.9
4
4
4
4
3
4
4
4
4
4
3
4
4
4
4
4
4
4
4
4
4
4
4
4
the grid point nearest
above ground
86
87
89
89
66
68
70
69
81
82
81
81
49
51
53
55
37
37
39
39
32
32
30
30
CLOUD
COVER
(tenths)
0
4
4
4
10
10
10
10
0
Q
0
0
0
0
0
0
release
(250)*
(250)*
(250)*
(100)*
8 (11)*
8 (16)*
8 (18)*
8 (13)*
4
8 (30)*
8 (24)*
4
site
71
-------
STATION 72429 DATE: 9/25/83 TIME 18 GMT
.5000
^ 2000
^E
I-
LU
o \f
-TT^
$>
i
i
K.--fr"
0 50 100 150 200 250 300
SPEED (m/s) , DIRECTION (°) bottom axis
I-
I
-------
layer averaged wind directions in Table 24 for this case. Another wind
profile from a rawinsonde launch at 20 GMT was also very similar to the
18 GMT profile, so the vertical variation in wind direction exhibited in
the 00 GMT sounding did not occur during the time period of the tracer
release. This case demonstrated the impact on model results when the
upper air data were not representative of the conditions at an earlier
time. It is apparent that better model results might have been obtained
if one of the upper air profiles taken during the release period was
substituted for the 00 GMT profile, or if the model was capable of using
rawinsonde data at more frequent time intervals.
The largest spatial difference in the location of plume impact
between the observed plume and modeled plume with the mixed layer averaged
winds occurred in CAPTEX #3 (Table 22). The exact location and magnitude
of the actual tracer peak concentration could not be accurately judged
in this case since the bulk of the tracer plume definitely traveled to the
west of this arc. Some of the tracer plume impacted site 320 located on
the western end of the arc. It is estimated from Table 22 results that the
peak concentration location from the model default run was shifted at least
30° to the right of the position where the observed plume may have passed
at this downwind distance. The 00 GMT wind profile at Dayton in Figure 15
reveals that significant vertical shear in direction existed within the
mixed-layer in this case, as wind directions varied from 200° at the
surface to almost 270° at Z^. This large vertical variation in
direction also explains the large differences in plume impact time and
location between the model runs with surface and 850 mb wind fields. The
release period for #3, being slightly later in the afternoon compared to
the other Dayton releases, may also have contributed to the variations
in the plume's trajectory and dispersion. In particular, tracer gas
emitted near the end of the release period at 17 LST had much less time to
be mixed through the boundary layer as convective turbulence is already
less vigorous and is rapidly decreasing during the late afternoon hours.
Tracer measurements from aircraft flights provided additional
valuable information about the impact of vertical direction shear coupled
with the timing of the release for CAPTEX #3 on the transport and
dispersion of the tracer plume. An aircraft flight leg was conducted
across northern Ohio at 1500 m around midnight. A tracer plume was
discovered at this level with the highest concentration being 1466 fl 1
in the same relative position as modeled peak concentration from the
mixed layer wind fields. However, the real suprise occurred from the tracer
measurements from aircraft flight legs over southeastern Pennsylvania
during the afternoon of the next day, October 3. The vertical cross-section
in Figure 16 showed evidence of a coherent tracer plume aloft with a
concentration as high as 2961 fl 1 . Clearly, this portion of the
tracer plume was transported by westerly winds which existed in the upper
part of the boundary layer. Additionally, for concentrations to be so
high, this plume section had experienced little boundary layer-based
turbulent activity, which would have dispersed it and fumigated it to the
surface. Surface sampling sites measured zero concentrations in this area.
The multi-layer Lagrangian model by Davis et al. (1987), which contained
seven vertical layers up to nearly 3000 m, displayed much greater success
in simulating the broad spread and decoupling of the tracer plume pattern
due to the large vertical shears during this episode.
73
-------
STATION 72429 DATE: 10/ 3/83 TIMEOO GMT
.3000
O 2000 -
c
h
O
5
- 1000 -
1.4
200 220 2*0 250 280 300 320
SPEED (m/s) , DIRECTION (°) bottom axis
STATION 72429 DATE: 10/ 3/83 TIMEOO GMT
.3000
-10 -5
Figure 15.
TEMPERATURE (°C)
a) Wind speed (0) and wind direction (X) profiles and
b) temperature profile at Dayton, Ohio from CAPTEX #3
74
-------
CONCENTRATION (FL/L)
RELEASE 3
3 OCT 21Z
E
E
u
Q
ID
28
24
26
16
8
4
88688
AVERAGE SURFACE HEIGHT
41
42
44
LATITUDE
Figure 16.
Tracer concentration pattern from aircraft flight legs at different
levels during the afternoon of October 3, 1983 during CAPTEX £3
(Ferber et al., 1986)
75
-------
The results for CAPTEX #4 in Table 23 reveal that the default model
plume exhibited relatively good correspondence in time and position with
the observed peak concentration. The modeled peak occurred within 10°
to the right of site where the observed plume impacted and the respective
travel times were in agreement. Both results also show that the
tracer plume was narrow in this case. The results for the single layer
wind fields were similar to those in #1 and #2; the modeled plume
using surface wind fields traveled to the right and slower than the
observed plume and the 850 mb winds caused the modeled plume to impact
to the right of the observed peak concentration position.
Although the experiments from Sudbury were not examined in these model
test runs, it is of interest to assess the plume impact locations for these
two cases, especially on the north side of Lake Ontario. For CAPTEX #5.
the observed and modeled plumes (Figure 8e) impacted sites 752 and 652,
respectively, which were both about 400 km from Sudbury and just north of
Lake Ontario (Figure 2). More importantly, the difference in azimuth angle
with respect to Sudbury between these sites was about 8°. In CAPTEX #7,
the observed plume traveled to the right of the modeled plume (Figure llf)
and the angle difference in azimuth angle for the sites in this case was
about 13°. Interestingly, in these cases and in some of the Dayton cases
the angular separation of the respective plumes was comparable to the
resolution of the surface wind direction observations.
The upper air wind profile data collected during the release periods
and at the regular 00 GMT launch time were analyzed to further explore
reasons why modeled speeds and directions differed from those inferred from
the actual tracer plume measurements out to the 300 km arc. The results
in Table 25 were determined from profiles at Dayton and for Buffalo (BUF)
to represent the Sudbury cases. The notable values are the ratio of the
mixed-layer averaged to surface wind speeds (WSm-j_/WS0) and difference
between the wind directions. In particular, most of the afternoon wind
speed ratios were lower than values at 00 GMT. Wind direction differences
in Table 25 also tended to be larger at 00 GMT than in the earlier profiles.
Stronger turbulent mixing in the afternoon period likely reduced vertical
shears in speed and direction within the mixing layer, however, vertical
variations appeared to become stronger by evening. The changes over the
time period of these profiles were also consistent with the results reported
by Shreffler (1982). From analysis of a large wind profile data base, he
found that the diurnal variations of wind speed ratio and wind direction
difference over the mixed layer exhibited minimum values during the
afternoon period with gradual increases during the evening to maxima at
night. Thus, the mixed layer averaged fields computed by the model are
believed to be slightly overestimating transport speeds during the daytime
period, however, this feature was difficult to assess with the 3-h averaging
period for these cases. On the otherhand, there was some evidence of an
effect on wind direction in the model as the previous results indicated
that the default model plumes were shifted slightly to the right of the
observed plumes in CAPTEX #4 (Table 23) and CAPTEX #1 (Table 20).
The other important feature to be examined is the significant model
overprediction of peak concentrations reported earlier. The same dispersion
method, as outlined in Section 2, was applied in these model runs with the
76
-------
TABLE 25
RESULTS OF ANALYSIS OF OBSERVED UPPER AIR WIND PROFILES
FROM THE CAPTEX RELEASE DAYS
EXPT
1
2
3
4
5
7
NOTE
*
SITE DATE -TIME
(MM/DD-GMT)
DAY
DAY
DAY
DAY
BUF
BUF
I?
WDQ
WDZi
zi '
9/18-1800
9/18-2000
9/19-0000
9/25-1800
9/26-0000
10/2 -1800
10/3 -0000
10/14-2000
10/15-0000
10/26-1800
10/27-0000
10/29-1800
10/30-0000
f ml ml
(m/s) (Deg)
12.5
13.1
12.7
5.3
7.5
4.9
7.3
9.2
8.0
9.8
11.7
8.5
6.2
229
239
221
220
237
258
226
249
239
290
315
335
330
.0
.0
.5
.0
.9
.3
.4
.2
.8
.1
.0
.5
.5
!?S»1
WSo
2.4
2.0
1.8
1.7
3.0
1.1*
2.4
1.5
2.0
1.4
1.7
1.2
3.1
WD -WDml WD -WDzi
(Deg) (Deg)
1
-19
11
-20
12
-26
-26
-9
-19
9
-25
14
-0
.0
.0
.5
.0
.1
.0*
.4
.2
.8
.9
.0
.5
.5
-12.
-27.
-22.
-21.
11.
-39.
-55.
-30.
-38.
12.
-30.
6.
-8.
0
0
0
6
8
0*
0
6
6
4
0
5
1
(nj
1254
1855
1675
1294
1439
1808
1732
1217
972
2293
2070
1254
1175
- mixed- layer mean wind speed from observed profile data
mixed- layer mean wind direction from observed values
- surface observed wind speed at the rawinsonde site
- surface observed wind direction at the rawinsonde site
- observed wind direction at the mixing height
observed mixing height; winds averaged up to this level
Missing observed surface
this result may slightly
wind; values at next higher level applied;
underestimate the actual condition.
77
-------
different fields. In the three Dayton cases where the peak concentration
was readily resolved (Tables 20, 21, and 23), model predictions with the
default wind scheme were greater than observed peak values by varying
amounts. Since dispersion is dependent on travel time after 100 km in
the model, modeled peak values differed for the various wind fields.
Focusing on the default model results in Tables 20-23, the overpredictions
were smaller for #1 (Table 20) and #4 (Table 23) where model transport
times agreed with observed values. The most sizeable overprediction of the
peak concentration by the default model run occurred in CAPTEX #2 (Table 21)
as the modeled plume arrived much earlier than the observed plume.
Differences in the width of modeled and observed plumes were difficult
to assess even for the site spacing of this network, however, the results
at 300 km suggested that modeled and observed plumes exhibited comparable
horizontal spread. Nevertheless, the consistent model overprediction
strongly implied that dispersion was greater for the observed plumes.
Since puff growth is controlled by dispersion parameters, whose variations
are dependent on stability class, the results for PGT class in Table 24
are illustrative of a possible cause for the model overpredictions. Neutral
stability (PGT class 4) was often found during these afternoon hours when
unstable conditions might be expected, although based on the observed wind
speeds the specification of a neutral stability class appears appropriate.
A further examination of processor output revealed stability class 4 was
indeed prevalent at the grid points in Ohio during the afternoon hours.
Consequently, the initial growth of the model plume, particularly in the
vertical dimension, was much more limited due to the significantly slower
increase in the dispersion parameters under neutral stability when compared
to that under the unstable PGT classes (Turner, 1970). In particular, the
coefficients of the power law curve for az for neutral stability in the
model were derived from a fit to the so-called D2 curve in Pasquill (1974),
which is the neutral curve for the vertical dispersion parameter in Turner
(1970). However, greater vertical dispersion in the model would have
been produced under the Dl curve for neutral stability in Pasquill (1974).
In any event, the principal cause for overprediction by the model in this
application was attributed to underestimation of vertical dispersion. More
evidence to support this explanation will be given next from additional
model test runs.
Diagnostic model test runs were also performed to assess the impact
on peak concentrations by applying optional variations to the default
dispersion method or selecting the uniform vertical mixing feature in the
model. Results in Table 26 for CAPTEX #1 were representative of those
found in model runs for the other Dayton cases. The peak concentrations
at the plume centerline with the uniform vertical mixing option were lower
than values from the default model run. Since the observed peak concentra-
tion was 1586 fl 1 l in CAPTEX #1 (see Table 20), the peak value using
the vertical mixing option was certainly more comparable and suggested
that distributing the plume vertically over the depth of the mixed layer
was more representative in this case. In fact, a further examination of
operational model output indicated that crz was much less than Z-
at the 300 km arc, which revealed the modeled plume had not yet become
well-mixed even out at this distance. Thus, it is evident that more rapid
growth of az in the model would have provided an improvement in the
prediction of peak concentrations.
78
-------
TABLE 26
PEAK CONCENTRATIONS FOR CAPTEX #1 USING VARIOUS DISPERSION OPTIONS
MODEL RUN
SITE NUMBER
320 318 316 314
312
DEFAULT METHODS* 0
UNIFORM VERTICAL 0
MIXING OPTION@
CROSSOVER DISTANCE* 0
AT 50 km
CROSSOVER DISTANCE* 0
AT 10 km
116 2367 1568 0
69 1446 980 0
133 2586 1722 0
126 3075 2020 0
* PGT dispersion curves applied out to 100 km, time dependent growth
according to Heffter (1965) at greater distances.
@ Puffs vertically distributed over the mixed-layer after release
instead of az growth.
# Downwind distance where puff dispersion changes from PGT curves to
time dependent method.
TABLE 27
COMPARISON OF OBSERVED AND
TIME
N
ZiO
ZiP
d
AAR
R
0>P
F2
CC -
F2 -
ALL
(GMT) 18
59
(m) 1456.
±533.
(m) 1145.
±317.
(m) 311.
(m) 468.
0.22
(%) 66.1
(%) 84.7
PAIRS
00
73
1479.
±496.
1166.
±326.
312.
471.
0.15
67.1
89.0
cloud cover
percent of observed
0 < CC
18
38
1368.
±437.
1244.
±270.
124.
336.
0.35
57.9
92.1
MODELED MIXING HEIGHTS
< 8/10
00
49
1369.
±431.
1191.
±339.
177.
402.
0.22
59.2
95.9
and modeled values
0 < CC < 4/10
18 00
18
1300.
±453.
1355.
±324.
-56.
283.
0.56
44.4
88.9
within a
28
1307.
±409.
1313.
±331.
-6.
340.
0.34
39.3
100.
factor of 2
79
-------
Additional results in Table 26 revealed that reducing the crossover
distance caused even higher peak concentrations compared to the default
distance (i.e. 100 km). Clearly, these results indicated that the time
dependent dispersion formulas provided even slower growth than that offered
by PGT class 4 in this application.
Since it was noted earlier that accurate specification of Z^ was also
important to both transport and dispersion in this model, a brief examin-
ation of the modeled mixing height is warranted. The diurnal variation
of Z^ at the grid point closest to Dayton over the two day period of
CAPTEX #1 is depicted in Figure 17 to illustrate the typical temporal
behavior and magnitude of mixing height derived by the MESOPAC II processor,
The most dramtic change in Z^ occurred around sunset when a switch in
methods takes place. Mechanical Z.;'s at night were generally much lower
than the daytime convective Z-'s which were primarily controlled by the
variation in the sensible heat flux. However, the term containing the
thermal gradient aloft in Eq. 1 also appears to be counteracting the
surface sensible heat flux term in this case. This effect is attributed
to the gradual decreases in Z^ in the late afternoon hours prior to the
substantial drop observed at sunset.
Of particular interest is how well the modeled mixing heights compare
to observed values. Therefore, a data set of observed and modeled Z^'s
was developed. An observed Z^, defined as the height of the elevated
temperature inversion base, was paired against a modeled value from the
closest grid point to a rawinsonde site at the same hour. Observed
values were easily obtained from the temperature profiles because a
strong i-nversion was present during the CAPTEX episodes (e.g. Figure 15).
Statisitical results were computed for Z^ pairs at 18 and 00 GMT. For
the latter time, however, it was decided that the modeled convective Z^
at 23 GMT was appropriate to compare with an observed value since sunset
had already occurred by 00 GMT at some sites.
The results in Table 27 contain the results for both time periods for
separate categories of cloud cover. The model performed best under clear
to partly cloudy conditions which generally prevailed, as the bias (d) was
smaller and the correlation was higher than for the mostly cloudy cases.
It is apparent that the underpredicted Z^_ for the few cases when overcast
conditions prevailed. Finally, no strong conclusions about the general
applicability of this method to model Z^ should be drawn because the data
set was relatively small and represented a rather limited set of temporal
cases and meterological conditions.
80
-------
MESOPAC II MODELED PARAMETER
00
zuuu
1500
T
IXiNG HEIGHT
o
0
0
500
<
0
1 1 1 1 | 1
<> <$>
i
/
i
i
M J
' \ r & I
- ? * \ ' \ ~
\
f
j ' q> <>
^ 9 \ / /
- ^ , ' i#^ ^ <>_
i^ ix ^ X V^ <*> / /
f i ' \ '
1 ^ i
i i t ill
o 10 20 oe 16 o;
Figure 17.
TIME OF THE DAY (HOURS)
Hourly variation of mixing height (ZjL) from the MESOPAC II
for Dayton, Ohio for September 18-19, 1983 of CAPTEX #1.
-------
SECTION 5
MODEL SENSITIVITY RESULTS
Flexibility built into the design of model code can be advantageous to
the testing of various model components. MESOPUFF II contains considerable
flexibility which allows certain changes in methods or variations in tech-
nical parameters to occur through input values (Scire et al., 1984b).
Nevertheless, the intent of this testing effort was not to conduct a comp-
rehensive sensitivity study -of the various model components. Practical
considerations of time and computer costs necessitated a limit to the
number of model test runs for selected model components. It was decided
to place major emphasis of this sensitivity analysis on the dry deposition
and chemical transformation algorithms, which were not exercised in the
CAPTEX evaluation effort. Furthermore, no evaluation of the methods for
these processes was possible during the development effort that led to
this version of the model. Finally, there was a desire to report for the
first time the impact from some of the options in this model.
The main purpose of this sensitivity effort was to investigate the
impact on surface concentrations for a typical short-term averaging
period (i.e. 24 h) from selected changes in methods available in the
model or from differences in particular key technical parameters involved
in the two model components. Differences in concentrations between the
test cases and a model base case are emphasized herein because Morris et al.
(1988) already examined the responses of deposition velocity and transform-
ation rates of the modeled pollutant species to ranges in atmospheric
parameter variations, land use types, and for differences in technical
parameters involved in these model components.
Although various types of sensitivity methods are available (Gelinas
and Vayk, 1979), the test procedure adopted for this effort consisted of
varying a single option or parameter in a model test run while holding all
other inputs and model features the same as those in a base case. This
testing strategy permitted a comparative assessment of the relative impact
of the variation in each test run to the base case results. Consequently,
all model test runs were performed with the same input meteorological fields;
namely, the gridded results produced by MESOPAC II in the default mode
for CAPTEX #1.
The identical source characteristics and emissions from a single
realistic elevated point source were input to each model run. Table 28
contains the source information and emission rates used in all model runs.
The location of this single source was set to the same grid coordinates
as the Dayton tracer release site. The elevated source height and strong
buoyancy of the plume also exercises plume rise and fumigation methods in
the model. A single source was chosen because the analysis and interpreta-
tion of the effects on ground-level concentrations from one point source are
more straightforward and model runs are less expensive than for multiple
sources. These test runs focused on computing SC^ and sulfate (SO/)
concentrations with the emission rates specified in Table 28.
82
-------
TABLE 28
SOURCE CHARACTERISTICS AND EMISSION RATES
FOR ALL MODEL SENSITIVITY TEST RUNS
Point source location - Grid coordinates of Dayton tracer release site
Stack height - 304 m AGL
Stack diameter - 8m
Exit velocity 22 m/s
Exit temperature 407 K
S02 emission rate - 7064 g/s
SO^ emission rate 53 g/s
83
-------
Several other aspects of the model base case and test runs should also
be identified. All model runs commenced at midnight and model execution
continued for a 48 h period. Source emissions in Table 28 were continuous
over the 2-day simulation period, which was of sufficient duration for the
source plume to travel across the entire model domain while being subjected
to the various physical processes. Additionally, pollutant concentrations
were computed at grid points on the model sampling grid which was set to
the same size as the meteorological grid in the CAPTEX application. It
was desireable to analyze concentrations on the denser, uniform model grid
in order to take advantage of the capabilities of the MESOFILE II postpro-
cessor program. Computational procedures were specially designed into this
program to perform various analyses on different gridded concentration
fields from independent model runs.
The methods described in Section 2 for dry deposition and chemical
transformation were applied in the model base case run. The 24-h average
concentration fields of SOo from the model base case run for Day 1 and 2
are displayed in Figures 18a and 18b, respectively. The results in Figure 18a
show that the plume had reached the eastern edge of the model grid by the
end of Day 1. The 24-h average plume patterns on both days are similar
because of prevailing southwesterly winds within the mixed-layer with stronger
westerly winds in the upper layer field. The highest SC^ concentrations
are found close to the source and values generally decrease with distance
downwind due to the dispersion, and both deposition and transformation
processes reducing SOo levels in the plume.
The 24-h sulfate concentration patterns for the two simulation days from
the base case run in Figure 19a and 19b are identical to those for SC^,
except SO^ concentration levels are considerably lower. However, in
contrast to the results for SC^, highest sulfate concentrations are found
several hundred kilometers downwind of the source on both days. This is
a consistent feature for some pollutants, such as sulfate, which is primarily
formed from chemical reactions during transport across a region.
Since the plume patterns from all model test runs were also identical
to the base case distributions because the same transport and dispersion
conditions were applied in all cases, concentration differences were
examined from the set of grid points defining the plume pattern. Thus,
analyses by the postprocessor was restricted to the grid points with nonzero
concentrations. The results consist of percent deviations of a test case
concentration (Ct) from a corresponding model base case value (CV) .
The results of peak and plume average concentrations for SOo and
SO^ over the two 24-h periods from initial model test runs are given in
Table 29. The peak concentration was defined as the highest value found
from the entire gridded plume pattern over the time period, while average
plume concentrations were computed from all nonzero values. These initial
test runs were conducted to examine the impact on concentrations when one
of these physical processes was not modeled, which represents an extreme
situation. The results in Table 29 show minor differences in peak SOo
concentrations when either deposition or transformation processes were
omitted. This is to be expected since peak SC^ concentrations were gener-
ally found close to the source where these processes have not yet had much
84
-------
DATA READ FROM MESOPUFF OUTPUT FILE UHIT! 20 RUIISTHEAtl: 1
VERSOII* 4.0 lEVEL=a70421 N5rR=83 IISQAY = 261 II5IIR; 0 HM>VTS= 48 IAVG* 1 IIPUF= 4 H5AttAO= 4 IELHET=30 JEIHET=19
DGRIOa 37000.0 IASTAH* 1 IASTOP=JO JASTAR= 1 JASTOP-19 ISASTR- 1 ISASTP=30 JSASTR* 1 JSASTPM9 nESIIOII- 1 IIPI5- 1
IIAREASi 0 HREC= 0 IPRIIIF = 2 tC»USS=T LCIIEM=T LDHY=T LWET=F LPHIIIT=T L3VL=T LVSAHP-T HSAHP= 4.00 L5GnlD=T
II5PEC= 2
XREC=
YREC=
ERIDOCO RECEPTOR COIICEHTRATIOIIS
MULTIPLY ALL VALUES BY 10 « -3
YEAR: 83 DAYI 262 ENDING HOUR I 0 POLLUTAHTt 1
19 \ 0 0 0
17 \ 0 0 0
15 \ 0 0 0
« UK 0 0 0
13 \ 0 0 0
12 \ 0 0 0
11 S 0 0 0
10 \ 0 0 0
a \ o o o
7 \ 0 0 0
6 \ 0 0 0
5 \ 0 0 0
4 \ 0 0 0
2ND 0 0
0000
0000
0000
0000
0000
0 0 0 0
0000
0000
000 6950
0 0 1999 46
0
0
0
0
0
0
0
0
0
0
0
6123
1366
0
0
0
0
0
0
0
0
0
0
0
0
299
5364
5138
686
0
0
0
0
0
0
0
0
0
0
0
344
2051
2700
2051
0
0
0
0
0
0
0
0
0
0
0
0
97
1622
1975
2415
615
0
0
0
0
0
0
0
0
0
0
9
1015
1753
1249
1108
666
0
0
0
0
0
0
0
0
0
1
292
1461
1950
1253
52S
707
0
0
0
0
0
0
0
0
51
941
1697
679
471
920
1259
4
0
0
0
0
0
0
0
223
1272
1170
195-
157
1339
6ZB
14
0
0
0
0
0
0
0
0
13
516
1257
428
295
77
1610
1219
593
21
0
0
0
0
0
0
0
75
B29
766
174
178
114
2086
2959
195S
159
0
0
0
0
0
0
0
213
1027
441
225
105
164
2687
2570
504
148
0
0
. 0
0
9
356
897
281
164
45
595
2984
1371
503
324
14
0
0
0
0
24
408
603
100
91
370
1999
2644
749
360
55
0
0
0
0
40
367
349
130
235
1519
3419
1775
470
143
0
0
0
0
0
76
363
232
129
429
2017
2775
10'i9
287
50
0
0
0
0
122
320
147
146
SOS
1401
1234
39S
eo'
0
0
000000
180 150 43 5 0 0
301 199 60 9 1 0
131 113 70 29 13 6
181 207 183 120 70 32
590 60S 467 283 155 69
1085 873 552 281 140 62
618 404 228 106 52 24
123 61 33 IS 8 4
0
0
0
2
»
19
IB
a
2
0
0
11
12
14
17
18
Figure 18.
Average 24-h SC>2 concentration fields from the model bas$
for a) Day 1 and b) Day 2. All values in units of pg m"-
case run
-------
GRIDDEO RECEPTOR COIICEIfTRATIOtia
MULTIPLY AIL VALUES BY 10 " -Z
TEAR" 85 DAY" 263 EIOIIIC HOUR! 0 POLLUTAJfn 1
19 \
10 \
17 \
15 \
14 \
1J N
00 12 \
11 \
10 \
9 \
a \
7 \
6 \
5 \
\
1 \
\
1 S
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
*
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
*
0
0
0
0
0
*
0
0
0
0
0
0
0
0
0
0
0
»
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
1376
0
0
0
0
0
0
*
0
0
0
0
0
0
0
0
0
0
0
795
2
0
0
0
0
0
0
*
0
0
0
0
0
0
0
0
0
398
365
0
0
0
0
0
0
*
0
0
0
0
0
0
0
2
353
251
309
0
0
0
0
0
*
0
0
0
0
0
0
7
286
171
229
121
0
0
0
0
0
0
0
0
0
0
7
100
153
109
366
196
0
0
0
0
0
0
0
0
0
2
63
95
42
273
319
225
0
0
. 0
0
0
0
0
0
0
16
63
26
101
440
329
171
0
0
0
0
0
0
0
0
0
2
22
47
78
307
379
277
152
1
0
0
0
0
0
0
0
0
0
2
17
87
180
365
310
202
123
0
0
0
0
0
0
0
0
0
0
1
25
124
262
355
302
305
89
0
0
0
0
0
0
0
0
0
0
2
54
140
218
263
321
329
58
0
0
0
0
0
0
0
*
0
0
0
10
86
130
160
231
359
320
33
0
0
0
0
0
0
0
0
0
1
23
94
110
132
229
420
242
16
0
0
0
0
0
0
0
0
0
3
34
ea
96
122
265
434
159
7
0
0
0
0
0
0
0
0
0
6
41
79
60
117
260
367
114
5
0
0
0
0
0
0
0
1
14
45
70
84
125
til
305
95
0
0
0
0
0
1
10
17
37
53
65
83
139
271
271
89
0
0
0
1
«
45
4S
46
44
65
11
156
261
222
71
0
a
28
55
41
32
40
78
11»
166
199
135
40
17
40
47
39
36
48
97
153
174
139
68
17
21
41
55
47
39
55
108
161
164
101
J7
7
20
37
47
41
42
65
111
"153
138
7J
2Z
4
5
15
26
36
38
45
67
104
133
110
51
13
2
0
2
6
14
22
26
31
45
70
88
69
30
7
1
0
11
16
24
25
26
Figure 18b. Same as 18a, except for Day 2.
-------
00
DMA RtlO MOH MLlOruF? OUTPUT MLl " UMITl 20 HUhlTREAHi 1
VfllOM* 4.0 LCveL-870421 HSTR-Bl HSOAT*2A1 HSHA* 0 NAOVIS* 41 IAVO 1 WPUP- 4 MSAHiD" 4 IfLHCV'10 J[LH(T"1t
oc«io- 17000.0 mm- 1 iiiiar-io JAUAI* t j»uor-i» HAITI- i UASTF-SO jusn- i jiisir*i> HMHOK- i NMJ- 1
lAlttl- 0 UICC- 0 IrtlHf 2 LCIUll-I LCHtlfl LOII-T LUtlT LriIHI-I L1VL-T LVSIKP-T V1INP- 4.00 UCIID-T
IMC* 2
III C-
MIC"
CIIODCD IfCtMOl CONCENIIITIOHS
HULIIPLT ILL VILUM IT 10 -4
ItAHl II D»l 242 [HOIKS HOUII 0 POUUIAHTl 2
If \
II V
17 V
It V
11 V
H V
II \
12 V
II V
10 V
» V
V
7 V
1 V
1 \
4 V
1 \
> V
1 V
0
0
a
0
0
0
0
a
0
a
0
a
a
a
0
0
a
a
0
a
0
0
a
a
0
0
0
0
a
a
a
a
a
a
0
a
a
a
a
a
a
o
0
a
a
a
0
a
a
0
a
0
o
a
0
c
a
0
a
a
a
a
a
0
0
0
a
a
a
a
D
0
0
0
a
0
0
a
a
0
0
a
a
a
0
0
0
a
0
0
0
0
0
a
0
0 0
0 0
a o
a a
o a
a a
o a
a a
a o
a o
o a
0 0
o o
0 0
0 2172
2«2 )
o a
a a
a o
a o
0 0
0 0
0 0
a o
a o
a o
0
a o
a a
0 0
a 119
0 111
21»l 21*
](» 10
a o
0 0
0 0
0 0
0
0
a
0
a
0
0
0
0
0
211
1441
1741
1210
a
0
0
0
0
a
0
0
0
a
a
0
0
0
10
127*
11»7
1047
174
0
0
0
0
0
0
0
a
0
0
0
0
0
i
91*
1141
1010
414
41*
a
0
0
a
0
0
a
0
0
a
0
0
i
301
1114
1144
14*
470
121
0
a
0
0
0
0
0
0
0
0
a
0
42
1041
117*
724
412
(11
101 1
1
0
0
0
0
0
0
0
0
a
0
0
271
11B1
11*7
270
227
1>21
194
12
0
0
0
0
0
0
0
0
0
0
21
791
11*4
111
«7I
112
1177
11*1
1*0
20
0
0
0
0
0
0
0
0
0
121
nti
1114
247
114
14*
2711
14**
2241
141
0
0
0
a
0
0
0
0
0
111
1471
1*2
414
202
271
4174
1*4*
417
144
4
0
a
a
0
0
0
0
14
541
1212
111
114
tl
114*
lltl
2411
4*7
111
1!
a
0
0
0
0
a
0
14
744
04
2*1
207
114
42*1
1111
1444
111
ri
0
0
0
o
- -o
0
0
0
112
110
414
211
401
J750
7*0*
l»»i
1011
217
7
' ' 0
0
0
0
- o
0
0
4
211
1*4
~ 411
2*1
1111
"ini
4127
2111
4*1
111
a
o
0
0
0
- - o
0
o
" 11
310
r«r
Jit
141
'11*1
'Jir7
1222
10(2
101
1*
0
' ~~ 0
0
0
0
0'
0
1
11
441
' III
~l»4
471
1111
21*4
1410
11*
24"
0
o ~
0"
0
0
0"
a
i
J71-
111
417
241 '
172
1T04
III*
1100
172
' 1 "
o
0
-o-
0
0
0
0
0
14
tt
111
171
14}
1J»7
1401
411
4
"4
0
o
.... 0
0
0
a
a
a
i
11
20
1*
1*1
til
*04
11*
10
._ j
0
0
0
0
0
0
0
0
0
0
I
41
117
III
102
lit
2*
' I
0
0
- o
0
0
0
0
0
0
0
0
24
121
211
214
tl
14
1
0
0
0
0
0
0
0
0
0
0
a
0
7
14
74
71
10
4
1
0
0
0
0
0
Figure 19. Average 24-h sulfate concentration fields from the model base case
run for a) Day 1 and b) Day 2. All values in units of /ig m .
-------
CO
00
HUlTTfLY III VilUM IT 10
It X 0 0 0 0
\4444
II \ 0 0 0 0
17 X
\
It V
\
1 J V
\
14 V
V
1) V
\
11 V
X
11 V
10 \
V
1 V
V
7 X
1 \
J \
\
4 \
X
1 X
7 X
X
1 X
-x
0 0
0 0
0 0
0 0
4 t
a a
0 0
4 4
0 0
0 0
o a
a o
0 0
0 0
0 0
0 0
0 0
0 0
»
0 0
1 I
- o
0
0
0
0
0
»
0
0
0
0
0
a
0
0
0
0
»
0
1
0
0
n
0
0
0
t
a
a
0
0
0
0
0
0
0
0
0
4
-4
0
0
0
0
0
Q
t
0
0
0
4
0
0
0
0
0
0
0
0
0
0
0
t
0
0
0
0
a
0
0
0
t
0
0
0
0
0
0
ion
0
0
a
0
0
0
4
0
0
a
a
0
*
0
0
0
t
0
0
0
«
ZOM
1
0
0
0
0
0
0
0
0
4
0
0
0
0
a
0
4
0
0
1421
2221
0
0
0
0
0
0
4
0
0 '
a
0
0
0
0
a
a
141
iro
1211
0
0
0
a
0
0
*
0
0
0
0
0
0
0
0
70
4
1 J»2
1)12
211)
t
go«
4
0
0
0
0
10
0
0
a
0
0
a
0
0
0
71
141f
1420
4
I4i4
322)
1419
4
0
0
0
0
11
0
0
0
0
4
0
a
0
a
21
410
10)2
4
7U
101.9
1110
1910
0
0
0
0
12
0
0
0
0
0
0
0
1
220
901
1)0
4
1171
4
4112
1 111
1211
)
0
0
0
1 1
0
0
0
0
0
0
0
1)
114
199
1297
4
)f»J
4
1IM
1 717
MM
1
0
0
0
It
0
0
4
0
0
4
0
0
1
4.2
IK
1171
2»92
1741
2)40
2071
II
2
0
0
0
11
0
0
0
0
4
0
0
0
20
112
1147
J71I
094
2911.
2))2
It)
0
0
4
0
0
11
0
0
0
0
4
0
0
0
14
111
2274.
2114
Jiff
)12f
>I47
4.10
0
0
0
0
17
0
0
0
0
0
0
1
Mt
1404
2174
2112
2114
4)10
1)10
111
0
0
0
0
11
0
Q
0
0
0
0
21
417
lilt
111)
2001
lilt
1111
1274
111
0
0
a
0
11
ii i
0
0
0
0
0
0
41
101
1411
1111
2017
4110
7771
0
0
0
0
0
1
ft
4
111
1711
1190
4
2411
1171
7111
2712' 74tO
111
a
0
0
0
10
Ml
0'
0
' 0
0
21
0
0
0
0
4
19
111
4
121
1U12
1112
1200
4
7009
4
9020
2111
204
o
0
0
0
22
0
0
12
<7
111
211
120
4
711
1011
19111
4197
noi
17)0
2»4»
114
2
0
0'
" " 0
21
1
4
21
151
444
712
714
721
4
790
I2»2
21(1
1144
tJ17
1212
2117
127
' ' I
0
0
0
14
72
1)9
)1»
H2
142
741
If t
1004
2000
1140
1112
4
714)
1401
11)1
1»7
- o
Q
a
-- o
- z>
72
1)7
714
f 11
99
9)
4
1421
29)0
4111
1102
4
5)27
2110
711
1)
o
0
0
a
21
124
417
1)1
1071
4
11)1
1121
111*
4
irgo
11)»
1111
19tl
4
l»t«
1)12
lit
21
0
0
0
0
27
147
47)
12
1021
110f
4
1192
4
1429
4
2204
1147
4
11)4
4tll
*
2142
ttl
171
1
o
0
0
0
21
111
111
12)
111
«2
I24t
«
1111
4
2172
1147
4414
1111
4
If 10
11)
(7
0
0
0
a
0
2f
4
If
)l
11
17
If
11)
111
2471
4
217
211
109
27
1
)
Figure 19b. Same as 19a, except for Day 2.
-------
TABLE
RESULTS OF MODEL SENSITIVITY
24-HOUR BASE CASE PEAK AND
MODEL RUN
VARIATION IN
PEAK SOo
(cb cj/cb (%)
DAY 1 DAY 2
BASE CASE
No Dep . only - 1
No Transf. only 1
No Vert. Gauss. 168
Distribution*
1- layer mode* -0
BASE CASE
No Dep. only -8.
No Transf. only 91.
No Vert. Gauss. 22.
Distribution*
1- layer mode* 0.
.6 -0.6
.2 -0.5
.5 -97.7
.1 0.0
PEAK S04
(cb - ct)/cb (%)
4 -28.5
3 89.3
6 21.5
0 0.0
29
RUNS: COMPARISONS WITH
MEAN CONCENTRATIONS
VARIATION IN
PLUME AVERAGE S02
(cb - ct)/cb (%)
DAY 1 DAY 2
-15.5 -24.6
-7.2 -8.8
3.2 25.9
-2.4 -2.9
PLUME AVERAGE S04
(cb - Ct)/Cb <%>
-8.6 -17.4
93.1 94.0
14.7 26.5
-1.7 1.9
NOTE: Cb - base case concentration
Ct - test case concentration
* Both deposition and transformation applied as in default mode
89
-------
impact on S02. However, there was a greater effect on the average SC>2
concentrations since these values were derived over the entire plume pattern
For example, in the model run where no deposition was permitted, the average
plume concentration on Day 2 was nearly 25% higher than the base case con-
centrations. The results in the case with no SC>2 loss by chemical trans-
formation indicated that the mean plume concentrations were less than 10%
greater than the base case. On the otherhand, it is evident from Table 29
that transformation has a much greater impact on SO^, as peak concentrations
without transformation were around 90% lower than those in the base case.
In this simulation, the peak SO^ concentrations were located from 500-
700 km downwind of the source position.
Other results in Table 29 are from the model test run with no vertical
Gaussian dispersion (i.e. uniform vertical mixing). In this case, peak S02
concentrations were particularly sensitive to the type of vertical
dispersion technique. Peak values were substantially higher than those in
the base case using the Gaussian vertical dispersion parameter because the
plume became distributed immediately through the entire mixed-layer under
this optional method. Since this option feature produced higher peak
concentrations and slightly greater deposition near the source, lower mean
plume concentrations were obtained. For Day 2, the impact on the mean
concentration is comparable, except in an opposite sense to the results
without deposition. Peak and mean SO^ concentrations were reduced
under this dispersion option. In particular, the peak SO^ concentration
for both days was more than 20% lower than the base case values. In contrast,
there appeared to be little impact on peak and mean concentrations in the
1-layer model mode where the source depletion method for dry deposition
was applied as pollutant mass over the entire vertical extent of the plume
was subjected to dry removal.
Additional model test runs were performed to determine the impact
of prescribed changes in rs for S02 and sulfate. The processor
results indicated that PGT class 4 (neutral) was often determined to
represent the stability during the afternoon periods. The surface
resistances for SC>2 f°r tne various land use categories is 100 s/m for
the unstable classes and 300 s/m for neutral stability. During the
daytime period, rs is generally the dominant resistance to pollutant
uptake since the sum of the two atmospheric resistances are often much
less than rs. A model run was performed with rs redefined to be
100 s/m for neutral class, which essentially was an attempt designate these
cases as being unstable in the model. However, the results in Table 30
show that the effects of this change are rather small on the peak and
plume average concentrations of S02 and SO^. On the otherhand,
the results for a test run where the default value of r for SO^ was
modified from its default value of 1000 s/m to 100 s/m reveal comparable
impacts on peak and mean SO^ concentrations. No differences occurred
in 862 values, but the peak and plume average SO^ concentrations
for this model run were between 15-20% lower than base case values. The
current default rs value limits the maximum Vj for sulfate to 0.1 cm s"^,
which might be too low based on some experimental results and the analysis
reported by Morris et al. (1988). Thus, a refinement in the formulation
for rs for SO^ appears to be warranted.
The background ozone concentration (03^^) is an important parameter
90
-------
TABLE 30
RESULTS OF SENSITIVITY RUNS FROM
IN SELECT KEY TECHNICAL PARAMETERS
VARIATION IN
MODEL RUN PEAK SOo
-------
in the determination of the transformation rate (k). Clearly, ozone
generally exbihits a strong diurnal variation and large differences
in ozone levels occur between air masses. Consequently, model test runs
were conducted to assess concentration differences to a ±50% variation
in 03b from its current default value of 80 ppb. In fact, k is larger
by a factor of two when 03^ - 120 ppb compared to 40 ppb with all other
parameters being equal (Morris et al., 1987). The results for these model
runs in Table 30 reveal little effect on SC>2, however, the changes in
SO^ peak and mean concentrations are more important. Rather clean and
polluted air masses are represented by 03130- being 40 and 120 ppb,
respectively. Peak and plume average SO^ concentrations in the test
cases appeared to be equally affected, as results with 0^^ - 40 ppb
were 18-19% lower and those with 03b = 120 ppb were 15-16% higher
than base case values. Thus, the sulfate concentration levels were fairly
sensitive to variations in Oj^p- values with this transformation method,
Therefore, representative ozone levels from measurements made within the
model domain should be input for particular applications.
The height of a pollutant release has a strong influence on the
ground-level concentrations; the rationale for higher stacks during the
past two decades was that peak concentration reductions would be produced
with higher stack heights (zs) . A model run was performed with
zs = 150 m, which is approxiately one-half the height of the stack used
in the base case. The final results shown in Table 30 reveal that
peak S02 concentrations indeed were considerably higher on both days
compared to base case values, while increases in the plume mean SC>2
concentrations were not as dramatic. A notable result with the lower
stack, however, is that peak sulfate values were 21% and 30% lower than
base case peak values on Day 1 and 2, respectively. Thus, the change in
stack heights may also have had an impact on secondary pollutant levels.
It is recognized that these results represent only a small subset of a
much larger group of possible model test runs which could explore the
effects more features and variations in the numerous parameters incorporated
into the current model. However, these limited set of model runs present
a clearer picture of the impact on modeled concentrations when an important
process is excluded or when certain key technical parameters in the dry
deposition and chemical transformation methods are modified. More evaluation
and testing efforts of this modeling system are certainly advocated.
92
-------
SECTION 6
SUMMARY AND CONCLUSIONS
The MESOPUFF II regional Lagrangian model has been evaluated and tested
with measurements from six full-scale CAPTEX episodes to assess its ability
to simulate long-range transport and dispersion of an inert, non-depositing
tracer plume. A neutrally-buoyant tracer gas was released over a 3-hour
period from ground-level during four cases from Dayton, Ohio and during two
cases from Sudbury, Ontario under fair weather conditions.
The results revealed that the model overpredicted both mean and peak
concentrations. The overpredictions were most pronounced for the Dayton
releases at the first two sampling arcs. Since differences in horizontal
plume spread were not evident during the first day of transport, it was
concluded that the primary cause for the model overpredictions was an under-
estimation of vertical plume growth for neutral stability, which was the
stability class most often specified during the afternoon release periods.
It is suggested that a different dispersion formula for the Gaussian vert-
ical dispersion parameter for neutral conditions (the Dl curve established
by Pasquill (1974)) be incorporated and tested within the current framework
of the model. This revision would provide for more rapid vertical plume
spread with the default Gaussian dispersion distance-dependent scheme under
neutral conditions, a stability class frequently selected from the widely-
used methodology by this model and others.
The statistical results also showed that large scatter and low corre-
lations occurred between modeled and observed concentrations paired in time
and space for both mean and peak values. This reflected the inability of
this model, like other similar models, to accurately replicate the speed
and/or direction of the tracer plume. Thus, the large concentration dif-
ferences where observed and modeled plumes did not overlap were primarily
attributed to trajectory errors. However, the statistical analyses did not
provide sufficient information for an assessment of the causes for model
errors in transport.
Graphical displays and a technique to determine plume centroid positions.
which indicated the size of the spatial plume differences, greatly assisted
in characterizing model trajectory errors. Spatial displacements of varying
amounts were evident from graphical maps of modeled and observed plume pat-
terns with time. Plume centroid positions, derived as the concentration-
weighted locations from non-zero values at 6-hour intervals, were used in
quantifying differences in the downwind distance and the separation distance
between observed and modeled plumes. Specifically, results from MESOPUFF II
indicated that the observed plume moved faster and/or along a different path
during the nocturnal period and was generally found further downwind of the
modeled plume by the second daytime period in these episodes. The plume
separation distance, the difference in centroid locations between the obs-
erved and modeled plumes, ranged from 100-300 km during these cases at down-
wind distances of 500-1000 km. The most rapid increase in plume separation
also occurred during the nocturnal periods.
93
-------
The CAPTEX tracer emissions consisted of non-buoyant, ground-level
releases. For this source type, puff heights in MESOPUFF II were always
less than the mixing height which caused modeled plumes to be continually
transported by the lower layer wind field. This feature is believed to be
a shortcoming of the model design, especially during the nocturnal period
when greater vertical shears in speed and direction often prevail. It was
concluded from analyses of the plume centroid positions that the modeled
plumes must have been advected by slower winds derived over the shallow
mixed layer at night, while observed plumes had been transported along a
different trajectory by faster winds at higher levels. In the case of an
elevated source emitting a buoyant plume which rises to even higher levels,
the model would have switched the plume transport at night to the upper
layer wind field when the mixing height dropped below the height of the puff
center. Based on these results, model applications for multi-day simulations
of ground-level, non-buoyant point emissions are not recommended. The model's
treatment for this source type could be improved if the height of the puff
center after release was redefined to be one-half of its vertical dimension
and this revised puff center height reached a limit when vertical puff growth
extended up to the mixing height.
Four other regional-scale models have also been applied on two or all
of the CAPTEX episodes with results being reported in the literature. These
models contained greater sophistication in transport (e.g. more vertical
layers) and different dispersion techniques. Unfortunately, comparable
statistical measures and plume displacement parameters were not reported
for three of the four models and an extensive comparison could not be made.
Nevertheless, the available results for the other models also displayed
large variabilities when observed and modeled concentrations were compared
and concentration differences were primarily attributed to spatial discre-
pancies in plume patterns due to trajectory errors.
The most extensive results with CAPTEX data were reported by Draxler (1987)
who tested a Lagrangian puff model which utilized several wind layers and
permitted puff splitting. Observed-modeled plume separation distances with
this model appeared to be smaller than those for MESOPUFF II, especially for
the cases exhibiting the greatest speed and directional shears. However,
plumes simulated by MESOPUFF II appeared to be displaced to the same side
with respect to the observed plumes (i.e. either to the right or left) as
those with the multi-layer model. .Draxler (1987) reported that average
spatial displacements for the 24-42 hour period after each release varied
from 90-189 km among the CAPTEX episodes, while individual separation dist-
ances with MESOPUFF II were found to be 100-300 km as noted above. It must
be noted, however, that Draxler's model runs included rawinsonde profile
data obtained at 6-hour intervals during CAPTEX. Better temporal resolution
of flow fields was translated into reduced trajectory errors when the more
frequent upper-air profile data and higher spatial resolution of observed
data were included in his model simulations. These supplemental profiles
were not used in the MESOPUFF II model simulations since they are not
available on a routine basis for model applications.
Qualitatively, the results from MESOPUFF II appeared to be competitive
with those of the other models. Therefore, radical changes of MESOPUFF II
and its processors are probably not warranted that could eliminate its
advantages of being computationally practical and easy to use for single
94
-------
sources. Indeed, the model should be expected to perform better for a
ground-level point source if revisions identified earlier are instituted.
The feasibility of puff splitting within the current model's framework
should also be explored as a step toward better addressing directional
shear-induced dispersion.
Any firm judgment of the comparative capabilites of the models for
general application was not possible based on the reported CAPTEX results.
If conclusions could have been made, they would necessarily be tempered by
the rather small number of cases, the specific source characteristics, and
the narrow set of meteorological scenarios during CAPTEX. Consequently, a
simultaneous evaluation of MESOPUFF II and a hierarchy of other models using
an agreed upon set of evaluation measures with a broader data base(s), which
should include elevated point source data, is certainly needed to provide
definitive evidence of relative model performance. In the interim, the
MESOPUFF II model should continue to be considered a useful modeling tool
for short-term, regional applications.
Diagnostic tests of optional features in the model provided distinct
differences in plume transport and dispersion from the default methods.
The mixed-layer averaged wind field (i.e. default lower level wind field)
was found to be superior to single level wind fields at the surface or
850 mb height in simulating the impact time and location of the tracer
plume at the 300 km arc. Results suggested that it should remain as the
preferred wind field for representing boundary layer transport when applying
this model. Model test run results with the uniform vertical mixing option
indicated that peak concentrations were much closer to observed values than
those from the default Gaussian dispersion parameter methods. With the
uniform vertical mixing method, puffs are completely dispersed through the
entire extent of the boundary layer after release. These test results gave
more evidence that modeled plumes required greater vertical mixing during
the afternoon release periods. In lieu of the suggested revision to the
Gaussian dispersion scheme noted earlier, the selection of this optional
dispersion method appears to be an attractive alternative since it produces
the desired vertical dispersion when neutral conditions are prevalent.
A limited group of model sensitivity runs was also performed in an effort
to examine the impact on 24-h peak and plume average SC^ and sulfate con-
centrations from variations in certain key parameters and changes in methods
in the dry deposition and chemical transformation components of the model.
The model base case and test case runs were exercised with the same meteor-
ological fields from CAPTEX #1 and the same emission rates from a large,
elevated point source. Results showed that sulfate concentrations were more
sensitive to the selected variation in a parameter or method. In particular,
sensitivity run results with differences in the surface SO^ resistance or
±50% differences in background ozone concentration produced negligible vari
ations in SOo peak and mean concentrations; however, the impact on sulfate
concentrations was from 15-20%. This finding is particularly relevant to
regional model applications since the higher concentration levels for sulfate
were found at much greater downwind distances than was SC^. It follows
that SO^ is influenced to a greater extent by transformation and deposition
processes. A more extensive evaluation of these modules is advocated with
suitable experimental data.
95
-------
REFERENCES
Benkley, C. W. and A. Bass, 1979: Development of mesoscale air quality
simulation models; Vol. 3. User's guide to MESOPUFF. EPA/600/7-80-058,
U.S. Environmental Protection Agency, Available as NTIS PB80-227796.
Brown, R. , M. Leach, G. Raynor, and P. Michael, 1985: A summary of the
weather documentation for the 1983 Cross-Appalachian Tracer
Experiments. Atmospheric Sciences Division, Brookhaven National
Laboratory, Upton, New York 11973.
Clark, T. L., R. Cohn, S. Seilkop, R. Draxler, and J. Heffter, 1988:
Comparison of modeled and measured tracer gas concentrations during
the Across North America Tracer Experiment (ANATEX). Paper at the
17th International Techn. Meeting of NATO-CCMS on Air Poll. Modeling
and Its Application, Cambridge, Engl., Sept. 19-22, 1988, 13 pp.
Davis., W. E., A. R. Olson, B.T. Didier, and D. W. Damschen, 1986:
A comparison of meso-alpha MLAM model results with observations from
CAPTEX. Preprints, Fifth Joint Conf. on Applications of Air Poll.
Meteorol., Amer. Meteorol. Soc., Boston, MA., p.69-72.
Draxler, R.R., 1987: Sensitivity of a trajectory model to spatial and
temporal resolution of the meteorological data during CAPTEX.
J. Glim, and Appl. Meteorol..26. 1577-1588.
Ferber, G. J., J. L. Heffter, R. R. Draxler, R. J. Lagomarsino,
F. L. Thomas, R. N. Dietz, and C. M. Benkovitz, 1986: Cross-
Appalachian Tracer Experiment (CAPTEX 83). Final Report NOAA
Tech. Memo. ERL ARL-142, Air Resources Laboratory, Silver Spring, MD.
Gelinas, R. J., and J. P. Vayk, 1979: Systematic sensitivity analysis of
air quality simulation models. EPA/600/4-79-035. U. S. Environmental
Protection Agency.
Hales, J. M., B. B. Hicks, and J. M. Miller, 1987: The role of research
measurement networks as contributors to federal assessments of acid
deposition. Bull. Amer. Meteorol. Soc.. 68. 216-225.
Heffter, J. L., 1965: The variation of horizontal diffusion parameters
with time for travel periods of one hour or longer. J. Appl.Meteorol. ,
4, 153-156.
Kao C. and T. Yamada, 1988: Use of the CAPTEX data for evaluations of a
long-range trnasport numerical model with a four-dimensional data
assimilation technique. Mon. Wea. Rev.. 116, 293-306.
Lagomarsino, R. J., F. L. Thomas, T. J. Weber, and L. R. Greenfield,
1987: Cross-Appalachian Tracer Experiment (CAPTEX 83) Quality Assurance
of Surface Samples. Environmental Measurements Lab., EML-462, New York,
New York 10014.
96
-------
Lee, I. Y., 1987: Numerical simulations of Cross-Appalachian transport
and diffusion. Boundary-Layer Meteorol.. 39, 53-66.
Morris, R. E., R. C. Kessler, S. G. Douglas, and K. R. Styles, 1988:
Rocky mountain acid deposition model assessment: Evaluation of meso-
scale deposition models for use in complex terrain. EPA/600/3-88-088.
U. S. Environmental Protection Agency. Available as NTIS PB88-167 481.
Page, S. H., 1980: National land and land cover inventory. U. S.
Environmental Protection Agency, Washington, B.C.
Pasquill, F., 1974: Atmospheric Diffusion (2nd Edition). Ellis Horwood
Ltd., Chichester, Engl. 228pp.
Policastro, A. J., M. Wastag, L. Coke, R. A. Carhart, W. E. Dunn, 1986:
Evaluation of short-term long range transport models (Vol I, II).
EPA-450/4-86-016a, b. U. S. Environmental Protection Agency.
Available as NTIS documents PB87-142337, PB87-142345.
Raynor, G. S., R. N. Dietz, and T. W. D'Ottavio, 1984: Aircraft
measurements of tracer gas during the 1983 Cross-Appalachian
Tracer Experiment. Proc. of Fourth Joint Conf. on Applications of
Air Pollution Meteorol., Amer. Meterol. Soc., Boston, MA, 162-165.
Scire, J. S., F. W. Lurmann, A. Bass, S. R. Hanna, 1984a: Development
of the MESOPUFF II dispersion model. EPA/600/3-84-057, Environmental
Protection Agency, Research Triangle Park, NC. Available as NTIS
PB No. 84-183 753.
Scire, J. S., F. W. Lurmann, A. Bass, and S. R. Hanna, 1984b: User's
Guide to the MESOPUFF II model and related processor programs.
EPA/600/8-84-013, Environmental Protection Agency, Research Triangle
Park, NC. Available as NTIS PB No. 84-182 775.
Sheih, C. M., M. L. Wesely, and B. B. Hicks, 1979: Estimated dry
deposition velocities of sulfur over the eastern United States
and surrounding regions. Atmos. Environ..13. 1361 1368.
Shreffler, J. H., 1982: Intercomparisons of upper air and surface
winds in an urban region. Boundary-Layer Meteorol..24. 345-356.
Stunder, B. J. B., J. L. Heffter, and R. R. Draxler, 1986: Long range
transport model evaluation using CAPTEX aircraft and surface data.
Vertical tracer distribution. Preprints, Fifth Joint Conf. on Appli-
cations of Air Pollution Meteorol., Amer. Meteorol. Soc., Boston, MA.,
65-68.
Turner, D. B., 1964: A diffusion model for an urban area. J. Appl.
Meteorol.,3, 83-91.
Turner, D. B., 1970: Workbook of atmospheric dispersion estimates.
97
-------
Publ. No AP-26, Office of Air Programs, U. S Environmental
Protection Agency., 84 pp. Available as NTIS PB 191 482.
Walcek, C. J., R. A. Brost, J. S. Chang, and M. L. Wesely, 1986: S02,
sulfate and HNO-j deposition velocities computed using regioanl land
use and meteorological data. Atmos. Environ..20. 949-964.
Wilson, W. E., 1981: Sulfate formation in point source plumes: A review
of recent field studies. Atmos. Environ.. 15, 2573-2590.
98
-------
APPENDIX A
MODEL TEST RESULTS FOR DIFFERENT PUFF RELEASE/SAMPLING RATES
A series of MESOPUFF II model runs for the CAPTEX #1 case was performed
with various puff release and sampling rates. The results presented below
were generated from peak tracer concentration values at the initial tracer
measurement arc at 300 km where the highest concentrations were found.
An average peak concentration value (CAVE) was computed from the results of
each model test runs and the individual peak concentrations were normalized
by this average value to determine the amount of difference among the test
runs. Differences over this set of puff release/sampling rates indicated
relatively little sensitivity to the peak concentrations at this distance
under the conditions of this experiment. Similar test runs are advocated
in other applications to determine sensitivity of modeled concentrations
to these rates.
TABLE 1A
DIFFERENCES IN PEAK VALUES AT THE INITIAL ARC
FOR DIFFERENT PUFF RELEASE/SAMPLING RATES
RUN RELEASE RATE SAMPLING RATE Cmax/CAVEmax CPU TIME
(puffs/h) ( #/h) (s)
12 2 0.997 12.7
22 4 0.966 22.1
34 2 1.020 22.4
44 4 0.984 41.0
5 8 2 1.023 41.8
6 12 2 1.024 61.6
7 12 4 0.986 117.6
99
------- |