United States
Environmental Protection
Agency
Office of Air Quality
Planning and Standards
Research Triangle Park NC 27711
EPA-450/4-86-016a
October 1986
Air
Evaluation of
Short-Term
Long-Range
Transport Models
Volume I.
Analysis
Procedures
and Results
-------
-------
EPA-450/4-86-016a
Evaluation of Short-Term Long-Range
Transport Models
Volume I. Analysis Procedures and Results
by
A. J. Policastro, M. Wastag, L. Coke
Argonne National Laboratory
Argonne, Illinois 60439
R. A. Carhart
University of Illinois
Chicago, Illinois 60680
W. E. Dunn
University of Illinois
Urbana, Illinois 61801
IAG No. DW 89930807
EPA Project Officer: Norman C. Possiel
Prepared for
U.S. ENVIRONMENTAL PROTECTION AGENCY
Office of Air and Radiation
Office of Air Quality Planning and Standards
Source Receptor Analysis Branch
Research Triangle Park, NC 27711
October 1986 U.S. En^^ental Protection A*er.cr
Eegion 5, - • ' •' ' . ,„
230 S. Le^^-^ ' • -*t, fi'^n ^ '
Cbicago, 1L 6U604
-------
-------
Disclaimer
This report has been reviewed by The Office of Air Quality
Planning and Standards, U. S. Environmental Protection
Agency, and has been approved for publication. Mention
of trade names or commercial products is not intended to
constitute endorsement or recommendation for use.
ii
-------
ABSTRACT
Eight short-term long-range transport models have been evaluated with
field data from two data bases involving tracer releases. The models tested
were MESOPUFF, MESOPLUME, MSPUFF, MESOPUFF II, MTDDIS, ARRPA, RADM, and
RTM-II. The Oklahoma data base encompasses two separate experiments with
measurements taken at 100 km and 600 km arcs downwind of a three-hour
perfluorocarbon tracer release. Time averages of 45 minutes for the 100 km
arc and 3 hours for the 600 km arc were obtained in the data. The spatial
detail along these arcs was very good providing an opportunity to evaluate
such predicted plume features as transport time, horizontal plume spreading,
and predicted/observed pattern offsets. The Savannah River Plant data base
encompasses 15 experiments with measurements taken over 2-5 days at distances
28-144 km downwind. Ten-hour time averages of the krypton-85 tracer were
taken at ground level. This data base involved a longer period of record and
meteorology from four seasons, yet included a fewer number of fixed samplers
than the Oklahoma experiments.
Model performance was evaluated by graphical and statistical methods.
The primary means of evaluating the performance of the models was the use of
the American Meteorological Society (AMS) statistics. These statistics
provided quantitative measures of model performance. Supplementary measures
included the use of isopleth plots of ground-level concentrations, scatter
plots, cumulative frequency distributions and frequency histograms of
residuals.
Model .performance was generally consistent between the two data bases.
General features of the predicted ground patterns included: (a) spatial offset
of predicted and observed patterns, (b) a time difference between the arrival
of the predicted and observed plumes at a particular receptor, (c) a definite
angular offset of predicted and observed plumes generally due to specific
assumptions made in the meteorological preprocessor. This angular offset is
frequently as much as 20-45 degrees. The ARRPA and MTDDIS models appear to
provide the most accurate trajectories of all eight models. The ARRPA model
employs the Boundary Layer Model of the National Weather Service for its
meteorological preprocessor, and the MTDDIS model employs wind directions at
the effective height of release. However, it should be stated that the ARRPA
111
-------
model could only be tested with the Oklahoma data base and the MTDDIS model
was applicable only to the Savannah River Plant data base. Further model
testing would be needed in order to generalize the above findings on plume
trajectory predictions.
The models also have a definite tendency to underpredict horizontal
spreading at ground level, along with a concomitant overprediction of plume
concentrations. This behavior is typical of the performance of most models.
The underprediction of horizontal spreading is usually due to the use of the
Turner curves at distances (50-100 km) far beyond their intended use.
The spatial and temporal offsets of the predicted and observed plumes
lead to predicted concentrations that correlate poorly with concentrations
observed at the same time and place. Values for standard deviation of
residuals, root mean square error, and absolute average residual are larger
than the average observed concentration for all eight models. However,
statistical comparisons of the peak values predicted by the models were
significantly better. For example, the highest 25 averaged predictions and
highest 25 averaged observations (unpaired in location and time) were within a
factor of two of each other for six of the eight models tested (MESOPUFF,
MESOPLUME, MESOPUFF II, MTDDIS, ARRPA, andRTM-II). Again, the ARRPA and
MTDDIS models were tested with only one data base. The observed tendency to
overpredict peak concentrations errs on the conservative side for regulatory
applications. However, this overprediction must be weighed against the
general tendency of the models to underpredict horizontal spreading and to
predict a pattern that is spatially offset from the data.
IV
-------
ACKNOWLEDGMENTS
The authors would like to thank Mr. Norman Possiel and Mr. Joseph Tikvart
of the U.S. EPA for their support and direction of this model evaluation work.
Their structuring of the project to permit periodic reviews of project
accomplishments by the model developers was important to the success of this
work.
The authors are also appreciative of the comments received from the
modelers and the discussions with them concerning the theoretical formulation
and performance of their models. We wish to thank:
Mr. Joseph Scire
Environmental Research and Technology
(now at Sigma Research, Inc.)
Mr. Martin Schock
Mr. Steven Weber
North Dakota Department of Health
Dr. I-Tung Wang
Combustion Engineering, Inc.
Dr. Steve Mueller
Dr. Tim Crawford
Tennessee Valley Authority
Mr. Doug Stewart
Mr. Gary Moore
Mr. Ralph Morris
— Systems Applications, Inc.
-------
TABLE OF CONTENTS
VOLUME I
ABSTRACT
ACKNOWLEDGMENTS V
TABLE OF CONTENTS vii
LIST OF FIGURES xiii
LIST OF TABLES xix
SECTION 1 INTRODUCTION 1-1
1.1. BACKGROUND 1-1
1.2. MODEL EVALUATION PROTOCOL 1-2
1.3. DEFINITION OF PROJECT TASKS 1-3
1.4. ORGANIZATION OF THIS REPORT 1-4
SECTION 2 REVIEW OF THEORETICAL FORMULATIONS OF THE MODELS 2-1
2.1. INTRODUCTION 2-1
2.2. DISCUSSION OF KEY FEATURES OF PREPROCESSOR MODEL THEORIES. ... 2-8
2.2.1. The Meteorological Preprocessor - MESOPAC 2-8
2.2.2. The Meteorological Preprocessor - MSPACK 2-9
2.2.3. The Meteorological Preprocessor - MESOPAC II 2-11
2.2.4. The Meteorological Preprocessor for MTDDIS 2-12
2.2.5. The Meteorological Preprocessor - The BLM/MDPP Model . . 2-14
2.2.6. The Application of MESOPAC II to the RADM Model 2-16
2.2.7. The Application of MESOPAC and MESOPAC II for RTM-II . . 2-17
2.3. DISCUSSION OF KEY FEATURES OF PLUME MODEL THEORIES 2-17
2.3.1. MESOPUFF Model 2-17
2.3.2. MESOPLUME Model 2-18
2.3.3. MSPUFF Model 2-21
2.3.4. MESOPUFF II Model 2-23
2.3.5. MTDDIS Model 2-25
2.3.6. ARRPA Model 2-26
2.3.7. RADM Model 2-30
2.3.8. RTM-II Model 2-31
SECTION 3 THE OKLAHOMA AND SAVANNAH RIVER PLANT DATA BASES 3-1
3.1. INTRODUCTION 3-1
3.2. METEOROLOGICAL DATA REQUIREMENTS OF THE MODELS 3-1
3.3. OKLAHOMA FIELD EXPERIMENTS 3-3
3.3.1. Description of Experiments 3-3
3.3.2. Ground Level Tracer Observations 3-4
3.3.3. The Oklahoma Modelers' Data Base 3-11
vii
-------
TABLE OF CONTENTS (CONTINUED)
3.3.3.1. The Meteorological Grid System 3-11
3.3.3.2. Surface Weather Observations /
Modelers' Data Base 3-14
3.3.3.3. Rawinsonde Observations /
Modelers' Data Base 3-14
3.3.3.4. Meteorological Tower Observations /
Modelers' Data Base 3-16
3.3.4. The Treatment of Missing Data 3-16
3.3.5. The Source Release / Modelers' Data Base 3-19
3.4. SAVANNAH RIVER PLANT KRYPTON-85 EXPERIMENTS 3-19
3.4.1. Description of Experiments 3-19
3.4.2. Ground-Level Krypton-85 Observations / Study Periods . . 3-20
3.4.3. The Savannah River Plant Modelers' Data Base 3-20
3.4.3.1. The Meteorological Grid System 3-20
3.4.3.2. Surface Weather Observations /
Modelers' Data Base 3-22
3.4.3.3. Rawinsonde Observations /
Modelers' Data Base ' 3-24
3.4.3.4. Meteorological Tower Observations /
Modelers' Data Base 3-25
3.4.3.5. The Source Release / Modelers' Data Base . . . 3-26
SECTION 4 OPERATIONAL EVALUATION BASED ON MODEL/DATA COMPARISONS ... 4-1
4.1. INTRODUCTION 4-1
4.2. APPLICATION OF AMS STATISTICS TO THE OKLAHOMA AND SRP DATA BASES 4-2
4.2.1. Introduction 4-2
4.2.2. Statistical Data Sets for Comparison of Observed
Predicted Concentrations 4-3
4.2.3. Peak Concentrations 4-5
4.2.4. Comparisons of All Concentrations 4-6
4.2.5. Statistical Analysis of Model Performance 4-7
4.2.6. Model Performance Results using AMS Statistics 4-12
4.3. GRAPHICAL EVALUATION OF THE EIGHT MODELS WITH THE OKLAHOMA
AND SAVANNAH RIVER PLANT DATA 4-23
4.3.1. Scatter Plots 4-26
4.3.2. Other Graphical Comparisons 4-36
4.3.2.1. Oklahoma Data Base 4-36
4.3.2.2. Savannah River Plant Data Base 4-42
Vlll
-------
TABLE OF CONTENTS (CONTINUED)
SECTION 5 DIAGNOSTIC EVALUATION BASED ON MODEL/DATA COMPARISONS. ... 5-1
5.1. INTRODUCTION 5-1
5.2. COMPARISON OF PREDICTED CONCENTRATION ISOPLETHS WITH
DATA AT OKLAHOMA 5-2
5.3. COMPARISON OF PREDICTED CONCENTRATION ISOPLETHS AT THE
SAVANNAH RIVER PLANT 5-19
5.4. PATTERN COMPARISON METHOD OF MODEL EVALUATION
(OKLAHOMA ONLY) 5-28
5.5. SUMMARY OF DIAGNOSTIC REVIEW OF MODEL/DATA COMPARISONS 5-34
5.5.1. MESOPUFF .' 5-40
5.5.2. MESOPLUME 5-43
5.5.3. MSPUFF 5-46
5.5.4. MESOPUFF II 5-48
5.5.5. MTDDIS 5-51
5.5.6. ARRPA 5-54
5.5.7. RADM 5-57
5.5.8. RTM-II 5-60
5.5.8.1. RTM-II Features Common to Both the Oklahoma
and the SRP Data Bases 5-61
5.5.8.2. RTM-II Features Specific to the Oklahoma
Data Base 5-62
5.5.8.3. RTM-II Features Specific to the Savannah River
Plant Data Base 5-63
SECTION 6 SUMMARY AND CONCLUSIONS 6-1
6.1 OVERVIEW 6-1
6.2. GENERAL PERFORMANCE FEATURES OF THE MODELS 6-2
6.3. QUANTITATIVE MEASURES OF MODEL PERFORMANCE 6-3
6.4. SPECIFIC PERFORMANCE FEATURES OF INDIVIDUAL MODELS 6-5
6.4.1. MESOPUFF 6-5
6.4.2. MESOPLUME 6-6
6.4.3. MSPUFF 6-6
6.4.4. MESOPUFF II 6-7
6.4.5. MTDDIS 6-7
6.4.6. ARRPA 6-8
6.4.7. RADM 6-8
6.4.8. RTM-II 6-8
6.5. OPTIONS AND PARAMETERS 6-9
REFERENCES R-l
IX
-------
TABLE OF CONTENTS (CONTINUED)
VOLUME II - APPENDICES
TABLE OF CONTENTS iii
LIST OF FIGURES vii
LIST OF TABLES xxix
APPENDIX A APCA PAPER/HARTFORD MEETING/OCTOBER 1983 A-l
APPENDIX B CHOICE OF MODEL-SPECIFIC INPUTS
'j
INTRODUCTION B-l
PART I: DATA BASE SELECTION AND INPUT PARAMETERS
FOR THE MTDDIS MODEL B-l
Oklahoma Tracer Study B-l
Savannah River Plant Krypton-85 Study. B-2
Source Information B-2
Meteorological Information B-2
Spatial and Temporal Grids B-3
Model-Specific Options B-5
PART II: DATA BASE SELECTION AND INPUT PARAMETERS
FOR THE ARRPA MODEL B-5
Oklahoma Tracer Study B-5
Source Information B-5
Meteorological Information B-6
Spatial and Temporal Grids B-6
Model-Specific Options B-6
Savannah River Plant Krypton-85 Study B-7
PART III: DESCRIPTION OF MESOPAC CHANGES AND OPTIONS B-7
PART IV: DATA BASE SELECTION AND INPUT PARAMETERS FOR THE
MESOPAC II WIND FIELD MODEL B-9
Description of MESOPAC II Options B-10
PART V: DATA BASE SELECTION AND INPUT PARAMETERS FOR THE
MESOPUFF, MESOPLUME, MSPUFF, AND MESOPUFF II MODELS . . B-13
Introduction B-13
Computational Considerations B-13
Decision Points B-14
PART VI: DESCRIPTION OF RTM-II CHANGES AND OPTIONS B-16
-------
TABLE OF CONTENTS (CONTINUED)
PART VII: DESCRIPTION OF RADM CHANGES AND OPTIONS B-17
APPENDIX C DESCRIPTION OF CODE MODIFICATIONS REQUIRED
FOR THE EIGHT MODELS
INTRODUCTION C-l
C.I. DESCRIPTION OF MTDDIS MODIFICATIONS C-l
C.2. DESCRIPTION OF ARRPA MODIFICATIONS C-3
C.3. DESCRIPTION OF MESOPAC (METEOROLOGICAL PREPROCESSOR)
MODIFICATIONS C-6
C.4. DESCRIPTION OF MESOPUFF MODIFICATIONS C-7
C.5. DESCRIPTION OF MESOPLUME MODIFICATIONS C-9
C.6. DESCRIPTION OF MSPACK (METEOROLGICAL PREPROCESSOR)
MODIFICATIONS C-ll
C.7. DESCRIPTION OF MSPUFF MODIFICATIONS C-ll
C.8. DESCRIPTION OF MESOPAC II (METEOROLOGICAL PREPROCESSOR)
MODIFICATIONS C-13
C.9. DESCRIPTION OF MESOPUFF II MODIFICATIONS C-17
C.10. RTM-II MODIFICATIONS (OKLAHOMA ONLY) C-18
C.ll. DESCRIPTION OF RADM MODIFICATIONS C-20
APPENDIX D COMPLETE STATISTICAL COMPARISONS OF THE EIGHT MODELS
WITH THE OKLAHOMA AND SAVANNAH RIVER PLANT DATA BASES
INTRODUCTION D-l
PART I: TABULAR LISTING OF THE PREDICTED AND OBSERVED GROUND-LEVEL
CONCENTRATIONS FOR EACH OF THE SRP AND OKLAHOMA DATA SETS .... D-2
PART II: PRESENTATION OF COMPLETE AMS STATISTICS RESULTS FOR THE
OKLAHOMA AND SRP DATA SETS D-23
APPENDIX E COMPLETE GRAPHICAL COMPARISONS OF THE EIGHT MODELS
WITH THE OKLAHOMA AND SAVANNAH RIVER PLANT DATA BASES
INTRODUCTION E-l
1. LOCATION OF AIR SAMPLERS AND SITE MAP FOR OKLAHOMA
EXPERIMENTS E-3
2. EVIDENCE OF LOW-LYING NOCTURNAL JET DURING THE
OKLAHOMA STUDY E-7
3. ISOPLETH PLOTS OF PREDICTED GROUND-LEVEL CONCENTATIONS FOR THE
OKLAHOMA EXPERIMENTS E-18
4. SUMMARY GRAPHICAL PLOTS COMPARING MODEL PREDICTIONS AND
FIELD DATA AT OKLAHOMA E-103
XI
-------
TABLE OF CONTENTS (CONTINUED)
5. LOCATION OF AIR SAMPLERS AND SITE MAP FOR SAVANNAH RIVER
EXPERIMENT E-132
6. SKETCHES OF PREDICTED AND OBSERVED GROUND-LEVEL CONCENTRATIONS
FOR TWO SAVANNAH RIVER PLANT EXPERIMENTS SUBCASES 4B AND 6C . E-134
7. SUMMARY GRAPHICAL PLOTS COMPARING MODEL PREDICTIONS AND FIELD
DATA AT SAVANNAH RIVER PLANT E-143
XI1
-------
LIST OF FIGURES
Figure Page
2-1 Schematic representation of puff superposition approach .... 2-2
2-2 Schematic representation of segmented plume approach 2-2
2-3 Sample meteorological, computational, and sampling grid .... 2-4
3-1 Location of the sequential air samplers (BATS) and aircraft
sampling path at 100 km from the Oklahoma tracer release site . 3-5
3-2 Location of sequential samplers (BATS), LASL samplers, and
aircraft sampling flight path at 600 km from the Oklahoma
tracer release site. The locations of rawinsonde stations are
also shown. 3-5
3-3 Average 45-min perfluorocarbon (PMCH) concentrations along the
100 km arc from the Oklahoma experiment No. 1 (July 8, 1980) . 3-6
3-4 Average 3-hour perfluorocarbon (PMCH) concentration along the
600 km arc for the Oklahoma experiment No. 1 (July 8, 1980). . 3-7
3-5 Average 3-hour PMCH concentrations along the 600 km arc for the
period July 9, 0800 GMT to July 11, 2000 GMT ... Oklahoma
experiment No. 1 (July 8, 1980) 3-8
3-6 Average 45-min PMCH concentrations along the 100 km arc from
the Oklahoma experiment No. 2 (July 11,1980) 3-10
3-7 Location of significant source, meteorological, and receptor
sites for the Oklahoma Cases No. 1 and 2. . 3-12
3-8 Location of significant source, meteorological, and receptor
sites for the Savannah River Plant data base (includes final
region choice; inner box is model solution region) 3-23
4-1 Scatter plot of observed and MESOPUFF predicted concentrations
at Oklahoma (points paired in space and time) 4-27
4-2 Scatter plot of observed and MESOPLUME predicted concentrations
at Oklahoma (points paired in space and time) 4-27
4-3 Scatter plot of observed and MSPUFF predicted concentrations at
Oklahoma (points paired in space and time) 4-28
4-4 Scatter plot of observed and MESOPUFF II predicted
concentrations at Oklahoma (points paired in space and time) . 4-28
4-5 Scatter plot of observed and ARRPA predicted concentrations at
Oklahoma (points paired in space and time) 4-29
Xlll
-------
LIST OF FIGURES (CONTINUED)
Figure Page
4-6 Scatter plot of observed and RADM predicted concentrations at
Oklahoma (points paired in space and time) 4-29
4-7 Scatter plot of observed and RTM-II predicted concentrations at
Oklahoma (points paired in space and time) 4-30
4-8 Scatter plot of observed and MESOPUFF predicted concentrations
at Savannah River Plant (points paired in space and time) . . . 4-31
4-9 Scatter plot of observed and MESOPLUME predicted concentrations
at Savannah River Plant (points paired in space and time). . . 4-31
4-10 Scatter plot of observed and MSPUFF predicted concentrations at
Savannah River Plant (points paired in space and time) ... . 4-32
4-11 Scatter plot of observed and MESOPUFF II predicted
concentrations at Savannah River Plant (points paired in space
and time) 4-32
4-12 Scatter plot of observed and MTDDIS predicted concentrations at
Savannah River Plant (points paired in space and time) . ... 4-33
4-13 Scatter plot of observed and RADM predicted concentrations at
Savannah River Plant (points paired in space and time) . ... 4-33
4-14 Scatter plot of observed and RTM-II predicted concentrations at
Savannah River Plant (points paired in space and time) . ... 4-34
4-15 Frequency distribution of predicted and observed concentrations
at Oklahoma for MESOPLUME based on points paired in space and
time ... concentration range: 0 to 100 parts per 1015 4-37
4-16 Cumulative frequency distributions of MESOPLUME predictions and
observed concentrations at Oklahoma based on points paired in
space and time 4-37
4-17 Frequency distribution of residuals at Oklahoma for MESOPLUME
based on points paired in space and time ... residual range:
-100 to 100 parts per 1015 4-38
4-18 Frequency distribution of predicted concentrations at Oklahoma
for MESOPUFF based on points paired in space and time ...
concentration range: 0 to 100 parts per lO1^ 4-38
4-19 Frequency distribution of predicted and observed concentrations
at Oklahoma for MESOPUFF II based on pints paired in space and
time ... concentration range: 0 to 100 parts per 1015 4-39
xiv
-------
LIST OF FIGURES (CONTINUED)
Figure Page
4-20 Frequency distribution of predicted and observed concentrations
at Oklahoma for RTM-II based on points paired in space and time
... concentration range: 0 to 100 parts per 1015 4-39
4-21 Cumulative frequency distributions of RTM-II predictions and
observed concentrations at Oklahoma based on points paired in
space and time 4-40
4-22 Frequency distribution of residuals at Oklahoma for RTM-II
based on points paired in space and time ... residual range:
-100 to 100 parts per 1015 4-40
4-23 Frequency distribution of predicted and observed concentrations
at Savannah River Plant for MESOPLUME based on points paired in
space and time ... concentration range: 0 to 100 pCi/m3 .... 4-43
4-24 Cumulative frequency distributions of MESOPLUME predictions and
observed concentrations at Savannah River Plant basd on
points paired in space and time 4-43
4-25 Frequency distribution of residuals at Savannah River Plant for
MESOPLUME based on points paired in space and time ... residual
range: -100 to 100 pCi/m3 4-44
4-26 Frequency distribution of predicted and observed concentrations
at Savannah River Plant for MTDDIS based on points paired in
space and time ... concentration range: 0 to 100 pCi/m3 .... 4-44
4-27 Cumulative frequency distribution of MTDDIS predictions and
observed concentrations at Savannah River Plant based on
points paired in space and time 4-45
4-28 Frequency distribution of residuals at Savannah River Plant for
MTDDIS based on points paired in space and time ... residual
range: -100 to 100 pCi/m3 4-45
4-29 Frequency distribution of predicted and observed concentrations
at Savannah River Plant for MESOPUFF II based on points paired
in space and time ... concentration range: 0 to 100 pCi/m3 . . 4-46
4-30 Frequency distribution of predicted and observed concentrations
at Savannah River Plant for RTM-II based on points paired in
space and time ... concentration range: 0 to 100 pCi/m3. . . . 4-46
5-1 Isopleth plot of ground-level concentrations for the Oklahoma
experiment of July 8, 1980 (2230 to 2315 GMT) ... (left)
MESOPUFF predictions, (right) MESOPLUME predictions 5-3
xv
-------
LIST OF FIGURES (CONTINUED)
Figure Page
5-2 Isopleth plot of ground-level concentrations for the Oklahoma
experiment of July 8, 1980 (2230 to 2315 GMT) ... (left) MSPUFF
predictions, (right) MESOPUFF II predictions 5-4
5-3 Isopleth plot of ground-level concentrations for the Oklahoma
experiment of July 8, 1980 (2230 to 2315 GMT) ... (left) ARRPA
predictions, (right) RADM predictions 5-5
5-4 Isopleth plot of ground-level concentrations for the Oklahoma
experiment of July 8, 1980 (2230 to 2315 GMT) ... RTM-II
predictions 5-6
5-5 Isopleth plot of ground-level concentrations for the Oklahoma
experiment of July 9, 1980 (0800 to 1100 GMT) ... (left)
MESOPUFF predictions, (right) MESOPLUME predictions 5-7
5-6 Isopleth plot of ground-level concentrations for the Oklahoma
experiment of July 9, 1980 (0800 to 1100 GMT) ... (left) MSPUFF
predictions, (right) MESOPUFF II predictions 5-8
5-7 Isopleth plot of ground-level concentrations for the Oklahoma
experiment of July 9, 1980 (0800 to 1100 GMT) ... (left) ARRPA
predictions, (right) RADM predictions 5-9
5-8 Isopleth plot of ground-level concentrations for the Oklahoma
experiment of July 9, 1980 (0800 to 1100 GMT) ... RTM-II
predictions 5-10
5-9 Isopleth plot of ground-level concentrations for the Oklahoma
experiment of July 12, 1980 (0315 to 0400 GMT) „.. (left)
MESOPUFF predictions, (right) MESOPLUME predictions 5-11
5-10 Isopleth plot of ground-level concentrations for the Oklahoma
experiment of July 12, 1980 (0315 to 0400 GMT) ... (left)
MSPUFF predictions, (right) MESOPUFF II predictions 5-12
5-11 Isopleth plot of ground-level concentrations for the Oklahoma
experiment of July 12, 1980 (0315 to 0400 GMT) ... (left) ARRPA
predictions, (right) RADM predictions 5-13
5-12 Isopleth plot of ground-level concentrations for the Oklahoma
experiment of July 12, 1980 (0315 to 0400 GMT) ... RTM-II
predictions ...... 5-14
5-13 Comparison of 10-hour averages of predicted plume and observed
data (in pCi/m^) for Savannah River Plant experiment of
November 18-19, 1976 (2200 to 0800 GMT) ... (top) MESOPUFF
predictions, (bottom) MESOPLUME predictions 5-20
xvi
-------
LIST OF FIGURES (CONTINUED)
Figure Page
5-14 Comparison of 10-hour averages of predicted plume and observed
data (in pCi/m3) for Savannah River Plant experiment of
November 18-19, 1976 (2200 to 0800 GMT) ... (top) MSPUFF
predictions, (bottom) MESOPUFF II predictions 5-21
5-15 Comparison of 10-hour averages of predicted plume and observed
data (in pCi/m3) for Savannah River Plant experiment of
November 18-19, 1976 (2200 to 0800 GMT) ... (top) MTDDIS
predictions, (bottom) RADM predictions 5-22
5-16 Comparison of 10-hour averages of predicted plume and observed
data (in pCi/m3) for Savannah River Plant experiment of
November 18-19, 1976 (2200 to 0800 GMT) ... RTM-II predictions. 5-23
5-17 Comparison of 10-hour averages of predicted plume and observed
data (in pCi/m3) for Savannah River Plant experiment of
February 17, 1977 (2200 to 0800 GMT) ... (top) MESOPUFF
predictions, (bottom) MESOPLUME predictions 5-24
5-18 Comparison of 10-hour averages of predicted plume and observed
data (in pCi/m3) for Savannah River Plant experiment of
February 17, 1977 (2200 to 0800 GMT) ... (top) MSPUFF
predictions, (bottom) MESOPUFF II predictions 5-25
5-19 Comparison of 10-hour averages of predicted plume and observed
data (in pCi/m3) for Savannah River Plant experiment of
February 17, 1977 (2200 to 0800 GMT) ... (top) MTDDIS
predictions, (bottom) RADM predictions 5-26
5-20 Comparison of 10-hour averages of predicted plume and observed
data (in pCi/m3) for Savannah River Plant experiment of
February 17, 1977 (2200 to 0800 GMT) ... RTM-II predictions . . 5-27
xvn
-------
-------
LIST OF TABLES
Table Page
2-1 Comparison of Major Features of Meteorological Preprocessors. . 2-5
2-2 Comparison of Major Features of Plume Models 2-6
3-1 Summary of Meteorological (and Land Use) Data Required by the
Eight Long-Range Transport Models 3-2 -
3-2 Timing Schedule of Releases, Ground-Level Samples, and
Computational Periods for Oklahoma Cases No. 1 and 2 3-15
3-3 Availability and Periods of Record for the Eight Rawinsonde
Locations for Oklahoma Cases No. 1 and 2 3-17
3-4 Sample Collection Periods and Model Calculational Periods for
10-Hr Samples ... Representing 15 Data Sets for the Savannah
River Plant Krypton-85 Data Base 3-21
4-1 Summary of Data Sets for Use with AMS Statistics (Examples
Given for SRP Data Base Only) 4-4
4-2 Statistics Recommended by AMS for Application to Data Sets
Representing All-Concentration Comparisons 4-8
4-3 Statistics Recommended by AMS for Application to Data Sets
Representing Peak Values Unpaired (25 Highest) 4-9
4-4 Statistical Estimators and Basis for Confidence Limits for
General Performance Measures Recommended by AMS 4-10
4-5 Statistical Data Set (A-l) for Oklahoma Data (parts per
1015 . (Compares highest observed value for each event with
highest prediction for same event, paired in time, not
location) 4-13
4-6 Statistical Data Set (A-l) for Savannah River Plant Data
(pCi/m^). (Compares highest observed value for each event
with highest prediction for same event, paired in time, not
location) 4-14
4-7 Statistical Data Set (A-2) for Oklahoma Data (parts per 1015).
(Compares highest observed value over all 21 experimental
periods at each monitoring station with the highest prediction
for those 21 experimental periods at the same station, paired
in location, not time) 4-17
xix
-------
LIST OF TABLES (CONTINUED)
Table Page
4-8 Statistical Data Set (A-2) for Savannah River Plant Data
(pCi/m^). (Compares highest observed value over all 65
experimental periods at each monitoring station with the
highest prediction for those 65 experimental periods at the
same station, paired in location, not time) . . 4-18
4-9 Statistical Data Set (A-4a) for Oklahoma Data .... (parts per
1()15) (Compares highest N (=25) observed and highest N
predicted values, regardless of time or location.) 4-20
4-10 Statistical Data Set (A-4a) for Savannah River Plant Data
(pCi/m^). (Compares highest N (=25) observed and highest N
predicted values, regardless of time or location.) 4-21
4-11 Statistical Data Set (B-3) for Oklahoma Data (parts per 1015)
(Compares observed and predicted values at all stati ons,
paired in time and location.) 4-24
4-12 Statistical Data Set (B-3) for Savannah River Plant Data
(pCi/m3) (Compares observed and predicted values at all
stations, paired in time and location.) 4-25
4-13 Comparison of Predictions of the Models Based on Percentage of
Pairs where Observed Values are Greater than Predicted Values
(points paired in space and time) 4-35
4-14 Summary of Results of Graphical Comparisons of Models with
Oklahoma Data 4-41
4-15 Summary of Results of Graphical Comparisons of Models with
Savannah River Plant Data 4-47
5-1 Frequency Distribution of Predicted and Observed Concentrations
for the Two Oklahoma Cases (Based on Predicted/Observed
Pairings in Space and Time) in parts per 1015 5-18
5-2 Frequency Distribution of Predicted and Observed Concentrations
for the 15 Savannah River Plant Cases (Based on
Predicted/Observed Pairings in Space and Time) in pci/m3. . . . 5-29
5-3 Pattern Comparison Results for the 100 km Arc ... July 8, 1980;
Oklahoma Experiment 5-32
5-4 Pattern Comparison Results for the 600 km Arc ... July 8, 1980;
Oklahoma Experiment 5-32
5-5 Pattern Comparison Results for the 100 km Arc ... July 11,
1980; Oklahoma Experiment 5-32
xx
-------
LIST OF TABLES (CONTINUED)
Table Page
5-6 Summary of Peak Predicted Concentrations for Oklahoma Data
Sets, July 1980 (parts/lOiS) 5-36
5-7 Summary of Peak Predicted Concentrations for Savannah River
Plant Data Sets, pCi/m3 5-37
5-8 Classification of Model Prediction Types for the Savannah River
Plant 5-39
5-9 Mixing height at plume center (rounded to 100 m) predicted by
MESOPAC II for the first release at Oklahoma 5-62
xxi
-------
-------
SECTION 1
INTRODUCTION
1.1. BACKGROUND
«
The U.S. Environmental Protection Agency (EPA) is currently evaluating
the performance of models in several categories for the purpose of updating
EPA's "Guideline on Air Quality Models".1 Long-range transport models are
needed in calculations made at distances beyond about 50 km, the limits of
applicability of straight-line Gaussian models. Argonne National Laboratory
(ANL) and the University of Illinois (UI) have been under contract with EPA to
assist in the evaluation of the eight long-range transport models submitted to
EPA in response to the March 27, 1980 Federal Register notice.2 The
evaluation was conducted by testing the models against field data with
performance defined largely through the use of statistical measures
recommended by the American Meteorological Society (AMS).3 Additional
statistical and graphical techniques were used to supplement the AMS
statistics in the performance evaluation.
The eight models submitted were all short-term long-range transport
models. These models predict short-term averages (3, 12, 24 hr) of S02,
sulfates, and particulates over distances of 20-2000 km from single or
multiple sources. The EPA required that data bases used to test the models
represent single sources in which concentrations above background can be
detected over mesoscale distances.
Two data bases were chosen as the basis for the evaluation of the models.
The first was the Oklahoma data base4 with two data sets and the second was
the Savannah River Plant (SRP) krypton-85 data base5 with fifteen data sets.
The Oklahoma field studies took place in July 1980 whereas the SRP data base
was obtained during the period October 1976 - July 1977. Both data bases
involved tracer releases; as a result, only the transport and diffusion
components of the models could be tested. Evaluation of the treatment of
chemical conversion and wet/dry deposition was beyond the scope of this study.
The above data bases contained ground-level concentrations measured with
averaging times of 45 minutes to 10 hours.
1-1
-------
The eight models submitted for evaluation were:
* MESOPUFF^/7 Environmental Research and Technology, Inc. (ERT)
* MESOPLUME^ Environmental Research and Technology, Inc. (ERT)
* MSPUFF^'iO North Dakota Department of Health.
* MESOPUFF II11"1^.... Environmental Research and Technology, Inc. (ERT)
* MTDDIS14 Rockwell International, Inc.
* ARRPA15'16 Tennessee Valley Authority (TVA)
* RADM17 Dames and Moore, Inc.
* RTM-II18 Systems Applications, Inc. (SAI)
These models are listed in the order that their theoretical development is
described in Section 2.
Except for ARRPA and MTDDIS, all models were evaluated with both data
bases. The evaluation of ARRPA was limited to the Oklahoma data base because
this model employs the Boundary Layer Model (BLM) of the National Weather
Service (NWS) for its wind field. BLM output is not available for most
periods earlier than December 1979. The MTDDIS model was evaluated only with
the SRP data base. The model is not applicable to the ground-level releases
of the Oklahoma experiments.
1.2. MODEL EVALUATION PROTOCOL
In the model evaluation presented here, the models were run as stated in
the user's manuals using data that the models ordinarily expected as input. No
modifications to the codes were made that might be interpreted as model
improvement. Decisions on model input parameters were made in a way that was
consistent with model theory and represented the likely choice of an informed
user.
The model evaluation protocol also involved the restriction on the
modelers from submitting theoretical improvements to their codes after testing
had begun, although the submission of corrections of computer coding errors
was permitted. In this way, no model had an advantage over any other model.
Within this general constraint, however, each modeler was permitted to provide
a "final" updated code for evaluation. Most of the models had undergone
1-2
-------
changes since their original submission to EPA based on the 1980 Federal
Register request. It was of interest to all involved to have the most current
version of each model evaluated.
A final aspect of the model evaluation protocol was the opportunity of
the modelers to comment on the work as it progressed. The project was
organized into three tasks (as described below) and each modeler had the
opportunity to comment on the procedure and accomplishments at each stage.
Modelers were asked for their approval of the results of each task before the
next task was undertaken.
1.3. DEFINITION OF PROJECT TASKS
The project was divided into three tasks of which the present report
provides the results of Task III. The three tasks were as follows:
TASK I. Preparation of Input for Eight Long-Range Transport Models
This task involved the acquisition and computerization of the pertinent
portions of the Oklahoma and SRP data bases. Software was developed to
extract from those data bases the required model inputs in the proper format.
The end result of Task I was 119 input data sets. The Task I report19
described the data bases and the methodology of model testing for review by
the modelers and by EPA. Comments by the modelers and by EPA were given in an
Addendum to the Task I report20 along with ANL/UI responses.
TASK II. Preparation of Sample Runs for Each Model
This task involved the preparation, for each model, of one sample run for
each of the two data bases. These runs were included as part of a Task II
report21 to EPA for review by the model developers. In this way, the model
developers verified that their computer codes were implemented on the ANL
Ridge-32 minicomputer correctly and that the codes were being run according to
the requirements of the user's manuals. Comments by the modelers and by EPA
were given in an Addendum to the Task II report22 along with ANL/UI responses.
1-3
-------
TASK III. Preparation of Model/Data Comparisons and Model Evaluation
This task, represented by completion of this report, provided the results
of the evaluation of the eight models with the 17 data cases. A draft of this
. report was submitted to the modelers and EPA for review. Comments were
received from the model developers and from the EPA and were used in
preparation of this final report document.
The project, therefore, proceeded in stages with reviews of the progress
of the project carried out at each step.
1.4. ORGANIZATION OF THIS REPORT
This report consists of two volumes. Volume I presents a summary of the
analysis and results. Volume II contains the supporting Appendices. An
overview of the contents of the remainder of Volume I now follows.
Section 2 of this volume summarizes the key theoretical differences among
the models. A discussion of the major assumptions employed in each
meteorological preprocessor and plume model is made. Section 3 describes the
characteristics of the Oklahoma and SRP data bases. Described there are all
the source, meteorological, and receptor data that were available. Section 3
also describes the subset of available data that were used in this model
evaluation study.
Section 4 presents an operational evaluation of the models. For the
operational evaluation, the model/data comparisons are evaluated by
statistical and graphical methods. Results for both data bases are discussed
simultaneously. The AMS statistics are supplemented with isopleth plots,
scatter diagrams, frequency histograms, cumulative frequency plots, and
pattern comparison results. Section 5 presents a diagnostic evaluation of the
models. This diagnostic evaluation involves tracing back model/data
discrepancies to theoretical assumptions in the models. Section 6 presents
the summary and conclusions of this evaluation work.
As noted above, Volume II of this report contains the Appendices A-E.
Appendices A, B, and C support the first several sections in Volume I.
Appendix A (also Ref. 23) reprints an earlier paper describing the model
1-4
-------
evaluation work done for the MESOPUFF and RTM-II models with the Savannah
River Plant data. Appendix B describes on a model-by-model basis the key
assumptions and choices that were made and problems faced in generating the
input data for each model. Appendix C describes the coding changes that were
required for each model to make it operational on our computer and to allow
the codes to be run with the Oklahoma and SRP data bases. Appendices D and E
contain auxiliary statistical tables and graphs that could not be placed in
Volume I due to space limitations.
1-5
-------
SECTION 2
REVIEW OF THEORETICAL FORMULATIONS OF THE MODELS
2.1. INTRODUCTION
This section provides a brief theoretical review of the eight candidate
models. Each of the eight models consists of a meteorological preprocessor
and a plume dispersion model. For most models, the meteorological
preprocessor computes the wind field on a two-dimensional array of grid points
that cover the area of interest. In general, the horizontal components of the
wind are computed on an hourly basis based on input meteorological data from
twice-daily rawinsonde soundings and/or hourly surface data within the region
modeled. For each hour, the meteorological preprocessor computes a two
dimensional array of mixing heights and Pasquill-Gifford-Turner (PGT)
stability classes.
The wind field models are generally spatial and temporal interpolation
schemes that operate on the meteorological data that are input to them. These
interpolation methods are not solutions to the continuity and momentum
equations. These interpolation schemes, although accurate only in a gross
sense, are designed to be a simple and practical compromise to the solution of
regulatory type problems.
The wind speeds, mixing heights, and PGT stability classes are input into
a plume dispersion code which computes plume transport and dispersion as a
function of time. Plume transport schemes used by the models are summarized
as follows:
(a) dispersion of a series of Gaussian puffs as illustrated in Figure 2-1
(MESOPUFF, MSPUFF, MESOPUFF II, and MTDDIS),
(b) dispersion of a series of contiguous plume segments where these
contiguous segments generalize the straight-line Gaussian plume, as
illustrated in Figure 2-2 (MESOPLUME and ARRPA),
2-1
-------
Figure 2-1. Schematic representation of puff superposition approach,
(Adapted from Scire, et al.12).
Figure 2-2. Schematic representation of segmented plume approach.
(Adapted from Benkley and Bass8).
2-2
-------
(c) a Lagrangian random walk process in which a large number of particles
are released that undergo transport and diffusion (RADM), or
(d) a finite-difference solution to the convective-diffusion equation
(RTM-II).
Figure 2-3 shows a sample meteorological grid for which a wind field
would be generated. This Figure also illustrates the computational grid for
which plume calculations would typically be made. This computational grid is
generally a subset of the meteorological grid. A sampling grid is defined as
a subset of (sometimes of finer resolution than) the computational grid in
which individual gridded and/or non-gridded receptors are located.
Table 2-1 provides a summary of the major features of the meteorological
preprocessors employed by the eight candidate long-range -transport models.
MESOPAC is the meteorological preprocessor for MESOPUFF, MSPACK is used by
MSPUFF, and MESOPAC II was developed for MESOPUFF II. The RTM-II model has no
fixed meteorological preprocessor and the user must decide which one is most
appropriate for the modeling problem at hand. MESOPAC was recommended by the
RTM-II model developers for the SRP data base, with MESOPAC II recommended for
the Oklahoma data base. For RADM, the model developers recommended the use of
MESOPAC II. It will be noted later from Table 3-1 that there is also a wide
variety of meteorological data needs among the competing models.
Table 2-2 compares the major features of the plume dispersion models.
References 24-41 support Tables 2-1 and 2-2. Described below are the
highlights of the theories of the eight candidate models along with the key
data requirements of each of the models. The emphasis in the discussion below
is in the transport and dispersion portions of the models. The use of tracer
releases (Oklahoma and Savannah River Plant data bases) in this evaluation
study precludes any testing of the components of the models that deal with
chemical transformation and removal processes.
2-3
-------
9.0
8.0
7.0
6.0
V 5.0
4.0
3.0
2.0
1.0
Meteorological
i
Computational
Grid
p
1
Sampling
"Grid
i
1.0 2.0 3.0 4.0 5.0 6.0 7.0 8.0 9.0 10.0 11.0
x
Figure 2-3. Sample meteorological, computational, and sampling grid.
(Adapted from Scire, et al. 12)
2-4
-------
to
hi
o
to
01
u
o
hi
CL
V
hi
p-
rH
CO
U
•H
BO
0
rH
O
hi
0
01
u
01
X
U-4
o
CO
01
hi
3
CO
01
tt.
hi
o
"3
2?
U-l
o
e
o
m
Ui
(0
i-
0
cj
,
_*
1
cv
01
rH
CO
H
•O
rH
CU
iH
to
TJ
e
3
u
o
IM
4J
3
a
e
rH
09
01
CO
a
CO
rH
U
X
•H
«H
-H
..O
to
W
4j
x
OC
•H
*
BO
C
•H
X
•H
£
CM
hi
01
X
3
hi
01
X
3
IM CO
O hi
CU
• X
.1,3
w
o
CO
CO
f^
CM
CO
hi
Aj
CO
B
01
>
«k
*C
CM
^-t
3
1
^
01
BO
CO
01
>
CO
CO
o
hi
01
>
•0
CU
BO
CO
01
>
a
x
rH
rH
«
U
-H
4J
hi
CM
M
rH
fj
0
to
£
•o
C CO
to e
o
Q -H
CO CO
•O 4J
CO
hi
01 01
t» 0
o to
U U-l
hi
•0 3
3 eg
o
rH 4J
(J (Q
c"S
O 01
•H a
W 03
to
•H TJ
•o c
eg -H
• . ih
w »»
pH
a
0 -0
•H 01
BO 4J
O CO CO
rH hi U
•0 O 0» CO
01 hi E -O
to o 01
CO 01 BO 01
X 4J CJ
CU CO CO
•a E hi IM
0) O 01 hi
•H hi U 3
UJ CJ 01 10
•H -H e
o B co E
01 hi O
CL C CO Ul
CO O CUM
1
01
4J - BO
X X C
BO B -H •
iH CO BO
oi o •
X O hi 01
r-. O <"
BO
•H x"x"c
TE o o
e oo m eu
o >
hi O hi 01
UH 4J O rH
O hi hi
U O O
CU ^N -X
O 4-1 CO B
O X T3
UH 60 B O
hi -H -H O
3 01 3 r-.
n f
*._' Q) A
BO CJ X
n C a B
•0 ^-1 U-
C X hi O
•H -H 3 m
x"
o
o
r-
•a
B"?
O B
0 0
m o
ao in
o
CO
•a
c
•H
B
O
O
in
1
•H BO
01 CL C
O -H -H
CO CJ T)
UH CU i-l
hi hi >
3 O. O
CO hi rH
BO CXrH
BO B -H
6 -H X 3
•H -O CO O*
CO 3 •O CO
3 rH CO
CJ IM CU
ECO
X iH U-l
4-1 01 O
•H n) B
hi 4J *H CO
O CO 4-1 OI
rH -O rH
to -a c co
3 10 >
rH O
CO rH B X
•H CJ O rH
CJ IH hi
01 -O 4J 3
a. B co o
tO CO 4-> X
X rH B
rH T3 O
O r 01 iH
X 01 CO BO
O CO O
6 iH X rH
03 0
4-1 CO U.
B 4-1 O
o u x cu
•H O BO 4J
4J Z -H 01
CO 01 E
O 0 01
CL BO U
hi CO E CO
01 "H iH IM
U CO X hi
B 10 -H 3
rH X E OJ
1
^
•H
13 -H U >01BC
BO *H X ^-^ CO *O 4-1 u*t CO CO E *O *H
4-1 BOX^UOIU x x cj — i o
Ufl 4) 00 O O. O CL CU CU
hlrHEhlEO>4J *OCO *O X hi X
*O Q) OICQO 3 01 OI^J OI4J
U T! 03 rH O rH O 4-1 O U X
O rH 01 01 CO 3U 3XCOG4-> CL CL O rH U
iMUBOEH-H-HB en E
•HXX 4-1 01 o -a o c hi E
?B OC^: Bhi-H OB U o 01 CO
033.001-0 IH -H ax
rHt-:ox£i-i>fg -o 3 *a u CL 4-1
rH *. L. O J 4J hi C C CO 3
•H ICOtOOlBOiH-OlHrH hi
3TIBOE MX 3 01 3 3 *O 01
CTOIEiHE-H4JOI BO O B X
CQ4J.HEOhl hi CyCO 01 rH CQ 4->
C03COOUOIC3 Xhi X CO to
> ^ 3
01 O > O O CO hi
C > o > cn C C 4)
cu -H o u a
» -0 ^- TJ CJ 01 E
OI ErHErH*4JO>
TJE COOICQCUU-lAJrH4J
CU CO > > O X CO
•HE 3CU30IBO ^ tO rH C 01 U
O u:EI--. u M-03COX a
UTI-OOB CMO CD CO OIBXXtO 1
OJBU-lOO — « CO-O COiHptJrH £
rH 0) CO iHOCOl SB 33301 J
COtO COjJOhlCM iH 4JEOI 00
CJXtU4J*HCMOI 0)3 OlrH X*
•H CJCOCOIXtO X XOIU-iOIUOIhl
tUBUH tO rH 41 01 01 UIMOO
EOhlhlpOX> •>> MrH4J OXX
3-H3-H4Jinsoi -OOI -OIBOI col
Z 4-1 CO CO : ^- CO rH OrH O 01 01 4-1 C CO
•HI -H rH E 3 O 01
M O> hlBOUaiHrHX
OlrH OIBCOEuXhl
a so a-HOiocococu
0) X-H X <-> IH 01
X rHCOrHOOXOI
CO IH -H 4J 41 4-1 E X O
Xr-4 COOICOXO4J4J
4J •OX'O'O'-'tOtJ
•H *J 1 4-1 1 4) 4-1 C T3
UCO 0) OIUIMCOXiHOI
rH-H iHhliHiH CQ-OrH
•HIM 3co3UCT34ia
•Hhi tO.CO-HBhliH OtOCO
> B co-HOieo-aoiXoi-H
OX O CO-OCJUhi
XMO BX-BiHtOto CL-O
JiOu-l OO^»O UHOO
SOII -H B^hiOIUUOIX
XXO 4-10) 4-1 01 X 3 U 4-i
OUNM* COrHOtOcOUCOBOIOI
UXO4J3 O»E
CO iH in CO 01 IM -H
09— 4) > O 4-1 to C
cuto 0) x -H a oi g
•U O ••OUOOIOCJE
O -O-C30IO3
09 OICOhiiH XXX
E'V-'BOI OIUUCJ
•H 01 CO X CO
CO-HXC03O tOcoX
*•* hiOBOhiOiH-OXOOl
OH X-HXUBi-lrH
OH xuoixcocohi oc j:
9 CJXCJ—UOIOE
SJ n o) COOIiHUCOrHU
^ I XOI-H-HB04JCO
-------
CO
eu
•a
jj
i
^_j
BL,
UH
o
CO
0)
kl
3
CO
01
kl
o
1-1
UH
O
B
O
CO
kl
ca
o.
B
s
0 ••
iHJ=
4J 4J
•H a.
CO 01
o u
a. B
28
B ki
O f
•H --*
U *4
CO CN
kl -
OJ=
UH kt
CO CO
B 01
CO B
kxrl
-7 CO
O iH
en
^^ 01
en IH
CO
01
3 —
0 B
kl
eu o
60
kl -H
2, v
CO
S 01
o u
ki B
UH CO
4J
NCO
O -1
T3
XU
O CO
CO
UH
UH
,2
B
CO
CO
CO
3
CO
u
E*
Ex
3
a,
O
M
£
o
eu
U CO
41 —
eo S
•s°
O -H
rH O
| |
CM -»
0 0
en en
• »
> >
CO
cu
u
E
kl CO
eu 4-1
U CO
UH «4
UH -0
S "
04-.
U CO
UH ca B
rH J4
DN10
» kl O
XO -H
O UH A.
eu
B
Z
X
•H
U
O
rH.
01
>
B
O ••
•H.C
•H a
09 01
0 U
a. B
<§S
S ki
O f.
iH ^
kl -
OJ=
ca co
B 01
CO B
-4* CO
0 iH
en
"^ 0)
CN4J
o ca
en IH
CO
01
ki
3 ••
o e
.*
eu o
E O
3 \/
CO
8 0)
0 U
kl B
UH CO
M eo
D -4
T3
^•Jpj
0 CO
CU
B
3
rH
CU
B
ca
•H
CO
CD
3
CO
U
2
>j
Qu
O
cn
u
£
0
01
O CO
eu ~^
co e
~mu
U —
*
-H O
M M
CN-*
O O
en en
?>"
CO
u
u
u c
0) (Q
W 4-1
M-4 «0
U-I *H
0
B B
O 01
4j ca
B 0
eu co
•O
01 B •• 1 1 1 1
a. coin o o o o
CU .____!
T> cue
U CO X X X X
I-H CO 4J
a b B o o o o
• > cu . . • •
§rH iH en rH rH _4
rH 0
eu ca IH n n » i
B UH CN -4* CNflL]
u IH cu o o o en
cu ca o en w z H
S M O •< •< -< •<
x u o o
4J eu oi eu
-i O CO CO CO
CJ CU ' — ^~-
O co B B B
rH -- O O O
01 B
> cj -H in in
B-H — ' O O O
o -
4J U
•H a, CM -a- CMP*
co cu o O O en
o o en en z H
a c . « » •
eu o -o -a TJ -o
Q 0 > > > >
B kl
4^
ctj CN
IH -
Of
m co
B eu
CO E
U -t
4J rH
•^ 09
0 T<
en
"-^ 01
O CO
en IH
CO
cu
ki
3 •-
UO 4J
S a
kl j!
V UH
E O CO B
ki in co *
3 V rH
co 1 m
§41 U A.
u o
kl E UH CO
UH CO CU
4J U 0
N CO CU B
O — 1 J-i CD
• UH ca
XU 01 — I
D CO X -0
CO
UH
£
E
ca
iH
CO
CO
3
ca
CJ
Dt,
s
cu
en
£
1
so i e B
B i-l - O O
B 01 UH X IH
O > <4H O 4-1 4J
ca eu z e ca
4-1 CJ O CU 4J
B co u --a -H
cu -»E a
•a -o o oi -H
BE • en o. o
0) CO 09 41 41 ^
a u -is ki B
01 CO B CN O. 41
"3 0) 01 O 0) N
4j -r* en ki UH o
rH CO O CO O kl
CO kl -H U UH
> UH O ni4
O rH UH u-i O 01 •
B rH 01 Z 4J TJ
01 CO O CO O iH
kt U4 U 4J » CO 3
B B en ki er
•J -r4 80 01 O CO 1-1
CU CO B IH Z X rH
9 ki iH 0 35 V^-
UH -O ki « B
o eu o -» o •«
CO UH O IH B
rH CO CO U CO
0) *fi 4J iH rH
•o a. - co
§B eu en o eu E
O B 0 O 0. ki O
01 -rt 4J O CO
X CO iH « CU 4J
COOCOX X • -H C
I-H ex o 4J o cnu cu
1 01 CL -H Z O -n TJ
01 TJ 01 O Z u B
0) -O O - O 01
kl X rH CN«O rH O.
guBcuOBcueu
01
•a i
<-« eu ki x
lHOO4JOCU.fi •
— - a. E -a oi
« xxScooeuTS
^ z z ej in 06
CO U-I *-, iH
01 • *O 01 01 ^ rH
4J -^ B kl 01 S 01 CO
too caco^iHkio
IH CA ki 4J C4.-H
-'-• 3 X B
O W «
eu o 01 u 4J
E X j: UH CO 4J r->
kl 4J UH iH rH •
3 s/ o> ^ 3 6
CO A 4J UH
B oi e co cu o
O O 13 O Q O
kl B CU k!4-l -w' rH
U-i « C u-i M
4-> -H CO B II
MCO B NrH X
0 ^ k. D 3 0
•a en S ox
4J « kl X
X4-1 CU XO UH
O CO -a D UH A. O
<0
UH
UH
I
B
ca
•H
CO
CO
3
ca
o
rH
Ex
Cx
3
o
en
b3
£
X
4J
iH
rH
iH
ft
ca
4J
CO
•o
E
ca
eu
CO
1
kl
o
CO
E
kl
4-1
4J
B
4J
B
u
rH
CO
CO
ca
rH
01
CO
cu
CO
01
CO
E
o
4J
§
CO
3
o
cu
B
jj
.E
•H
BO
E
kl
3
TJ
-a
>
I
iH
CO
O
a
oi
TJ CO
01
4-1 CO
0) 3
rH 10
ID J=
U 4J
iH
kl rH
iH 01
CL-O
s o
IH B
6 O
CU IH
» 4-1
iH CU
ca cu
o o
cu c
cu o
o u
E
O ki
<3 **
B en
kl
O •
U4 kl
CO CO
C 01
CO E
U -H
O IH
CO
~- 0)
CN4J
O CO
en M
-H S «
CO o i— t
u-i NO
U-t O u-i
u-i en
B CO kl
O rH CU
kl 3 rH
U-I B X
kl CO
>o i;
O UH 0
E 4J
O CO
01 -H ki
CO 4J
CO CO 4J
f> u 3
•H O
TJ o. j:
3 iH CO
O O CO
rH 0) S
U ki
a. w
•n ca eu
u -a "E
B cu
T3 CO O.
01 01
e£-°
cu so cu
CD -r4 4J
J3 01 CO
O £ kl
O
a
H
£
2-6
-------
«1
V
3
~
jj
1
a
ts|
i
CM
C
H "H
iH
01
•o
^
01
73
35
01
c
jj
X73
JJ C
•H to
O
O SM
•H O
0) tO C
> X O
kl JJ
e o -H x
O uj 3 w X
•H C JJ
JJ JJ 01 01 -H
•H o. jj 73 -H
09 4) B C i-l
o o u-i oi x
o. e i-i o. B
01 O 3 Of JJ
O o oo 73 01
01
C X B
O JJ
•H BC
JJ X C
fJJ IH i-H
i-l > B
ki 3 B 3
o «c c
» ffl JJ 10
C 01 C
to c to -o
kl -H JJ C X
JJ — 1 03 10 O
c e
-»00 O -H 01
O -H O B 73
to c c
— OI 01 ki 01
CMJJ jj 3 a.
O B 10 i-4 01
CO ki ki 73 73
u
•o
o
X
JJ
01
E
01
JJ
iH
00
o
&
u
0)
B
a
Bu
c
B
01
01
a
a
o
1
<
01
c
o
2
*>^
£
Ov-,
O CM
-H 0
oi en
>
kl
c o
O UJ
1-4
JJ JJ
•"4 O.
09 0)
O 0
a. e
01 O
Q 0
e
o
-H
JJ
m
kl
Clf"
Uj ki
to B
C 01
to e
kl 1-4
JJ i-H
•7 0)
O -H
en
-~ 01
CMJJ
O B
CO ki
I a
01 U
*-H «H
OJ jj
klBO
x u e
JJ > O
•H iH
3 a jj
™
JJ 00 iH
c « a
0] NO
C ^ UJ
O
U - 41
C -H
a o 1-4
JJ O
x a ^
x > a
_«/
B B
•H ^
BO
c B
0) 0
k" -a
BO e
JSol
|
PC
73
C 41
8 OI C B • V ki
BOJJkiOX309 kijJ to • B
X C S tO iH rH .H 4) Qt B JJ
iHiHlOJJi-HklrH XJJ U-l X *J "
iH 01 B C 09 "H B X B B 01 O • BO X JJ
B 3 N i-4 OI 09 kl iH X 1—4 C *H BO X
•HIJ.BO.OO>H • xe 41 o oi -HBO
jjoikiktxo.a.i-iijj 73*0 B N -H x oi -H
BiHOIJJOlBSO OIB-H i-ljj 01
o>x jj'^'aaio* xojbo ma BO «— ix
naoiCTjJJOie — i jj c -H c IH
•HBBStM O e 03 TO 4) T3 i-4 TO JJ
CkiBOO-OBOi-l ki XB X «
OBkiJJfHCOCCJJ UjM BO JJ ki N>H |J X
«H > B 3 O CO -H B^l OOIB *^B OJJ
JJ CU O 09 Ncn ^ i-J O kl V U-4
iHXXOI>X3i-lkiXC • « TO Z -O - C
03 1-4-9 03 01 T3 ^ U 41 JJ 3 0 B IH3CX 09 O
O r4C"> fOkiOI^HO>OIJJ O> - 01 OiJBJJ 01
O. TO 003XJJBr-4JJJJ OlvO -O O 3 01O.1-I 3 JJ
0)ki*- JJBUXOJ3 *Ocn kiU"iO OXU *H C
^j o JJ •- iH-HBBin *-> o M iHuufo a 01
O.JJCB3JJ1HB iHOI U-l /^BO BU r-l > *^
-4'EoooBaktki'^ IH JJ 41
ki BO C » -0 3 01
O oi -OO-HOI "OB C S T)
oi jjueoioioc i-i u-i
XX HTOXXCtO&.'3'4)ol
JJJJXI*O BOB' JJO &-— ' JJ "W 00
— i jj a c c 3: 03 u x =>« oi x o
O 60 'S 00 *XOI >BJJOI*4OkiBiH 01
|HX3 03 kl -HO kl O C JJ -O VI TO UJ 4) JJ 0)
OlSOki-OBO Be 3-HOloo -HUBkiX'C 01
> T70101 kl O
iHCLJJCJO>BJJ*O BJJ JSOOOiBO 09
01 4) 3 TO ki 11 C JJ ki H jj c to C D U 6
O O auj OCBOOI CO 01 iJ 41 TO TO XX'OB ki
CUCBkiaiOOIO. BX *O-OX |HJJ*BCJJQ3 3
a)OO3J3u>l) JJ 0) 41 k. uj JJ 01 OOO 3 6 to O
aouootOjj>-'-a oo eo >-. csr o ^ c
•HX xxotjj «ooi coo
JJ BB>C 01 i-IOlO UJ JJ
kl OHkllJ OlBklO-HO) X
OX U 3 B to to JJ 4) 03 BO
C JJ •U'OiHOl-Uj-HOOI'H
o C ki JJ ^0N O O > JJ
B*HJJ Oo>OOBBO*H(ne B
ki 3 i-l u ki O *X 01 -H ki O 41 CD JJ X
*JJkiB <*4B So^xcCX'OXx's ^
03 10 -J 01 VQ-HBIHCncBB^
COI U*O /^1J JJOOIJJ09UJO
TO C T3 >H XCOa.UiHX4l «-H
fn 13X COIOOSSBf-'ki X'-'
x jj--toxTOoa-.oa.jj
1 JM; — 1 S -H W O S X 03 -H-Q
300 BO bou oicnojj>oiuo
U-l C UOOOCOI C W ki 60 X 0) JJ O -H
UJi-l «-l OBJJ* OZUJCJJDC-HU
^4ki JJ S O ki C 0) N OI-H30IOIOI
•OO klji — BO 01 UJ O»Ci-IOi-4i-4 >a
BO 01 VoBOu hOIOBO
xa > o J3T3CL-^:>'-i>-Heaj
•OS U-l XOIO.CSBO4IUJOB
•O t/3 el D 00 E 3O.jJX>JJUj,-l^
01 -HO -3COU3j«;-H4)JJJJ
X |j-a to O 0901
B OBOUJOOl^kiBQ-Huj OX
JJOO *HC UjktCOCJOTjJOIBOO-JJ
CD JJBOOU-lOSBUJ NC O T3 C 01
NJJ ja > o o to c > jj 6o e
ki> ux g-oiuj oaefiooioi e*
xoo to kieuoi4ia*o
•ne ki o 3 01 o ;gi- oi ki u jj oi
•O-H oiouoi 1-1 x u: jj x a. 01 XB
01 CO-HJJJJJJOIB03
O CT)ki-kikiO»J03XkiT3.— 101
e BOI3OOIOICiHJJBO«ColOI09
kl 09UJ C UJXOIOJJX3
01 011-4 VJ309 Bi-lbi-ieXiHOIX
UJ 3T3"HtOtOJJOXOBJJtOOjJ
u-i aoex oiiHjjo^>e*>t30IUJiHU-4UOIU
Q -HCWklUjUJXOI-HOklklO
UJ 4) 01 V O 03 BOOIJJ>'-'iHO!OliH
01 Okiki-HB 0 -H D 1 uj 41
JJ 01 OIOOOU OOCtJ-JCTl>B41>
•H C3O.Ok.rfllo-HO4l3k.
e o o9ooioijj-Hex-Hj->eooc
•H 1-409i-4vOakl TO JJ O 10 JJ C O 09O
Cb JJ J) 75 05 klO-HjjOll-Hllll^
O iJ 01 ^J O •« -H . C 3 kl kl 4) -H JJ ^H ^
3eau^ iHOUJOieujkia CLOD
to o o jj tHCiHiH o. o a. o uj a o
tO •HC--«O4l01ffltO 01 O-fHSO,
XlJjJQXO-HOk. U73O01DTO03D
OJ-HNkl-HOl OJ0973CBOIT)X 73
M oikiekiaiki^iHe o>o.kii-4ajQ)4ioi
I asooxoiBkio NX D 3 B x c a x
g SHOXJJ>UB(lUUSH73to>HoaJJ
Pi BXU 730IUioOXi-4i-i
2-1
-------
2.2. DISCUSSION OF KEY FEATURES OF PREPROCESSOR MODEL THEORIES
2.2.1. The Meteorological Preprocessor - MESOPAC
A meteorological preprocessor to the MESOPUFF and MESOPLUME model called
MESOPAC employs twice-daily rawinsonde soundings to provide a gridded field of
winds (u,v), mixing heights, and Pasquill-Gifford-Turner (PGT) stability
classes on an hour-by-hour basis. MESOPAC provides only a single-layer wind
field; i.e. u and v are constant with height but vary in the horizontal plane
and with time.
The user's manual to MESOPAC leaves to the user the determination of the
most appropriate method of defining the layer for computing the wind field for
his application. This determination is crucial to the application of the
model. Suggestions are made that the user consider employing the 850 mb winds
(if mandatory pressure levels are only considered) or the 700 mb winds (if the
terrain is high). Once the user has decided how to define his single-layer
wind field, the user must then program that feature into MESOPAC. In this
sense, the model is not a simple black box but requires significant
engineering judgment in applying the model correctly, with accurate model
predictions dependent in good part upon the accuracy of the judgment made.
Winds (and mixing heights) are first determined (twice daily) at each
rawinsonde station; from those values, 1/r2-interpolation (where r is the
distance between a rawinsonde station and a grid point) in space and linear
interpolation in time is used to compute the winds (and mixing heights) at
each grid point (x,y) at any hour in the day. The mixing depth algorithm is
derived from simplifying and generalizing, for regional scales, the site
specific algorithm of Benkley and Schulman.2* At each nodal point on the
meteorological grid, the mixing depth is computed as the larger of two
independent values: a mechanical value and a convective value. MESOPAC also
produces gridded fields of PGT stabilities for each model time step at each
grid point. The normal method for computing PGT stability (Turner25) relies
on wind speed and solar insolation data during the day and wind speed and
cloud cover data at night. Wind speed at each grid point is readily available
in MESOPAC, but solar insolation and cloud cover are not. Assumptions are
made to try to develop realistic stability classes from the data available:
(a) wind speed at 10m, (b) computed mixing depth, and (c) the daytime or
2-8
-------
nightime period. In MESOPAC, then, hourly gridded values are computed for the
stability class, mixing height, and the two horizontal wind components.
For the evaluation of MESOPUFF with the Oklahoma and SRP data bases, the
model developers recommended that, of the wind field options available, the
wind field be computed based on a vertical average of wind speed and direction
up to 1500 m height for both the OOZ and 12Z rawinsonde soundings. This
assumption was used in MESOPAC for the MESOPUFF and MESOPLUME runs (Oklahoma
and SRP sites), and for the RTM-II runs (SRP site only).
The principal data requirements for the MESOPAC model are described
below. Winds are predicted using information obtained only from twice-daily
rawinsonde data. The soundings used must be for locations inside the
meteorological grid being modeled and the data employed from these rawinsondes
are wind components, temperatures, and pressures. Due to the way the mixing
depth algorithm works, pressure and temperature data are only necessary for
the morning (12Z) soundings. MESOPAC is coded to account for missing data.
For example, the code will ignore a missing sounding unless all soundings are
missing for a 12-hour time step.
2.2.2. The Meteorological Preprocessor - MSPACK
MSPACK is a revision of Version 2.0 of MESOPAC. Version 2.0 of MESOPAC
was specially prepared by ERT for a regulatory compliance case in North
Dakota.
The major difference between the versions of the two models evaluated
here is that MSPACK employs surface wind data in addition to the twice-daily
upper air data. However, the surface wind data are not used to predict the
wind field, but are used in generating the mechanical mixing heights in the
Benkley-Schulman algorithm used by MSPACK.
The key technical changes made to MESOPAC (in developing MSPACK) were:
(a) meteorological stations located outside the meteorological grid may
be considered in predicting the wind field. MESOPAC did not permit upper air
stations outside the MESOPAC meteorological grid,
2-9
-------
(b) the Benkley-Schulman mixing height routine was modified to include
turbulent dissipation of convection. Rather than allowing an immediate
collapse in the convective mixing height after hour NHRZ (reference hour of
afternoon maximum temperature), and the resulting discontinuity in the diurnal
mixing height pattern, MSPACK interpolates between the actual convective
mixing height at hour NHRZ and a value of zero four hours later. This allows
a smoother transition to the mechanical-dominated nocturnal mixing height.
(c) related to (b), the MSPACK program computes the convective mixing
depth on the basis of the temperature at hour NHRZ (user input), which is not
required to be OOZ (as in MESOPAC). However, NHRZ must be a daytime hour
between 13Z and OOZ, inclusive. The model authors commonly use 1600 hours
local time for NHRZ for runs in North Dakota,
(d) for each hour, the surface temperature data is saved on disk along
with other MSPACK output so that it can be used to compute the buoyancy flux
on a hour-by-hour basis in MSPUFF. In MESOPAC, a single value of the buoyancy
flux was read for each source and that value remained constant with time,
(e) hourly surface winds have been included in the calculation of
mechanical mixing depth; however, the 850 mb (or other user-selected height)
wind data are now required in determining the u and v wind fields for plume
advection, and
(f) an hourly precipitation field is now produced for each grid point and
passed on to the input of MSPUFF. Wet deposition was not treated in MESOPUFF.
The principal data requirements for both MESOPAC and MSPACK are the same
except that MSPACK also requires
(a) hourly surface wind speed, wind direction and ambient temperature
data, and
(b) precipitation data used for the treatment of wet deposition.
2-10
-------
2.2.3. The Meteorological Preprocessor - MESOPAC II
One of the major differences between MESOPAC and MESOPAC II is that
MESOPAC II has a two-layer wind field based on both hourly surface data as
well as twice-daily rawinsonde data. The lower layer simulates boundary layer
flow and the upper layer represents flow above the boundary layer. The user
has several options to choose from in defining the winds in either of these
two layers. The options are: (1) surface winds (2) vertically averaged winds
(surface to mixing height, mixing height to 850 mb, 700 mb, or 500 mb); or (3)
winds at mandatory rawinsonde levels (850 mb, 700 mb, or 500 mb). Any
combination of choices is available to the user for the lower and upper
layers.
The mixed-layer averaged wind and the mixed-layer to 700 mb averaged wind
are the current defaults for the lower and upper level winds, respectively.
Mixed-layer averaged wind fields are constructed from both hourly surface data
and twice-daily rawinsonde data using a scheme adapted from Draxler32. in
this method, mixed-layer averaged winds at the rawinsonde stations are used to
adjust surface winds that have been interpolated to grid point values.
MESOPAC II predicts the mixing layer height as a function of hour in the
day from micrometeorological parameters. The daytime (convective) mixing
height, the neutral (shear-produced) mixing height, and the stable boundary
layer mixing height are computed from such variables as the sensible heat
flux, the convective velocity, and the friction velocity. These latter
variables are computed from surface data that have been input to the model.
MESOPAC II also has an option to process precipitation data for use in
wet deposition calculations in MESOPUFF II.
Three types of principal data input are required for the meteorological
preprocessor, MESOPAC II.
The first type are the twice-daily rawinsonde soundings available from
the National Climatic Center (NCC). At each sounding level, the following
variables are extracted: pressure, height, temperature, wind speed and wind
direction.
The second type of data required are the hourly surface meteorological
data encompassing (from the CD-144 format): cloud cover, ceiling height,
precipitation type, wind speed, wind direction, surface pressure, temperature,
2-11
-------
and relative humidity. Surface data are required to be input at hourly
intervals.
The third type of data allows a classification of the typical surface
characteristics in each grid square. Most users will input land use
categories rather than having the specific data requested; i.e., roughness
lengths and canopy resistances.
2.2.4. The Meteorological Preprocessor for MTDDIS
The meteorological preprocessor for MTDDIS employs only surface wind
data. Also, the MTDDIS model is not applicable to surface releases as in the
Oklahoma experiment.
The model is unique among the eight evaluated in this report due to its
exclusive use of surface wind data in predicting plume transport. The
justification given by the model authors for the exclusive use of surface wind
data to transport the plume are:
(a) hourly observations of surface winds from a much denser monitoring
network are available as opposed to the upper air sounding network. Upper air
data are available only at 12-hour intervals from stations that are sparsely
distributed, and
(b) for short distances of model application (no more than a few hundreds
of kilometers), surface data may be adequate for wind field determination.
Consequently, no twice-daily rawinsonde data are used in the MTDDIS
meteorological preprocessor. The preprocessor does, however, require
twice-daily mixing heights as available from the National Climatic Center
(NCC). The model provides a special interpolation scheme in space and time
using the NCC mixing heights (different from that used in the CRSTER
Model42) for the computation of hourly values of the mixing height.
The single-layer MTDDIS wind field is derived from surface station data
by extrapolation upwards of the hourly wind speed and wind direction using
wind speed power law exponents and wind directional shear coefficients. Wind
direction is assumed to vary linearly with height with the shear coefficient
2-12
-------
providing the rate of increase (or decrease) above the 10-m surface level.
The wind speed power law exponent and the wind directional shear coefficient
are determined at each surface station for each hour based on the local
stability class. A special algorithm using wind speed, ceiling height,
precipitation factors, and time of day is used to provide hourly values of the
Pasquill stability class A-F (G stability is treated the same as F). The wind
speed power law exponents and shear coefficients are then determined based on
stability class from tabular values of these coefficients. A default table is
provided in the computer code, but the user may optionally use his own
site-specific tabular data.
The speed of the single-layer wind field (at each surface station at each
hour) is determined by evaluating the wind-speed power law formula at the
t
effective height of rise of the plume segment released for that hour (one
segment per hour ). This height of rise equals the stack height plus the
buoyant plume rise. Buoyant plume rise is determined from the Briggs formula.
The hourly wind direction at each station is taken as the value computed from
a linear relationship that includes the 10-m wind direction and the shear
coefficient. This formula is evaluated not at the effective height of rise
but at the stack height. Interpolation in space and time leads to the
development of a two-dimensional wind field that varies in time.
Mixing heights play an important role in the MTDDIS model because they
are used to determine the lid used in dispersion calculations. From the NCC
mixing heights, a model-specific algorithm is used to derive mixing heights at
each station for each hour. The algorithm takes into consideration ceiling
height, cloud data, stability, and precipitation in order to perform
interpolation between the available morning and afternoon heights. The derived
hourly mixing heights are subsequently used in trajectory and dispersion
calculations. The MTDDIS treatment of mixing height differs from the CRSTER
method largely in its treatment of stable conditions. MTDDIS limits the
mixing during all hours of stable conditions rather than only in the hour
prior to sunrise.
Since nearly all field data contain some gaps due to missing data,
adjustments for missing data are important to the model. MTDDIS provides for a
degree of adjustment in such cases. First, incomplete data from hourly
surface station records are permitted in order to make use of all available
data. Surface station records are flagged as missing for a given hour by
2-13
-------
assigning a -1.0 value to the wind speed. Such records are subsequently
skipped in computations requiring wind speed, and therefore do not count in
the trajectory weighting. In some cases, such as for temperature, default
values are provided by the code based upon a known reasonable mean value for
the study period. Missing mixing heights are patched by selecting the next
closest available station with data when making concentration calculations.
The principal data requirements for the meteorological preprocessor to
MTDDIS are as follows. The model uses four general types of data for model
input. Meteorological data come primarily from National Weather Service
surface stations (WBANs). Wind speed, wind direction, temperature, ceiling
height, and precipitation are among the variables contained in the surface
station data. In addition, the model uses mixing heights available from the
National Climatic Center (NCC) from tape input. As an option, the user may
also provide analyzed mixing heights to complement the NCC values.
2.2.5. The Meteorological Preprocessor - The BLM/MDPP Model
A unique feature of the ARRPA Model is its reliance on the National
Weather Service (NWS) Boundary Layer Model (BLM) for the prediction of the
wind field. The BLM is a large-scale, atmospheric boundary-layer prognostic
model. Horizontal (u,v) and vertical (w) components of the wind field are
predicted on an hourly basis on a fixed 80 km x 80 km grid. The BLM model is
currently applicable only to the eastern two-thirds of the United States.
Predictions of the BLM model are available on tape only through the National
Weather Service Techniques Development Laboratory and the Air Quality Branch,
Tennessee Valley Authority.
The meteorological preprocessor to ARRPA is actually a postprocessor to
BLM. This interface is a program called MDPP. First, a short summary of the
BLM model and its output are presented. The BLM model has 10 computational
levels, or nine layers. The lowest layer (0-50 m) is simulated using Obukhov
similarity theory. The remaining eight layers (50-2000 m) are simulated
numerically using approximately 65 upper air observations, along with a number
of surface station reports. Numerical computations for mass, momentum,
energy, and moisture are solved in the upper transition layer (50-2000 m)
using finite difference methods. Only six BLM output parameters are used to
2-14
-------
determine the information needed by the ARRPA model. These are u, v, w, 8
(potential temperature), u* (friction velocity) and IT1 (L is the
Monin-Obukhov stability parameter). Roughness lengths for each grid point are
maintained in a DATA statement in the ARRPA meteorological preprocessor code,
MDPP.
Mixing heights are computed in the meteorological preprocessor using a
modified form of the Benkley-Schulman method. The modification involves the
application of the method to every 3-hour BLM-predicted wind and temperature
profile rather than to the original rawinsonde data.
Pasquill-Gifford categories are used in ARRPA to simulate near-source
plume dispersion due to small-scale processes. The stability classes are
computed from the Golder^O diagram using roughness lengths and Monin-Obukhov
lengths. These stability classes are used for vertical dispersion (plume
center height) less than 50 m. More elevated dispersion in ARRPA (plume
center height greater than 50 m) requires the Brookhaven stability classes.
The Brookhaven parameterization is computed from vertical potential
temperature gradients obtained from the BLM output.
Some additional meteorological preprocessing is required to prepare the
input to the ARRPA plume code. ARRPA manipulates the BLM output and carries
out three types of interpolation. The purpose of the interpolation is to make
the meteorological data available on a finer grid than the 80 km x 80 km grid
used by the BLM model. Two of these interpolations involve vertical profiles
and one involves horizontal profiles. The first type is logarithmic
interpolation, and it applies to the vertical interpolation of the horizontal
wind components below 191 meters. Above that level, a linear interpolation is
used for the horizontal wind components. The linear interpolation in the
vertical direction is also applied at all levels for wind direction, vertical
wind component, and temperature. In this way, meteorological data are
available on a finer grid than the 80 km x 80 km grid used by the BLM model.
A four-point inverse-square distance weighting scheme is used by the
ARRPA code to interpolate winds within a given horizontal plane. The
algorithm takes the four adjacent values obtained from the BLM grid points
that are nearest neighbors to the local position and applies inverse
distance-squared weighting to obtain the local value.
2-15
-------
2.2.6. The Application of MESOPAC II to the RADM Model
The treatment of the wind field for RADM is commonly made by means of a
meteorological preprocessor developed by Dames and Moore, Inc. The company
had three meteorological preprocessors developed in-house, DEPTH, WINDSRF, and
WIND3D. One of these models, WIND3D, predicted the wind (providing all three
components, u, v, and w) and mixing heights as a function of time and space.
However, all three models were not available to outside users since they had
not passed through company quality assurance. A very simple wind and mixing
height algorithm was available with the model. This version would have used
an internally-computed wind velocity profile for the advection of parcels
based only on surface meteorological data. Considering the circumstances
surrounding the lack of an adequate/available wind field model for RADM plume
predictions, Dames and Moore, Inc. representatives indicated that they wished
MESOPAC II to be used with their model. The MESOPAC II model was similar to
their best in-house model in terms of the interface to the RADM plume model.
As noted above, the MESOPAC II preprocessor was chosen for the current
study by Dames and Moore, Inc. MESOPAC II was programmed by ANL to provide
the lower-layer (Layer 1) wind field for use in RADM. The Layer 1 field was
obtained by averaging the wind from the ground to the mixing height at each
grid point (an option in MESOPAC II). This method of obtaining -the wind field
and mixing heights for input to RADM was consistent with the usual way Dames
and Moore, Inc. applies one of their wind field models in that the modified
surface wind includes the true surface wind data, suitably adjusted by the
upper air data. It should be recognized that the actual surface wind data
(unaltered by upper air data) were still used directly to develop Pasquill
stability classes in RADM. As a result, two wind fields were output from
MESOPAC II for RADM input; the modified surface winds for plume transport and
the true surface wind speed data for Pasquill stability class determination.
Mixing heights and surface roughnesses were also output from MESOPAC II for
input to RADM. Further details concerning the interfacing of the MESOPAC II
meteorological preprocessor with the RADM plume model are discussed in
Appendix C.
2-16
-------
2.2.7. The Application of MESOPAC and MESOPAC II for RTM-II
There is no single wind-field and mixing height model recommended by the
developers, Systems Applications Inc. (SAI), for the RTM-II model. Several
wind field models have been used by them in previous model studies. The
choice of wind field is dependent upon the nature of the modeling problem.
For the Savannah River Plant data base, SAI suggested the use of MESOPAC.
Those computer runs were carried out at ANL in 1983. For the more recent
application to the Oklahoma experiments (in 1985), MESOPAC II was now
available in the literature and was the choice of SAI for that data base.
Both these meteorological preprocessors are discussed above. The computer
codes for MESOPAC and MESOPAC II were both modified to provide the binary
input files on an hourly basis as required by RTM-II.
2.3. DISCUSSION OF KEY FEATURES OF PLUME MODEL THEORIES
2.3.1. MESOPUFF Model
MESOPUFF is a variable trajectory Gaussian puff superposition model. In
MESOPUFF, the emission is divided into a series of puffs emitted continually
over time. The Gaussian puffs are advected by the wind and dispersed using ay
and az values from Turner25 up to 100 km, and from Heffter^l for distances
beyond 100 km. At each hour, the contributions of each puff to the
concentration at each sampler is determined. MESOPUFF treats dry deposition
and chemical conversion yet does not treat wet deposition.
A Gaussian distribution is assumed for each puff upon exit from the
source. MESOPUFF permits the user to specify one of two possible algorithms
for the vertical dispersion: (a) a uniform vertical distribution below the
mixing height, or (b) a Gaussian, multiple reflection algorithm considering
reflection from the ground and the mixing height. Puffs emitted below the
local hourly mixing height will contribute to ground-level concentrations.
Puffs emitted above the mixing height will have no ground impacts until the
height of the mixing layer rises above the puff center. In MESOPUFF, the
appropriate mixing height to use in the dispersion algorithm for any puff is
2-17
-------
the highest mixing height since the release of that puff, in spite of the fact
that the mixing height might have been lowered since that particular hour.
The computational scheme of the MESOPUFF model has three distinct
functional elements: (Da Lagrangian puff trajectory function, (2) a puff
dispersion function, and (3) a puff sampling function. The Lagrangian
trajectory function is used to advect the centerpoint of each puff during a
basic time step. The radius of each puff is determined by the puff dispersion
function. Given the size and location of every puff, the puff sampling
function computes the concentration exposure received at each grid point. This
calculation is done for every time step and grid point by summing up the
individual puff contributions at that grid point.
Key input requirements for MESOPUFF are the buoyancy flux, the stack
height, and the pollutant emission rates for each source in addition to
MESOPAC predictions of wind components, mixing heights and stabilities. Use
of the buoyancy flux as input does not allow for changing buoyancy with time
resulting from emission ,or ambient temperature changes on an hour-by-hour
basis. There are available, however, 24 multiplicative factors that allow for
hourly changes in buoyancy flux on a daily basis. In order to run MESOPUFF, a
puff release rate and a puff sampling rate must be chosen. Described in the
user's manual is an algorithm for the choice of these two parameters based
upon wind field predictions from MESOPAC.
2.3.2. MESOPLUME Model
The MESOPLUME Model is a regional-scale variable trajectory Gaussian
plume segment model. The MESOPLUME model is very similar to the MESOPUFF
model. First, both employ the same meteorological preprocessor, MESOPAC. In
addition, the MESOPUFF and MESOPLUME dispersion models are structured
similarly. However, the MESOPUFF model treats the plume as a series of
superimposed Gaussian puffs whereas the MESOPLUME model treats the plume as
divided into contiguous segments. MESOPLUME is, therefore, a generalization
of the conventional straight-line Gaussian plume model to regional
applications. Each segment of the plume in MESOPLUME describes a portion of
the plume between successive time periods; the end points of each segment are
advected in a Lagrangian sense. The representation of a continuous plume by
2-18
-------
the segmented-plume approach is depicted in Figure 2-2. As with MESOPUFF, the
MESOPLUME model treats linear conversion of sulfur dioxide (S02) to sulfate
(504), and dry deposition of S02 and 804.
In MESOPLUME, a continuous plume is simulated by subdividing the plume
into a number of contiguous "plume segment" elements. A simple equation of
conservation of pollutant mass for each plume segment helps define the
concentration of that plume segment in its transport from the source. The
gain or loss by chemical conversion and the loss by dry deposition are treated
in that mass conservation equation. The user has the option of specifying two
possible vertical distributions for the plume segment: (1) a vertical Gaussian
profile, ignoring any effects of the mixing lid, or (2) a uniform vertical
distribution below the mixing lid. In Case 1, reflection from the ground is
assumed. In Case 2, no ground-level concentrations are calculated if the
plume centerline lies above the mixing lid. The relevant mixing height at any
time, for a given segment, is the maximum mixing height that the segment has
encountered during its travel time from the source.
MESOPLUME requires the following fields (usually at hourly intervals) at
each grid point (from MESOPAC):
* horizontal (u,v) wind components
* mixing depth, and
* Pasquill-Gifford-Turner (PGT) stability classes.
The computational scheme of the MESOPLUME model has three distinct
functional elements:
(a) a Lagrangian trajectory function...This function is used to advect
the endpoints of each plume segment during a basic time step. The resultant
distance between consecutive endpoints defines the length of that plume
segment,
(b) a plume dispersion function...The widths at the upwind and downwind
ends of each plume segment are determined by this function. The plume
dispersion parameters, ay and
-------
distances greater than 100 km, the plume growth rates given by Heffter^l are
used, and
(c) a plume sampling function...Given the size and location of each plume
segment, the plume sampling function computes the concentration exposure
received during that time step for each grid point that lies under the plume
segment.
The treatment of the mixing layer is the same as for MESOPUFF with respect to
mixing height interaction with the plume segments (rather than puffs).
The MESOPLUME model has three potential problem areas that require
attention. First, in the current version of MESOPLUME, no reflection from the
mixing height is assumed in Case 1 listed above. Consequently, in Cases 1 and
2, MESOPLUME will not give correct results for distances where the plume has
not yet approached a uniform vertical distribution. This problem has the
greatest effect under stable conditions because the plume will not approach a
uniform vertical distribution until many kilometers downwind. The model
developers recommend that the Case 1 option only be used in MESOPLUME at and
beyond distances for which the assumption of uniform mixing is appropriate -
often at distances of 100 km or more. It is actually a simple modification to
incorporate an optional Gaussian reflected vertical distribution so as to make
MESOPLUME suitable for near-field computations. (The MESOPUFF model had
already been augmented by the model authors to handle reflection from the
mixing lid in order to provide meaningful near-field impacts.)
The second area relates to the shrinking of the segment size when the
wind flow along the plume axis rapidly decelerates. Due to the inverse wind
speed dependence in the concentration algorithm, the predicted concentration
for ground-level points beneath the elevated segment can actually increase.
This problem is more fundamental and arises from the basic advective-diffusive
scheme of the model.
The third area relates to the way in which adjacent plume segments are
actually juxtaposed in a curvilinear flow. In the presence of strongly
sheared flows (i.e., recirculating flows), adjacent segments may not be
perfectly contiguous. Some portions of the segments will overlap causing grid
points below to receive two concentration doses during one time step. Other
grid points may be passed over without impact at all due to the break of
2-20
-------
continuity between adjacent segments. This potential problem is also one that
is fundamental to the Gaussian segment approach.
The key input requirements to MESOPLUME are the same as for MESOPUFF
except that a puff release rate and the puff sampling rate in MESOPUFF are no
longer needed. The MESOPLUME user must, however, define a basic time step
which determines the plume segments. The MESOPLUME model releases plume
segments as a function of time and, as with the release of puffs in MESOPUFF,
each segment will eventually pass outside the computational grid and will no
longer be considered in the computation.
2.3.3. MSPUFF Model
The MSPUFF Model was developed by the North Dakota State Department of
Health (NDSDH) as a proposed improvement of an earlier version of the MESOPUFF
Model. The MSPACK/MSPUFF modeling system encompasses a revised MESOPAC
(MSPACK) and a revision of MESOPUFF (MSPUFF). The MESOPUFF model evaluated in
this report is Version 6.0. MSPUFF is a modification of Version 2.1 of
MESOPUFF. (Version 2.0 of MESOPAC and Version 2.1 of MESOPUFF were specially
prepared by ERT for a regulatory compliance case in North Dakota.) The version
of MSPUFF used by Argonne/UI includes all revisions of MESOPUFF from Version
2.0 through Version 6.0. The MSPUFF Model is commonly used in the State of
North Dakota for licensing situations involving long-range transport to Class
I areas. The MSPUFF Model has nearly identical input requirements to MESOPUFF
as described above and very similar theoretical underpinnings. Major
differences between MESOPUFF and MSPUFF are described below.
The key changes to the MESOPUFF Model (in the development of MSPUFF) are:
(a) the changeover in the functional form (Turner curves to Heffter
formulas) for ay and az occurs at 50 km in MSPUFF as opposed to 100 km as used
in MESOPUFF and MESOPLUME. Furthermore, the value of the vertical
diffusivity in the Heffter form of az is varied for each stability class.
Previously, a constant value of 5 m2/sec was used in MESOPUFF and MESOPLUME,
(b) the a and b parameters in the power law form of av and az have been
adjusted for downwind distances less than 50 kilometers. The revised power
2-21
-------
law forms now closely agree in the range 2 to 50 kilometers with the values
from the steady-state MPTER model.36 This revision was done to assure
agreement between the two models often used in regulatory applications by the
North Dakota Department of Health. In North Dakota, the analysis of air
quality for short-range distances (less than 50 km) is often made with
steady-state models, such as MPTER. Mesoscale range predictions (between 50
km and 250 km) are commonly made with MSPUFF. Extensive model testing has not
been carried out to compare MSPUFF and MESOPUFF predictions due to this change
in oy, az values for MSPUFF. However, in one case comparison (release from a
short stack), the MSPUFF predictions were greater than the MESOPUFF
predictions for distances less than about 10 km, and lower than MESOPUFF
predictions for further distances, and
(c) for those puffs released above the mixed layer, the dispersion has
been constrained to the use of dispersion coefficients for stability classes E
or F until the mixed layer depth becomes greater than the puff height,
(d) the dependence of dry deposition velocity on stability has been added
to the model,
(e) the buoyancy flux, F, for plume rise is computed hourly for each
source from the surface temperature at that meteorological station nearest the
source. In MESOPUFF, a single buoyancy flux value, F, was read in for each
source and was assumed to be applicable to the entire period of model
predictions, and
(f) the treatment of sulfur dioxide wet deposition has been added to the
model; the washout coefficient for sulfur dioxide is a function of the
rainfall rate.
As noted earlier, the input requirements for MSPUFF are nearly identical
to MESOPUFF. For MSPUFF (as compared with MESOPUFF), hourly stack dynamic
operating parameters and hourly ambient temperatures are employed in
predicting source buoyancy flux as a function of hour using the surface
station nearest the source.
2-22
-------
The model can handle up to four pollutants simultaneously: SC>2, sulfates,
NOX, and particulates. Hourly source emissions data are required for each
pollutant (as compared with only S02 and sulfates for the MESOPUFF model).
All particulates are assumed to be fine particles and are treated as gases.
2.3.4. MESOPUFF II Model
The MESOPUFF II model was developed to enhance the capabilities and
flexibilities of the MESOPUFF model including the prediction of secondary
aerosols. Extensive modifications had been made to improve the treatment of
advection, vertical dispersion, transformation and removal processes.
The MESOPUFF II dispersion model is a Gaussian variable-trajectory puff
superposition model that was designed to account for the spatial and temporal
variations in advection, diffusion, transformation, and removal mechanisms on
regional scales. A continuous plume is simulated as a series of discrete
puffs. Each puff is subject to space and time-varying wet removal, dry
deposition, and chemical transformation.
In MESOPUFF II, the basic equation of dispersion of a puff is the
Gaussian distribution with reflection from the ground and the mixing height.
As with MESOPUFF, the dispersion parameters ay and az are calculated for puff
travel distances less than a user-input distance XQ (100 km default value)
with plume growth functions fitted to the curves of Turner.25 The
time-dependent puff growth equation used for distances greater than XQ are
those given by Heffter.31 in the latter equations, the vertical diffusivity,
Kz, is a function of stability class.
MESOPUFF II allows three options for determining growth rates for puffs
above the boundary layer: (1) E stability rates, (2) F stability rates, or (3)
boundary layer stability rates. The default instructions are to use the E
stability growth curves for puffs above the boundary layer.
In terms of chemical transformations, a parameterization of the
conversion of S02 to 504 and NOX to N03 is used. These parameterizations
include the chemical equilibrium of the HN03/NH3/NH4N03 system. Dry
deposition is included by means of resistance modeling including options for
source or surface depletion. Time and space-varying wet removal is included
2-23
-------
based on the precipitation rate (liquid or frozen) with a scavenging
coefficient based on precipitation type and intensity.
A change from MESOPUFF was in the treatment of vertical dispersion as it
relates to dry deposition. Once the plume has become vertically mixed in the
far field, the user has an option to employ a three layer treatment of dry
deposition. In this option (as a consequence of vertical mixing), az is no
longer applicable and stability classes are no longer used; parameterization
of vertical dispersion is made in terms of micrometeorological variables with
inclusion of a shallow surface layer in order to predict dry deposition from
j
the plume.
A change involves the treatment of spatially and temporally variable
rates of chemical transformation, dry deposition, and wet removal. MESOPUFF
simulated the transport and chemical conversion of only S02 and sulfates and
treated only dry deposition. MESOPUFF II treats up to five pollutants: S02,
S04/ N0x/ H^OS/ ant* N03 ani^ can predict wet deposition as well in a spatially
and temporally variant manner.
A change involves improvement in the treatment of the puff sampling
function. In MESOPUFF, there were some problems of poor resolution of puffs
in the near field if the puff sampling rate were not specified high enough.
The revised sampling function tends to reduce the error in the near field and
makes it less likely that the user obtains poor results with the model if
incorrect choices are made of the puff release rate and puff sampling rates.
The key data required are emissions from point (and/or area) sources.
For each point source are required: location, stack height, stack diameter,
exit velocity, stack gas exit temperature, and the emission rate of each
pollutant.
One potential problem for the user is that missing data are not treated
within the code. The user is required to input replacement values for all
missing data. This can be a time-consuming process for a large modeling
region with significant amounts of missing surface data.
2-24
-------
2.3.5. MTDDIS Model
The MTDDIS Model (Mesoscale Transport and Dispersion Model for Industrial
Plumes) is a variable-trajectory Gaussian puff model. The model is an
adaptation of the ARL trajectory model by Heffter.43 MTDDIS was developed to
predict short-term impacts from elevated industrial releases over distances
that range from tens to a only a few hundreds of kilometers. It does not apply
to surface releases as a result of its special treatment of the single-layer
wind field. The model also has an option to treat dry and wet deposition.
MTDDIS uses the common elevated Gaussian puff algorithm to predict
concentrations. Reflections due to the presence of the ground and mixed layer
are assumed. The Lagrangian evolution of trajectories is achieved by tracking
the endpoints of each plume segment. Segments are further sub-divided into a
variable number of puffs when better resolution and continuity in the
concentration field are required. Horizontal dispersion grows linearly with
time as in the ARL Heffter model. Vertical dispersion is treated as a function
of stability class and time following the procedure of Draxler.44
The modeled region is the distance between the ground and the local
mixing height. Ground-level concentrations are computed by summing puff
contributions at each grid point for each hour. Vertical mixing is treated as
uniform beyond downwind distances where oz exceeds 1.6 times the mixing
height. When the effective height exceeds the local lid, the contribution to
the ground-level concentration field is counted as zero. This situation would
occur, for instance, if the stack height exceeded the mixing height.
Trajectories are generated by initiating one new trajectory from the
source at the beginning of each hour. From that point, the trajectory segment
endpoints are updated once each hour. A composite trajectory segment for each
hour is obtained by station weighting as adapted from the ARL-Heffter Model.42
Each trajectory segment is first propagated independently based upon data from
each individual surface station that has data available for the hour. The
cumulative segment motion is derived through a weighted averaging of the
independent motions. The weighting algorithm accounts for distance using
inverse square weighting and for directional alignment using angular
weighting. Recall that the hourly wind speed and direction used in the
trajectory computations are based upon the predicted values of the wind speed
2-25
-------
and wind direction at the effective plume height and stack height,
respectively, and not simply on the 10-m surface values.
The principal data requirements for the MTDDIS model are as follows.
Source-related input data include stack height and diameter, updraft velocity,
exit temperature, and hourly-averaged emission rate for each hour. Also, the
mean sea-level elevation at each station including source latitude and
longitude are required. It should be noted that MTDDIS also provides for
input of certain deposition-related parameters, but the SRP data base using
passive tracers did not employ the deposition modules.
Inputs relating to the modeling domain include surface elevations,
roughness heights, boundaries for wind fields, and boundaries for plume
predictions.
Finally, inputs are required to prescribe user options for the
predictions on a grid including grid origin, grid boundaries, grid spacing,
and locations of specially placed samplers. These inputs are options chosen
by the user to permit predictions for the desired time-averaging periods and
geographical area.
2.3.6. ARRPA Model
The ARRPA Model (Air Resources Regional Pollution Assessment Model) is a
single-source segmented-plume model designed to compute pollutant transport,
dispersion, and deposition over the regional and interregional scale. It
predicts dry deposition and chemical transformation processes for S02 and
sulfate.
ARRPA uses a segmented Gaussian plume approach to predict airborne
concentrations. The Lagrangian evolution of trajectories is achieved by
tracking the segment endpoints. A new segment is emitted from the source
during each hour. The endpoints of the trajectory segments are updated once
each hour.
Terrain topography is inherited from the BLM model. ARRPA uses 1/r2
interpolation between BLM grid points in order to obtain terrain elevations at
individual receptors. Because of the BLM mesh spacing, ARRPA cannot
adequately resolve complex terrain features. However, by specifying receptor
2-26
-------
heights at particular locations, concentrations are adjusted to simulate
terrain effects to some degree.
The divergence adjustments resulting from dynamic features within the BLM
model are inherited by ARRPA, and therefore adjustments for shear are
necessary. ARRPA computes the vertical wind field deformation (change in w
velocity with elevation) over the plume top and bottom edges to arrive at an
adjusted az. The horizontal divergence is adjusted by varying the centerline
concentration subject to the constraint that it cannot be increased by shear
effects only (that would violate entropy).
Plume rise is predicted by classical one-dimensional entrainment theory
following the method of Slawson et al.45 Vertical transport of segment
endpoints uses the interpolated wind at the plume centerline. The ARRPA plume
model provides two methods as user options for computing horizontal plume
transport. Horizontal components may be computed by averaging the horizontal
transport computed over the 2a vertical region about the plume centerline.
The other option involves taking the level nearest the centerline to derive a
computationally-faster horizontal transport approximation.
Each plume segment is treated as a continuous Gaussian plume. Horizontal
dispersion is simulated using an algorithm which has four growth regimes.
Turner's curves^S are used to model the rapid growth regime; i.e. distances
downwind to where av becomes greater than 1000 m. When the size of av exceeds
1000 meters at the start of a plume step, a transition is made modeling ay by
means of a combination of the Turner curves and Gifford's Lagrangian theory-*?.
When the starting oy for a plume step is between 6000 and 10000 meters,
Gifford's Lagrangian theory is used. In the slow growth region beyond 10000
meters, Taylor's statistical theory is used.
The vertical sigmas are computed at low levels (plume center height less
than 50 meters) using Turner's curves. For elevated regimes, Smith's
Brookhaven curves^ are used.
The ARRPA plume model limits diffusion between the ground and an upper
atmospheric layer according to set of six rules. These rules limit mixing to
an upper level of 2000 meters and selects the vertical wind level to use for
computing the transport of a plume segment during a time step; such rules are
required because the height of the plume segment varies during the step and
could possibly rise above the upper lid.
2-27
-------
Although ARRPA advances segment endpoints once per hour in order to
improve computational efficiency in transport calculations, an internal
stepping routine is used to obtain more refined temporal adjustments for
ground-level plume predictions. Concentrations are computed from a number of
discrete segment positions based on the size of the horizontal sigma as
compared to the amount transported perpendicular to the centerline. When the
predominant plume transport is parallel to the centerline, the concentration
can be computed for the one-hour interval at a given receptor as a simple
straight-line Gaussian plume since the relative motion allows use of a single
midpoint for time-averaging. However, perpendicular transport must be treated
with smaller time steps since the relative motion is two-dimensional; in this
case, the one-hour period is divided into steps with plume parameters (such as
az and oy) varying linearly between steps. Contributions from each step are
summed using multiple reflection terms following the method recommended by
Turner.25 After the plume becomes completely vertically mixed, the simpler
conventional horizontal Gaussian (vertically uniform) plume formula is used.
The other segmented Gaussian plume model in this evaluation study is the
MESOPLUME model. As discussed in Section 2.3.2., the theoretical approach of
a Gaussian plume segment model has two potential problems that should be
treated. The first relates to the shrinking of the segment size when the wind
flow along the plume axis rapidly decelerates. Due to the inverse wind-speed
dependence in the concentration algorithm, the predicted concentration for
ground-level points beneath the elevated segment can increase. ARRPA
addresses the problem caused by the shrinking segment by means of two
mechanisms: (a) it expands the horizontal or vertical width during these
periods of decelerated winds to prevent increasing concentrations, and (b) it
uses the largest segment length in the segment's history for concentration
calculations and not the actual reduced segment length.
The other problem relates to the way in which adjacent plume segments are
actually juxtaposed in a curvilinear flow. Adjacent segments that overlap can
lead to grid points below those segments receiving double doses during a time
step. ARRPA treats that problem by first checking to determine whether the
angle between adjacent segments is greater than 90 degrees. (It is 180 degrees
for a constant-direction flow.) If greater than 90 degrees, the overlapping
portion of the more downwind segment of the two is not included in
concentration calculations when overlap occurs. If the angle of adjacent
2-28
-------
segments is less than 90 degrees, the contributions from both segments are
included since this case is thought to represent a physical situation (e.g.
sudden wind shift from north to south) causing overlapping puffs. This latter
case is not considered a modeling problem.
One difficulty that a user should be aware of with ARRPA is in the
assignment of mass to the segments. In ARRPA, at the start of each hour, mass
is assigned to the leading edge of a new segment based on the release rate for
that hour. Segment masses are determined based on masses assigned at the
upwind (hour N) and downwind sides (hour N+l) of the segment. A problem
arises, however, at the leading and trailing edges of the continuous plume
release. The leading segment, for example, will have a zero assigned mass at
its upwind edge and a nonzero assigned mass at its downwind edge.
Consequently, nonzero concentrations can result from the leading segment (and
final segment as well) where zero contributions would otherwise be expected.
In that sense, ARRPA is not quite mass conserving. The problem can be
eliminated to large degree by adjusting the emission rate at the beginning and
end of the continuous release to insure low emission rates at the very start
and very end of the'period.
The model uses three general types of data as input information. The
meteorological input data to the ARRPA plume model is a subset of the output
from the National Weather Service Boundary Layer Model. The BLM grid is 3-
dimensional (35 by 30 horizontal, 10 vertical) and is defined by a polar
stereographic projection. Only six BLM output parameters are used to
determine the information needed by the ARRPA model (for each hour): these are
u, v, w, 0, u* and L"1. 0 is the potential temperature, u* is the friction
velocity, and L"1 is the inverse of the Monin-Obukhov length. The latter two
variables are only stored once per horizontal grid point since they are
surface layer values. Since the BLM output contains all meteorological data
required by the ARRPA plume model, the effort in running the ARRPA model is
reduced to the simple task of selecting the grid section and time period that
are required for the particular application.
Source-related input data include stack height, diameter, updraft
velocity, exit temperature, and hourly-averaged emission rate for each hour.
Also, the mean sea level (MSL) elevation at each station including source
latitude and longitude are required. ARRPA also provides for the input of
certain deposition-related parameters, but these parameters were not relevant
2-29
-------
to the current model evaluation program since data from tracer releases were
used.
As with the other plume models, inputs involving the characteristics of
the study area include receptor elevations, lateral boundaries for wind field
model input, and the boundaries for plume model prediction. The ARRPA plume
model contains an internal storage of the mean sea level (MSL) elevations for
the BLM grid points. Inputs are required to prescribe user options for
computation and prediction grid characteristics including grid origin, grid
boundaries, grid spacing, and locations of specially placed samplers. The
model allows the user to choose a rectangular grid with arbitrary spacing and
directional orientation.
2.3.7. RADM Model
The RADM model employs a Lagrangian random walk approach to dispersion.
In this method, the mass release is divided up into a number of parcels and
these parcels are advected and transported by the wind. Puff advection is
treated as in the Gaussian puff models, yet the diffusion employs a
displacement by means of a random walk that is scaled by horizontal and
vertical dispersion coefficients.
The random walk procedure transports and disperses parcels after release
into the meteorological grid region. The horizontal and vertical eddy
diffusivities are used to scale the random contribution to each parcel's
motion, whereas the layer-averaged winds are used to provide the advective
contribution to each parcel's motion. The horizontal eddy diffusivity is
calculated at each grid point, but is not a function of height. The vertical
eddy diffusivity is also gridded, and is determined for each of up to 8
layers. In the calculations for the Oklahoma and Savannah River Plant data
sets, six layers were used for diffusivity computation; each layer requires a
complete grid of diffusivity values to be stored. Additional layers would not
add important detail. Parcel contributions to pollutant concentrations are
summed at the end of each hour for each grid point based on the fractional
overlap of the receptor volume with an equal volume centered on each parcel.
In the transport of parcel mass, RADM assumes a simple model of chemical
conversion and of dry deposition as it follows the dispersion of two
2-30
-------
pollutants (typically SC>2 and 504). A user-supplied linear decay rate
(independent of space and time) is used to simulate chemical conversion of
pollutant 1 to pollutant 2 and for the decay of pollutant 2. For parcels
sufficiently near the ground, a user-supplied deposition velocity independent
of space and time is used to subtract deposited mass fractions from the
parcels. No wet deposition is simulated in the model. In the current
evaluation, neither the decay nor deposition portions of the model were tested
since passive tracers were used in the field experiments.
Since the model employs the MESOPAC II meteorological preprocessor, the
input requirements for RADM will include those for MESOPAC II. A discussion
of the input requirements and the modeling methodology for MESOPAC II are
presented in Section 2.2.3. The RADM plume model requires a one-time entry of
gridded surface roughness heights determined from land use categories as in
MESOPAC II.
The RADM plume model requires an average value per hour (determined from
available surface station data) of the surface temperature, the total opaque
sky cover, the cloud ceiling height, the potential temperature lapse rate in
the mixed layer and the potential temperature lapse rate in the inversion. For
the latter two variables, the model developers suggested that the values of
0.0 and 0.005 deg C/100 meters may be used. Both variables enter the Briggs
plume rise formulas; the effects of the choice of those two parameters was
insignificant for the Savannah River Plant case runs and were not used for the
ground-level release at Oklahoma. Hourly emissions data with appropriate
stack and ambient conditions must also be supplied. Finally, a table of 10000
properly selected random numbers must be read in for the random walk
calculation.
2.3.8. RTM-II Model
The RTM-II model uses a puff-on-grid hybrid technique. The emission is
released as a series of puffs and these puffs are transported and dispersed
using a Gaussian puff formulation. When a puff expands to the size of either
dimension of a grid cell (Wx, Wy or Wz = mixing height), the puff is released
into the grid system and then disperses according to the finite-difference
solution to the convective-diffusion equation. The numerical scheme used is
2-31
-------
from Boris and Book.47 This method is characterized by a two-stage, flux
corrected transport algorithm in which a correction of pollutant fluxes in the
second stage counteracts the numerical diffusion arising from the first stage.
Horizontal eddy diffusivities are parameterized in the model using a formula
proposed by Smagorinsky33, where the horizontal diffusivities are equal to an
empirical constant (TJ, based on scale considerations) multiplied by the
magnitude of the velocity deformation tensor. In this formulation, % = Kx =
Ky = Tj|Def|. Furthermore, the horizontal diffusivities are constrained by
maximum and minimum values chosen from regional-scale field studies.
The wind field in the model is represented by a single-layer; however, an
upper layer is retained to hold plume mass that escapes the mixing layer when
the mixing height decreases. Later, when the mixing height increases again,
either due to diurnal variation or to advection into a region with a larger
mixing height, this plume mass can re-enter the mixing layer. However,
transport in this upper layer proceeds according to the same wind field as is
used in the lower layer. Although the wind field is single-layer, the model
makes some attempt to include advection of air in the vertical when the
single-layer winds show either divergence or convergence. At a grid point
where the wind field has a net divergence, the air mass needed to conserve
mass is considered advected down from the upper layer, bringing part of the
upper-layer concentration with it. At a point where the wind field has a net
convergence, some mixing layer concentration is removed into the upper layer,
representing an upward advection of air. Since MESOPAC II and MESOPAC both
predict wind fields without adjusting them to be divergence-free, this
theoretical feature involving conservation of mass plays a role in the
predictions reported in this study. That feature may represent important
physics that could have an impact on model predictions, especially when fronts
approach the area.
The convective-diffusion equation involves concentration explicitly, so
no special procedure is needed to derive concentration values. The basics of
the method were described in the introduction to the RTM-II model discussion.
RTM-II allows for the oxidation of S02 and 804 by a reaction rate
equation with a coefficient that varies diurnally and with latitude. Further,
the model simulates a complex set of interrelated physical processes. It
allows for absorption by hydrometeors (802), condensation (804), wet
deposition, liquid phase oxidation (S02 to $04 in hydrometeors), and dry
2-32
-------
deposition. Dry deposition depends on surface roughness, a Reynolds number
appropriate to the flow in the roughness layer and the ratio of the kinematic
viscosity of air to the molecular diffusivity of the pollutant gas. A
deposition velocity formulation is used with allowance for vegetation type and
diurnal variation. Wet deposition is driven by the amount of rainfall (not an
input for the present study because of the use of passive tracers). Again,
diurnal variations affect the choice of coefficients in the rate equations.
The input meteorological data required for the model are exactly those
required of the MESOPAC (for SRP) and MESOPAC II (for Oklahoma) meteorological
preprocessors. Meteorological input data to the RTM-II model includes hourly
gridded values of average wind speed and direction (averaged over 1500 meters
for the Savannah River Plant cases and over the mixing height for Oklahoma),
mixing height, horizontal diffusion coefficient (adjusted within user-supplied
upper and lower limits), region top (mixing height plus 100 meters) and
exposure class (from which stability is inferred). Both the MESOPAC and
MESOPAC II models compute gridded values of stability class; the exposure
class is actually only a means by which stability classes are re-created in
the RTM-II model. Hourly emissions data are also required over the study
period in the model (as modified by ANL/UI, see Appendix C) whereas the
original model only allowed for a set pattern of diurnal variation with only
the total daily emissions allowed to vary day by day.
2-33
-------
SECTION 3
THE OKLAHOMA AND SAVANNAH RIVER PLANT DATA BASES
3.1. INTRODUCTION
This section provides a summary of the data available as part of the
Oklahoma and Savannah River Plant data bases. The field experiments are first
described along with the data that are relevant to this model evaluation
program. Next, the development of the modelers' data base for the Oklahoma
and Savannah River Plant experiments is presented. The modelers' data base
represents the ensemble of source, receptor, and ambient meteorological
(surface and upper air) data that were required by the eight models. By
receptor data are meant receptor locations; actual measured ground-level
concentrations were listed in the data reports and had been entered in a
special file for use by the statistical and graphical evaluation programs.
The largest portion of each data base is the meteorological data acquired
during the experiments. It is helpful to understand the meteorological data
requirements of each model before a presentation is made of the data bases
themselves.
3.2. METEOROLOGICAL DATA REQUIREMENTS OF THE MODELS
All of the models under evaluation compute hourly varying wind fields and
mixing heights. To provide input for the calculation of such quantities, the
models require some or all of the following: surface data (typically at NWS
stations), twice daily rawinsonde profiles, NCC mixing heights (MTDDIS only),
and surface land use data. The ARRPA model was an exception to this pattern
in that it used the wind field as computed by the NWS Boundary Layer Model. A
typical user acquires the wind field already computed by NWS on a magnetic
tape from either the National Weather Services Techniques Development
Laboratory or from the Air Quality Branch, Tennessee Valley Authority. Table
3-1 presents a summary of the meteorological needs of each of the models.
3-1
-------
Table 3-1. Summary of Meteorological (and Land Use) Data Required
by the Eight Long-Range Transport Models
Model
MESOPUFF
MESOPLUME
MSPUFF
MESOPUFF II
MTDDIS
ARRPA
RADM
RTM-II
a
b
Meteorological Twice Daily Input NCC
Preprocessor Surface Rawinsonde Mixing
Used . Data Profile Heightsa
MESOPAC NO Yes NO
MESOPAC NO Yes No
MSPACK Yes Yes No
MESOPAC II Yes Yes No
MTDDIS Yes No Yes
BLM/MDPP No15 Nob Nob
MESOPAC II Yes Yes No
MESOPAC (SRP) Yesc Yes No
MESOPAC II (OKL)
All models compute mixing heights internally to the
codes. Only MTDDIS requires mixing heights as a
starting point for its own hourly mixing height
calculation.
The ARRPA model is driven by BLM-generated fields of
wind components (u, v, and w), potential temperature
(0), surface-layer friction velocity (u*) and
surface-layer inverted Monin-Obukhov length (L"1). BLM
forecasts are based on initialized fields that are
dependent on surface and rawinsonde observations. BLM
Land Use
Data
No
No
No
Yes
Yes
Nob
Yes
Yesc
output is archived by TVA and the NWS.
For the Savannah River Plant experiment, RTM-II uses a
wind field produced by MESOPAC, which does not employ
surface data; for the Oklahoma experiment, RTM-II uses
the MESOPAC II wind field which does take surface winds
into account.
3-2
-------
An agreement was made between ANL, EPA and the modelers that
meteorological tower data, when available, were to be used to provide
surrogate surface stations. Data not available from each tower directly were
supplied from the nearest NWS station. Wind speed and direction were always
obtained from the tower either from a data recording level at 10 m or by using
a "l/7"-power law to extrapolate down to 10 m from the lowest available tower
wind values. Only if temperature were actually measured at the 10 m level on
the tower was the surface temperature for the surrogate surface station
obtained from the tower. Otherwise, the value at the nearest surface.station
was used. No downward extrapolation of temperatures to the 10 m level was
undertaken from elevated locations on the tower. Employing this method, more
"surface stations" were available to provide surface wind values to the wind
field models. As a result, more refined wind field predictions should have
resulted. It is expected that a user (typically a source or EPA) would carry
out this same procedure in a regulatory setting.
3.3. OKLAHOMA FIELD EXPERIMENTS4
3.3.1. Description of Experiments
A long-range tracer experiment was conducted on July 8, 1980 with the
release of the perfluorocarbon, PMCH, from the National Oceanic and
Atmospheric Administration (NOAA) National Severe Storms Laboratory (NSSL) at
Norman, Oklahoma. Samplers were deployed to measure tracer concentrations
along arcs at 100 km and 600 km north of the release point. A second
experiment was conducted on July 11, 1980 with samplers located only along the
100 km arc. Each experiment involved the release of the perfluorocarbon tracer
over a 3-hr period with concentrations measured 100 km downwind. In the
primary experiment, the perfluorocarbon was measured at a distance of 600 km
as well as 100 km. The region of plume dispersion to the north and northeast
of the release site is typical of the Great Plains, including minor river
valleys.
For the July 8 experiment, the PMCH tracer was released over a 3-hr
period from 1900 to 2200 GMT (1400-1700 CDT) from an open field. The release
nozzle was about one meter above ground level. The flow rate was monitored to
3-3
-------
assure a nearly constant release rate. The amount of perfluorocarbon released
was calculated to produce concentrations well above the detection limit at the
600 km sampling arc. The background level of PMCH is extremely low (2.4 parts
per 1015) and also, very few points of poor data were identified from the
measurements.
For the 100 km sampling arc, thirty sampling sites were selected at 4-5
km intervals (see Figure 3-1). Only seventeen samplers were available, so,
based on expected winds after release, the sequential samplers, known as
Brookhaven Atmospheric Tracer Samplers (BATS), were placed at sites 12-28
only. The tracer release began at 1900 GMT (1400 CDT) and the samplers were
set to take ten 45-minute samples starting at 2100 GMT, before the tracer was
expected to arrive.
The 600 km sampling arc consisted of 38 sampling sites through Nebraska
and Missouri (see Figure 3-2). These sites represent locations of the NOAA
National Weather Service substation network. All samplers on that 600 km' arc
started automatically at 0800 GMT (0300 CDT) on July 9. Of these 38 stations,
valid data were collected for all but three sites (numbers 14, 21, and 22).
On the evening of July 8, based on the latest wind data and forecasts,
the samplers were deployed to the sites indicated by double circles in Figure
3-2. Five sequential samples were taken at these locations at 3-hr intervals
beginning at 1100 GMT (0600 CDT) on July 9.
3.3.2. Ground Level Tracer Observations
The sampling results for the July 8 experiment appear in Figures 3-3 to
3-5. The sampling sites are plotted as a function of azimuth from the release
site. The perfluorocarbon (PMCH) concentrations for the first 6 sampling
periods are shown in Figure 3-3 for the July 8 experiment. The peak plume
concentrations along the 600 km arc arrived at about the same time sampling
commenced at that arc. The entire record of PMCH concentrations at all sites
on the 600 km arc is shown in Figure 3-4. The initial plume probably arrived
at all sites on the 600 km arc just before sampling began at 0800 GMT on July
9. Plume passage had a duration of about 15 hours before background levels
were seen at all locations. Background concentrations are seen for the next
15 hours, whereupon the July 11 (1400- 1700 GMT) samples show a secondary
3-4
-------
35.5*
T-JT
+ .SiiHactSMnpfin|SitM
•9 Aiteraft Flight F»th
KHenwtWl
07J*
Fig\ire 3-1. Location of the sequential air samplers (BATS) and aircraft
sampling path at 100 km from the Oklahoma tracer release site.
43'N
100*W
to*
Figure 3-2. Location of sequential samplers (BATS), LASL samplers, and
aircraft sampling flight path at 600 km from the Oklahoma tracer
release site. The locations of rawinsonde stations are also
shown.
-------
350°
Azimuth From Release Site
000° 010°
020°
5000
1000
500
jo
"o
I
I
I
§
3
o
100
50
10
100 KM Arc
JulyS. 1980
Sampling Period (GMT)
2100-2145
2145-2230
2230-2315
2315-0000
0000-0045
0045-0130
0130-0215
12 13 14 15 16 17 18
Sampling Sittt
19 20 21
Figure 3-3. Average 45-min perfluorocarbon (PMCH) concentrations along the
100 km arc from the Oklahoma experiment No. 1 (July 8, 1980).
3-6
-------
352°
Azimuth From Raleasa Sitt
356°
000°
004°
008°
012°
016°
020°
024°
028°
1000
500
o
I
100
10
T I
600 KM Arc July 9.1980
Sampling Period (GMT)
1 0800-1100
2 1100-1400
3 1400-1700
4 1700-2000
5 2000-2300
6 2300-0200
J L
'
-L_J L_L
'
J_
J L
45 679 1011 1213
Sampling Sittt
14
1516 17 18 19
Figure 3-4. Average 3-hour perfluorocarbon (PMCH) concentration along the 600
km arc for the Oklahoma experiment No. 1 (July 8, 1980).
3-7
-------
°0
•l!S atM|0u UJQJj
°0
n n
V A
O O
JJ
S 2
i §
3 Q
in
*o
»»
i
o
•9-
-.•0-
^ >
I
00
in
04
00
in
0)
4->
U)
0) o
58
CM
S1
5^r
A rH
W
C ><
J-, CO
•P O O^
C 4J «H
: H co
' 5 >i
rt
! 3
CO*"
,
en o
i >,*
m i-t 4J
0)
(0
IT)
ro
009
3-8
-------
plume arriving at the 600 km arc (see Figure 3-5). The maximum concentration
of this secondary plume is about two orders of magnitude lower than the
initial plume but they cover a much larger area. The duration of this
secondary plume on the arc was about 30 hours. It is suspected that a
night-time low-level jet transported a portion of the plume initially, and
then the remainder of the tracer plume appeared on the arc the day following
the arrival of the first portion. (It is also possible that this secondary
plume is a return of the initial plume.) A discussion of the meteorological
data during the first Oklahoma experiment is presented in Appendix E along
with the evidence of the presence of a nocturnal jet. A second theory
proposed by Dr. Ray Dickson and Dr. Gene Start of NOAA/Idaho Falls is that
there was nighttime capture of the tracers on vegetation with a subsequent
daytime release. This theory is not accepted by collaborator Gil Ferber of
NOAA/ARL since subsequent plant studies in the laboratory have indicated no
such absorption/release mechanisms. All agree that the cause of the secondary
plume is uncertain and that none of the above theories has been ruled out.
A second, more limited tracer experiment was conducted on July 11, 1980.
The perfluorocarbon, PMCH, was released over a 3-hr period (1900-2200 GMT)
using the same release system and the same site as in the first experiment.
The release amount was calculated to produce concentrations well above the
detection limits at the 100 km arc.
In this experiment, sampling was done only at 100 km downwind of the
release site, using the same array as in the first experiment. As in the
first experiment, the BATS sequential samplers were deployed at sites 13-30
(as compared with sites 12-28 for Experiment No. 1). The tracer release began
at 1900 GMT (2PM CDT) and the samplers were set to start at 2200 GMT (5 PM
CDT) and take nine 45-minute samples.
The PMCH results are plotted in Figure 3-6. The initial sampling period
(2200-2245 GMT) showed concentrations near background at all sampling
locations. The next sampling period (2245-2330 GMT) shows concentrations at
sites 14 through 24 at about 50 times background levels. During the third
sampling period (2330-0015 GMT) peak plume concentrations are reached at sites
14 through 21 with decreasing concentrations to the east. During subsequent
sampling periods, an orderly decrease in the PMCH concentration occurs at all
sampling sites and by the eighth sampling period (0315-0400 GMT) the
concentrations were approaching background levels again.
3-9
-------
000°
010°
Azimuth From Release Site
020° 030°
040°
500
100
50
|
u
T
r
~~r
100 KM Arc
July 11,1980
T
T
1
Sampling Period (GMT)
1 2200-2245
2245-2330
2330-0015
0015-0100
0100-0145
0145-0230
0230-0315
03154)400
I
I I I I I
I
I I III
I
14 15 16 17 Iff 19 20 21 22 23 24 25 26 27 28 29 30
Sampling Sites
Figure 3-6. Average 45-min PMCH concentrations along the 100 km arc from the
Oklahoma experiment Mo. 2 (July 11,1980).
3-10
-------
Since there was no sampling west of site 14, the plume width could not be
determined. Analysis of trajectories suggests that the plume did not extend
much beyond site 14. *
These experiments provide useful case studies for verification of the
eight short-term long-range dispersion models. The time-dependent tracer
concentrations at a large number of points at these distances (100 and 600 km)
for the July 8 experiment enhances the value of verification with these data.
Unfortunately, only two experiments were made and, as a result, no long-term
data base is available which would have provided an excellent opportunity for
a solid statistical evaluation of the models.
3.3.3. The Oklahoma Modelers' Data Base
3.3.3.1. The Meteorological Grid System
The grid system for the Oklahoma modelers' data base was expanded from
that given in the original report by Ferber et al.* Figure 3-7 provides a
sketch of the 24 surface stations and eight upper air stations with respect to
the source and the 54 active receptors. In the original data report, 4 upper
air stations were planned for use based on data listed there. In the region
outlined in Figures 3-1 and 3-2, there were 14 surface stations, although none
were identified in the report. However, by enlarging the region over which
the meteorological wind field models would be run (to that of Figure 3-7), an
additional 9 surface stations and 4 more upper air stations were included.
This expansion of the wind field modeling region should lead to a more
accurate prediction of the prevailing wind field and thereby a more accurate
prediction of plume transport.
The meteorological grid selected for wind field predictions was 1000 km
(E-W) by 800 km (N-S). Development of this grid was based on the following
factors:
(a) As many surface and upper air stations as possible should
be included in regions that might affect transport from
the source to the receptors. In the two Oklahoma cases,
prevailing southwesterly winds occurred. These factors
3-11
-------
LU
Q
I
OKLAHOMA EXPERIMENT
101° W
99° W
97° W
95° W
93° W
LONGITUDE
89° W
Source:
100 Km Arc Receptors:
600 Km Arc Receptors:
Rawlnsonde Stations:
VBAN Surface Stations:
-f"
Figure 3-7.
Location of significant source, meteorological, and receptor
sites for the Oklahoma Cases No. 1 and 2.
3-12
-------
dictated expanding the area from the region presented in
the Ferber report (Figures 3-1 and 3-2) to the area
covered in Figure 3-7.
(b) Several of the models require a meteorological "cushion"
of at least '3 grid points around the region of
concentrat ion calculat ions.
(c) It is not desirable to keep expanding the region beyond
the boundaries of the active region, because the
influence of a surface station or upper air station is
weighted usually by the inverse square of the distance to
the computational point.
(d) Long range transport models that use a grid or mesh are
seldom run with more than 1200-1600 grid points. The
grids used in model development and application usually
range from 30 by 30 to 40 by 40 grid cells. Further,
there is little point in achieving a spatial resolution
that is much finer than the average spacing between
surface weather stations or sources, since conditions and
variations in meteorology cannot be resolved on the finer
scale.
For the Oklahoma experiment, these four criteria were met easily by
expanding the grid to 51 by 41 with 20 km square grid spacing. This choice
yields over 2000 mesh points, and has small enough grid spacing to accommodate
variations between surface stations.
As a result of this expanded meteorological base, more accurate wind
field predictions result from the models, thereby permitting a fairer test of
model dispersion capabilities. A total of 25 surface meteorological stations
were used for the two Oklahoma tests. Of these, only Salem, Illinois had
highly incomplete surface data for the study period (July 7-12, 1980) and
Salem, Illinois had no surface data (only upper air data). This left a total
of 24 surface stations. None of the surface station data were included in the
Ferber data report.
3-13
-------
3.3.3.2. Surface Weather Observations / Modelers' Data Base
NWS surface station data for 24 stations within the selected modeling
grid region were ordered from the National Climatic Center (NCC) in Asheville,
North Carolina. Five of the eight models (MSPUFF, MESOPUFF II, MTDDIS,
RTM-II, and RADM) require surface data as input.
All of the surface meteorological data were acquired from the NCC in the
form of photocopies of the original sheets on which full hourly data were
recorded. Data from such sheets are normally entered into the NCC computer
for the development of the TDF-14 and CD-144 magnetic tapes. However, these
archived NCC tapes only contained 3-hourly data during the study period. By
obtaining only the sheets needed, it was possible to acquire all of the hourly
data. For relatively short study periods, this procedure is reasonable for
any model user to follow, but for periods longer than a week, the effort and
cost entailed may not be justified. The user might then have to rely on the
3-hourly data normally provided by the NCC in the CD-144 or TDF-14 format.
(More recently, the NCC has been archiving hourly data.)
3.3.3.3. Rawinsonde Observations / Modelers' Data Base
Rawinsonde data from Oklahoma City, Oklahoma; Monett, Missouri; Topeka,
Kansas; and Omaha, Nebraska; were presented as part of the original data
base.4 These observations were from July 8, 0000 GMT to July 12, 1200 GMT.
The data included pressure, altitude (m above sea level), air temperature, dew
point depression, wind direction, and wind speed. In addition, special
rawinsonde observations were taken at Tinker Air Force Base about 20 km NNE of
the release site, starting on the morning of July 8. These data (height,
temperature, wind speed and direction) were available in tabular form.*
The upper air data contained in the Ferber report* were also augmented by
obtaining from NCC the computer printouts of the original soundings for
additional times (at stations reported in Ref. 4) and at all times (for
stations not reported in Ref. 4) within the study period. The final study
period for upper air data was selected to run from 1200 GMT on July 7 to 1200
GMT on July 12. Table 3-2 provides the timing of releases, sampling and
periods of computation for Cases No. 1 and 2 for Oklahoma.
3-14
-------
OS
ce
at
O
oc
o
u
CM
CJ
CM
U
CM
U U
~ CM
O CJ
~* CM
U U
-• CM
CJ CJ
-« CM
U CJ
—• CM
U U
"• CM
CJ U
-- CM
CJ CJ
-« CM
U O
—i CM
(J CJ
— CM
CJ CJ
-• CM
CJ CJ
—• CM
U CJ
-H CM
CJ CJ
— CM
CJ CJ
— (M
U U
-> IM
CJ O
-• CM
U U
— CM
CJ CJ
a) o
IV*
W<«
V-*
fi
M-l
4-1
O Ul
5
(U'lH
•3S
UJ
M
<
y
O
t-
i*
3
(L. "^
ZH
oz
uu
0
ui E —
CD
CO
tO
.e
to
•4J
to
tO
o
0) co
CO tO
tO Ol
Ol iH
i-l CU
01 M
>-t
00 -H
>ik ^k
^ P"!
iH rH
3 3
o o *->•->
h M
tO tO >4 I-l
o o
JJ'H'"
•O "Q
O O O O
O O i-! i-l
•-H VO Vl I-l
01 Ol
4J 4J O. D.
T3 CO tO
O i-t i-H
•H co co to to
IJ 4-1 4J C C
cu e c o o
O- Ol Ol i-l -H
e e •>-> j-i
0) Ol CU CO tO
CO M M 4-J -U
CO 3 3 3 3
cu co co p. a.
i-i to to S E
O) CO Ol O O
M E £3 O O
II II H II II
pd ^-l O i-l CM
0 U
,c
u
t-l
o
H-J
CO
•Q ..
O CO
._ 1 1^
T-l ?
»4 0
Ol i-H
aiH
o
bo v-i
c
•H CO
be to
to
!-i CO
01 -H
>
tfl g
IM o
O 73
^
h tO
cu o>
JS M
€ ^
C Ol
-C
•H P
to
u
o •
4-) •-•*
tS
01
f to
H -H
r-~ ys
•• •• ••
x-^
/^ /-N. CJ
CJ O 1-4
W tj 03
to to
s
e s x
x 3.
o
o o o
O O — i
I— 1 \O ^~S
^-^ ^^
01
01 o> co
CO CO tO
to to cu
0) Ol i-J
i— 1 i-l Ol
^ >-. >>
i— 1 iH i— 1
333
"o •-: ^
(N
CO
0)
i-H
I
rtf
u.
O
e
0)
be
o
z
3-15
-------
Of the eight upper air stations employed in this model evaluation study,
data from at least six are available during each 12-hour period. Table 3-3
summarizes availability and periods of record for the eight rawinsonde
locations. During the Oklahoma experiment, special upper air soundings were
taken at three of the eight stations to enrich the detail of the upper air
data during the time period when perfluorocarbon tracer was expected to be
spreading toward the samplers. It was found, however, that the extra
rawinsonde soundings at the special 6-hourly intervals could not be used by
the models. The models, in their present form, accept only 12-hourly
rawinsonde data at all stations. As a result, the extra 9 soundings could not
be used by the models.
3.3.3.4. Meteorological Tower Observations / Modelers' Data Base
Wind speed and direction are available from the KTVY tower located about
40 km north of the release site. The tower is instrumented at seven levels
between the surface and 444 meters. The wind data provided at these levels
were averaged over 15-minute periods. Data from the KTVY tower are available
at heights of surface, 24, 45, 89, 177, 266, and 444 m. The period of
measurement is from 1815 GMT July 8, 1980 to 0200 GMT July 9, 1980. No tower
data are available for later periods of the July 8 experiment or for the July
11 experiment. A surrogate surface station was made from the KTVY tower data.
3.3.4. The Treatment of Missing Data
The treatment of missing data was resolved on a model-by-model basis for
the Oklahoma (and SRP) data based on instructions in the user's manuals. For
MESOPAC, the code handles missing data internally. MESOPAC II, however,
requires all missing data (surface and upper air) to be filled in with
replacement data either by (a) interpolating in time using data at the same
station, or (b) by using data from a nearby station if the period of data loss
was large (more than just a few hours). Subjective judgment was required for
many cases of missing data for MESOPAC II in order to determine the best
alternative. Judgments were made primarily on the basis of how conditions
3-16
-------
Table 3-3. Availability and Periods of Record for the Eight Rawinsonde
Locations for Oklahoma Cases No. 1 and 2.
(1) SAL is Salem, IL — 03879
(2) UMN is Monett, MO — 03946
(3) LTR is Little Rock, AK — 13963
(4) OKC is Oklahoma City, OK — 13967
(5) DGC is Dodge City, KA — 13985
(6) TOP is Topeka, KA — 13996
(7) PEO is Peoria, IL — 14842
(8) OMA is Omaha, NB — 94918
SULODTPO
DATE YYJJJ TIME TIME AMTKGOEM NO. SEQ.
GMT CST GMT GMT GST LNRCCPOA TAKEN NO.
7/07 7/07 80189 1206AAAAAAAA 8 0
7/08 7/07 80190 00 18 A A' A A A A A A 8 1
7/08 7/08 80190 1206AAAAAAAA 8 2
7/08 7/08 80190 18 12 [ ] A [ ] [ ] [ ] A [ ] A 3 3
7/09 7/08 80191 00' 18 [X] A A [X] A A A A 7 4
7/09 7/09 80191 06 00 [ ] A [ ] [ ] [ ] A [ ] A 3 5
7/09 7/09 80191 12 06 [X] A A A A A A A 7 6
7/09 7/09 80191 18 12 [ ] A [ ] [ ] [ ] A [ ] A 3 7
7/10 7/09 80192 00 18 [X] A A A A A A . A 7 8
7/10 7/10 80192 12 06 [Xj A A A A A A A 7 9
7/11 7/10 80193 00 18 [X] A A A A A A A 7 10
7/11 7/11 80193 12 06 [X] A A A A A A A 7 11
7/12 7/11 80194 00 18 [X] A A A A [X] A A 6 12
7/12 7/12 80194 12 06 [X] A A A A A A A 7 13
TOTALS 3 14 11 11 11 13 11 14 88
A = Available and included in the data base
[ ] - Not available
[X] = Not available, but filled in from nearest neighbor rws
for some models
There are 88 available soundings out of 112 possible. With
the "fill-ins" there are still 88 soundings; the soundings
at 06Z and 18Z cannot be used by any of the models.
3-17
-------
were changing for the variable of interest (e.g. wind speed, wind direction,
opaque cloud cover, etc.) at the station and for nearby stations. For the
MSPACK preprocessor, missing data are treated internally within the code.
The following additional adjustments were made due to missing data based
on the requirements of the meteorological processors:
(a) At Salem, Illinois, eight of the eleven total rawinsonde
soundings were missing. MESOPAC II requires a continuous record from any
rawinsonde site that is utilized. The three models that employ MESOPAC II are
MESOPUFF II, RADM, and RTM-II (Oklahoma data only). For MESOPAC II therefore,
it was necessary to augment the three available Salem soundings with eight
soundings from Peoria, Illinois, the nearest station that had upper air data.
f
(b) At Topeka, Kansas, one upper air sounding was missing, and for
MESOPAC II based models, it was replaced by the nearest one from Omaha,
Nebraska.
(c) The addition of one surface station was made. This station was
created from the 10-m level winds of the KTVY tower and the temperature and
cloud data from the nearby Oklahoma City surface station. Only nine hours of
data were obtained from this station, since winds were measured for only a
9-hour period between July 8 and 9, 1980.
(d) In the surface data sheets obtained from NCC, the lowest layer
cloud height was only recorded every three hours, although the ceiling was
noted every hour at most stations. To provide hourly values of lowest layer
height, a modified rule was followed. A given lowest layer height was used
for the next two hours unless additional cloud height data recorded in the
ceiling column indicated an intermediate change. (The ceiling column showed
several layers, and usually reflected the presence of lowest layer clouds in
agreement with the lowest layer height column at three-hourly intervals.)
3-18
-------
3.3.5. The Source Release / Modelers' Data Base
One further issue was the nature of the tracer release in the Oklahoma
experiment. All of the models expect the emissions from a point source stack
with buoyancy or from an area source. At Oklahoma, the neutrally buoyant
tracer was released into the prevailing wind from a nozzle at a height of 1 m
above the ground. The area over which the release was made is far too small to
fit the area source category, and normal type exit conditions did not fit the
tracer source configuration. Exit conditions were devised that gave a large
densimetric Froude number and a small plume rise—on the order of 10 m under
most conditions. This allows the models to calculate a plume rise value
without numerical difficulties. As recommended by the model developers, each
code was modified by resetting the plume rise to 1 m above ground after the
artificial plume rise was calculated.
3.4. SAVANNAH RIVER PLANT KRYPTON-85 EXPERIMENTS5
3.4.1. Description of Experiments
The second data base used for model testing represents observations taken
routinely in the vicinity of the Savannah River Plant during the period
October 1976 through July 1977. Krypton-85 is a noble gas which is released
to the atmosphere during the chemical separation of nuclear fuel from target
material at the Savannah River Plant (SRP) in Aiken, South Carolina. The
release comes from two 62 m stacks located 4 km apart. The release is
strongly time dependent. The krypton-85 release rate is based on calculations
which are estimated to be accurate to within about a factor of two for an
individual hour and to within 10% for daily averages.
The terrain within 150 km of the SRP is represented by gently rolling
hills ranging in elevation from 150 m above sea level to the northwest to
about 25 m toward the southeast. The SRP is covered with mixed hardwood and
pine forests; the surrounding area consists of equal amounts of mixed forests
and cleared farm land.
3-19
-------
3.4.2. Ground-Level Krypton-85 Observations / Study Periods
The krypton-85 samplers were located at 13 sites surrounding the Savannah
River Plant. The nearest sampler was 28 km from the stacks and the farthest
was 144 km. Twice-daily samples representing 10-hr averages were used in this
study for the testing of models.
Fifteen data sets were chosen for model evaluation. The fifteen were the
"standard" data sets defined by the Savannah River Laboratory (SRL) for use in
the Model Validation Workshop held in Hilton Head, South Carolina in November
1980. Table 3-4 lists the time periods over which 10-hour samples were
collected for inclusion in the 15 data sets. At least three data sets are
available for each season of the year. Each data set represented roughly 3-6
days of between one to twelve 10-hour sampling periods. In general, 1-1.5
days of plume transport calculations were carried out in advance of the start
of ground sampling measurements to assure that the plume is sufficiently
transported downwind to be predicted accurately for the same periods as the
samples were measured.
The fifteen periods were defined by SRL after careful evaluation of all
the data measured during the 1975-1977 period of study. The quality of
meteorological and ground-level data, and the presence of sufficient
above-background data readings were key considerations in the choice of these
data sets by SRL. Table 3-4 lists the sampling period and computational period
for each of the 15 SRP data sets.
3.4.3. The Savannah River Plant Modelers' Data Base
3.4.3.1. The Meteorological Grid System
The locations of the rawinsonde stations, sampler locations, surface
stations, and meteorological tower locations are presented in Figure 3-8.
Their locations are superimposed on the grid system used in model testing.
The meteorological grid is 33 by 56 with 10 km by 10 km grid cells (total grid
size is 320 km by 550 km). This is slightly larger than the grid system used
in the previous work with MESOPUFF and RTM-II (see Ref. 23, Appendix A),
3-20
-------
Table 3-4. Sample Collection Periods and Model Calculational Periods for
10-Hr Samples ... Representing 15 Data Sets for the Savannah
River Plant Krypton-85 Data Base.
Sample Collection Periods
Start
Case
No.
1.
2.
3.
4.
5.
6.
7.
8.
9.
10.
11.
12.
13.
14.
15.
Hr
2200
2200
0900
1000
1000
2200
1000
2200
1000
2200'
0900
0900
0900
0900
0900
Date
10-05-76
10-14-76
10-29-76
11-18-76
02-02-77
02-16-77
02-22-77
04-05-77
04-11-77
04-17-77
04-27-77
07-11-77
07-15-77
07-18-77
07-25-77
End
Hr
1200
1200
0700
0800
0800
0800
0800
0800
0800
2000
0700
0900
0700
0900
0700
Model
Calculation Periods
Start
Date
10-06-76
10-16-76
10-30-76
11-20-76
02-04-77
02-19-77
02-23-77
04-09-77
04-16-77
04-22-77
04-29-77
07-12-77
07-16-77
07-20-77
07-27-77
Hr
0000
0000
1200
1200
1200
1200
1200
1200
1200
0000
1200
1200
1200
1200
1200
Date
10-05-76
10-14-76
10-28-76
11-17-76
02-01-77
02-15-77
02-21-77
04-04-77
04-10-77
04-17-77
04-26-77
07-10-77
07-14-77
07-17-77
07-24-77
End
Hr
0000
0000
1200
1200
1200
1200
1200
1200
1200
0000
1200
1200
1200
1200
1200
Date
10-07-76
10-17-76
10-30-76
10-20-76
02-04-77
02-19-77
02-23-77
04-09-77
04-16-77
04-23-77
04-29-77
07-12-77
07-16-77
07-21-77
07-27-77
Note: The total number of 10-hour averaging periods for the SRP data base
is 65. The breakdown is as follows into subcases:
Experiment
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
No. of Subcases
1
3
2
4
4
5
2
7
10
10
4
2
2
5
4
3-21
-------
This region and grid interval choice satisfies the four criteria listed in the
discussion of grid choice for the Oklahoma data previously.
3.4.3.2. Surface Weather Observations / Modelers' Data Base
Meteorological data for the krypton-85 Savannah River Plant experiments
had been placed on magnetic tapes for the region within and surrounding the
Savannah River Plant by the Savannah River Laboratory. These meteorological
data bases will be referred to below.
The Savannah River Plant data base contains surface weather observations
at stations between 100 W and 60 W, 50 N and 20 N. These weather observations
were originally provided to SRL on magnetic tape from the NOAA National
Climatic Center. About 600 surface weather stations report each hour in this
general area. As many as 35 items, mostly meteorological variables, are
recorded for each station. Of these variables, wind speed, wind direction, air
temperature, ceiling, total opaque sky cover, and height of lowest cloud layer
were used in the data base. Figure 3-8 identifies the surface stations within
200 km of the SRP source. Surface weather stations located between 86 W and
77 W, 37 N and 30 M are included in the four tapes which contain the SRP data
base. Ten parameters per station had been added to the original four SRP
tapes by the Savannah River Laboratory (SRL). These ten parameters are the
World Meteorological Organization (WMO) block station number, station
longitude and latitude, station elevation above mean sea level, wind
direction, wind speed, station pressure, dry bulb temperature, dew point
depression, and the previous 6-hour precipitation amount. In addition, the
Pasquill stability category (A to G represented by 1 to 7) as defined by
Turner has been added to each surface observation.
One problem area that required work was the need to append the cloud data
(ceiling, lowest layer height and total opaque sky cover) to each hour for
each surface station. Such cloud data were not on the SRL meteorological
tapes. To acquire the additional cloud data, the full surface meteorology
tapes containing data from 600 stations, including the 19 in the SRP study
region, were obtained from Dr. Roland Draxler of NOAA/ARL. The meteorological
records for each station at each hour as contained in the original SRP data
base were augmented with the needed cloud data to produce new surface
3-22
-------
SAVANNAH RIVER EXPERIMENT
m^w O. DO
till IS\1 IJ'lllllllllllflllllllTII" V '
I I i I I I I m I I i i i I I I i> I I i i i i I I i i i i I
84 W
82* W 81° W
LONGITUDE
80° W
Source: 53
Ravinsonde Stations: Q
Kr-85 Samplers: V
WBAN Surface Stations:
Meteorological Towers:
Figure 3-8.
Location of significant source, meteorological, and receptor
sites for the Savannah River Plant data base (includes final
region choice; inner box is model solution region).
3-23
-------
meteorological records. A number of models use cloud cover and ceiling, but
lowest layer cloud height is used only by MTDDIS. The Air Force surface
station data on these tapes define lowest layer cloud height differently than
they are defined by the NWS. A consistent interpretation of these NWS and Air
Force definitions was developed which was later approved by Dr. I-Tung Wang
of Rockwell International (for MTDDIS). For the NWS data, which was the
prototype data for the development of MTDDIS, the lowest layer cloud height is
usually reported every 3 hours, while ceiling and total opaque sky cover are
reported hourly. The lowest layer height for non-reported hours was obtained
by using the last reported measurement for two more hours. However, the
ceiling entry often recorded multiple • layers, including the lowest layer.
Where available, these data were used in preference to using the last 3-hourly
lowest layer value. In the Air Force data, "lowest layer" is defined as
clouds lying from the ground to a fixed height, while in the NWS data it is
defined as the lowest layer present. The Air Force data reports lowest layer,
middle layer and high clouds. The smallest of the numbers recorded at an hour
for the lowest layer and the largest for the ceiling were used, thus achieving
the closest commonality possible between the definitions.
3.4.3.3. Rawinsonde Observations / Modelers' Data Base
Upper air soundings in the SRP data base are available from four stations
at 12-hour intervals: Waycross and Athens, Georgia; Greenville and Charleston,
South Carolina. Temperatures were linearly interpolated to a wind level where
no corresponding temperature existed. Temperatures at levels where no winds
were available were excluded.
Related to the rawinsonde data is the need for NCC mixing heights as
input to the MTDDIS model. These mixing heights were obtained for the
Charleston, South Carolina rawinsonde station. (For the distance scale of
this SRP data base, the mixing heights for only one station were adequate for
input to MTDDIS.)
3-24
-------
3.4.3.4. Meteorological Tower Observations / Modelers' Data Base
An additional problem area related to whether the available meteorological
tower data should be used or not. It was decided to use these data by adding
seven surrogate surface locations to the 19 actual surface stations.
Meteorological data collected at power plant sites by seven
meteorological towers in the area near the Savannah River Plant were provided
by local utility companies (Carolina Power and Light Co., Duke Power Co.,
Georgia Power Co., South Carolina Electric and Gas Co.) as part of the SRP
data base. The Newberry, SC tower never recorded data during the experimental
period (only before), and the Southport, NC tower is outside the
meteorological grid. In effect, there were only five. With the addition of
two "special towers" described below, that totals seven for use as surrogate
surface stations. The data from the seven meteorological towers were
reformatted to simulate seven surface stations. Added to the 19 actual
surface stations in the grid, the total becomes 26.)
In most cases the measurements on the five power plant towers included
hourly wind speed and wind direction at one to three levels and ambient
temperature, dew point, vertical temperature gradient, precipitation and solar
radiation when available. All power plant towers contained surface level
data.
In addition to the five remaining power plant towers, meteorological data
from the WJBF television tower, 21 km from the SRP source, are included with
the other data as the sixth tower. The TV tower is instrumented at seven
levels between 2 and 335 m above ground with temperature sensors and
turbulence-quality wind sensors. Data at three levels (10 m, 91 m and 243 m)
were averaged over 15 minute periods and tabulated at one-hour intervals by
SRL.
Adjacent to the main SRP operating areas, seven on-site towers with a
wind sensor at 62 m only were located in pine forests within a 10 km radius of
the SRP source. The data from these were combined by SRL to make a seventh
tower due to (a) the close proximity of the on-site towers to each other, and
(b) the problems with the availability of data from all stations at the same
time. To this end, the 15 minute average wind speeds and directions were
tabulated at one hour intervals. Because any individual 62 m tower usually
did not have a continuous record, an hourly arithmetic average of the wind
3-25
-------
velocity from all on-site towers was then calculated from the available data
at one-hour intervals. This arithmetic average was assumed to be
representative of the wind conditions at 62 meters (source area stack height)
at a location midway between the two source areas, F and H. The seven towers
at the SRP are shown in Figure 3-8. In summary then, seven surrogate surface
stations were created for the Savannah River Plant data sets: five from
off-site power plant towers, one from the WJBF tower and one from the combined
seven on-site towers at the SRP.
3.4.3.5. The Source Release / Modelers' Data Base
At the Savannah River Plant there were two identical stacks emitting
krypton-85 located about 4 km from each other. The emissions data that were
available from the Savannah River Laboratory provided only the sum of the
emissions from the sources, and individual source emission data were not
available. • Further, in light of the 10 km grid spacing and the fact that the
nearest sampler location was 28 km from the sources, the two sources were
replaced with a single effective source whose diameter was enlarged from the
actual stack diameter by a factor equal to the square root of two. It would
have been desirable to use a two-source simulation with a 4-km separation.
However, discussions with Dr. Allen Weber of the Savannah River Laboratory
revealed that the separation of emission rates between the two sources is
classified information. The emissions sum was carefully evaluated before
being unclassified to assure that it would not be possible to separate out the
two individual emission rates. The split of krypton-85 emissions between the
two site areas (one stack for each area) of SRP is not to be revealed. The
only calculations done previously with the separate source emission rates were
made for samplers very close to the sources. Those calculations are
classified as well but it is known, however, that the two stacks do not have
the same emission rates. It was felt that given combined emissions data, the
single effective source was a consistent and better choice than the use of
half the total for each of the two separate sources. Any errors introduced by
this combined source schematization should occur only at the nearest samplers.
The actual exit velocity was available and was used, but the exit
temperature was not measured and varies from a few degrees below to a few
3-26
-------
degrees above ambient temperature. A uniform value of 1 degree Centigrade
above ambient was used for exit temperature to provide a slightly buoyant
plume. Although the plume rise calculated using these values will differ
somewhat from the actual rise (within the validity of the formulas used for
plume rise) because of ignorance of the actual positive or negative buoyancy
of the plume, the differences appear to be less than 10% of the rise.
Further, the nearest sampler was 28 km distant, and under most conditions the
plume would occupy most of the mixing layer by the time it had transported to
the first sampler. As a result, this assumption is not considered a
significant source of error.
3-27
-------
-------
SECTION 4
OPERATIONAL EVALUATION BASED ON MODEL/DATA COMPARISONS
4.1. INTRODUCTION
This section provides a summary presentation of the model/data
comparisons made with the eight models for the Oklahoma and Savannah River
Plant data bases. An operational evaluation of the models is presented using
several methods of comparing model predictions with the field data. In
Section 5, a diagnostic evaluation will be presented which is aimed at tracing
back the causes of model/data discrepancies to model assumptions. Additional
methods of testing model performance are presented there to provide added
insight into the causes of model/data discrepancies.
The operational evaluation relies mainly on the use of the American
Meteorological Society (AMS) statistics^ recommended to EPA by the AMS for the
evaluation of the performance of air quality models. Additional quantitative
insight into systematic differences between model predictions and data, is
obtained by examining a set of graphs for each model, which compare predicted
and observed concentration values in several special ways. This section will
provide only the highlights of the numerous graphs and statistical tables that
were generated during the preparation of the model/data comparisons. A more
complete summary of the tables and figures used in the evaluation appears in
Appendices D and E.
In Section 4.2 the results of the AMS statistics are applied to each of
the eight models. Tables of the various statistical measures are presented
separately for the Oklahoma and the Savannah River Plant data sets. The AMS
statistics provide a method of model evaluation commonly used by the EPA in
evaluating air quality models. Additional graphical methods of presenting
model/data comparisons have been developed and are presented in Section 4.3.
These graphs primarily confirm the evaluations based on the AMS statistics,
but provide a visual means of examining the predictions of each model as
compared with the data. They also provide some additional insight into the
performance of the models.
4-1
-------
One class of the AMS statistical measures in Section 4.2 involves a
comparison of the set of predicted concentrations to the set of observed
concentrations, regardless of any time relationship between them. These
•measures involve points not "paired in space and time." A second class
involves "points paired in space and time," which means that predicted and
observed values are compared at the same sampler for the same averaging
period. In general, the set of predicted and observed values was found to
agree best when they are not paired in space and time because the most common
inaccuracy in the predicted plume patterns is a shift in the overall pattern
away from the observed one either in ground position or in time of arrival or
in both respects. In Section 4.3, all of the graphs and tables involve points
paired in space and time.
4.2. APPLICATION OF AMS STATISTICS TO THE OKLAHOMA AND SRP DATA BASES
4.2.1. Introduction
The primary means of evaluating model performance in this report is through
the use of the American Meteorological Society statistics, prepared from
predicted and observed concentration values for each model on a paired basis.
These statistics are described in general form in Reference 3. The AMS
workshop recommended that performance evaluations be based on comparisons of
the full set of observed/predicted data pairs, of the highest observed and
predicted concentration per event (e.g., 1, 3, or 24 hour time period) and of
the highest N values (unpaired in space or time). In addition, if the data
are sufficient, comparisons of observed and predicted concentrations may be
carried out on data subsets representing individual monitoring stations or
selected meteorological conditions.
The guidance provided in the Woods Hole Workshop report^ is quite general
and broad. The statistics were aimed at air quality model evaluation
exercises in general; no specific guidance was given on the application to any
model category such as long-range transport models. However, much has been
learned in the last few years regarding application of the generalized
statistics to the evaluation of the rural4**, urban4^, and complex terrain50
models. Much of that experience is transferable and was used in the
4-2
-------
evaluation of long-range transport models with the Oklahoma and SRP data sets.
The format of the discussion below closely follows that of the previous
EPA/TRC evaluations48-50 using the AMS statistics. In this way, it would be
possible to compare easily the performance of models of different categories,
e.g. urban and long-range transport.
4.2.2. Statistical Data Sets for Comparison of Observed and Predicted
Concentrat ions
Table 4-1 summarizes the data sets to which the AMS statistics in their
most general form may be applied. The data sets listed in Table 4-1 represent
the different types of comparisons recommended by the AMS workshop. To
provide a common basis for the comparison of model predictions and data, it
was necessary to account for background concentration. The background for the
Oklahoma data sets was 3.0 parts/1015 and for the SRP data sets, 16.0 pCi/m3.
The main assumptions made in the creation of the statistical data sets are
described below.
(a) The Oklahoma and SRP model/data comparisons are evaluated separately
in terms of the AMS statistics.
(b) The Oklahoma data sets have averaging periods of 45 minutes
(100 km arc) and 3 hours (600 km arc). Model/data comparisons for
both averaging periods at Oklahoma were not separated for two
reasons. First, it is expected that the usual damping due to time
averaging should not be very significant in comparing 3 hour averages
and 45 minute averages5^. Second, if a split in model/data pairs
were made, there would not be a sufficient number of points in either
set to carry out meaningful statistics.
(c) A similar issue to (b) arises for the Savannah River Plant data sets.
Most of the observed values represent 10 hour averaging periods. For
roughly 10% of all points, the averages were over a different period
with 14 hours the most common alternative. These different averaging
periods were included together with the 10-hour averaging periods in
the statistical data sets since the numbers of points involved and
the differences in time periods were not significant.
4-3
-------
Table 4-1. Summary of Data Sets for Use with AMS Statistics
(Examples Given for SRP Data Base Only).
A. Peak Concentration
(A-l) Compare highest observed value
for each event with highest
prediction for same event (paired
in time, not location).
(A-2) Compare highest observed value
over all 65 experimental periods
at each monitoring station with
the highest prediction for those
65 experimental periods at the
same station (paired in location,
not time).
(A-3a) Compare maximum observed value
for the 65 experimental periods
with highest predicted values
representing different time or
space pairing (fully unpaired;
paired in location; paired in
time; paired in space and time).
(A-3b) Compare maximum predicted value
for the 65 experimental periods
with highest observed values
for various pairings, as in
(A-3a).
(A-4a) Compare highest N (=25) observed
and highest N predicted values,
regardless of time or location.
(A-4b) Compare highest N (=25) observed
and highest N predicted values,
regardless of time, for a given
monitoring location. (A total of
13 data sets.)
(A-5) Same as (A-4a), but for subsets
of events by meteorological
conditions (stability and wind
speed).
B. All-Concentrat ions
(B-l) Compare observed and predicted
values at a given station,
paired in time (a total of 13
data sets).
(B-2) Compare observed and predicted
values for a given time
period, paired in space (not
appropriate for data sets with
few monitoring sites).
(B-3) Compare observed and predicted
values at all stations,
paired in time and location
(one data set) and by time
of day.
(B-4) Same as (B-3), but for subsets
of events by meteorological
conditions (stability and wind
speed) and by time of day.
4-4
-------
(d) In order not to dilute the statistics with numerous points that had
both the predicted value at background and the observed value at or
below background, such pairs were discarded from all statistical (and
graphical) data sets.
In this application of the AMS statistics, no outliers were defined and
discarded. This treatment is consistent with the application of, the AMS
statistics to the EPA evaluation of rural, urban, and complex terrain models.
The significance of outliers was considered to be less important due to the
often counterbalancing effect of allowing both (a) large predicted values with
zero observed, and (b) large observed values with zero predicted. The elim-
ination of any outliers would be inappropriate, in any case, for the category
A statistical data sets (see Table 4-1) since these data sets relate to an
evaluation of the prediction of peak concentrations.
4.2.3. Peak Concentrations
For peak concentrations, comparisons are made to determine model perfor-
mance both on an unpaired basis and for various pairings in time and space.
The first two items in Table 4-1 represent a comparison of the highest
observed and highest predicted concentrations paired in time (A-l) and paired
in location (A-2).
The (A-3a) comparison includes the highest observed concentrations,
regardless of time and space, and predicted values representing different time
and space pairings. Item (A-3b) is directly analogous to (A-3a), but starts
from the highest predicted value. Data sets (A-3a) and (A-3b) were not
examined as part of this evaluation because it was felt that their inclusion
would not provide additional useful information based on the nature of the
database. The data sets (A-3a) and (A-3b) were dropped from further
consideration.
Items (A-4) and (A-5) involve comparisons of the "N" highest observed and
predicted values, unpaired in time or space. The AMS workshop recommended
that such comparisons be based on the upper 2 to 5 percent of concentrations,
rather than on one or two extreme values. In the previous three evaluations
(rural, urban, and complex terrain), this percentage was replaced by a small
4-5
-------
number (N=25) of points. This number was expected to be appropriate in
representing the set of highest observed and predicted values, and at the same
time in still providing a statistical basis for the determination of
confidence limits. In the present evaluation, the minimum of either 25 or 25%
of the observed or predicted values is used instead. In preparing statistical
calculations, any statistical data set with fewer than 10 elements was not
considered further.
One difference between the present and past use of the AMS statistics by
EPA is the neglect in this study of spatial and temporal correlation that may
exist in the data, especially over a period of a few hours. In the complex
terrain evaluations^, employing 1-hour and 3-hour periods separately, the
highest 25 values were screened to eliminate cases with two or more high
values from the same period, or with two consecutive high values at the same
location. That screening was intended to reduce the effects of
autocorrelation and to avoid double-counting a single event. Although some
correlations were anticipated in the Oklahoma and SRP data, it was expected
that auto-correlation corrections would be small. Such corrections^? were,
therefore, not applied. Moreover, it was not expected that serial correlation
would be a serious enough problem in the Oklahoma and SRP cases to lead to a
significant impact on the results. The probabalistic statements often made on
the basis of the AMS statistics have much greater uncertainty than any
correction autocorrelation would provide.
Comparisons of the highest 25 observed and predicted values were
performed for all stations combined (A-4a), but not for each station
individually or for selected meteorological conditions. Further separation of
the data points into meaningful subcategories would produce data sets that
would be too small to provide meaningful statistics.
4.2.4. Comparisons of All Concentrations
As requested by the AMS workshop, comparisons were also made based upon
all observed and predicted concentration values. Item (B-l) is the comparison
of observed and predicted values at a given monitoring station (for all data
pairs where the predicted or observed concentration was above background).
Item (B-3) represents comparisons based on the set of values from all stations
4-6
-------
combined. Item (B-4) was omitted as it represents subsets of (B-3) that do
not contain enough points for meaningful statistics.
4.2.5. Statistical Analysis of Model Performance
The AMS workshop report recommended two somewhat different lists of
performance measures for comparing model predictions with air quality data,
one appropriate for data sets representing pairs of observed and predicted
values (Table 4-2), the other appropriate for unpaired data sets (Table 4-3).
Paired data sets provide a means for assessing how well a model predicts on an
event-by-event basis, whereas unpaired sets do not. Table 4-4 summarizes the
basic list of performance measures, and the statistical methods recommended
for establishing confidence limits on each measure. At the head of each
column (paired and unpaired) are listed the data sets from Table 4-4 to which
each list of measures and statistical indicators has been applied in the
present study.
The data sets from item (A-l) (highest observed and predicted values for
each event) and from items (B-l), (B-3), and (B-4) all represent observed and
predicted values paired in time. For these sets, statistical analyses based
on the residual (i.e., the differences between each pair of observed and
predicted values) are appropriate for measuring model performance. If the
time pairing for these data sets is ignored, however, it is also possible to
assess model performance (in aggregate) by comparing the features of the
composite set of all observed values to those of the predicted values.
Consequently, both paired and unpaired comparisons were recommended by the AMS
workshop for these data sets. Data sets representing comparisons of the
highest 25 values, regardless of time and space, provide no basis for paired
analysis. For these sets ((A-4), (A-5)), only unpaired comparisons are to be
performed. Item (A-2) represents a comparison of the single highest observed
and predicted values from each of the N stations. Only the paired comparison
performance measures were computed for this case. No statistics were computed
for the single-value comparisons in item (A-3).
For paired comparisons, as noted above, the performance measures are
based on an analysis of residuals. Model bias is indicated by the average
and/or the median residual, with a value of zero representing no bias. The
4-7
-------
Table 4-2. Statistics Recommended by AMS for Application to Data Sets
Representing All-Concentration Comparisons.
Highest Highest All Events
per Event per Station Paired in
Paired in Paired by
Time
(A-l)
Number of events A
Average observed A
Average difference A(CI)
Fraction 0 > P x
Characteristic
discrepancies
ad A(a )
RMSE x
AAR x
Correlation
coefficients
Pearson R x
Spearman p x
Kendall T x
Variance comparison x
Maximum frequency A(CI )
difference
A = test is applicable
A(CI) = test is applicable
Location
(A-2)
A
A
A(CI)
A
A(CI)
A
A
A
A
A
A
A(d)
with confidence
Time and
Location
(B-3)
A
A
A(d)
A
A(CI)
A
A
A
A
A
A
A(d)
interval
All Events
at Each
Station
Paired in Time
(B-l)
A
A
A
x
A
x
X
X
X
X
X
X
Subsets of
Events Paired
in Time
and Location
(B-4)
A
A
A
x
A
x
x
X
X
X
X
X
x = test is not applicable
0 = observed concentration
P = predicted concentration
o& = standard deviation
of bias
RMSE = root mean square error
AAR = average absolute residual
4-8
-------
Table 4-3. Statistics Recommended by AMS for Application to Data Sets
Representing Peak Values Unpaired (25 Highest).
Average Average Difference Difference Variance Freq. Dist.
Observed Predicted of Averages of Medians Ratio Comparison
All stations/ A
all events
(A-4a)
By station/ A
all events
(A-4b)
Subsets by met. A
conditions
(A-5)
A A(CI) A(CI) A(CI) A(CI)
A A xxx
A A xxx
A = test is applicable
A(CI) = test is applicable with confidence interval
x = test is not applicable
4-9
-------
Table 4-4. Statistical Estimators and Basis for Confidence Limits for General
Performance Measures Recommended by AMS.
Performance
Measure
Estimator
Basis for Confidence Interval
Paired Comparison
e.g. (B-l), (B-3)
Unpaired Comparison
e.g. (A-l), (A-4a)
Bias
Average
Median
One sample "t," with
adjustment for serial
correlation
Two sample "t"
Wilcoxon matched pair Mann-Whitney
Noise/Scatter
Standard
deviation of
residues
Chi-square test on
standard deviation
of residuals
F test on variance
ratio
Correlation
Gross
variability
Average
absolute
residual
Pearson
correlation
coefficient
None
None
None
Not applicable
Not applicable
Not applicable
Spearman's None
rank-corr elat ion
coefficient, p
Kendall's r
None
Not applicable
Not applicable
Frequency
distribution
comparison
Maximum dif-
ference between
two cumulative
distribution
functions
Not applicable
Kolmogorov-Smirnov
(K-S) test on
versus fpred.
4-10
-------
characteristic magnitude of the residuals is an indicator of the scatter
between observed and predicted values on an event-by-event basis. Three
measures of noise or scatter were computed:
* Variance ? (di - d)2
N-l i
1 T- "5
* Gross variability ~ *: &i
N i
1 _ . .
* Average absolute residual ~ r 'di1
y Hi1
where di is the residual (observed (0) minus predicted (P)) for data pair i, d
is the average residual, and N is the number of data pairs. The correlation
of paired observed and predicted values is measured by the Pearson correlation
coefficient.
For unpaired comparisons, the list of performance measures is shorter.
Model bias is indicated by the difference between the average (or median)
•observed value and the average (or median) predicted value. A ratio of the
variances of the observed and predicted values is provided to indicate whether
the distribution of values in the two data sets is comparable. Similarly, the
frequency distribution of observed values is compared with the frequency
distribution of predicted values.
Standard statistical methods have been used to estimate confidence limits
for each of the performance measures. For paired comparisons, the confidence
interval on the average residual was estimated using a one-sample t test.
This parametric test incorporates the assumption that the residuals follow a
normal distribution; however, for large N, departures from normality are not
critical. Under certain circumstances, serial correlation can affect results
significantly. The AMS workshop recommended the adjustment of confidence
limits for serial correlation. As noted earlier, the correction for serial
correlation has been ignored since (a) serial correlation is not expected to
be a significant problem with the Oklahoma and SRP model/data comparisons, and
(b) the correction for serial correlation is not a complete one in any case.
A nonparametric indicator of model bias, analogous to the t test, is the
median residual. The statistical method of estimating a confidence interval
on the median residual is provided by the Wilcoxon matched-pairs test.
4-11
-------
A confidence interval for the variance of the residuals is calculated
using a chi-square test. No adjustment was made for serial correlation. No
standard method is available for estimating confidence intervals for the gross
variability or average absolute residual measures.
Comparison of two cumulative distribution functions is accomplished using
the Kolmogorov-Smirnov (K-s) test. For this test, the two distribution
functions are compared across the full range of concentration (or residual)
values, and the maximum frequency difference between the two functions is
identified.
For unpaired comparisons, two bias measures are computed. The average of
the observed values is compared with the average of the predicted values. The
confidence interval on the difference of the averages is estimated with a two
sample t test. The median difference is also computed, and the confidence
interval is estimated using the Mann-Whitney nonparametric test.
The variance of observed values is compared with the variance of
predicted values for unpaired data sets. .The performance measure is the ratio
of the variances; the F test provides confidence limits on the ratio. The
frequency distribution comparison for unpaired data sets provides a measure of
the difference between the observed and predicted distribution functions. The
K-S test is again used to assess the statistical significance of the maximum
frequency difference.
4.2.6. Model Performance Results using AMS Statistics
Statistics comparing observed and predicted concentrations have been
generated for each of the eight short-term long-range transport models and the
two data bases. The results for each data base are presented at the same time
for each statistical data set.
(A-l) ... Statistics for Highest Concentration by Event
Statistics were prepared for the data set consisting of the highest
observed and predicted concentrations over the entire monitoring network for
each sampling period, paired in time. For this statistical data set, pairing
is in time and not space. Tables 4-5 and 4-6 present the results for the
4-12
-------
Table 4-5. Statistical Data Set (A-l) for Oklahoma Data (parts per
10^5). (Compares highest observed value for each event with
highest prediction for same event, paired in time, not location).
Model
Number
of
Events
Average
Observed
Value
Average
Difference*
(Obs-Pred)
Standard
Deviation
of Residuals*
Maximum
Frequency
Difference*
MESOPUFF 21 753.1 -437.9 1023.6 0.29
( -903.8, 28.1) ( 783.1, 1478.2) (0.420)
MESOPLUME 21 753.1 -836.1 2181.4 0.24
(-1829.1,- 156.9) (1668.9, 3150.0) (0.420)
MSPUFF
ARRPA
RADM
RTM-II
21 753.1 -3130.1 4955.8 0.33
(-5386.1, -874.2) (3791.5, 7156.5) (0.420)
MESOPUFF II 21 753.1 -363.9 1171.0 0.29
( -896.9, 169.2) ( 895.9, 1691.0) (0.420)
21 753.1 -827.2 2024.0 0.33
(-1748.5, 94.2) (1548.5, 2922.8) (0.420)
21 753.1 -1756.1 2740.5 0.33
(-3003.6, -508.6) (2096.6, 3957.4) (0.420)
21 753.1 404.6 1322.5 0.24
( -197.4, 1006.6) (1011.8, 1909.8) (0.420)
* 95 percent confidence in parenthesis.
4-13
-------
Table 4-6. Statistical Data Set (A-l) for Savannah River Plant Data (pCi/m3).
(Compares highest observed value for each event with highest
prediction for same event, paired in time, not location).
Model
Number Average Average
of Observed Difference*
Events Value (Obs-Pred)
Standard Maximum
Deviation Frequency
of Residuals* Difference*
MESOPUFF
MESOPLUME
MSPUFF
MESOPUFF II 63
MTDDIS
RADM
RTM-II
63 228.4 -104.0 903.5 0.30
( -331.5, 123.6) ( 768.3, 1096.5) (0.242)
63 228.4 -87.5 744.5 0.37
' ( -275.0, 100.0) ( 633.1, 903.6) (0.242)
63 228.4 -54.2 616.9 0.46
( -209.6, 101.1) ( 524.6, 748.7) (0.242)
228.4 -26.1 644.3 0.21
( -188.3, 136.2) ( 547.9, 782.0) (0.242)
63 228.4 -168.0 531.2 0.27
( -301.8, -34.2) ( 451.7, 644.6) (0.242)
63 228.4 -247.8 853.0 0.29
( -462.6, -32.9) ( 725.3, 1035.2) (0.242)
63 228.4 -27.3 489.6 0.19
( -150.6, 96.0) ( 416.4, 594.2) (0.242)
4-14
-------
Oklahoma and SRP data sets respectively. The number of events for Oklahoma is
21 which includes the events on the 100 km and 600 km arcs and data from both
experiments. The number of events for the Savannah River Plant data base
is 63. Two events were dropped from consideration because there were fewer
than three stations reporting measured concentrations above background for
those events. The formula for the 95% confidence interval for the maximum
frequency difference is a function of the number of cases only, 1.36/(2/N).
The value, therefore, is the same (0.420 or 0.242) for all models since the
number of cases is always 63 or 21.
The average differences displayed in both tables indicate that each of
the models for each data base tends to overpredict. The only exception is the
RTM-II model at Oklahoma. For that data set, the large spreading observed
with the RTM-II model (with low peak values) led to the inclusion of more
predicted/observed pairs with lower predicted values. The largest over-
prediction occurs for MSPUFF for Oklahoma and for RADM for the SRP data base.
RADM is also the only model that overpredicts at the 95% confidence level for
both data sets (based on the number of samples available).
The standard deviation of residuals is an indicator of the range of
residual values encountered for each model. The smallest standard deviation
was observed from MESOPUFF at Oklahoma and RTM-II at the Savannah River Plant.
The largest occurred with MSPUFF at Oklahoma and MESOPUFF at the Savannah
River Plant.
Comparisons of frequency distributions of observed and predicted
concentrations ignore any time pairing between predicted and observed values.
The cumulative distribution function f(C) represents the fraction of the data
set (in this case, the fraction of either 21 or 63 data points) with concen-
tration values less than or equal to C. The value presented in this column is
the largest absolute difference between the observed and predicted
distribution functions (for the same concentration value) when the two
functions are compared over all concentration values. The value given in
parentheses is the maximum difference which is significantly different from
zero, at a 95% confidence level, as given by the Kolmogorov-
Smirnov (K-S) test. This confidence interval is a function of the number of
cases as presented above.
There are two ways to interpret the results of the statistical estimator
entitled "maximum frequency difference." First, one may say that the
4-15
-------
difference between the frequency distribution of predicted concentrations and
the frequency distribution of observed concentrations was not significantly
different from zero for all the models at Oklahoma and only for MESOPUFF II
and RTM-II for the SRP data base. Second, this may be interpreted as
indicating that for all the models at Oklahoma and for MESOPUFF II and RTM-II
at SRP, the predicted and observed values in this statistical data set come
from the same distribution at the 95% confidence level.
(A-2) ... Statistics on Highest Concentration at each Station
Tables 4-7 and 4-8 present performance statistics which compare the
maximum observed and predicted concentration values at each monitoring
station. These statistics involve a different number of stations for each
model at Oklahoma and SRP (dependent on the location and spreading patterns of
the models). The fact that at Oklahoma there are only 28 stations available
•for RADM, and 37 and 38 for RTM-II and MSPUFF, respectively, is indicative of
the differences in spreading characterisics and correct positioning of the
predicted plumes simulated by the models.
The statistics provided in these two tables compare observed and
predicted values for the number of data pairs shown in the first column. The
next two columns present the average of the observed concentrations and the
average differences between observed and predicted values. The 95 percent
confidence interval is given in parentheses, as calculated with a one-sample
t test. For the SRP data base, this average difference indicates over-
prediction for all models; the overprediction is significant at the 95%
confidence level for MESOPUFF and MESOPLUME. For the Oklahoma data base,
overprediction occurs for four of the seven models tested (MESOPLUME, MSPUFF,
RADM, and ARRPA). In fact, for RTM-II, the model shows a significant under-
prediction at the 95% confidence level.
The fourth column in these tables displays the fraction of positive
residuals. This performance measure indicates the fraction of predicted/
observed data pairs for which the observed concentration is larger than the
predicted concentration. The results indicate underprediction at Oklahoma for
all models except MSPUFF (50%) and ARRPA (42%). The ARRPA results indicate
overprediction. It is interesting that at Oklahoma, the two measures of bias
(average difference and fraction of positive residuals) lead to generally
4-16
-------
^4
8.
w
td
«j
0) <1>
r~~
i
I
UIOO
> oo
u —
DfU.
UlU O
X O
a: eg u.
K cc u.
4 O UJ
.
(ft
a cc u.
4 O UI
UI U O
a. u
ui _i
Ulf- 4
1933
> a ui
r-Z a O
0440:
o ui 3 a:
0££ C7UI
0 0 <
CC 1-1 3
< r- O
4 > ui
r- UI EC
(no
Z ui en ui
~ > -ice
1-140.
r-U.f-3
U O r-l O A
O CO in
OL i^) ^3
CLO
UICJO
IO Z UI
4 LU a:
a a a.
u ui i
> u. en
4 u. ca
r-l O
UI O
4 > UI
a a: 3
> in 4
: u, f— ri
1044
•~ »~ in
«M to &•
K> to «-
«SI
in CM
CM *-
O 0
in
K>
r-. «»• o
r- CS» CM K)
CM
K)
o
•0 «0 o
o o o
•- in
l>> C9
o r
in
o
K> WS «-
tf »0
•£> in o> vo
CM «- in K>
vO
CM
«- 10 •-«-
CM
CM w
O
in in CM
tr* i~* •&• CM 10 CM in
r«- r» to vo •»• o ^
«- «- «- 10 ^- CM CM 0 «- in -0 «0 CM
0 • rO • S^ • ^~
o ro >± •— in e^ >o
«4> -K> «rO • e^
• IN. •!> .o -
• »»• -CM
co K> -a t> ^ e^
o r» in *o in r»
»-
^> 0
ui in
o o o
m «
o o
CM
o» in * in ro CM o»
•- in in in CM «- r^
o --a- -to -«o -ro .^- .CM
•— i «- i in »o»->o i CM
I •£ cr> in i eo «o
I «- I CSJ w-
* «o o ro '«o**
«n >o K) co »o r«
•- r» M to
K» IO CM IO
.
* it e s
8 £ 8 i
£ £ £ £
4-17
-------
• W 0)
*-> -O «o
o o
S O ui
Ul U O
o. u
(A
W K U.
o: CE u.
tO Ul
Ul K
H-Z K O
O < < (*
o uj n a;
o: z or uj
o o <
o: i-i n
<£ I- Q
Z M OT
•a > ui
>— Ul K
tn o
z ui en ui
o > -in:
i-i w < a.
t- U. K 3
U O l-l O A
< t/1 •-!
o cn w
a. ui co
a: o
*
LU —
LU u a
o s ui
< ui n:
tt ce a.
ui o
o ui
Q: a: Z)
Ul Ul —I
< £0 >
O
: u. t- M
I O < <
: aa.
CJ T-
>o «-
CO O
00
CM
w-
sD
o
CM CM
«O ••—
CM P»
r~ »e> »in »»c
oo * CM • c^ • ^r • in
o
1
I *
I ^»
'O
CM »-
«*•
M3
in
«- ift
in cj
CM M LD
CM
<*
00
K>
•*•
«- >O PJ CM
I*. CM
o in
in <-
CO
rO vO
o «-
•£>»-«-
o« so in
vD
CM
r-»
in
CM
CM
SO
cr-
in
eo
K>
eo
°l
CM
in
10
oo
_ o
in c3 m «— c5 o in
o •- ^>"— O\«-O»-\O^-O»-CMO
CM
oo in \r
P) in oo
•a- CM IO
en
UJ
(/)
in
•- sr
oo co
CM
eo «-;
in
»o
CM
in
in
o >r oo sr •*> *~ CM «— oo i »—
ooir^iCMiroiin
•- r~^ o in in
•— i »- i i i
i i
IO Kl O K) rO K)
CM CM CM CM
*O *O NO »O
K( K)
CM CM
«O sO
-
i. C-
0
.
u. 3 t/> «-i
U. Q- I-H t-l
p O O I
tu OT o 3:
CO Ul h- ^-
£
D
4-18
-------
contradictory findings. At SRP, both measures of bias (average difference and
fraction of positive residuals) indicate overprediction.
The next three performance measures provide estimates of scatter. They
include the standard deviation of residuals (with 95% confidence limits
calculated from the F test), root mean square error, and average absolute
residual. MESOPUFF has the smallest values for Oklahoma, and RTM-II,
MESOPUFF II and MTDDIS have the smallest values for SRP. RADM has the largest
values at both sites. A single large outlier predicted by RADM at SRP may, to
a large degree, be causing the very large scatter indicated by these
statistics.
The Pearson correlation coefficient of observed and predicted concen-
tration pairs and the nonparametric Spearman correlation of ranked sets of
observed and predicted concentrations provide indications of the spatial
correlation of the maximum concentration values at each station. Pearson's
coefficient varied greatly at Oklahoma (-0.05 to 0.91) and at SRP (0.25 to
0.8). The single large outlier for RADM in the SRP data base is the cause of
the poor Pearson's coefficient at SRP; this outlier is effectively removed
upon consideration of the nonparametric correlation coefficients, Spearman's
p and Kendall's r.
The last column in Tables 4-7 and 4-8 (variance comparison) presents the
ratio of observed variance divided by the predicted variance, with 95 percent
confidence bounds in parentheses as calculated by an F test. The variance
comparisons for the highest predicted and observed concentrations by station
generally reveal values less than unity for all models except RTM-II, which
has variance comparison ratios greater than unity for both the Oklahoma and
SRP sites. (For the Oklahoma site both MESOPUFF and MESOPUFF II exhibit
values slightly above unity.) Values of variance comparison and confidence
intervals less than unity indicate the large magnitude and range of predicted
values compared to the observed values.
(A-4a) ... Statistics for Highest 25 Values
Statistics on the set of 25 highest observed and 25 highest predicted
concentrations are presented for each model in Tables 4-9 and 4-10. The first
two columns of results are simply the average of the 25 highest predicted
values for each data set. Comparing the average observed and average
4-19
-------
uT 8
^H £)
O r-l
+•• "a
V4 0)
& tj
&
id *
Jj ti
Q 0)
^ 43
O > —»
«M 5-i .
- I I
fl O *J
•P E
•-i o
•H JC
id co
o to u
•«-i 0) 0)
§ 8
CO *^
0)
>• o
>H 3 UI
X Ou.
< UI U>
2: o: IH
U. O
ui d o
u (n ui
Z M Qi
< QC a.
IH< N
Q: a. (/>
< Z (O
>0 O
u —
UJ * «
Z = S
ui < o;
o: MO.
ui o i
u. uj en
u.s.ea
M o
o u. —
o
X
LU tn —
u u o
z u ui
ui < ac
a: a: a.
UJ LU I
u, > tfl
u.
a.
ID > UI
*I CK 3
o: ui _i
UJ CO <
> CO >
CO
«j o;
UI
> -o •<
o o
•10
> •<
o
in
sj-
CM
K)
oo
in in
K) 0
«o
0 CVJ
10 in
in
W>
r>-
S*£>
-
r^ •&• "T
•CM
K>
K) »
O^
m »-
O
00 CM CM
CM
CM
"- «»•
CM
CM
m i CM i ro
00
«?.
fx CM
O hO
in CM
3*
o »o -o -r«. «r^
•on -
f~» CM
«- CM
1 1
lT. CM \O 1 >— 1 O
CM *~ CM lf> <7>
in «- i xj- «-
1 1 II
r^
CM ro
)iTKirOI
• to . »-
CM
o r^
CM «T- .r« -CM
. m«-«* pjro i
o CM K) i CM r- oo i r-.
op_.o-^in_.^-^. .<=> .K>
oo ^
S«-
in
i
in
0
i
T- oo
<7> Cf
K> &
ITl CM
II
K>
•a-
CM
O>
00
-
KJ CM CM Op «-
vo CM «~ in KJ
oo
O.
O
oo
rs.
^
o
00
oo oo
C?>
O
oo
in
CM
m
CM
m
CM
in
CM
u. z:
u. 3
_i u.
a. u.
O
co
.
3 M
a. «-i
O 1
to z
in
CM
5C
o
m
CM
cc
a.
UJ
a.
4-20
-------
I in
M <>
O
2
0)
Ill
Ul » -«
omo
2 Ziu
UK K
oe t-to.
ui o i
u. uj en
in c*
•- -oo • <* -«
^- -CM • *~ • 1 • i r* i •• «~CM*~ i m i in
*
ui en —
yujo
Z to uj
uj < oj
ocata.
u uj i
u. > en
s<§
o; t-i-j
u/ a
UJ UJ
o > ui
< 0:3
a: uj _j
uj en <
en
.
O
m
n«- o »~
OO W5 «- in «- CM O
>J -CM «
-------
predicted value indicates that all models except RADM (both data sets) and
MSPUFF (Oklahoma only) are within a factor of two of the high 25 observed
data. The first performance measure, presented in column 3, is the difference
between the two averages. Negative values exist for all models except RTM-II
for Oklahoma; the negative values indicate overprediction. The confidence
intervals being negative as well for MESOPUFF, MESOPLUME, MSPUFF, RADM, and
ARRPA at Oklahoma indicate that at the 95% confidence level, these models
overpredict with respect to this statistic. The confidence intervals were
determined using the two-sample Student t test.
The second performance measure relating to average values is the median
difference (313th highest value) between all 625 possible pairings of the
25 highest observed and predicted concentrations. The 95% confidence interval
is determined with the nonparametric Mann-Whitney test. Results for the
median difference are very similar to those for the difference of averages.
Here, all models exhibit overprediction for both data sets. With all models
for Oklahoma (except RTM-II), overprediction is significant at the 95%
confidence level.
The third performance measure is the variance ratio. The variance of the
25 highest observed values was divided by the variance of the 25 highest
predictions. The F test was used to calculate the 95 percent confidence
levels for these comparisons. Results indicate that, at SRP, the scatter of
the 25 highest predictions is much larger than the scatter of the 25 highest
observations, except for the RTM-II predictions. For Oklahoma, the scatter of
the 25 highest predictions is much larger than the scatter of the 25 highest
observations for the MESOPLUME, MSPUFF, RADM and ARRPA models. However, the
MESOPUFF, MESOPUFF II and RTM-II models exhibit more scatter in the 25 highest
observations than in the 25 highest predictions. The RTM-II model is extreme
in having less scatter of its predictions than the observations indicate.
The last performance measure presented in Tables 4-9 and 4-10 is the
frequency distribution comparison. The confidence interval is a function of
the number of cases. The value is, therefore, the same (0.385) for all
models, since the number of cases is always 25. For Oklahoma and for all
models, one can reject the hypothesis that the predicted and observed peak
values came from the same distribution (at the 95% confidence level). At SRP,
the opposite is true (except for RADM and MTDDIS); one cannot reject the
hypothesis that the peak values came from the same distribution.
4-22
-------
(B-3) ... Statistics for All Concentrations Paired in Space and Time
Tables 4-11 and 4-12 present the comparison of all observed and predicted
concentration values paired in time and space for the Oklahoma and SRP data
bases, respectively. Results for Oklahoma show that in terms of average
differences, an overprediction exists for all models except RTM-II. MSPUFF
and RADM show significant overpredictions at Oklahoma and the RADM model does
so again at SRP. For the MESOPUFF, MESOPLUME, MSPUFF and RADM models at
Oklahoma, and for the MTDDIS at SRP, the entire confidence interval for
average differences is negative. This indicates overprediction at the
95% confidence level for those four models at Oklahoma and for MTDDIS at SRP.
Values for the fraction of positive residuals, the standard deviation of
residuals, the root-mean-square error, and the average absolute residual all
exceed the average observed values. The largest values for the measures of
scatter at Oklahoma occur for the MSPUFF and RADM models. The RADM model has
the largest values for the measures of scatter at SRP. Significant scatter
exists for all models at the 95% con-fidence level.
Correlations of observed and predicted concentrations are extremely low
and negative in many cases. Variance ratios indicate that variances for
predicted values are generally much greater than variances for observed
values. The only exception is RTM-II (for both data sets) with a variance
comparison exceeding 1.0.
4.3. GRAPHICAL EVALUATION OF THE EIGHT MODELS WITH THE OKLAHOMA AND SAVANNAH
RIVER PLANT DATA
Graphical comparisons of predicted and observed concentrations for each
data base are used to provide insight into systematic differences between
model predictions and data. Four useful types of graph are:
(a) scatter plots of observed and predicted concentrations for all 15 SRP
data sets and (separately) both Oklahoma data sets. Points paired in
space and time are used in this scatter plot. A related graph
presents a scatter plot of residuals (observed minus predicted)
versus the average of the predicted and observed value for that pair,
4-23
-------
^ -g
in
is .
Q, W
CO O
«_ w
0)
rH
•8
E-"
Ul
Z U 2
Z ui or
I-* ID Ul
X O U.
< Ul u.
z a: M
u. a
i_
CJ CO Ul
s: w or
or a. co
>S2
_J
< or u.
Z o ui
Ul U O
^ u
z
z or u.'
< O Ul
Ul CJ O
a. u
CO
en or u.
or o: u.
< o ui
u u O
o. u
UJ _l
Ul 1- <
< _) Q
or o M
UJ CO CO
> O ui
< _i or
>-i i-i<( a.
>— u. »— 3
CJ O w O A
< co 1-1
or o co co
a: o
*
UJ CJ O
CD Z Ul
or or a.
Ul Ul t
> U. CO
r>s oo o xr o oo
f- 0 0 CM CM •- 0
o o o o o o o
1 1
oo r^- in xr CM «- m
o o o o o o o
1 1 1
xr CM in «- o ro CM
o o oo r^ m in o
o oo o oo in o sr
CM »o fs. in 1*1 CM so
o ^o so r** r^ CM *^
NO o CM «- r* in in
«- CM «~ CM «-
cr* *"" in sf firt o ^*
%r in CM o so o o
o in «- in CM ro «- xo K>CMO»-
•^ -vr .0 -r* -co -fJ "•"> •
*— ^-inCM^*^"r^ ^-CMC?*"^
so CF* o ^ r*^ CM ^
CM «- CM «-
vT «- CM O «- «- 00
in «o in in \r so in
O O O O C3 O O
10 so ^D so v7 r^ rs>
r^ ca CM in o
i i
o 1-1 in o vr o in
- 'O* 'PS •<> • 'CM ' O •
CM d OO CM 1 0 xT
1 1 1 1 «— 1
1
i*» rsl r-J «0 o* -a- «0
vT in Kl %O K) CO sO
CM CM CM CM CM CM £3
CM CM CM CM CM «~ CM
1-1
h-t
Ul
u. Z u.
u. 3 u.
3 — J u. 3 t-t
n. o. u. o. u-t <
o o 3 o i z o,
co co o. co z a or
uj ui to ui >- «t a:
c £ z z or or «t
4-24
-------
Cti i— I
5rH
(0 rd
Q
-P -P
c (d
SS
fel
Savannah Riv
predicted v
/ •
G
•o o
J-i C5 -in
O (d -P
4-1 (0
*rt o
^ ® -H
m J-i -a
1 M -P
W 0)
b c
aST
SB'S
SoS
v-*.H
S.&
••H/— . CO
-pm c
w s o
•H X. -iH
-P -H -P
5HS
W— ' W
•
CM
rH
4.
CD
rH
•a
E-<
UI
r-H Z> LU
~i
u. O
*
UJ O O
<_) en uj
iS£
F-l< N
a: cu in
> 0 0
u —
_i
_j • •
< cc u.
Z O LU
uj <_> O
^ U
< • •
a: cc u.
4 O UJ
ui <_> O
0. u
in
in ex u.'
o: K u.
< o ui
UJ O O
0- 0
UI — 1
UJ h- CD ui
<
to
Z —i
o o <
CC M O
§< M
1-1 en
< > ui
r- UI CC
en Q
u.
o
z ui en ui
0 > -1 K
M M < Q.
I- u. t- 3
O O M 0 A
< en i-i
eg o in en
IX 0
*
m —
UI U O
tf\ ~^ i.t
ct o: a.
UI UI 1
> u. in
< u. ro
S2
CO UJ
Ct £K 3
UJ UI _l
> IS)
o:
ui en
"u. ** K
r> o y«~
00000000000000
ro e\j o> 10 »• >3- 00 0
o o e o o o o
>O sT K) o «— K» «O
O ^ (M C3 K> O «-
o ea o o o o o
i i i
00 *f O »- K> Ift *-
o o ro o >3- o (NJ
o o o o o o o
i i i
r> in CNJ «- in v if>
o o o cs ro o CJ
es ^ o o o o o
i i
K) K) CJ «- 00 0 (M
T- «- »— r- «— K>
«- ^) o oo «0 K> W
CM ro in eo »- f-» irt
sj- <7> 0 CM !>- <7> CO
sT K) K) K) tM e\) 0
CM
•j- o co <* ui ro >o
o eo r^ in «o M r-
oo CM PJ tn 55 rj «-
>T >f •&• fO CM IO <7-
«o «- •* CM r» i^ -J-CM
CMc\i'<*T-f.»jo«-m
. o . (\) .»o . <7« .CO •
>J-OrOlfl»Ol><.IO«OC\IK)CM^5'OO
«- o <>o o in r^pjco
±
CM
>J- CO 0 K) CO »O sj-
r^ r^ r«- in CM -a r^
o o e» o o o o
o ^ c\j •— <»• r~ r-
r^ «- c\j >o «- r^ •a-
PO K> >O CNJ «- CM CT>
1
«* o PJ o sj- in a'v»'OOrO^-
»-.«—.«-.! .f»j.| .^-.
»-ir-iM <» i «— «opj'-
^ m m si- ^ K> i tM
i i i i > i in
i
iTl \O CM K> .0 -3- CO
o •— in >± co o o \o ^ -a t~-
•M r-. rs c\J CNJ »- \J-
r- O K> ^- in K> <7>
»-4
i«i
S
3 u.
_i u. zi in >-i
Q. U. O. r-l IH
0300 1 £
en a. en o £ o
ui en ui »— »— «X
s: z ac z CK ct
4-25
-------
(b) frequency histogram of residuals for each model and each data base.
Points paired in space and time are used to determine residuals.
This type of graph illustrates whether a balance exists between over-
prediction and underprediction,
(c) frequency histograms showing the distribution of predicted values and
observed values for a model. The determination of whether the model
predictions and observed data (unpaired) have roughly the same
frequency of high and low values is of interest, and
(d) cumulative frequency distribution of predicted and observed concen-
trations over all data sets for the SRP and (separately) the Oklahoma
data bases. Points paired in space and time are used to create the
residuals here as well. This kind of graph is related to the
Kolmogorov-Smirnov statistic (presented in Section 4.2) which is
aimed at identifying whether the predicted and observed distributions
are statistically the same.
A complete set of these graphs for both data bases and all eight models
appears in Appendix E. A subset of those graphs are presented below which
illustrate special or unique facets of the performance of the models. The
full set of graphs in Appendix E support even more strongly the character-
istics described below.
4.3.1. Scatter Plots
The scatter plots for all eight models are presented in Figures 4-1 to
4-7 for the Oklahoma data base. Each box represents one predicted/observed
pair; the horizontal axis is the observed value in parts per 1015, and the
vertical axis is the predicted value in the same units. Perfect model/data
agreement is represented by the dotted line. A point on the vertical axis has
a nonzero predicted value, but a zero observed value. A point on the
horizontal axis has a zero predicted concentration and a nonzero observed
concentration. For the Savannah River Plant data base the scatter plots for
the eight models are presented in Figures 4-8 to 4-14, where units are in
pCi/m3.
Each scatter plot has most of the predicted/observed pairs on the axes.
Predicted values greater than zero (background) are often associated with
4-26
-------
OBSERVED VS PREDICTED CONCENTRATIONS
MESOPUFF MODEL - OKLAHOMA
i
* 1000
BNTRATION (PAH
§
1
Q tO
»•*
1
1
0 QO •(' 0
o X°
• o «."'•' °
J*
o o *3°D ir'"' o
0 " ° X' o
°o ° B X °
»'' a
° 0
o
xx' -
s'£ °
a
10 too , 1000 , 10000
OBSERVED CONCENTRATION (PARTS PER 10"15)
Figure 4-1. Scatter plot of observed and MESOPUFF predicted concentrations at
Oklahoma (points paired in space and time).
OBSERVED VS PREDICTED CONCENTRATIONS
MESOPLUME MODEL - OKLAHOMA
*4
b
K
U
^ tooo
1
fe
S5
0
S 100
g
fe
2
0
U
g l°
^
a
j,
i
a a g
D a ''' o
o ,-''
a o x °
ft*
a Q °a ^** °
O n D *
°o 8 .•' °
a ° o 9-''
a
-•••'''a <* °
a '"'
° .-• a " °
x'
etr'
,' a
o x'
x'
10 100 1000 10000
OBSERVED CONCENTRATION (PARTS PER 1O*"15)
Figure 4-2. Scatter plot of observed and MESOPLUME predicted concentrations
at Oklahoma (points paired in space and time).
4-27
-------
OBSERVED VS PREDICTED CONCENTRATIONS
MSPUFF MODEL - OKLAHOMA
§
b
u
•*• 1000
B
1
0
6 lo°
g
§
I
0 ">
1
t
0 °
o *'
o X
o o _ ,-
n O " *'
0 00 ,-'0
° 0°
On 00
0 0 ° ° 0
O ° _ ?''
B °x °
a oX
° oo ,Xo
o o „,' o
O _ ,' O
0 °° ,X' o
o ..'•'' °
.'"'' a
,.-''
10 100 . 1000 v 10OOO
OBSERVED CONCENTRATION (PARTS PER 10-15)
Figure 4-3. Scatter plot of observed and MSPUFF predicted concentrations at
Oklahoma (points paired in space and time).
OBSERVED VS PREDICTED CONCENTRATIONS
MESOPUFF II MODEL - OKLAHOMA
*4
L
i
*• 1000
£2
«
o.
I
P 100
^!
p
z
s
2
O
Q lo
_
Q
0.
1
1
1 0 _,-''
°0 ° o - ° .,'••'"
10 0 0 ..'
o o 03 a..'
° °° *a 0 n/
O n ^^ ^^
a .a- a
o o o x'''D,5
0° D .''O °
o 4*0
Q Q ,,« ^ O _
o a n''
a
o
° Xo 0° a o
o ,s a
,
-------
OBSERVED VS PREDICTED CONCENTRATIONS
ARRPA MODEL - OKLAHOMA
1
NTRATION (PARTS PER
i J
B
q to
S
t
D "
i o °
' o 0 o a
a
a 4 o
° ° ° 0° .••'''
-------
OBSERVED VS PREDICTED CONCENTRATIONS
RTM-II MODEL - OKLAHOMA
3
&>
*• 1000
&
o
5 too
o
u
a 10
M
1
1
ff''
fif
, •« .00 /
- ° ?
-------
OBSERVED VS PREDICTED CONCENTRATIONS
MESOPUFP MODEL - SAVANNAH RIVER PLANT
f?
I
CONCENTRATION (PCI/
i I
G to
i
i
D ,''
0
O
a a a
a a
° ° a ° ••''°
a .•'' ° a
O o-'
0 ,a a a
o /" -
° a
o oa ,-' a °
a ° ,,-'' oo a
a *'' D
o> o° .-••'' ° ° o °
aa a -•' 03 ^
°-* °
j.'' „ ° a
10 100 10OO 10000
OBSERVED CONCENTRATION (PCI/M"3)
Figure 4-8. Scatter plot of observed and MESOPUFF predicted concentrations at
Savannah River Plant (points paired in space and time).
OBSERVED VS PREDICTED CONCENTRATIONS
MESOPLUME MODEL - SAVANNAH RIVER PLANT
«"
Eg
S- 1000
o
1 tofl
8
D 10
a
I
i
i
a
o
0 D
° ,'"'
o o ° •'
/a °
.• o
" ° ,H'° a
a a " ,.-'a
^° o X
° .-' o ° o a
o a o a,-'
.•'
v'°
,•'' °
10 100 looq 10000
OBSERVED CONCENTRATION (PCI/M''3)
Figure 4-9. Scatter plot of observed and MESOPLUME predicted concentrations
at Savannah River Plant (points paired in space and time).
4-31
-------
OBSERVED VS PREDICTED CONCENTRATIONS
10000
c
f loo°
i
1 io°
8
G 10
S
u
g
1
MSPUFF MODEL - SAVANNAH RIVER PLANT
o
o °
o
o
o o S
o - ' x."'
a .•' a
i o o ° x as o
O y' °
OH ° ,'*'
0°
0 O 0 ,.'' 0 °
0 ° .4
0
.•'' a o
0°
10 100 , IOOO 10000
OBSERVED CONCENTRATION (PCI/M"3)
Figure 4-10. Scatter plot of observed and MSPUFF predicted concentrations at
Savannah River Plant (points paired in space and time).
OBSERVED VS PREDICTED CONCENTRATIOMS
MESOPUFF II MODEL - SAVANNAH RIVER PLANT
5|
S- 1000
Z
o
p
1 to°
2
8
|
G 10
S
g
t
1
a ° °
0
a
a "
•° '• „,.•••""
On •• »
n B O '*
/a _ o ° Ox' o °
,°0 ° °o Qx' „ ^
a .-'
° . ° '*' _°0 n n D
• 0 0,,'' 00 ^
l°°0 ° o 0 .''''* °
0 "
M ,* O O
D f _ W
** o o
0 X' tt
,-' ° 0 0
D
D a
10 100 IOOO 100OO
OBSERVED CONCENTRATION (PCI/M'^3)
Figure 4-11. Scatter plot of observed and MESOPUFF II predicted
concentrations at Savannah River Plant (points paired in space
and time).
4-32
-------
OBSERVED VS PREDICTED CONCENTRATIONS
MTDDIS MODEL - SAVANNAH RIVER PLANT
f
j? t000
I
5
E 100
W
2
8
8
b to
Q
u
«
D,
1
1
D 0 f.-''
o a,-''
§ o°8 a an° 0 /'"
°o a
a a o °o jo a * ° o*' °
a n S 8 aa a o /' n °
aa n.n a ia a Hi tfd a
'r T3 a ifp ,' a
& a ° o a a F' DO a
aa ° a * * -''' ° H° °
O, D ° .' u 0 D
* ana a a a °a a.-''6lSD D
a o ^^ -'' D „ ^ a
n _' °nn
i a -a D D
o a ,,-'o °
D '' ^
a
,O^ o
.-0 0
10 too loog 10000
OBSERVED CONCENTRATION (PCI/M"3)
Figure 4-12. Scatter plot of observed and MTDDIS predicted concentrations at
Savannah River Plant (points paired in space and time).
OBSERVED VS PREDICTED CONCENTRATIONS
RADM MODEL - SAVANNAH RIVER PLANT
P"
§
yt
Q 1000-
o
2
O
P
2 10°
O
2
8
b 10
a
u
ffi
0.
t
i
o ,.''
0 ° „
0 °
O rj
a °^ °
0 a
o ° a
a o ° o .''
B a *•' o
HO M n O O *""
o °S ° °8 ° "o .••'
°0 ° ° : /"-„ , -
o a ,.-' ° *
'* D
0 0
0 o " /'°
a ,"''
'"' ^ D
a
*'' O
fe
n,-' 0
,' °
10 100 10OO 1OOOO
OBSERVED CONCENTRATION (PCI/M"3)
Figure 4-13. Scatter plot of observed and RADM predicted concentrations at
Savannah River Plant (points paired in space and time).
4-33
-------
OBSERVED VS PREDICTED CONCENTRATIONS
10000
0 100°
CONCENTRATION (P
i
Q
b 10
i
i
i
RTM-II MODEL - SAVANNAH RIVER PLANT
" ,""""
o
a ° o a a
a a
oo ° 0 /^
o° ° O|D./ ° a
8 o./ °
• • • °-*x . • . •
0 n 5fe °
X* D ° a "
= «
,-'' ° D CP o
00 .''' D 0
^»' O . O
a n
to 100 ~ looq 10000
OBSERVED CONCENTRATION (PCI/M"3)
Figure 4-14. Scatter plot of observed and RTM-II predicted concentrations at
Savannah River Plant (points paired in space and time).
4-34
-------
observed values that are zero and vice versa. This is clear evidence that the
predicted and observed plume patterns on the ground are offset from one
another. Table 4-13 summarizes these behaviors.
Table 4-13. Comparison of Predictions of the Models Based on Percentage of
Pairs where Observed Values are Greater than Predicted Values
(points paired in space and time).
Model Oklahoma SRP
(% of pairs where obs > pred)
MESOPUFF
MESOPLUME
MSPUFF
MESOPUFF II
MTDDIS
ARRPA
RADM
•RTM-II
69 -
56
50
61
*
54
64
42
74
'78
70
53
28
*
63
74
* Model not applied to this data base.
As may be seen, most of the percentages are larger than 50% indicating
that the plumes have generally not spread out over the observed data due
either to (a) an underestimation of lateral spreading of the predicted plume,
and/or (b) a poor location of the predicted plume with respect to the observed
plume. The situation where a predicted plume underestimates spreading may be
viewed as one where the observed plume is expanding wider than the predicted
plume, leading to many points where the observed values are greater than zero
and the predicted value are zero.
It is interesting to note that the RTM-II model for Oklahoma (with a 42%
value) tends to overpredict plume spreading, whereas for SRP (with a 74%
value) it tends to underpredict spreading. An examination of the complete set
of contour plots (Appendix E) reveals this to be the case. The cause of this
4-35
-------
behavior is simply in the choice of the 77-parameter (and its upper and lower
bounds) in the lateral diffusivity for RTM-II. In this study predictions of
horizontal spreading were found to be sensitive to the choice of these
parameters. Apparently, too large a value of rj and its lower bound were
selected by the model developers for Oklahoma and too small a value of 17 and
its upper bound were recommended by them for the SRP runs.
The other interesting result is for MTDDIS, which predicts that 28% of
all predicted/observed pairs (points paired in space and time) have
observations greater than predictions. One may infer from this comparison
that the MTDDIS predicted plume overpredicts horizontal spreading. That
conclusion will be supported further by the results of the pattern-comparison
studies in Chapter 5.
The general tendency to underpredict spreading shown in Table 4-13 also
coincides with the general 'tendency noted above of the models to overpredict
peak values for the Oklahoma and SRP data sets.
4.3.2. Other graphical comparisons
The other types of graphical comparisons presented are the following:
(a) frequency distribution of predicted and observed concentrations,
(like Fig. 4-15)
(b) cumulative frequency distributions of predicted and observed
concentrations, (like Pig. 4-16)
(c) frequency distribution of residuals for points paired in space
and time, (like Fig. 4-17)
(d) Average of observed and predicted concentrations versus
residuals, (not discussed, like Fig. E-99)
Conclusions drawn from these four types of comparison are surprisingly
similar among the models for a given data base. A short summary of trends
follows:
4.3.2.1. Oklahoma Data Base
Table 4-14 provides a summary of the behaviors of the models for the
Oklahoma graphical comparisons. In this Table, "P" represents predicted
4-36
-------
FREQUENCY DISTRIBUTION OF PREDICTED AND OBSERVED
CONCENTRATIONS - RANGE: 0 - 100 PARTS PER 10"15
MESOPLUME MODEL - OKLAHOMA
EZ1 OBSERVED
K3 MESOPLUME
30 40 SO 60 70 SO
CONCENTRATION (PARTS PER 10"1S)
«o
Figure 4-15.
Frequency distribution of predicted and observed concentrations
at Oklahoma for MESOPLUME based on points paired in space and
time ... concentration range: 0 to 100 parts per
TO
CUMULATIVE FREQUENCY DISTRIBUTION
OF PREDICTED AND OBSERVED CONCENTRATIONS
MESOPLUME MODEL - OKLAHOMA
CONCENTRATION
Figure 4-16. Cumulative frequency distributions of MESOPLUME predictions and
observed concentrations at Oklahoma based on points paired in
space and time.
4-37
-------
FREQUENCY HISTOGRAM OP RESIDUALS
MESOPLUME MODEL - OKLAHOMA
40-
fe*"
E 20<
10-
i ft.
V7/\ T77* WW//
%
''//.
-IDO-aO -80 -70 -8O -50 -40 -30 -80 -10 0 10 21
v/<
D 3
'//,
a «
^77\ X//\ vtjvsf
0 50 <0 70 80 00 100
RESIDUAL (OBSERVED - PREDICTED)
Figure 4-17. Frequency distribution of residuals at Oklahoma for MESOPLUME
based on points paired in space and time ... residual range:
-100 to 100 parts per 1015.
FREQUENCY DISTRIBUTION OF PREDICTED AND OBSERVED
CONCENTRATIONS - RANGE: 0 - 100 PARTS PER 10"15
MESOPUFF MODEL - OKLAHOMA
M-
TO-
M-
bM
30-
EO-
10
0
(
Z21 OBSERVED
ESI MESOPUFF
y////
1
•
•
m
m
vvvvv
) 10 Z
0 30 4O 50 00 70 80 M 100
CONCENTRATION (PARTS PER 10"15)
Figure 4-18.
Frequency distribution of predicted concentrations at Oklahoma
for MESOPUFF based on points paired in space and time ...
concentration range: 0 to 100 parts per
4-38
-------
FREQUENCY DISTRIBUTION OP PREDICTED AND OBSERVED
CONCENTRATIONS - RANGE: 0 - 100 PARTS PER 1(T*15
MESOPUFF II MODEL - OKLAHOMA
OBSERVED
S3 MESOPUFF II
30 40 SO 60 70 80
CONCENTRATION (PARTS PER 10"15)
M 100
Figure 4-19. Frequency distribution of predicted and observed concentrations
at Oklahoma for MESOPUFF II based on points paired in space and
time ... concentration range: 0 to 100 parts per
FREQUENCY DISTRIBUTION OF PREDICTED AND OBSERVED
CONCENTRATIONS - RANGE: 0 - 100 PARTS PER 10"15
M-
80-
TO-
FREQUENCY
£ 8 8
30-
ZO-
10-
0-
(
RTM-II MODEL - OKLAHOMA
E3 OBSERVED
E3 RTM-II
W?
m
m
1
•
^^^^C(i(i^^Avvv^jl'>X»Xi'.v«.vvi"^J^AJu ••""
) 10 ZO 30 40 SO 80 70 80 90 100
CONCENTRATION (PARTS PER 10"15)
Figure 4-20. Frequency distribution of predicted and observed concentrations
at Oklahoma for RTM-II based on points paired in space and
time ... concentration range: 0 to 100 parts per 1015.
4-39
-------
CUMULATIVE FREQUENCY DISTRIBUTION
OF PREDICTED AND OBSERVED CONCENTRATIONS
RTM-II MODEL - OKLAHOMA
100
BO
•o
70
M
M
40
30
20
10
OBSERVED
RTM-II
to too 1000
CONCENTRATION (PARTS PER 10"15)
IOOOO
Figure 4-21. Cumulative frequency distributions of RTM-II predictions and
observed concentrations at Oklahoma based on points paired in
space and time.
FREQUENCY HISTOGRAM OF RESIDUALS
so-
40-
FREQUENCY
H> M U
0009
RTM-II MODEL - OKLAHOMA
7771 , VX
//%S/W//fy77y//y//9(//h-rr!777///
I
10O-BO -«0 -70 -«0 -90 -40 -30 -ZO -10 0 10 80 30 40 50 «0 70 80 M 100
RESIDUAL (OBSERVED - PREDICTED)
Figure 4-22.
Frequency distribution of residuals at Oklahoma for RTM-II based
on points paired in space and time ... residual range:
-100 to 100 parts per 10*5.
4-40
-------
Table 4-14. Summary of Results of Graphical Comparisons of Models with
Oklahoma Data.
Model Frequency Distribution
of P and 0
(trend for 0-10 range)
Cumulative
Frequency Plot
(initial trend)
Frequency Histogram
of Residuals
(position of peak range)
MESOPUFF
MESOPLUME
MSPUFF
MESOPUFF II
ARRPA
RADM
RTM-II
0 :
P :
P :
P :
P :
P :
0 :
> P
> 0
> 0
> 0
> 0
> 0
> P
P :
P :
P :
P :
P :
P :
0 :
> 0
> 0
> 0
> 0
> 0
> 0
> P
0
0
0
0
0
0
-10
- 10
- 10
- 10
- 10
- 10
- 10
- 0
concentrations in parts per 1015 and "0" represents observed concentrations in
the same units.' Graphs like Figure 4-15 were examined to provide the first
column, which reports whether the number of predicted ground concentrations
exceeded the number of observed concentrations in the range of 0-10
parts/1015 (the "initial trend" of the frequency distribution). This entry
determines whether a model predicts as many nonzero small concentrations as
were observed. The cumulative frequency plots, like Figure 4-16 give the
number of predicted and observed concentrations that were equal to or less
than a given value on the horizontal axis. They indicate in what range of
concentration values overprediction is occurring, and in what range the model
tends to underpredict. The "initial trend" of this plot encompasses roughly
the range of concentrations from 0 to 50 or 100 parts/1015. The histogram of
residuals (excess of observed over predicted without regard to magnitude of
concentration) has a strong peak near zero for all models. Figure 4-17 is an
example of this type of graph. Whether this peak occurs for small positive
residuals or small negative residuals determines whether the model tends to
underpredict ground concentrations regardless of the magnitude of the concen-
tration (positive) or whether it tends to overpredict these concentrations
(negative). The summary behavior in this table of all of the models except
4-41
-------
RTM-II for the first two types of graphical comparisons indicates
overproduction of plume ground concentrations for values in the range of 0-50
parts/1015. However, the positive values for the peak of the frequency
histogram of residuals for these models indicates overall underprediction of
values when larger observed concentrations are included. RTM-II clearly
exhibits the opposite behavior. It predicts too few small concentrations, but
too many large ones. The first column for MESOPUFF indicates that this model
tends to underpredict plume ground concentrations for the smallest range of
observed values, but the second column entry for MESOPUFF indicates over-
prediction of concentrations in the range of 10-50.
The predictions of MESOPLUME are presented first in Figures 4-15 to 4-17
to show the typical prediction of the models. The other graphs present the
sole deviations from the trends represented by the MESOPLUME predictions.
(Appendix E contains the full set of graphs for each model.) Note that the
RTM-II graphs show opposite trends to the "standard" MESOPLUME graphs.
4.3.2.2. Savannah River Plant Data Base
Table 4-15 has been developed for the SRP graphs to parallel Table 4-14
for the Oklahoma graphs. The "standard behavior" graphs for the SRP data
cases are Figures 4-23 to 4-25, and graphs of deviations from the standard are
presented in Figures 4-26 to 4-30.
Three of the models (MESOPLUME, MSPUFF and RADM) exhibit the same
behavior in all three types of graphical comparisons as they did for the
Oklahoma data. The MTDDIS model underpredicts plume ground concentrations as
shown by the first two types of graphs, but over-predicts large observed
values. The mixed results summarized in Table 4-15 for MESOPUFF II and RTM-II
show that these two models have predicted and observed ground concentration
distributions that are more nearly in agreement. RTM-II predicts the same
number of concentrations as observed in the 0-10 range, but the cumulative
number of predictions above 10 parts/101^ exceeds the observed number.
However, for RTM-II, the peak of the residuals is above zero indicating some
underprediction of large concentrations. MESOPUFF II predicts fewer small
concentrations than were observed, but its cumulative frequency distribution
also exceeds the observed. Its tendency toward some underprediction of large
concentrations is shown by the occurrence of a positive value for the peak of
4-42
-------
FREQUENCY DISTRIBUTION OF PREDICTED AND OBSERVED
CONCENTRATIONS - RANGE: 0 - 100 PCI/M'*3
90-
80-
70-
00-
I80'
g 40-
30-
eo-
10-
0-
(
MESOPLUME MODEL - SAVANNAH RIVER PLANT
^§
H
i
•
•
OOOOQ
II
ffl
/%^
E2 OBSERVED
S3 MESOPLUME
> 10 20 30 40 50 80 70 80 90 100
CONCENTRATION (PCI/M"3)
Figure 4-23. Frequency distribution of predicted and observed concentrations
at Savannah River Plant for MESOPLUME based on points paired in
space and time ... concentration range: 0 to 100 pCi/m3.
I
I
e
100
90
•0
70
•0
90
40
30
20
10-
CUMULATIVE FREQUENCY DISTRIBUTION
OF PREDICTED AND OBSERVED CONCENTRATIONS
MESOPLUME MODEL - SAVANNAH RIVER PLANT
OBSERVED
MESOPLUME
10 100 , IOOO
CONCENTRATION (PCI/M"3)
10000
Figure 4-24. Cumulative frequency distributions of MESOPLUME predictions and
observed concentrations at Savannah River Plant based on points
paired in space and time.
4-43
-------
FREQUENCY HISTOGRAM OF RESIDUALS
MESOPLUME MODEL - SAVANNAH RIVER PLANT
FREQUENCY
? ? 8 6 S
-1OO-80 -8O -70 -«0 -50 -40 -30 -20 -
V'
'%.
I
^
VTA
'jm
10 0 10 ZO 3
%
0 4
^
0 SO 60 70 80 M 100
RESIDUAL (OBSERVED - PREDICTED)
Figure 4-25. Frequency distribution of residuals at Savannah River Plant for
MESOPLUME based on points paired in space and time ...
residual range: -100 to 100 pCi/m3.
FREQUENCY DISTRIBUTION OF PREDICTED AND OBSERVED
CONCENTRATIONS - RANGE: 0 - 100 PCI/M"3
BO-
eo-
70-
00-
i-
33 40-
30-
tO-
10 -
0
(
MTDDIS MODEL - SAVANNAH RIVER PLANT
W,
8$
1
•
1
XS999
y$$$&
eZ) OBSERVED
K3 MTDDIS
$$^$
> 10 ZO 30 40 90 80 70 80 9O 100
CONCENTRATION (PCI/M"3)
Figure 4-26. Frequency distribution of predicted and observed concentrations
at Savannah River Plant for MTDDIS based on points paired in
space and time ... concentration range: 0 to 100
4-44
-------
CUMULATIVE FREQUENCY DISTRIBUTION
OF PREDICTED AND OBSERVED CONCENTRATIONS
MTDDIS MODEL - SAVANNAH RIVER PLANT
10 100 , 100O
CONCENTRATION (PCI/M"3)
10000
Figure 4-27. Cumulative frequency distribution of MTDDIS predictions and
observed concentrations at Savannah River Plant based on points
paired in space and time.
FREQUENCY
so-
40-
FREQUENCY
588
HISTOGRAM OF RESIDUALS
MTDDIS MODEL - SAVANNAH RIVER PLANT
I
\
L
loo-eo -io -TO -ao -so -io -30 -ko -io 6 ib 20 so *o so a'o TO e'o eo too
RESIDUAL (OBSERVED - PREDICTED)
Figure 4-28. Frequency distribution of residuals at Savannah River Plant for
MTDDIS based on points paired in space and time
residual range: -100 to 100
4-45
-------
FREQUENCY DISTRIBUTION OF PREDICTED AND OBSERVED
CONCENTRATIONS - RANGE: 0 - 100 PCI/M"*3
90-
80-
FREQUENCY
£883
30-
M-
10
0
I
MESOPUFF II MODEL - SAVANNAH RIVER PLANT
n
1
•
I
^^
EZJ OBSERVED
S3 MESOPUFF II
> 10 ZO 30 40 SO 60 70 SO BO 10O
CONCENTRATION (PCl/M"3)
Figure 4-29. Frequency distribution of predicted and observed concentrations
at Savannah River Plant for MESOPUFF II based on points paired
in space and time ... concentration range: 0 to 100 pCi/m3.
FREQUENCY DISTRIBUTION OF PREDICTED AND OBSERVED
CONCENTRATIONS - RANGE: 0-100 PCI/M'*3
M-
•0-
70-
60-
55 SO-
9 4O-
3O-
ZO
10
0
RTM-n MODEL - SAVANNAH RIVER PLANT
EZ OBSERVED
S3 RTM-H
1
•
1
•
1
9$$$$ ffiftfflty M/S&
I 10 ZO 30 40 50 80 70 80 90 iOO
CONCENTRATION (PCI/M"3)
Figure 4-30. Frequency distribution of predicted and observed concentrations
at Savanna River Plant for RTM-II based on points gaired in
space and time ... concentration range: 0 to 100
4-46
-------
Table 4-15. Summary of Results of Graphical Comparisons of Models with
Savannah River Plant Data.
Model Frequency Distribution Cumulative
of P and 0
(initial trend,
MESOPUFF
MESOPLUME
MSPUFF
MESOPUFF II
MTDDIS
RADM
RTM-II
0-10
P
P
P
0
0
P
P
range)
> 0
> 0
> 0
> P
> P
> 0
= 0
Frequency Plot
(initial
P >
P >
P >
P >
0 >
P >
P >
trend)
0
0
0
0
P
0
0
Frequency
Histogram
of Residuals
(position of
0
0
0
0
-10
0
0
peak range)
- 10
- 10
- 10
- 10
- 0
- 10
- 10
the residuals. MESOPUFF predicted fewer concentrations in the 0-10 parts/1015
range than were.observed for the Oklahoma data; however, the opposite trend
was found when comparisons were made with the SRP data.
Again, the MESOPLUME graphs that are presented in Figures 4-23 to 4-25
represent the common behavior of the models. The remaining graphs present
deviations from that common behavior. As may be seen, the MTDDIS model
provides the opposite behavior to the "standard" MESOPLUME prediction
characteristics. MTDDIS reveals a much greater plume spreading than shown by
the other six models and also has a tendency to overpredict with respect to
the ground-level concentration data as well. No other noteworthy deviation
can be systematically identified from Table 4-15.
4-47
-------
CTHIS PAGE INTENTIONALLY LEFT BLANK)
4-48
-------
SECTION 5
DIAGNOSTIC EVALUATION BASED ON MODEL/DATA COMPARISONS
5.1. INTRODUCTION
In this section, additional graphical and tabular comparisons of model
predictions with observed data will be presented. The specific features of
the predictive behavior of the models will be traced to model assumptions and
experimental constants. Although the AMS statistics provide an objective
method for evaluating the predictive accuracy of the models, they do not
provide information that can be used to identify the causes of model/data
discrepancies. For example, the predicted plume from a model might be
approximately the right shape and might contain realistic concentration
values, but its ground concentration pattern might simply be shifted from the
observed plume. Such a directional inaccuracy of the predicted plume could
arise from a combination of errors in the predicted wind field from the
meteorological preprocessor and from the coarse spatial resolution of the
input meteorological data. In such a case the AMS statistics will only show
model/data disagreement, but cannot indicate that the pattern is correct,
yet in an incorrect location. A further analysis of the model predictions has
been carried out to help provide more insight into the causes of the
model/data discrepancies.
Section 5.2 presents a subset of the plume concentration isopleth plots
that have been prepared for each of the 21 sampling periods in the Oklahoma
data base. Section 5.3 also presents a representative sample of plume concen-
tration graphs prepared for the 65 sampling periods in the Savannah River
Plant data base. These graphs provide a clear comparison of the location and
width of the predicted plumes relative to the data. They allow several
general conclusions to be made concerning the relationship between model
predictive performance and modeling assumptions.
In Section 5.4, a quantitative pattern comparison method is applied to
the Oklahoma data. This method was not applied to the Savannah River Plant
data cases since the samplers are too widely spaced to allow the same degree
5-1
-------
of quantitative analysis. Tables have been prepared comparing predicted and
observed values of (a) plume width, (b) azimuth angle of transport through
sampler arcs and (c) time of arrival at the arcs. These comparisons provide
new insight into the AMS statistical results for each model. In that section,
the accuracy of several common modeling assumptions is discussed on the basis
of the pattern comparison results.
The final section, Section 5.5, presents a simplified yet quantitative
pattern comparison method for the Savannah River Plant data, consistent with
the wide spacing of samplers in that experiment. Tables of maximum predicted
concentrations for each averaging period for both data bases are then
presented. These tables allow the models to be intercompared on the basis of
whether they tend to predict ground concentration patterns that are always
higher or lower than those of the other models. Finally, the theoretical
assumptions for each of the eight models are examined in light of the
predictive performance of the model. Section 5.5 ends with a discussion
synthesizing all the diagnostic comparison methods described in this Section
5. For each model, the specific assumptions responsible for the model's
predictive performance relative to observations are discussed, and causes of
inaccuracies are identified where possible.
5.2. COMPARISON OF PREDICTED CONCENTRATION ISOPLETHS WITH DATA AT OKLAHOMA
Figures 5-1 through 5-12 illustrate the type of isopleth plot prepared
for each model for each of the 21 sampling periods in the Oklahoma data base.
The remainder of these graphs appear as Figures E-12 through E-95. The model
name, date and time-averaging period are printed at the top of the graph. The
20 km grid steps are indicated by tick marks on the axes, and the source is
located at the origin, (0,0). The contour interval used for the isopleth map
is printed in the upper left corner in parts/101^, and isopleth values are
presented in those units in the figure. The maximum predicted concentration
found on the prediction grid of size 34 x 34 (20 km by 20 km grid spacing) is
listed beneath the designation of contour interval. (Since a constant contour
interval is used, and no more than 10 contours were permitted on a graph, the
peak concentration is often much larger than that of the innermost contour
shown.) The black dots refer to samplers that were turned on during the
5-2
-------
2
LJ
-z.
0
CK CM
H- o
Z H
LLJ ^
z" 8
o 8
u
I
u
Q.
Ld
2
.0.
o
00
LJ
o
03
00
0
« a
OOSi 000$ OOSZ 0
•DNOD 'S8O
u
4".
> 2
Ct UJ
UJU
.O
z x
o<
02
08£
a
LJ
o
cc
I
o
o
HS
o
-oo
0
•(N
UJ
o
a
o
£
z u
O ,p
b rn
I?
LJ O
(J n
Z ^
O "
U
-r °
i\ °°
CJ
2 >.
_
U_
D
Q.
O
GO
o
«-y
a.
2
OOSi OOOS 00«Z o
•DNOD 'SBO
.
O fN
a u
UJ (J
.O
-^
8^
a:
UJ
o
o:
I
o
o
o
-------
I/}
UJ
5
N^X
CD
UJ
S ^
^ s
00 U
I *
b? ™
I P
z d
UJ n
^ a
o
U O
-I- m
5 ?
^ >:
a- D
1
^
^
•<^.a
OOSi OOOC 00n
•<* f"*
^ n
II *
i*
a i |
l-fl
UJ (J
£g
_ o
.u
\- .
zx
0<
02
08Z OZZ
oo
o:
a UJ
"NX _l
z °-
• * "5
«a -5
UJ <
.0 _i tO
"a
2 u
•S< cc
(0 <
01 I
5
^
1
o
o
"
II
V
0 n
, m O
x' » 0
/' \ 0 UJ
/x'^^V ^ a
XN\ J "0
091 001 0* 02-
o
-*
r<
o
•act
y*N
0 "5
(N -^
52 *
\.x
(_
to
UJ
o
"<0
-o
o
vO
• 1
HldON
(NQJ
UJ
kl
-------
CO
h- 6
Z m
UJ (N
u ™
o
u
X
u
Q
<
o:
UJ
'<
cc
O
u
r
u
o:
o:
O
00
03
O
n
(N
o
(N
o
03
0>
03
*2;
oosi ooos oosz o
•DNOD 'SBC
O 03
O / \ 0 „' \ a
• ^ ,'^i^m^ ^* Q
o
(N
O
-oo
o
•M
o
x>
fH
25 ^j
U Q)
in o.
rH
CO S
CM Q
0)
rH
4
I
4->
O
a
0)
I
in
0)
OH
O 4->
ro J3
OJ CJ»
-! *4J
0) <4-l
& 5
0) %_
OZZ
091 001
HldON
02-
5-5
-------
rd
I
x*\
LO
L»J
Vy^
z
o
o:
h-
z
UJ
U
z
o
u
r
u
^c
Q.
™"~
1
^>
j
o:
H
o
tf}
£v
•^^ — ^^^X!^wiiVx
x *"• ^ '^—^^ s* '
\ ^^008^2^^
1
b
••* M-l
fN
w
G
o
• rH
4J
ID
-I S
1
§
^
o Q
-§i §5
<^ iH
1
1- -g
^ §
o OT
"«o
4-1
o
^
rH
-O
5
t5
rH
Q
O [fi
<« h-l
•
•
•
**-»
E™*
jg
e>
in
rH
CO
CJ
0
CO
rg
CM
o
CO
a
g
E
•r-t
>-l
0)
^
CO
w
o
• rH
*J
U
• rH
•s
^
0,
1—4
1— t
I
32
EH
«
082
OK
091 OOt
HldON
02-
•*
^
0)
5-6
-------
LQ
LJ
k 9
Q, .-
LJ
U
z
O
u
I
u
LJ
a.
o
CO
LJ
O
o
00
o
o
GO
a\
O CD
OOSl 0001 DOS 0
•DNOD -sao
OSS
09*
OfrC
oot
.
u .
XZ
<0
Su
LJ
U
o:
D
O
CO
00
<
LU
HldON
iQ
LJ
CO
D
Q_
O
CO
LJ
l-
Q d
H O
OH ^
LJ
U
z
o
u
T °
i11 m
u o»
o
o
00
o
D
0 CD
OOSl 0001 OOS 0
•DNOD -sao
OSS
09*
OtC
001
2 u
OS
X z
-------
CO 5
2^t
CO
o .
iZ O
< °
o: ^
£ e
LU •
¥ §
^ 00
8 °
X o
U OD
s °*
Q- r~
D
Q.
O
to
U
CO O
I §
p *
•z. o
UJ O
(j 00
z °
o
U O
T ^
-^- ul
u -
2 5
Q.
CO
0V
o
O 00
C OZ2
HldON
001
.o
uio
-
iS
UJQ:
5
u .
X2
IS
u
cc
D
o
00
-o
to
<
UJ
o
- O »i
0) x_x ^^
I O »
•a oo w
c cr> c
3 fH O
o ••-«
U » 4->
en o
S -8
^ a
O «M
r—I O
O4
4J
A §
« -S
•H W»
O, Q)
O Oi
n 0>
in
0)
4J
IM
0)
5-8
-------
UJ
y~
Q.
^•^
RATIONS
1-
z
UJ
U
O
U
X
u
Q.
2
Q
o:
^
in
LJJ
s
to
Z
0
(_
<
a:
\-
z
LJ
U
0
L)
X
u
^c
Q_
ARRPA
I-
5
O
d
0
d
o
CD
0
0
(D
OJ
V
D
-3
•
O
• O 03
• .
* "'o
* 00 *~
Ps °^ "
•' o O 1 1 _J
• . 55 z>
UJ UJtt
• -Sgi u^
* OT .(- '
• oo'ei oo'ot oos o
.* 'DNOD'Sao 2u
.
- (N
« * *~ li 1
• u
. o:
• * D
• /* V^^ O
• ° ' ^Ls°j) ^
CL.X U
-•
o
0
• o ^
(/I
UJ
0
00
o
/—.
D
J
^^
O
• 0 OD
* .
* * * • r>
* ^O
•* o *~
L » n II
PI *
'n Z rf
• -«ce Z>
^ UJ UJtt
• "•• Q_ U UJ
-=^mi3^< §2
• *" .1-'
• OOSl 0001 005 0 < Q
/ -3NOD -sao 20
*
-%
-^^*" ~s
.•^^^^v^ s - -» i * i
* ^^P^^\N ^
;:;':"^^ f
-•
* ...
o
••*
in
o
"Q!
~
-o ^
1-
UJ
0
-oo
.0
o
vO
I
I
Q*
r*
in
0)
fc
oes
09*
O*E on
HlfclON
001
02-
5-9
-------
•i
rH
*
0)
* 2
\_> ^
Z o
O o
a H
z- d
5 S
U .0
o
X
u
0
00
O
= o>
I o
2
h-
a:
o oo
<*>
Si
OOSI 0001 OOS 0
'DNOD 'S80
in
OD ^
00 II
oss
09*
o*e 022
HldON
ooi
> o
o
CO
cr>
u
c
o
rH O •<-«
a « H
5 I "
0) •<-! hH
fH 5-i 1-1
CX "
s
CO
in
0)
i
£
5-10
-------
LO
u
(A
2
O
I-
LJ
U
2
O
U
(J
Q_
UJ
^
D
0.
o
o
o
o
o
CO
o
o
00
b
-3
(N
UJ
Q d
U— O
< «•
o: o
I- o
L) ^
O
U
u
o
CD
O>
D 2
CL
O
00
UJ
. _
"a
oi s o
'DNOD'SaO
00
a:
u
o:
I
o
o
o
2 X
0<
08Z
OZZ
091 001
HldON
01 « 0
•DNOD '200
o .
m CD
DC. UJ
UJ O
.u
5x
81
082
OZZ
091 001
HldON
u
a.
UJ
Oi-
LJ
U
a:
D
O
I/)
(/I
<
UJ
02-
I
I
§
•1-1
-p
u
O 33 §
•^ u iJ
jj (X,
(d o S
cox
° 4J
in "&
O
-------
o:
I-
LJ
U
U
2
Q.
O
O
o
o
S
°
_
CL
O
CO
UJ
{/I
UJ
«(L
-•*
o
-co
01 C 0
•DNOO 'S80
I
5
I
o
o
.
m co
tt UJ
Ul (J
*Z
.u
2 X
o<
082
022
091
OOi
UJ
D
02-
o
i §
U m
Z o
O
U o
T.
(J
CL.
D
CL
to
oo
(N
01 S 0
•DNOO 'sao
m
S
u S
CK UJ
UJ (J
.u
082
o
O
03
UJ
u
CL
D
O
UJ
-o
022
091 001
HldON
02-
o
-------
z
o
Qi
UJ
U
Z
o
u
I
(J
Q
a
LJ
a:
I-
z
UJ
u
z
o
o
I
u
a:
a:
o
o
o
o
0
CO
O
i-
o
C)
o
o
o
n
o
o
CO
O
D
-5
fN
a:
UJ
Q.
i « a
01 s p
'DNOD'Sao
10 O*
UJU
.o
z x
o<
08Z
Oil
091 001
HldON
to
UJ
o
ot
01 s o
•DNOD'SQO
CO
a
UJ
00
u
a:
I
o
o
> z
tt III
UJ U
z z
~. u
z x
0<
(J 5
UJ
u
«
o
"
o ^
ci
to
UJ
-o
08Z
OZZ
091 001
HldON
02-
rH
^
O
5-13
-------
I
0)
LO
UJ
Z
O
H
t-
O
O
o
o
o
^- m
UJ •-
u /-.
u -
2 §
-a
T O
i-
o:
C 0
•DNOD -sao
UJ
a.
u
CL
I
o
o
O
II
_l
a
LJ
I-
z
o
u
UJ
u
o
o
08Z
OZZ
091 001
HldON
o
a:
-o
w
nl
V-i
0)
§5
o
O
n
u in
CO
O
00
0 -3 §
*J ^ 3
O «w O
i-l O •<-<
.c c
0) -H M
7
in
0)
-------
measurement period cited in the figure title; the unfilled dots refer to
sampling stations that were not in operation during that period. Only a few
samplers are labeled to avoid obscuring important contour features. The
identification number of intermediate samplers is determined by counting from
the labeled ones. The dotted line in the isopleth plots refers to the
smallest curve that encloses non-zero predictions made by the model. (It
passes through the points with zero predicted concentration nearest the region
with non-zero values.) Finally, in the upper right corner is drawn a graph of
the observed concentration values at each active sampler across the 100 or
600 km arc, joined by straight lines. The x-axis is the sampler
identification number. (Equal spacing on this axis does not indicate equal
distances between samplers.) The y-axis units are parts/1015.
The interpolation scheme used to prepare the isopleth plots assumes, for
any point inside a grid cell, a linear variation of concentration based on
predictions made at the four nearest grid nodes. This method produces concen-
tration values within a cell that are always smaller than the largest value at
the four neighboring grid points. It should be recognized, however, that
model predictions with a finer grid spacing would likely reveal a larger value
within a cell.
Figures 5-1 to 5-4 present isopleth plots of the seven model predictions
for the time period representing the peak observed concentration at the 100 km
arc for the July 8 experiment. This time period represents the 45-minute
averaging period 2230 to 2315 GMT; the perfluorocarbon release took place from
1900 to 2200 GMT. The actual transport time from the source to the 100 km arc
was about five hours. Figures 5-5 to 5-8 present similar isopleths for the
600 km arc during the period expected to be the time of peak observed concen-
tration, 0800 to 1100 GMT on July 9, 1980. (As the data indicated, the
concentration peak arrived earlier before the samplers were turned on. This
may have been due to rapid transport by an elevated nocturnal jet.) On
July 11, 1980 a second three-hour release was made from 1900 to 2200 GMT.
Figures 5-9 to 5-12 present predictions for the concentration averaging period
0315 to 0400 GMT on July 12, 1980. Peak observed concentrations on the 100 km
arc did not occur during this time period. That period represents the last
one for which measurements at the 100 km arc were taken. Isopleth plots for
that time period were chosen for presentation here because it best illustrates
the differences in the concentration patterns predicted by the models.
5-15
-------
Two special aspects of the above graphs should be explained. In Figures
like 5-3 and 5-4, some of the isopleths pass upwind of the source (with the
wind from the southwest). This is an non-physical feature of the isopleth
program resulting from linear interpolation between the source grid point and
the zero concentration values at the nearest upwind grid points. Such nonzero
upwind values should be ignored as an artificial result of the interpolation
method used. Also, in Figure 5-9, the 350 parts/1015 contour crosses the 400
parts/1015 contour near the peak of the concentration pattern. This is also
obviously not a physical effect, but results from the isopleth program logic
that develops each isopleth without checking nearby ones to eliminate
crossings or to make other adjustments. Although more sophisticated
isopleth-generating programs are available, the method used provides adequate
interpretive information for this study.
An examination of the above figures along with those in Appendix E
reveals the following conclusions:
(a) All sampler values representing maximum predicted values are offset
in time from the observed maximum values. For the July 8
experiment, a 1-3 hour time offset is typical at the 100 km arc and
a 2-8 hour time offset is representative of transport to the 600 km
arc. Some models predict plumes that transport slower and others
faster than the observed data at the 100 km arc for the July 8
experiment. All models predict a slower transport to the 600 km arc
as compared with the observed plume. The time lag/lead problem
also exists for the 100 km arc in the July 11 experiment. The
MESOPUFF and MESOPLUME model predictions lead the observed plume.
Both models employ MESOPAC as the meteorological preprocessor. For
the other models, the time lag is about one hour to the 100 km arc.
Isopleths for the peak observation period for the second Oklahoma
experiment (not presented here) may be found in Appendix E.
(b) All model predictions are offset from the data in space.
Differences are largely dependent on the choice of meteorological
preprocessor. Models with the same preprocessor (MESOPUFF and
MESOPLUME using MESOPAC; and MESOPUFF II, RADM, and RTM-II using
MESOPAC II) have the same general orientation of their isopleth
5-16
-------
pattern both in space and time. The horizontal and vertical
dispersion algorithms used by each plume model largely create the
differences in patterns within the same wind field.
(c) Comparisons of the predicted peak concentrations and horizontal
spreading among the models reveal large differences. For these two
Oklahoma cases, the RTM-II model tends to predict the greatest
horizontal spreading with the RADM model predicting the least
spreading. In terms of peak concentrations, the RADM model appears
to predict the largest ground concentrations and the RTM-II model
predicts the lowest peak values. Conservation of mass principles
lead us to expect that models that predict the highest ground levels
would have the least spreading. Clearly, a third dimension (the
vertical) must play a role, yet, the vertical dimension may have a
lesser effect since full mixing up to the mixed layer probably is
occurring in each model at these 100 km and 600 km distances.
Some of this systematic behavior can be supported by means of a table of
predicted and observed concentrations. Table 5-1 presents a listing of the
predictions of the models and the observed ground-level data for both Oklahoma
experiments as functions of ranges of values in parts/1015. All time periods,
both arcs, and both experiments are considered together here. All
predicted/observed pairs (based on an original pairing in space and time) were
employed in the preparation of this table. (All pairs with both members equal
to zero were eliminated, however).
The large spatial and temporal offset of the patterns leads to large
numbers of predicted or observed values that are zero in this table. If
values greater than 300 parts/1015 are considered, the RADM model has many
more of these large values than the other models, supporting the conclusion
that RADM tends to overpredict the high concentrations. In fact, examination
of Table 5-1 reveals that 42 observed values are greater than 200 parts/1015.
However, all models predict more than 51 values greater than 200 parts/1015,
with RADM predicting 68. Note also that RTM-II has the fewest number of
predicted zeroes of all the models (52). This is likely due to its over-
prediction of horizontal plume spreading; a wide plume is more likely to cover
many grid points leading to a fewer number of predicted zeroes at receptor
5-17
-------
Table 5-1. Frequency Distribution of Predicted and Observed Concentrations
for the Two Oklahoma Cases (Based on Predicted/Observed Pairings
in Space and Time) in parts per 1015.
MESOPUFF
=
( 0 -
( 100 -
( 200 -
( 300 -
( 400 -
( 500 -
( 600 -
( 700 -
( 800 -
( 900 -
( 1000 -
( 1500 -
( 2000 -
( 3000 -
( 4000 -
( 5000 -
( 6000 -
( 7000 -
( 8000 -
( 9000 -
(10000 -
(15000 -
(20000 -
(25000 -
>
0
100)
200)
300)
400)
500)
600)
700)
700)
900)
1000)
1500)
2000)
3000)
4000)
5000)
6000)
7000)
8000)
9000)
10000)
15000)
20000)
25000)
30000)
30000
Totals:
DBS
47
122
13
11
4
2
2
2
1
2
3
5
1
6
1
1
1
0
0
0
0
0
0
0
0
0
224
PRED
94
56
18
10
10
10
0
1
1
0
0
5
4
7
3
5
0
0
0
0
0
0
0
0
0
0
224
MESOPLUME MSPUFF MESOPUFF
OBS
39
122
13
11
4
2
2
2
1
2
3
5
1
6
1
1
1
0
0
0
0
0
0
0
0
0
216
PRED
114
33
17
6
5
7
5
3
2
3
1
3
1
5
2
5
2
1
0
0
0
1
0
0
0
0
216
OBS
53
122
13
11
4
2
2
2
1
2
3
5
1
6
1
1
•1
0
0
0
0
0
0
0
0
0
230
PRED
109
53
14
6
4
4
5
0
4
2
2
3
6
6
1
1
1
1
0
0
0
6
1
1
0
0
230
OBS
26
122
13
11
4
2
2
2
1
2
3
5
1
6
1
1
1
0
0
0
0
0
0
0
0
0
203
II ARRPA
PRED OBS
88
46
14
4
7
3
6
3
5
1
4
3
7
6
4
2
0
0
0
0
0
0
0
0
0
0
203
31
122
13
11
4
2
2
2
1
2
3
5
1
6
1
1
1
0
0
0
0
0
0
0
0
0
208
PRED
80
60
9
9
8
7
3
1
3
1
4
7
1
4
5
3
0
0
1
1
0
1
0
0
0
0
208
RTM-II
OBS
51
122
13
11
4
2
2
2
1
2
3
5
1
6
1
1
1
0
0
0
0
0
0
0
0
0
228
PRED
52
89
25
13
7
2
4
8
6
2
0
16
4
0
0
0
0
0
0
0
0
0
0
0
0
0
228
1
RADM 1
OBS
15
122
13
11
4
2
2
2
1
2
3
5
1
6
1
1
1
0
0
0
0
0
0
0
0
0
192
PRED
97
23
4
6
5
2
1
0
5
3
1
9
7
7
3
7
5
0
2
1
1
3
0
0
0
0
192
5-18
-------
locations. Note also that MSPUFF and HESOPLUME have the largest number of
observed zeroes. An examination of the patterns reveals that MSPUFF generally
underpredicts horizontal spreading and MESOPLUME tends to have a combination
of less spreading and less directional accuracy than the other models at the
100 and 600 km arcs.
5.3. COMPARISON OF PREDICTED CONCENTRATION ISOPLETHS WITH DATA AT THE
SAVANNAH RIVER PLANT
The small number and large separation distances between sampling
locations in the Savannah River Plant data base give poor spatial concen-
tration gradient resolution, making such contour plots less useful for
comparing model predictions with observed data. The added information that
might be obtained from such isopleth plots did not justify the effort needed
to prepare the large number of plots for the Savannah River Plant data base.
Instead, computer printouts of average concentrations at each grid point were
prepared for each model for each sampling period, with observed values printed
below the grid points nearest them. In Section 5-4, the 65 predicted patterns
for each model are broken down into 8 categories giving different
relationships between the predicted and observed patterns. In this section,
detailed plume outline graphs were drawn in the spirit of the isopleth plots
for two 10-hour time periods represented in Figures 5-13 through 5-20. The
graph and axis labels provide the same information as do the isopleth plots
for Oklahoma. Each figure presents measured 10-hr averaged concentrations
(above background) at each sampler (in pCi/m^) along with values of the
predicted plume concentrations within the outlined plume. A number within a
circle represents the identification number (2 through 13) of a fixed sampler.
Case 4B (Figures 5-13 to 5-16) illustrates one of the two common problems
revealed with the model predictions for the SRP cases. All of the models
appear to have a rotation angle error of the predicted plume as compared with
the data. The models that use MESOPAC have a definite clockwise rotation to
their predicted plumes. The models that use MESOPAC II exhibit the same
rotation error, but by a smaller rotation angle.
Figures 5-17 through 5-20 illustrate the second problem found upon
examining the patterns from the SRP case runs. This problem is an
5-19
-------
HESOPUFF — CASE IB MY 18 (2200 H«) TO HOY 19 (MOO ml
250 KM-
200 KM-
100 KM-
0 -
-21
V520 1 2 3 11 26 ,
119 108 30 2 1
239 331 76 18
29 262 131 61 21
1
9 3
230
138 93
15
1 2
12
68 17
91 210 218 ,-431 89 70 60
18 117 21
-1
191 131 91
16 111 152 151 127
1
100 KM
200 KM
HESOPLHHE — USE <» HOY 18 (2200 H») TO Hoy 19 (MOO ml
100 KM-
0 -
100 KM
200 KM
Figure 5-13. Comparison of 10-hour averages of predicted plume and observed
data (in pCi/m3) for Savannah River Plant experiment of
November 18-19, 1976 (2200 to 0800 GMT) ...
(top) MESOPUFF predictions, (bottom) MESOPLUME predictions.
5-20
-------
HSPUFF — CASE IB HOY 18 (2200 H«l TO HOY 19 (0800 ml
0 -\
100 KM
200 KM
HESOPBFF II — CASE M HOY 1» (2200 ml TO MV 19 (MM ml
230 m-l
200 KM J
100 KM-I
0 H
100 KM
200 KM
Figure 5-14. Comparison of 10-hour averages of predicted plume and observed
data (in pCi/m^) for Savannah River Plant experiment of
November 18-19, 1976 (2200 to 0800 GMT) ...
(top) MSPUFF predictions, (bottom) MESOPUFF II predictions.
5-21
-------
HTDDIS - CASt 18 »OV 18 (2200 H») TO IIOV 19 (MM Hit)
230 KM-
200 KM-
100 KH-
2 3"
9 10 9 8 6
23 20 16
83 69 52
*-2« 36 24
fl 40 118 144 148 124 94 66 44 28
99 269 248 203 156 112 74 47 28 17
k 380 367 287 215 Q.^ 74 44 25 J
272 238 171 61 36 19 (£)-
68 66 54 37 22 12 6
15 20 20 16 10 6
1
100
—I
200 KM
tABH - CASE « HOT 1» (?MO H«) TO 10V 19 (0»QO Ml
100 KM
200 KM
Figure 5-15.
Comparison of 10-hour averages of predicted plume and observed
data (in pCi/m^) for Savannah River Plant experiment of
November 18-19, 1976 (2200 to 0800 GMT) ...
(top) MTDDIS predictions, (bottom) RADM predictions.
5-22
-------
mn-11 -- OSE n urn i« tnaa »«i to HOY is (MM ml
100 KM-
0 -
100 KM
200 KM
Figure 5-16. Comparison of 10-hour averages of predicted plume and observed
data (in pCi/m^) for Savannah River Plant experiment of
November 18-19, 1976 (2200 to 0800 GMT) ...
RTM-II predictions.
5-23
-------
HE50PUFF - CASE 6C fit 17 (2200 mi) TO FEB 18 (0800 H.)
230 KM-
200 KM-
100 KH-
0 -
12V1
CO-0
13>0
69 29 18 14 12 " W •
10 52 25 21 18 15 14 12 X 10
4 17 29 32 26 20 17 15
1 7
1
100 KM
1
200 KM
IffSOPlUHE -- CASE 6C FEB 17 (2200 m) TO FEi II (MM H«)
230 KM-
200 Kn-
100 KH-
0 -
C2M
96 618 274 116 63 22
42 38 29 19 19
1
100 KM
1
200 KM
Figure 5-17. Comparison of 10-hour averages of predicted plume and observed
data (in pCi/m^) for Savannah River Plant experiment of
February 17, 1977 (2200 to 0800 GMT) ...
(top) MESOPUFF predictions, (bottom) MESOPLUME predictions.
5-24
-------
HSPUFf -- CASE 6C FEB 17 (7700 H.) TO fit H (0800 m)
230 KM-
200 KH-
100 KM -
0 -
C9>16
J219 331 66 21 5
6 2HI|
—I
100 KM
1
200 KM
HE50PUFF II -- CASE 6C FEB 17 (7MO H.) TO FEi IH IMM
230 wi-
200 KH-
100 KM-
C2)-l
Ci>o
i3>0
100 KM
1
200 KM
Figure 5-18. Comparison of 10-hour averages of predicted plume and observed
data (in pCi/m^) for Savannah River Plant experiment of
February 17, 1977 (2200 to 0800 GMT) ...
(top) MSPUFF predictions, (bottom) MESOPUFF II predictions.
5-25
-------
HtDIIIS - CASE 6C fit 17 (2200 Mil) To FED 18 (0800 Mil)
230 KM-
200 KM-
100 KM-
'2 14 61 106 135
'3 11 37 88 131
fl 3 12 33 70
68 40 20 116 4
116 72 39 21 (7) 8 5
128 92 57 35
116 98 72 51
244
107
14 11 29 57 86 101 98 83 65 18
ll 3 10 23 45 70 88 94 88 75 61 35
7 15 28 46 64 78 84 83 77 67
5 10 19 31 46 60 ®~M 79 81 78
4 8 15 26 38 52 64 73 78 80
—r
100 KM
—I
200 KM
yam - «SE V ff* 1?
M«) TO FEI II <0«W «)
230 KM
200 KM
100 KM
i3>0
-24
100 KM
200 KM
Figure 5-19. Comparison of 10-hour averages of predicted plume and observed
data (in pCi/m^) for Savannah River Plant experiment of
February 17, 1977 (2200 to 0800 GMT) ...
(top) MTDDIS predictions, (bottom) RADM predictions.
5-26
-------
ma-11 — CASE 6C fit 17 (2200 H«) TO FEI 18 (0800 H«)
230 w-
200 KM-
100 wi-
12 VI
Cl>0
JQ)-1
20
2 1
208 31 19 11 9 97 6
58 26 17 12 ^v
211
8 21 21 11
10 13
2 9 13
7 1211
1 91 12
100 KM
—I
200 KM
Figure 5-20. Comparison of 10-hour averages of predicted plume and observed
data (in pCi/m^) for Savannah River Plant experiment of
February 17, 1977 (2200 to 0800 GMT) ...
RTM-II predictions.
5-27
-------
underprediction of the horizontal spreading of the plume by the models.
Recall here that the RTM-II model which overpredicted spreading in the
Oklahoma cases now underpredicts spreading. Two factors are relevant here for
RTM-II. First, at SRP, the MESOPAC meteorological preprocessor is used rather
than MESOPAC II. Second, and more importantly, the dispersion constant TJ and
the minimum allowed value of the horizontal diffusivity were chosen
differently in each case. RTM-II model predictions are very sensitive to the
values used for these parameters. More discussion of these parameters and the
RTM-II model is presented in Section 5.5.
Table 5-2 presents in similar format to Table 5-1 a breakdown of
predicted and observed concentrations into a number of ranges. Whereas all
models overpredicted peak values for Oklahoma, there is only a slight
overprediction (in terms of numbers of large concentration values) for the SRP
cases. Note that the MTDDIS and RADM models have a definite tendency to
overpredict peak values and that RADM has one very large prediction greater
than 30,000 pCi/m3. Examination of the distribution of 10-hr concentration
predictions greater than 200 pCi/m3 for MTDDIS reveals a large number of
predictions between 200-700 pCi/ra3, but few above 700 pCi/m3. This may be
explained by the large spreading observed with MTDDIS (as compared with other
models); i.e., the predicted plume covers more samplers than other models
leading to a larger number of predictions in each frequency category.
Also, relatively few very large 10-hr concentrations are predicted with
MTDDIS. An examination of the table (and the set of predicted ground patterns
of the models) also reveals the following: MSPUFF and MESOPLUME have a
combination of the least spreading and the least correct directional
orientation of the models. One may verify this by noting the large number of
predicted zeroes (214 for MESOPLUME and 206 for MSPUFF) in the first row of
Table 5-2. The possible causes of the model/data discrepancies are discussed
in Section 5.5.
5.4. PATTERN COMPARISON METHOD OF MODEL EVALUATION (OKLAHOMA ONLY)
It is interesting to evaluate the accuracy of the ground-level patterns
predicted by the models as compared with the ground-level patterns observed in
the data. It is clear from Figures 5-1 through 5-20 that the predicted and
5-28
-------
Table 5-2. Frequency Distribution of Predicted and Observed Concentrations
for the 15 Savannah River Plant Cases (Based on Predicted/Observed
Pairings in Space and Time) in
MESOPUFF MESOPLUME MSPUFF MESOPUFF II ARRPA RTM-II RADM
OBS PRED OBS PRED OBS PRED OBS PRED OBS PRED OBS PRED OBS PRED
1
( 0
( 100
( 200
( 300
( 400
( 500
( 600
( 700
( 800
( 900
( 1000
( 2500
( 2000
( 3000
( 4000
( 5000
( 6000
( 7000
( 8000
( 9000
(10000
(15000
(20000
(25000
0
- 100)
- 200)
- 300)
- 400)
- 500)
- 600)
- 700)
- 700)
- 900)
- 1000)
- 1300)
- 2000)
- 3000)
- 4000)
- 5000)
- 6000)
- 7000)
- 8000)
- 9000)
- 10000)
- 15000)
- 20000)
- 25000)
- 30000)
> 30000
Totals:
27
240
21
8
4
2
0
3
1
0
0
3
2
1
0
0
0
0
0
0
0
0
0
0
0
0
224
193
84
8
9
2
2
2
1
3
0
0
5
1
1
0
0
1
0
0
0
0
0
0
0
0
0
224
22
240
21
8
4
2
2
3
1
0
0
3
2
1
0
0
0
0
0
0
0
0
0
0
0
0
216
214
54
12
1
9
2
2
1
2
2
1
3
1
2
1
0
0
0
0
0
0
0
0
0
0
0
216
52
240
21
8
4
2
2
3
1
0
0
3
2
1
0
0
0
0
0
0
0
0
0
0
0
0
230
206
89
18
6
1
4
3
1
1
1
0
3
0
3
1
0
0
0
0
0
0
0
0
0
0
0
230
57 133
240 154
21 30
8 6
4 4
2 4
2 2
3 1
1 3
0 0
0 0
3 2
2 1
1 2
0 0
0 0
0 0
0 0
0 0
0 0
0 0
0 0
0 0
0 0
0 0
0 0
203 203
167
240
21
8
4
2
2
3
1
0
0
3
2
1
0
0
0
0
0
0
0
0
0
0
0
0
208
46
328
34
15
8
1
3
7
3
1
0
2
2
2
0
0
0
0
0
0
0
0
0
0
0
0
208
46 146
240 138
21 15
8 10
4 5
2 3
2 4
3 3
1 2
0 1
0 0
3 2
2 1
1 1
0 0
0 0
0 0
0 0
0 0
0 0
0 0
0 0
0 0
0 0
0 0
0 0
228 228
9
240
21
8
4
2
2
3
1
0
0
3
2
1
0
0
0
0
0
0
0
0
0
0
0
0
192
196
35
23
13
5
3
2
0
1
1
0
8
2
3
0
1
0
0
0
0
0
0
0
0
0
1
192
5-29
-------
observed plumes are generally offset spatially from one another. This feature
is complicated by the fact that there is an offset in time as well, as the
main bulk of the predicted plume arrives earlier or later at a given arc than
the observed plume. Furthermore, it is theoretically possible for a model to
predict excellent concentration patterns at the ground (as compared to similar
patterns for the data), yet for the predicted and observed patterns to be
simply offset in space and time from each other. The positioning problem in
space and time that exists creates difficulties in interpreting the AMS
statistics (Section 4.2) since those statistics provide an evaluation of the
models based only on pointwise comparisons of model predictions and data. The
AMS statistics do not separate out dispersion errors from directional errors.
In order to provide a simple estimate of the accuracy of the predicted
patterns, an analysis was performed of the model predictions as they were
available for the Oklahoma cases. In order to provide the maximum spatial
resolution, the 34 x 34 set of gridded concentrations were not used because
the interval between grid points was 20 km. Instead, the predicted hourly
concentration values for the nongridded receptors were utilized, since they
were generally much closer together, especially on the 100 km arc. The
analysis was not performed for the SRP data sets since it was found that no
clearly-outlined pattern in the observed data could readily be identified for
most 10-hour averaging periods because of the sparcity of receptors and their
wide spacing. However, a simplified yet quantitative pattern analysis of the
SRP data cases will be presented in the next section.
The pattern comparison at Oklahoma was carried out by analyzing the
average plume ground concentration pattern during the period of peak observed
concentration as the plume passed each of the following arcs:
* 100 km arc for the July 8, 1980 experiment at Oklahoma
* 600 km arc for the July 8, 1980 experiment at Oklahoma
* 100 km arc for the July 11, 1980 experiment at Oklahoma
The peak periods were: 2230-2315 GMT for the 100 km arc of the July 8
experiment, 0800-1100 GMT for the 600 km arc of the July 8 experiment and
2330-0015 GMT for the 100 km arc of the July 11 experiment. Mote that the
observed data has averaging periods of either 45 minutes or 3 hours, whereas
for the predicted values at the nongridded receptors, the hourly average
5-30
-------
values were used to provide greater temporal resolution for plume passage. In
the case of the 45 minute averaged data at the 100 km arc, the difference is
minimal; however, for the 3 hour averaged data at the 600 km arc, the observed
arrival time was taken as the start of the period. Therefore, the predicted
arrival times for the 600 km arc should be considered as uncertain within
1.5 hours. Fixing these time periods of peak concentrations, three variables
were extracted from the model predictions and the observed data. The first
two (centerline azimuth and transport time) describe the location of the
plume, whereas the third variable (plume width) describes the spreading about
that plume location. These variables are described in more detail below:
* centerline azimuth....the angular direction measured clockwise from
north where the peak of the predicted or observed concentrations
passes the 100 or 600 km arc of interest. The peak in the predicted
plume usually occurred during a different time averaging period than
for the observed data.
* transport time the time from initial release to the time that
the first traces of the plume passed the arc of interest.
* plume width the width of the plume (edge to edge) as the
peak concentration passes the 100 or 600 km arc. Each predicted plume
has a different transport time (not shown) for this peak to reach the
arc of interest.
Tables 5-3 to 5-5 present the results of the these pattern comparisons
for the 7 applicable models. (MTDDIS was not run for the Oklahoma
data base.) Conclusions that may be drawn from an examination of these tables
are:
(a) At the 600 km arc, the observed plume widths were underestimated by
all the models. The observed width of the plume at the 100 km arc
could not be estimated accurately due to horizontal spreading beyond
the samplers. (The extreme sampler to the east or to the west
showed high observed concentrations during the peak period.) The
underestimation of predicted plume widths is consistent with the
conclusions in Section 5.2 and with the general overprediction in
plume concentrations found with the models. Considering that small
predicted values were being compared with observed values near
background in determining plume width, detailed width comparisons
5-31
-------
Table 5-3. Pattern Comparison Results for the 100 km Arc
Oklahoma Experiment.
July 8, 1980;
Model/Obs.
Observed
MESOPUFF
MESOPLUME
MSPUFF
MESOPUFF II
ARRPA
RADM
RTM-II
Centerline Azimuth
358 degrees
000
004
356
020
005
020
015
Transport Time
3.0 hours
2.5
1.5
2.0
3.0
3.0
3.0
4.0
Plume Width
> 50 km
35
40
25
35
60
40
75
Table 5-4. Pattern Comparison Results for the 600 km Arc
Oklahoma Experiment.
,. July 8, 1980;
Model/Obs.
Observed
MESOPUFF
MESOPLUME
MSPUFF
MESOPUFF II
ARRPA
RADM
RTM-II
Centerline Azimuth
010 degrees
024
024
017
Oil
019
024
024
Transport Time
< 13.0 hours
15.0
16.0
13.0
20.0
15.0
22.0
21.0
Plume Width
200 km
140
140
100
160
100
50
180
Table 5-5. Pattern Comparison Results for the 100 km Arc
Oklahoma Experiment.
,. July 11, 1980;
Model/Obs.
Observed
MESOPUFF
MESOPLUME
MSPUFF
MESOPUFF II
ARRPA
RADM
RTM-II
Centerline Azimuth
019 degrees
002
002
002
002
015
002
352
Transport Time
3.75 hours
2.5
2.5
3.5
4.0
4.0
4.0
4.0
Plume Width
> 60 km
40
45
35
60
60
40
60
5-32
-------
among models here are not justified. On the other hand, the models
that predict plumes that are generally much too narrow (such as
MSPUFF and RADM) could be distinguished during at least two of the
three periods.
(b) The centerline azimuth results indicate that the accuracy of the
direction of the peak concentration as it passes the arcs is quite
variable between arcs and experiments. Direction predictions for
MESOPUFF II, RADM, and RTM-II were significantly shifted to the east
for the 100 km arc on the July 8 experiment, and showed the same
shift during passage through the 600 km arc.
(c) The transport time comparisons indicate that the predicted plumes of
three models reach the 100 km arc more rapidly than the observed
plume. However, for the 600 km arc, all predicted plumes lag the
observed plume.
(d) No model shows a consistently superior performance (as compared with
the other models) with respect to the spreading and location
variables presented in Tables 5-3 through 5-5. More accurate and
more definitive pattern comparison results could be obtained in an
experiment where sampler arcs are always wide enough to include all
of the observed plume, where receptors are operational during the
entire episode of plume passage, and where sampling utilizes shorter
and more frequent averaging times. The models would also show more
precise arrival times if they were modified to provide shorter-term
averages than every hour.
(e) Although MESOPUFF II, RADM and RTM-II share the same meteorological
preprocessor, their centerline azimuth, travel time and plume width
values are not identical in these tables. MESOPUFF II differs more
significantly from the other two models, because it alone uses the
upper-level wind field generated by MESOPAC II in transporting the
plume. The other two models rely solely on the wind field averaged
over the mixing layer. The small differences between their
predicted values for these parameters stem primarily from the
5-33
-------
differences in how they handle exchange between the upper layer and
the mixing layer (RADM has only one layer), and in how they compute
surface concentration values from elevated plume material.
5.5. SUMMARY OF DIAGNOSTIC REVIEW OF MODEL/DATA COMPARISONS
In this section, the predictive performance of each of the eight models
is related to the theoretical formulation of the model. The primary tools for
interpretation, in addition to the statistical measures already presented,
are:
>
(a) the contour plots of averaged concentration for each of the 21 time
averaging periods of the two Oklahoma experiments (as shown in
Figures 5-1 to 5-12 and Figures E-12 to E-95), and
(b) the printed panels of 10-hour-average gridded concentrations and
observed values for the 65 time averaging periods for the Savannah
River Plant experiment (as shown in Figures 5-13 to 5-20).
The graphical comparisons cited above for Oklahoma (not available for
MTDDIS) have good spatial resolution at the two arcs of samplers, and good
time resolution for passage of the plume through these arcs. Those graphs
also allow evaluation of model performance over distances as large as 600 km.
However, those releases took place under fairly steady summertime conditions,
and model performance under a variety of meteorological conditions cannot be
inferred from that data. The graphical comparisons for the SRP data sets (not
available for ARRPA) offer much less spatial resolution, with only 13 samplers
located within about 150 km of the source. The SRP comparisons have much
coarser time resolution, since the averaging periods are all at least
10 hours; however, those comparisons include approximately equal numbers of
nighttime and daytime averaging periods, and includes cases from all four
seasons of the year.
Three supplementary tables have been prepared to provide additional
quantitative measures of model predictive performance at each of the two
sites. Table 5-6 lists the maximum concentration that each model predicts (in
5-34
-------
parts/1015) over the whole prediction grid for each of the 21 averaging
periods during the Oklahoma experiment. For each of the averaging periods, a
ranking of the seven models was also prepared, from 1 (lowest maximum concen-
tration) to 7 (highest maximum concentration). The small subtable at the
bottom of Table 5-6 shows the average ranking for each model over the 21
Oklahoma averaging periods. Table 5-7 presents the same analysis of model
predictions in pCi/m^ for the 65 Savannah River Plant data cases (with MTDDIS
predictions replacing those of ARRPA).
The third table (Table 5-8) requires more description. Each of the 65
Savannah River Plant cases for each model is classified with respect to two
major features. First, each subcase was counted in one of 4 categories that
indicates roughly the comparison of the predicted plume trajectory with the
observed one. Some subjective judgment was required in assigning predicted
plume patterns to directional categories, because sometimes only one or two
samplers had observed values above background. Other times all of the
observations were within several pCi/m^ of the background of 16, and one or
more of those samplers could have been recording fluctuations in the
background concentration. Secondly, each case was counted in one of 3
categories that indicates whether the plume spread too much, too little or
about the right amount (implied by category (1) with no entry in (7) or (8).)
Each of these three tables will be referred to in the summary discussion below
of each model whenever the table is useful in drawing inferences concerning
the degree of validity of a model's theoretical formulation.
Issues to receive special mention are (a) the adequacy of the wind field
and the gridded mixing heights predicted by the meteorological preprocessor,
and (b) the accuracy of the spreading rate implied by the model's formulation
of plume dispersion. Also to receive some mention, where possible, is the
method each model uses to compute ground concentrations from its
representation of the three-dimensional spatial distribution of plume
material.
5-35
-------
Table 5-6. Summary of Peak Predicted Concentrations for Oklahoma Data Sets,
July 1980 (parts/1015).
Day/Time(GMT) MESOPUFF MESOPLUME MSPUFF MESOPUFF II ARRPA RADM RTM-II
08 2100-2145
08 2145-2230
08 2230-2315
08 2315-0000
09 0000-0045
09 0045-0130
09 0130-0215
09 0800-1100
09 1100-1400
09 1400-1700
09 1700-2000
09 2000-2300
09 2300-0200
11 2200-2245
11 2245-2330
11 2330-0015
11 0015-0100
11 0100-1045
11 0145-0230
11 0230-0315
11 0315-0400
9185
8014
6272
5180
4079
2804
2860
1017
869
767
662
546
335
542
438
439
407
418
355
296
318
4807
3904
4173
4217
6574
4918
4578
1147
976
956
941
961
1033
653
566
557
780
831
572
489
569
180365
124975
64854
20190
12558
9635
7798
924
896
626
0
0
0
5367
4538
3302
3110
1323
1238
1198
1053
822884
274294
4375
7609
3447
3222
2115
1575
825
387
421
419
214
812
838
997
1149
762
677
470
478
7139
26494
26494
7713
8426
5531
5695
3072
2462
2258
217
0
0
935
945
694
773
725
909
845
723
147472
63413
28028
29047
17227
9373
7074
9831
4974
1026
2232
1296
418
4105
2154
2310
' 2442
1836
1730
1343
1295
4576
4040
4845
5262
2876
1790
1206
88
44
12
3
1
0
748
673
637
709
315
165
125
111
Average rank among models of maximum concentration by case:
(1 = lowest max. cone., 7 = highest max. cone.)
Average
RTM-II
MESOPUFF
MESOPLUME
MESOPUFF II
ARRPA
MSPUFF
RADM
1.86
2.76
3.67
3.81
4.38
5.24
6.38
5-36
-------
Table 5-7. Summary of Peak Predicted Concentrations for Savannah River Plant
Data Sets, pCi/m3.
Case
MESOPUFF MESOPLUME MSPUFF MESOPUFF II MTDDIS RADM RTM-II
1 — —
2A —
2B —
2C —
3A —
3B —
4A —
4B —
4C —
4D —
5A —
5B —
5C —
5D —
6A —
6B —
Cjr
OV»
6D —
oE ~™"
7A —
7B —
8A —
QQ _
OD
8C —
8D —
8E —
8F —
8G —
9A —
9B —
9C —
9D —
9E —
9F —
9G —
9H —
91 —
9J —
10A —
10B —
IOC —
1740
19209
1609
12936
1675
1467
240
872
1747
0
3978
10509
3373
4794
9753
1949
199
2937
292
399
0
141
3551
288
1053
1607
98
0
12771
832
303
7383
4532
247
19
3
2975
1279
5066
6957
3877
2241
10604
322
11581
936
1227
236
391
374
0
1995
13622
1693
1596
2657
1390
618
2042
359
339
0
160
586
258
566
198
66
0
3782
4824
1514
144
6847
238
24
0
1331
2230
13141
1813
5686
4049
7451
183
17269
891
735
509
746
613
0
- 1487
1201
3348
3446
7186
3757
3219
51641
12513
251
0
366
2564
886
2273
477
351
0
1564
40
0
900
1914
183
101
1644
2465
20004
20590
5275
17688
12820
2398
747
3867
39425
1660
21040
8253
26014
0
23495
10606
26222
20559
19999
45342
574
7830
17398
2078
0
1756
56844
7700
46643
5378
1355
0
49340
1503
412
6153
31018
1096
759
6839
48820
4214
10507
16676
79640
28581
3694
959
51
20874
3299
3467
380
2753
103
3077
245
1883
217
154
1599
135
5159
1147
552
34
0
3147
3818
6466
1446
1262
26
4051
339
298
3370
2879
440
83
907
1537
1968
27851
4124
11369
8691
16875
16185
11250
3163
5276
623
3522
2924
0
2367
13557
32283
4658
8958
10408
4886
11531
3058
6605
0
2950
5708
2075
2290
1352
700
0
12969
3102
3294
9109
6764
4961
1409
3263
4508
4327
35005
30533
13089
1940
3595
281
5446
558
455
254
508
319
0
964
1921
1470
540
2193
1310
208
1533
280
245
0
180
801
187
222
174
57
0
3885
606
481
1447
484
178
49
75
724
1198
1328
5055
4946
5-37
-------
Table 5-7. (continued) Summary of Peak Predicted Concentrations for Savannah
River Plant Data Sets, pCi/m3
Case
MESOPUFF MESOPLUME MSPUFF MESOPUFF II MTDDIS RADM RTM-II
10D —
10E —
10F —
10G —
10H —
101 —
10J —
11A —
11B —
11C —
11D —
12A —
12B —
13A —
13B —
14A ~
14B —
14C —
14D —
14E —
ISA —
15B —
15C —
15D —
6825
160
26188
424
7054
822
383
369
499
3409
357
3210
3547
5562
162
0
51
74518
5293
315
8554
1263
344
119
4159
713
2257
987
3304
753
242
404
646
3392
270
. 2579
7844
3856
220
0
1114
13106
1656
923
7214
2025
284
141
11311
666
5724
1578
7468
1429
171
123
809
8251
494
3115
18072
554
0
0
118
10631
11802
343
23214
23003
690
0
38669
6459
67818
12682
82569
19341
19465
4928
16072
38785
9069
53283
15054
18265
60
0
0
74888
37869
871
79738
48624
1547
193
1820
207
66666
1968
17589
1763
3511
244
4440
65168
8938
3201
10123
4787
625
0
16
16859
1357
388
14603
2965
754
315
10326
5925
22897
2231
13269
4346
2954
7356
1902
6720
1232
5312
17664
6081
1273
0
904
37242
2904
1306
10744
3565
2813
2531
1048
189
4363
620
1869
615
138
139
587
1750
306
920
4125
588
96
0
27
3604
1177
258
1378
1400
251
42
Average rank among models of maximum concentration by case:
(1 = lowest max. cone., 7 = highest max. cone.)
Average
RTM-II
MESOPLUME
MESOPUFF
MSPUFF
MTDDIS
RADM
MESOPUFF II
1.80
3.05
3.56
3.72
4.11
5.79
5.97
5-38
-------
Table 5-8. Classification of Model Prediction Types for the Savannah River
Plant.
Predicted Plume MESOPUFF MESOPLUME MSPUFF MESOPUFF II MTDDIS RADM RTM-II
1
2
3
4
5
6
7
8
- Right
direction
- Rotated
clockwise
Categories
30
14
- Rotated counter- 9
clockwise
- Rotated appx.
180 degrees
- Complex,
confused
- Pred. all 0;
Observed not
- Narrow
- Wide
6
2
4
Categories
26
0
related
25
18
9
6
1
6
related
28
0
to plume
15
24
9
7
1
9
to plume
24
1
trajectory:
28
20
10
1
1
5
diffusion
10
8
47
5
9
1
1
2
rate:
9
' 24
28
20
11
1
1
4
22
0
31
13
9
6
2
4
12
1
NOTES
(1) The predicted plume direction was correct within about 10-20 deg. as
compared with the observed plume.
(2) The predicted plume was rotated clockwise (as viewed from above) at least
20-45 deg. from the observed plume.
(3) The predicted plume was rotated counterclockwise (as viewed from above by
at least 20-45 deg. from the observed plume.
(4) The predicted plume direction appeared to lie about 180 deg. from that
indicated by the available data.
(5) No simple relationship existed between the predicted and observed plumes,
but there was significant disagreement between them.
(6) The predicted plume concentrations were zero at all grid points, but
there was at least one observation above background.
(7) The predicted plume was narrower than the observed plume.
(8) The predicted plume was wider than the observed plume.
5-39
-------
5.5.1. MESOPUFF
In the evaluation of the performance of MESOPUFF, it is useful to try to
separate the causes of model/data discrepancies between the meteorological
preprocessor MESOPAC and the plume dispersion model MESOPUFF. The wind field,
mixing heights, and stability classes are provided by the meteorological
preprocessor MESOPAC. Plume growth and the calculation of ground concen-
trations are features of MESOPUFF. The separation of effects is impossible to
carry out completely, since mixing heights and stabilities (computed in
MESOPAC) influence plume spreading and concentration calculations (computed in
MESOPUFF). One distinction, however, is that the wind field alone affects
plume advection leading to the predicted trajectories. Obviously, conclusions
concerning model/data discrepancies relating to the plume trajectory and times
of arrival of the plume at fixed receptors are easier to make than conclusions
relating to plume model performance (e.g., the rate of diffusion and the
magnitude of predicted ground concentrations).
The discussion now focuses on the performance of MESOPAC in predicting
the location and time of arrival of the predicted plume. The summary of the
behavior of plume trajectories at SRP shown in Table 5-8 indicates that in
nearly half of the cases (30 cases), the direction of plume travel is
approximately correct. However, for a substantial fraction of the cases (29
cases) the plume was rotated either clockwise (14 times), counterclockwise (9
times) or advected in nearly the opposite direction from the data (6 times).
The cause of the mild preference for clockwise rotation can be suggested.
Since the wind field predicted by MESOPAC for SRP (and Oklahoma) is computed
by averaging twice-daily rawinsonde winds over 1500 m above ground (and then
interpolating in space and time), some of the effects of the Ekman spiral will
be observed. The Ekman spiral involves rotation clockwise of the wind
direction with elevation (as viewed from above) relative to the surface winds.
It appears from examination of the meteorological data and results of the
model runs that the surface wind direction is more appropriate for
characterizing transport of the ground-level plume, even though the plume
material is likely to be vertically mixed throughout the mixing layer roughly
20-50 km from the point of release. It is likely that wind direction shear
causes mixing to occur in a skewed sense, following the Ekman spiral.
It is interesting that 6 cases occurred where the predicted plume was
5-40
-------
advected in nearly the opposite direction from the observed plume. The
probable reason is the single-layer nature of the MESOPAC wind field. Under
some meteorological conditions, the wind direction in an elevated layer is
opposite to the surface wind, and a single-layer model like MESOPAC with
1500 m averaging will obtain a wind direction that includes the elevated
layer. If the surface layer up to the inversion is shallow when the winds in
the two lowest layers are opposite, the average wind direction will be nearly
that of the elevated layer. Then the plume material that leads to ground
concentrations will advect in the direction of the winds in the surface mixing
layer, whereas the predicted plume will advect in the opposite direction.
This interpretation is strengthened by noting from Table 5-8 that MESOPUFF II
(whose wind field is provided by the two-layer MESOPAC II model) produces only
one such case.
A second result of the simplifying assumptions used in the MESOPAC
meteorological preprocessor is that during the nighttime averaging periods at
SRP, new puff releases are likely to remain within two stack heights (124 m)
of the ground, and a 1500 m vertical averaging of the horizontal winds will
not correctly represent transport of these new puffs. Within MESOPAC, one
could attempt a 1500 m averaged wind for daytime transport and a 150-m
averaged wind for nighttime transport. MESOPAC, as formulated, requires an
input wind value from the soundings before a mixing height is calculated.
That procedure for averaging the winds over the local mixing height would lead
to too low a wind speed at night for transport of the plume puffs already
diffused aloft. The development of MESOPAC II was, in part, aimed at solving
the inherent problems in defining a single layer wind field.
Examination of the Oklahoma results reveals that the plume passes through
the 100 km arc at about the right location, but somewhat early. It passes the
600 km arc clockwise of the point where the samplers show maximum concen-
trations. The predicted plume then veers quite strongly to the east, possibly
in response to an approaching front. The plume's arrival at 600 km is
somewhat late (by at most 2-3 hours), probably due to the presence the night
before of a strong elevated nocturnal jet (documented in Appendix E) which is
only partially represented by a 1500-m averaged wind. This behavior tends to
confirm the interpretation given above for the SRP results. The plume angular
offset caused by using a wind field that is averaged over the Ekman spiral
does not lead to a large lateral offset distance at the 100 km arc.
5-41
-------
However, in general the angular error will lead to a large lateral offset
distance at the 600 km arc.
The remarks which follow summarize the findings with respect to the plume
model MESOPUFF. The pattern comparison analysis in Tables 5-3 to 5-5
indicates that the spread predicted by MESOPUFF is less than that of the
observed plume at the 100 km arc, (as is the case for all of the models except
RTM-II and ARRPA). However, examination of the entries in Table 5-6 for
MESOPUFF shows that its predicted maximum concentrations at Oklahoma are the
second lowest for the seven models applied to that data base.
The model apparently predicts insufficient lateral spreading as may be
seen in the SRP model/data comparisons in Table 5-8. However, the maximum
concentrations predicted by MESOPUFF usually rank in the middle of the
maximum concentrations predicted by the other six models (an average rank
order of 3.56 on a scale of 1 to 7), as demonstrated by Table 5-7. For very
short-term averages (2-3 hours), such small spreading can be interpreted as
inadequate treatment of small-scale processes, e.g. too small a value for av.
For 24-hour averages, too small a lateral spreading would most likely be due
to wind shear and plume meander, i.e. inadequate resolution in the
meteorological preprocessor. For 10-hour averages, as in the present case, a
combination of both deficiencies is likely to be present.
Concerning small-scale processes, the use in MESOPUFF of the Turner
curves out to 100 km does not provide sufficient spreading, especially under
stable conditions. The Turner oy values are more applicable in the range up
to 10 km and for flat level terrain. The transition point for the Heffter
formulation should probably be no farther out than 10 km. Additional
numerical experiments with these SRP data cases can better verify that point.
The results of the model/data comparisons for the first Oklahoma release
tend to confirm, with higher spatial resolution, the observations of plume
spreading evident in the SRP predictions. As the plume passes the first arc
(100 km), its width and location are approximately correct, although predicted
plume arrival is a little late. Very early plume dispersion should be
adequately represented by the Turner curves. In MESOPUFF, the transition to
the Heffter formulation occurs at 100 km, which is probably at too long a
distance because large-scale processes have already become important; in any
case, the effect of this late transition from the Turner curves to the Heffter
formulation has not had significant effects on the predicted concentrations
5-42
-------
and spreading. As the plume arrives at the 600 km arc, it is still narrow by
about 30%, due to the presence of early spreading rates that were too small.
5.5.2. MESOPLUME
The theoretical formulation of the MESOPLUME model is very similar to the
MESOPUFF model, and the predictions of both models have many similarities.
Both models use the same meteorological preprocessor, MESOPAC, with identical
inputs and outputs. As a result, the trajectories and times of arrival of the
plumes should be about the same for both models. In the treatment of plume
dispersion, both models use the Turner curves to obtain stability-dependent
Gaussian plume growth rates out to 100 km. Beyond 100 km, the formulation by
Heffter is employed. The predictive differences that do emerge between the
models must therefore derive from
(a) their different geometrical representations of the shape of the
region that is spreading according to the Gaussian widths, and
(b) their formulas for predicting ground-level concentrations.
In MESOPLUME, the plume is viewed as a series of cylinder-like plume
segments. In MESOPUFF, the plume is represented by spherical "puffs." In
both cases the growth of lateral and vertical dimensions is governed by the
Turner curves and the Heffter formulation, but in calculation of the ground
concentrations, the geometrical differences in plume shape will lead to
differences in model predictions. Thus, comparisons of the predictions of the
two models sheds light on the differences that result from the two theoretical
approaches to defining plume shape and the calculation of ground-level concen-
trations.
In spite of the numerous similarities in the models, an examination of
the predictions for the two Oklahoma releases reveals that MESOPLUME and
MESOPUFF have significantly different ground concentration predictions. For
the first release (from 1900 to 2200 GMT on July 8, 1980), the maximum concen-
trations listed in Table 5-6 for MESOPLUME are nearly a factor of two less
than the MESOPUFF predictions for the first 1.5 hours (from 2100 GMT).
5-43
-------
However, the predictions of the two models become equal after 3 to 4 hours.
Thereafter, MESOPLUME predicts higher maximum concentrations than does
MESOPUFF. During the second release (from 1900 to 2200 GMT on July 11, 1980),
the maximum concentrations predicted by MESOPLUME are higher than those
predicted by MESOPUFF for the entire 6 hours after the release, beginning with
a factor of about 1.2 and ending with a factor of 1.8.
At both arcs for the first release, and at the 100 km arc for the second
release, the width of the MESOPLUME predicted plume, however, is equal to or
greater than that of the MESOPUFF predicted plume. This behavior offers an
interesting insight into the geometrical differences between these models.
Unlike the SRP data sets, the Oklahoma data sets feature a ground-level
release with the 100 km arc relatively close to the source. Apparently, the
use of cylindrical plume segments in MESOPLUME as contrasted with the
spherical plume puffs in MESOPUFF, and the use of slightly different
methodologies for predicting ground concentrations from the elevated puffs,
combine to provide an opposite trend when the material is released near the
ground, particularly near the source. The larger concentrations listed for
MESOPLUME in Table 5-6 for the first release (before passage through the
100 km arc) are, thus, coupled with a wider plume spread at the ground as
compared with MESOPUFF. These differences come solely from differences in the
method of calculating ground concentrations from plume parcel shape and not
from differences in rates of plume diffusion. Although both models are
examples of puff models, the methods used in MESOPUFF yield more accurate
predictions in this study than the methods used in MESOPLUME.
The pattern analysis results in Tables 5-3 to 5-5 show that the times of
arrival at the Oklahoma sampling arcs are identical to those of MESOPUFF
within 1 hour. The 1 hour difference for the first release is due to
different amounts of predicted spreading in the ground-level concentration
pattern, because the underlying wind field is the same for both models.
Inspection of any corresponding pair of concentration isopleth graphs for
Oklahoma shows this differing amount of spread at ground-level. These
differences in the spreading of the ground-level plume cannot be attributed to
the use of different plume dispersion coefficient formulations. Rather, they
come from differences in plume parcel geometry and differences in the method
of computation of ground-level concentrations.
It is more difficult to draw definitive conclusions about the respective
5-44
-------
merits of the concentration averaging formulations of the two models for the
SRP model/data comparisons because of the larger averaging periods for which
these comparisons were made (10 hours) and due to the large separation between
receptor locations. An examination of the data show that for those 10-hour
averaging periods when receptors have concentrations above zero, it is usual
that only one to three of them report the presence of the plume, and these
points are not arranged in clearly defined arcs giving good spatial
resolution. As mentioned previously, this blurs distinctions between the
effects of the dispersion formulations and the effects resulting from the
plume trajectory positioning. However, in general, one may say that MESOPLUME
shows some tendency to overpredict the observed concentrations when the
observed values are above background. This is evident both from the scatter
plots of observed data versus predictions, the graphs of frequency of
residual, the graphs of cumulative frequency distributions of observed and
predicted values, and the graphs of average versus residual. The statistical
results reveal this conclusion as well. There is only a moderate difference
in predictions between the two models for the SRP data. More significant is
the fact that MESOPUFF predicts nonzero concentrations 68 times when the
observed concentration is nonzero, whereas MESOPLUME does so only 56 times.
From that information, one may surmise that MESOPLUME leads to a smaller
spreading at ground level than does the MESOPUFF model. Results in Table 5-8
support these conclusions. In 6 cases, MESOPLUME predicted zero concen-
trations at every grid point when at least one sampler showed the presence of
tracer material, whereas MESOPUFF did so in only 4 cases. The MESOPLUME
predicted plumes were in the right direction for 5 fewer cases than were the
MESOPUFF predicted plumes. The directional error in the MESOPLUME predicted
plume was a clockwise rotation from the observed plume in 4 of the 5 cases for
which the plume predicted by MESOPUFF was in the right direction.
The conclusion that the MESOPLUME predicted ground patterns were narrower
than those of MESOPUFF for the SRP experiment should be contrasted with the
conclusion for the Oklahoma experiment that MESOPLUME seemed to have broader
ground concentration patterns. By mass conservation, a narrower predicted
plume should contain higher maximum concentrations. However, an examination
of Table 5-7 shows that MESOPLUME predicts lower peak concentrations on the
average than does MESOPUFF. This difference, opposite to what is expected
with plume material uniformly distributed in the mixing layer, must be a
5-45
-------
result of the different methods for computing ground concentrations from
elevated parcels. For the Oklahoma data sets, the effects of the cylindrical
and spherical puffs are being compared with the centers of both types of puff
relatively near the ground, whereas for the Savannah River Plant data sets
they are being compared with the centers of both types of puff elevated by at
least 62 m. Apparently, the effects of these differences in puff geometry on
ground concentrations are greatest when the puff centers are near the ground.
In summary, for the Oklahoma experiment, the plume at ground level predicted
by MESOPLUME is wider and has larger concentrations than the plume predicted
by MESOPUFF. On the contrary, the MESOPLUME predicted plumes for the Savannah
River Plant data base were narrower and had lower concentrations than the
plumes predicted by MESOPUFF.
5.5.3. MSPUFF
I.t might be expected that the similarity of the formulation of MSPUFF to
that of MESOPUFF would lead to very similar predictive behaviors for both
models. However, significant differences emerge from a comparison of both
models with the data. For this reason, the discussion of the predictive
performance of MSPUFF as it relates to its formulation will focus on (a) the
differences between the performance of MSPUFF and the performance of MESOPUFF,
or (b) the predictive accuracy of MSPUFF with respect to observed data. An
examination of Tables 5-3 to 5-8 yields the following performance
characteristics of MSPUFF:
(a) The MSPUFF predicted ground concentration patterns often appear
similar to the observed pattern in shape, but seem to be rotated
about the source through some angular offset, and
(b) The ground-level concentrations are overpredicted with the width of
the pattern usually narrower than the observed pattern.
Point (a) is evident from the centerline azimuth comparisons in
Tables 5-3 and 5-4 for Oklahoma, and in categories (1) through (3) in
Table 5-8 for the SRP data sets. At SRP the MSPUFF plume was in the right
5-46
-------
direction for only 15 of 65 cases, and was rotated clockwise in 24 cases. In
the same tables, MESOPUFF shows a higher degree of directional accuracy.
Since MSPACK, (the meteorological preprocessor for MSPUFF) employed 850 mb
winds for its single-layer wind field, and the user option of averaging the
winds over 1500 m was used for the MESOPAC runs, the wind directions predicted
by MSPACK tended to be rotated clockwise from those of MESOPAC. This
increases the clockwise rotational errors already present in MESOPAC due to
the Ekman spiral effect. Model/data discrepancies involving the predicted
trajectories may be explained largely by the choice of these 850 mb winds in
MSPACK for both the Oklahoma and SRP data sets. For the SRP data sets, MSPUFF
showed the greatest tendency of any model to predict a plume that has been
rotated clockwise as compared to the data. This effect should be most
pronounced during the nighttime hours, when convective mixing does not tend to
eliminate the spiral. In only 15 of the 65 cases was the direction of the
predicted plume judged to be correct, fewer than for any other model.
In support of point (b), Table 5-8 shows that plumes narrower than the
observed plume were predicted in 24 cases. The tendency toward overprediction
was documented in the statistical analysis of Section 4.2. Also, Table 5-6
indicates that peak ground concentrations predicted by MSPUFF for the Oklahoma
data usually exceed those of MESOPUFF and MESOPLUME. Only RADM predicts
larger peak concentrations. (Table 5-7 shows that peak ground concentrations
predicted by MSPUFF for the SRP data exceed those of both MESOPUFF and
MESOPLUME.) This physical behavior is currently an unresolved issue.
Several possibilities have been examined, however. First, as discussed in
Section 2, the Turner curves employed by the MSPUFF model have been altered
slightly in the range less than 50 km in order that steady-state predictions
of MSPUFF would agree with the MPTER model. This difference between
dispersion coefficients can lead to some differences in plume concentration
predictions for MSPUFF as compared to MESOPUFF. No extensive comparisons have
been made of the differences, however. Second, the Heffter formulation for
dispersion was used beyond 50 km from the source rather than 100 km as in
MESOPUFF. This change should lead to lower concentrations in MSPUFF as
compared to MESOPUFF beyond 50 km. The differences in dispersion coefficients
are not, therefore, considered to be the main cause of the behavioral
differences between the MESOPUFF and the MSPUFF predictions. By a process of
elimination, it is suspected that the larger concentration predictions at
5-47
-------
ground level given by MSPUFF as compared to MESOPUFF are due to changes in the
mixing height formulation. The two key changes made there in MSPUFF were:
(a) surface winds were used to compute the mechanical mixing height,
and
(b) turbulent dissipation of convection was added to the computation
of the convective mixing height in the Benkley-Schulman method.
It is not clear at this time what the precise effects of these changes
were on the prediction of mixing heights during the diurnal cycle.
The pattern comparison results presented in Tables 5-3 to 5-5 also
support these conclusions. The width of the MSPUFF predicted plume was
consistently less than that of MESOPUFF or MESOPLUME when the plume passed the
sampler arcs. The MSPUFF plume time of arrival, however, was better predicted
than that of MESOPUFF and MESOPLUME in two of the three tables because the
850 mb winds tend to be larger than the 1500 m layer-averaged winds.
5.5.4. MESOPUFF II
The MESOPUFF II model contains a methodology for predicting plume
dispersion and ground-level concentrations that may be considered as an
extension or improvement of that in MESOPUFF. Plume transport occurs by means
of a wind field generated by MESOPAC II. This meteorological preprocessor
also provides gridded values of mixing height, stability and other
micrometeorological variables. To be discussed first is the predictive
accuracy of the model for plume trajectories and its implication for the
accuracy of the plume model MESOPUFF II.
The accuracy of the MESOPAC II wind field prediction was determined by
examining the accuracy of the predicted plume trajectories, as listed in
Table 5-8 for SRP. In the prediction of wind fields for the Oklahoma and SRP
data bases, MESOPAC II differs from MESOPAC in two key ways:
(a) MESOPAC II utilizes a two-layer wind field as compared to the single-
layer wind field of MESOPAC, and
5-48
-------
(b) MESOPAC II utilizes surface wind directions to help adjust the
direction of the layer-averaged wind (lower layer).
The fact that the predicted plume often follows the Ekman spiral as it
disperses vertically is supported by the statistics in Table 5-8, where the
model predicted clockwise rotated plumes in 20 cases, but counterclockwise
rotated plumes in only 10 cases. If a model shows both directions of
rotation, that may reflect the naturally poor spatial and temporal resolution
of the input wind data. If the clockwise direction occurs most of the time,
it suggests an effect of the Ekman spiral, as discussed above. The 28 cases
of correct plume direction compared favorably with the other models, with only
MSPUFF predicting markedly fewer and MTDDIS predicting markedly more cases of
correct plume orientation. For plume direction, the predictions of MESOPAC II
appear to have the same accuracy as those predicted by MESOPAC. Both models
showed a clear increase in clockwise rotated plumes as compared to
counterclockwise rotated plumes.
For MESOPUFF II, categories (7) and (8) in Table 5-8 indicated that the
rate of lateral plume spread is more accurately predicted than for MESOPUFF.
Whereas the predicted plumes from MESOPUFF were never too wide (either about
right or too narrow), those from MESOPUFF II were more balanced between too
wide and too narrow (8 too wide versus 10 too narrow) for the SRP data. The
main source of this difference is the use in MESOPUFF II of the user-input
distance of only 10 km at which a transition takes place from dispersion rates
based on the Turner curves to rates based on Heffter's curves. Since
spreading rates implied by the Heffter curves are higher than those based on
the Turner curves at large distances from the source, allowing the former to
take effect closer to the source leads to considerably more spreading. The
slight excess of too little spreading (5) over too much spreading (6),
suggests that 10 km may still be too far from the source to make the
transition, or that other scale considerations need to be included in
specifying the lateral spreading rate.
Results shown in Table 5-7 do not contradict this point even though
MESOPUFF II has the largest maximum concentrations of any of the models for
the SRP runs. These very large concentrations occur at points near the source
at times when the source is emitting krypton-85. However, beyond 20 km (the
distance to the nearest sampler), the maximum concentration predictions of
5-49
-------
MESOPUFF II are usually smaller than those of the other models.
The predictive results for the two Oklahoma releases yield a different
comparison. The predictions of MESOPUFF (driven by MESOPAC) are uniformly
more accurate than those of MESOPUFF II (driven by MESOPAC II). As Table 5-6
illustrates, maximum concentrations over the grid for MESOPUFF II were larger
than those for MESOPUFF, both for the first few hours after release and during
peak convective mixing hours; those for MESOPUFF were more nearly equal to the
maximum observed concentrations.
The plume predicted by MESOPUFF II for the Oklahoma data sets seemed
spatially more compact. After the first release, the plume passed to the east
of the appropriate samplers, moved too slowly, arrived late and remained for
1-2 hours after the samplers ceased indicating tracer material above
background. On the contrary, the MESOPUFF plume passed through the active
part of both arcs, arrived on time at the 100 km arc, and had approached the
600 km arc more closely than MESOPUFF II during the averaging period when the
samplers were first activated and measured nonzero plume concentrations. It
appears that the predictive advantages that MESOPUFF II and MESOPAC II
possessed over MESOPUFF and MESOPAC for SRP became disadvantages at Oklahoma.
The meteorological conditions for the Oklahoma experiment were markedly
different than those for the SRP experiment. The latter data cases
represented a variety of conditions from four seasons, whereas the former
represented only hot summertime conditions with steady wind flow from the
south or southwest at most locations for most of the experimental periods.
The maximum concentrations listed in Table 5-7 show lower values predicted by
MESOPUFF II at distances beyond 100 km than those predicted by MESOPUFF. This
behavior further confirms that the dispersion formulation of MESOPUFF II
embodies more rapid plume spreading beyond 100 km than does the formulation in
MESOPUFF.
However, in drawing preliminary conclusions on the relative merits of the
theoretical formulations of the two models, it must be remembered that their
meteorological preprocessors have sufficiently large theoretical differences
and that considerable variability in predictive behavior can be expected. In
MESOPAC, which provides the wind field and mixing heights for MESOPUFF, the
wind field is simply interpolated from the layer-averaged winds at rawinsonde
locations to grid locations, whereas in MESOPAC II a more detailed weighting
scheme is used involving both the layer-averaged winds and surface winds from
5-50
-------
reporting stations. In MESOPAC, the mixing heights are obtained as the
maximum of the convective and mechanical values computed without the use of
measured cloud cover and ceiling height data. (Constant typical values are
assumed for cloud cover and ceiling height.) In MESOPAC II, cloud cover and
ceiling height data are used in connection with micrometeorological variables
in obtaining both mixing heights and stability classes. Resulting values of
stability and mixing height at grid points often differ substantially,
affecting the predicted rates of plume spreading. In MESOPUFF II, unlike
MESOPUFF, a two-layer wind field is also used, with puffs in the upper layer
following a different trajectory than puffs in the mixed layer. When mixing
heights rise in MESOPUFF II, either as a function of space or time of day,
puffs in the upper layer can influence ground concentration values once again.
Our conclusion is that the results of this limited comparative study
cannot discriminate between the effects of -these differences in model
formulation. Further and more detailed study would be needed to determine
whether each of these differences produces a net improvement or net worsening
of model predictive performance.
5.5.5. MTDDIS
The MTDDIS model uses considerably different theoretical assumptions than
the four Gaussian puff models considered above. Because it is formulated only
for elevated releases, it was applied solely to the Savannah River Plant
cases. It contains its own meteorological preprocessor, which employs only
wind data at the release height, rather than layer-averaged winds or two-layer
winds. Its mixing heights are computed by a methodology that begins with NWS
mixing height values from a single location in the experimental region. Other
differences are noted in Section 2. The MTDDIS theoretical formulation is
sufficiently distinct from the other models that it becomes more difficult to
separate the effects of a single assumption in comparing the model's
predictions with those of the other puff models. The important
characteristics of the model/data comparisons of the MTDDIS model with the
Savannah River Plant data base are (see Tables 5-7 and 5-8):
5-51
-------
(a) plume trajectories appear to be the most accurately predicted of the
seven models compared,
(b) plume concentrations tend to be overpredicted, and
(c) the model tends to make slight overpredictions of horizontal
spreading.
A major distinction between the MTDDIS predictions and those of the other
models is that MTDDIS overpredicts spreading whereas the other models
underpredict spreading (except RTM-II for Oklahoma). The expected theoretical
causes of these physical characteristics of the MTDDIS predictions are:
(a) Plume trajectories are well predicted by MTDDIS due to the use of
wind directions that are representative of the effective height of
release of the source at SRP. Other models employ wind directions
taken from the 850 mb level/ or computed as an integrated mixing
layer average, or computed as an average to 1500 m. Apparently, for
the modeling of the relatively short mesoscale distances at SRP (less
than 150 km), wind directions at the effective height of release for
industrial stack applications (not ground release applications) may
be appropriate. Predictions for much longer distances (i.e., longer
transport times) may require wind directions that represent puff
movement at higher effective elevations from the ground than are used
by MTDDIS.
(b) Plume concentrations tend to be overpredicted. All eight models
exhibit this trend for both Oklahoma and SRP data sets, except RTM-II
for Oklahoma. This is apparently a symptom that the vertical and
horizontal mixing are not properly handled in any of the models.
This overprediction occurs despite a considerable range of
assumptions for modeling dispersion. For the dispersion methodology
used in MTDDIS, as for each of the other approaches, there seems to
be a need for better determination of the dispersion coefficients.
5-52
-------
(c) The slight overprediction of horizontal spreading is a very
interesting feature of the MTDDIS predictions. Two possible causes
may be at work here. First, the use of the horizontal dispersion
coefficient formulation taken from Heffter^l applied directly from
the release point undoubtedly adds more lateral spreading to the
plume than the Turner curves (at least for the neutral and stable
classes). Second, the use of hourly wind directions and speeds taken
from the effective height of release provides, in general, a wider
variation in transport time and direction of travel than winds with
directions taken from the 850 mb level, or directions obtained from
averaging over the mixing height or a fixed height, such as 1500 m.
These greater variations lead to the transport of puffs to a wider
range of directions, and cause the plume material to be spread out
more along the direction of travel. In essence, then, the Heffter
formulas appear to provide a greater spreading due to small-scale
processes, and the use of wind .directions at the effective height of
release appears to provide greater plume meander and longitudinal
stretching.
In summary, the MTDDIS predicted plumes cover a larger ground area than
do those of the other models, and show a slight overprediction of lateral
spreading when compared with the data.
A logical question arises then as to the accuracy of the mixing height
algorithm since an overprediction of both horizontal spreading and 10-hour
averaged concentrations occurs at ground level. It appears that mixing
heights may be under-predicted based on simple conservation of mass principles.
A detailed examination of this possibility was beyond the scope of the
project. A re-examination of the choice of roughness height was made to
determine whether values that were too small were used, producing a small
mixing height. It was found that the MTDDIS computer runs had used values
between 10 and 30 cm which were, indeed, characteristic of the land use in the
modeled region.
5-53
-------
5.5.6. ARRPA
Because the National Weather Service Boundary Layer Model wind field
predictions were not available for the time period of the Savannah River Plant
experiment, the ARRPA model could only be run for the Oklahoma experiment.
Comments here on the model's predictive behavior are based on Tables 5-3 to
5-6. Evaluation of the results in these tables and of the statistical
measures presented in Section 4 reveals the following characteristics of the
ARRPA predictions:
(a) the BLM wind field provides plume transport that is quite accurate.
The ARRPA plume arrives slightly later than the observed plume
(generally by one averaging period). On the other hand, the ARRPA
plume transport is more rapid than any of the predictions of the
other six models tested at Oklahoma, and
(b) the ARRPA plume predictions are characterized by overprediction of
concentrations and underprediction of lateral spreading.
These features characterize both ARRPA runs for the Oklahoma data base.
Table 5-6 reveals that the ARRPA predictions for maximum concentrations were
larger than the predictions of all of the models except MSPUFF and RADM, and
tended to decrease less rapidly with downwind distance than those of MSPUFF.
The same characteristic features become evident also in Tables 5-3 to 5-5.
The predicted azimuthal angle of plume passage through both arcs compares
favorably with the observed azimuthal angle of passage in all three tables.
Plume widths are also close to the observed widths at the 100 km arc, but are
too narrow at the 600 km arc. The transport times predicted by ARRPA are also
close to the observed times.
The wind field is quite well predicted based on an examination of the
location of the concentration distribution. Specific comparisons of wind
components with upper air or surface data have not been carried out since such
a comparison, although valuable, was beyond the scope of the present effort.
The BLM model would be expected to do well in these Oklahoma cases since there
was a well-defined pressure gradient present. The BLM model has difficulty in
cases of light and variable winds52, common in the summer season. In such
5-54
-------
situations, the surface topography becomes important, and the 80 km x 80 km
grid provides too little spatial resolution.
The well-predicted plume transport for the Oklahoma runs is, however,
surprising considering that a ground-level release was used in the experiment.
The BLM model has difficulty resolving winds in the lower layers of the
boundary layer52. The vertical velocities in the surface layer are essen-
tially negligible. Apparently, the richer set of upper air and surface
stations along with the more sophisticated physical treatment of the
meteorological interpolation (based on conservation equations) explains the
greater wind field prediction accuracy. It should be kept in mind that this
evaluation has been only with two cases, and that further evaluation of the
BLM model over a wide range of meteorological conditions (with ARRPA) is
required before more conclusive comments can be made about the BLM wind field
model.
The cause of the overprediction of concentrations and the underprediction
of lateral spreading is likely due to the choice of Ky for the lateral
dispersion coefficient. This coefficient, dependent on stability class, is
used in the dispersion of plumes whose oy is greater than 1000 m. At the time
step when ay becomes greater than 1000 m, a matching is made between Turner's
curves and Gifford's Lagrangian theory to determine the Ky value in Gifford's
theory. (In Gifford's theory, the lateral dispersion is a function of time
and Ky.) In this way, a continuous plume width is maintained during
transition between the two theories of lateral spreading. The value of Ky
determined by this process was found to be on the lower end of the range of
values found in the literature for dispersion in the mesoscale.31 Moreover, a
sensitivity analysis carried out by TVA indicated that model predictions were
very sensitive to the choice of this parameter.53 A choice of the higher
values (based solely on measured values in the literature and not on a
matching process) should lead to greater lateral spreading and reduced peak
concentrations. This hypothesis as to the cause of model/data discrepancies
is likely to be correct, but remains to be confirmed by means of actual
computer simulations.
The following discussion is offered to explain the physical reason for
the underprediction of lateral spreading. The Turner curves were developed
from small-scale diffusion experiments, with the dispersion relationships
generated from them often extrapolated to larger distances. On the other
5-55
-------
hand, the scale of the field of turbulence is quite different in the mesoscale
with eddies involved in the mixing being kilometers in size, considering that
plume widths are in that size range. It is conceivable then that the matching
procedure does not provide a Kv that represents the true scales of turbulence
that are occurring at these large distances.
It is expected that the conversion from Turner's curves to the Gifford
Lagrangian theory has occurred early in the dispersion due to the presence of
strongly convective conditions on the hot summer days of the releases. It is
possible that the ay value became greater than 1000 m and the matching took
place after the first hour of transport. Further numerical experiments with
Ky including a rerun of the two Oklahoma cases, printing out the location and
details of the matching (stability class, downwind distance, cy, time, etc.),
would be very instructive.
If Ky is the problem, one possible improvement might be to continue the
matching procedure as before, while adding a scale-dependency to Ky. Such a
dependency would allow Ky to increase as the plume and the scales of
turbulence which affect it grow.
It is interesting that recent validation tests of ARRPA carried out by
TVA54 suggest a tendency for underprediction of plume width and an over-
estimate of plume vertical extent. This validation work covered downwind
distances only from about 10 to 70 km. An examination of plume cross-
sectional areas in that study revealed no significant problem, undoubtedly
because the underprediction of width and the overprediction of plume vertical
extent offset each other. However, for larger distances downwind, the under-
prediction of lateral spreading continues (as shown here with the Oklahoma
runs), whereas the overprediction of vertical spreading does not since it is
limited by the mixing height. As a result, that compensation does not occur
beyond these larger distances and the bias in the lateral spreading becomes
more evident in the predictions. It is expected that this underprediction of
lateral spreading is the main cause of the overprediction of concentrations at
the ground.
Finally, the ARRPA model was one of four evaluated recently with the
CAPTEX data sets.54 in these seven data sets, the models were tested for
their predictive capability out to distances of about 1000 km. The ARRPA
model revealed a very significant underprediction of lateral spreading at the
ground and a significant overprediction of plume concentrations at ground
5-56
-------
level. Consequently, the systematic behaviors observed in the current study
are consistent with that longer distance testing.
5.5.7. RADM
One of the most obvious features characterizing the predictions of RADM
for the Oklahoma data base and, except for MSPUFF, is its tendency to over-
predict plume ground-level concentrations. This characteristic was observed
with both the Oklahoma and SRP data bases. RADM predicts the highest maximum
concentration values (over the whole grid) of any model within the first
100 km of the source (see Tables 5-6). In fact, even the MSPUFF peak
predictions eventually fall below those of RADM during the first Oklahoma
case. Its maximum concencentration predictions are the sixth highest of the
seven models (on the average) for the Oklahoma data sets, exceeded only by
MESOPLUME. Tables 5-3 to 5-5 also confirm this behavior in that plume widths
on passage through the sampler arcs are among the smallest. When the RADM
predicted plume passes the 600 km arc after the first release, the width is a
factor of 4 smaller than the observed, and a factor of 2 below the next-
smallest predicted width (by MSPUFF). Table 5-7 confirms this tendency to
overestimate ground concentrations in that RADM predicts the second-highest
maximum concentration values over the entire grid at SRP. (It was explained
in Section 5.5.4. above that MESOPUFF II predicts very large near-source
concentrations while the source is emitting, which causes it to predict the
largest maximum concentrations much of the time at SRP.) Two reasons can be
identified that are responsible for the predictive behavior of RADM described
above.
(a) The eddy diffusivities used in the RADM plume dispersion calculations
may be too low. Previous experience with the model at Dames & Moore,
Inc, has only involved the modeling of plume dispersion in regions
not exceeding 100 km square with a plume whose maximum lateral extent
seldom exceeded 10 km. It is within this region where the RADM
formulation of lateral and vertical spreading using the lateral and
vertical diffusivities, % and Kv, agrees roughly with the Turner
curves. However, it is known that at some distance less than 100 km
5-57
-------
from the source, one has to begin using growth rates of plume
dimensions that take account of the effect of larger-scale eddies.
One such formulation used by the other models is that of Heffter. If
one continues to use the Turner curves to mesoscale distances, too
little dispersion will result at large distances, as observed in the
predictions of RADM. The experience with the MESOPUFF models in this
study suggests that the transition region is more likely to be less
than 10 km rather than 100 km or more. In addition, an increase in
the diffusivities with scale may be needed in predicting plume
dispersion for many hundreds of kilometers.
(b) For distances exceeding 100 km, it may be necessary to select the
parcel mass so that more parcels are used to represent the plume. In
these computer runs, the parcel mass was adjusted so that for each
case at least 5000 parcels were present at some time during the run.
The 5000 parcel figure was adopted in accord with the RADM user's
manual guidelines; however, for mesoscale distances a larger number
may be needed. Using too few parcels can lead to artificially high
concentrations and an underestimation of the lateral plume extent.
This behavior is due to the small number of widely dispersed parcels,
which reduces the likelihood that a sampler will be impacted by one.
(Of course, with more parcels, each one has a smaller mass.)
Concerning the positioning of the predicted plume, the RADM plume follows
a trajectory determined by the wind field from MESOPAC II. Its trajectory was
seen to agree with plume trajectories predicted by the RTM-II model (at
Oklahoma) and the MESOPUFF II model. The winds predicted by MSPACK (for
MSPUFF) are regularly less accurate than those of either the MESOPAC model
(for MESOPUFF and MESOPLUME) or the MESOPAC-II model (for MESOPUFF II, RADM
and RTM-II at Oklahoma). However, both MESOPAC models produce winds that are
not as accurate in direction or speed as those from the BLM model (for ARRPA).
In the RADM model runs, parcels were released every 10 minutes, which is
well within the guidelines offered in the user's manual. For some runs,
marked differences were observed between the RADM model predictions and those
from MESOPUFF II. This is interesting because both employed the
meteorological preprocessor MESOPAC II. For case 13-A at SRP (July 15, 1977
5-58
-------
from 0900 to 1900 GMT), RADM predicts a compact plume to the northwest of the
source, extending about 70 km downwind, whereas MESOPUFF II predicts a much
longer plume extending to the west-southwest. However, during the next
averaging period for case 13-B, the ground level plume regions for the two
models roughly coincide, extending to the southwest. Such differences
emphasize the combined effect of differences in plume diffusion formulations
and methods of computing concentration averages. The effects of these
different factors are difficult to separate. Case 13-A illustrates another
interesting point. During that averaging period, only two samplers, one 20 km
from the source and one 50 km from the source, recorded plume concentrations
significantly above background. Both models predict zero concentration at the
farther sampler, and both predict a concentration within a factor of 2.0 of
the observed concentration at the nearer one. However, the plume patterns are
quite different. Even the statistical measures paired in space and time will
be affected equally by these predictions, because the SRP sampler network does
not provide enough spatial resolution to exhibit the difference in plume
patterns. This case clearly shows the benefit' of augmenting the AMS
statistics with other methods for comparing predicted plume patterns.
The method of computing plume concentrations from parcel location and
"size" may be responsible for some of these large prediction differences
between RADM and MESOPUFF II. The horizontal and vertical a's are used in
RADM to define a rectangular "box" around each parcel, and the fractional
overlap of each parcel's "box" with a user-defined receptor volume (also
rectangular) is used to assign part of that parcel's mass as a contribution to
the receptor concentration. (The parcel's volume is constrained in all three
dimensions to be less than or equal to that of the receptor.) This
interesting method differs markedly from the use of a Gaussian distribution of
plume material to compute concentrations from puffs, and may partially explain
the wide predictive differences between RADM and MESOPUFF II. With the
present determination of its diffusion coefficients, RADM exhibits less
predictive accuracy than does MESOPUFF II.
Finally, from Table 5-8 for the 65 Savannah River cases, one can draw
several conclusions concerning the performance of RADM. The assignment of
plume trajectories to direction categories yields 28 correct trajectories,
which places RADM among the four most accurate models. When the direction of
the trajectory is found to be offset from the observed trajectory (31 times),
5-59
-------
a clockwise rotation is twice as frequent as a counterclockwise rotation. As
expected, these figures are almost identical to those for MESOPUFF II.
However, because of the interplay of the model's Lagrangian random walk
formulation, the stability-dependent diffusion coefficients and the concen-
tration averaging method, the model appears to predict much narrower plumes
with less lateral spread than does MESOPUFF II. The RADM model never predicts
a plume wider than has been observed, and predicts a too narrow plume in 22
cases. On the other hand, MESOPUFF II predicts a plume that is too narrow
only 10 times and a plume that is too wide 8 times, suggesting an
approximately correct spreading rate. The reasons for the prediction of too-
narrow plumes and too-high ground concencentrations by the RADM model for the
SRP data base are the same as discussed in (a) and (b) above.
5.5.8. RTM-II
Of the six models that were applied .to both data bases, RTM-II showed
major predictive differences between the data bases. These differences arise
from the differences in model application between the SRP and the Oklahoma
data bases. The two differences in model application were the following:
(a) The meteorological data were provided from MESOPAC for the SRP
predictions and from MESOPAC II for the Oklahoma predictions. Thus,
trajectories are predicted for the plumes at SRP that match those for
MESOPUFF and MESOPLUME, whereas at Oklahoma, the predicted
trajectories match those for MESOPUFF II and RADM.
(b) The model developers selected values for the coefficient TJ, defined
as the coefficient that multiplies the magnitude of the velocity
field deformation tensor in the formulation of the gridded horizontal
diffusion coefficient, KJJ. Separate 17 values were selected by the
model developers for the Oklahoma and SRP data base runs. They also
selected different upper and lower bounds for this diffusion
coefficient for each experiment. Since scale considerations enter
into the choice of %, differences in these values between
experiments are, a priori, reasonable. In RTM-II the 17 coefficient
5-60
-------
is a crucial user input that requires considerable engineering
judgment.
5.5.8.1. RTM-II Features Common to Both the Oklahoma and the SRP Data Bases
The RTM-II model predictions are subject to large day-night variations in
surface concentrations as the mixing layer follows its diurnal cycle. Since
the mass of pollutant present is spread out over the mixing layer, mass
conservation implies that significant decreases in the layer depth can cause
large increases in surface concentrations, rather than the normal decrease
with time due to diffusion. Also, significant increases in mixing height can
cause decreases in ground concentrations that are more rapid than diffusion
alone would cause. For the SRP cases studied, the averaging periods (10 hour)
had been selected to represent primarily either daytime or nightime
conditions, and the mixing heights did not show large diurnal variations
within these periods. The long averaging times also tend to mask time-
depehdent effects in the observed concentrations. At Oklahoma, however, the
first release was at 1900 GMT and sampling took place until 0300 GMT on the
second day following, thus including both daytime and nighttime conditions.
Table 5-9 shows the changes in the predicted MESOPAC II mixing height over
this period for the location at the center of the predicted plume. During the
period from 1100 to 1700 GMT, the predicted plume concentrations drop sharply,
which seems to be explained by the rapid increase in mixing height and the
spreading out of the available mass over a larger volume. However, the drop
in concentration exceeds that which can be explained by the degree of increase
in predicted mixing height. Upon examination, one other factor is probably
responsible. In RTM-II, when the wind field shows divergence, mass that is
being carried in the upper layer is transferred to the lower layer. The
concept is that the divergent mass flow must come from air aloft. On the
other hand, when the wind field is convergent, mass is transferred from the
mixing layer to the upper layer. MESOPAC II does not compute a divergence-
free wind field. When the wind field is convergent or divergent, the plume
predicted by RTM-II loses mass to the upper layer or gains mass from it. In
fact, during the period when the RTM-II concentrations are dropping so
quickly, a front is approaching from the northeast, which normally produces
5-61
-------
Table 5-9. Mixing height at plume center (rounded to 100 m) predicted by
MESOPAC II for the first release at Oklahoma.
Day Time Mixing Height Grid Location
July 08
July 08
July 08
July 09
July 09
July 09
July 09
1500 LST (2100 GMT)
1900 LST (0100 GMT)
2300 LST (0500 GMT)
0300 LST (0900 GMT)
0700 LST (1300 GMT)
1100 LST (1700 GMT)
1500 LST (2100 GMT)
2000 m
1100 m
600 m
600 m
400 m
1200 m
1300 m
(15,11)
(15,11)
(15,11)
(18,20)
(19,22)
(26,31)
(41,34)
convergent winds in the mixing layer. This behavior of the RTM-II model may
be physically correct under the particular meteorology that was present after
the time of plume passage (as determined from the measurements) through the
600 km arc. However, since there were no additional samplers active beyond
600 km, this conjecture cannot be verified by the data. A further
investigation of how much of the physical convergence in the wind field caused
by this approaching front is actually being predicted by MESOPAC II is beyond
the scope of this evaluation.
5.5.8.2. RTM-II Features Specific to the Oklahoma Data Base
The plume trajectory predicted by RTM-II during the first release
(July 8, 1980 at 1900 GMT to July 10, 1980 at 0200 GMT) stays eastward of the
observed plume location and does not transport rapidly enough to arrive at the
600 km arc when the tracer was first observed on July 9, 1980 at 0200 GMT.
However, there is a predicted wind speed gradient that is quite strong near
this arc (from MESOPAC II); prior errors in wind speed prediction become
5-62
-------
amplified in this region. The predicted plume actually sweeps far to the east
and exits the region moving east-southeast, whereas the data suggest a plume
that should exit moving northward approximately in the center of the grid.
The plume also arrives too late by more than an hour at the 100 km arc.
During the second case, the predicted plume arrives at the 100 km arc at the
correct time, but it passes 20-40 km to the west of samplers recording
concentrations above ambient. It is also slow to leave this arc, producing
sizable concentrations at samplers up to 2 hours after the data shows none.
The apparent slow movement may be caused by too-rapid spreading, rather than
errors in the wind speeds predicted by MESOPAC II. The directional problems
do stem from methodologies in MESOPAC II.
For both cases, RTM-II predictions show extremely rapid spreading as
illustrated in the isopleth graphs. Consistent with its rapid spreading, its
peak concentrations, (listed in Table 5-6) are below those of any other model
on an average over the 21 time periods. (The plumes predicted by the MSPUFF
and ARRPA models yield zero maximum concentrations during the last two periods
on July 9 because they exit the grid too soon.) This behavior of RTM-II
described above is undoubtedly due to values for 17 (thus the horizontal
diffusivity, Kg) and the minimum horizontal diffusivity, (KH)min, that are too
large. (No value of % computed from the wind field deformation was found to
be as large as the specified maximum value in these Oklahoma runs, but the
minimum value was often assigned to grid points.) Refinements in the procedure
for selecting 77 and (%)m^n as a function of scale seem desirable to improve
the predictive accuracy of RTM-II. From Tables 5-3 to 5-5, one can see that
the predicted widths on plume passage through the sampler arcs equal or exceed
the observed widths, a feature found only in the predictions of RTM-II, and
only for the Oklahoma experiment. The arrival of the predicted plume at the
600 km arc is late by about 8 hours, a feature common to the MESOPAC II-based-
models.
5.5.8.3. RTM-II Features Specific to the Savannah River Plant Data Base
Plume trajectories in this case follow those for MESOPUFF and MESOPLUME.
In Table 5-8, categories (l)-(4) characterize the plume trajectory and
categories (5)-(6) describe rate of plume spreading relative to the data (with
65 minus the sum of categories (5)-(8) giving the number of predicted plumes
5-63
-------
that had approximately correct widths). For nearly half of the cases (31),
the direction of the predicted plume is appropriate. There are slightly more
plumes rotated clockwise than rotated counterclockwise. One would expect
layer-averaged winds to show more clockwise rotation if, indeed, the surface
wind direction were the correct one to use in predicting ground concentrations
at these distances, as appears from the performance of the MTDDIS model. The
RTM-II model also shows some complete misses, such as 6 plumes that were
directed nearly opposite to the observed one, and the 4 cases of all zero
predictions when some sampler recorded the presence of the plume. Plume
spreading comes from only two factors in this model. First, if the wind were
to change direction significantly during the averaging period, the plume would
appear wider than diffusion alone would predict, because the plume would sweep
across more grid points as its transport direction changes. Since the
MESOPUFF model should also show this type of spreading, we can largely rule it
out as a factor. The second factor is the magnitude of the horizontal
diffusion coefficient. The tendency of plumes to be too narrow indicates an rj
coefficient that is somewhat too small. In this model, stability class values
at the grid points are not used to influence the horizontal diffusion
coefficient. It may be physically unrealistic to rely entirely on the
deformation of the wind field to obtain the magnitude of the horizontal
diffusion coefficient. The fact that the spreading of the plume for the
Savannah River Plant data sets is too small and at Oklahoma it is much too
rapid, underscores the need to develop a more accurate way of choosing these
parameters for RTM-II once the grid size and scale of the calculation are
defined. Despite the fact that the RTM-II model predicted plumes that spread
too little with respect to the observed plumes, they spread more rapidly than
those of any other models. As a result, the maximum predicted concentrations
over the entire grid, as listed in Table 5-7, are less than those of any of
the other models on an average over the 65 sampling periods.
5-64
-------
SECTION 6
SUMMARY AND CONCLUSIONS
6.1. OVERVIEW
Eight short-term long-range transport models have been evaluated using
two data bases representing different transport distances and averaging times.
The Oklahoma data base contained two data sets for which the source was
located at the ground and the emissions were 3-hour continuous releases (one
for each experiment) of perfluorocarbon tracer. The concentration data
encompassed detailed spatial measurements of the tracer at the 100 km and
600 km arcs with ground-level measurements made over a total of 21 time-
averaging periods. The measured data were taken as 45-minute averages at the
100 km arc and 3-hour time averages at the 600 km arc. The very fine spatial
resolution at 100 km and 600 km arcs provided a unique opportunity to evaluate
the accuracy of the models in predicting the time of travel of the observed
and predicted plume to these arcs along with the location, spreading, and
pointwise concentration values of the predicted and observed plumes as they
crossed those arcs. An interesting feature of this data base is the presence
of a strong nocturnal jet that transported the plume very rapidly downwind.
The Savannah River Plant data base contained 15 data sets representing
spatial measurements over distances of 28 to 144 km. There were 13 fixed
samplers placed about a 62-m stack emitting a tracer gas, krypton-85.
Measurements were available over several 10-hour averaging periods for each
data set. These fifteen episodes complement the Oklahoma data sets in that
they involve records of 2-5 days, but at a fewer number of fixed samplers. In
spite of the differences in character of the data bases used in the
evaluation, the performance of the models was generally consistent between the
data bases.
Several methods were used to evaluate the models:
(a) statistical comparisons using the American Meteorological Society
(AMS) statistics. These statistics employ pointwise pairs of
predicted and observed concentrations,
6-1
-------
(b) graphical comparisons of predicted/observed pairs including scatter
plots, frequency histograms of residuals, cumulative frequency
plots, etc. for each model for each data base,
(c) quantitative comparisons of the predicted and observed patterns for
the Oklahoma data sets in order to evaluate their similarity.
Compared were the peak concentrations, the angular offset of the
pattern centerlines, and the time lag/lead of the predicted patterns
from the observed, and
(d) qualitative comparisons of the ground-level concentration patterns
predicted by the models and observed data for each averaging period.
Concentration isopleths were presented for Oklahoma and sketches of
the ground pattern were presented for SRP.
The results of these comparisons were largely consistent and depict the
model performance described below.
6.2. GENERAL PERFORMANCE FEATURES OF THE MODELS
The key feature of the ground-level plume patterns predicted by the
models is that they are frequently offset from the data by as much as 20-45
degrees. The treatment of the meteorological data within the meteorological
preprocessor employed by a long-range transport model is the primary factor in
the offset of the predicted and observed patterns. Often, the uncertainty in
plume spreading is of secondary importance as compared to the uncertainties in
characterizing the wind field. Due to the relatively short mesoscale
distances involved in this model evaluation study (most of the data used in
the evaluation were taken at distances less than 150 km), the accuracy of the
wind field prediction (especially the initial direction of the plume) was
crucial to the accuracy of the plume model predictions.
The BLM model (preprocessor to ARRPA) and the MTDDIS meteorological
preprocessor appear to provide the most accurate trajectories of all eight
models. However, the BLM model testing was carried out only for the two
Oklahoma cases; more testing with that model is required before definitive
6-2
-------
conclusions can be reached. The MTDDIS preprocessor employs wind directions
estimated at the effective height of release based on surface data. The use of
low elevations for wind direction determination (rather than the 850 mb level,
or the integrated average over 1500 m, or an average to the mixing height)
appear to have provided good estimates for predicted trajectories at ground
level in the distance range tested here, i.e. less than 150 km. However, the
MTDDIS model was not tested with the Oklahoma data base since the model was
considered by its developers to be inapplicable to that site due to the
presence of a ground-level source. Definitive conclusions about the
predictive accuracy of the MTDDIS-predicted trajectories also should await
testing with data having greater spatial and temporal resolution and longer
distances of transport than available from the SRP data base.
As noted above, the second most important cause of model/data
discrepancies is the treatment of plume spreading in the models. This
spreading may be viewed as superimposed about the trajectory given by the
predicted wind field. Except for MTDDIS (at SRP) and RTM-II (at Oklahoma),
the models underpredict horizontal spreading at both sites. In addition, the
evaluation of predicted and observed ground-level concentrations revealed that
each of the models (except RTM-II for SRP) overpredicts ground-level plume
concentrations. (An approximate conservation of mass principle at ground
level appears to be operational here; the vertical dimension is probably not
as significant considering that after a short distance downwind, full mixing
throughout the mixed layer has taken place.) The main cause of the
underprediction of lateral spreading appears to be the often-used assumption
that lateral spreading is given by the Turner curves for distances downwind
out to 50-100 km. The models affected by that assumption are MESOPUFF,
MESOPLUME, MSPUFF, ARRPA, and RADM. The more correct distance is about 10 km.
6.3. QUANTITATIVE MEASURES OF MODEL PERFORMANCE
The AMS statistics were applied to the predicted/observed pairs for the
models at both sites. Statistical results are generally consistent between
the two sites. The spatial and temporal offsets of the predicted and observed
plumes lead to predicted concentrations that correlate poorly with concen-
trations observed at the same time and place. On average, all models
6-3
-------
overpredict (pairing in space and time) at both sites except for RTM-II which
underpredicted only at Oklahoma. The largest overprediction is by RADM at
both sites. Values for standard deviation of residuals, root mean square
error, and absolute average residual are larger than average observed concen-
trations for all eight models. The largest values for all three measures occur
with RADM. Predicted concentrations also are found to correlate poorly with
concentrations observed at the same time and place. Pearson's correlation
ranges among the models from -0.14 (MESOPUFF II) to 0.68 (MESOPUFF) for
Oklahoma and -0.02 (MSPUFF) to 0.35 (MTDDIS) for SRP. Variance ratios are
consistently less than 1.0 for all models except RTM-II.
On the other hand, statistical comparisons of the peak values predicted
by the models are significantly better. Statistics for highest concentration
by event (unpaired by location) reveal an overprediction by all models except
RTM-II at Oklahoma. Statistics for highest concentration at each station
(unpaired in time) also reveal overprediction by all models except MESOPUFF
and RTM-II at Oklahoma. The best statistics are achieved through 'impairing in
both space and time of the highest 25 predictions and highest 25 observations.
Based on the results presented in Tables 4-9 and 4-10, the ratios of the
averages of the highest 25 predictions to the highest 25 observations are:
Model Oklahoma SRP
MESOPUFF
MESOPLUME
MSPUFF
MESOPUFF II
MTDDIS
ARRPA
RADM
RTM-II
1.4
1.8
3.5
1.2
M/A
1.8
3.3
0.7
1.4
1.4
1.3
1.1
1.4
M/A
4.7
1.0
Clearly, then, as more impairing is accomplished, the statistical results
improve significantly. Notice that in the above table, most models tend to
overpredict peak concentrations. However, the advantage of an overprediction
of the peak values in a regulatory setting must be weighed against (a) the
6-4
-------
spatial offsets that occur between the predicted and observed patterns, and
(b) the underprediction of lateral spreading that has been shown to occur for
these models (exceptions are MTDDIS for SRP and RTM-II for Oklahoma).
A pattern comparison method was used at Oklahoma to quantify the
differences in ground-level configurations of the observed and predicted
plumes. The large numbers of samplers at these arcs made quantification
possible to a limited degree. At the 600 km arc for example, it was found
that:
(a) the predicted plume transport time lags the observed time of
transport for all models with the slowest plume (RADM) arriving at
least 9 hours late,
(b) all models underpredict plume widths as the peak concentration passed
over the arc. The range is 50 km width (RADM) to 180 km width
(RTM-II) as compared to a measured width of 200 km,
(c) an angular offset of the location of the predicted and observed peaks
on the arc is usually found, and that offset varied from 11°
(MESOPUFF II) to 24° (MESOPUFF, MESOPLUME, RTM-II, and RADM). For
this first Oklahoma experiment, the observed peak crossed at 10°
clockwise from north.
Mixed results occurred for the 100 km arc. Some predicted plumes (from
MESOPUFF, MESOPLUME, and MSPUFF) arrive earlier than the observed plume at
that 100 km distance. The remaining models (MESOPUFF II, ARRPA, RADM, RTM-II)
show actual plume arrival later than predicted plume passaage.
6.4. SPECIFIC PERFORMANCE FEATURES OF INDIVIDUAL MODELS
6.4.1. MESOPUFF
The MESOPUFF model frequently reveals errors of 20 to 40 degrees in the
direction of the predicted plume (predictions usually clockwise relative to
observed) along with an underprediction in plume spread. Model errors are
6-5
-------
most likely due to (a) the single-layer nature of the wind field analysis
based on twice-daily rawinsonde data used in MESOPAC (inadequately treating
the diurnal cycles of mixing depth and vertical shear in the horizontal wind,
as well as not treating meandering), and (b) an underprediction of small-scale
processes; i.e., too small oy. Concentrations are generally overpredicted
as compared with the other models and with the data.
6.4.2. HESOPLUME
The MESOPLUME model employs the same wind field as does MESOPUFF; as a
result, the same relative ground positioning and times of arrival are found
with both models. However, the models show significantly different ground
concentration predictions. In general, the MESOPLUME concentration
predictions are up to a factor of two larger than the MESOPUFF predictions,
with a smaller lateral spreading as compared to MESOPUFF. Compared with the
data, MESOPLUME generally overpredicts concentrations and underpredicts widths
at ground level.
6.4.3. MSPUFF
This model predicts plume trajectories that are usually oriented
clockwise of the trajectories predicted by MESOPAC-based-models. As noted
above, the trajectories predicted by MESOPAC are themselves rotated clockwise
with respect to the data. The explanation for MSPUFF relates to the use of
850 mb winds in the meteorological preprocessor MSPACK (as compared with the
averaging of winds over a 1500 m depth, as done in MESOPAC). The concen-
trations at ground level from MSPACK are usually higher than for MESOPUFF and
higher than the observed concentrations as well. Predicted plume widths are
generally more narrow than indicated from the data. Possible interpretations
of the causes of model/data discrepancies relate to the treatment of the
mixing height (treatment of turbulent dissipation of convection for the
convective mixing height in the Benkley-Schulman method and/or the use of
surface winds in modeling the mechanical mixing height).
6-6
-------
6.4.4. MESOPUFF II
The MESOPUFF II model also tends to overpredict plume concentrations and
has a tendency to underestimate plume widths. At the SRP site, the MESOPUFF
II model predicted about an equal number of plumes that were too wide or too
narrow, whereas at the Oklahoma site an underprediction of spreading occurred.
The model tends to overpredict concentrations at both sites. The use of 10 km
as the transition distance between the use of the Turner curves and Heffter
formula for lateral dispersion definitely improved concentration predictions
over MESOPUFF. The two-layer wind field model appeared to be an advantage at
SRP but a disdavantage at Oklahoma. The advantage of having two wind fields
that are essentially uncoupled may be in providing a better treatment of
nocturnal wind shear. At Oklahoma, a single-layer wind field model appeared
to perform better due to the fairly uniform flow there from the south or
southwest.
6.4.5. MTDDIS
This model was only applicable to the SRP data sets. The model provided
the best prediction of plume trajectories for this data base, yet revealed a
slight overprediction in horizontal spreading and an overprediction in
predicted concentrations. The use of a single-layer wind field with the wind
direction taken at the effective height of release apparently provides good
transport directions for the puffs at least for the short mesoscale distances
in the SRP data base. The wider range in wind directions that occurs at such
a low level (at the effective height of release from the 62-m stack as
compared, say, to directions at the 850 mb level) enhance lateral spreading.
The use of the Heffter formula for lateral dispersion from the source also
provides more lateral spreading under most conditions than in the other models
that use the Turner curves. These latter two effects are the likely causes of
the overprediction of horizontal spreading. The cause of the overprediction
in the concentration predictions may be the result of the mixing height model.
6-7
-------
6.4.6. ARRPA
This model was applied only to the Oklahoma data sets. The BLM wind
field model provides the most accurate trajectories of the competing wind
field models at Oklahoma. The physical model used in interpolating the
surface and upper air data appears to have been quite good for these two cases
in which strong pressure gradients dominated the flow. In spite of the good
trajectories, the lateral dispersion was underpredicted and the plume concen-
trations were overpredicted. The formulation for Ky in the Gifford Lagrangian
model for lateral dispersion (ay in the range between 1000 and 6000 m) appears
to have been underestimated.
6.4.7. RADM
This model uses the MESOPAC II meteorological preprocessor and thus has
the same features in terms of trajectories as does MESOPUFF II. The
horizontal spreading is underpredicted and the ground-level concentrations are
overpredicted. The use of KH and Kv as horizontal and vertical dispersion
coefficients (tied closely to the Turner curves) are likely causes of this
behavior. For mesoscale applications, a new choice of coefficients % and Kv
that leads to plume growth rates consistent with the presence of the larger-
scale eddies should be considered for future model improvement. It is also
possible that the method of computing ground concentrations from parcel
masses and locations is producing some of the occasional wide differences in
the predicted plume pattern relative to MESOPUFF II for the same cases.
6.4.8. RTM-II
The predictions of this model were quite different between the SRP and
Oklahoma sites due to (a) the different choice of t\ (the constant that
multiplies the deformation of the wind field to produce the horizontal
diffusivity) for the lateral dispersion coefficient for both sites, and (b)
the different choice of meteorological preprocessors. At SRP, the model
overpredicted plume concentration and underestimated plume spreading. At
Oklahoma, the model overpredicted plume spreading and underpredicted plume
concentrations. Considerable judgment is required in the choice of this
6-8
-------
lateral spreading coefficient. The grid spacing used in the plume modeling as
well as the regional scale of the modeling problem are both inputs into this
judgment process.
6.5. OPTIONS AND PARAMETERS
Finally, as alluded to earlier, most of these models (MESOPUFF,
MESOPLUME, HSPUFF, MESOPUFF II, RADM, and RTM-II) require special judgment in
their application. The choice of options is often crucial to the prediction
accuracy. For RTM-II, the choice of the constant rj determines how much
horizontal spreading will occur in the predictions. For MESOPUFF II, the
choice of the transition distance between the use of the Turner curves and the
Heffter formulas also determines the amount of horizontal spreading that will
occur in a particular application. In addition, the choice of wind field
option in many of the preprocessors is equally if not more important. As a
result, the accuracy of these models in future applications depends to a large
degree on intelligent user choices for such key parameters and options.
6-9
-------
-------
REFERENCES
1. United States Environmental Protection Agency. "Guideline on air quality
models", Report EPA-450/2-78-027. Office of Air Quality Planning and
Standards, Research Triangle Park, North Carolina, 1978.
2. Federal Register, "Guideline on air quality models," 45:20157-20158,
March 27,1980.
3. D.G. Fox, "Judging air quality model performance. A summary of the AMS
workshop on dispersion model performance. Woods Hole, Mass., 8-11
September 1980. "Bulletin American Meteorological Society,
62:599-609, May 1981.
4. G.J. Ferber, K. Telegadas, J. Heffter, C. Dickson, R. Dietz, and P.
Krey "Demonstration of a long-range atmospheric tracer system using
perfluorocarbons," Air Resources Laboratories, NOAA Technical Memorandum
ERL ARL-101, Silver Spring, Maryland. April 1981.
5. K. Telegadas, G. Ferber, R. Draxler, M. Pendergast, A. Boni, J.
Hughes, and J. Gray. "Measured weekly and twice-daily krypton-85
surface air concentrations within 150 km of the Savannah River Plant
(March 1975 through September 1977) -final report, " NOAA Technical
Memorandum ERL ARL-80. Air Resources Laboratories. Silver Spring,
Maryland. January 1980.
6. C.W. Benkley and A. Bass. "User's guide to MESOPUFF (Mesoscale Puff)
Model," Environmental Research and Technology, Inc., Concord,
Massachusetts. September 1979.
7. C.W. Benkley and A. Bass. "Development of mesoscale air quality
simulation models. Volume 6. User's guide to MESOPAC (Mesoscale
Meteorology Package)," EPA-600/7-79-XXX, Environmental Research and
Technology, Inc., Concord, Massachusetts. September 1979.
R-l
-------
8. C.W. Benkley and A. Bass, "Development of mesoscale air quality
simulation models. Volume 2. User's guide to MESOPLUME (mesoscale plume
segment) model," EPA-600/7-79-XXX, Environmental Research and Technology,
Inc., Concord, Massachusetts. September 1979.
9. M.R. Schock and S.F. Weber. "Modification of the MSPUFF model (a
version of MESOPUFF) treatment of plume dispersion,™ Division of
Environmental Waste Management and Research. North Dakota State
Department of Health. Bismarck, North Dakota. December 1982.
10. M.R. Schock and S.F. Weber. "MSPUFF mesoscale air quality computer
modeling system," Division of Environmental Waste Management and
Research. North Dakota State Department of Health. Bismarck, North
Dakota. July 1, 1984.
11. J.S. Scire, F.W. Lurmann, A. Bass, and S.R. Hanna. "Development of the
MESOPUFF II dispersion.model," Environmental Research and Technology,
Inc. Concord, Massachusetts. Prepared for Office of Research and
Development. U.S. Environmental Protection Agency. Research Triangle
Park, North Carolina. 1984.
12. J.S. Scire, F.W. Lurmann, A. Bass, and S.R. Hanna. "User's guide to
MESOPUFF II model and related processor programs," Environmental Research
and Technology, Inc., Concord, Massachusetts. Prepared for Office of
Research and Development. U.S. Environmental Protection Agency.
Research Triangle Park, North Carolina. 1984.
13. J.S. Scire and F.W. Lurmann. "Development of the MESOPUFF II
dispersion model." Paper presented at the AMS sixth Symposium on
Turbulence and Diffusion. Boston, Massachusetts. March 22-25, 1983.
14. I.T. Wang, T.L. Waldron, "User's guide for MTDDIS, mesoscale transport,
diffusion, and dispersion model for industrial sources," Report
EMSC6062.1UG(R2), Rockwell International, Inc., Newbury Park, California.
December 1980.
R-2
-------
15. S.F. Mueller and R.J. Valente. "Meteorological data preprocessing
manual for the Air Resources Regional Pollution Assessment Model (generic
version)," Office of Natural Resources, Air Quality Branch. Tennessee
Valley Authority. Muscle Shoals, Alabama. September 1983.
16. S.F. Mueller, R.J. Valente, T.L. Crawford, A.L. Sparks, and L.L.
Gautney, Jr. "Description of the Air Resources Regional Pollution
Assessment (ARRPA) Model." Tennessee Valley Authority. Air Quality
Branch. Muscle Shoals, Alabama. September 1983.
17. D.I. Austin, A.W. Bealer, W.R. Goodin. "Random-walk advection and
dispersion model (RADM)," Dames & Moore, Inc., Los Angeles, California.
December 1981.
18. M. Yocke, R. Morris, M. Liu. "Revised user's guide to the Regional
Transport Model," Publication No. 82120. Systems Applications, Inc.,
San Rafael, California. April 1982.
19. A.J. Policastro, M. Wastag. L. Coke, R.A. Carhart, and W.E. Dunn,
"Evaluation of eight short-term long-range transport models with field
data, Task I Report: Preparation of input for eight long-range transport
models", Prepared by Argonne National Laboratory, Argonne, Illinois and
the University of Illinois at Chicago and Champaign-Urbana, for U.S.
Environmental Protection Agency, Office of Air Quality Planning and
Standards, Research Triangle Park, NC, January 1985.
20. A.J. Policastro, M. Wastag, L. Coke, R.A. Carhart, and W.E. Dunn,
"Evaluation of eight short-term long-range transport models with field
data, ADDENDUM to Task I Report: Preparation of input for eight
long-range transport models", Prepared by Argonne National Laboratory,
Argonne, Illinois and the University of Illinois at Chicago and
Champaign-Urbana, for U.S. Environmental Protection Agency, Office of
Air Quality Planning and Standards, Research Triangle Park, NC, May 1985.
R-3
-------
21. A.J. Policastro, M. Wastag, L. Coke, R.A. Carhart, and W.E. Dunn,
"Evaluation of eight short-term long-range transport models with field
data, Task II Report: Preparation of test cases and proposed
statistical/graphical evaluation methods (revised)", Prepared by Argonne
National Laboratory, Argonne, Illinois and the University of Illinois at
Chicago and Champaign-Urbana, for U.S. Environmental Protection Agency,
Office of Air Quality Planning and Standards, Research Triangle Park, NC,
July 1985.
22. A.J. Policastro, M. Wastag. L. Coke, R.A. Carhart, and W.E. Dunn,
"Evaluation of eight short-term long-range transport models with field
data, ADDENDUM to Task II Report: Preparation of test cases and proposed
statistical/graphical evaluation methods (revised)", Prepared by Argonne
National Laboratory, Argonne, Illinois and the University of Illinois at
Chicago and Champaign-Urbana, for U.S. Environmental Protection Agency,
Office of Air Quality Planning and Standards, Research Triangle Park, NC,
February 1985.
23. A.J. Policastro, M. Wastag, J.D. Shannon, R.A. Carhart, and W.E.
Dunn, "Evaluation of two short-term long range transport models with
field data". Transactions, APCA Specialty Conference. The Meteorology
of Acid Deposition. Perry Samson, Ed. October 6-9, 1983. Hartford,
Connecticut.
24. C.W. Benkley and L.L. Schulman, "Estimating hourly mixing depths from
historical meteorological data", Journal of Applied Meteorology,
18:772-780, 1979.
25. D.B. Turner, "Workbook of atmospheric dispersion estimates", U.S.
Department of H.E.W., Public Health Service, Publ. 999-AP-26, 1970.
26. P.R. Maul, "Atmospheric transport of sulfur compound pollutants",
Central Electricity Generating Bureau MID/SSD/80/0026/R, Nottingham,
England, 1980.
R-4
-------
27. A. Venkatram, "Estimating the Monin-Obukhov length in the stable
boundary layer for dispersion calculations", Boundary Layer
Meteorology, 19:481-485, 1980.
28. A. Venkatram, "Estimation of turbulence velocity scales in the stable
and unstable boundary layer for dispersion applications", IN: Eleventh
NATO-CCSM International Technical Meeting on Air Pollution Modeling and
its Application, p. 54-56, 1980.
29. D.B. Turner, "A diffusion model for an urban area", J. Applied
Meteorology. 3:83-91, 1964.
30. D. Golder, "Relations among stability parameters in the surface layer",
Boundary Layer Meteorology. 3:46-58, 1972.
31. J.L. Heffter, "The variations of horizontal diffusion parameters with
time for travel periods of one hour of longer", Journal of Applied
Meteorology. 4:153-156, 1965.
32. R.R. Draxler, "Modeling the results of two recent mesoscale dispersion
experiments", Atmospheric Environment. 13:1523-1533, 1979.
33. J. Smagorinsky, "General circulation experiments with the primitive
equations: I. The basic experiment", Mon. Wea. Rev.. Vol 91:
99-164, 1963.
34. B.C. Scott, "Parameterization of sulfate removal by precipitation",
J. Applied Meteorology. 17:1375-1389, 1978.
35. J.M. Hales and S.L. Sutter, "Solubility of sulfur dioxide in water at
low concentrations," Atmospheric Environment. 7:997-1101, 1973.
36. T.E. Pierce and D.B. Turner, "User's guide for MPTER", Environmental
Sciences Research Laboratory, U.S. Environmental Protection Agency,
Research Triangle Park, NC, 1980.
R-5
-------
37. F.A. Gifford, "Horizontal diffusion in the atmosphere: a
Lagrangian-dynamical theory", Atmospheric Environment, 16:505-512,
1982.
38. G.J. McRae, W.R. Goodin and J.H. Seinfeld, "Development of a
second-generation mathematical model for urban pollution", Atmospheric
, Environment, 16:679-696, 1982.
39. C.C. Shir, "A preliminary numerical study of atmospheric turbulent flows
in the idealized planetary boundary layer", J. Atmos. Sci.,
30:1327-1339, 1973.
40. J. O'Brien, "On the vertical structure of the eddy exchange coefficient
in the planetary boundary layer", J. Atmos. Sci.. 27:1213-1215,
1970.
41. J.C. Wyngaard, "Modeling of the planetary boundary layer - extension to
the stable case", Bound. Laver Met.. 9:441-460, 1975.
42. Environmental Protection Agency/ "User's manual for single source
(CRSTER) model", EPA-450/2-77-013, Office of Air Quality Planning and
Standards,U.S. Environmental Protection Agency, Research Triangle Park,
North Carolina, 1977.
43. J.L. Heffter, "Air Resources Laboratories Atmospheric Transport and
Dispersion Model (ARL-ATAD)", NOAA Technical Memorandum ERL ARL-81, Air
Resources Laboratories, Silver Spring, Maryland, February 1980.
44. R.R. Draxler of Air Resources Laboratory, NOAA. Personal communication
with Dr. I.T. Wang of Rockwell International Corp., Newbury Park,
California, 1979.
45. P.R. Slawson, J.H. Coleman and J.W. Frey, "Natural draft cooling tower
plume behavior at Paradise steam plant: Part II", Division of
Environmental Planning, Tennessee Valley Authority, Publication No.
TVA/EP-78/01, February 1978.
R-6
-------
46. M.E. Smith, "Recommended guide for the prediction of the dispersion of
airbourne effluents", American Society of Mechanical Engineers, 1968.
47. J.P. Boris and D.L. Book, "Flux corrected transport — I. SHASTA, a
fluid transport algorithm that works", J. Comput. Phys., Vol II:
38-69, 1973.
48. R.J. Londergan, D.H. Minott, D.J. Wackter, T. Kincaid, and L.
Boutata, "Evaluation of rural air quality simulation models." TRC
Environmental Consultants, Inc. Report EPA-450/4-83-003. October 1982.
49. R.J. Londergan, D.H. Minott, D.J. Wackter, and R.R. Fizz, "Evaluation
of urban air quality simulation models." TRC Environmental Consultants,
Inc, EPA-450/4-83-020, July 1983.
50. D. Wackter and R. Londergan, "Evaluation of complex terrain air quality
simulation models", TRC Environmental Consultants, EPA-450/4-84-017, June
1984.
51. C.S. Hirtzel and J.E. Quon, "Estimating precision of autocorrelated air
quality measurements", IN: Summary of Proceedings of Environmetrics
81:200-201, 1981.
52. Mueller, S. of Tennessee Valley Authority. Personal Communication with
A.J. Policastro, Argonne National Laboratory, March 1986.
53. Mueller, S.F. and L.M. Reisinger, "Evaluation of the air resources
regional pollution (ARRPA) model", Office of Natural Resources Air
Quality Branch, Tennessee Valley Authority, Muscle Shoals, Alabama,
February 1986.
54. Shannon, J. of Argonne National Laboratory. Personal Communication with
A.J. Policastro, Argonne National Laboratory, November 1985.
R-7
-------
55. C.M. Shieh, M.L. Wesely, and B.B. Hicks, "A guide for estimating dry
deposition velocities of sulfur over the eastern United States and
surrounding regions", Argonne National Laboratory, ANL/RER-79-2, Argonne,
Illinois, April 1972.
R-8
-------
TECHNICAL REPORT DATA
(Please read Instructions on the reverse before completing)
1. REPORT NO.
EpA-450/4-86-Q16a,
2.
3. RECIPIENT'S ACCESSION NO.
s. REPORT DATEdate of prepare, t Ion
Qr.toSer 19R6
4. TITLE AND SUBTITLE
Evaluation of Short-Term Long-Range Transport
Models—Volume I Analysts Procedures and Results
6. PERFORMING ORGANIZATION CODE
7. AUTHOR(S)
A. 0. Policastro, M. Wastag, L. Coke,
R. A. Carhart, and W. E. Dunn
8. PERFORMING ORGANIZATION REPORT NO.
9. PERFORMING ORGANIZATION NAME AND ADDRESS
U. of 111,
Argonne National Laboratory ":,"'* ":„ "; "' *'{ 1VJ
b*r,nnno Tii-inn-ie cn/iia Chicago, 111 Uroana, 111.
Argonne, Illinois 60439 cncon ciom
10. PROGRAM ELEMENT NO.
B24A2F
60680
61801
11. CONTRACT/GRANT NO.
DW89930807
12. SPONSORING AGENCY NAME AND ADDRESS
U.S. Environmental Protection Agency
Office of Air Quality Planning and Standards
Research Triangle Park, NC 27711
13. TYPE OF REPORT AND PERIOD COVERED
Final r
14. SPONSORING AGENCY CODE
15. SUPPLEMENTARY NOTES
16. ABSTRACT
Eight short-term long-range transport models (MESOPUFF, MESOPLUME, MSPUFF,
1ESOPUFF II, MTDDIS, ARRPA, RADM, and RTM-II) have been evaluated with field data from
wo data bases involving tracer releases. The primary quantitative means of evaluating
lodel performance was the use of the American Meteorological Society statistics. Sup-
lementary measures included the use of isopleth plots of ground-level concentrations,
catter plots, cumulative frequency distributions and frequency histograms of residuals
General features of the model performance included: (a) spatial offset of predicted
md observed patterns, (b) a time difference between the arrival of the predicted and
bserved plumes at a particular receptor, and (c) an angular offset of as much as 20-45
egrees between predicted and observed plumes. The models also tended to underpredict
lorizontal spreading at ground level, along with overprediction of plume concentrations
s a result, predicted concentrations correlated poorly with concentrations observed at
he same time and place. However, statistical comparisons of the peak values predicted
the models were significantly better. For example, the highest 25 averaged predic-
ions and highest 25 averaged observations (unpaired in location and time) were within
factor or two of each other for six of the eight models tested (MESOPUFF, MESOPLUME,
1ESOPUFF II, MTDDIS, ARRPA, and RTM-II).
7.
KEY WORDS AND DOCUMENT ANALYSIS
DESCRIPTORS
b.lDENTIFIERS/OPEN ENDED TERMS C. COSATI Field/Group
Air Pollution
Long Range Transport Models
Meteorology
Model Evaluation
Statistics
18. DISTRIBUTION STATEMENT
Release unlimited
19. SECURITY CLASS (ThisReport)
Unclassified
21. NO. OF PAGES
221
20. SECURITY CLASS (Thispage)
Unclassified
22. PRICE
EPA Form 2220-1 (Rev. 4-77) PREVIOUS EDITION is OBSOLETE
------- |