vvEPA
United States
Environmental Protection
Agency
Office of Research and
Development
Washington, DC 20460
^DA 600 3-90 051
June 1990
The Across North
America Tracer
Experiment (ANATEX)
Model Evaluation Study
-------
EPA/600/3-90/051
June 1990
THE ACROSS NORTH AMERICA TRACER EXPERIMENT (ANATEX)
MODEL EVALUATION STUDY
by
Terry L. Clark*
Atmospheric Sciences Modeling Division
National Oceanic and Atmospheric Administration
U.S. Department of Commerce
Research Triangle Park, NC 27711
and
Richard D. Colin
Analytical Sciences, Incorporated
100 Capitola Drive, Suite 106
Durham, NC 27713
Chicago,, JL 6060-.
*On assignment to the Atmospheric Research and Exposure Assessment Laboratory,
U.S. Environmental Protection Agency
ATMOSPHERIC RESEARCH AND EXPOSURE ASSESSMENT LABORATORY
OFFICE OF RESEARCH AND DEVELOPMENT
U.S. ENVIRONMENTAL PROTECTION AGENCY
RESEARCH TRIANGLE PARK, NC 27711
-------
NOTICE
The information in this document has been funded wholly or in part by the U.S. Environmental
Protection Agency, the U.S. Air Force, and the National Oceanic and Atmospheric Administration of the
U.S. Department of Commerce under U.S. Department of Commerce Contract No. 50-EANR-8-00025 to
Analytical Sciences, Incorporated. It has been subjected to the EPA peer and administrative review, and it
has been approved for publication as an EPA document. Mention of trade names or commercial products
does not constitute endorsement or recommendation for use.
-------
FOREWORD
The U.S. Environmental Protection Agency, the National Oceanic and Atmospheric Administration,
and the U.S. Air Force have completed an evaluation of 11 operational models to assess the performances of
state-of-the-science, long-range transport and diffusion models. The model calculations were compared to
observations of surface concentration data compiled during the Across North America Tracer Experiment
(ANATEX).
Before the distribution of the ANATEX data, modelers applied their models in a "blind" applications
mode using required meteorological input data and the actual periodic 3-hour (h) ANATEX tracer emission
rates according to the prescribed schedule during the first 3 months of 1987. Some of these models were
very similar to others in this study; the only differences were variations of the modeling assumptions or the
selection of modeling options.
Model performance measures were developed on the basis of the features of the surface sampling
network and the sampling protocol. These measures were quantified using either ensemble concentration
means or relative distances of the centroids of tracer "footprints"~composite tracer plumes defined by the
24-h-mean measurements.
Several aspects of the evaluation study are discussed in this report. First, the performances of the
three genres of models-(l) single-layer Lagrangian, (2) multiple-layer Lagrangian, and (3) multiple-layer
Eulerian~are compared to each other to relate model performance to model approach. Secondly, the
performances of the various model versions are related to the differences in the modeling codes to relate
model performance to model assumptions/options. Thirdly, model performance is related to three
meteorological scenarios to relate model performance to the degree of complexity of the air flow.
111
-------
ABSTRACT
During the first three months of 1987, three perfluorocarbon tracer gases were released at 2.5-day or
5.0-day intervals from two sites in central North America (Glasgow, Montana and St. Cloud, Minnesota) and
sampled for 24-h periods at 77 surface sites. The source-receptor distances ranged from less than 30 km to
3,000 km. These Across North America Tracer Experiment (ANATEX) data serve as a unique evaluation
data set with which to evaluate the long-range transport and diffusion simulations of acid deposition models
and to establish a range of uncertainty for various model genres.
The performances of three single-layer Lagrangian, six multiple-layer Lagrangian, and two multiple-
layer Eulerian models are assessed using quantifiable measures based on comparisons of ensemble mean
concentrations and plume widths as well as trajectory errors expressed as a function of transport time. In
general, the multiple-layer Lagrangian models performed best in simulating the transport of the tracers, while
the Eulerian models performed best in simulating the ensemble concentration frequency distributions.
After 0.5 day of transport, trajectory errors ranged from 100 km to 400 km; after 2.5 days, the errors
ranged from 300 km to 800 km. Beyond 2.5 days, errors from four Lagrangian models plateaued, while
errors for the other models continued to increase, peaking at nearly 1,100 km after 3.5 days. Of the three
single-layer models, one performed as well as any other model in simulating the transport; the other two
single-layer models performed worst. Thus, a slight difference in the method of calculating single-layer
transport vectors can yield a significant difference in model performance. For six of the eleven models, the
greatest errors in transport speed and location tended to occur when the tracer was intercepted by cyclones
and/or fronts.
The ensemble mean concentrations along three bands of sites 1,000 to 2,300 km downwind from the
release sites were nearly always overpredicted by the single-layer Lagrangian models, in some cases by as
much as a factor of 5; overpredictions by a factor of 3 were common. The two Eulerian models tended to
underpredict these ensemble means by about 40%, especially along the nearest band. Finally, the multiple-
layer Lagrangian models tended to underpredict the ensemble means for the tracer released from Montana,
yet tended to overpredict the ensemble means for the tracer released from Minnesota.
Horizontal spreading of the tracer plumes beyond 1,000 km transport distance occurred at a rate of
30% to 60% per 600 km. Generally the models replicated this spreading rate, although there was a wide
range of plume widths. The plume widths of the Eulerian models tended to be the greatest, which partially
explained their tendency to underpredict mean concentrations. In virtually all cases, the plume widths of the
Lagrangian models were equivalent to or less than the actual plume widths; for some models, plume widths
were much less, by factors of 2 and 3.
iv
-------
CONTENTS
Foreword iii
Abstract iv
Figures vii
Tables xiii
Abbreviations and Symbols xiv
Acknowledgments xvii
1. Introduction 1
2. Previous Model Evaluation Studies 8
3. Model Evaluation Protocol 10
3.1 The Evaluation Approach 10
3.2 Model Performance Measures 16
3.2.1 Ensemble Concentration Means and Distributions 16
3.2.2 Ensemble Mean Horizontal Diffusion 18
3.2.3 Transport Speed of Discrete Tracer Footprints 19
3.2.4 Tracer Footprint Location Error 22
4. Assessment of the Model Performances Based on Ensemble Data 23
4.1 Comparison of Concentration Magnitudes 23
4.1.1 Ensemble Concentration Distributions 23
4.1.2 Mean Concentrations Along Bands 29
4.2 Comparison of Horizontal Diffusion of Tracer Puffs 35
5. Assessment of the Model Performances Based on Comparisons of Individual
Tracer Releases 43
5.1 Performance as Related to Meteorological Scenario 43
5.2 Mean Separation Error 49
5.3 Footprint Transport Speeds and Centroid Locations 52
5.3.1 Summary of Model Performances 54
5.3.2 Single-Layer Lagrangian Models 58
5.3.3 Multiple-Layer Lagrangian Models 66
5.3.4 Multiple-Layer Eulerian Models 82
-------
CONTENTS (CONCLUDED)
6. Summary and Conclusions 86
6.1 Single-Layer Lagrangian (SLL) Models 88
6.1.1 Summary 88
6.1.2 Conclusions 89
6.2 Multiple-Layer Lagrangian (MLE) Models 90
6.2.1 Summary 90
6.2.2 Conclusions 92
6.3 Multiple-Layer Eulerian (MLE) Models 93
6.3.1 Summary 93
6.3.2 Conclusions 94
7. Recommendations 95
8. References 96
Appendix A. Model Descriptions A-l
Appendix B. Time Series Plots of 24-h Mean Concentrations of PTCH and PDCH Tracers
Along Bands of Sites B-l
VI
-------
FIGURES
Number Page
1 The 77-site ANATEX surface network for 24-h-average tracer concentration
measurements and the location of the tracer release sites at Glasgow, Montana (GGW)
and St. Cloud, Minnesota (STC) 4
2 Illustration of the 6 January 1987 "footprint", or a 24-h composite of instantaneous
tracer puffs near the surface, and the centroid Cm, for the measured PDCH
concentrations from release 1 12
3 The 24-h locations of ARL (—) and actual ( ) footprint centroids for the
first release of PTCH tracer 14
4 The bands of sites along which ensemble-mean calculated and measured horizontal
diffusion were estimated for (a) PTCH and (b) PDCH tracers 17
5 The model performance measures characterizing (a) the agreement between
calculated and measured transport speed and (b) footprint location errors 20
6 Box plots of cumulative frequency distribution of uncertainty ([/) versus excess
concentration (x) for PMCH, PDCH, and PTCH 21
7 Box plots comparing quartiles and means of calculated and measured PTCH
tracer concentrations at the 77 surface sites during the periods:
(a) 5 January - 16 February 1987 and (b) 17 February - 29 March 1987 25
8 Frequency distributions of calculated and measured PTCH tracer concentrations at
the 77 surface sites during the periods: (a) 5 January - 16 February 1987 and
(b) 17 February - 29 March 1987 26
9 Box plots comparing quartiles and means of calculated and measured PDCH
tracer concentrations at the 77 surface sites during the periods:
(a) 5 January - 16 February 1987 and (b) 17 February - 29 March 1987 28
10 Frequency distributions of calculated and measured PDCH tracer concentrations at
the 77 surface sites during the periods: (a) 5 January - 16 February 1987 and
(b) 17 February - 29 March 1987 30
vn
-------
FIGURES (CONTINUED)
Page
11 Mean calculated and measured PTCH concentrations along three bands of sites
downwind of GGW during the periods: (a) 5 January - 16 February 1987 and
(b) 17 February - 29 March 1987 32
12 Mean calculated and measured PDCH concentrations along three bands of
sites downwind of STC during the periods: (a) 5 January - 16 February 1987
and (b) 17 February - 29 March 1987 36
13 Mean horizontal diffusion of calculated and measured tracer puffs as indicated by
the average distance of nonzero PTCH concentrations along three bands of sites
downwind of GGW during the periods: (a) 5 January - 16 February 1987 and
(b) 17 February - 29 March 1987 38
14 Mean horizontal diffusion of calculated and measured tracer puffs as indicated by
the average distance of nonzero PDCH concentrations along three bands of sites
downwind of STC during the periods: (a) 5 January - 16 February 1987 and
(b) 17 February - 29 March 1987 41
15 Comparison of the mean separation errors of PTCH and PDCH footprint centroids,
(D e) as a function of transport time 51
16 The comparison of trajectory error ranges determined from ANATEX and CAPTEX
model evaluations 53
17 Percentages of footprint-days when the models deviate by 20% and 50% from the
actual (a) transport speed, (b) centroid location within 24-h,
(c) between 24 h and 48 h, and (d) after 48 h of transport 56
18 Errors in SRL transport speeds for (a) PDCH and (b) PTCH footprints; and
centroid locations for (c) PDCH and (d) PTCH for the designated tracer releases 59
19 Errors in TCAL transport speeds for (a) PDCH and (b) PTCH footprints; and
centroid locations for (c) PDCH and (d) PTCH footprints for the designated
tracer releases 61
vm
-------
FIGURES (CONTINUED)
Page
20 Errors in VCAL transport speeds for (a) PDCH and (b) PTCH footprints; and
centroid locations for (c) PDCH and (d) PTCH footprints for the designated
tracer releases 63
21 Comparison of TCAL and VCAL (V) errors in transport speeds for
(a) PDCH and (b) PTCH footprints; and centroid locations for (c) PDCH and
(d) PTCH footprints common to both sets for the designated tracer releases 65
22 Errors in ARL transport speeds for (a) PDCH and (b) PTCH footprints; and
centroid locations for (c) PDCH and (d) PTCH footprints for the designated
tracer releases 67
23 Errors in BAT transport speeds for (a) PDCH and (b) PTCH footprints; and
centroid locations for (c) PDCH and (d) PTCH footprints for the designated
tracer releases 69
24 Errors in GAMUT transport speeds for (a) PDCH and (b) PTCH footprints; and
centroid locations for (c) PDCH and (d) PTCH footprints for the designated
tracer releases 71
25 Errors in HY-SPLIT transport speeds for (a) PDCH and (b) PTCH footprints; and
centroid locations for (c) PDCH and (d) PTCH footprints for the designated
tracer releases 73
26 Comparison of ARL and HY-SPLIT (H) errors in transport speeds for
(a) PDCH and (b) PTCH footprints; and centroid locations for (c) PDCH and
(d) PTCH footprints common to both sets for the designated tracer releases 75
27 Errors in MLAM-FINE transport speeds for (a) PDCH and (b) PTCH footprints; and
centroid locations for (c) PDCH and (d) PTCH footprints for the designated
tracer releases 77
28 Errors in MLAM-COARSE transport speeds for (a) PDCH and (b) PTCH footprints;
and centroid locations for (c) PDCH and (d) PTCH footprints for the designated
tracer releases 79
29 Comparison of MLAM-COARSE and MLAM-FINE (F) errors in transport speeds for
(a) PDCH and (b) PTCH footprints; and centroid locations for (c) PDCH and
(d) PTCH footprints common to both sets for the designated tracer releases 81
IX
-------
FIGURES (CONTINUED)
Number Page
30 Errors in ADOM transport speeds for (a) PDCH and (b) PTCH footprints; and
centroid locations for (c) PDCH and (d) PTCH footprints for the designated
tracer releases 83
31 Errors in ADPIC transport speeds for (a) PDCH and (b) PTCH footprints; and
centroid locations for (c) PDCH and (d) PTCH footprints for the designated
tracer releases 85
B-l Time series plots of 24-h mean-measured and calculated PTCH concentrations
along the 1,000-km band for January 1987 B-2
B-2 Time series plots of 24-h mean-measured and calculated PTCH concentrations
along the 1,000-km band for February 1987 B-3
B-3 Time series plots of 24-h mean-measured and calculated PTCH concentrations
along the 1,000-km band for March 1987 B-4
B-4 Time series plots of 24-h mean-measured and calculated PTCH concentrations
along the 1,600-km band for January 1987 B-5
B-5 Time series plots of 24-h mean-measured and calculated PTCH concentrations
along the 1,600-km band for February 1987 B-6
B-6 Time series plots of 24-h mean-measured and calculated PTCH concentrations
along the 1,600-km band for March 1987 B-7
B-7 Time series plots of 24-h mean-measured and calculated PTCH concentrations
along the 2,300-km band for January 1987 B-8
B-8 Time series plots of 24-h mean-measured and calculated PTCH concentrations
along the 2,300-km band for February 1987 B-9
B-9 Time series plots of 24-h mean-measured and calculated PTCH concentrations
along the 2,300-km band for March 1987 B-10
B-10 Time series plots of 24-h mean-measured and calculated PDCH concentrations
along the 300-km band for January 1987 B-ll
B-ll Time series plots of 24-h mean-measured and calculated PDCH concentrations
along the 300-km band for February 1987 B-12
-------
FIGURES (CONTINUED)
Number Page
B-12 Time series plots of 24-h mean-measured and calculated PDCH concentrations
along the 300-km band for March 1987 B-13
B-13 Time series plots of 24-h mean-measured and calculated PDCH concentrations
along the 700W-km band for January 1987 B-14
B-14 Time series plots of 24-h mean-measured and calculated PDCH concentrations
along the 700W-km band for February 1987 B-15
B-15 Time series plots of 24-h mean-measured and calculated PDCH concentrations
along the 700W-km band for March 1987 B-16
B-16 Time series plots of 24-h mean-measured and calculated PDCH concentrations
along the 700E-km band for January 1987 B-17
B-17 Time series plots of 24-h mean-measured and calculated PDCH concentrations
along the 700E-km band for February 1987 B-18
B-18 Time series plots of 24-h mean-measured and calculated PDCH concentrations
along the 700E-km band for March 1987 B-19
B-19 Time series plots of 24-h mean-measured and calculated PDCH concentrations
along the l,OOOW-km band for January 1987 B-20
B-20 Time series plots of 24-h mean-measured and calculated PDCH concentrations
along the l,OOOW-km band for February 1987 B-21
B-21 Time series plots of 24-h mean-measured and calculated PDCH concentrations
along the l,OOOW-km band for March 1987 B-22
B-22 Time series plots of 24-h mean-measured and calculated PDCH concentrations
along the l,OOOE-km band for January 1987 B-23
B-23 Time series plots of 24-h mean-measured and calculated PDCH concentrations
along the l,OOOE-km band for Febuary 1987 B-24
B-24 Time series plots of 24-h mean-measured and calculated PDCH concentrations
along the l,OOOE-km band for March 1987 B-25
B-25 Time series plots of 24-h mean-measured and calculated PDCH concentrations
along the 1,400-km band for January 1987 B-26
XI
-------
FIGURES (CONCLUDED)
Page
Time series plots of 24-h mean-measured and calculated PDCH concentrations
along the 1,400-km band for February 1987 B-27
B-27 Time series plots of 24-h mean-measured and calculated PDCH concentrations
along the 1,400-km band for March 1987 B-28
B-28 Time series plots of 24-h mean-measured and calculated PDCH concentrations
along the 1,800-km band for January 1987 B-29
B-29 Time series plots of 24-h mean-measured and calculated PDCH concentrations
along the 1,800-km band for February 1987 B-30
B-30 Time series plots of 24-h mean-measured and calculated PDCH concentrations
along the 1,800-km band for March 1987 B-31
xn
-------
TABLES
Number Page
1 The ANATEX schedule and amounts of GGW releases of PTCH and STC
releases of PDCH and PMCH tracers 3
2 Number of "site-days" in the ANATEX database for each tracer 5
3 The 11 long-range transport and diffusion models evaluated in the ANATEX
Model Evaluation Study (AMES) 6
4 Model performance measures 15
5 The mean calculated and measured tracer concentrations along bands of samplers
for the two halves of the ANATEX period 33
6 The mean distance of nonzero calculated and measured concentrations and
mean spread rates along bands of samplers 39
/ Number of calculated and measured footprint pairs and the type of
meteorological scenario for each tracer release 44
8 The three types of meteorological scenarios for which model performance was assessed . 46
9 Number of footprint-days intercepted (I) and not intercepted (NI) by cyclones
and/or fronts after 10-h of transport 46
10 Relationship between model performance and meteorological scenario 48
11 Percent of Model A footprint-days common to both Models A and B 50
12 Number of calculated and measured footprint pairs for each model
as a function of transport time 50
13 A summary of model performances in calculating transport speeds and footprint
centroid locations 55
14 Overall biases in the model results as indicated by the model performance measures .... 87
A-l SRL attributes A-2
A-2 TCAL and VCAL attributes A-3
A-3 ARL attributes A-4
A-4 BAT attributes A-5
A-5 GAMUT attributes A-6
A-6 HY-SPLIT attributes A-7
A-7 MLAM (COARSE and FINE) attributes A-8
A-8 ADOM attributes A-9
A-9 ADPIC attributes A-10
B-l List of 24-h-mean concentrations exceeding 50 dfL/L B-32
xiii
-------
ABBREVIATIONS AND SYMBOLS
ABBREVIATIONS
ADOM
ADPIC
AFGWC
AGL
AMES
ANATEX
ARL
BAT
CAPTEX
dfL/L
GAMUT
GGW
GMT
h
HIRAS
HY-SPLIT
LST
MESOPUFF
METREX
MLAM
MLAM-COARSE
MLAM-FINE
MLE
MLL
NGM
NOAA
PDCH
PMCH
PTCH
rms
SLL
SRL
STC
TCAL
VCAL
Acid Deposition and Oxidant Model
Atmospheric Diffusion Particle-In-Cell model
Air Force Global Weather Center
above ground level
ANATEX Model Evaluation Study
Across North America Tracer Experiment
Air Resources Laboratory model
Branching Atmospheric Trajectory model
Cross-Appalachian Tracer Experiment
decifemtoliters per liter (1(T16 L/L)
Global Atmospheric Multilayer Transport model
Glasgow, Montana
Greenwich Mean Time
hour
High Resolution Analysis System
HYBRID Single-Particle Lagrangian Integrated Trajectories model
local standard time
Mesoscale Puff model
Metropolitan Tracer Experiment
Multilayer Air Mass Model
a coarse-resolution version of MLAM
a fine-resolution version of MLAM
Multiple-Layer Eulerian model type
Multiple-Layer Lagrangian model type
NOAA's Nested Grid Model
National Oceanic and Atmospheric Administration
perfluoro-ortho-dimethylcyclohexane
perfluoro-methylcyclohexane
perfluoro-trimethylcyclohexane
root-mean-square
Single-Layer Lagrangian model type
Savannah River Laboratory Adjusted Geostrophic Model
St. Cloud, Minnesota
Trajectory Calculation model
Variable Layer Trajectory Calculation model
xiv
-------
ABBREVIATIONS AND SYMBOLS (CONTINUED)
SYMBOLS
Cc centroid of the calculated footprint
Cm centroid of the measured footprint
D. cumulative distance between C and the tracer source
O O
Dc/Dm transport speed ratio
DQ distance between Cc and Cm
DQ/Dm normalized separation distance between Cc and Cm
D cumulative distance between C and the tracer source
d{ distance from site / to trajectory starting point
Kh horizontal turbulent eddy diffusivity coefficient
Kz vertical turbulent eddy diffusivity coefficient
L Monin-Obukhov length
N number of footprint days
p atmospheric pressure
R, gradient Richardson number
T temperature
Tv virtual temperature
t time
A< time step
Uk advection velocity
UD diffusivity velocity
u eastward wind speed component
u* friction velocity
v northward wind speed component
w vertical wind speed component
wt vertically-averaged wind at site /
or weight for site / used in calculating foortprint centroids
z height
Az thickness of model layer
z0 surface roughness height
V gradient operator ( d/dx + d/dy + d/dz )
d potential temperature of atmospheric layer
0B potential temperature of bottom layer
#j- potential temperature of top layer
xv
-------
ABBREVIATIONS AND SYMBOLS (CONCLUDED)
SYMBOLS
0W wet bulb potential temperature
0* scaled potential temperature
A0 potential temperature difference across a layer
crh horizontal dispersion parameter
CTy crosswind distance from plume centerline
CTZ vertical dispersion parameter
ae standard deviation of wind direction
-------
ACKNOWLEDGMENTS
The authors acknowledge the valuable assistance received during this project from Roland R.
Draxler and Jerome L. Heffter of the National Oceanic and Atmospheric Administration, U.S. Department
of Commerce. Also acknowledged are the other modelers who participated in this model evaluation study,
without whom the study would not have been possible:
William E. Davis and Anthony R. Olsen, Pacific Northwest Laboratory
Stephen E. Masters, ENSCO, Inc.
Marvin P. Olson and Keith J. Puckett, Environment Canada
Malcolm Pendergast, Savannah River National Laboratory
Daniel J. Rodriguez, Lawrence Livermore National Laboratory
The technical assistance of Steven K. Seilkop, Analytical Sciences, Inc.; the report-preparation
assistance of Christine Maxwell, Computer Sciences Corporation; and the financial support of the
U.S. Air Force Technical Applications Center were also greatly appreciated.
xvn
-------
SECTION 1
INTRODUCTION
Computer models have been developed to simulate the fate of gases and particles in the atmosphere
over spatial scales ranging from the urban scale to the global scale. These models are applied to study such
phenomena as acidic deposition, Arctic haze, photochemical oxidants, accidental releases of radioactive
material, etc. Each of these models is comprised of a series of modules, each simulating a physical process
(e.g., vertical diffusion, horizontal diffusion, transport, and for some, wet and dry deposition) relating to the
atmospheric concentrations of gases and particles. In addition, some of these models simulate chemical
transformations while still others simulate radioactive decay.
It is essential to assess the limitations and reliability of these models, as well as the uncertainty limits
of their results, in both an operational mode (i.e., performance assessments of the final model results), as
well as a diagnostic model (i.e., performance assessments of the results of each module of the model).
However, the scarcity of adequate measurements (due to the enormous costs of obtaining them) prohibits a
comprehensive diagnostic model evaluation and limits the simpler operational evaluation.
The concentration data from two field studies-the 1983 Cross-Appalachian Tracer Experiment
(CAPTEX) [Ferber et al., 1986] and the 1987 Across North America Tracer Experiment (ANATEX)
[Draxler and Heffter, 1989]~provide virtually the sole means to evaluate the transport and diffusion
modules of long-range models. Additional tracer data are required to minimize the uncertainties of the
evaluation results. The other modules cannot be evaluated until appropriate measurements are available.
The CAPTEX study measured 3-h and 6-h mean concentrations of an inert perfluorocarbon tracer
gas released on seven occasions during September and October from either Dayton, Ohio or Sudbury,
Ontario. The sampling network consisted of 86 sites located within 300 km to 1,100 km of Dayton. The
results of model evaluations using this data set are summarized in Section 2.
The more recent Across North America Tracer Experiment (ANATEX) provided additional tracer
concentration data for many more days and over a much larger region (Draxler and Heffter, 1989). During
the 12-week period between 5 January and 29 March 1987, three different inert perfluorocarbon tracers
-------
were released periodically during thirty-three 3-hour (h) periods from near-surface sources near Glasgow,
Montana (GGW) and St. Cloud, Minnesota (STC). Two of these tracers, PTCH 1 and PDCH2, were
uniformly released for 3 h every 2.5 days from GGW and STC, respectively (Table 1). The third
tracer-PMCH 3 -was released from STC simultaneously with every other PDCH release. Figure 1 shows
the 77-site surface network locations where 24-h-average concentrations of each tracer were measured for
the period 1400-1400 GMT. The sites are oriented along eight bands, each of which is to the east of
Glasgow by the approximate designated distances (km). Table 2 indicates the number of "site-days" when
concentrations were available and passed the screening criteria.4
The ANATEX Model Evaluation Study (AMES) relies on these 24-h surface concentrations to
assess the performance of the 11 long-range transport and diffusion models listed in Table 3 and described
in Appendix A. Because of resource limitations, MLAM-FINE and ADOM were applied only in the first-
half period (i.e., 5 January - 16 February 1987). All but two models use routinely available wind data;
HY-SPLIT and ADOM use higher resolution data generated by diagnostic meteorological models. The
primary objective of each model is to simulate the fate of atmospheric pollutants emitted from a single
source in the form of either a single puff or a continuous plume. In the AMES, these models were applied
in a prognostic mode by the respective modelers independent of each other and without benefit of the
evaluation data set—that is, the models were also applied in a "blind-test" mode. Trajectories and
concentrations computed by several of these models, as well as others, have already been compared to the
CAPTEX data to assess model performance. However, with the ANATEX measurements, much more can
be learned about the performance of these models, since:
(1) the ANATEX sampling domain was much larger than the CAPTEX network (enabling us to
assess model performance over longer spatial and temporal scales);
(2) the ANATEX tracers were periodically released, as opposed to the CAPTEX tracer that was
released only for two simple meteorological scenarios (enabling us to evaluate the models over
a much larger range of meteorological scenarios); and
(3) there were 66 discrete tracer releases in ANATEX, as opposed to only 7 CAPTEX tracer
releases (enabling us, for the first time, to base the model performance assessment on a
statistically meaningful data set).
Perfluoro-trimethylcyclohexane.
2
Perfluoro-ortho-dimethylcyclohexane (oPDCH), hereafter referred to as HPDCH".
3
Perfluoro-methylcyclohexane.
Concentrations were excluded when the QA/QC flags indicated that the sample was suspect (Draxler and Heffter, 1989).
-------
TABLE 1. THE ANATEX SCHEDULE AND AMOUNTS OF GGW RELEASES OF PTCH
AND STC RELEASES OF PDCH AND PMCH TRACERS
Release amount, ke
Release
number a
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
31
32
33
Average
Dateb
Jan 5
8
10
13
15
18
20
23
25
28
30
Feb 2
4
7
9
12
14
17
19
22
24
27
Mar 1
4
6
9
11
14
16
19
21
24
26
Time
period, GMT
1700-2000
0500-0800
1700-2000
0500-0800
1700-2000
0500-0800
1700-2000
0500-0800
1700-2000
0500-2000
1700-2000
0500-0800
1700-2000
0500-0800
1700-2000
0500-0800
1700-2000
0500-0800
1700-2000
0500-0800
1700-2000
0500-0800
1700-2000
0500-0800
1700-2000
0500-0800
1700-2000
0500-0800
1700-2000
0500-0800
1700-2000
0500-0800
1700-2000
PTCH
Glasgow
76.6
88.8
83.5
83.9
83.4
83.9
83.5
83.7
83.6
83.6
83.7
83.6
83.7
83.7
83.7
83.7
83.7
83.7
84.1
83.7
83.7
83.7
83.7
83.7
83.7
83.6
83.8
83.6
83.6
83.7
83.7
83.6
75. 1
83.4
PDCH
St. Cloud
50.7
48.8
50.8
50.6
51.1
51.5
50.0
51.1
50.5
51.0
50.8
51.0
50.7
47.2
46.8
47.0
46.8
47.0
47.1
47.4
47.1
47.6
46.9
47.2
47.1
51.3
47.1
51.4
47.0
49.9
46.8
47.2
44.6
48.7
PMCH
St. Cloud
49.5
0.0
49.7
0.0
50.2
0.0
48.6
0.0
49.3
0.0
49.6
0.0
49.6
2.9
52.4
2.9
52.5
2.9
53.0
2.9
53.0
3.0
52.6
2.9
53.0
0.0
53.0
0.0
52.7
0.1
52.4
2.9
49.0
51.2 c
a. Odd-numbered releases are daytime releases, while even-numbered releases are nighttime releases.
b. Year = 1987.
c. Not including releases of 2.9 kg.
-------
DC
O
LU
Z
CD
CL
cn
x
LU
LU
E
^
o
u
•o H
c °
II
s.s
r
»-. CO
4— *
C T3
d> c
O c- o
I .J1
"f o
TT 00
tN «°
o3
|5
4^
-^ 03
(D 03
O *U
•Ss
3 u
>v> O
r> 03
W J3
*-,
< o
iu a
• - 2
to 'B
.
f-
I-
-------
TABLE 2. NUMBER OF "SITE-DAYS" IN THE ANATEX DATABASE
FOR EACH TRACER
Site-days
PMCH
ANATEX tracer
PDCH
PTCH
Total possible
Missing sample
Failing screening criteria
6,468
-892
-558
6,468
-944
-348
6,468
-830
-342
Total in database
5,018
5,176
5,296
-------
TABLE 3. THE 11 LONG-RANGE TRANSPORT AND DIFFUSION MODELS EVALUATED IN
THE ANATEX MODEL EVALUATION STUDY (AMES)
Model
acronym
Model name
Model developer
SRL
TCAL
VCALa
SINGLE-LAYER LAGRANGIAN (SLL) MODELS
Adjusted Geostrophic Model Savannah River National Laboratory
Trajectory Calculation Model ENSCO, Inc.
Variable Layer Trajectory ENSCO, Inc.
Calculation Model
MULTIPLE-LAYER LAGRANGIAN (MLL) MODELS
ARL
BAT
GAMUT
HY-SPLITb
MLAM-FINE
and
MLAM-COARSE
Air Resources Laboratory Model
Branching Atmospheric
Trajectory Model
Global Atmospheric Multilayer
Transport Model
Hybrid Single-Particle
Lagrangian Integrated
Trajectories Model
Multilayer Air Mass Model
National Oceanic and Atmospheric
Administration
National Oceanic and Atmospheric
Administration
ENSCO, Inc.
National Oceanic and Atmospheric
Administration
Pacific Northwest Laboratory
MULTIPLE-LAYER EULERIAN (MLE) MODELS
ADOM
ADPIC
Acid Deposition and
Oxidant Model
Atmospheric Diffusion
Particle-In-Cell Model
Environment Canada--
Atmospheric Environment Service
Lawrence Livermore National
Laboratory
a. A model very similar to the TCAL model.
b. A model very similar to the ARL model.
-------
As indicated by the objectives listed below, this study focuses on the performance of the models
during the first three months of 1987. Explanations of model behavior are limited in this study in that
additional analyses are required to substantiate the observations. The modelers are encouraged to conduct
these analyses using the results of this study as a starting point.
The objectives of AMES are fourfold:
(1) to assess the overall performance, as well as the model errors on temporal scales of 24 h, of
prognostic long-range atmospheric models with respect to transport and diffusion as a
function of transport time and distance,
(2) to intercompare the model performances and relate performance to fundamental modeling
approaches,
(3) to identify the periods and associated meteorological conditions when each model
performed best and worst, and
(4) to compare and contrast the AMES conclusions with those of similar studies using
CAPTEX data.
Section 2 summarizes the conclusions of previous model evaluation studies. Section 3 discusses the
model performance measures and analyses that were developed specifically for the ANATEX database.
The model evaluation results based on ensemble statistics are presented and interpreted in Section 4, while
the model evaluation results based on the 24-h concentration patterns of individual tracer releases are
presented and interpreted in Section 5. Section 6 presents the conclusions of this study and compares the
final results to those of similar studies.
-------
SECTION 2
PREVIOUS MODEL EVALUATION STUDIES
Past studies have assessed the reliability and uncertainties of long-range transport and diffusion
models by comparing model-calculated concentrations to measured concentrations of inert tracer gases.
One of the early studies compared calculations of a Lagrangian puff model to twice-daily averages of
Krypton-85 concentrations 1,500 km downwind from the point source (Draxler, L982). This study
concluded that, unless a model simulates wind shear effects, surface concentrations will be overestimated.
More recent studies used the 3- and 6-h surface and 6- to 10-minute aircraft measurements of an
inert tracer gas during the 1983 CAPTEX (Ferber et al., 1986) and the concentrations of radioactive
material from the 25 April 1986 Chernobyl reactor accident (Persson et al., 1986). Haagenson et al. (1987)
concluded that the root-mean-square (rms) separation distances between all CAPTEX tracer and transport
model trajectories for all time periods ranged from 220 km to 260 km. They also concluded that absolute
trajectory errors were 120 km and 200 km after 12 h and 24 h of transport, respectively. These estimates
were within 20% of those of Kahl and Samson (1988). For the MESOPUFF-II model, Godowitch (1989)
calculated a mean error of approximately 150 km after 24 h to 50 h of transport.
Draxler (1987) assessed the mean transport errors of a simple puff trajectory model (Draxler, 1982)
using routine meteorological data and the tracer data from all CAPTEX time periods. Mean trajectory
errors for this model were slightly lower than those of others-approximately 80 km and 150 km after 12 h
and 24 h of transport, respectively. Between 30 h and 42 h of transport, the error was virtually constant at
approximately 180 km. The mean ratios of predicted to measured concentrations also were a function of
time after release; for the first 24 h, the model tended to underpredict by 50%, but after 30 h, the
underprediction decreased to 10% or less.
Using data from each CAPTEX tracer release, Lee (1987) assessed mean transport errors of a
three-dimensional particle-in-cell model. The author concluded that the position of the simulated plumes in
general corresponded well to those of the measured plumes. The frequency histogram of residuals
computed from spatially- and temporally-paired concentrations illustrated a modal value near zero-72% of
the pairs of concentrations fell between ±400 decifemtoliters per liter (10~16 L/L or dfL/L), and 59% of the
-------
time, the model overpredicted the concentrations. Because the frequency of concentrations in excess of
50 dfL/L was greater for the predictions, the author concluded that the simulated horizontal diffusion was
overpredicted.
For two CAPTEX tracer releases, Kao and Yamada (1988) computed plume trajectories and surface
tracer concentrations using a random-particle statistical method and four-dimensional data assimilated wind
fields. For these two cases--(l) a light-wind, fair-weather situation with a wide surface plume and (2) a
cold frontal situation with a rather narrow plume-the simulated concentration patterns corresponded well
with those of the measurements in that the center lines of the composite simulated and measured plumes
coincided after 54 h.
These studies indicated that after 12 h and 24 h of transport, absolute model trajectory errors ranged
from 80 km to 150 km and 150 km to 220 km, respectively. One study showed that trajectory errors
approximated 180 km after 30 h to 42 h of transport. After 72 h of transport, another study indicated that,
50% of the time, the error would exceed 350 km. In Section 5.2, these errors are compared to those
computed in the ANATEX Model Evaluation Study. A range of concentration prediction errors could not
be established on the basis of the published results.
-------
SECTION 3
MODEL EVALUATION PROTOCOL
3.1 THE EVALUATION APPROACH
The AMES objectives are to:
(1) assess the overall performance (for the first three months of 1987), as well as the model
errors on temporal scales of 24 h, of prognostic long-range atmospheric models with respect
to transport and diffusion as a function of transport time and distance,
(2) intercompare the model performances and relate performance to fundamental modeling
approaches,
(3) identify the periods and associated meteorological conditions when each model performed
best and worst, and
(4) compare and contrast the AMES conclusions with those of similar studies using CAPTEX
data.
These objectives are accomplished by first developing a set of model performance measures relating the
basic attributes of the model-calculated transport and diffusion to those of the measurements. The values
of these performance measures are determined from comparisons between model calculations and
ANATEX measurements of the 24-h surface tracer concentrations at each of the 77 sites in the network.
From the values of these measures, the performance of each model is compared to that of a perfect model
as well as to each other.
Basic attributes of the transport and diffusion characterized from the ANATEX surface data are:
(1) magnitude of tracer concentrations,
(2) extent of horizontal diffusion,
10
-------
(3) transport speed, and
(4) transport direction.
The most accurate evaluation of a predictive model is carried out on an individual event, rather than
average behavior over an ensemble of events. The reason for this is that a model can err on individual
events, but if those errors are unbiased, it may still replicate an ensemble of measurements on the average.
Thus, it may appear to be accurate, but in fact it may not be accurate for the right reasons. If, on the other
hand, a model successfully replicates each individual event, it is likely that it has done so through accurate
representation of the relevant meteorological factors.
Therefore, one would ideally evaluate each of the characteristics listed above through the thorough
analysis of each individual release as it is carried through time and space. The 24-h-mean surface
concentrations from a single tracer release define a geometric entity of concentrations exceeding threshold
values (see page 19), heretofore referred to as a "footprint." This entity differs from an instantaneous
tracer puff in that the footprint is a composite of the surface manifestations of all instantaneous puffs within
the 24-h sampling period (Figure 2).
In the ANATEX database, it is not always possible to associate a concentration with a particular
tracer release. Footprints from separate releases have a tendency to be merged together in both the
measurements and model calculations. While many individual footprints remain identifiable for the
measurements and for each model, it is extremely rare that one particular release can be followed for the
measurements as well as for ajl of the models. Thus, the intercomparison of models using the ANATEX
database is somewhat difficult.
An additional complication with regard to tracer concentration levels and horizontal diffusion is the
spatial and temporal resolution of the sampling network. That is, the determination of horizontal diffusion
requires knowledge of the temporal changes in the width of the footprints, yet the separation of monitors
along a band is generally greater than 100 km. Thus, the potential error in estimating the width of a single
footprint can be as great as 200 km. Similarly, the uncertainty of the estimated mean tracer concentration
of a footprint can also be unacceptably large for one 24-h period, since the higher concentrations can often
elude the monitoring sites.
For these reasons, measures of the magnitude and diffusion of concentrations are assessed using
average behavior over time, despite the acknowledged weakness therein. Specifically, they are characterized
for two ensembles (both halves of the ANATEX period: 5 January - 16 February 1987 and 17 February -
29 March 1987), and for neither individual 24-h periods nor individual footprints. The splitting of the
ANATEX period was necessitated by the fact that the sets of meteorological scenarios of the two halves
were generally quite different. Due to the higher frequency of days with frontal systems stretching across
11
-------
LAT
5250
4750
5750
2500 +
-11000
' R
3 '
PDCH-1
-10600 -10200 -9800 -9400 -9000
-8600
LONG
-8200 -7800 -7400 -7000 -6600
x = missing data
R = STC release site
Figure 2. Illustration of the 6 January 1987 "footprint", or a 24-h composite of instantaneous tracer puffs
near the surface, and the centroid, Cm, for the measured PDCH concentrations from release 1.
12
-------
the northcentral U.S. after 16 February 1987 (24 days versus 14 days in the first half period [Draxler,
1988a]), the transport and horizontal spreading of the tracer tended to be different than that in the first
half. Consequently, the model performances would be expected to differ for each half.
In interpreting these measures of magnitude and diffusion, it is important to realize that any
conclusions must be somewhat one-sided. That is, if a model performs poorly on average, it can be
inferred that it must perform poorly for individual events as well. However, if a model performs well on
average, that does not necessarily mean that it successfully replicates individual events.
With regard to transport speed and direction, one must base any meaningful analysis on
concentrations associated with a specific tracer release. Such an approach is used in this evaluation,
acknowledging the weakness inherent in utilizing different subsets of the data for each model as well as the
measurements. In order to minimize this weakness, certain analyses of transport characteristics are
performed on the ensemble database. While not as powerful as the analyses of individual releases, the
ensemble results can be used to lend support to those more specific analyses.
Describing attributes of transport first requires a tracer footprint reference point so that a distance
can be calculated as a function of elapsed time. Options include the farthest edge from the source, the
maximum concentration, the concentration-weighted center, and the geometric center.
Haagenson et al. (1987) considered the first three options, or variations thereof, in their CAPTEX
evaluation of isentropic, isobaric, and sigma models. They concluded that the third option~the
concentration-weighted center-was best, since it consistently yielded the smallest rms separation for all
models. Furthermore, this option is more suitable than the geometric center option, since it is more closely
related to the core of the footprint (i.e., the highest concentrations) and is less affected by one-sided
shearing of the footprint. Also, the concentration-weighted center is minimally affected by uncertainties in
defining the footprint edges. Therefore, the concentration-weighted center or centroid was selected to
represent both calculated and measured ANATEX footprint locations.
For each of the calculated and measured footprints, the coordinates (X, Y) of the centroid were
computed from the set of n nonzero 24-h concentrations (x j ) and the coordinates of the sampling sites
(*i> y\) as:
x = £ w\ x\ i = !» - . n
where the weights, w( , were defined as-
w\ = Xj / [EX,- ] / = 1,
13
-------
As an example, Figure 3 illustrates the location of the centroids for ARL and actual footprints of PTCH
tracer at 24-h intervals up to five days after release number 1.
For each of the features of transport and diffusion listed in Section 3.1, one or more model
performance measures were adopted. These measures, identified in Table 4 and discussed in the following
sections, could be either quantitative or qualitative, where the latter would take the form of maps or plots.
Note that, due to the spatial and temporal dependence of the measurements, formal statistical inference
(i.e., testing the null hypothesis that calculated and measured parameters are statistically similar) is
generally not possible.
CENTROIDS AND DAYS SINCE RELEASE
RELEASE NUMBER 1
Figure 3. The 24-h locations of ARL (—) and actual (-
PTCH tracer.
14
-) footprint centroids for the first release of
-------
TABLE 4. MODEL PERFORMANCE MEASURES a
Ensemble statistics
Concentration
magnitude (x)
Horizontal
diffusion
24-h concentration patterns of
individual tracer releases
Transport
speed
Footprint
location
Box plots of
nonzero x's
[Section 4.1.1]l
Mean distance
of nonzero x's
along bands
[Section 4.2]
Transport
speed ratio: °
[Section 5.3]
Normalized
location
error:
[Section 5.3]
Distributional
bar charts
[Section 4.1.1]
Mean separation
error:
[Section 5.2]
Mean x
along bands
[Section 4.1.2]
a. Used to describe the agreement between characteristics of calculated and measured tracer transport and diffusion.
b. Results are discussed in the section number given in brackets.
c. The ratio of the cumulative distance (D) between the centroid of calculated footprints (Cc) and the tracer source over the
cumulative distance between the centroid of measured footprints (C,,,) and the tracer source.
d. The ratio of the cumulative distance separating the centroids of calculated and measured footprints over the cumulative
distance between the centroid of measured footprints (C,,,) and the tracer source.
e. The mean distance between the centroids of the calculated and measured footprints.
15
-------
No calculated concentration was considered for those cases when the measured concentration was
either missing or ignored on the basis of the screening criteria, thus, avoiding an inconsistency that has the
potential of dominating the centroid locations. For all but one model, there were 5,296 and 5,176 pairs (or
82% and 80% of the maximum possible pairs) for the PTCH and PDCH tracers, respectively. Since
HY-SPLIT did not calculate concentrations at three sites along the border of the network, the number of
pairs decreased by approximately 230.
32 MODEL PERFORMANCE MEASURES
3.2.1 Ensemble Concentration Means and Distributions
One commonly-used approach of assessing model performance regarding transport and diffusion is
to compare calculated concentrations to measured concentrations paired in space and time. This approach
is not suitable for an evaluation using the ANATEX data, since the temporal and spatial resolutions of the
ANATEX network is too coarse to resolve individual plumes. Furthermore, this approach has a serious
disadvantage in that the simulated mass transport could differ only slightly from the actual transport (i.e.,
be well-within the expectations of the model), yet a statistical comparison of concentrations paired both in
space and time would infer that the model performed dismally. For example, suppose after 48 h of
transport a model calculates a concentration of 20 dfL/L at site A and 0 dfL/L at site B, 30 km from
site A, while the measurements indicate that the opposite is true. A transport error of only 30 km after
48 h is hardly significant and one could conclude that the model performed very well. However, a statistical
measure based on concentrations paired in space and time (e.g., the mean square error) would falsely imply
total failure and would not by itself inform the evaluator of the proximity of the calculated plume to the
measured plume.
As an alternative to this approach, a model performance measure was adopted to assess model
performance on the basis of the differences between calculated and measured ensemble mean
concentrations along six bands of sites (Figures 4a and 4b) for each half of the ANATEX period. The sites
within each band were approximately equidistant to the tracer release site; 1,0(30-, 1600-, and 2,300-km for
the PTCH tracer released from GGW and 300-, 700-, and 1,400-km for the PDCH tracer released from
STC. Supplementing these analyses are time-series plots of 24-h mean calculated and measured
concentrations (Appendix B). Only the three bands were selected for PTCH because the sampler
spacing-and thus, our uncertainty of the true mean concentration-was considerably larger for the 1,300-
and 2,000-km bands. Some sites located near the band end points were not considered when calculating
this measure to preserve the angle from source to band end points and to exclude some of the sites that
often reported no data.
16
-------
a
o aookm
o aookm
Figure 4. The bands of sites along which ensemble-mean calculated and measured horizontal diffusion
were estimated for (a) PTCH and (b) PDCH tracers.
17
-------
With this approach, differences in mean concentrations (i.e., model biases) can be related to
transport distances. Furthermore, the ensemble-mean-concentration approach avoids a comparison of 24-h
concentrations when, for example, the centerline of the measured plume avoids the sampling sites, while
that of the calculated plume passes over a sampling site. By comparing ensemble mean concentrations, the
influence of such cases on the model performance measure can be minimized,
The ensemble-mean concentrations (x ) were calculated as follows:
where: n is the total number of concentrations along the band for every day in the ensemble period
(i.e., the number of site-days), and
X j is the concentration at site-day i.
The mean concentrations were based on only those site-days for which both a calculated and a measured
concentrations were simultaneously available. For this measure, both zero and nonzero concentrations
were used.
In addition to the ensemble mean concentration comparisons, the ensemble distributions of
measured and calculated concentrations for each half period are compared via box plots and distributional
bar charts. These graphic tools quickly display any biases that may exist. The distributional box plots
consider only nonzero concentrations to focus on the concentrations in the tracer plumes. Meanwhile, the
distributional bar charts were restricted to those calculated concentrations corresponding to available
measured concentrations passing the screening criteria. This was done to present a better comparison of
the distributions of calculated and measured concentrations.
322 Ensemble Mean Horizontal Diffusion
As mentioned earlier, it is difficult to adequately calculate the horizontal diffusion of the tracer
footprints, especially for the first 24 h of transport, due to the spatial resolution of the monitoring network.
Rather than attempting to estimate the horizontal spread for each 24-h period, we used a measure based
on the average distance along the 1,000-, 1,600-, and 2,300-km bands for PTCH tracer and the 300-, 700-,
and 1,400-km bands for PDCH tracer (Figures 4a and 4b) of nonzero concentrations. Specifically, the
mean horizontal spread is approximated by the ensemble product of the mean daily number of samplers
along a band with nonzero concentrations and the average distance between samplers for that band.
18
-------
Only the three bands were selected for PTCH because the sampler spacing--and thus, our
uncertainty of the true footprint width-was considerably larger for the 1,300- and 2,000-km bands. Some
sites located near the band end points were not considered when calculating this measure to preserve the
angle from source to band end points and to exclude some of the sites that often reported no data.
The model performance measure is the ratio of the average distance for the model calculations over
the average distance for the measurements. The averaging periods are the first and second halves of the
ANATEX sampling period. For a model that, on the average, underpredicts the horizontal diffusion, the
ratio will be less than 1.
323 Transport Speed of Discrete Tracer Footprints
The agreement between calculated and measured transport speed is quantified by the transport
speed ratio defined as: the cumulative distance between the centroid of calculated footprints and tracer
source, DC , divided by the cumulative distance between the centroid of measured footprints and tracer
source, Dm , for the same period (Figure 5). The cumulative distance is measured along the line segments
connecting the centroids of each footprint series (from one 24-h centroid to the next), as opposed to a
straight-line distance.
In calculating the centroids for the measured and calculated footprints, only concentrations at or
above a specified threshold were considered. The threshold concentrations-defined as 6 dfL/L for PTCH
and 9 dfL/L for PDCH-were necessary since: (1) many of the sites with concentrations below the
threshold values were isolated and located where actual tracer was not expected, and (2) uncertainties of
concentrations below the threshold values tended to be at least ±25% (Figure 6). These threshold
concentrations were defined so that:
(1) nearly all concentrations greater than the threshold are statistically significant at the p = 0.1
significance level, and
(2) as many discrete footprints as possible could be identified with minimal disturbance of the size
and shape of the footprints.
A measure based on cumulative distances, as opposed to one based on discrete segment distances
for single 24-h periods, was better suited for model evaluation, as illustrated by the following example.
Suppose that a model underpredicts the transport speed for one 24-h period and overpredicts the transport
speed by the same magnitude for the next 24-h period. The transport speed ratio for the first period would
Uncertainties were quantified by Draxler and Heffter (1989) for two terms: (1) the uncertainty in the gas chromatograph background,
or noise level, and (2) the uncertainty in the concentration from the analytical precision of measuring reference standard tracer volumes
for each of the three analytical laboratories.
19
-------
a
Transport Speed Comparison
D
m
m
Centroid Location Error
D
D
m
DC = distance between calculated centroid and tracer source
D m = distance between measured centroid and tracer source
Cc = centroid of the calculated footprint
Cm = centroid of the measured footprint
£> = distance between C. and C_
6 c n I
Figure 5. The model performance measures characterizing (a) the agreement between calculated and
measured transport speed and (b) footprint location errors.
20
-------
100
10
_ f
10
U (dfL/L)
10 —
i I i | f i 111 I I i i i' it
T
1 1
I M III J_
0
IiW
1
T
K
A -
A
\_
i ~
0
6^
i i i i i 11
PMCH
o PDCH
PTCH
10
X (dfL/L)
100
Note: boxes indicate the 25th and 75th percentile values, while circles show the 10th, 50th, and 80th
percentile values.
Figure 6. Box plots of cumulative frequency distribution of uncertainty (U) versus excess concentration
for PMCH, PDCH, and PTCH (from Draxler and Heffter, 1989).
21
-------
be less than 1, while the ratio for the second period would be 1-a perfect score. Since for the 48-h period
the calculated transport speed was identical to that measured, the model receives credit yet the model error
for the first 24-h period is considered. If discrete distances were used, the model would be penalized for
both 24-h periods.
For a perfect model, the ratio would equal 1 for every 24-h measurement period. If the value
usually exceeded 1, this measure would indicate that the model tended to overpredict the transport speed.
On the other hand, if the value usually is less than 1, the measure would indicate that the model tended to
underpredict the transport speed.
3.2.4 Tracer Footprint Location Error
Two performance measures describe the agreement between calculated and measured transport
direction. The primary measure is the normalized separation distance, Z>e / £>m, between the centroids of
the calculated and measured footprints (Figure 5). The normalization factor is the cumulative distance
between the centroid of the measured footprint and tracer source. By normalizing the separation distance,
less emphasis is placed on smaller model errors after long transport distances. This measure was selected
over one based on angle differences, since the separation-distance measure is commonly used to determine
calculated transport errors (Haagenson et al., 1987; Draxler, 1987; Kahl and Samson, 1988; Godowitch,
1989).
Unlike the transport speed ratio, this measure has a value of zero for a perfect model. When the
calculated centroid is to the right (left) of the measured centroid-relative to the source location—the value
of this measure is defined to be positive (negative). As was the case of the transport speed ratio, a model
can be properly rewarded for calculating the correct location of the centroid after 48 h of transport even if
it incorrectly calculated the centroid after 24 h of transport.
The second measure is the mean separation error expressed as a function of cumulative transport
time. In addition to providing a useful description of model performance, this measure is often calculated
by other model evaluators. It, therefore, affords us the means to compare the results of this model
evaluation with those of others.
22
-------
SECTION 4
ASSESSMENT OF THE MODEL PERFORMANCES
BASED ON ENSEMBLE DATA
4.1 COMPARISON OF CONCENTRATION MAGNITUDES
4.1.1 Ensemble Concentration Distributions
Graphical displays of the concentration distributions in the form of box plots and distributional bar
charts were constructed for the calculated and measured concentrations for two periods-the first half and
the second half of the ANATEX period--and for both tracers. For the box plots, the range of
concentrations, excluding those exceeding the 99th-percentile, is indicated by the two short horizontal bars.
The first, second, and third quartiles are indicated by the longer horizontal bars, where the second quartile
is the median. Means are represented by the asterisks. Since only nonzero concentrations are considered
in the box plots, a somewhat different sample is considered for each model.
When interpreting the box plots of measured concentrations, one should understand that the
distribution could slightly overstate the frequencies of very low concentrations. This is a consequence of
those cases when some low measured concentrations were actually zero. Although the data uncertainties
were assessed and a screening procedure was executed, some false occurrences of nonzero concentrations
probably remained.
The distributional bar charts were constructed from every pair of calculated and measured
concentrations. Each of these charts considers four classes:
forPTCH- OdfL/L6, 1-5 dfL/L, 6-99 dfL/L, and >99 dfL/L, and
for PDCH-- 0 dfL/L, 1-8 dfL/L, 9-99 dfL/L, and >99 dfL/L.
The ranges differed slightly for the two tracers to distinguish the concentrations below and above the
thresholds. Natural background concentrations are <0.3 dfL/L for PTCH and <1.0 dfL/L for PDCH,
both of which are below current analytical sensitivity.
6 1 dfL/L = ID"16 L/L
23
-------
PTCH (GGW) Concentration Distributions--
The PTCH box plots (Figures 7a and 7b) indicate that the median measured concentrations were
5 dfL/L and 6 dfL/L for the first and second halves, respectively. The third quartiles approximated the
measured means (denoted by the asterisks in Figures 7a and 7b) and ranged from 10 to 15 dfL/L.
The PTCH distributional bar chart (Figures 8a and 8b) shows that approximately 60% of the time,
the measured PTCH concentration was zero. Concentrations exceeded 5 dfL/L 20% of the time and
exceeded 99 dfL/L less than 2% of the time.
In the first half, the median and mean PTCH concentrations for HY-SPLIT, SRL, TCAL, and
VCAL were considerably higher than those of the measurements (Figure 7a). The medians of the first
three models exceeded 10 dfL/L and their first quartile concentrations approximated the median of the
measurements, indicating a tendency for these models to overpredict. The quartiles of the other models
were very similar to that of the measurements.
Compared to the distribution of measurements (Figure 8a), HY-SPLIT, SRL, TCAL, and VCAL
also calculated fewer occurrences of nonzero concentrations, but more occurrences of concentrations
exceeding 99 dfL/L in the first half. Thus, providing that the calculated transport speeds are accurate, the
horizontal and/or vertical diffusion calculated by these models might be underestimated. The ARL, BAT,
GAMUT, and MLAM-COARSE also calculated fewer occurrences of nonzero concentrations, but unlike
the four models mentioned above, they did not calculate more occurrences of concentrations in excess of
99 dfL/L. While the ADOM and MLAM-FINE distribution closely resembles the distribution of the
measurements, the ADPIC distribution shows considerably more occurrences of nonzero concentrations,
suggesting the possibility that this model's horizontal diffusion rate is too high and/or the calculated
transport speed is too slow.
In the second half, the SRL median and mean concentrations were an order of magnitude greater
than those of the measurements (Figure 7b). Its first quartile exceeded 10 dfL/L and was higher than the
third quartile measured concentration. Figure 8b illustrates a serious discrepancy between the SRL
distribution and the distribution of the measurements: 98% of the time, SRL calculates a zero
concentration. Apparently, the diffusion rates in this half were much too low and differed from those of
the first half.
Unlike the SRL median, the other medians were within ±3 dfL/L of the median of measurements
(Figure 7b). Again HY-SPLIT and VCAL concentrations were higher than the measurements, but less so
than in the first half. The second quartiles for ARL and BAT were considerably lower than those for the
measurements. The distributions for the remaining models were very similar to the distribution of the
measurements. Note that ADOM and MLAM-FINE results were unavailable for the second half.
24
-------
a
BOXPLOTS OF NONZERO PTCH CONCENTRATIONS
ACTUAL DATA AND MODEL PREDICTIONS
FOR THE PERIOD JANUARY 5 THROUGH FEBRUARY 16
ACTUAL ADOM ADPIC ARL
BAT COARSE FINE GAMUT HYSPLIT SHL
MODEL
TCAL VCAL
BOXPLOTS OF NONZERO PTCH CONCENTRATIONS
ACTUAL DATA AND MODEL PREDICTIONS
FOR THE PERIOD FEBRUARY 17 THROUGH MARCH 29
T T
ACTUAL AOPIC
COARSE GAMUT
MODEL
Figure 7. Box plots comparing quartiles and means (*) of calculated and measured PTCH tracer
concentrations at the 77 surface sites during the periods: (a) 5 January - 16 February 1987
and (b) 17 February - 29 March 1987.
25
-------
DISTRIBUTION OF PTCH CONCENTRATIONS
ACTUAL DATA AND MODEL PREDICTIONS
FOR THE PERIOD JANUARY 5 THROUGH FEBRUARY 16
100-
90-
80-
70-
P 60-
E
fl
C SO-
E
N
T 40-
30-
20-
10-
0
ACTUAL ADOM ADPIC ARL BAT COARSE FINE GAMUT HYSPLIT SRL TCAL VCAL
MODEL
PTCH CONCENTRATION DFL/L I I 0 L\VO 1-5 E£33i 6-99 •• 100 +
DISTRIBUTION OP PTCH CONCENTRATIONS
ACTUAL DATA AND MODEL PREDICTIONS
FOR THE PERIOD FEBRUARY 17 THROUGH MARCH 29
90-
80-
70-
P 60-
E
R
C 50-
E
N
T 40-
ACTUAL AOPIC ARL
PTCH CONCENTRATION DFL/L
BAT COARSE GAMUT HYSPLIT SRL
MODEL
TCAL VCAL
LVXXI 1-5
6-99
Figure 8. Frequency distributions of calculated and measured PTCH tracer concentrations at the
77 surface sites during the periods: (a) 5 January - 16 February 1987 and (b) 17 February
29 March 1987.
26
-------
Nonzero concentration occurred much more often for MLAM-COARSE, TCAL, and VCAL than
for the measurements during the second half (Figure 8b). This suggests that the horizontal diffusion rates
calculated by these models may be too high for this half, which has a higher frequency of cyclonic and/or
frontal passages and accompanying wind shears.
It is noteworthy to contrast the poor agreement of TCAL and VCAL (both SLL models) with the
good agreement of GAMUT. For the complicated wind patterns common to the second half, the
multilayer model produced a distribution much closer to that of the measurements than did the single-layer
models. In addition to the GAMUT distribution, the ARL distribution closely resembled that of the
measurements. The remaining distributions (i.e., ADPIC, BAT, and HY-SPLIT) showed a higher
frequency of zero concentrations. For both BAT and HY-SPLIT, this was consistent with the first half;
however, for ADPIC, the first-half distribution showed a much lower frequency of zero concentrations.
It is also noteworthy to mention that HY-SPLIT-a version of ARL that used finer resolved
meteorological input data-predicted PTCH concentrations in excess of 99 dfL/L more often than ARL,
suggesting that the HY-SPLIT diffusion rate was slower than that of ARL. This difference was also
observed in the first half.
PDCH (STC) Concentration Distributions--
Distributional differences were detected between PDCH and PTCH. As the second pair of box plots
(Figures 9a and 9b) illustrate, low concentrations were measured more often for PDCH than for PTCH.
The median concentrations for both halves were 3 dfL/L, about a factor of 2 lower than the PTCH
medians. Also lower were the third quartiles-9 dfL/L and 8 dfL/L for the first half and the second half,
respectively. The higher occurrence of low PDCH concentrations can be explained by the lower PDCH
emission rates-on the average, 41.6% lower than those for PTCH.
For both halves, all mean calculated concentrations exceeded the mean measurements and, in some
cases, exceeded them by an order of magnitude. (This was not observed for the PTCH concentrations
since the mean concentration, which is greatly influenced by the highest values, was based on concentrations
at sites at least 500 km downwind of the tracer source-much farther than the closest sites to the PDCH
source.) In addition, nearly all the models calculated maximum concentrations considerably higher than
those measured; some exceeded 500 dfL/L-more than double those measured. As was the case with the
first-half PTCH concentrations, first-half PDCH quartiles were highest for GAMUT, HY-SPLIT, SRL,
TCAL, and VCAL. However, all model medians were higher than the median of the measurements. In
addition, the third quartiles were at least twice those of the measurements for all models but ADOM and
ADPIC. The ADOM and ADPIC quartiles closely resembled those of the measurements.
27
-------
a
BOXPLOTS OF NONZERO OC-PDCH CONCENTRATIONS
ACTUAL DATA AND MODEL PREDICTIONS
FOR THE PERIOD JANUARY 5 THROUGH FEBRUARY 16
0 1000-
c
p
D
c
H
C
0 100-
N
C
T
H
A
T
I
0 10-
N
F
L
/
L 1-
T
'
1
1
ACTUAL
ADOM
i
1
•
1 1
T
1 ... . — , , _, j !
ADPIC AOL BAT COARSE FINE GAMUT HYSPLIT SRL TCAL VCAL
MODEL
BOXPLOTS OF NONZERO OC-PDCH CONCENTRATIONS
ACTUAL DATA AND MODEL PREDICTIONS
FOR THE PERIOD FEBRUARY 17 THROUGH MARCH 29
0 1000
C
P
D
C
ACTUAL ADPIC
"T
ARL
—T—
BAT
COARSE GAMUT
MODEL
Figure 9. Box plots comparing quartiles and means (*) of calculated and measured PDCH tracer
concentrations at the 77 surface sites during the periods: (a) 5 January - 16 February 1987
and (b) 17 February - 29 March 1987.
28
-------
The differences between the model and measured PDCH distributional bar charts (Figures lOa
and lib) for the first half resembled those in the first-half PTCH charts. For example, SRL calculated
concentrations in excess of 99 dfL/L most often, while ADPIC calculated nonzero concentrations most
often. In addition, ARL, BAT, GAMUT, HY-SPLIT, TCAL, and VCAL nonzero concentrations occurred
less often than the nonzero measurements--15% to 25% versus nearly 40%, respectively. More importantly,
however, the frequency of PDCH concentrations exceeding the threshold for these six models was about the
same as that for the measurements. The ADOM distribution closely resembled the distribution of
measurements. The distributions of the final two models-MLAM-COARSE and MLAM-FINE--were
similar to the distribution of measurements in that the frequencies of nonzero concentrations were virtually
the same, but their frequencies of concentrations above the threshold were nearly double that for the
measurements.
In the second half. SRL quartiles were once again the highest (Figure lOb). Also, as the case for
PTCH, the SRL first quartile approximated the median measured concentration. The medians of the other
models were within ±2 dfL/L of the median of the measurements. With the exception of TCAL, the model
third quartiles were higher, but not nearly as much as in the first half.
As was the case for its PTCH concentrations, SRL PDCH concentrations were zero much more
often and MLAM-COARSE, TCAL, and VCAL concentrations of PDCH were nonzero more often than
those measured indicating a problem with diffusion. To a lesser degree, BAT and HY-SPLIT
concentrations also were zero more often, as was the case in the first half; this suggests that these two
models slightly underestimate horizontal diffusion. In contrast, ADPIC, ARL, and GAMUT concentration
distributions resembled those of the measurements.
4.1.2 Mean Concentrations Along Bands
One test assessing the combined effects of model transport speeds and diffusion rates is the
comparison between model and measured mean concentrations at specific distances downwind of GGW
and STC. If, for example, the model mean concentrations are appreciably greater than the measured mean
concentrations, one or more of the following is true about the model: (1) the transport speeds are too
high, (2) the horizontal diffusion rates are too low, or (3) the vertical diffusion rates are too low.
Unfortunately, the evaluation data set cannot distinguish between these error sources. Across three bands
of samplers-where samplers along a band are quasi-equidistant from the tracer sources-the ensemble
mean concentrations of the calculations and measurements are determined and compared for each tracer
for each half of the ANATEX period.
29
-------
a
P 60-
E
R
C 50-
E
N
T 40-
DISTRIBUTION OF OC-PDCH CONCENTRATIONS
ACTUAL DATA AND MODEL PREDICTIONS
FOR THE PERIOD JANUARY 5 THROUGH FEBRUARY 16
ACTUAL ADOM ADPIC AHL BAT COARSE FINE GAMUT HYSPLIT SHL TCAL VCAL
MODEL
OC-PDCH CONCENTRATION DFL/L I I 0 [V\V1 1-3 tSSSggj 9-99 ••• 100+
P 60
E
R
C 50
E
N
T 40
DISTRIBUTION OF OC-PDCH CONCENTRATIONS
ACTUAL DATA AND MODEL PREDICTIONS
FOR THE PERIOD FEBRUARY 17 THROUGH MARCH S3
\V
ACTUAL ADPIC AHL BAT COAHSE GAMUT HYSPLIT SHL TCAL VCAL
MODEL
OC-PDCH CONCENTRATION DFL/L
i\vo i-a
9-99
100+
Figure 10. Frequency distributions of calculated and measured PDCH tracer concentrations at the 77
surface sites during the periods: (a) 5 January - 16 February 1987 and (b) 17 February -
29 March 1987.
30
-------
Mean PTCH (GGW) Concentrations-
The mean PTCH concentration plot for the first half (Figure lla) shows a wide range of model
mean concentrations as the tracer plumes were transported from their sources to the 1,000-, 1,600- and
2,300-km bands. The mean measured concentration decreased sharply from nearly 18 dfL/L at 1,000 km to
5 dfL/L at 1,600 km and then decreased at a much slower rate to more than 3 dfL/L at 2,300 km
(Table 5). The models tended to smooth through the steepest segment of the gradient; that is, the model
mean concentrations generally were lower closest to the source, while the model mean concentrations were
more comparable to the mean measured concentrations at the mid-range. Most models were within
±1 dfL/L of the mean measured concentration at the 2,300-km band.
Most obvious is the fact that SRL mean concentrations in the first half were the highest for each
band and are 2 to 3 times higher than the measured mean concentrations. In contrast, SRL mean
concentrations in the second half (Figure lib) were the lowest and approach zero at the 1,600- and
2,300-km bands. Thus, as seen before, this model's diffusion rates and/or transport speeds appeared to be
inconsistent with reality.
ADPIC, HY-SPLIT, TCAL, and VCAL replicated the measured mean concentration gradients
rather well for the first half, but of these models only ADPIC and HY-SPLIT replicated the gradients for
the second half. ADOM also replicated the gradient in the first half, while that of MLAM-FINE was much
flatter. The mean concentrations of both ADOM and MLAM-FINE tended to be lower than the mean
measured concentrations, especially at the 1,000-km band, where calculated concentrations were
approximately 50% lower. In the second half, both TCAL and VCAL mean concentrations were much
higher than the mean measured concentrations. This concurred with the distributional bar charts discussed
earlier (Figure 8b). Unlike the mean concentrations of the single-layer TCAL and VCAL, GAMUT mean
concentrations closely resembled the mean measured concentrations with the exception of the first-half
mean at the 1,000-km band, where its concentration was a factor of 3 too low.
ARL, BAT, and MLAM-COARSE oversmoothed the gradient as well as underestimated the mean
concentrations for both halves. This suggests that these models did one or both of the following:
(1) transported the tracer too slowly to allow more dilution before the tracer plume leaches
each band, and/or
(2) diffused the tracer too quickly in the horizontal and/or vertical to overstate the dilution of
the tracer plume at the surface.
The analyses in subsequent sections will shed more light on the causes of the apparent underestimates of
mean concentrations.
31
-------
OBSERVED AND PREDICTED MEAN PTCH, BY BAND
FOR THE PERIOD JANUARY 5 THROUGH FEBRUARY 16
28 H
M
E 26-
A
N 24-
C 22-
Sa
-------
TABLE 5. THE MEAN CALCULATED AND MEASURED TRACER CONCENTRATIONS ALONG
BANDS OF SAMPLERS FOR THE TWO HALVES OF THE ANATEX PERIOD
PTCH
Tracer concentration, dfL/L
(Ratio of calculated to measured)
Model
Measured
SLL
MIL
MLE
SRL
TCAL
VCAL
ARL
BAT
GAMUT
HY-SPLIT
MLAM-FINE
HLAM-COARSE
ADOM
ADPIC
Measured
SLL
MLL
MLE
SRL
TCAL
VCAL
ARL
BAT
GAMUT
HY-SPLIT
MLAM-FINE
MLAM-COARSE
ADOM
ADPIC
1,000 km
17.7
43.7
23.3
21.6
8.3
9.8
5.3
15.8
6.9
7.3
10.7
13.6
10.1
1.2
16.9
26.8
7.3
4.7
9.2
12.8
--
19.3
--
3.8
(2.5)
(1.3)
(1.2>
(0.5)
(0.6)
(0.3)
(0.9)
(0.4)
(0.4)
(0.6)
(0.8)
(0.1)
(1.7)
(2.7)
(0.7)
(0.5)
(0.9)
(1.3)
(1.9)
(0.4)
1,600
First
4.9
11.8
7.6
7.6
2.9
4.7
2.8
8.1
4.2
4.0
2.7
4.9
Second
2.8
0.0
8.9
9.3
1.5
1.4
2.7
1.3
--
9.8
--
1.0
km
half: 1/5/87
(2
(1
(1
(0
(1
(0
(1
(0
(0
(0
(1
.4)
.6)
.6)
.6)
-0)
.6)
.7)
.9)
.8)
.6)
.0)
half: 2/17/87
(0
(3
<3
(0
(0
(1
(0
(3
(0
.0)
.2)
.3)
-5)
.5)
.0)
.5)
.5)
.4)
2,300 km
Concentration ratios
between bands
1,000-
1,600 km
1,600-
2,300 km
- 2/16/87
3.1
10.6
3.9
7.7
1.7
1.5
3.9
4.6
4.5
4.3
1.9
3.0
(3.4)
(1.3)
(2.5)
(0.5)
(0.5)
(1.3)
(1-5)
(1.5)
(1.4)
(0.6)
(1.0)
3.
3.
3.
2.
2.
2.
1.
2.
1.
1.
4.
2.
6
7
1
8
9
1
9
0
6
8
0
8
1.6
1.1
1.9
1.0
1.7
3.1
0.7
1.8
1.9
0.9
1.4
1.6
- 3/29/87
1.8
0.0
6.2
5.3
2.6
0.3
3.3
0.6
--
6.1
--
0.5
(0.0)
(3.4)
(2.9)
(1.4)
(0.2)
(1.8)
(0.3)
(3.4)
(0.3)
3.
-
1.
2.
4.
3.
3.
9.
-
6
-
9
9
9
4
4
8
-
2.0
-
3.
-
8
1.6
--
1.4
1.8
0.6
4.7
0.8
2.2
--
1.6
--
2.0
(continued)
33
-------
TABLE 5. (CONCLUDED)
PDCH
Tracer concentration, dfL/L
(Ratio of calculated to measured)
Model
Measured
SLL
MIL
HLE
SRL
TCAL
VCAL
ARL
BAT
GAMUT
HY-SPLIT
MLAM-FINE
MLAM-COARSE
ADOM
ADPIC
Measured
SLL
MLL
MLE
SRL
TCAL
VCAL
ARL
BAT
GAMUT
HY-SPLIT
MLAM-FINE
MLAM-COARSE
ADOM
ADPIC
300
8.7
46.1
13.4
32.2
9.4
21.7
19.1
16.1
46.9
44.2
17.0
18.2
15.1
10.6
12.4
18.5
8.0
3.6
12.2
6.5
--
14.7
--
7.6
km
(5.3)
(1.5)
(3.7)
(1.1)
(2.5)
(2.2)
(1.9)
(5.4)
(5.1)
(2.0)
(2.1)
(0.7)
(0.8)
(1.2)
(0.5)
(0.2)
(0.8)
(0.4)
(1.0)
(0.5)
700 km
First
4.2
20.8
13.7
15.8
5.2
10.5
12.8
7.1
15.3
16.3
13.2
7.9
Second
5.0
5.4
10.3
14.2
4.7
4.5
8.2
3.8
--
9.0
--
3.5
half:
(5.0)
(3.3)
(3.8)
(1.2)
(2.5)
(3.0)
(1.7)
(3.6)
(3.9)
(3.1)
(1.9)
half:
(1.1)
(2.1)
(2.8)
(0.9)
(0.9)
(1.6)
(0.8)
(1.8)
(0.7)
1
,400
km
Concentration ratios
between bands
300
700
-
km
700 -
1,400 km
1/5/87 - 2/16/87
3
17
4
5
3
2
5
4
13
9
1
4
2/17/87 -
2
8
3
6
2
1
3
2
5
1
.9
.5
.7
.9
.6
.7
.1
.1
.4
.8
.8
.5
(4.5)
(1.2)
(1-5)
(0.9)
(0.7)
(1.3)
(1.1)
(3.4)
(2.5)
(0.5)
(1.2)
2.
2.
1.
2.
1.
2.
1
2
0
0
8
1
1.5
2.
3.
2.
3
1
7
1.3
2.3
1.1
1.2
2.9
2.7
1.4
3.9
2.5
1.7
1.1
1.7
7.3
1.8
3/29/87
.0
.4
.8
.4
.4
.1
.5
.8
--
.3
--
.9
(4.2)
(1.9)
(3.2)
(1.2)
(0.6)
(1.8)
(1.4)
(2.7)
(1.0)
3.
2.
1.
1.
1.
0.
1.
1.
-
1.
-
2.
0
0
2
3
7
8
5
7
-
6
-
2
2.5
0.6
2.7
2.2
2.0
4.1
2.3
1.4
--
1.7
--
1.8
34
-------
Mean PDCH (STC) Concentrations--
For the first half, the mean measured concentrations (Figure 12a) peaked to nearly 9 dfL/L at
300 km and decreased at a much slower rate than those for PTCH to approximately 4 dfL/L at both
700 km and 1,400 km. In contrast, the second-half mean concentrations (Figure 12b) peaked at 15 dfL/L
at 300 km, decreased sharply to 5 dfL/L at 700 km, and then decreased at a slower rate to 2 dfL/L at
1,400 km.
The ARL model replicated these characteristics to the greatest extent, especially during the first half.
However, for the second half, like the other models, it oversmoothed the gradient between 300 km and
700 km. In addition, its 300-km mean concentration was about half that of the mean measured
concentration.
Decreases of ADPIC, BAT, GAMUT, and HY-SPLIT mean concentrations with distance more or
less paralleled those of the mean measurements in the first half, but the model means were considerably
higher. All four models oversmoothed the gradient in the second half. However, for both halves, the
model mean concentrations at 1,400 km resembled those of the measurements. At the closer bands, the
ADOM mean concentrations exceeded the mean measured concentrations by factors of 2 and 3, but at
1,400 km, the ADOM mean resembled that of the measurements.
The MLAM-FINE, MLAM-COARSE, SRL, and VCAL means were much higher than measured
means and showed much steeper gradients between 300 km and 700 km for the first half. The relatively
high concentrations near STC were also illustrated by the distributional bar charts (Figure lOa). Also
during this half, the TCAL mean at 700 km was comparable to its 300-km mean. In the second half, based
on the few nonzero concentrations, the SRL mean at 1,400 km was greater than its mean at 700 km. The
VCAL and MLAM-COARSE means tended to be higher than the mean measured concentrations,
especially at 1,400 km, where calculated concentrations were 2-to-3 times greater.
42 COMPARISON OF HORIZONTAL DIFFUSION OF TRACER PUFFS
Given the distribution of samplers in the ANATEX network, horizontal diffusion of a tracer could
not be determined precisely. As a means of assessing the differences in ensemble-average diffusion rates of
the models and the actual tracer, the average distances of nonzero concentrations calculated by each model
along three bands for each half of the ANATEX period were compared to those of the measurements. If a
model's average horizontal diffusion rate differed from the actual average rate, the differences should
become evident from these comparisons.
35
-------
OBSERVED AND PREDICTED MEAN OC-PDCH. BY BAND
FOR THE PERIOD JANUARY 5 THROUGH FEBRUARY 16
E 44
A
N 40
C
0 36
N
C 32
E
N 28
2"
T
I 20
0
N IE
DISTANCE OF BAND FROM SOURCE. KM
•6 OBSERVED
V VCAL
S SHL
-A ARL
-G GAMUT
-P ADPIC
-B BAT
-F FINE
-H HYSPLIT
-T- TCAL
C COARSE
-0 ADOM
OBSERVED AND PREDICTED MEAN OC-PDCH, BY BAND
FOR THE PERIOD FEBRUARY 17 THROUGH MARCH S3
C
0 IB
N
C 16
E
N jj
T
>
T
I 10
0
N B
- -S
B-
-0 OBSERVED
•V VCAL
-P ADPIC
DISTANCE OF BAND FBOM SOURCE. KM
-A AHL
-G GAMUT
-H HYSPLIT -
BAT
COARSE
-T
-S
TCAL
SRL
Figure 12. Mean calculated and measured PDCH concentrations along three bands of sites downwind
of STC during the periods: (a) 5 January - 16 February 1987 and (b) 17 February -
29 March 1987.
36
-------
PTCH (GGW) Concentrations--
As Figures 13a and 13b illustrate, the mean distance of nonzero measured PTCH concentrations
along three bands of sites-approximately 1,000 km to 2,300 km downwind of GGW-increased from 590 km
to 1,441 km during the first half and from 546 km to 1,091 km during the second half (Table 6). The mean
rate of change relative to the transport distance (i.e., the "mean spread rate") approximated 0.42 km/km for
both half periods, with the exception of the 0.83 km/km first-half-period spread rate between 1,600 km and
2,300 km.
Compared to the mean distances for the measured concentrations, the first-period means for the
calculated concentrations tended to be lower by 10% to 70%. ADPIC was a noticeable exception; its mean
distances were 20% to 60% greater. Mean distances for ADOM, MLAM-FINE, and SRL most closely
resembled those for the measurements; all were within 30% for the three bands.
First-period spread rates also tended to be lower than those for the measurements. MLAM-FINE,
the one exception, had spread rates of nearly 1.0 km/km. The spread rates for ARL and its sibling,
HY-SPLIT, were much lower, approximately 0.15 to 0.25 km/km.
During the second period, the mean distances for TCAL, VCAL, and MLAM-COARSE
concentrations tended to be factors of 2.0 to 2.5 greater than those for the measurements. In contrast to
the first period, ADPIC mean distances were 50% to 70% lower than the means for the measurements.
The mean distances for ARL and GAMUT were within 10% and 30% for those of the measurements,
respectively. The extremely low SRL mean distances indicated a serious modeling problem during the
second period.
The range of spread rates during the second period varied much more than that of the first period.
In fact, some models, particularly BAT and HY-SPLIT, indicated a decrease in the mean distances with
increasing transport distances, while ADPIC mean distances changed little. ARL most closely resembled
the spread rates of the measurements.
PDCH (STC) Concentrations--
The mean distances of nonzero measured PDCH concentrations (Figures 14a and 14b) ranged from
approximately 650 km at 300 km downwind to 1,350 km at 1,400 km downwind of STC during each period
(Table 6). Meanwhile, the spread rates ranged from 0.37 km/km to 0.92 km/km, similar to the range for
PTCH.
During the first period, the mean distances for the calculated concentrations again tended to be
lower than those for the measured concentrations. In addition, the ratios of calculated-to-measured mean
distances for PDCH were very similar to those for PTCH. As before, ADOM, MLAM-FINE, and SRL (in
37
-------
OBSERVED AND PREDICTED NONZERO PTCH AS MEASURE OF DISPERSION
GIVEN AS AVEHAGE DISTANCE ALONG WHICH NONZERO
CONCENTRATIONS ARE PRESENT ON A BAND.
FOR THE PERIOD JANUARY 5 THROUGH FEBRUARY 16
D 2200
I
s 2000-
T
1800-
0
F 1600-
N 1100
0
N 1200-
Z
E 1000-
R
0 BOO
p 600-
T
C 400
H
200
0-
1600
DISTANCE OF BAND FROM SOURCE. KM
-6 OBSERVED
V VCAL
S SHL
-A AHL
-G GAMUT
-P ADPIC
-8
-f
-H
BAT
FINE
HYSPLIT
-T TCAL
C COARSE
-B AOOM
OBSERVED AND PREDICTED NONZERO PTCH AS MEASURE OF DISPERSION
GIVEN AS AVEHAGE DISTANCE ALONG WHICH NONZERO
CONCENTRATIONS ARE PRESENT ON A BAND.
FOR THE PERIOD FEBRUARY 17 THROUGH MARCH 29
2400-
I
s 2000-
T
1800
0
F 1600-
N 1400-
0
N 1200-
Z
E 1000
R
0 800
P 600
C 400
H
200
K °
M
DISTANCE OF BAND FROM SOURCE. KM
-9 OBSERVED
V VCAL
f ADPIC
•A ARL
-G GAMUT
•H HYSPLIT
BAT
COARSE
TCAL
SRL
Figure 13. Mean horizontal diffusion of calculated and measured tracer puffs as indicated by the
average distance of nonzero PTCH concentrations along three bands of sites downwind of
GGW during the periods: (a) 5 January - 16 February 1987 and (b) 17 February - 29 March
1987.
38
-------
TABLE 6. THE MEAN DISTANCE OF NONZERO CALCULATED AND MEASURED
CONCENTRATIONS AND MEAN SPREAD RATES ALONG
BANDS OF SAMPLERS
PTCH
Mean
distance of nonzero concentrations,
(Ratio of calculated to measured)
km
Mean spread rates
between bands, km/km
1,000-
Model
1,000
km
1,600
km
First half: 1/5/87 -
Measured
SLL
MIL
MLE
SRL
TCAL
VCAL
ARL
BAT
GAMUT
HY-SPLIT
MLAM-FINE
MLAM- COARSE
ADOM
ADPIC
590
575
412
363
427
340
182
314
420
337
760
923
(1.0)
(0.7)
(0.6)
(0.7)
(0.6)
(0.3)
(0.5)
(0.7)
(0.6)
(1.3)
(1-6)
861
714
527
627
520
496
317
401
999
718
960
1,269
(0.8)
(0.6)
(0.7)
(0.6)
(0.6)
(0.4)
(0.5)
(1.2)
(0.8)
(1.1)
(1.5)
Second half: 2/17/87
Measured
SLL
MLL
MLE
SRL
TCAL
VCAL
ARL
BAT
GAMUT
HY-SPLIT
MLAM-FINE
MLAM-COARSE
ADOM
ADPIC
546
—
1,266
1.279
559
478
520
482
...
1.092
...
256
(2.3)
(2.3)
(1.0)
(0.9)
(1.0)
(0.9)
(2.0)
(0.5)
800
—
1,931
1.830
708
423
671
303
...
1.752
...
198
(2.4)
(2.3)
(0.9)
(0.5)
(0.8)
(0.4)
(2.2)
(0.3)
(continued)
39
2,300
km
1,600 km
1,600-
2,300 km
2/16/87
1,441
930
686
1,167
702
717
808
511
1,678
1,144
1,190
1,716
(0
(0
(0
(0
(0
(0
(0
(1
(0
(0
(1
.7)
.5)
.8)
.5)
.5)
.6)
.4)
.2)
.8)
.8)
.2)
0.
45
0.23
0.
0.
0.
0.
0.
0.
0.
0.
0.
0.
19
44
16
26
23
15
97
64
33
58
0.83
0.30
0.23
0.77
0.26
0.32
0.70
0.16
0.97
0.61
0.33
0.64
- 3/29/87
1,091
—
2,281
1,782
974
325
766
275
...
2.148
...
325
(2
(1
(0
(0
(0
(0
(2
.1)
.6)
.9)
.3)
.7)
.3)
.0)
(0.3)
0.
-
1.
0.
0.
-0.
0.
-0.
-
1.
-
-0.
42
--
11
92
25
09
25
03
--
10
--
10
0.42
—
0.50
-0.07
0.38
-0.14
0.14
-0.04
...
0.57
—
0.18
-------
TABLE 6. (CONCLUDED)
PDCH
Model
Mean
distance of nonzero concentrations, km
(Ratio of calculated to measured)
300 km
700 km
First half: 1/5/87 -
Measured
SLL
MIL
MLE
SRL
TCAL
VCAL
ARL
BAT
GAMUT
HY-SPLIT
MLAM-FINE
MLAM-COARSE
ADOM
ADPIC
674
547
213
369
351
288
225
253
530
536
805
927
(0.8)
(0.3)
(0.6)
(0.5)
(0.4)
(0.3)
(0.4)
(0.8)
(0.8)
(1.2)
(1.4)
820
934
442
606
499
385
414
349
906
834
1,050
1,412
(1.1)
(0.5)
(0.7)
(0.6)
(0.5)
(0.5)
(0.4)
(1.1)
(1.0)
(1.3)
(1.7)
Second half: 2/17/87
Measured
SLL
MLL
MLE
SRL
TCAL
VCAL
ARL
BAT
GAMUT
HY-SPLIT
MLAM-FINE
MLAM-COARSE
ADOM
ADPIC
658
139
700
1.086
531
266
676
428
...
1.261
...
676
(0.2)
(1.1)
(1.7)
(0.8)
(0.4)
(1.0)
(0.7)
(1.9)
(1.0)
1,026
230
1,093
1.668
815
441
978
585
...
1.841
...
882
(0.2)
(1.1)
(1.6)
(0.8)
(0.4)
(1.0)
(0.6)
(1.8)
(0.9)
1,400 km
Mean spread rates
between bands, km/km
300 -
700 km
700 -
1,400 km
2/16/87
1,368
1,275
661
1.144
847
577
930
512
1,610
1,600
1,015
2,205
(0
(0
-------
a
V
G
2400-
D
I 2200 •
S
T 2000-
0 1800-
F
0 m
c
_ 600-
P
D 400-
C
H 200-
0-
OBSERVED AND PREDICTED NONZERO OC-PDCH AS MEASURE OF DISPERSION
GIVEN AS AVERAGE DISTANCE ALONG WHICH NONZERO
CONCENTRATIONS ARE PRESENT ON A BAND.
FOR THE PERIOD JANUARY 5 THROUGH FEBRUARY IB
APPHOX DIST OF BAND FROM SOURCE. KM
-6 OBSERVED
-V VCAL
-S SRL
-A ARL
-G GAMUT
-P ADPIC
B BAT
-F FINE
-H HYSPLIT
-T- TCAL
C COARSE
-B ADOM
OBSERVED AND PREDICTED NONZERO OC-PDCH AS MEASURE OF DISPERSION
GIVEN AS AVERAGE DISTANCE ALONG WHICH NONZERO
CONCENTRATIONS ARE PRESENT ON A BAND.
FOR THE PERIOD FEBRUARY 17 THROUGH MARCH 29
2400-
2200-
2000-
1800-
1600-
1400-
1200-
1000-
800-
600-
400-
200-
0-
APPHOX DIST OF BAND FROM SOURCE. KM
•9 OBSERVED
V VCAL
-P ADPIC
-A ARL
-G GAMUT
-H HYSPLIT
BAT
COARSE
-T TCAL
S SRL
Figure 14. Mean horizontal diffusion of calculated and measured tracer puffs as indicated by the
average distance of nonzero PDCH concentrations along three bands of sites downwind of
STC during the periods: (a) 5 January - 16 February 1987 and (b) 17 February - 29 March
1987.
41
-------
addition to MLAM-COARSE) mean distances were within ±30% of those for the measurements, while
ADPIC mean distances were 40% to 70% greater. BAT, HY-SPLIT, and TCAL mean distances were
consistently lower by 50% to 70%.
Although first-period mean distances for ARL, GAMUT, and VCAL were lower than those for the
measurements, their spread rates (0.37 km/km to 0.77 km/km) most closely resembled those for the
measurements. BAT and HY-SPLIT spread rates (approximately 0.25 km/km) were the lowest, while
ADPIC, MLAM-FINE, and MLAM-COARSE spread rates were the greatest (approximately 0.75 km/km
to 1.20 km/km). ADOM, SRL, and TCAL spread rates were lower between the 300 km and 700 km
bands, but greater between the 700 km and 1,400 km bands.
Similar to the second-period mean distances for PTCH calculations, the mean distances for VCAL
and MLAM-COARSE PDCH calculations were generally 60% to 90% greater than those for the
measurements while SRL mean distances were much (80%) lower. ARL and GAMUT mean distances
again best replicated those for the measurements (within ±20% at all three bands). For PDCH, TCAL
mean distances were within ±30% of those for the measurements. As with the first period, second-period
mean distances for BAT were low (by 60% to 70%) and for HY-SPLIT (by 30% to 40%). In contrast to
those of the first period and the two PTCH periods, ADPIC mean distances for PDCH in the second
period were within ±10% of those for the measurements.
As before, the mean spread rates for ARL and GAMUT (ranging from 0.24 km/km to 0.76 km/km)
most closely resembled those for the measurements. ADPIC mean spread rates (approximately
0.50 km/km) also resembled those of the measurements. BAT and HY-SPLIT mean spread rates again
were low (ranging from near 0.00 km/km to 0.44 km/km), as were those for SRL (<0.25 km/km). TCAL
and VCAL mean spread rates were greatest (0.85 km/km to 1.46 km/km). MLAM-COARSE spread rates
were much greater between the 300 km and 700 km bands, but much lower between the 700 km and
1,400 km bands.
42
-------
SECTION 5
ASSESSMENT OF THE MODEL PERFORMANCES BASED ON
COMPARISONS OF INDIVIDUAL TRACER RELEASES
5.1 PERFORMANCE AS RELATED TO METEOROLOGICAL SCENARIO
From a set of discrete 24-h tracer footprints (i.e., those that did not merge with the footprints of
other releases of the same tracer), 24-h mean transport speeds and 24-h locations of the footprint centroids
were determined. The speeds and locations of all footprints common to both the sets of measurements and
the calculations for each model were compared to assess the accuracy of calculated transport speeds and
directions. Table 7 identifies the number of 24-h footprints common to both model and measured sets for
each tracer release. The relatively small number of ADPIC footprints results from the fact that this model
often showed long residence times of tracer across the grid, making it difficult to distinguish discrete
footprints. The time-series plots in Appendix B of the 24-h mean calculated and measured concentrations
at various transport distances, often demonstrate this unique behavior.
Also, the relationship between model performance and meteorological scenario was investigated.
This section discusses the results of these comparisons and the behavior of the models with respect to the
meteorological scenarios. Section 5.2 discusses the mean model location errors for ANATEX and
compares the range of errors with that for CAPTEX. Section 5.3 summarizes ANATEX performances by
model genres and assesses the performances of individual models.
Three types of meteorological scenarios were considered. Each type was based simply on the timing
of cyclonic and/or frontal interceptions of the measured footprints, if any, relative to the transport period
(Table 8). The purpose was to distinguish the simpler flow patterns from the more complicated flow
patterns. That is, cyclonic and frontal interceptions usually introduce complexities in the horizontal and
vertical wind flow patterns that may accelerate vertical mixing and horizontal diffusion. Moreover, tracer
looping in the horizontal plane can result from these phenomena.
The approach of relating model performances to meteorological scenarios was quite simple. Firstly,
each footprint-day beyond 10 h of transport was labeled as either being or not being intercepted by a
cyclone and/or front since the tracer release ("/' and "NT, respectively) [Table 9]. Model performances
43
-------
TABLE 7. NUMBER OF CALCULATED AND MEASURED FOOTPRINT PAIRS AND
THE TYPE OF METEOROLOGICAL SCENARIO FOR EACH TRACER RELEASE3
PTCH
Release
number
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
31
32
33
Total
Type of
met.
scenario
I
I
II
III
II
III
III
I
II
III
III
III
II
II
II
II
I
I
III
I
I
I
II
III
III
I
III
I
II
III
I
II
III
SLLb
SRL TCAL
3
3 1
5
2 2
1 1
5 2
1 1
3
2
1
2
1
2
5
12 31
VCAL
3
3
2
2
2
2
2
1
3
1
21
ARL
2
1
3
3
1
2
1
4
2
1
2
1
2
2
3
2
32
BAT
2
3
5
2
1
2
1
2
1
3
1
2
25
MLL MLE
HY- HLAM MLAM-
GAMUT SPLIT FINE COARSE ADOM ADPIC
2122 2
1 33
5533 23
2222 32
11 11
15 1
1
12 1
21 2
1 11
3333 1
111 1
2
1
1 2
1
1
23 21 17 18 12 8
(continued)
44
-------
TABLE?. (CONCLUDED)
PDCH
Release
number
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
31
32
33
Total
Type of
met .
scenario
I
III
I
III
II
III
II
II
II
III
II
III
I
I
III
III
I
I
III
I
II
III
I
III
III
I
III
I
I
II
III
III
III
SRL
3
2
3
2
2
1
2
1
I
1
3
1
2
24
SLLb
TCAL
4
1
3
2
3
2
2
4
2
1
2
3
3
2
2
1
37
MLL MLE
VCAL
5
2
2
4
2
5
4
2
3
2
2
1
2
2
2
2
1
43
ARL
6
1
5
2
2
3
4
2
1
2
3
1
2
2
2
2
40
BAT
5
2
2
4
3
2
2
2
2
3
1
2
2
1
2
1
36
GAMUT
5
2
5
2
2
5
1
2
2
1
3
1
1
2
3
1
38
HY- MLAM HLAM-
SPLIT FINE COARSE ADOM ADPIC
555 42
122 2
523 21
2 1
22 2
352 23
223 2
221 2
1
222
1 1
132 21
111
2
1
1
3
1
1
2
2
2
3
2
1
35 27 26 16 19
a. See Table 8 for descriptions of the types of meteorological scenarios.
b. SLL = Single-Layer Lagrangian; MLL = Multiple-Layer Lagrangian; MLE = Multiple-Layer Eulerian.
45
-------
TABLE 8. THE THREE TYPES OF METEOROLOGICAL SCENARIOS FOR WHICH MODEL
PERFORMANCE WAS ASSESSED
Type of
met.
scenario
I
II
III
Meteorological scenario
with respect to actual
footprints
Actual footprint is in or
near surface pressure ridge
and is not intercepted by a
cyclone or front
Actual footprint is inter-
cepted by a cyclone or front
only after 48 h of transport
Actual footprint is inter-
cepted by a cyclone or front
within the first 48 h of
transport
First half:
Second half:
First half:
Second half:
First half:
Second half:
PTCH
release
numbers
1 o a
1, Z, »
17, 18, 20, 21,
22, 26, 28, 31
3, 5, 9, 13,
14, 15, 16
23, 29, 32
4, 6, 7,10,
11, 12
19, 24, 25, 27,
30,33
PDCH
release
numbers
1, 3, 13, 14
17, 18, 20, 23,
26, 28, 29
5, 7, 8, 9
11
21,30
2, 4, 6, 10,
12, 15, 16
19, 22, 24, 25,
27, 31, 32, 33
TABLE 9. NUMBER OF FOOTPRINT-DAYS INTERCEPTED (/ ) AND NOT INTERCEPTED (NI)
BY CYCLONES AND/OR FRONTS AFTER 10-h OF TRANSPORT
PDCH
Model3
SLL:
MLL:
MLE:
SRL
TCAL
VCAL
ARL
BAT
GAMUT
HY- SPLIT
MLAM-FINE
MLAM- COARSE
ADOM
ADPIC
I
11
13
16
15
13
15
11
11
7
5
8
NI
6
16
18
19
16
16
18
9
14
7
5
PTCH
I
8
17
13
19
15
13
15
9
12
6
4
7/1
2
11
7
11
7
8
4
7
5
4
2
Total
I, %
70
53
54
53
55
54
54
56
50
50
63
SLL = Single-Layer Lagrangian; MLL = Multiple-Layer Lagrangian; MLE = Multiple-Layer Eulerian
46
-------
were related to meteorological scenarios only for transport times beyond 10 h, since initial performances
were often dissimilar to performances beyond this time. Daily surface weather maps and tracer
concentration maps were superimposed to determine interception occurrences. Secondly, the footprint-days
were identified for which transport speed errors and centroid location errors were least and greatest.
Finally, the number of / and .NI cases were compared for the least and greatest errors to establish the
presence or absence of a relationship for all but two models (Table 10). SRL performances and
meteorological scenarios were not related for SRL, since an inordinate percentage (70%) of its footprint-
days were intercepted by cyclones and/or fronts, and ADPIC, since there were only 19 footprint-days while
a high percentage of these (63%) were intercepted by cyclones and/or fronts. For the other nine models,
the number of / cases was comparable to the number of NI cases (50% to 56% of the cases were /
cases).
For four of the models (TCAL, ARL, BAT, and HY-SPLIT), the greatest errors in transport speed
and centroid location tended to correspond to cyclonic and/or frontal interceptions. The same is true for
VCAL and ADOM, but for these two models, the least errors also tended to correspond to cyclonic and/or
frontal interceptions. Meanwhile, the least MLAM-FTNE errors tended to correspond to cyclonic and/or
frontal interceptions.
The model performance assessments discussed in this section rely primarily on three plots. The first
plot compares mean separation errors (i.e., the mean distance between the centroids of the measured and
calculated footprints) as a function of transport time. The second plot compares model and actual
transport speeds by displaying for each case the ratio of calculated-to-measured transport speeds
(D c I D m ) as a function of transport time. Finally, the last plot displays the model error in footprint
centroid locations via the ratio of separation distance over measured transport distance (Z>e / D ) also as
a function of transport time.
A word of caution when interpreting the transport speed comparison plots: model errors are not
necessarily greatest for the lowest ratios. It is possible for the calculated centroids to be very close to the
measured centroids, yet at the same time, the transport speed ratios are <0.5 or >2.0. In some cases when
both the model and actual wind speeds are low, the ratios can be very small. For these cases, one must
also view the corresponding cases on the plots of footprint centroid separation.
Since each model has a unique set of discrete footprints, model performances can be intercompared
only if the sample sizes are large enough to reveal a true assessment of model performance. That is, a
comparison of the performances of two models may not always be appropriate, since the two models could
perform equally well for the same set of footprints (if one were available).
47
-------
TABLE 10. RELATIONSHIP BETWEEN MODEL PERFORMANCE
AND METEOROLOGICAL SCENARIO
Number of footprints
Transport speed errors
Least Greatest
Model3
SLL:
MLL:
MLE:
a. SLL =
b. I =
SRL
TCAL°
VCAL
ARLC
BAT0
GAMUT
HY- SPLIT0
MLAM-FINEd
MLAM- COARSE
ADOM
ADPIC
Ib
3
9
5
8
5
8
7
9
3
Single-Layer Lagrangian; MLL
footprint intercepted 1
by a cycloa
NIb I NI
4
6
8
6
5
5
3
5
2
= Multiple-Layer
7
4
9
7
5
8
2
2
6
-
Lagrangian;
5
1
3
1
4
2
2
1
2
-
MLE =
Centroid location errors
Least
I
6
10
5
7
7
6
10
9
6
NI
6
4
5
7
5
7
4
7
3
Greatest
I
7
3
7
5
2
8
0
2
4
NI
2
2
4
2
2
2
0
0
1
Multiple-Layer Eulerian.
e and/or front since tracer release;
NI = footprint not intercepted by a cyclone and/or front since tracer release.
c. Greatest errors appeared to be related to the occurrences of cyclonic and/or frontal interceptions.
d. Least errors appeared to be related to the occurrences of cyclonic and/or frontal interceptions.
48
-------
52 MEAN SEPARATION ERROR
For each of the 11 models, the mean distance separating Cc and Cm for the combined set of PTCH
and PDCH footprints was calculated as a function of transport time. Ideally, for a model intercomparison,
the set of footprints for each model should be identical so that the performance of each model is assessed
with a standard data set. However, due to the inconsistencies of the sets of footprints between models
(Table 11), a standard set would not be useful: the number of footprints common to each model would be
too small to draw statistically meaningful conclusions about model performance intercomparisons.
There are enough footprints within the first 61.5 h of transport (32-60 for the Lagrangian models and
27 for the Eulerian models) to characterize the mean model error (Figure 15), while minimizing the
possibility that differences between model performances were due to the differences in footprint sets. The
number of cases for each model expressed as a function of transport time (Table 12), can be used as a
qualitative confidence factor when interpreting Figure 15; that is, the greater the number of cases, the
greater the probability that the results presented in Figure 15 are representative. A precautionary note:
the ADOM mean error decreases with increasing transport time from 37.5 h to 61.5 h, probably reflects the
small sample size at 61.5 h (i.e., 6) and not the true behavior of the model.
There appears to be two clusters of model behavior in Figure 15. The first cluster of models—BAT,
MLAM-COARSE, MLAM-FTNE, and VCAL-show relatively low mean errors that gradually increase with
increasing transport time within the first 85.5 h of transport. These mean errors range from 100 km to
400 km. The second cluster of models-ADOM, ADPIC, ARL, GAMUT, HY-SPLIT, SRL, and
TCAL—show higher mean errors than the other four models. These mean errors also increase at a higher
rate as a function of transport time, from 180 km to 380 km after 13.5 h of transport to 750 km to 1,080 km
after 85.5 h of transport.
49
-------
TABLE 11. PERCENT OF MODEL A FOOTPRINT-DAYS COMMON TO BOTH
MODELS A AND Ba
Model
B
SRL
TCAL
VCAL
ARL
BAT
GAMUT
HY-S
MLAM-F
MLAM-C
ADOM
ADPIC
SRL
34
41
32
44
38
45
45
45
57
56
SLLb
TCAL
64
77
79
79
74
73
82
82
89
74
VCAL ARL
67
72
--
68
87
74
68
91
93
89
78
64
82
77
—
79
85
79
80
91
96
78
BAT
75
71
83
67
72
82
91
95
86
81
Model A
MLL
GAMUT HY- SPLIT
64
66
70
72
72
73
82
82
93
70
67
60
63
61
74
67
70
72
79
78
MLE
MLAM-
FINE
56
53
63
49
66
59
55
75
86
63
MLAM-
COARSE ADOM
56
53
64
56
69
59
61
75
75
59
44
37
39
38
39
43
39
55
48
- _
41
ADPIC
42
29
33
33
36
31
38
39
36
39
a. Range = 29% - 96%; mean = 64%.
b. SLL = Single-Layer Lagrangian; MLL = Multiple-Layer Lagrangian; MLE = Multiple-Layer Eulerian.
TABLE 12. NUMBER OF CALCULATED AND MEASURED FOOTPRINT PAIRS
FOR EACH MODEL AS A FUNCTION OF TRANSPORT TIME3
Total
Model number 13 . 5
SLL:b
MLL:
MLE:
SRL
TCAL
VCAL
ARL
BAT
GAMUT
HY- SPLIT
MLAM-FINE
MLAM- COARSE
ADOM
ADPIC
36
68
64
72
61
61
56
44
44
28
27
a. Number of pairs for 6 h preceding and
b. SLL
= Single-Layer Lagrangian;
MLL
12
14
14
14
12
13
13
10
8
9
11
Transport time , h
37.5
14
23
26
26
25
24
19
16
17
12
9
61.5 85
6
20
14
20
16
15
14
12
15
6
7
.5 109.5
2
6
6
6
6
5
5
4
3
1
1
3
4
5
2
4
4
2
1
133.5
1
2
1
1
following designated transport times.
= Multiple-Layer
Lagrangian; MLE
= Multiple-Layer
Eulerian.
50
-------
MEAN CENTROID LOCATION ERROR
PTCH and PDCH
1200
2 600 -|
UJ
13.5
37.5 61.5
Transport time, h
85.5
Note: The number of footprint pairs decreases sharply beyond 61-5 h (see Table 12);
plot includes those errors with at least 5 cases.
Figure 15. Comparison of the mean separation errors of PTCH and PDCH footprint centroids (D e),
as a function of transport time.
Two of the three single-layer models--SRL and TCAL-showed the greatest mean errors. However,
the mean errors of the third single-layer model-VCAL-were considerably less. For example, the VCAL
mean error after 85.5 h of transport approximated the mean errors of SRL and TCAL after only 13.5 h of
transport. Since the TCAL and VCAL codes are very similar to each other, the magnitude of these
differences was unexpected. These differences can be attributed to two possible causes. First, VCAL uses
a varying mixing depth based on potential temperature profiles, while TCAL uses a fixed mixing depth of
1,500 m. The VCAL approach seems more reasonable, especially since the TCAL mixing depth of 1,500 m
is very high for winter applications. Holzworth (1972) empirically determined that typical midday winter
mixing depths across the central United States range from about 500 m in January to 900 m in March. The
VCAL approach would more likely reflect these representative values. Secondly, the differences in the
results may be attributed, in part, to the dissimilarity of the two samples of footprints. However, as shown
in Table 7, the two sets are quite similar with respect to the release numbers.
51
-------
The ARL and HY-SPLIT sibling models produce similar results, despite the HY-SPLIT model's use
of the higher resolution Nested Grid Model (NGM) meteorological data. One possible explanation for this
is the rather turbulent K2 profiles used by HY-SPLIT, which could exaggerate vertical diffusion (Draxler,
personal communication, 1989). Likewise, the MLAM-COARSE and MLAM-FINE family produced
similar results, indicating that the higher frequency of puff mass redistribution in the vertical (and hence,
additional computational time) did not improve the model performance.
The average trajectory errors of the two MLE models--ADOM and ADPIC-- resembled those of the
MLL models at 13.5 h, but deviated significantly (i.e., increased sharply) from them either at 37.5 h
(ADOM) or 61.5 h (ADPIC). Since the number of footprint pairs exceeded six for both models at these
times, the sudden deviation did not appear to be caused by a very small sample size. However, more data
are required to substantiate this. Based on the results illustrated in Figure 15 and discussed in Sections 4
and 5.3, the MLE models appeared to excel in replicating the frequency distribution of the ensemble
concentrations, but did not perform as well as the MLL in trajectory accuracy. This seems to indicate that,
in general, the MLE simulated the diffusion process better than the transport process.
Compared to the range of model trajectory errors during CAPTEX, as reported in the literature
(Kahl and Samson 1988; Haagenson et al. 1987; Draxler 1987; and Godowitch 1989), the range of model
trajectory errors during ANATEX was not only greater, but the magnitude of the errors tended to be
greater also (Figure 16). The simpler flow patterns of CAPTEX were a definite factor.
53 FOOTPRINT TRANSPORT SPEEDS AND CENTROID LOCATIONS
The model performances based on footprint transport speeds and locations are discussed in two
formats. First, in Section 5.3.1 the overall performances are summarized for the combined sets of PTCH
and PDCH tracer footprints. Secondly, the performances for individual footprints are discussed separately
for each model in Section 5.3.2 through Section 5.3.4. These sections also compare the performances of the
three model genres: the single-layer Lagrangian (SLL), the multiple-layer Lagrangian (MLL), and the
multiple-layer Eulerian (MLE) models, as well as compare the performances of sibling models.
52
-------
MODEL TRAJECTORY ERRORS
1000-
0 13.5
25.5 37.5 49.5
Transport time, h
61.5 73.5 85.5
Note: The single least and greatest mean ANATEX errors are ignored for each transport time interval.
Figure 16. The comparison of trajectory error ranges determined from ANATEX
and CAPTEX model evaluations.
53
-------
53.1 Summary of Model Performances
Regarding transport speeds, the MLL models as a group performed slightly better than the SLL and
the MLE models (Table 13 and Figure 16a-d). With one exception (MLAM-FINE), the transport speeds
of approximately half of the MLL footprints were within ±20% of the speeds of the measured footprints.
In contrast, the transport speeds of only one-third of the footprints for all but one of the SLL models
(VCAL) and both MLE models were within ±20% of the measured speeds. Likewise, the MLL models
tended more often to calculate transport speeds within ±50% of the measured speeds (64% to 89% for the
MLL models versus 61% to 78% and 68% to 70% for the SLL and MLE models, respectively).
Although the range in the MLL model performances in calculating footprint centroid locations
increased over that for transport speeds, the MLL models again tended to perform slightly better than the
others. With increasing transport time, no consistent trend in the performances appeared. That is,
performances did not systematically improve or worsen as a function of transport time. However, the
calculated centroids generally were within ±20% x Z>m more frequently for T < 24 h than for T > 48 h.
The following comments highlight the results and conclusions of Section 5.3:
(1) The model performances in calculating the location of the centroids paralleled those for the
transport speeds (Table 13 and Figures 17a-d).
(2) Compared to the range of model trajectory errors during CAPTEX, as reported in the
literature (Kahl and Samson 1988; Haagenson et al. 1987; Draxler 1987; and Godowitch 1989),
the range of model trajectory errors during ANATEX was not only greater, but the magnitude
of the errors tended to be greater also (Figure 16). The simpler flow patterns of CAPTEX
were a definite factor.
(3) The MLL models as a group tended to perform as well or better than the SLL and MLE
models (Figure 15).
(4) The performances were worst for two of the three SLL models-SELL and TCAL; both use
simple approximations to the transport wind (SRL uses surface pressure fields to calculate
geostrophic wind vectors, while the TCAL model calculates mean-wind vectors for a constant
mixing depth of 1,500 m).
(5) The VCAL performance paralleled those of many of the MLL models, while the MLAM-FINE
performance paralleled those of SRL and TCAL--SLL models.
54
-------
TABLE 13. A SUMMARY OF MODEL PERFORMANCES IN CALCULATING
TRANSPORT SPEEDS AND FOOTPRINT CENTROID LOCATIONS
A. Percentage of N footprint-days when calculated transport speed was within ±20% and ±50%
of the measured transport speed
tf=
±20%:
±50%:
SRL
36
36
61
SLLa
TCAL
68
35
62
VCAL ARL
64
55
78
72
47
75
BAT
61
57
80
GAMUT
61
46
72
MLL
HY- SPLIT
56
46
68
MLE
MLAM-
FINE
44b
35
82
MLAM-
COARSE ADOM
44
59
89
28b
32
68
ADPIC
27
30
70
B. Percentage of N footprint-days when calculated centroids were within ±20% x Dm and ±50% x Dm
of measured centroids as a function of transport time (T )
For
N=
±20%
±50%
For
N=
±20%
±50%
For
N=
±20%
±50%
SRL
T < 24
12
: 25
: 42
24 h <
14
: 21
43
I > 48
10
: 0
: 30
SLLa
TCAL
h,
14
14
21
T < 48
23
17
43
h,
31
13
52
VCAL ARL
14
50
57
h,
26
35
58
24
21
79
14
7
64
26
31
77
32
25
53
BAT
12
50
67
25
32
52
24
42
88
GAMUT
13
54
69
24
17
54
24
21
58
MLL
HY- SPLIT
13
54
77
19
32
42
24
13
58
MLE
MLAM-
FINE
10"
30
50
16b
19
81
18b
39
89
MLAM-
COARSE ADOM
8
50
88
17
41
88
19
42
95
9"
44
56
12b
0
42
7"
14
71
ADPIC
11
27
64
9
22
44
7
0
71
a. SLL = Single-Layer Lagrangian; MLL = Multiple-Layer Lagrangian; MLE = Multiple-Layer Eulerian.
b. Through February 16, 1987, only.
55
-------
,D
S* 3
' H§
1SB'
I
II
,0
ffl < K W •*
•a
a>
a
2
-
r g.
*•> etf
U £
2
-
O et>
""> ^~.
•o 3
rt 73
-------
(6) Although the TCAL and VCAL model codes are nearly identical, the varying mixing depth in
VCAL appeared to yield a better transport simulation. Since many of the TCAL and VCAL
footprint-days correspond to the same days (Table 7), this conclusion is likely valid.
(7) Although MLAM-FINE and MLAM-COARSE are virtually identical models, the
more-frequent mass redistribution in MLAM-FINE apparently afforded no benefit; in fact,
consistently the performance was degraded. Although the MLAM-FINE footprint set was
limited to the first half of the ANATEX period (unlike the MLAM-COARSE footprint set),
few of the MLAM-COARSE footprints pertain to the second half~6 out of a total of 44
MLAM-COARSE footprints (Table 7); therefore, this conclusion also is likely valid.
(8) With few exceptions, the ARL and HY-SPLIT model performances were similar; use of the
high-resolution NGM meteorological data in HY-SPLIT did not appear to substantially improve
its performance.
The remainder of Section 5.3 discusses in more detail the performances of each of the 11 models.
The discussions are ordered by model genre.
57
-------
53.2 Single-Layer Lagrangian Models
SRL--
Approximately one-third (36%) of the 36 SRL footprint-days were within ±20% of the measured
transport speed (Figures 18a and 18b). For seven footprint-days, the SRL transport speeds were more than
double the measured speeds. After 48 h of transport, only 3 of 10 footprint-days were within ±50% D m of
the measured centroids (Figures 18c and 18d). SRL tended to overestimate transport speeds and place
centroids to the right of the measured centroids.
There were relatively few SRL footprint-days when PTCH and PDCH footprints could be related to
a single release. Because this model diluted the surface concentrations much more slowly than the
measurements indicated, many of the SRL footprints blended with others. For example, after 150 h of
transport, the SRL surface concentrations for PTCH-3 (the third release of PTCH tracer) exceeded 100
dfL/L (compared to measurements <10 dfL/L).
For two of the three PDCH footprint-days when SRL significantly underpredicted the transport
speeds (PDCH-20, -27), low wind speeds and reversals in wind directions (complicated wind patterns)
occurred. In contrast, winds for PDCH-8 had straight-line flow at a moderate speed of nearly 20 km/h.
SRL overpredicted transport speeds by at least a factor of 2 for PDCH-6 and PDCH-24, both of which
were intercepted by cyclones and/or fronts.
For the footprint-days with the largest separation distances beyond 40 h (i.e., PTCH-3 and
PDCH-15, -24, -27), cyclonic and/or frontal interceptions occurred. The transport speed was overestimated
for each of these footprint-days.
58
-------
CO
Q
LU
LU
CL
CO
CC
o
CL
CO
z
cc
u_
0 ;
z
o
CO
cc <
CL
^^
^J
f *\ 1
^J
i
_i O
en Q
co a.
2
j ^r
tA i
« <
^" ,J- CO (y
1 Z °» '
1 1 1 1 1 1 1 1 1
^_
,-
« 8
*~ OJ
1 r i ~i
- *r
.8 CO
cc
L O
cc
-8 CC
LU
-8 Q
J2 i
- 1 §
-§ I 3
- ^ Q
Q
"? CC
I—
^
LU
-CM Q
- **—**.
•^^ ^
•^^.
-0
_i O
CO CL
,-
s
10 r- in
i- N *~ CJ
5 S « w
a
^j w eoj
f 1 1
£
v
to ^r
^
! i
i i i
*-
_
-8
-1 1
"S
*"
u
-g g
"111
las?
£ C "
• *^ 4>
Q.
g *-• 4>
"9 §5
**"* u
~ ^i ^i
OH •-
r •~*> CL
i .0 S1
•^^ o
° "Q <2
f\ (r>CO(OVCMC\JtO''J-CMO ^ W OJ T- 0 -T- "^
o
o
I
-i O
CC H
co CL
o r»>
m
r- O
§O L-
a o
a <~
S c
*- o
h§ Si '^
S on _o
I .S T3
-s | e '§
^ t Q
W o
-° oo
~u
^
-a .2?
UH
co
rv w oj
LU 0
a/ a
LU 9
a/ a
59
-------
TCAL (Sibling of VCAL)-
Only 24 of the 68 TCAL footprint-days (35%) were within ±20% of the measured transport speed
(Figures 19a and 19b). The transport speeds for the majority of footprint-days (62%) were overestimated;
40% of these were overestimated by more than a factor of 2. After 48 h of transport, 17 out of 30
footprint-days (57%) were within ±50% Z>m of the measured centroids (Figures 19c and 19d). The
greatest centroid location errors tended to occur for cyclonic and/or frontal interceptions (Table 10).
The model particularly showed problems with PDCH-4, -5, -6, -7, -14 and -15, where all but
PDCH-5 and -14 were affected by cyclonic and/or frontal interceptions. In the cases of PDCH-5 and-14,
the model substantially underestimated and overestimated, respectively, the transport speed during the first
24 h, when the actual boundary layer winds were very light.
For the footprint-days when the centroid separation distance was great, TCAL tended to locate its
centroids to the right of the measured centroids. Because TCAL is a single-layer model that computes a
layer-average transport vector, it may have overemphasized the higher-level winds for these footprint-days.
60
-------
CO
Q
LU
LU
DL
CO
DC
O
Q.
CO
Z
^
\- !
LJL
O :
Z
O <
CO
DC !
H -
O .
0 '
JD *
— '
CO
Q
LU
LU
Q.
CO
DC
0
Q_
CO
~
DC
U.
0
~p
0
CO
DC
^c
Q_
IsE
O
O
IT? «
-2- °
. -r
1 JT
O Q
h- Q-
o>
Ol£ -
or- »-
10 *
01
«t 0>j
^
1 1 1 I 1 1 1 1 1
>*C\IOIIOVN
LU 0
a/ Q
< 0
O h-
o
. *.
a
6
u» ^
0
, ^
r~
CO CO
1 1 1
\j in T- w <
^ d
LU
C
a>
T- &
T- in r-
88
•- !
8 »
i •
1 i i i
> y w n
a
/ a
8
8
CM
8 P
8
a
*£ 2
*
i i i
3 to T- in c
0 ' »p
9
/ 0
~ V
-.
K§
«
-8 «
u«
~ ^
jo v
~8 o g
~ <-«
~8 II. 1
W U< crt
C O. «
- H °
53 u.
^ **"
-8 ^|
r ^ o-
T3 ^
G o
0 CO <*H
r EC ic
uu
H^ d
^^
o o-O
* <2 G
4j K
-| g- H
IH *^^\
o o
Q. v~x
-o CO
*^ C
i < -S
- S o QJ w
- g c S2 '§
1 § I
-5 ^ cK
0)
IM
g
-8 -SP
PU
_
s)
61
-------
VCAL (Sibling of TCAL)-
Slightly more than half (55%) of the 64 VCAL footprint-days were within ±20% of the measured
transport speed (Figures 20a and 20b). Like TCAL, VCAL showed a tendency to overestimate transport
speeds. However, there were only 6 footprint-days when the overestimates exceeded a factor of 2,
compared to 24 for TCAL. After 48 h of transport, 19 out of 24 footprint-days (79%) were within ±50%
Dm of the measured centroids (Figures 20c and 20d).
The model performed particularly well for PTCH-1, -3, -5, -11, -13, arid -15 and for PDCH-1, -8, -9,
and -13; these represent a mixture of meteorological scenarios. Beyond 80 h of transport, the VCAL
centroids for these footprints were well within ±50% D of the measured centroids. VCAL tended to
place its PTCH centroids to the right of the measured centroids. However, such a tendency was not
apparent for PDCH.
Compared to the 47 TCAL footprint-days common to the set of VCAL footprint-days, the VCAL
transport speeds and centroids in nearly all cases were as close or closer to the measured transport speeds
and centroids than those of TCAL (Figures 21a - 21d). The VCAL centroids were slightly farther from the
measured centroids for only three footprint days.
Like TCAL, the greatest VCAL errors tended to correspond to cyclonic and/or frontal interceptions.
In contrast, however, the cases when VCAL errors were least tended also to correspond to cyclonic and/or
frontal interceptions, suggesting that VCAL is better suited than TCAL to simulate transport during the
more complicated flow patterns. In general, the single-layer wind vector computed by VCAL to represent
the flow within its varying mixed depth appeared to be more representative of the actual flow than the
single-layer wind vector computed by TCAL to represent the flow in the fixed 300-1,500 m layer.
62
-------
CO
T SPEED
EC
O
a.
CO
^^
£E
LJL
0
Z
o
CO
DC :
£
^>
o ,
0 '
~2 "
< o
> CL
O)
I- fl
u>
0) £
(0
a - s
2 33
-
a><0*N«OD ^
_j i
> Q.
CT
CO
CO)
!£ o>
(D CO
,,,
^ » «
i 01]
I 1 I
r « « *- c
-
^0)
2 -
S
_« K w ^
S ^
0
1 1 [
i~ CM n *
*~
-a
e/j
-8 |
-
CO 4)
-So §
-C _ J3
• HI
- 1 1 1
co u Jo
C Q, CO
2 Q T)
_o *" O «>
^^ JZ^
U o
rS S £2
C!
* 'a
•o S
W M
s^^*^ r \ ^^
U (J
LU Q LU a f— ' P\
a/ a a/ a a, £
9 9 S TI
CO
Q
LU
LU
Q.
CO
|—
or
O
Q_
CO
^»
t
r~ n
= "'
O
^ t
n
*
O
p.-*
*
;
st in i- m c
i^
^
01
"*
1 I 1
> 10 -r- m c
<*- G
CO «
CU X
- £i M £!
-*-• ^^
O o
ex ^*^
CO ._
- § c O
^- rt c*-
u
4^ (/]
" co ^ §
^ "^ "Z3
- O Pj f) eg
^ > 0
ar _, — •
-i -s ."S
" * o o*3
Q. U C
CO V" CD
c py o
-° o
*«
LN
-° j|
- o
•J
^^ ' ^~ ° 9 ' T '
LU 3 LU 8
a/ a a/ a
63
-------
(This page is intentionally left blank.)
64
-------
Q
LU
LU
Q_
CO
DC
O
Q.
CO
Z
I :
LL.
o
z
O <
CO
DC J
Q.
O i
0 '
S 5
> Q
P » s
o> >,- >
w/ T- a
o> U?~ T-
« S 1
s ^
i CD S 0,0 S
: "> " N J
>
*~
! § i*s
! 1 1 1 1 1 1 I 1
g>
i
I?
S*
1 1 1 1
- ^
100 120
rs
>N ERRORS
LS 1 Q
I <
8 ! 3
CO
§ Q
? H O
DC i
J-
8 LU
o
^_^ i
T3
_i _
*^ £
H
s 1 s s
« s
»>l
! • > »?
,- >
««sj
r o « ^ c
»
1 -
1
^ ^ > L
a •< « ,
p^
1 i
i i i
-8
:, |j
3" o
**~s *^
I -a a
-s £ s §
' 1" B 1
-s i fc 5
S" "^ .2
co v~"' Q.
1 >- « -4-*
•O !-H
(oTeMCM^CMO ^"^ ^
C J wrMrwoii-T-'»-*-aooo r* rP
LU 8 •-> rj
UJ 0 Q/ Q £ H
a/ a o OH
£ s
CO
Q
LU
LU
Q_
CO
1-
DC
O
a.
CO
z
^^
cc
1 1
U_
0
z
0
CO
DC
Q.
O
n
rf i
> H
_J ^"
^
1—
" § £
0 g
- 5 n 5 ^
<* |^
"
-fc^
S
I? a
: •> S
^
»
- ^
o
- CM
"
-8
e
Q 7^
C O
8l
•o o >-
C "" u
* ."S 8
d 2 2
l^-§
II ¥1
s g o t:
(N
CD
CM CM CO (O
LU 3
a/ a
a/ a
65
-------
533 Multiple-Layer Lagrangian Models
ARL--
Approximately half (47%) of the 72 ARL footprint-days were within ±20% of the measured
transport speed (Figures 22a and 22b). When the error exceeded ±20%, the model tended to overpredict-
24 overpredicted footprint-days versus 14 underpredicted footprint-days. After 48 h of transport, 44% of
the footprint-days were within ±50% D m of the measured centroids (Figures 22c and 22d).
ARL was particularly successful with the footprint-days involving PDCH-1 and -4; both sets were
characterized by minimal wind direction changes with time. For each case, ARL's transport speeds and
centroid locations were very similar to those of the measured footprints, even after 100 h of transport.
The greatest errors in transport speed and centroid location tended to correspond to footprint-days
with cyclonic and/or frontal interceptions (Table 10), which introduce more complicated wind patterns and
vertical shears. During these meteorological scenarios, it is possible that ARL either relied too heavily on
the greater wind speeds at higher levels or did not replicate the looping of the actual trajectories (and
thereby, transporting the tracer farther downwind).
For the PTCH footprint-days of overpredicted transport speeds, the ARL footprint centroids tended
to be to the right of the measured centroids. Thus, higher-level wind speeds could have been
overemphasized. However, for the PDCH footprint-days of overpredicted transport speeds, the opposite
tended to be true. The cause of the difference warrants further investigation by the modeler.
The model underpredicted the transport speeds for PTCH-26 by 60% to 80% and, as a result, erred
in locating the centroids by at least 60% D . ARL PTCH-26 centroids were always to the left of the
measured centroids. Since cyclonic and frontal interceptions were not factors, it is possible that ARL
overemphasized the lower-level winds for this case.
66
-------
CO
Q
LLJ
LLJ
Q.
CO
DC
O
CL
CO
1
h-
o
z
o
CO
cc
U5
O s
o s
,— V
_j O
CC Q
< CL
j *
« 5
III!
- ^
-S CO
cc
o
-1 cc
LLJ
e ?
-8 g O
.c | —
»- <;
1 0
-8 ? O
Q. __l
- 1 Q
-? ^ o
DC
1-
-8 §
L^
g '
3 ^
X
< £
0) 12 »•
*-
* N ?
S »
r m ™ ^ t
.
*
* 0)
-
»
3?.
t m f- 2
» * ?
r-
-
-8
v5
8C3
qj
"s
§ 8
-So «
| -0 T3
"8 | J3 1
(» e •£
" | 1-8
r— 4_»
<2 *
53 .0
u ^
-S Q £2
o^.g
^ o
^ r «,,«,
CD
8
8
IO
o
I
— i O
CC (—
< L
0
*"
tn co m
T- T- T- CM
O
P a ,
ix m
to o
j w> T- m c
S
S
^
t- in
CM no
CM O (M
<*- T- CM
^
i in y in c
^T3
o S
^ !rl
-S "S u
<3U UH
Q, ft^
eo ^
!„» ^™*^
tn 0
-8 0 ---
a >-
C/3 O
C <•->
2 2 g
o 2 *t o
"g 0 J •£
rtf j O
40 60
Transport time
22. Errors in A
centroid loi
a>
0 £j
.2?
UH
- o
4
"aT " - - - - - ~ -<
* LU 0 LU 3
ol a at a
67
-------
BAT-
More than half (57%) of the 61 BAT footprint-days were within ±20% of the measured transport
speed (Figures 23a and 23b). When the error exceeded ±20%, BAT (like ARL) tended to overpredict-
19 overpredicted footprint-days versus 7 underpredicted footprint-days. After 48 h of transport, all
12 PTCH cases and all but one PDCH case were within ±50% £>m of the measured centroids (Figures 23c
and 23d). After 40 h, BAT exhibited no tendency to place its centroids to the left or right of the measured
centroids; however, for the greatest errors, BAT tended to place its centroids to the right of the measured
centroids. The greatest BAT errors in both transport speeds and centroid locations virtually always
correspond to cyclonic and/or frontal interceptions (Table 10).
This model performed much better for PTCH than PDCH. For only one PTCH case was the
calculated transport speed in error by more than ±40%, compared to 12 PDCH footprint-days. One may
think that this performance difference is related to the fact that the wind patterns tend to be simpler for
PTCH releases that are transported to the sampling domain. However, all but the PTCH-1 and -8
footprint-days had cyclonic and/or frontal interceptions. Therefore, the better performance could not be
attributed to simpler wind patterns.
BAT appeared to have problems with PDCH-6, -10, -14, and -15 in the initial stages of transport.
BAT transport speeds during the first 32 h were at least double the measured speeds; directional bias was
inconsistent. All of these footprint-days but PDCH-10 were affected by cyclonic and/or frontal
interceptions during the first 48 h. In contrast, even beyond 100 h, BAT performed well for PDCH-1-the
case with minimal wind direction changes.
68
-------
CO
Q
LU
LU
Q_
CO
1 —
CC
O
CL
CO
1
LJL
0
Z
O
CO
cc -
Q_
o ,
o
_— .
.,.
1- O
;§£
U> ,_ CO
«> s<^
1 1 1
*-
S
2
«
«t a
*8 '°
^ <
*~
i i i
— ^
-8
tn
8ra
tp
c/> tu
- f -si
^8I 15
l~ §" "*
«2 "^
X °
-S Q §
o- .S
^a
^ 0
o "O O
c <*-
""^r1 ncOtDTCJCUffltD^CVJ-i-cOCO^-INO V <•) CM T- O -7 OJ « ^ 53 .^
E U
LU 0 LU 8 OQ
a/ a a/ a £ £
CO
Q
LU
LU
CL
CO
1
CC
0
a.
CO
z
^^
cc
U_
o
Z
O
CO
DC
£
^>
O
o
Hg
m t
^
0
(M t
(
* '
i i i i i i
'
*- Ul
W ^10
?"
1 1 1
o
"" CO
cc
h o
-8 DC
CC
LU
2 z:
-s | o
Q) 1^^
:'t 2
en — J
jo Q
l ^^
9 O
CC
h-
z
8 LU
O
"Q"
- O
,_5
CD Q-
*
»
t~
ttt « n *
i
t
<
i
1 1 1
o
«"
i--
» ^ (V 0
, *
l I l
r|
t<-1
- S "O fi
S H
Q. OH
8 ^" ^
0. g
<« rt oo
3 *J *^
Q)" *JJ O
- | CQ o
"s ! el
- 1 §£
H Wo
(N
Oi
-8 3
00
E
- o
TtOiO (y y) ,- y> Q j/j ,_ t/> f\j
X-^. WOJ(NT-T-»-i-0000 ,- O O ' T- '
CO
' LU 0 UJ 8
a/ a a/ a
69
-------
GAMUT-
Approximately half (46%) of the 61 GAMUT footprint-days were within ±20% of the measured
transport speed (Figures 24a and 24b); there was no apparent tendency to under- or overestimate transport
speeds (29 underestimates and 29 overestimates). After 48 h of transport, 58% of the footprint-days were
within ±50% £>m of the measured centroids (Figures 24c and 24d). The GAMUT centroids showed a
strong tendency to be to the left of the measured PTCH centroids, but showed no tendency for the PDCH
centroids; the high frequency of PTCH footprints affected by cyclones and/or fronts (Table 9) could have
been a factor. The least and greatest errors did not correspond to cyclonic and/or frontal interceptions.
GAMUT transport speeds and centroid locations compared favorably with the measurements for the
PDCH-1, -5 and -8 footprint-days, even out to 100 h. For the latter two sets of footprint-days, cyclonic
and/or frontal interceptions occurred after 48 h of transport. The model performed very well relative to
the other models for PTCH-4, which was released at the time of a frontal passage; transport speeds
differed by less than ±10% between 10 and 110 h of transport. After 100 h, the calculated centroid was
within 30% D m. However, for the case of PTCH-25, released 24 h prior to a frontal passage, the model
did not perform as well.
There were five footprint-days beyond 12 h of transport when the GAMUT transport speeds were at
least double the measured speeds. For four of these, cyclonic and/or frontal interceptions occurred during
the first 48 h of transport. The two outlier footprint-days of PTCH-15 (released 24 h prior to a frontal
passage) apparently involved improper timing of a wind direction shift; however, with time, the accumulated
error decreased.
70
-------
CO
Q
LU
LU
Q.
CO
CC
0
CL
CO
Z
|
u_
0
Z
O "
CO
cc
CL
O s
O
t- i
2 ^
CD "•
T
^
V «
in
» s
! r- c> £
2 * ',
•
«p>
in
)
- 5-
8-
t«
! «
0
IA r
S<0
in ,- ;
8 «><*
* .
-
*~ i-
^.
»>
8 »
.. -
32 "•
r •
*~
100 120
releases.
2 S
-§o "2 *
"*". « **
- 1 if8
•^= a *;
-s g 'C a
S- S--2P
- S o g
2 <2 -8
-« s|
Q Ui
QL, (2
"8 ~l
"c a
W 0
- o hr* O
<*^^w (O CO <£> ^ CM
Cj e\i w CM w
o o o o
LU 0
a/ a
Ul 3
Q/ Q
-
eS
CO
Q
LU
LU
Q.
CO
CC
o
Q.
CO
cc
I-
Ll_
O
O
CO
cc
O
O
< CL
C5
-8
81
*8
8.
CO
-8
CO
oc
O
cc
cc
LU
LOCATION
Q
0
cc
\—
z
LU
O
2
,_ ^
^ O
2 h"
^f Q_
C*5
s
- ^
8!
(0
•»
8!
" ^" 8
^Y i~
»
- 1
^
- w
"
g
- o
60 80
sport time, hours
CD
-8
T i i r i i r w
"""Q £VIO u
W o
•
CJ
3
.2?
LU 0
a/ a
LU a
01 Q
71
-------
HY-SPLIT (Version of ARL)--
Approximately half (46%) of the 50 HY-SPLIT footprint-days were within ±20% of the actual
transport speed (Figures 25a and 25b). Model performance varied significantly with the tracer; that is, for
19 of the 23 PTCH footprint-days-compared to 9 of 29 PDCH footprint-days--the calculated transport
speed was in error by less than ±25%. This is further evidence that the PTCH data-by virtue of the
location of the GGW release site-may be biased toward simpler flow patterns for which the models are
more apt to perform well. Similar to ARL, approximately half (58%) of the HY-SPLIT footprint centroids
beyond 48 h of transport were within ±50% £>m of the actual centroids (Figures 25c and 25d).
HY-SPLIT tended to overpredict PDCH transport speeds and place its centroids to the left of the
actual centroids (especially PDCH-4, -10, -15, -20, and -27). However, for PTCH, this model was generally
within ±25% of actual transport speeds and within ±50% D m for centroid locations.
Determining the relationship between HY-SPLIT errors and types of meteorological scenarios was
hindered by the bias of PTCH cases towards cyclonic and/or frontal interceptions (17 with and only 4
without); PDCH cases were more balanced (15 with and 20 without). The cases when HY-SPLIT transport
speeds differed most from the actual transport speeds tended to occur when cyclones and/or fronts
intercepted the tracer (PTCH-7, -15; PDCH-4, -5, -6, -10, -15, -27). In two other cases HY-SPLIT
overpredicted for PDCH-20 and underpredicted transport speeds for PDCH-26 footprints, both within
ridges of high pressure characterized by light and variable winds. For the footprints of 8 of these 10 tracer
releases, HY-SPLIT transport speeds were greater than the actual transport speeds.
Similarly, HY-SPLIT location errors tended to be greatest for cyclonic and/or frontal interceptions
(PTCH-7, -10, -15; PDCH-4, -6, -10, -15, -27). Location errors for PDCH-18 and PDCH-20 were also
relatively large; although these cases involved no cyclonic and/or frontal interceptions, light winds prevailed.
PDCH location errors tended to be the least for cases when there were no cyclonic and/or frontal
interceptions (PDCH-1, -8, -13, -17, -21, -26). Since all but four PTCH footprint-days were affected by
cyclones and fronts, HY-SPLIT performance could not be definitively related to types of meteorological
scenarios. However, for the four footprint-days not affected by cyclones or fronts, the location errors were
the lowest (PTCH-1, -5, -13 during the first 48 h of transport). After 48 h, the location errors for PTCH-5
and PTCH-13 increased as fronts intercepted the tracer. There were other cases when HY-SPLIT location
errors were small, despite cyclonic and/or frontal interceptions (PTCH-4, -13, -30), demonstrating that the
model could perform well under such circumstances.
72
-------
CO
Q
LU
LJJ
Q_
CO
h-
DC
O
D_
CO
2
^*
DC
U_
O
2
O
CO
CE
<
0.
^
0
0
t i
% g
^
^
a>
0 to,
! ft 2 » £
•» .ig «e
i
1 ! 1 1 1 1 1 1
8
J2 «>
s
u>
s
^
CM
<0
1 I 1 1
**
-
-8 CO
DC
O
-S oc
LJJ
2 2
-s § O
•*- L__
- I S
•I o
-« S O
Q. _/
§ Q
H Q
DC
J-
•z.
o LU
O
S-
- o
t I
|g
1
a
.
<»o ;
2 » S
,.
<\
i i i
8 »
5.
S »
N i
0 ° «
0 V
1 1 1
*"
_
-8
CO
CU
"S
-
2 T3 S
S 3 C Ji
CD O rt «
^ is
ffl 42 _-•
E c U
*: 'C "£!
"8 I f |
o> o '35
i ^^
*^ ^*^ j.
Q •£
CU t.
S <~
- S ^ "^
•o e
a -C
CS Q_
K S
P £
•^^ (OffltDtCMCJeOUDirCJi-OOtfl^'CNJO ^ - O ^j
M «2
2 fc G
Q 3 ai o
rg o i •£
-*= >- «
- | a 8
-si -s:s
| 22
Sbrt fli
H Wo
^•3
JQ
-8 5
.Sf
u,
- 0
(DVcywcofOTcyi-BOio^rwo «N m •>- m o in *- in oj
^ „ « «, oooo , o 9 • v •
* ' LU 0 LU 8
al a a/ a
73
-------
Figures 26a and 26b compare the transport speed performances of HY-SPLIT and its sibling model
ARL for only those footprint-days common to both models. The numbers accompanied by an "H" refer to
the HY-SPLIT results, while those without refer to the ARL results. For the PTCH tracer, HY-SPLIT
tended to outperform ARL for 10 of the 15 footprint-days; while for the PDCH tracer, neither consistently
outperformed the other. Figures 26c and 26d show similar results for location errors in that HY-SPLIT
tended to outperform ARL for the PTCH tracer, but for the PDCH tracer, no model consistently
performed better than the other.
74
-------
CO
Q
LU
LU
D_
CO
^^^
CC
O
CL
CO
Z
cc
h-
u_
O
Z J
o s
s
OL
O J
O 1
t j;
CL o
CO Q.
1
cr
X
T
S i
1 * »
> " is!
¥ i ^
T
I -1 to ^
*"
1 1 1 J 1 1 t J 1
-
1 *z
^
WE5 5"
S 52
.5 «
^
III!
~ ^
-
-s co
cc
0
cc
-s CC
LU
£2 Z
-g g O
••= P
I §
^§ H O
& -1
§ Q
3^0
cc
\-
-^
LU
0
--v. :
S-
°
t I
5± g
co a
>-
cc
^
» S 5
5
^
X co^ ([
i ° 5S>o!
1 1 1
-
'5
^
« 5 »
^ i
t- - ;
S' 1? 5
i
^
i i i
*~
-
£ CO
- M rj £2
g S
-8 §1
T3 o
G **
-s i§ u 6
-11 H S
0)" O- O
f~ ^
~ ^ o ^ .S
a. o 1-1
0) C<_| Q,
,: |||
-' fi
o ^i, ^^
S ^
2 ^
0 .S §
r - ~* COaSCO*?c\JCMCOCO T ^ 1
I 1 1 i 1 1 1
^
in
i i i \
•-
-
-8 CO
cc
O
DC
"s DC
LU
w "7
^§ i o
•^ L_
fl) j^1
i ^
*- O
h§ ? O
% -J
" § Q
-? H O
cc
i —
i
o LU
0
-5"
- o
t i
s! £
CO CL
X
OL
"S 2^
X
"• i
CD J
1 1 1
f *
§§ 2 S
"co ^ *-
*
1 1 1
C «2
tT" V5 yi
^o o ^
o ' -5 g
HH O
-------
MLAM-FINE (Sibling of MLAM-COARSE)-
Of the 44 MLAM-FINE footprint-days in the first half of the ANATEX period, transport speeds for
only about one-third (35%) were within ±20% of the measured transport speed (Figures 27a and 27b).
However, virtually all transport speeds beyond 12 h of transport were within ±40% of the measured
transport speed. The strong tendency for the model to underpredict transport speeds was apparent.
Despite this tendency, after 48 h of transport, 16 of the 18 footprint-days (89%) were within ±50% D m of
the measured centroids (Figures 27c and 27d).
The success of this model in calculating centroid locations beyond 48 h was not attributable to a
predominance of simple wind patterns; all but PTCH-1 and -8 and PDCH-1 footprint-days involved cyclonic
and/or frontal interceptions. In fact, for the footprint-days with the least errors in transport speeds and
centroid locations, the occurrences of cyclonic and/or frontal interceptions out-numbered the
nonoccurrences by a 2:1 margin. This suggests that the model is, relative to the other models, well suited
to simulate transport for complex wind patterns.
MLAM-FINE appeared to have the most difficulty in simulating PTCH-4, which was affected by a
cyclonic and/or frontal passage as the tracer was released. It also appeared to have difficulty in calculating
the transport speeds for PTCH-5, which was released in the center of an anticyclone (a region of low wind
speeds).
76
-------
CO
Q
UJ
UJ
Q.
CO
DC
O
Q_
CO
Z
DC
U_
O
Z
o
CO
DC
Q.
o ,
0
UJ X
Z 0
u_ D
r
-
0)
«D
-
; o T CM c
i T~ \ ~i i "~i r n r "
"^T"1 cocowvcMWcowvtMT-
Cj CM CM OJ CM t-T-t~»-
•
a
in
2 «i
CM
- T
.
•8 CO
CC
O
§ DC
UJ
-.| Q
40 60
Transport time,
ROID LOCAl
i-
z
-a LU
8 Q
^^
2,
i i ] r I "
03 (O f CJ O T
O O 0 0
UJ I
o
JV £
s
in
(0
CO
CJ
«
"
r « CM r- c
-
GO
PJ ^
0)
9*
*
*_
1 \ i
3 y^ <\I
"8 8
i S8
S-J rl. %i
O ~M TO
40 60
Transport time,
>) PDCH footpri
the designated '
~, Ui
•o «*-
-a g |
r 0 o
o fc I
' -2- DC
"~" V" •
' 1 1 1
in
,
^ V
•»
"" 1 1 1
CX -o
« G
u, "*
O «JH
r§ « U
c8 g^
^ ^v
"? Z ^
E «2
:s|i|
-1^1
35 C _.
- s o '« '3
O. *^ fcn
CO O *••
c t e
i? W o
-s
U
"8 1
~ O
(o^cgo M in T- in o in t~ in m
n**"^*. ftitXICM »-»•»-*- OOOO *- O O'T-'
CC
" uj o we
a/ a a/ a
77
-------
MLAM-COARSE (Sibling of MLAM-FINE)--
Of the 44 MLAM-COARSE footprint-days in both halves of the ANATELX period, slightly more
than half (59%) were within ±20% of the measured transport speed (Figures 28a and 28b). Like MLAM-
FINE, MLAM-COARSE exhibited a strong tendency to underpredict transport speeds. Similarly, after 48
h of transport, 18 of the 19 footprint-days (95%) were within ±50% D of the measured centroids; most
footprint-days were within ±25% D m (Figures 28c and 28d). Both model versions exhibited a slight
tendency to place centroids to the right of measured centroids.
For most of the footprint-days common to both sets, there was little if any difference in model
performance (Figures 29a - 29d). Neither model version consistently performed better than the other for
the remaining footprint-days. This indicated that the more-frequent, vertical mass redistribution in MLAM-
FINE (i.e., twice daily versus once daily in MLAM-COARSE) did not improve the transport performance.
Moreover, the more-frequent, vertical mass redistribution did not appear to improve the diffusion
performance, as Table 5 indicates. That is, MLAM-FINE mean concentrations along bands of sites tended
to differ from the mean measured concentrations more so than those of MLAM-COARSE. Relative to
other sources of variation in this study, the differences between the MLAM-F1NE and MLAM-COARSE
code versions appear unimportant.
78
-------
CO
Q
111
111
Q_
CO
1-
DC
0
Q_
CO
2
DC
h-
U_
O
2
O
CO
DC
|
O ,
O
2> .
— '
CO
Q
UJ
UJ
OL
CO
DC
0
CL
co
2
^^
ft"*
l
LL
O
2
O
CO
DC
^^
OL
^>
O
o
\«^
*'^
UJ I
CO O
u> «
8*
t « «2
®OJ (0
1 1 1 1
CD {£> t CM <:
1
n m
0
r
l i i 1
CO (O ^ CM C
0000
*•
-
-s co
DC
O
-1 DC
HI
o) 2
-g i o
-I §
-s B 0
s- -1
' 5 Q
-s ^ O
DC
h-
2
HI
o
S
3 ^
?
^
-8 CO
DC
0
1 DC
HI
e 2
-8 o O
0)" V^
C ^^^
^ O
~8 0 O
Q. 1
co — '
" i Q
J~~ P^
DC
2
o LLJ
0
'o"
s"™1*'
(\
UJ I
| Q
8 t
1
.,
12 e
<°fl
atj
^^
i i i
r « cy »- <
UJ
a
UJ I
< CL
o
±
5
s
-om
8
in Jh 7
n
1 | |
in i- in c
T- 0
UJ
a
-
-O)
n
s.
>
1
> o
1 1 !
3 y CVI PI ^
a
/ a
^
w fc
2 ^
"? 81
-------
(This page is intentionally left blank.)
80
-------
CO
Q
UJ
LU
D_
CO
I—
DC
0
Q.
CO
Z
DC
h-
U_
O
2£
o
CO
cc
1
O .
O
^_^
UJ x
— o
4- Q
I"
uJ
CO
cc
O U-
i
&»
sfe * •*
1 1 1 1 1 1 1 1 1
-
^-^
^fe2
c^i
F *
' ' ' '
^-
§ CO
DC
O
§ DC
LU
g | O
1 §
§i o
a. |
(0
§ Q
§ H O
CC
1—
Z
o LU
0
o
Hi 33
— O
^ °
uJ
CO
i
1
*p
^ j£
"fe
W
i i i
t
L
~&
&
t&P
b ^
i i i
- TT
(fl
•g
0
o
o CJ
^ Q
-So 8 "o
i §i
-g t: a 33
"° O «" r<
D. C U
co rt ^H
i ~ E
0 H .C ^
* " ^
o o
" CM f~^ O
H § S
^, — ' 4>
u, 3 "?>
_Q'«<>icic\i"~^^^o5oc>~ * ™ N *~ ° T *)" "? T WH ° ""
3 if
p o c "5
CO
Q
LU
LU
DL
CO
1-
DC
O
Q_
CO
Z
^
tr
LJL
0
0
CO
DC
"^
~
O
UJ j;
i °"
^.
CO
cc
o
^ fe
|
^5 H tf5
^u.
.» .
^
* ^
- co
DC
O
§ ff
"" DC
LU
8 Z
-So O
a>" ^^
1 °
-s | O
§ Q
5 ^ O
DC
|_
Z
- S 1 1 1
o
-5*
- t-1
i i i i i i i i i i i i -"
^••••s. NCSIcO T^T-T-I- O O O O
UJ X
5 a.
2
CO
cc
Q
u to
1
*
ft &v
1
" <
j in *- in ic
fe
*
',
**
i i i
> I/) -^- i/> c
- ? T3 CO *5
C • *. rt
« « C
W c .2?
K 'C K
-8 C4 |^
o ° -S
o « ^
g M n <£
>- ^ Q
J fc «
-s| 5|i
E O 2
~8 8. 1. H §
- 8 S ^ 6
o i? u A 8
*«
-S 3
.2?
u.
- O
cc Y
"*•— -'' LU 0
Ml Q
a/ a mQ,*Q
81
-------
5.3.4 Multiple-Layer Eulerian Models
ADOM--
For less than one-third (32%) of the 28 footprint-days in the first half of the ANATEX period were
the ADOM transport speeds within ±20% of the actual transport speeds (Figures 30a and 30b). Another
one-third were in error by at least a factor of 2. Although its greatest errors were overpredictions,
ADOM showed a strong tendency to underpredict the transport speeds. After 48 h of transport, the
calculated centroids for 5 of the 7 footprint-days (71%) were within ±50% D m of the measured centroids
(Figures 30c and 30d).
ADOM transport tended to be to the right of the actual transport (19 of 28 footprint-days). Unlike
many of the other models, for those cases when the separation distance was greatest, the model did not
consistently underpredict or overpredict the transport speeds.
ADOM performance in predicting transport speeds tended to be related to the meteorological
scenarios. The greatest ADOM transport speed errors tended to pertain to cyclonic and/or frontal
interceptions when the model overpredicted (PTCH-4, -5, -15; PDCH-4, -10, -15). Exceptions were
PDCH-1, which followed, but never was intercepted by, a cold front, and PDCH-8, which remained in a
pressure ridge with light wind speeds. The model performed best for PTCH-1. -7, -11 and PDCH-1, -2;
only PTCH-7 and PTCH-11 were intercepted by cyclones and/or fronts.
Likewise, the greatest centroid location errors tended to pertain to cyclonic and/or frontal
interceptions (PTCH-15; PDCH-4, -10, -15). ADOM performed well for PTCH-1 and PDCH-1-both
unaffected by cyclones and/or fronts-but also performed well for cases pertaining to interceptions (PTCH-
4, -5, -7, -11; PDCH-2, -5). Therefore, ADOM location errors appeared to be less related to
meteorological scenario than did its transport speed errors.
82
-------
CO
Q
LU
LU
Q.
CO
CC
O
Q_
CO
z
^£
cc
o
z
O !
CO
cc
DL
o .
o
5 .j.
§o
Q
< o-
2 *
i i i i i f i ( i
*~
-
- »
oj a)
-
0) CM 0
fill
— ^
-
-S co
DC
O
-1 DC
LU
en ^^
-8 1 O
-C i
- 1 S
-8 ? O
s- -1
- § Q
DC
h-
Z
"8 O
X—.
^3
s I
§o
Q
< °-
*~
in ^
•tA
CO CM
-
CMCO
1 1 1
t :
° *
o
"
1 1 I
— ^
-
-0
CO
-g |
to fe
, o 3 0
i~ co O W
-c TJ t
" E" s,l
-| IS
CO f^ ffl
5 " eu
2 o -a
*"" ® 4>
X "f
Q <2
o CU co
el
ts o
o ClJ O
. . ^4-(
(3 CJOJOJW«-*-*-T-OOOO .•.• fcH NH
LU o LU a H if
a/ a a/ a ^ cu
9 9 ^ "O
CO
Q
LU
LU
Q_
CO
h-
CE
O
F TRANSP
0
z
O
CO
cc
«^
Q.
^>
o
o
0 0
< CL
s ^
f
1
1 1 1 1 1 1 1
u*-
m
3 V
\ \ I
0
™ CO
cc
o
o fy^
° DC
LU
1 Z
-80 O
1 I "
"8 I O
CO —1
2 Q
-s H O
DC
Z
-S LU
O
"o*
~ o
^ g
§ E
U> i-
m
10
•«)
i 1 !
? S
1 1 1
«2-g
§ 8 *
51 S" u
•*-* Q«
l«
O '"^
o Q. 3.
- 0 CO ^^
(0 O
u CM
•*-»
60 80
port time, hours
rs in ADOM
roid locations
co 0 S
• i s s
H W 0
o
*«
- ° 3
00
E
- o
(D^FCNJC\ICD(D^'CMi-CO
-------
ADPIC-
Similar to SRL, ADPIC diluted surface concentrations much slower than the measurements
indicated. Consequently, there were only 27 ADPIC footprint-days for 15 releases when footprints could be
related to a single release. This was the lowest total for any model-even those that were applied only
during the first half of the ANATEX period [ADOM (28) and MLAM-FINE (44)]. Since there were so
few footprint-days, relating ADPIC errors to type of meteorological scenario was not attempted.
As Table 12 indicates, ADPIC performance mirrored that for ADOM, the other Eulerian model.
Less than one-third (30%) of the ADPIC footprint-days were within ±20% of the actual transport speed
(Figures 31a and 31b). Like ADOM, ADPIC tended to underpredict the transport speeds. After 48 h of
transport, 5 of the 7 footprint-days (71%) were within ±50% D of the measured centroids (Figures 31c
and 31d), identical to ADOM. Unlike ADOM, however, ADPIC located its centroids to the right of the
actual centroids for virtually all 19 PDCH footprint-days; the opposite was true for the PTCH footprint-
days.
84
-------
CO
Q
LU
LU
D_
CO
i
1
DC
O
Q_
CO
Z
^f
DC
1-
U_
O
Z
o
CO
DC
Q.
O ,
O
-5- -
0 I
58
0)
T
CM
-
5°
S
11111)111
CM CM (M OJ T- i- *- i-
-
: o
™ «
s
> S 5
1 1 1 1
- T
-
-S co
DC
O
-i DC
LU
2 Z
-§ § O
•*• 1
- 1 fe
1 0
-8 ? O
s- -1
" | Q
DC
h-
Z
"8 0
g -
> T
a i
^ Q-
«
8
in o,
« ^
CT *"
8CCM- a
r<\l
II 1
(0
3 v OJ «
^
-
-8
V)
-8 g
1-1
2 »
-So «
-c -0 £
- I a,
"III
c a «
2 o-o
-s H «2 w
M •£
S Ui
Q <2
0 &H VI
~ OJ •*—*
^ G
*U *^N
' v&
C 0
53 *~
==- U Q
01 0 LU 3 ^"* X
a/ a a/ a eu
S •*""
9 Q . I^T,
FEEDS
CO
DC
O
Q_
CO
DC
\—
O
z
o
CO
DC
Q_
O
O
0 I
E 0
O
(1
1 t 1 i 1 1 1
in
„
in
n
i i I 1
o
^ CO
DC
O
DC
h° DC
LU
2 Z
-So O
£ —
E *^
-8 i o
Q- _J
w — '
" 1 Q
-9 ^ O
DC
Z
_ o ill
-Q-
- o
0 I
E O
(0
T
in
43
^
»
i i i
v ^ -~-
' ?K
-S £> Q
« H
IM
O 'TT*
-8 « ^
C k<
ta o
Ui <<-
' i 5.1
" _c 2? *~*
- Q «
- 1 < 3
10 O e/a O
Q_ ^H LJ
W O -t-1
- s t g
(± W u
ff-j
-8 3
.2?
UH
- 0
-------
SECTION 6
SUMMARY AND CONCLUSIONS
The transport and horizontal diffusion calculated by 11 atmospheric models were compared to the
transport and horizontal diffusion determined from 24-h tracer concentrations across a 77-site surface
sampling network encompassing central and eastern North America. Three of these models are single-layer
Lagrangian (SRL, TCAL, and VCAL), six are multiple-layer Lagrangian (ARL, BAT, GAMUT,
HY-SPLIT, MLAM-FINE, and MLAM-COARSE), and two are multiple-layer Eulerian (ADPIC and
ADOM). Tracer concentrations were measured during the period 5 January - 29 March 1987.
The limitations of the ANATEX data (e.g., virtually no vertical tracer distributions beyond 300 km of
the release sites, the spacing between sites, and 24-h integrated sampling at surface sites), limited the scope
of the model evaluation study. Firstly, evaluations based on point-to-point comparisons of simultaneous
tracer concentrations were not practical. Secondly, model errors could not be related to specific model
processes. Consequently, this model evaluation study focused on identifying model biases for whatever
reason. When appropriate, possible problems with modeled processes were offered as explanations for the
observed biases. However, a more resolved data base and additional model applications are required to
reveal the actual causes of these errors.
The model performance assessment-summarized in Sections 6.1 through 6.3 for each of the three
model genres-was based on seven performance measures and charts using either both halves of the entire
data set or a subset of this data set relating concentrations to specific tracer releases. Table 14 succinctly
summarizes the model performances via either mean errors or mean ratios of calculated-to-measured
values derived from some of the figures and tables presented in preceding sections. These performance
terms, appearing across the top row of Table 14, are described below.
(1) Third quartile-ratios calculated from the frequency distributions of nonzero concentrations
(Figures 7a, 7b, 9a, and 9b).
(2) Frequency of x's > threshold-ratios calculated from the percentage of concentrations
exceeding the thresholds of 5 dfL/L for PTCH and 8 dfL/L for PDCH (Figures 8a, 8b, lOa,
and lOb).
86
-------
TABLE 14. OVERALL BIASES IN THE MODEL RESULTS AS INDICATED BY THE MODEL
PERFORMANCE MEASURES
Ensemble Measures: Ratios of c/m f
(first half, second half)
Model
Third
quart i le
Frequency
of x 's >
threshold
Footprint
Mean of Mean lateral Mean ratio
X for diffusion of speed
3 bands for 3 bands ( >7.5 h)
Measures
Mean ratio
of location
error (De/Dn)
( >7.5 h)
Mean
location
error (km)
( <61.5 h)J
Single-Layer Lagrangian:
SRL
TCAL
VCAL
T
D
T
D
T
D
*3.7, 23.3
*6.0. 9.3
2.8, 0.8
4.1. 0.8
*2.2, 1.6
*3.8. 1.8
1.2,
*2.0.
0.8,
0.9.
0.9,
*1.5.
0.1
0.5
1.9
0.9
2.3
2.5
*2.8, -
*5.0. 2.0
1.4, 2.8
*2.0. 1.6
*1.8, 3.0
*3.0. 2.4
0.8,
1.0.
0.6,
0.5.
0.7,
0.7.
--
0.2
2.3
1.2
2.1
1.7
1.4
1.4
1.1
2.8**
1.1
1.4
+0
+0
-0
+0
+0
+0
.44
.64
.03
.58
.17
.22
536
570
283
Multiple-Layer Lagrangian:
ARL
BAT
GAMUT
HY-SPLIT
HLAH-FINE
MLAH-COARSE
T
D
T
D
T
D
T
D
T
D
T
D
1.1, 0.8
2.3. 1.3
0.9, 0.7
*2.7. 1.6
1.1, 1.2
3.2. 1.4
*3.5, 2.3
*4.9, 2.1
0.7, --
*3.4. --
0.9, 0.9
*2.8. 1.5
0.6,
1.0.
*0.5,
0.6.
0.4,
1.0.
0.8.
0.8.
0.8,
*1.9.
0.7,
*1.7.
0.6
0.9
0.4
0.5
0.8
1.1
0.6
0.8
--
-.
1.7
2.0
0.5, 0.9
1.1. 0.9
0.7, 0.4
1.9. 0.6
0.7, 1.2
2.2, 1.4
1.4, 0.7
1.6. 0.9
0.9, -
*4.1, --
0.9, --
*3.8. 1.8
0.6,
0.6.
0.6,
*0.4,
0.4,
0.5.
*0.5,
0.4.
1.0,
1.0.
0.7,
1.0,
0.9
0.8
0.6
0.4
0.8
0.9
0.5
0.6
--
.-
2.1
1.7
1.1
1.1
1.0
1.5**
1.1
1.3
1.1
1.4
0.9
0.8
0.9
0.8
+0
-0
-0
+0
-0
-0
+0
-0
.03
.26
.05
.44
.27
.27
.05
.40
+0.09
0
+0
+0
.00
.06
.02
370
298
417
366
270
241
Multiple-Layer Euterian
ADOM
ADPIC
T
D
T
D
0.6, --
*1.6. --
0.9, 1.0
1.2. 1.6
0.8,
1.3.
1.4,
1.8.
--
--
0.3
1.2
0.6, --
*1.9. --
0.9, 0.4
1.7. 0.7
1.1.
1.1.
1.4,
1.6.
--
.-
0.3
0.9
0.9
1.2
0.9
0.9
-0
-0
-0
+0
.02
.93
.21
.68
467
457
T: PTCH; D: PDCH
f c/m: calculated/measured.
J Mean location error for combined PTCH and PDCH.
* Calculated values for both half-periods differed by at least ±50% of measured values (first half for MLAM-FINE and ADOM).
** Calculated values differed by at least ±50 of measured values.
87
-------
(3) Mean of \ for 3 bands-mean ratios of the ensemble mean concentrations for 3 bands of sites
(ranging from 300 km to 2,300 km downwind of the tracer release sites) (Figures lla, lib, 12a,
and 12b).
(4) Mean lateral diffusion for 3 bands-mean ratios of the ensemble mean lateral diffusion for the
same 3 bands of sites (Figures 13a, 13b, 14a, and 14b).
(5) Mean ratio of speed-determined from the ratios of the 24-h mean transport speeds of all
footprint pairs beyond 7.5 h of transport (Figures 18-31); ratios for PTCH and PDCH
footprints are provided.
(6) Mean ratio of location error-determined from the ratios of 24-h separation distances (£>e)
normalized by actual transport distance (£> ) of all footprint centroid pairs beyond 7.5 h of
transport (Figures 18-31); ratios for PTCH and PDCH footprints are provided.
(7) Mean location error-determined by averaging the mean absolute centroid separation distances
for the combined set of PTCH and PDCH footprints at transport times of 13.5 h, 37.5 h, and
61.5 h (Figure 15).
For most of the terms, values were calculated for both PTCH and PDCH for both half periods. Those
models that did not perform as well as the others (i.e., when the ratios for both half periods exceded ± 50%
of measured values) are identified by an asterisk preceding the values in the table.
6.1 SINGLE-LAYER LAGRANGIAN (SLL) MODELS
6.1.1 Summary
Box Plot Distributions-TLacli of the three models of this genre (SRL, TCAL, and VCAL) revealed a
tendency to overestimate the frequencies of higher concentrations. During the first half-period, the
medians and third quartiles of each model were 2-to-6 times greater than those of the measurements. For
the second half-period, the same was true for SRL only; TCAL and VCAL values were much closer (i.e.,
within a factor of 2) to those of the measurements.
Frequency Distributions—During the first half-period the SLL models generally approximated the
frequency of concentrations above the thresholds (5 dfL/L and 8 dfL/L for PTCH and PDCH,
respectively). However, SRL overestimated the frequencies of concentrations exceeding 99 dfL/L by
approximately a factor of 4. TCAL, and to a lesser degree, VCAL closely approximated the distributions.
During the second half-period SRL greatly underestimated the frequency of PTCH concentrations above
88
-------
the threshold (<1% versus 18%), as well as the frequency of nonzero concentrations (1% versus 36%).
Meanwhile, the percentage of sites with nonzero concentrations calculated by both TCAL and VCAL was
much greater than that for the measurements of both tracers (55% to 80% versus approximately 40%).
Mean Concentrations as a Function of Transport Distance-For transport distances ranging from
300 to 2,300 km, the mean TCAL and VCAL concentrations during both half-periods and the mean SRL
concentrations during the first half-period along several bands of sites tended to be higher than those of the
other models, as well as the measurements. Deviations of factors of 2 and 3 from the measured means
were common; some SRL deviations were as great as a factor of 6.
Mean Lateral Diffusion-The comparison of the model and measured mean plume widths showed an
inconsistency between half-periods. For the first half-period, the model mean plume widths were generally
underestimated, but within an average of 20% of the measured width for TCAL, 30% for VCAL, and 50%
for TCAL. For the second half-period, TCAL and VCAL mean plume widths were greater than the
measured mean plume widths especially for PTCH, by as much as 130%. SRL mean plume widths for this
period (less than 250 km) were much less than the actual widths~in fact, nearly zero-indicating a serious
problem with the diffusion.
Footprint Transport Speeds and Centroid Locations-Of these three models, VCAL performance was
clearly best in calculating the transport speeds and centroid locations of tracer footprints. Although VCAL,
as well as TCAL, demonstrated a tendency to overestimate the transport speeds (10% and 40 % for VCAL
PTCH and PDCH, respectively; 10% and 180% for TCAL PTCH and PDCH, respectively), the VCAL
overestimates exceeded a factor of 2 for only 6 footprint-days, compared to 24 footprint-days for TCAL. In
addition, VCAL showed little bias in placing its PTCH and PDCH footprint centroids (mean ratios ~ +20%
Z>m). TCAL was less consistent, showing no bias for PTCH footprints, but a large positive bias (i.e., its
centroids generally were to the south of the measured centroids) for PDCH footprints. TCAL also tended
to place its centroids to the right of the actual centroids when the transport speeds were overestimated.
SRL tended to both overestimate transport speeds by +40% and place the footprint centroids to the right
of the actual centroids.
Mean Trajectory Errors—The mean centroid location errors of SRL and TCAL were greater than any
other model. These errors increased linearly with transport time from approximately 350 km to 800 km
after 13.5 h and 61.5 h, respectively. On the other hand, the mean centroid location errors for VCAL were
among the least, half those of SRL and TCAL.
6.1.2 Conclusions
The SRL transport vectors, based on surface pressure gradients, clearly are biased: speeds tend to
be overestimated and directions tend to be to the right of the actual vectors. This bias, as well as its
89
-------
direction, is not surprising given the nature of geostrophic wind vectors. The high bias in the transport
speed can explain SRL's tendency of overestimating the mean concentrations and the frequency of high
concentrations, as well as underestimating the number of sites with nonzero concentrations and the lateral
diffusion. That is, for a model that overestimates transport speed, the plumes will be narrower and the
concentrations will tend to be greater for fixed distances and transport times.
The definition of the height of the mixed layer-the only difference between TCAL (fixed height at
1,500 m AGL) and VCAL (variable height based on potential temperature profiles)~has a large influence
on the performance of a single-layer model. This underscores the need to choose carefully the layer
through which wind vectors are to be calculated. The low mean centroid location errors for VCAL
indicates that a single-layer model can perform as well as the models of the other genres.
The general tendency of TCAL, and to a lesser degree VCAL, to overestimate the transport speeds
can explain their tendencies to overestimate the mean concentrations, as a consequence of slower tracer
diffusion relative to transport distance. Since wind speeds generally increase 'with height, the higher TCAL
transport speeds could have resulted from a mixed layer height that was too high; climatological data
suggest that the 1,500-m height is a factor of 2 too high.
62 MULTIPLE-LAYER LAGRANGIAN (MLL) MODELS
6.2.1 Summary
Box Plot Distributions--In general, with the exception of HY-SPLIT, the means, medians, and third
quartiles of the MLL models more closely corresponded to those of the measurements than did those of
the single-layer Lagrangian models. This was especially true for the PTCH concentrations, where the
means, medians, and third quartiles were within ±50% of those for the measurements. Those for the
PDCH concentrations were generally greater than those for the measurements by a factor of 2. HY-SPLIT
means, medians, and third quartiles tended to be greater than those of the other MLL models, greater than
those of the measurements by as much as factors of 3 and 4, and more closely resembled those of the SLL
models. During the first half-period, HY-SPLIT medians were comparable to the third quartiles of the
measurements.
Frequency Distributions—The frequency distributions of the MLL models generally corresponded
more favorably to those of the measurements than did the SLL models. With the exceptions of BAT,
MLAM-FINE, and MLAM-COARSE, the MLL model frequencies of concentrations above the thresholds
approximated those of the measurements; MLAM-FINE and MLAM-COARSE frequencies tended to be
higher by a factor of 2. HY-SPLIT concentrations and MLAM-FINE and MLAM-COARSE PDCH
concentrations exceeding 99 dfL/L during the first half-period occurred at least twice as often as those
90
-------
measured; the opposite was true for ARL PTCH concentrations. ARL and GAMUT distributions of
PDCH concentrations were virtually identical to those of the measurements.
Mean Concentrations as a Function of Transport -D/sto/ice-Although the mean concentrations along
the 300-m and 1,000-m bands for the MLL models tended to be lower than the actual means by factors
of 2 to 4, the mean concentrations along the bands for the MLL models tended to more closely resemble
those of the measurements than did the SLL models. This was especially true at distances farther
downwind of the release sites, where the means of all MLL models but MLAM-COARSE were within
±2 dfL/L of the actual means; MLAM-COARSE means generally were high by a factor of 2 to 3 at all
distances.
Mean Lateral Diffusion—Dwvag the first half-period the mean dispersions of MLAM-FINE and its
sibling, MLAM-COARSE, were within ±30% of the actual dispersion. GAMUT and HY-SPLIT
dispersions were the lowest of all models, factors of 2 to 3 lower than the actual dispersions. The
dispersions for the other MLL models were 50% to 100% lower than the actual dispersion. During the
second half-period both BAT and, once again, HY-SPLIT dispersions were the lowest of any model (lower
than the actual dispersions by factors of 2 to 3). MLAM-COARSE dispersions were very high, generally
double those of the measurements. The dispersions of the remaining MLL models were lower than the
actual, but within 30%. For both half- periods, BAT and HY-SPLIT plume widths showed virtually no
change with transport distance.
Footprint Transport Speeds and Centroid Locations-Each of the MLL models demonstrated skill in
simulating the tracer transport. With two exceptions (i.e., BAT and HY-SPLIT PDCH footprints), mean
transport speeds were within an average of 30% of actual mean speeds and mean relative location errors
were within 30% of the actual transport distance. Furthermore, only VCAL-a SLL model—had a lower
mean absolute location error than the MLL with the greatest error (GAMUT: 417 km).
The performances of half of the MLL models in simulating footprint transport speeds and centroid
locations did not vary for each tracer. Performances for the three exceptions-ARL, BAT, and HY-SPLIT-
were better for the PTCH data, partly because the PTCH data set tended to be dominated by simpler,
northwest flows, while the PDCH data set included a wider range of flow patterns. For example, BAT
showed the lowest speed and location errors and no significant biases for the PTCH footprints; however,
for the PDCH footprints, BAT speeds strongly tended to be lower than actual speeds and its centroids
tended to be to the right of the actual centroids. MLAM-FINE and MLAM-COARSE speeds tended to be
lower while GAMUT speeds tended to be greater, but for each model more so for the PDCH footprints.
Although MLAM-COARSE showed minimal location errors for both PTCH and PDCH sets, MLAM-
FINE locations tended to be to the right of the actual locations for both tracers while GAMUT locations
tended to be to the left of the actual PTCH footprints. Both ARL and HY-SPLIT showed the most scatter
91
-------
for speeds and location errors; speeds tended to be greater than the actual speeds. ARL centroid locations
for the PDCH footprints tended to be to the left of the actual centroids.
Mean Trajectory Errors—The six MLL models were divided into two types of behavior. BAT,
MLAM-FINE, and MLAM-COARSE mean errors peaked to 400 km after 3.5 days of transport and
showed no significant additional increase beyond that. On the other hand, ARL, GAMUT, and HY-SPLIT
errors increased more sharply with transport time, reaching 530 ±30 km after 2.5 days and 920 ±160 km
after 3.5 days, or about 3 times greater than the other MLL models. Errors decreased to 750 ±100 km
after 4.5 days.
62.2 Conclusions
The MLL models clearly outperformed all others except VCAL in simulating the transport of tracer
footprints. With the possible exceptions of ARL and GAMUT, none of the MLL models clearly
outperformed any other in its genre. For most of these models, the majority of the performance measures
indicated relatively good performance, but the remaining performance measures indicated biases in the
model results. For instance, ARL showed little if any bias in its results as did GAMUT (with the exception
of its high bias in the transport speeds), but the mean location errors were relatively great. This indicated
that although their mean errors were rather substantial, their centroids, in general, were to the left of the
actual centroids as often as they were to the right. The relatively good comparison of its distribution
statistics with those of the measurements appears to indicate that ARL and GAMUT simulated rather well
the lateral/vertical diffusion; however, they both appeared weak in simulating the transport.
BAT's underestimates of the mean lateral diffusion and the frequencies of occurrence of
concentrations above the thresholds would appear to be related to each other. That is, a model that
underestimated plume widths will show fewer cases of nonzero concentrations and concentrations above the
thresholds. The tendency to overstate the PDCH footprint transport speeds and to place its PDCH
footprint centroids to the right of the actual centroids indicated that BAT's vertical mixing could be
overstated, effectively giving more influence to the higher-altitude winds, which tend to have greater speeds
and directions to the right of the lower-altitude winds. The very low mean centroid location errors,
however, indicate that BAT simulates well the transport.
HY-SPLIT's tendencies to understate the plume widths and overstate its third quartiles could be
symptomatic of its algorithm for calculating atmospheric stability from the NGM results, as opposed to the
algorithm of ARL (its sibling), which interpolates surface and rawinsonde data. Like ARL, HY-SPLIT
mean centroid location errors were relatively great, yet no substantial biases were evident in its calculation
of footprint speeds and locations. The relatively high-turbulent K2 profiles used by HY-SPLIT may have
exaggerated the vertical diffusion, adversely affecting its performance.
92
-------
MLAM-FINE's footprint transport speeds tended to be lower and to the right of the actual speeds
and locations for the first half-period, the only period for which it was applied. These slower speeds could
explain its other tendency to overestimate the frequency of concentrations exceeding the thresholds. That
is, its footprint widths would be wider relative to transport distance. Furthermore, at any one site the
concentrations from one release can be nonzero for two days rather than one day.
For both half-periods, MLAM-COARSE showed the same tendencies as MLAM-FINE. In addition,
MLAM-COARSE tended to overstate the frequency of concentrations exceeding the thresholds as well as
the mean concentrations at all distance downwind of the release sites. Especially during the second half-
period, MLAM-COARSE plume widths tended to be greater than those of the actual widths. All but one
of these biases could be explained by the slower MLAM-COARSE transport speeds; the high bias in the
mean concentrations could be symptomatic of a low bias in the vertical mixing, causing concentrations near
the surface to be biased high.
63 MULTIPLE-LAYER EULERIAN (MLE) MODELS
63.1 Summary
Box Plot Distributions--The correspondence between model and measurement distributions varied
with tracer. The ADOM median and third quartile for PTCH concentrations were 30% to 60% less than
those of the measurements, while the opposite was true for the PDCH concentrations. The PTCH box plot
distribution for ADPIC was virtually identical to those of the measurements; the ADPIC median and third
quartile for PDCH concentrations were within 60% of those for the measurements.
Frequency Distributions-The ADOM distributions for the first half-period corresponded very closely
to those of the measurements; ADOM was not applied for the second half-period. The comparisons of the
ADPIC with the actual distributions were inconsistent; while ADPIC distributions were quite similar to the
actual distributions for PTCH during the first half-period and PDCH concentrations for the second half-
period, the ADPIC frequencies for concentrations exceeding the thresholds deviated by 75% for the
remaining half-periods for each tracer.
Mean Concentrations as a Function of Transport Distance-Tlie mean PTCH concentrations for three
bands of sites for both ADOM and ADPIC tended to be lower than those of the measurements. Ratios of
calculated-to-predicted means for all three bands ranged from 0.4 to 0.9. However, the opposite was true
for first-period PDCH concentrations; these ratios approximated 1.8 for both ADOM and ADPIC. Second-
period ADPIC mean PDCH concentrations were lower by an average of 30%.
Mean Lateral Diffusion-During the first half-period the lateral dispersion of both models was
generally greatest of all the models and greater than the actual dispersion an average of 10% for ADOM
93
-------
and 50% for ADPIC. During the second half-period, ADPIC dispersion for PTCH was low by factors of 2
to 3 for some bands and varied little with transport distance; however, its PDCH dispersion was within
±10% of the actual dispersion at all distances.
Footprint Transport Speeds and Centroid Locations-ADOM speeds for both tracers tended to be
lower than actual speeds by 20% to 40%. For its largest location errors, ADOM speeds were greater than
actual speeds by at least a factor of 2 and its centroids tended to be to the left of the actual centroids.
However, in general, its centroids tended to be to the right. Similarly, ADPIC speeds tended to understate
the actual speeds, sometimes by as much as 60% and 80%; ADPIC centroids for the PDCH footprints were
to the right of the actual centroids in nearly every case and to the left for the PTCH footprints.
Mean Trajectory Errors--Both ADOM and ADPIC mean errors were among the greatest of all
models. The ADOM mean error after 1.5 days was low, approximately 200 km, but quickly increased to
600 km after 2.5 days (greatest of all models), then decreased to 500 km after 3.5 days. Similarly, the
ADPIC mean error increased sharply from 300 km at 1.5 days to 850 km after 2.5 days, among the greatest.
632 Conclusions
In general, the MLE models performed quite similarly and better for the ensemble measures than
they did for the footprint comparison measures. This implies that these two models performed relatively
well for the average, but performed relatively poorly for individual cases. The only substantial bias
observed in the ensemble measures was for lateral diffusion; both ADOM and ADPIC tended to overstate
the footprint widths in the first half-period, the only period for which ADOM was applied. However, both
models tended to understate the footprint speeds, which could by itself explain the high bias in lateral
diffusion, as well as the large mean centroid location errors.
The strong relationship between the large ADOM centroid location errors and overestimated
transport speed errors could indicate a problem with its vertical diffusion for several cases (PTCH-15,
PDCH-4, -10, and -15), all of which were intercepted by cyclones or fronts. Thai is, the model could have
overestimated vertical diffusion and, as a consequence, relied more on the faster wind speeds at higher
levels.
The strong ADPIC tendency to understate the transport speeds and place footprint centroids to the
right of the actual footprint centroids demonstrates its weakness in simulating the transport for the 27
footprints of this study. Additional data are needed to substantiate this conclusion. The reason for
relatively few ADPIC footprints related to the fact that ADPIC concentrations very often did not return to
zero days after actual tracer footprints were transported across regions.
94
-------
SECTION 7
RECOMMENDATIONS
The results presented in this report are the product of the first step in evaluating these long-range
transport and diffusion models. It is recommended that the developers of these models continue the
evaluation process by using the ANATEX data and the AMES results to better understand the behavior of
their models. This can be accomplished by sensitivity analyses in tandem with comparisons of calculated
and measured concentrations. These sensitivity analyses should focus on the changes in model calculations
and performances with respect to changes in such model variables as vertical mixing, horizontal diffusion,
and transport speeds and directions. Publication of these latter results is strongly recommended.
It is also recommended that the ANATEX aircraft and tower data be used to evaluate the multiple-
layer models with respect to the vertical distribution of tracer and the effects of vertical wind shears on the
tracer plumes. Regrettably, these data were unavailable at the time this study was conducted.
Finally, it is strongly recommended that the sampling protocol for future tracer studies include
aircraft sampling at greater distances downwind of the release site than the 0-300 km of ANATEX. (The
problem of prescribing the sampling flight patterns in a prognostic fashion is recognized.) Vertical
distribution of tracer at greater distances must be known to identify specific causes of model errors.
Without these distributions supplementing the ground-base measurements we can only state that there
appears to be a model error, but we would not be equipped to relate the error to a specific modeling
process.
95
-------
REFERENCES
Atchison, M.K. and C.R. Parks. 1986. Evaluation of a single-layer trajectory and diffusion model using the
Metropolitan Tracer Experiment (METREX) data base. Fifth Joint Conference on Applications of
Air Pollution Meteorology, Chapel Hill, NC, November 1986.
Businger, JA. 1973. Turbulent transfer in the atmospheric surface layer. Workshop in Micrometeorology,
American Meteorological Society, Boston, MA. pp. 67-100.
Businger, JA., J.C. Wyngaard, Y. Izumi, and E.F. Bradley. 1971. Flux-profile relationships in the
atmospheric surface layer. /. Atmos. Sci., 28:181-189.
Businger, JA. and S.P.S. Arya. 1974. Height of the mixed layer in the stably stratified planetary boundary
layer. Advances in Geophysics, ISA, Academic Press, pp. 73-92.
Byers, H.R. 1974. General Meteorology, McGraw-Hill Book Company, New York, NY. 461 pp.
Davis, W.E., A.R. Olsen, and B.T. Didier. 1989. MLAM assessment of radionuclide air concentration and
deposition for the Chernobyl Reactor accident. Air Pollution Modeling and Its Application VII.
H. van Dop, Editor. Plenum Press, New York, NY. pp. 123-136.
Draxler, R.R. 1982. Measuring and modeling the transport and dispersion of Krypton-85 1,500 km from a
point source. Atmos. Environ., 16:2763-2776.
Draxler, R.R. 1987. Sensitivity of a trajectory model to the spatial and temporal resolution of the
meteorological data during CAPTEX. /. of dim. andAppl. Meteor., 26:1577:1588.
Draxler, R.R. 1988a. Across North America Tracer Experiment (ANATEX) weather maps and tracer
concentrations. NOAA Technical Memorandum ERL ARL-165, Air Resources Laboratory, Silver
Spring, MD. 87 pp.
96
-------
Draxler, R.R. 1988b. HYBRID Single-Particle Lagrangian Integrated Trajectories (HY-SPLIT): model
description. NOAA Technical Memorandum ERL ARL-166, Air Resources Laboratory, Silver
Spring, MD. 23 pp.
Draxler, R.R. and J.L. Heffter. 1989. Across North America Tracer Experiment (ANATEX), Volume I:
Description, ground-level sampling at primary sites, and meteorology. NOAA Technical
Memorandum ERL ARL-167, Air Resources Laboratory. Silver Spring, MD. 83 pp.
Eimutis, B.C. and M.G. Konicek. 1972. Derivations of continuous functions for the lateral and vertical
atmospheric dispersion coefficients. Atmos. Environ., 6:859-863.
Ferber, G.J., J.L. Heffter, R.R. Draxler, RJ. Lagomarsino, F.L. Thomas, R.N. Dietz, and C.M. Benkovitz.
1986. Cross-Appalachian Tracer Experiment (CAPTEX '83) Final Report. Technical Memorandum
ERL ARL-142, National Oceanic and Atmospheric Administration, Silver Spring, MD. 60 pp.
Gifford, FA. 1987. The time-scale of atmospheric diffusion considered in relation to the universal diffusion
function, f-f. Atmos. Environ., 21:1315-1320.
Godowitch, J.M. 1989. Evaluation and sensitivity analyses results of the MESOPUFF II model with
CAPTEX measurements. EPA-600/3-89/056, NTIS No. PB 89-198253, U.S. Environmental
Protection Agency, Research Triangle Park, NC. 99 pp.
Haagenson, P.L., Y-H Kuo, M. Skumanich, and N.L. Seaman. 1987. Tracer verification of trajectory models.
/. ofClim. andAppl. Meteor., 26:410-426.
Heffter, J.L. 1965. The variation of horizontal diffusion parameters with time for travel periods of one hour
or longer. /. Appl. Meteor., 4:153-156.
Heffter, J.L. 1980. Transport layer depth calculations. Second Joint Conference on Applications of Air
Pollution Meteorology, New Orleans, LA, March 24-27, American Meteorological Society, Boston,
MA. pp. 787-791.
Heffter, J.L. 1983. Branching Atmospheric Trajectory (BAT) Model. NOAA Technical Memorandum ERL
ARL-121, Air Resources Laboratory, Silver Springs, MD. 19 pp.
Holzworth, G.C. 1972. Mixing heights, wind speeds, and potential for urban air pollution throughout the
contiguous United States. Technical Report AP-101, U.S. Environmental Protection Agency,
Research Triangle Park, NC. 118 pp.
97
-------
Kahl, J.D. and P.J. Samson. 1988. Trajectory sensitivity to rawinsonde data resolution. Atmos. Environ.,
22:1291-1299.
Kao, C.J. and T. Yamada. 1988. Use of CAPTEX data for evaluations of a long-range transport numerical
model with a four-dimensional data assimilation technique. Mon. Wea. Rev., 116:293-306.
Lange, R. 1978. ADPIC~a three-dimensional particle-in-cell model for the dispersal of atmospheric
pollutants and its comparison to regional tracer studies. /. Appl. Meteor., 17:320-329.
Lee, I.J. 1987. Numerical simulations of cross-Appalachian transport and diffusion. Bound. Layer Meteor.,
39:53-66.
Masters, S.E. and M.A. Kienzle. 1989. The impact of vertical mixing on large-scale atmospheric transport.
Sixth Joint Conference on Applications of Air Pollution Meteorology, Anaheim, CA, January 1989.
O'Brien, J.J. 1970. A note on the vertical structure of the eddy exchange coefficient in the planetary
boundary layer. /. Atmos. Sci., 27:1213-1215.
Olson, M.P., K.J. Puckett, D. Davies, and K.K. Oikawa. 1989. A comparison between the ADOM model
and the ANATEX tracer data. San Francisco, CA, November 1989.
Pasquill, F. 1976. Atmospheric dispersion parameters in Gaussian plume modeling: Part II. Possible
requirements for change in the Turner Workbook values, EPA-600/4-76-030b, U.S. Environmental
Protection Agency, Research Triangle Park, NC.
Persson, C., R. Henning, and L. DeGeer. 1986. The Chernobyl accident~a meteorological analysis of how
radionuclides reached Sweden. SMHI Reports, Meteorology and Climatology, No. 55, X-60176,
NorrkOping, Sweden.
Rodriquez, D.J. and K.R. Peterson. 1989. Simulating the venting of radioactivity from a Soviet nuclear test.
Atmos. Environ., 23:953-958.
Scholtz, M.T., B. Weisman, L. Mahrt, and A.D. Christie. 1986. Generation of meteorological data fields for
the ADOM Eulerian regional model. Fifth Joint Conference on Applications of Air Pollution
Meteorology, Chapel Hill, NC, November 18-21, American Meteorological Society, Boston, MA.
pp. 145-150.
Szepesi, D.J. 1989. Compendium of Regulatory Air Quality Models. Akademiai Kiado es Nyomda Vallalat,
Budapest, Hungary. ISBN 963 05 46752. 516 pp.
98
-------
Venkatram, A., P.K. Karamchandani, and P.K. Misra. 1988. Testing a comprehensive acid deposition model.
Atmos. Environ., 22:737-747.
Yamartino, R.J. and J.S. Scire. 1984. ADOM/TADAP Model Development Program Volume 3: The
transportation and diffusion modules. ERT Document No. P-B980-210, ENSR Inc., 1220 Avenida
Acaso, Camarillo, CA 93010.
99
-------
APPENDIX A
MODEL DESCRIPTIONS
A-l
-------
TABLE A-l.
SRL ATTRIBUTES
Savannah River Laboratory Adjusted Geostrophic Model (SRL)
Model genre:
Reference(s):
Resolution:
Horizontal--
Vertical--
Temporal - -
Advection:
Diffusion:
Horizontal--
Vertical--
Me t. parameters:
Mixing depth--
Stability--
Eddy diffusivity-
Winds--
single-layer segmented plume
none
variable
single layer3
3 h
based on curve fits to empirical and
theoretical plume growth
(Gifford, 1987)
function of z0, u, and Pasqui11-Turner
stability class
u,v calculated from 3-h surface p fields
generated by an objective scheme
forcing temporal continuity;
w estimated from terrain slope
wind direction, T advection, and
positive vorticity advection
Can be applied using two layers.
A-2
-------
TABLE A-2. TCAL AND VCAL ATTRIBUTES
Trajectory Calculation Model (TCAL) and
Variable Layer Trajectory Calculation Model (VCAL)
Model genre:
Reference(s):
Resolution:
Horizontal--
Vertical--
Temporal--
Advection:
Diffusion:
Horizontal--
Vertical--
Met. parameters:
Mixing depth-
Stability--
Eddy diffusivity-
Winds--
single-layer Lagrangian puff
Atchison, M.K. and C.R. Parks (1986)
rawinsonde network resolution (~ 300 km)
single layer
1 h
layer-mean winds;
TCAL: 300-1,500 m AGL;
VCAL: top determined by local mixing height
Gaussian
none
TCAL: constant at 1,500 m AGL;
VCAL: potential temperature technique
based on vertical changes in the
potential temperature; concentrations
based on a fixed depth of 1,500 m AGL.
N/A
N/A
Primary source: rawinsonde data;
Secondary source: AFGWC HIRAS data
interpolated via Cressman technique;
no surface data
A-3
-------
TABLE A-3. ARL ATTRIBUTES
Air Resources Laboratory Model (ARL)
Model genre:
Reference(s):
Resolution:
Horizontal--
Vertical--
Temporal--
Advection:
Horizontal--
Vertical--
Diffusion:
Horizontal--
Vertical--
Met. parameters:
Mixing depth--
Stability--
Eddy diffusivity-
Winds--
multilayer Lagrangian puff
Draxler (1987)
200 km
10 layers; 0-3,000 m AGL
1 h for trajectories; 3 h for meteorology
first-order approximation
conservation of 0W;
puffs split in the vertical dimension when
they overlap more than one grid cell
dah/dt = 1,853
( daz )2 /dt = 2 K2
height of the base of first elevated inversion
(Heffter, 1980)
surface layer flux relations of Businger
(1973)
interpolation formula of O'Brien (1970)
spatially and temporally interpolated surface
and rawinsonde data;
vertical velocity based on conservation of
0B (Byers, 1974)
A-4
-------
TABLE A-4. BAT ATTRIBUTES
Branching Atmospheric Trajectory Model (BAT)
Model genre:
Reference(s):
Resolution: a
Horizontal--
Vertical--
Temporal - -
Advection:
Horizontal--
Vertical--
Diffusion:
Horizontal--
Vertical--
Met. parameters:
Mixing depth--
Stability--
Eddy diffusivity-
Winds--
multilayer Lagrangian puff
Heffter (1983), Szepesi (1989)
rawinsonde network resolution (~ 300 km)
3 layers (300 m; critical inversion; 3,000 m)
3 h for advection and 1 h for diffusion
Modified Eulerian: "Z w, At d,'2 / S d,'2
N/A
Gaussian: ah - 0.5 t
uniform mixing within each layer;
branching puffs at day/night transitions
(0300 GMT and 1500 GMT)
where A0/Az > 0.005 K/m and
0T - 0B > 2 K
N/A
N/A
6-h temporally interpolated rawinsonde data
at sites (no spatial interpolation)
No grid is used.
A-5
-------
TABLE A-5.
GAMUT ATTRIBUTES
Global Atmospheric Multilayer Transport Model (GAMUT)
Model genre:
Reference(s):
Resolution:
Horizontal-
Vertical- -
Temporal -•
Advection:
Horizontal-
Vertical- -
Diffusion:
Horizontal-
Vertical--
Het. parameters:
Mixing depth--
Stability--
Eddy diffusivity-
Winds--
multilayer Lagrangian puff
Masters, S.E. and M.A. Kienzle (1989)
280 km
4 layers between 0 and 3,000 m AGL
3 h
layer-mean winds; predictor-corrector
advection method
stable lofting/sinking usLng moist static
energy as air mass tracer; vertical mixing
between layers according to computed depth
Gaussian, dah/dt = 1.0 nmi/h
trajectory splitting between layers controlled
by mixing depth (may reach 100 active
trajectories/release); mass loss to free
troposphere by stable lofting & convective
mixing
potential instability technique; mixing depth
is the level of the "critical inversion"
identified by the moist-adiabatic lapse rate
N/A
N/A
Primary source: rawinsonde data;
Secondary source: AFGWC HIRAS data over most
areas at 0600 and 1800 GMT; over oceans and
areas/times of missing rawinsonde data, no
surface data
A-6
-------
TABLE A-6. HY-SPLIT ATTRIBUTES
Hybrid Single-Particle Lagrangian Integrated Trajectories Model (HY-SPLIT)
Model genre:
Reference (s) '.
Resolution:
Horizontal-
Vertical- -
Temporal--
Advection:
Diffusion:
Horizontal-
Vertical- -
Met. parameters: a
Mixing depth--
Stability--
Eddy diffusivity--
Winds--
multilayer Lagrangian- -Eulerian hybrid
Draxler (1988b)
180 km
8 layers up to 0.721 of surface pressure
1 h for trajectory calculations;
2 h for meteorological data
first-order approximation; three dimensions;
puffs split in horizontal and vertical when
diffusive growth exceeds grid dimensions
dah / dt = 1,853
( daz y /dt = 2
K2
where x is uniform within plume of radius
1.54 crh and depth of 3.08 O2
calculated by the NGM
calculated from NGM boundary layer variables
(u* , rv , 0* )
calculated from NGM boundary layer variables;
Kz = 30 • ( 1 + JJ?,| ); minimum of 1 m2/s
near surface: normalized wind profiles
(similar to Businger et al . , 1971);
0;
at or above NGM a- level 1: calculated by
the NGM
Gridded analytical and prognostic output from NOAA's Nested Grid Model (NGM).
A-7
-------
TABLE A-7. MLAM (FINE AND COARSE) ATTRIBUTES
Multilayer Air Mass Model (MLAM)
Model genre:
Reference(s):
Resolution:
Horizontal--
Vertical--
Temporal--
Advection:
Diffusion:
Horizontal/
Vertical
Met. parameters:
Mixing depth--
Stability--
Eddy diffusivity--
Winds--
Vertical redistribution:
multilayer Lagrangian puff
Davis et al. (1989)
190 km (user defined)
9 layers with tops of:
100; 400; 700; 1,000; 1,500; 2,000;
2,600; 3,400; and 4,400 m
(user defined)
1 h
three-dimensional; based on gridded fields of
0T . ". v, w
<100 km: function of distance, OQ and AT;
based on regression analysis
(Eimutus and Konicek, 1972)
>100 km: ay2 / 2t (Heffter, 1965)
diurnal; maximum depth at 1600 LST
hourly; function of t (LST)
not used
site measurements or gridded
redistribution of mass twice daily
(FINE: 1000 LST and 1600 LST) or
once daily (COARSE) during the first
3 days to create new sets of puffs
A-8
-------
TABLE A-8. ADOM ATTRIBUTES
Acid Deposition and Oxidant Model (ADOM)
Model genre:
Reference(s):
Resolution:
Horizontal-
Vertical- -
Temporal - -
Advection:
Diffusion:
Met. parameters:
multilayer Eulerian (no-cloud version)
Venkatram et al. (1988); Olson et al. (1989)
~ 100 km
11 unevenly-spaced layers from 0 to 10 km with
greater resolution in the planetary
boundary layer
1 h
numerical solution of advection-diffusion
equation using Blackman cubic spline
(Yamartino and Scire, 1984)
numerical solution of advection-diffusion
equation using the Smagorinsky scheme
derived from a diagnostic module combining
information from the Canadian
Meteorological Centre's large-scale
numerical weather prediction model
(upper-boundary conditions) with
information from a high-resolution
boundary layer model
(Scholtz et al., 1986)
A-9
-------
TABLE A-9. ADPIC ATTRIBUTES
Atmospheric Diffusion Particle-In-Cell Model (ADPIC)
Model genre:
Reference(s):
Resolution:
Horizontal-
Vertical- -
Temporal - -
Advection:
multiple-layer, Lagrangian-Eulerian,
particle-in-cell
Lange (1978); Rodriguez and Peterson (1989)
381 km
150 m
dependent on the maximum wind speed across the
model domain
flux -conservative form of the advection -
dif fusion equation:
vo [X
UD)] = 0,
where x is a scalar concentration,
Uk is an advection velocity, and
UD ( = -Kvx/x ) is a diffusivity velocity
Diffusion:
Horizontal-
Vertical- -
Me t. parameters:
Kh = av • (dffy/dt),
where ay = a (ut)b
K; = (J2 • (daz/dt),
where az = c (ut)d (Pasquill, 1976)
u- and v-wind components on a 47 x 51 x 15
grid based on the polar stereographic
projection and vertical interpolation of
AFGWC's HIRAS data
A-10
-------
APPENDIX B
TIME SERIES PLOTS OF 24-H MEAN CONCENTRATIONS OF
PTCH AND PDCH TRACERS ALONG BANDS OF SITES
LIST OF FIGURES
Band
(km)
1,000
tl
11
1,600
"
H
2,300
ft
"
300
tl
It
700W
H
tf
Tracer
PTCH
"
"
II
II
H
11
It
11
PDCH
It
II
II
fl
tf
Month
Jan
Feb
Mar
Jan
Feb
Mar
Jan
Feb
Mar
Jan
Feb
Mar
Jan
Feb
Mar
Page
no.
B-2
B-3
B-4
B-5
B-6
B-7
B-8
B-9
B-10
B-ll
B-12
B-13
B-14
B-15
B-16
Band
(km) Tracer
700E PDCH
II 11
II 11
1 , OOOW
ft II
If II
l.OOOE
tt II
It II
1,400
tt tt
II tl
1,800
II tt
II tt
Month
Jan
Feb
Mar
Jan
Feb
Mar
Jan
Feb
Mar
Jan
Feb
Mar
Jan
Feb
Mar
Page
no.
B-17
B-18
B-19
B-20
B-21
B-22
B-23
B-24
B-25
B-26
B-27
B-28
B-29
B-30
B-31
B-l
-------
24-H AVERAGE CALCULATED AND MEASURED PTCH CONCENTRATIONS
ALONG THE 1000 KM BAND
JANUARY 1987
u.
D
O
O
O
MLAM FINE
MLAM COARSE
567
9 10 11 12 13 14 15 16 17 18 192021 22 23 24 25 26 27 28 29 30 31
DAY OF MONTH
B-2
-------
24-H AVERAGE CALCULATED AND MEASURED PTCH CONCENTRATIONS
ALONG THE 1000 KM BAND
FEBRUARY 1987
u.
D
O
O
O
O
123456
7 8 910111213141516171819202122232425262728
DAY OF MONTH
B-3
-------
LL
Q
O
O
O
I
O
24-H AVERAGE CALCULATED AND MEASURED PTCH CONCENTRATIONS
ALONG THE 1000 KM BAND
MARCH 1987
MEASURED
SRL
ARL
BAT
GAMUT
HY-SPLIT
MLAM
MLAM
FINE
COARSE
ApOM
ADPIC
45678 91011121314151617181920212223242526272829
DAY OF MONTH
B-4
-------
24-H AVERAGE CALCULATED AND MEASURED PTCH CONCENTRATIONS
ALONG THE 1600 KM BAND
JANUARY 1987
50
40
30
20
10
0
50
40
30
20
10
0
50
40
30
20
10
0
50
40
30
20
10
0
50
40
30
20
10
0
50
40
30
20
10
0
MEASURED
SRL
TCAL
VCAL
LL
D
o
o
O
X
o
ARL
BAT
GAMUT
HY-SPLIT
MLAM FINE
MLAM COARSE
ADOM
ADPIC
678
9 1011 12 13 14 15 16 17 18 192021 222324.252627282930 31
DAY OF MONTH
B-5
-------
LL
Q
O
O
O
I
O
50
40
30
20
10
0
50
40
30
20
10
0
50
40
30
20
10
0
50
40
30
20
10
0
50
40
30
20
10
0
50
40
30
20
10
0
24-H AVERAGE CALCULATED AND MEASURED PTCH CONCENTRATIONS
ALONG 1600 KM BAND
FEBRUARY 1987
MEASURED
SRL
TCAL
VCAL
ARL
BAT
GAMUT
HY-SPLT
MLAM FINE
MLAM
COARSE
ADQM
ADPC
12345678
9 1011 12 131415161718192021 22232425262728
DAY OF MONTH
B-6
-------
24-H AVERAGE CALCULATED AND MEASURED PTCH CONCENTRATIONS
ALONG 1600 KM BAND
MARCH 1987
LL
Q
o
o
O
I
MLAM FIN£
MLAM COARSE
1 2 3-4 5678 91011121314151617181920212223242526272829
DAY OF MONTH
B-7
-------
50
40
30
20
10
0
50
40
30
20
10
0
50
40
30
20
§ 0
O 50
O 40
°- 30
20
10
50
40
30
20
10
0
50
40
30
20
10
0
24-H AVERAGE CALCULATED AND MEASURED PTCH CONCENTRATIONS
ALONG THE 2300 KM BAND
JANUARY 1987
MEASURED
SRL
TCAL
VCAL
ARL
BAT
GAMUT
HY-SPLIT
MLAM FINE
MLAM COARSE
ADOM
ADPIC
567
9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31
DAY OF MONTH
B-S
-------
O
§
O
I
24-H AVERAGE CALCULATED AND MEASURED PTCH CONCENTRATIONS
ALONG THE 2300 KM BAND
FEBRUARY 1987
50
40
30
20
10
0
50
40
30
20
10
0
50
40
30
20
10
0
50
40
30
20
10
0
50
40
30
20
10
0
50
40
30
20
10
0
MEASURED
SRL
TCAL
VCAL
ARL
BAT
GAMUT
HY-SPLIT
MLAM FINE
MLAM
COARSE
ADOM
AdPIC
12345678
9 1011 12131415 161718192021 22232425262728
DAY OF MONTH
B-9
-------
u.
D
O
O
O
24-H AVERAGE CALCULATED AND MEASURED PTCH CONCENTRATIONS
ALONG THE 2300 KM BAND
MARCH 1987
50
40
30
20
10
0
50
40
30
20
10
0
50
40
30
20
10
0
50
40
30
20
10
0
50
40
30
20
10
0
50
40
30
20
10
0
MEASURED
SRL
TCAL
VCAL
GAMUT
HY-SPLIT
MLAM FINE
MLAM
ARL
BAT
COARSE
ADOM
ADPIC
1234567
8 9 1011 12131415161718192021 2223242526272829
DAY OF MONTH
B-10
-------
24-H AVERAGE CALCULATED AND MEASURED PDCH CONCENTRATIONS
ALONG THE 300 KM BAND
JANUARY 1987
MLAM FINE
MLkM COARSE
7 8
9 10 11 12 13 14 15 16 17 18 192021 2223 24 2526 27 2829 30 31
DAY OF MONTH
B-ll
-------
o
o
Q
a.
40
30
20
10
0
50
40
30
20
10
0
50
40
30
20
10
24-H AVERAGE CALCULATED AND MEASURED PDCH CONCENTRATIONS
ALONG THE 300 KM BAND
FEBRUARY 1987
TCAL
VCAL
GAMUT
HY-SPLT
2345678
9 1011 12 131415161718192021 22232425262728
DAY OF MONTH
B-12
-------
24-H AVERAGE CALCULATED AND MEASURED PDCH CONCENTRATIONS
ALONG THE 300 KM BAND
MARCH 1987
LL
Q
O
O
O
O
Q
Q_
23-45678
9 1011 12131415161718192021 2223242526272829
DAY OF MONTH
B-13
-------
LL
Q
O
O
O
I
O
O
CL
50
40
30
20
10
0
50
40
30
20
10
0
50
40
30
20
10
0
50
40
30
20
10
0
50
40
30
20
10
0
50
40
30
20
10
0
24-H AVERAGE CALCULATED AND MEASURED PDCH CONCENTRATIONS
ALONG THE 700W KM BAND
JANUARY 1987
MEASURED
SRL
TCAL
VCAL
ARL
BAT
GAMUT
HY-SPLIT
MLAM FINE
MLAM COARSE
ADOM
ADPIC
5678 910111213141516171819202122232425262728293031
DAY OF MONTH
B-14
-------
24-H AVERAGE CALCULATED AND MEASURED PDCH CONCENTRATIONS
ALONG THE 700W KM BAND
FEBRUARY 1987
u.
0
O
O
O
I
O
D
Q_
123456
7 8 910111213141516171819202122232425262728
DAY OF MONTH
B-15
-------
u_
Q
O
Z
O
O
I
O
0
Q_
50
40
30
20
10
0
50
40
30
20
10
0
50
40
30
20
10
0
50
40
30
20
10
0
50
40
30
20
10
0
50
40
30
20
10
0
24-H AVERAGE CALCULATED AND MEASURED PDCH CONCENTRATIONS
ALONG THE 700W BAND
MARCH 1987
MEASURED
SRL
TCAL
YCAL
GAMUT
\HY-SPLIT
MLAM FINE
MLAM
ARL
BAT
COARSE
ADOM
ADPIC
12345678 91011121314151617181920212223242526272829
DAY OF MONTH
B-16
-------
u.
Q
O
o
o
o
Q
CL
50
40
30
20
10
0
50
40
30
20
10
40
30
20
10
0
50
40
30
20
10
0
50
40
30
20
10
24-H AVERAGE CALCULATED AND MEASURED PDCH CONCENTRATIONS
ALONG THE 700E KM BAND
JANUARY 1987
MEASURED
SRL
TCAL
VCAL
GAMUT
HY-SPLIT
MLAM FINE
MLAM COARSE
5678 910111213141516171819202122232425262728293031
DAY OF MONTH
B-17
-------
24-H AVERAGE CALCULATED AND MEASURED PDCH CONCENTRATIONS
ALONG THE 700E KM BAND
FEBRUARY 1987
u.
Q
O
O
O
O
Q
Q.
MLAM FINE
MLAM COARSE
12345
678 910111213141516171819202122232425262728
DAY OF MONTH
B-18
-------
24-H AVERAGE CALCULATED AND MEASURED PDCH CONCENTRATIONS
ALONG THE 700E KM BAND
MARCH 1987
LL
D
O
o
O
8
D.
23456
7 8 91011121314151617181920212223242526272829
DAY OF MONTH
B-19
-------
U-
D
O
O
O
O
Q
D.
50
40
30
20
10
0
50
40
30
20
10
0
50
40
30
20
10
0
50
40
30
20
10
0
50
40
30
20
10
0
50
40
30
20
10
0
24-H AVERAGE CALCULATED AND MEASURED PDCH CONCENTRATIONS
ALONG THE 1000W KM BAND
JANUARY 1987
MEASURED
SRL
TCAL
VCAL
ARL
BAT
GAMUT
HY-SPLIT
MLAM FINE
MLAM COARSE
ADOM
ADPIC
5678 910111213141516171819202122232425262728293031
DAY OF MONTH
B-20
-------
24-H AVERAGE CALCULATED AND MEASURED PDCH CONCENTRATIONS
ALONG THE 1000W KM BAND
FEBRUARY 1987
ou
40
30
20
10
50
40
30
20
10
50
40
30
3 20
D 10
o
Zf\
_ 0
8 50
0 40
0
a- 30
20
10
40
30
20
10
40
30
20
10
n
-
-
—
—
— /\
-
—
~
—
-
—
—
-
—
-
L
—
—
-
(
_
-
—
—
—
-
-
—
—
*
m
/ \
/•-~^_
/ ^--— ^
__*— ^ T t- ..
•y
1
1
/
y
,«.
%^x. -">-
\
\
\ \
/ ••*•
/
^^—, l^ ,
MEASL
SRL
T
V
RED
^^
hAi
:AL
API
BAT
_^ _^_ .*-• ••
GAMUT
HY-SPLJT
__x
MLAM FIN^
MLAM COA
• • •
AD
Ad
RSE
H- — «
DM
PIC
123456
78 910111213141516171819202122232425262728
DAY OF MONTH
B-21
-------
24-H AVERAGE CALCULATED AND MEASURED PDCH CONCENTRATIONS
ALONG THE 1000W KM BAND
MARCH 1987
ou
40
30
20
10
0
50
40
30
20
10
0
50
40
30
^ 20
LL
0 10
z o
8 50
0 40
Q
o- 30
20
10
0
50
40
30
20
10
0
50
40
30
20
10
n
—
-
— \
—
—
-
'"*.
-
-
—
- .
—
- \
— \ ^^^
-
—
-
' *
—
—
—
—
• /
••-••' "T ""?
;* /
• """ "•* •- ,•
t'
A
\
V.
\-'
*••
'"•-
\
" • •
''•-.
—-t~-~r—r^~~~i
' •-• ..
• .._
MEA
SRL
— ^T"
~1
I.
-.^/'
__x<
Gfi
M
A
/ / •.
SURED
TpAl
VSJAL
\
-*-.^\
'" "^
Ani
BAT
^
MUT
-SPLIT
\^
MLAMF
MLAM.C
•
NE '
3ARSE.
*
•
AfM-MH
ADPIC
'.
'• • .
123456
78 91011121314151617181920212223242526272829
DAY OF MONTH
B-22
-------
u_
Q
O
O
O
O
D
Q_
50
40
30
20
10
0
50
40
30
20
10
0
50
40
30
20
10
0
50
40
30
20
10
24-H AVERAGE CALCULATED AND MEASURED PDCH CONCENTRATIONS
ALONG THE 1000E KM BAND
JANUARY 1987
MLAM FINE
MLAM COARSE
I51EASURED
TCAL
VCAL
ARL
BAT
GAMUT
HY-SPLIT
7 8
910111213 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31
DAY OF MONTH
B-23
-------
LL
Q
O
o
O
8
0_
50
40
30
20
10
0
50
40
30
20
10
0
50
40
30
20
10
0
50
40
30
20
10
0
50
40
30
20
10
0
50
40
30
20
10
0
24-H AVERAGE CALCULATED AND MEASURED PDCH CONCENTRATIONS
ALONG THE 1000E KM BAND
FEBRUARY 1987
MEASURED
SRL
TCAL
VCAL
BAT
GAMUT
HY-SPLIT
MLAM FINE
MLAM COARSE
AQOM
ADPIC
123456
7 8 910111213141516171819202122232425262728
DAY OF MONTH
B-24
-------
LL
D
O
Z
O
O
O
D
n_
50
40
30
20
10
0
50
40
30
20
10
0
50
40
30
20
10
0
50
40
30
20
10
0
50
40
30
20
10
0
50
40
30
20
10
0
24-H AVERAGE CALCULATED AND MEASURED PDCH CONCENTRATIONS
ALONG THE 1000E KM BAND
MARCH 1987
MEASURED
SRLJ
MLAMI
MLAMi
ARL
BAT
GAMUT
HY-SPLIT
FfNE
COARSE
ADOM
1 2 3 "4 5 6
7 8 91011121314151617181920212223242526272829
DAY OF MONTH
B-25
-------
24-H AVERAGE CALCULATED AND MEASURED PDCH CONCENTRATIONS
ALONG THE 1400 KM BAND
JANUARY 1987
u_
D
O
O
O
O
D
D.
N LAM FINE
\ LAM COARSE
7 8
9 10 11 12 13 14 15 16 17 18 192021 2223242526272829 30 31
DAY OF MONTH
B-26
-------
o
o
D
D.
50
40
30
20
10
0
50
40
30
20
10
0
50
40
30
20
10
50
40
30
20
10
0
50
40
30
20
10
0
50
40
30
20
10
0
24-H AVERAGE CALCULATED AND MEASURED PDCH CONCENTRATIONS
ALONG THE 1400 KM BAND
FEBRUARY 1987
MEASURED
SRL
TPAL
VCAL
GAMUT
HY-SPLIT
ARL
feAT
MLAM FIN£
MLAM COARSE
AQOM
ADPIC
12345678
9 1011 12131415161718192021 22232425262728
DAY OF MONTH
B-27
-------
50
40
30
20
10
0
50
40
30
20
10
0
50
40
30
!} 20
0 10
o
o
o
o
O
Q.
50
40
30
20
10
0
50
40
30
20
10
0
50
40
30
20
10
0
24-H AVERAGE CALCULATED AND MEASURED PDCH CONCENTRATIONS
ALONG THE 1400 KM BAND
MARCH 1987
' .
MEASURED.
SRL
MI-AM i
MLAM
TCAL
VCAL
ARL
BAT
GAMUT
HY-SPLIT
FNE
COARSE
ADOM
ADPIC
2345678 91011121314151617181920212223242526272829
DAY OF MONTH
B-28
-------
O
z
O
O
O
D
D_
50
40
30
20
10
0
50
40
30
20
10
0
50
40
30
20
10
0
50
24-H AVERAGE CALCULATED AND MEASURED PDCH CONCENTRATIONS
ALONG THE 1800 KM BAND
JANUARY 1987
40 -
30
20
10
MLAM FINE
M'LAM COARSE
MEASURED
SRL
TCAL
VCAL
ARL
BAT
GAMUT
HY-SPLIT
7 8
91011 1213 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31
DAY OF MONTH
B-29
-------
50
40
30
20
10
0
50
40
30
20
10
0
50
40
30
*J 20
fe 10
O
z
O
O
X
O
Q
Q_
50
40
30
20
10
50
40
30
20
10
0
50
40
30
20
10
0
24-H AVERAGE CALCULATED AND MEASURED PDCH CONCENTRATIONS
ALONG THE 1800 KM BAND
FEBRUARY 1987
MEASURED
SRL
TCAL
VCAL
ARL
BAT
GAMUT
HY-SPLIT
MLAM FINE
MLAM
COARSE
ADOM
AQPIC
12345678
9 1011 12131415161718192021 22232425262728
DAY OF MONTH
B-30
-------
24-H AVERAGE CALCULATED AND MEASURED PDCH CONCENTRATIONS
ALONG THE 1800 KM BAND
MARCH 1987
ou
40
30
20
10
0
40
30
20
10
0
50
40
30
^ 20
0 10
Z1 0
0 50
5 40
g 30
20
10
0
50
40
30
20
10
0
40
30
20
10
n
—
—
-
~ ./_
—
—
—
—
-
—
-
—
—
r-/v.
—
—
—
—
—
—
—
,-•' '•» . .
..»'
•-^-l^-^*^~^-^
• • • * .
" ,9 * •••<
L, . . t
".' " '* • •
.— * .-•— . * - 1 I
-
""*" '"* '• •-•
-~^t
MEA
SRL
G^
M
MLAMF
MU\MC
. . •
SURED
f^^"* * ~~t
TPAI
VCA.L
^^
ARI
BAT
MUT
-SPLIT •
..- \
NE
DARSE
* • r
ADHM
ADPIC
•t * •*
1 23-45678 91011121314151617181920212223242526272829
DAY OF MONTH
B-31
-------
TABLE B-l. LIST OF 24-H MEAN CONCENTRATIONS EXCEEDING 50 dfL/L
Band
(km) Tracer
1 , 000 PTCH
tt tf
It tt
II It
tl II
tt It
tt tt
II tt
It tt
tt tt
1 , 600
300 PDCH
II tt
tt It
tl II
700W
tt tt
tt tt
II tt
700E
tt II
tt tl
tt tt
1 , OOOW
II II
1000E
tt It
Page
no .
B-2
"
ii
B-3
tt
II
B-4
tt
tt
tt
B-6
B-10
B-ll
tt
II
B-14
B-15
"
tt
B-17
It
B-18
1!
B-19
B-21
B-22
ii
Date
Jan
Jan
Jan
Feb
Feb
Mar
Mar
Mar
Mar
Mar
Feb
Jan
Feb
Feb
Feb
Feb
Mar
Mar
Mar
Feb
Feb
Mar
Mar
Jan
Mar
Jan
Jan
8
26
28
4
5
22
2
4
5
21
7
6
2
3
4
28
9
14
23
2
5
28
29
19
25
9
20
Model
MLAM-
SRL TCAL VCAL, COARSE ADOM ADPIC
65
54
57
53
69
76
73
54
51
54
50
57
66
53.0 70.9
55.9
.8
.6
.5
56.0
50.5
55.0
51.9
62.8
.4
64.8 78.1
.3
.0
.2
51.4
.7
55.0
.0
.0
.8
58.0
51.8
.6
56.8
56.8
.7
*U.S. GOVERNMENT PRINTING OFFICE; 19 9 0 .7«. e. i* » 2 0 * s n
B-32
------- |