& EPA
United States
Environmental Protection
Agency
Office of Air Quality
Planning and Standards
Research Triangle Park, NC 27711
EPA-454/R-95-005
March 1995
Aii
TESTING OF METEOROLOGICAL
AND DISPERSION MODELS FOR USE
IN REGIONAL AIR QUALITY
MODELING
-------
BPA-454/R-95-005
TESTING OF METEOROLOGICAL AND DISPERSION
MODELS FOR USE IN REGIONAL
AIR QUALITY MODELING
U.S. Environmental Protection Agency
Emissions, Monitoring and Analysis Division (MD-14)
Research Triangle Park, North Carolina 27711
National Park Service
Air Quality Division
Denver, Colorado 80225
USDA Forest Service
Office of Air Quality
Fort Collins, Colorado 80526
U.S. Fish and Wildlife Service
Air Quality Branch
Denver, Colorado 80225
March 1995
-------
DISCLAIMER
The information in this document has been reviewed in its entirety by the U.S.
Environmental Protection Agency (EPA), and approved for publication as an EPA document.
Mention of trade names, products, or services does not convey, and should not be interpreted
as conveying official EPA approval, endorsement, or recommendation.
EPA-454/R-95-005
-------
PREFACE
The Interagency Workgroup on Air Quality Modeling (IWAQM) was formed to
provide a focus for development of technically sound, regional air quality models for
regulatory assessments of pollutant source impacts on Federal Class I areas. Meetings were
held with personnel from interested Federal agencies, viz, the Environmental Protection
Agency, the U.S. Forest Service, the National Park Service, and the U.S. Fish and Wildlife
Service. The purpose of these meetings was to review respective regional modeling
programs, to develop an organizational framework, and to formulate reasonable objectives
and plans that could be presented to management for support and commitment. The
members prepared a memorandum of understanding (MOU) that incorporated the goals and
objectives of the workgroup and obtained signatures of management officials in each
participating agency. Although no States are signatories, their participation in IWAQM
functions is explicitly noted in the MOU.
This report documents results from a series of sensitivity analyses of several
meteorological models and puff dispersion models suitable for application to air pollution
dispersion involving long range transport. One of the primary objectives was to assess
whether the introduction of the hourly, gridded 80-km mesoscale meteorological data with
improved spatial and temporal resolution over typical observational networks would improve
the quality of the characterization of the transport and dispersion. Based on the results
obtained there was a noticeable improvement in the trajectory of the transport when
mesoscale meteorological data were employed in comparison to results obtained using
O diagnostic winds developed using available wind observations. This report is the fourth
s^ document published by the IWAQM in an effort to provide the sponsoring agencies and other
^ interested parties information on appropriate "off-the-shelf" methods for estimating long
<' range transport impacts of air pollutants on Federal Class I areas and impacts on regional
-v visibility. The IWAQM members anticipate issuing additional publications related to
\ progress toward meeting the IWAQM goals and objectives, the results of model evaluation
^ studies, proposed and final recommendations on modeling systems for regulatory
applications, and other topics related to specific objectives in the MOU.
-------
ACKNOWLEDGEMENTS
Several groups collaborated in the developing the results summarized in this report.
All of the simulations involving NUATMOS were conducted by the National Forest Service.
The EPA Office of Research and Development provided the mesoscale meteorological
modeling results and observational data used in the model evaluation comparisons with the
CAPTEX tracer data. EARTH TECH coordinated various activities, conducted various
sensitivity analyses and the CAPTEX model simulations, and developed the summary report
of the results.
The members of IWAQM acknowledge the special efforts of Joseph S. Scire,
Elizabeth M. Insley, David G. Strimaitis, Joseph C. Chang and ChungChin (Ed) Chang of
EARTH TECH, formerly Sigma Research, Inc. This report was prepared by The CADMUS
Group, Inc. under EPA Contract No. 68-DO-0095 with John S. Irwin as the Work
Assignment Manager.
IV
-------
Table of Contents
1. Introduction 1-1
2. Meteorological Model Sensitivity Tests 2-1
2.1 Description of CALMET 2-1
2.1.1 Grid System 2-3
2.12 CALMET Wind Field Module (DWM) 2-3
2.12.1 Step 1 Formulation 2-3
2.1.2.2 Step 2 Formulation 2-9
2.1.3 Micrometeorological Model 2-14
2.2 Description of NUATMOS 2-22
2.3 Grid and Episode Selection of Sensitivity Tests 2-25
2.4 Use of MM4-FDDA Wind Fields 2-32
2.4.1 Terrain Weighting Factor 2-32
2.4.2 Sensitivity to the Method of Use of MM4-FDDA Data 2-36
2.5 Statistical Measures 2-49
2.5.1 Trajectory Analysis 2-49
2.5.2 Point-by-Point Comparisons 2-51
2.6 Meteorological Sensitivity Tests 2-53
2.6.1 CALMET 2-53
2.6.2 NUATMOS 2-80
2.6.3 Statistical Results 2-80
3. Dispersion Models 3-1
3.1 Sensitivity Tests of Model Components 3-4
3.1.1 Dispersion Parameters 3-4
3.1.2 Dry Deposition 3-23
3.1.3 Chemical Transformation 3-29
4. Model Evaluation with CAPTEX Tracer Data 4-1
4.1 Model Application to CAPTEX 4-1
4.2 Model Comparison Measures 4-3
4.3 Results 4.4
5. References 5_1
Appendix. Forest Service Notes on NUATMOS Modeling Exercise A-l
I:\t314\a314rev\tocwph
-------
List of Illustrations
Figure Page
2-1. Flow diagram of the diagnostic wind model in CALMET 2-2
2-2. Schematic illustration of the CALMET horizontal grid system for
a 7 x 4 grid showing the grid origin (X,, Y0) and grid point location 2-4
2-3. Digitized terrain contours of the northern portion
of the Shenandoah National Park 2-27
2-4. The MM4-FDDA domain for the 54-km nested grid (outer box) and
the proposed IWAQM 54-km subgrid (inner box) 2-28
2-5. The MM4-FDDA domain for the IS-km nested grid (outer box) and
the proposed IWAQM 18-km subgrid (inner box) 2-29
2-6. The proposed IWAQM 5-km subgrid (inner box) 2-30
2-7. Terrain weighting factor, W0, for the surface level on the 5-km grid 2-37
2-8. Terrain weighting factor, W0, for the 50 m level on the 5-km grid 2-38
2-9. Terrain weighting factor, W0, for the 100 m level on the 5-km grid 2-39
2-10. Terrain weighting factor, W0, for the 200 m level on the 5-km grid 2-40
2-11. Terrain weighting factor, W0, for the surface level on the 18-km grid 2-41
2-12. Terrain weighting factor, W0, for the 50 m level on the 18-km grid 2-42
2-13. Terrain weighting factor, W0, for the 100 m level on the 18-km grid 2-43
2-14. Terrain weighting factor, W0, for the 200 m level on the 18-km grid 2-44
2-15. Terrain weighting factor, W0, for the surface level on the 54-km grid 2-45
2-16. Terrain weighting factor, W0, for the 50 m level on the 54-km grid 2-46
I:\»314\«314rev\tocwph
-------
List of Illustrations (Continued)
Figure Page
2-17. Terrain weighting factor, W,,, for the 100 m level on the 54-km grid ......... 2-47
2-18(a). Vector plot of CALMET winds. Level 4 (200 m), 54-km grid,
August 4, 1988, 'small* radius of influence parameters .................... 2-55
2-18(b). Vector plot of CALMET winds. Level 4 (200 m), 54-km grid,
August 4, 1988, "large" radius of influence parameters .................... 2-56
2-18(c). Vector plot of 80-km grid MM4-FDDA winds, interpolated
to 200 m level August 4, 1988 ...................................... 2-57
2-19(a). Vector plot of CALMET winds. Level 5 (400 m), 54-km grid,
August 6, 1988, "small" radius of influence parameters .................... 2-58
2-19(b). Vector plot of CALMET winds. Level 5 (400 m), 54-km grid,
August 6, 1988, "large" radius of influence parameters .................... 2-59
2-19(c). Vector plot of 80-km grid MM4-FDDA winds, interpolated
to 400 m level, August 6, 1988 ...................................... 2-60
2-20. Vector plot of CALMET winds. Level 5 (400 m), 18-km grid,
August 6, 1988, "small" radius of influence parameters .................... 2-62
2-2l(a). Vector plot of CALMET winds. Level 4 (200 m), 54-km grid,
August 4, 1988, 80 km MM4-FDDA data as initial guess field .............. 2-63
2-2 l(b). Vector plot of CALMET winds. Level 4 (200 m), 54-km grid,
August 4, 1988, 80-km MM4-FDDA data as Step 1 field .................. 2-64
2-21(c). Vector plot of CALMET winds. Level 4 (200 m), 54-km grid,
August 4, 1988, 80-km MM4-FDDA data as observations ................. 2-65
2-21(d). Vector plot of nested 54 km grid MM4-FDDA winds, interpolated
to the 200 m level, August 4, 1988 ................................... 2-66
l:\a314\i3J4rev\toc.wph
-------
List of Illustrations (Continued)
Figure Page
2-22(a). Vector plot of CALMET winds. Level 5 (400 m), 18 km grid,
August 4, 1988, "large" radius of influence parameters
and 80 km MM4-FDDA data as initial guess field 2-67
2-22(b). Vector plot of CALMET winds. Level 5 (400 m), 18 km grid,
August 4, 1988, "large" radius of influence parameters
and 80 km MM4-FDDA data as Step 1 field 2-68
2-22(c). Vector plot of CALMET winds. Level 5 (400 m), 18 km grid,
August 4, 1988, "large" radius of influence parameters
and 80 km MM4-FDDA data as observations 2-69
2-22(d). Vector plot of nested 18 km grid MM4-FDDA winds, interpolated
to the 400 m level, August 4, 1988 2-70
2-23(a). Trajectories at 10 m on 54 km grid produced by CALMET (large, using
observations only) and 54 km MM4 winds 2-72
2-23(b). Trajectories at 10 m on 54 km grid produced by CALMET (large, using
80 km MM4 winds as initial guess) and 54 km MM4 winds 2-73
2-23(c). Trajectories at 10 m on 54 km grid produced by CALMET (large, using
80 km MM4 winds as Step 1 field) and 54 km MM4 winds 2-74
2-23(d). Trajectories at 10 m on 54 km grid produced by CALMET (large, using
80 km MM4 winds as observations) and 54 km MM4 winds 2-75
2-24. Time series plot of CALMET mixing heights for grid point (10,12)
on the 54-km IWAQM grid (Dayton, OH) 2-76
2-25. Time series plot of CALMET friction velocity (u*) for grid point (10,12)
on the 54-km IWAQM grid (Dayton, OH) 2-77
2-26. Time series plot of CALMET convective velocity scale (w*) for grid
point (10,12) on the 54-km IWAQM grid (Dayton, OH) 2-78
I:\«314\k314KV\totwph j^
-------
List of Illustrations (Continued)
Figure Page
2-27. Time series plot of CALMET stability class (1 * A stability, 2 = B stability,
3 « C stability, 4 « D stability, 5 « E stability, 6 * F stability) for grid
point (10,12) on the 54-km IWAQM grid (Dayton, OH) 2-79
2-28(a). Vector plot of NUATMOS winds. Level 4 (200 m), 54 km grid,
August 4, 1988. Observations only 2-81
2-28(b). Vector plot of NUATMOS winds. Level 4 (200 m), 54 km grid,
August 4,1988. Uses 80 km MM4-FDDA data as input 2-82
2-28(c). Vector plot of NUATMOS winds. Level 4 (200 m), 54 km grid,
August 4, 1988. Terrain weighted 80 km MM4-FDDA data as input 2-83
2-29(a). Vector plot of NUATMOS winds. Level 5 (400 m), 54 km grid,
August 4, 1988. Observations only 2-84
2-29(b). Vector plot of NUATMOS winds. Level 5 (400 m), 54 km grid,
August 4, 1988. Uses 80 km MM4-FDDA data as input 2-85
2-29(c). Vector plot of CALMET winds. Level 5 (400 m), 54 km grid,
August 4, 1988. Terrain weighted 80 km MM4-FDDA data as input 2-86
3-1. oy vs. x for ISC and MESOPUFF, rural 3-7
3-2. ot vs. x for ISC and MESOPUFF, rural 3-8
3-3. oy vs. x for urban areas 3-9
3-4. at vs. x for urban areas 3-10
3-5(a). oy vs. x based on Draxler's formula, one frame for each PG class,
Class A equivalent 3-11
3-5(b). oy vs. x based on Draxler's formula, one frame for each PG class,
Class B equivalent 3-12
I:\a314\a314rev\locvpb X
-------
List of Illustrations (Continued)
Figure Page
3-5(c). oy vs. x based on Draxler's formula, one frame for each PC class,
Class C equivalent 3-13
3-5(d). oy vs. x based on Draxler's formula, one frame for each PG class,
Class D equivalent 3-14
3-5(e). oy vs. x based on Draxler's formula, one frame for each PG class,
Class E equivalent 3-15
3-5(f). oy vs. x based on Draxler's formula, one frame for each PG class,
Class F equivalent 3-16
3-6(a). oz vs. x based on Draxler's formula, one frame for each PG class,
Class A equivalent 3-17
3-6(b). or vs. x based on Draxler's formula, one frame for each PG class,
Class B equivalent - 3-18
3-6(c). oz vs. x based on Draxler's formula, one frame for each PG class,
Class C equivalent 3-19
3-6(d). oz vs. x based on Draxler's formula, one frame for each PG class,
Class D equivalent 3-20
3-6(e). oz vs. x based on Draxler's formula, one frame for each PG class,
Class E equivalent 3-21
3-6(f). oz vs. x based on Draxler's formula, one frame for each PG class,
Class F equivalent 3-22
3-7(a). Deposition velocities for SO2 as a function of hour, Bowen ratio,
and temperature 3.39
3-7(b). Deposition velocities for SO2 as a function of T*v vegetation
state, and albedo 3.31
I:\«3M\i314rev\loc.wph y^_
-------
List of Illustrations (Continued)
Figure Page
3-7(c). Deposition velocities for SO2 as a function of wind speed,
cloud cover, and LAI 3-32
3-8(a). Deposition velocities for SOJ as a function of hour, Bowen ratio,
and temperature 3-33
3-8(b). Deposition velocities for SOJ as a function of Z0, vegetation
state, and albedo 3-34
3-8(c). Deposition velocities for SOJ as a function of wind speed,
cloud cover, and LAI 3-35
3-9(a). Deposition velocities for NO, as a function of hour, Bowen ratio,
and temperature 3-36
3-9(b). Deposition velocities for NO, as a function of Z0, vegetation
state, and albedo 3-37
3-9(c). Deposition velocities for NO, as a function of wind speed,
cloud cover, and LAI 3-38
3-10(a). Deposition velocities for HNO3 as a function of hour, Bowen ratio,
and temperature 3-39
3-10(b). Deposition velocities for HNO3 as a function of Z0, vegetation
state, and albedo 3-40
3-10(c). Deposition velocities for HNO3 as a function of wind speed,
cloud cover, and LAI 3-41
3-1 l(a). Deposition velocities for NO3 as a function of hour, Bowen ratio,
and temperature 3-42
3-ll(b). Deposition velocities for NO3 as a function of Z0, vegetation
state, and albedo 3-43
l:\i3M\«3Mrev\toewph
-------
List of Illustrations (Continued)
Figure Page
3-1 l(c). Deposition velocities for NO3 as a function of wind speed,
cloud cover, and LAI ............................................. 3-44
3-12(a). SO2 loss rate for CALPUFF for NO, « 7 ppb .......................... 3-48
3-12(b). S02 loss rate for CALPUFF for NO, « 350 ppb ........................ 3-49
3-12(c). S02 loss rate for CALPUFF for NO, « 1400 ppb ....................... 3-50
3-13(a). SO4 formation rate for CALPUFF for NO, « 7 ppb ..................... 3-51
3-13(b). SO4 formation rate for CALPUFF for NO, *= 350 ppb ................... 3-52
3-13(c). SO4 formation rate for CALPUFF for NO, « 1400 ppb .................. 3-53
3-14(a). NO, loss rate for CALPUFF for NO, = 7 ppb ......................... 3-54
3-14(b). NO, loss rate for CALPUFF for NO, = 350 ppb ........................ 3-55
3-14(c). NO, loss rate for CALPUFF for NO, = 1400 ppb ....................... 3-56
3-15(a). TNO3 formation rate for CALPUFF for NO, = 7 ppb .................... 3-57
3-15(b). TNO3 formation rate for CALPUFF for NO, « 350 ppb .................. 3-58
3-15(c). TNO3 formation rate for CALPUFF for NO, « 1400 ppb ................. 3-59
3-16. SO2 loss rate for ARM3 ........................................... 3-60
3-17. SO4 loss rate for ARMS ........................................... 3-61
3-18. NO, loss rate for ARM3 .......................................... 3-62
3-19. TNO3 formation rate for ARM3 ....................... : ............ 3-63
I:\i314\«314rrv\toc.wpb
-------
List of Illustrations (Continued)
Figure Page
3-20(a). Diurnal distribution of conversion rates for SO* SO4. NOr and TNO3
predicted by the CALPUFF model assuming (1) concentrations of O, and
NOX equal 50 ppb and 7 ppb, respectively, (2) ambient temperature = 303°K,
and (3) relative humidity « 50% 3-64
3-20(b). Diurnal distribution of conversion rates for SOj, SO4. NOr and TNO3
predicted by the CALPUFF model assuming (1) concentrations of O3 and
NOX equal 50 ppb and 7 ppb, respectively, (2) ambient temperature = 303°K,
(3) relative humidity = 50%, and (4) the nighttime conversion rate for
NO, is set to 0.093% hour as the ARM3 model 3-65
3-21. Diurnal distribution of conversion rates for SO2, SO4. NOr and TNO3
predicted by the ARM3 model assuming (1) concentrations of O3 and NOX
equal 50 ppb and 7 ppb, respectively, (2) ambient temperature = 303°K,
and (3) relative humidity = 50% 3-66
4-1. Distribution of peak observed concentrations (fL/L) during CAFTEX #3 4-5
4-2. Distribution of peak concentrations (fL/L) produced by MESOPUFF n
using MESOPACII wind fields for CAPTEX #3 4-6
4-3. Distribution of peak concentrations (fL/L) produced by MESOPUFF II
using CALMET wind fields for CAPTEX #3 4-7
4-4. Distribution of peak concentrations (fL/L) produced by CALPUFF
using CALMET wind fields for CAPTEX #3 4-8
4-5. Distribution of peak concentrations (fL/L) produced by MESOPUFF H
using CALMET/MM4 wind fields for CAPTEX #3 4-9
4-6. Distribution of peak concentrations (fL/L) produced by CALPUFF
using CALMET/MM4 wind fields for CAPTEX #3 4-10
4-7. Distribution of peak observed concentrations (fL/L) during CAPTEX #5 .... 4-11
I:\a314\«314rav\locwph
XIV
-------
List of Illustrations (Continued)
Figure Page
4-8. Distribution of peak concentrations (fL/L) produced by MESOPUFF n
using MESOPAC H wind fields for CAPTEX #5 4-12
4-9. Distribution of peak concentrations (fL/L) produced by MESOPUFF n
using CALMET wind fields for CAPTEX #5 4-13
4-10. Distribution of peak concentrations (fL/L) produced by CALPUFF
using CALMET wind fields for CAPTEX #5 4-14
4-11. Distribution of peak concentrations (fL/L) produced by MESOPUFF n
using CALMET/MM4 wind fields for CAPTEX #5 4-15
4-12. Distribution of peak concentrations (fL/L) produced by CALPUFF
using CALMET/MM4 wind fields for CAPTEX #5 4-16
4-13. Distribution of peak observed concentrations (fL/L) during CAPTEX #7 4-17
4-14. Distribution of peak concentrations (fL/L) produced by MESOPUFF n
using MESOPAC H wind fields for CAPTEX #7 4-18
4-15. Distribution of peak concentrations (fL/L) produced by MESOPUFF II
using CALMET wind fields for CAPTEX #7 4-19
4-16. Distribution of peak concentrations (fL/L) produced by CALPUFF
using CALMET wind fields for CAPTEX #7 4-20
4-17. Distribution of peak concentrations (fL/L) produced by MESOPUFF n
using CALMET/MM4 wind fields for CAPTEX #7 4-21
4-18. Distribution of peak concentrations (fL/L) produced by CALPUFF
using CALMET/MM4 wind fields for CAPTEX #7 4-22
4-19. Peak modeled and observed concentration as a function of distance
for CAPTEX #3 4.32
I:\i314\i314rcv\tocwph
-------
List of Illustrations (Concluded)
Figure Page
4-20. Peak modeled and observed concentration as a function of distance
for CAPTEX #5 4-33
4-21. Peak modeled and observed concentration as a function of distance
for CAPTEX #7 4-34
A-l. Schematic plan view of the computational grid used by NUATMOS A-2
I:\i314\4314rev\toc-wph
-------
List of Tables
Table Page
2-1. Computation of the Angle, P", Used in the Computation of the Slope
Row Vector (from Douglas and Kessler, 1988) 2-8
2-2. Net Radiation Constants (Holtslag and van Ulden, 1983) 2-18
2-3. Diagnostic Model Run Matrix 2-33
2-4. Statistical Summary of Summer Episode CALMET vs. Nested Grid
MM4-FDDA Trajectories Combined Over Three Levels and All Sites 2-87
2-5. Statistical Summary of Summer Episode NUATMOS vs. Nested Grid
MM4-FDDA Trajectories Combined Over Three Levels and All Sites 2-88
2-6. Summary of Statistics for All Hours and All Levels, 54-km Grid Results 2-91
2-7. Summary of Statistics for All Hours and All Levels, 18-km Grid Results 2-92
3-1. Summary of Puff/Plume Model Features 3-2
4-1. Performance Measures for All CAPTEX Concentrations Paired
in Time and Space 4-25
4-2. Performance Measures for Peak CAPTEX Concentrations Paired in Space ... 4-28
I:\«314\«314rrv\tocwph XV11
-------
1. INTRODUCTION
An official Memorandum of Understanding (MOU) among the Environmental
Protection Agency (EPA), USDA National Forest Service (FS), the National Park Service
(NFS), and the United States Fish and Wildlife Service (FWS) was signed and completed
November 25, 1991. As part of the agreement an Interagency Workgroup for Air Quality
Modeling (IWAQM) was formed. The objectives of the work-group include: review, identify
and intercompare candidate air quality simulation modeling techniques that can be used to
estimate pollutant concentrations on a regional scale, including the individual and cumulative
impacts of proposed and existing sources on Air Quality Related Values (AQRVs), Prevention
of Significant Deterioration (PSD) increments, and National Ambient Air Quality Standards
(NAAQS), with emphasis on Federal Class I areas. The output from such models is needed by
Federal land managers, and others, to make informed decisions regarding the protection of
natural resources.
Sigma Research has conducted for IWAQM sensitivity analyses of several meteorological
models and puff dispersion models suitable for application to regional problems. The analysis
has been conducted in three phases:
• Sensitivity testing of diagnostic meteorological models;
• Sensitivity tests of dispersion model components; and
• Evaluation of the recommended meteorological and dispersion models
with tracer data collected during the Cross Appalachian Tracer
Experiment (CAPTEX) field study.
The meteorological sensitivity tests included simulations with the CALMET (Scire et aL,
1990) and NUATMOS (Ross et al., 1988; Smith and Ross, 1988) diagnostic wind field models.
These models were applied in two ways: (1) using observed surface and upper air wind data as
input and (2) using gridded wind fields produced by the Penn State/NCAR Mesoscale Model
with four-dimensional data assimilation (MM4-FDDA) (Anthes et al., 1987; Stauffer et aL, 1990)
on an 80-km grid as "data" in addition to the actual observed wind data. One of the primary
objectives of the analysis was to assess whether the introduction of the hourly, gridded 80-km
MM4 data with its improved spatial and temporal resolution over typical observational networks
would improve the quality of the wind fields produced by the diagnostic wind field models on
smaller grid scales (e.g., 5-50 km).
Several groups were involved in the meteorological modeling. All of the simulations
involving NUATMOS were conducted by the NFS. The EPA provided the MM4-FDDA data
i:\i314\t314rev\iectl.wph 1-1
-------
and observational data used in the modeling. Sigma Research was responsible for coordinating
the activities of the other groups, conducting the CALMET modeling, and summarizing the
statistical performance of the models.
Section 2 contains a description of the meteorological models and a discussion of the
meteorological sensitivity test results. Based on the superior ability of CALMET to reproduce
fine grid scale wind fields predicted by nested-grid simulations of MM4-FDDA, and its ability to
predict other meteorological fields required by the dispersion models such as surface heat and
momentum fluxes, mixing heights, and gridded precipitation fields, CALMET was selected by
IWAQM for use in the model evaluation study with the CAFTEX tracer data. A review of the
features and capabilities of several non-steady-state dispersion models (i.e., MESOPUFF-n,
CALPUFF, CITPUFF, ARMS) and the regulatory plume model, ISC2, is presented in Section 4.
Based on the model capabilities and dispersion model sensitivity test results, CALPUFF model
was selected by IWAQM to serve as the refined model for CAPTEX evaluation effort.
The CALMET/CALPUFF models were evaluated both with and without using 80-km
grid MM4-FDDA winds to supplement actual observed wind data in the construction of the
wind fields. In addition, the MESOPAC-H/MESOPUFF-n models, recommended by IWAQM
as the interim (Phase I) models for simulating long range transport, were tested against the
CAPTEX tracer data. Also, spatially-interpolated MM4-FDDA winds were tested as well. The
results of the model evaluation effort, which suggest that the improved resolution offered by the
MM4-FDDA fields enhances the performance of the dispersion modeling, are discussed in
Section 4. A summary of the main results and conclusions of the study is provided in Section 5.
fc\«314\«314rev\ieetl.«ph 1-2
-------
2. METEOROLOGICAL MODEL SENSITIVITY TESTS
2.1 Description of CALMET
The wind field module in CALMET is based on the Diagnostic Wind Model (DWM),
which was developed by Douglas and Kessler (1988). It contains parameterizations for slope
flows, kinematic terrain effects, terrain blocking, divergence minimization, smoothing, and
objective analysis. The diagnostic wind field model in CALMET uses a two step procedure in
developing the final wind fields, as illustrated in Figure 2-1. An initial guess field, based on
domain-average winds, is adjusted for terrain effects and divergence minimization to produce a
"Step 1" wind field. Observations are introduced into the Step 1 wind field to produce a final
("Step 2") wind field. An objective analysis procedure is used to incorporate the observational
data into the Step 1 field in such a way as to weight the observations heavily in portions of the
domain with observational data and weight the Step 1 field heavily in regions without
observations.
A series of modifications have been made to the DWM to make it more suitable for
regional applications. These include options to use a spatially variable initial guess field based
on observations or coarse-grid prognostic wind model data, optional Lambert conformal map
factors, and the ability to use hourly MM4-FDDA data as a supplement to observational data.
CALMET also contains algorithms to compute boundary layer parameters such as
surface friction velocity (u.), Monin-Obukhov length (L), convective velocity scale (w.), and
mixing heights (zj, which are described in Scire et al. (1990). CALMET uses an energy balance
technique over land surfaces and a profile technique in the overwater boundary layer to
determine the surface heat and momentum fluxes.
In the current study, MM4-FDDA wind fields have been used to supplement the
observational data normally used by CALMET. The MM4 data can be introduced into
CALMET in three different ways:
• as a replacement for the initial guess wind field (pathway <§) in Figure 2-1).
• as a replacement for the Step 1 field (pathway ©); or
• as "observations" in the objective analysis procedure (pathway O)-
Each of these methods of using the MM4 data was tested as part of the meteorological
sensitivity analysis of CALMET.
lA*314\«314irv\ieel2.wph 2-1
-------
SttUpUtial
(D«MiB Scak Winds)
CM^VH Terrain
EBMts
SttplWiMl
FWd
Nrfom Object**
Aaaljpsis Pntecdarc
FtaWJ
Apply CXBricn
Pim«d»r» and
(OfriowO)
FMd
Figure 2-1. Flow diagram of the diagnostic wind model in CALMET. The MM4 80-km
winds can be introduced as the initial guess field ®, Step 1 field ®, or as
"observations" 0-
2-2
-------
2.1.1 Grid System
The CALMET model uses a grid system consisting of NZ layers of NX by NY square
horizontal grid cells. Figure 2-2 illustrates one layer of grid cells for a 7 x 4 grid. The "grid
point" refers to the center of the grid cell in both the horizontal and vertical dimensions. The
"cell face" refers to either the horizontal or vertical boundary between two adjacent cells. In
CALMET, the horizontal wind components (u and v) are defined at each grid point. The
vertical wind component (w) is defined at the vertical cell faces.
The position of the meteorological grid in real space is determined by the reference
coordinates (X,,, Y0) of the southwest corner of the grid cell (1,1). Thus, grid point (1,1) is
located at (^ + AX/2, Y0 + AX/2), where AX is the length of one side of the grid square.
The CALMET model operates in a terrain-following vertical coordinate system.
Z = 2 - h,
(2-1)
where Z is the terrain-following vertical coordinate (m)
z is the Cartesian vertical coordinate (m), and
h, is the terrain height (m).
The vertical velocity, W, in the terrain-following coordinate system is defined as:
a/i, aft,
W = w - u— - v— '-
dx dy
(2-2)
where w is the physical vertical wind component (m/s) in Cartesian coordinates, and
u,v are the horizontal wind components (m/s).
2.1.2 CALMET Wind Field Module (DWM)
2.1.2.1 Step 1 Formulation
The CALMET diagnostic wind field model uses a two-step approach in the computation
of the wind fields (Douglas and Kessler, 1988). In Step 1, an initial guess wind field is adjusted
for:
• kinematic effects of terrain
• slope flows
I:\t314\a314rev\iecl2.wph 2-3
-------
Grid Cell
I
Grid
Point
G
R
I
D
C
E
L
I
N
D
E
X
1 X
AX
(X0, Y0)
X GRID CELL INDEX
Figure 2-2. Schematic illustration of the CALMET horizontal grid system for a 7 x 4 grid
showing the grid origin (X0> Y0) and grid point location (•). Winds are defined
ut the grid points.
I:\a314\«314it!v\iBd2.*pb
2-4
-------
• blocking effects
• three dimensional divergence minimization
In the original DWM, the initial guess field was assumed to be spatially constant
throughout the grid. However, CALMET has been modified to allow spatially variable winds as
an initial guess field. This option is considered to be important for regional domains, since the
assumption that a single domain average wind can characterize the entire domain is not
appropriate on regional scales.
For small modeling domains, spatially-constant domain mean wind may be adequate. In
CALMET, the domain mean wind components can be computed internally in the model by
vertically averaging and time-interpolating upper air sounding data or simply specified by the
user. If the domain mean winds are computed, the user specifies which upper air station is to
be used for determining the domain mean wind and the vertical layer through which the winds
are to be averaged.
The parameterization used in the computation of a Step 1 wind field, i.e., simulation of
kinematic effects of terrain, slope flows, blocking effects, and divergence minimization, are
described in the following sections. This discussion is largely derived from Douglas and Kessler
(1988).
Kinematic Effects
CALMET parameterizes the kinematic effects of terrain using the approach of Liu and
Yocke (1980). The Cartesian vertical velocity, w, is computed as:
w = (K-vT,,) exp(-te)
(2-3)
where V is the domain-mean wind,
h, is the terrain height,
k is a stability-dependent coefficient of exponential decay, and,
z is the vertical coordinate.
The exponential decay coefficient increases with increasing atmospheric stability.
N
\y\
(2-4)
I:\t314\t314rev\iect2.wph 2-5
-------
(2-5)
where N is the Brunt-Vaisala frequency in a layer from the ground through a user-input
height,
6 is the potential temperature,
g is the acceleration due to gravity, and,
tV I is the speed of the domain-mean wind.
The initial guess wind is then used to compute the terrain-forced Cartesian vertical
velocity, w, into a terrain-following vertical velocity, W (Eqn. 2-2). The kinematic effects of
terrain on the horizontal wind components are then evaluated by applying a divergence-
minimization scheme to the initial guess wind field. The divergence minimization scheme
iteratively adjusts the horizontal wind components until the three-dimensional divergence is less
than a user-specified maximum value.
Slope Flows
CALMET uses an empirical scheme to estimate the magnitude of slope flows in complex
terrain. The direction of the slope flow is assumed to be oriented in the drainage direction.
The slope flow vector is added into the Step 1 gridded wind field in order to produce an
adjusted Step 1 wind field.
(2-6)
v,' • v, + v,
(2-7)
where (u,,Vj) are the components of the Step 1 wind field before considering slope flow
effects,
(u,,v,) are the slope flow wind components, and
(Uj'.Vj') are the components of the Step 1 wind field after considering slope flow
effects.
The direction of the slope flow is determined by the following empirical procedures
suggested by Allwine and Whiteman (1985). First, an angle, p' is computed based on the slope
of the terrain.
I:\«3M\«314rtv\iKl2.%ph 2-6
-------
P' « tan'1 [ (», / dy) I (dht I 3r) ]
(2-8)
where h, is the height of the terrain.
A second angle, P', is computed as shown in Table 2-1. The drainage direction, 0* is
defined as:
/ 90 - p" 0 * P" £ 90
' * \450 - p" 90 < p"<360
(2-9)
where Pd is in degrees.
By definition, the magnitude of the slope flow is:
(2-10)
which is parameterized as a function of time of day, domain-scale temperature lapse rate,
terrain height, and terrain slope as:
where S is the speed of the slope flow (m/s),
Sj is a measure of the steepness of the terrain
(2-11)
dh,
dx
dh
1/2
Y is the domain-scale temperature lapse rate,
h^ is the maximum terrain height within the radius of influence of terrain features,
T0 is the domain-averaged air temperature and,
P2 is a function which depends on time of day. It has a value of -I- 1 for upslope flows
and - 1 for downslope flows.
Blocking Effects
The thermodynamic blocking effects of terrain on the wind flow are parameterized in
terms of the local Froude number (Allwine and Whiteman, 1985).
I:\i314\i314frv\xa32, wph
2-7
-------
Table 2-1
Computation of the Angle, P", Used in the Computation of the Slope Flow Vector
(from Douglas and Kessler, 1988)
P" (degrees)
i.o
ay
dh,
dht
ly"*0
dh, dh, - dh,
— I =0 — - < 0 — - > 0
cbc ctx cbc
P' + 180 p' + 360
270 p' + 180 P' + 360
90 P' + 180 p'
* Flat terrain. Drainage direction is undefined.
I:\>314\a314rev\iec(2.wpb
2-8
-------
-
ATAA,
H • (^ - (Z)*
(2-12)
where Fr is the local Froude number,
V is the wind speed at the grid point,
N is the Brunt, Vaisala frequency as defined in Eqn. (2-5),
Ah, is an effective obstacle height,
(hguJy is the highest gridded terrain height within a radius of influence of the grid point
(ij), and
(z)p is the height of level k of grid point (ij) above the ground.
The Froude number is computed for each grid point. If Fr is less than a critical Froude
number and the wind at the grid point has an uphill component, the wind direction is adjusted
to be tangent to the terrain. The wind speed is unchanged. If Fr exceeds the critical Froude
number, no adjustment is made to the flow.
2.1.2.2 Step 2 Formulation
The second step in the processing of the wind field by the diagnostic model is the
introduction of observational data into the Step 1 gridded wind field. The Step 2 procedure
consists of four substeps (Douglas and Kessler, 1988).
• Interpolation
• Smoothing
• O'Brien adjustment of vertical velocities
• Divergence minimization
Interpolation
An inverse-distance method is used to introduce observational data into the Step 1 wind
field.
L\«314\«314rev\«ect2.wph 2-9
-------
JL
R\
(2-13)
where (uote,vobt)k are the observed wind components at station k,
(u,v)i are the Step 1 wind components at a particular grid point,
(u,v)2 are the initial Step 2 wind components,
Rk is the distance from observational station k to the grid point, and
R is a user-specified weighting parameter for the Step 1 wind field.
This interpolation scheme allows observational data to be heavily weighted in the vicinity
of the observational station, while the Step 1 wind field dominates the interpolated wind field in
regions with no observational data. The weighting procedure described by Eqn. (2-13) is applied
independently to each vertical layer. Surface observations are used only for lowest wind field
layers except when the option for vertical extrapolation of the observational data is selected.
The user specified parameter, R, determines the relative weighting given to the Step 1
wind field. Different values of R are used in the surface layer (Rj), and layers aloft (R2).
An observation is excluded from interpolation if the distance from the observational
station to a particular grid point exceeds a maximum radius of influence. Three separate
maximum radius of influence parameters can be specified by the user:
• Radius of influence over land in the surface layer (RMAX1)
• Radius of influence over land in layers aloft (RMAX2)
• Radius of influence over water (RMAX3)
The region influenced by an observation can be limited by user-specified "barriers."
These barriers consist of line segments which define the boundaries of the region of the grid
which can be influenced by a particular observation. Any time a barrier exists between a grid
point and an observation site, the observational data is omitted for the interpolation. For
example, user-specified barrier segments can be defined to prevent observational data from a
station in a well-defined valley from being applied outside the valley region.
I:\»314\i314rw\ieet2.wph 2-10
-------
Smoothing
The intermediate Step 2 wind field resulting from the addition of observational data into
the Step 1 wind field is subject to smoothing in order to reduce resulting discontinuities in the
wind field. The smoothing formula used in CALMET is:
"u " O-5 « * °-125
(2-14)
where (u^)" is the u wind component at grid point (ij) after smoothing, and
is the u wind component before smoothing, as determined by Eqn. (2-13)
A similar equation is applied for the v component of the wind.
Computation of Vertical Velocities
Two options are available for computing vertical velocities in CALMET. With the first
method the vertical velocities are computed directly from the incompressible conservation of
mass equation using the smoothed horizontal wind field components. The second method
adjusts the vertical velocity profile so that the values at the top of the model domain are zero.
The horizontal wind components are then readjusted to be mass consistent with the new vertical
velocity field.
The initial vertical velocity is determined from the incompressible mass conservation
equation:
dx dy dz
(2-15)
where w, is the vertical velocity in terrain-following coordinates, and
u",v" are the horizontal wind field components after smoothing.
This mass-consistent vertical velocity is used as the final vertical velocity (i.e., w = wj if
Method 1 is selected. Also, with this method, no further adjustment is made to the horizontal
wind components. The final horizontal winds are the smoothed winds resulting from Eqn. (2-
14).
I:\i314\«314rev\ieet2.«pb 2-11
-------
Godden and Lurmann (1983) suggest that this procedure may sometimes lead to
unrealisticalty large vertical velocities in the top layers of the grid. In order to avoid this
problem, CALMET provides an option to use a procedure suggested by O'Brien (1970) to adjust
The O'Brien procedure forces the vertical velocity at the top of the model domain to be
zero. Because the horizontal winds are not mass consistent with the adjusted vertical velocities,
the horizontal winds are subject to adjustment by the divergence minimization scheme described
below. The divergence minimization procedure iteratively adjusts the u and v components to
within a user-specified divergence threshold while holding the vertical velocity field (w = w2)
constant.
There are situations where the use of the O'Brien procedure is not warranted. For
example, if the top of the modeling grid is within a sea-breeze convergence zone, the large
vertical velocities resulting from application of Eqn. (2-15) may be realistic. Therefore, the use
of the O'Brien procedure is an optional feature of CALMET.
Divergence Minimization Procedure
Three-dimensional divergence in the wind field is minimized by a procedure described by
Goodin et al. (1980). This procedure iteratively adjusts the horizontal wind components (u,v)
for a fixed vertical velocity field so that at each grid point, the divergence is less than a user-
specified maximum value.
dx dy dz.
(2-17)
where u,v are the horizontal wind components,
w is the vertical velocity in terrain following coordinates, and
e is the maximum allowable divergence.
In CALMET, the horizontal wind components are defined at the grid points. Vertical
velocities are defined at the vertical grid cell faces. Therefore, the divergence, D, at grid point
is:
I:\i3M\i314rev\tea2.»ph 2-12
-------
24)1 (MS)
where Ax and Ay are the sizes of the grid cell in the x and y directions, respectively.
For each grid point, divergence is computed. The u and v wind components at the
surrounding cells is adjusted so that the divergence at the grid point is zero. The adjustments
are:
4>
(2-19)
(B-4-u* s •'-«* " "•*
W (2-20)
(V"«W = Vu * v^
(2-21)
(V-~)y-U = Vw ' V*
(2-22)
where the adjustment velocities (u^v^) are:
(2-23)
(2-24)
Each time the divergence is eliminated at a particular grid point, divergence is created at
surrounding points. However, by applying the procedure iterativeh/, the divergence is gradually
reduced below the threshold value, c, throughout the grid.
I:\«314\a314icv\Md2.«!>b 2-13
-------
2.1.3 Micrometeorologjcal Model
Surface Heat and Momentum Flux Parameters
A number of significant advances have been made in recent years in our understanding
and characterization of the structure of the planetary boundary layer (PEL) (e.g., see Weil,
1985; Briggs, 1985). As noted by van Ulden and Holtslag (1985) and others, the use of the
appropriate boundary layer scaling parameters can improve the quality or dispersion predictions.
The principal parameters needed to describe the boundary layer structure are the surface heat
flux, surface momentum flux, and the boundary layer height. Several additional parameters,
including the friction velocity (u.), convective velocity scale (w.), and the Monin-Obukhov length
(L), are derived from these.
Hanna et al. (1986) have evaluated several models for the prediction of these boundary
layer parameters from routinely available meteorological observations. Two basic methods are
commonly used to estimate the surface heat and momentum fluxes. The first method is referred
to as the profile method. It requires at a minimum the measurement of the wind speed at one
height and the temperature difference between two heights in the surface layer, as well as
knowledge of the air temperature and roughness characteristics of the surface. Monin-Obukhov
similarity theory is then used to solve for the surface fluxes by iteration. The second approach,
called the energy budget method, computes the surface heat flux by parameterizing the unknown
terms of the surface energy budget equation.
Hanna et al. (1986) tested the following four energy budget models and two profile schemes:
Energy Budget Models
Holtslag and van Ulden (1983)
Weil and Brower (1983)
• Berkowicz and Prahm (1982)
Briggs (1982)
Profile Schemes
Two-level tower method
Four-level tower method
I:\43M\a314rrv\iediwph 2-14
-------
The major conclusion drawn from the comparison of the six schemes was that the energy
budget methods were superior because of the sensitivity of the profile method to small errors in
the measured temperature difference. However, as discussed below, this conclusion does not
apply to the marine boundary layer, where a profile method based on the air-sea temperature
difference is recommended. The relative performance of all of the energy budget methods was
similar. An intercomparison of the u. predictions of each of the energy budget methods showed
a very high correlation with the other energy budget schemes (r2 from 0.98 to 0.99 and RMS
errors from 0.027 to 0.055 m/s). The correlation coefficient of the energy budget schemes with
observed u* ranged from 0.63 to 0.65 and RMS errors from 020 to 021 m/s.
Overland Boundary Layer
A energy budget method, based primarily on Holtslag and van Ulden (1983), is used over
land surfaces in the CALMET micrometeorological model The energy balance at the surface
can be written as:
Q. + Qf*QB + Qt + Qt (2-25)
where, Q. is the net radiation (W/m2),
Qf is the anthropogenic heat flux (W/m2),
Qh is the sensible heat flux (W/m2),
Qe is the latent heat flux (W/m2), and,
Qg is the storage/soil heat flux term (W/m2).
The ratio of the sensible heat flux to the latent heat flux is defined as
the Bowen ratio.
B = — (2-26)
-------
is the solar elevation angle (deg.).
I:\a314\i314icv\iecl2.wph 2-16
-------
The last term in Eqn. (2-30) accounts for the reduction of incoming solar radiation due
to the presence of clouds. The values for the empirical constants q, Cj, c,, alt a^ blt and bj
suggested by Holtslag and van Ulden (1983) are used (see Table 2-2), although it may be
beneficial to develop site-specific values. The solar elevation angle is computed each hour using
equations described by Scire et aL (1984).
Using Eqns. (2-25) to (2-30), the daytime sensible heat flux can be expressed in terms of
only known quantities:
Once the sensible heat flux is known, the Monin-Obukhov length and surface friction
velocity are computed by iteration.
«. * *n/[ln(z/z.) - t.fc/I) «• *«(Z./I)] (2-32)
where, z0 is the surface roughness length (m),
ym is a stability correction function [e.g., see Dyer and Hicks (1970)],
k is the von Karman constant, and,
u is the wind speed (m/s) at height z.
The Monin-Obukhov length is defined as:
L - -pC'r"! (2-33)
* *
-------
Table 2-2
Net Radiation Constants (Holtslag and van Ulden, 1983)
Constant Value
q 531 x ID'13 W/m2/deg K*
Cj 60 W/m2
GS 0.12
ax 990 W/m2
a2 -30 W/m2
bi -0.75
b 3.4
I:\«314\t314rev\ieet2-wph 2-18
-------
c -1 -
C
c
(2-35)
(2-36)
where, C!>N is the neutral drag coefficient
Y is a constant (• 4.7), and,
z,, is the measurement height (m) of the wind speed, u.
The temperature scale, 8., is computed as the minimum of two estimates:
6. = mm[6.,, e,2]
The estimate of 8. is based on Holtslag and van Ulden (1982):
e.t = 0.09 (1 - 0.5 N*)
and 8.2 is:
'•2
(2-37)
(2-38)
(2-39)
The heat flux is related to u. and 6. by:
- p cf «. e.
(2-40)
and L is computed from Eqn. (2-33).
The daytime mixing height is computed using a modified Carson (1973) method based
on Maul (1980). Knowing the hourly variation in the surface heat flux from Eqn. (2-31) and the
venical temperature profile from the twice-daily sounding data, the convective mixing height at
time t+dt can be estimated from its value at time t in a stepwise manner:
2 2
-------
where, ^ is the potential temperature lapse rate in the layer above hp
d6 is the temperature jump at the top of the mixed layer (*K), and,
E is a constant (- 0.15).
The neutral (mechanical) boundary layer height is estimated by Venkatram (1980b) as:
*"'
(2-43)
where, f is the Coriolis parameter (- 10"* s'1)
B is a constant (- 21/2), and,
NB is the Brunt- Vaisala frequency in the stable layer aloft.
The initial estimate of the daytime mixing height is taken as the maximum of the
convective and mechanical values predicted by Eqns. (2-41) and (2-43). However, the cell-to-cell
variations in mixing heights resulting from an independent computation of h, and h can result in
unreasonably large variations. Such an independent, cell-by-cell computation would also not
include important advectrve effects on the mixing depths, such as the significant reduction of
inland mixing depths during sea or lake breeze conditions.
Several researchers (e.g., Wheeler, 1990; Tesche et al., 1988; Steyn and Oke, 1982) have
suggested various upwind-looking mixing depth averaging schemes involving estimation of back
trajectories or computation of lateral advection of heat fluxes. As the model is explicitly
marched in time, a rather simple scheme has been incorporated which approximates the back
trajectory methodology. For a given grid cell (ij), the most upwind grid cell which could directly
impact cell (iu,ju) during the time step, dt, is computed as (i,, * i - u»dt, ju = j - v«dt), where
(u,v) are the wind components at cell (ij). An upwind-looking cone, originating at (ij) and
having a user-selected, half-opening-angle of HAFANG (i.e., a full cone opening angle of twice
HAFANG), is then generated such that grid point (^jj sits at the middle of the base of the
triangular cone. For each grid cell (ik jk) lying within or on the boundaries of the triangular
region, upwind and crosswind distances, du and dp respectively, are computed in units of number
of grid cells, and a weighting factor,
(2-44)
is computed. Normalized weights are then computed as,
w.
(2-45)
I:\«314\«314rev\»eea.wpb 2-20
-------
where the sum on n extends over all the grid points encompassed by the triangle. In addition,
Eqn. (2-44) weights are also computed for a square box of user-defined half-width of
MNMDAV grid cells and centered on cell (ij). The purpose of including this supplementary
square box region is to allow some intercell averaging to occur even when the mean advective
wind goes to zero. Hence, a reasonable value for MNMDAV would be of order ov«dt/dx,
which is usually of order unity in many mesoscale applications. For those cells which are
actually downwind, such that d, < 0, the quantity d, in the Eqn. (2-44) weight is replaced by the
quantity d,,' - e - d., where e is the Courant number or the height of the triangle from its base
at (i,ju) to the vertex at (ij). This ensures that downwind cells receive rather small weighting
but ensures complete azimuthal symmetry as the wind speed (and e) go to zero.
The weights, w/ appropriately normalized via Eqn. (2-45) for all points lying in the
triangular or square box regions, are then applied to the fields of convective and effective
daytime (i.e., the maximum of h, and h) mixing depths to produce smoothed equivalents, and
these fields are stored for use in the current hour. In addition, it is the spatially smoothed
convective h, which is used for the next hour's computation using Eqn. (2-41). Thus, there is a
cumulative effect on the convective h, calculation, comparable to the effect of computing a
multiple time step, back trajectory.
In the stable boundary layer, mechanical turbulence production determines the vertical
extent of dispersion. Venkatram (1980a) provides the following empirical relationship to
estimate the stable mixing height.
h = B2 u? (2-46a)
where B2 is a constant (- 2400).
Ziiitinkevich (1972) developed the following relationship for the mixing height under
stable conditions:
0.4
N
(2-46b)
where f is the Coriolis parameter. In CALMET, the stable mixing height is taken as the
minimum of h for Eqn. (2-46a) and (2-46b). The mixing heights are then spatially averaged
using the scheme described in (2-44) and (2-45).
In the convective boundary layer, the appropriate velocity scale is w., which can be
computed directly from its definition using the results of Eqns. (2-28) and (2-38).
I:\t314\«314fcv\iect2.«i>t> 2-21
-------
*. - \*
-------
Hosker's result is based on the analysis of Kitaigorodskii (1973) showing z,, « u.2 and the
logarithmic wind speed profile relating wind speed and u..
2.2 Description of NUATMOS
The NUATMOS model (Ross et aL, 1988; Smith and Ross, 1988) uses a variational
calculus method to produce three-dimensional mass-consistent wind fields. The method makes
the numerical adjustments necessary to an initial, interpolated wind field to produce a
divergence-free field. This is accomplished by minimizing the function
£(«,v,w) = HI [(u - uf + (v - v.f + a'2 (w - w.f ] dV
(2-52)
subject to the constraint
H = * + * «. ^L « o.
dx dy dz
(2-53)
where x,y are the horizontal coordinates,
z is the vertical coordinate,
V0 = (u0>vo»wo) are tne initial (interpolated) velocity components,
V = (u,v,w) are the final velocity components, and
a is the ratio of the adjustment of the relative vertical wind component to the horizontal
components.
Ross et al. (1988) express the constrained minimization function as:
ef * (v - vef + a'2 (w - w.
u - ue
where X(x,y,z) is a Lagrange multiplier, and
2(«-".) = f
(2-55)
(2-56)
I:\i314\«314rev\wd2.wph 2-23
-------
" w-> &
(2-57)
subject to the boundary conditions
X (u - « \ - 0
(2-58)
X (v - v,) * 0
(2-59)
(MO)
Eqs. (2-55) - (2-57) and Eq. (2-53) can be combined to yield
- 2V-F
«
^j ^^
(2-61)
NUATMOS user solves this equation with the assumption X = 0 at the free boundaries
and X(w - wc) * 0 at the surface.
The solution is performed in a terrain following coordinate system:
Jt = x
(2-62)
(2-63)
(2-64)
where, z^ is the height of the top of the domain,
z, is the height of the surface, and
the terrain-following velocity components are:
il - u
(2-65)
v = v
(2-66)
I:\«3M\«314rev\ieca.wph 2-24
-------
* = Jl |W + OB ^1
(2-67)
Ross et al. (1988) express the final equation for X as:
JL - £.!*_£.) (J*. - £. Ji Ji) + f J. - i J* _L
ax «axdoj(ax itdxdoj \dy it dy do
ax o a« ax MP #x
do J
(2-68)
with the boundary conditions
X (« - *,) = 0
(2-69)
X (v - v\ = 0
(2-70)
X (w - w \ = 0
(2-71)
on the x,y,and a boundaries.
A preliminary expression for the a parameter suggested by Ross and Fox (1991) is:
a'2 = 1 + ?
(s2 - l)Fr2
(2-72)
where s is the speedup of the flow over the terrain, and
Fr is the Froude number of the flow.
2.3 Grid and Episode Selection of Sensitivity Tests
CALMET and NUATMOS were tested using three-dimensional wind fields produced by
MM4-FDDA on an 80-km grid as "data" for the diagnostic models. The testing of the diagnostk
wind field models was conducted for two episodes-one summer episode and one winter episode.
The episodes were between 6 to 8 days in duration. The summer period (August 1-6, 1988)
was characterized by light wind, stagnating conditions. The winter period (December 3-10,
I:\i314\i314rev\tedlwpb 2-25
-------
1988) was an active period that included the passage of a front and low-pressure system through
the domain.
The simulations were conducted at three different grid sizes: 54-km, 18-km, and 5-km.
The selection of the grid sizes of 54-km and 18-km was based on the availability of special
nested-grid MM4-FDDA data during the summer episode on these grids. The MM4-FDDA
nested-grid results were not used to drive the diagnostic models but rather were used to test the
ability of the diagnostic models to interpolate the 80-km grid MM4-FDDA results down to finer
grid scales. Only the large grid 80-km grid MM4-FDDA data were used as input into the
diagnostic models. It is envisioned that an operational analysis will require the combination of
results from course-grid simulations to assess the impact of large, distant sources over hundreds
of kilometers from the Class I area, and fine scale simulations to evaluate near-field impacts
over several tens of kilometers. Computational and computer memory limitations are likely to
require a course model grid spacing in the 10-40 km range for the distant-source impact
analysis. However, a grid spacing of order 5-km or less may be necessary to resolve terrain
features and flow fields important for near-field sources. For example, Figure 2-3 compares
terrain contours in the northern part of the Shenandoah National Park with those obtained after
averaging the terrain elevations to a 5.5-km grid. The figure suggests that a grid scale of
approximately 5-km can resolve significant features of the topography in this region. Large grid
sizes (e.g., 20-40 km) do not resolve the terrain in the Shenandoah area adequately for near-
field source impacts. Thus, in many applications, it would be necessary to combine the results
from both fine-grid and course-grid simulations in order to produce a cumulative impact
assessment.
The MM4-FDDA nested grid domain and the subgrids for the IWAQM simulations are
shown for the 54-km and 18-km grids in Figures 2-4 and 2-5, respectively. In the staggered grid
system used in MM4-FDDA, the wind components are defined at "dot points." The MM4-
FDDA data provided by EPA has been interpolated to the grid cell locations, so the
diagnostic model grids have been designed to correspond to a subset of the interpolated MM4-
FDDA grid data.
The 5-km grid shown in Figure 2-6, consists of 93 x 57 grid points. The 5-km grid
domain includes both the Shenandoah National Park and the coastal region of Virginia in order
to provide information on both coastal and mountainous flow field patterns. The ability to
simulate transport and dispersion in both regions is considered critical because most Class I
areas are located in high terrain areas or coastal regions.
I:\«314\i314icv\iect2.wph 2-26
-------
CSIwf&tM
Iiiiiuin iiiniiii iiiiiiiu minim iiiiiiiii iiiiiiin iiiiiiiu inn
-IIOM -(1600 -IMXM -10JOO
-IHCOO -J.OOO
Figure 2-3. Digitized terrain contours of the northern portion of the Shenandoah National
Park. Contour Elevations (ft) are 1000, 1200, 1400, 1600, 2000, 2500, and 3000.
-------
140 *
130 V 120 » 110 ¥ 100 T 90 » 80 * 70
•0 T 60 V
40
4O M
30 K
20 N
110 ¥
1OO V
Figure 2-4. The MM4-FDDA domain for the 54-km nested grid (outer box) and the
proposed IWAQM 54-km subgrid (inner box). The MM4-FDDA domain consists
of 125 x 92 grid points. The IWAQM 54-km subgrid consists of
34 x 30 grid points.
I:\*314\a314rev\MCi2.wph
2-28
-------
90 V
06 T
10 V
43 N !
46 II
40 •
M V
Figure 2-5. The MM4-FDDA domain for the 18-km nested grid (outer box) and the
proposed IWAQM 18-km subgrid (inner box). The MM4-FDDA domain consists
of 103 x 91 grid points. The IWAQM subgrid consists of 38 x 36 grid points.
I:\U14\m314rev\ieeawpb
2-29
-------
M V
TO t
70 V
4ft M
40 M
aa M :
00 V
Figure 2-6. The proposed IWAQM 5-km subgrid (inner box). The outer box is the
18-km grid domain for MM4-FDDA nested grid simulations. The IWAQM 5-km
subgrid consists of 93 x 57 grid points.
I:\i314\«314wv\teel2.wph
2-30
-------
The modeling domain grids were defined as:
54-km 18-km
a (-27,-594) (657,-432) (866.5.-240.5)
u. (-54.-621) (648.-441) (864.-24S)
nx,ny 3430 3836 93,57
A (km) 54 18 5
where (X^YJ,,. are the coordinates for the center of cell (1,1), (X^YJ^ are the coordinates
for the lower left corner of cell (1,1), nx and ny are the number of cells in the grid, and A is the
size of one cell (DX «= DY = A).
A total of 7 vertical levels were used in the CALMET diagnostic model simulations. The
vertical levels are:
10 m, 50 m, 100 m, 200 m, 400 m, 800 m, and 1240 m
The vertical structure used in NUATMOS is defined by surfaces of constant o, where a
represents the height of the cell interfaces between the top of the domain and ground level.
The o levels were chosen to yield approximately the same vertical levels as modeled with
CALMET. For example, the NUATMOS layer closest to the surface has its center between 8 m
and 10 m.
The CALMET model was run using meteorological observations from 119 surface
stations and 25 upper air stations for the summer episode and 23 upper air stations for the
winter episode.
CALMET was run with a small radius of influence and a large radius of influence in
order to demonstrate the influence of the domain mean wind on the flow fields. The radii used
were defined as:
I:\»314\«314rev\i«ct2.wpb 2-31
-------
Smajj Large
RI
R2
RMAXl
RMAX2
RMAX3
SO km
100km
100km
200km
1000km
500km
1000km
500 km
1000km
1000km
where, Rlt R2 « the distances at which the Step 1 wind and observations have equal weight in
the surface layer (RJ and upper layer (R2),
RMAXl, RMAX2 = the maximum radius of influence of observations in the surface
(RMAXl) and upper layers (RMAX2), and
RMAX3 « the overwater radius of influence.
Table 2-3 contains the matrix of diagnostic model runs made in the meteorological
testing phase of this project. A total of 41 model simulations were performed, i.e., 14 CALMET
runs using small radius of influence parameters, 14 CALMET runs using large radius of
influence parameters, and 13 NUATMOS runs.
2.4 Use of MM4-FDDA Wind Fields
Each of the models studied was designed to accept observed wind data at a number of
heights and locations within a modeling domain. Therefore, the output from the MM4-FDDA
simulations could be treated as "observations", and processed along with the true observed
winds. The potential drawback to this approach is that no distinction is made in the relative
confidence we may have in the MM4-FDDA simulations and the observed wind data. For
example, when winds are interpolated to the modeling grid, nearby wind observations are
treated in the same way as nearby MM4-FDDA winds, even though local circulations embodied
in the observed winds may be "missed" by the 80 km resolution of the MM4-FDDA simulation.
Therefore, two important elements in the sensitivity analysis included (1) altering the
manner in which the simulated MM4-FDDA wind fields from the coarse grid are combined with
either the observed point-value winds and/or the winds generated by the diagnostic wind field
models and (2) altering the relative weights assigned to each in the combination process. The
representativeness on a fine-scale grid of the observed point-value winds as compared with
winds derived from the MM4-FDDA on a coarse grid is expected to depend on such factors as
the height above the surface, subgrid-scale terrain variations, and the ratio of the coarse-grid to
fine-grid size. For example, the 80-km MM4-FDDA winds will not reflect potentially important
t\i314\i314iw\iea2.wph 2-32
-------
Table 2-3
Diagnostic Model Run Matrix
Model
CALMET
CALMET
CALMET
CALMET
CALMET
NUATMOS
NUATMOS
NUATMOS
NUATMOS
NUATMOS
Run No.
1
2
3
4
5
6
10
11
12
13
14
15
16
17
1
2
3
4
5
6
7
8
9
10
11
12
13
Radius of Episode**
Influence*
S.L summer
S.L
S.L
S,L summer
SO.
SJ.
S,L summer
S,L
S.L
S,L summer
S,L
S,L
S,L winter
S,L
summer
summer
summer
winter
winter
Grid Size
5km
18km
54km
5km
18km
54km
5km
18km
54km
5km
18km
54km
5km
18km
5km
18km
54km
5km
18km
54km
5km
18km
54km
5km
18km
5km
8km
Run Description
observations only
observations with 80 km MM4 data
used as initial guess field
observations with e weighted
80 km MM4 data used as Step 1
field
observations with e weighted
80 km MM4 data used as
'observations*
observations only
observations only
observations with 80 km MM4 data
observations with a weighted
80 km MM4 data
observations only
observations with o weighted
80 km MM4 data
S * Small radius of influence parameters
L « Large radius of influence parameters
summer episode: August 1,1988 7 am - August 6,1988 7 am
winter episode: December 3, 1988 7 am - December 10, 1988 7 am
I:\«314\«314rev\«ed2.wpb
2-33
-------
local features of the surface flow field induced by terrain variations which can not be resolved by
this coarse MM4-FDDA grid in areas such as the Shenandoah National Park. On the other
hand, the point-value "snapshot* observations in such areas do not necessarily represent larger-
scale flow fields as well as the MM4-FDDA fields. Therefore, a weighting factor based on the
subgrid-scale terrain variations within each grid cell has been derived.
2.4.1 Terrain Weighting Factor
Although the use of MM4-FDDA winds are expected in many circumstances to improve
the diagnostic model's wind fields, MM4-FDDA may not produce winds "near" the surface that
are representative if much terrain is poorly resolved by the scale of the grid used for the MM4-
FDDA simulations. When this is the case, local observations might be given more weight than
the MM4 winds in interpolating winds to the grid used for the diagnostic models. The
methodemployed for altering weights involved (1) computing ot, the standard deviation of the
departure of the "actual" terrain elevations from the grid-average terrain elevation, (2) defining
a weight W0 that is a function of ov and (3) weighting observed wind by W0, and MM4 winds by
(1 - W0) when performing the interpolation process.
A program was developed to quantify differences between the terrain as represented by
a "coarse" grid, in this case the 80-km grid used in one series of MM4-FDDA simulations, and
the terrain as represented on the original 0.9-km grid provided with the NPSAQMS. In this
program, we calculate the root-mean-square (RMS) of the difference between the original (0.9-
km) terrain and the "coarse-grid" terrain elevations within a region about each point in the
"coarse" grid. The difference in elevation, (hori - h^), is calculated with a resolution equal to
that of the original gridded terrain data, where h,^ is the elevation of a point contained in the
original terrain file and bilinear interpolation is used to find hm at the same location. A similar
procedure is also used to calculate RMS(hfiD - h,,,), where h^ denotes elevations in the "fine-
grid" used by the diagnostic models. The difference in elevation (h^ - hm) is found at the same
locations used for (h^ - hm), using bilinear interpolation within both the fine and coarse grids.
Therefore, RMS(hto - h^) is zero if the same grid is used by both the MM4-FDDA and the
diagnostic models.
The RMS for each grid-point represent a measure of the sub-grid terrain within each
coarse-grid-scale cell. It carries no information about the organization of the sub-grid terrain.
A second method developed for calculating the RMS uses bilinear interpolation to obtain (h -
hOT) along a line through the grid-point. The RMS is calculated for a number of orientations of
this line, and the largest RMS obtained is reported along with the orientation associated with it.
I:\i3M\i3J4fw\ted2.wph 2-34
-------
Future weighting techniques may be able to use this additional information, but for now, we will
focus on the use of the "grid-square" RMS in defining the weights, W0.
A simple formulation that allows near-surface adjustments to the use of MM4-FDDA
winds, is a product relationship:
w. s wt w.
(2-73)
where W, is the weighting factor near the surface, and Wz is a height-dependent modifier. Wz
tends toward zero if the model-layer being processed is well above the terrain, or if there are no
sub-grid variations in the terrain (e.g., if the terrain is fiat). Using the mean elevation of the
layer above the surface, denoted as Zj, and the RMS(hfln - h^), denoted simply as RMS^,
W, « [MI^RMS^ , 1.0)f
has the desired properties. When the terrain resolved by the fine-scale grid used by the
diagnostic model has a characteristic departure from the coarse-grid terrain (quantified as
RMSgJ that is less than the height of the layer, Wri will be less than 1, which will reduce the
magnitude of W0, indicating that the subgrid terrain is less important for this layer than for any
closer to the surface. As higher layers are processed, W^ approaches zero, which emphasizes
the use of the MM4-FDDA winds in the diagnostic model. If the fine-scale grid should have the
same resolution as the coarse grid, RMS^ «= 0 and Wr « 0, so that the MM4-FDDA winds are
used in preference to the observed winds at all levels. Note that the choice of the power of 2
hastens the rate at which W^, approaches zero at levels further from the surface.
The near-surface factor, W,, makes use of both RMSto and RMS^, where
- hm]
(2-75)
The scale of the departure of the original terrain from that resolved by the coarse-grid,
is used to scale the departure of the terrain resolved by the find grid from that resolved by the
coarse grid. The ratio RMS^/RMS^, has a range of 0 to 1.0, provided that RMS^ is not zero.
When RMS^ is zero, or when RMS^/RMS^ is nearly zero, W§ should be nearly zero, thereby
indicating that the MM4-FDDA winds should be preferred over any observed winds (the
observed winds have already been "used" within MM4-FDDA). On the contrary, when
RMSfe/RMSorf approaches 1.0, local subgrid terrain could be important, and local observations
or diagnostic wind estimates near the surface should be emphasized. Hence, W, can be given by
Wt = IRMS^KMS^ «• RMSJ)n
(2-76)
I:\a3M\i314rev\ieel2.wpb 2-35
-------
where the power V has been set to 2.0 for this study. For n > 1, smaller values of W, will be
produced, thereby making it more •difficult* to ignore the MM4-FDDA winds in favor of
observed winds. For n < 1, the opposite trait is favored.
RMS0 is added to RMS^ in the denominator to avoid a problem that arises if terrain
variations are "small". W, may be nearly 1.0 (which emphasizes the observed winds) in some
cases in which terrain variations are small enough that the MM4 winds are indeed
representative in the surface-based layer, in spite of W,. To address this case, a condition that
the terrain variations be "significant" is added. That is, the denominator is never allowed to fall
below some specified length-scale, RMS0. Because the center of the surface-based layer is 10 m
in these applications, a length scale of 10 m has been adopted for "significance". All cells in the
80 km grid that are so characterized as having insignificant terrain variation from that resolved
by the 80 km grid will thereby promote the use of MM4 winds in preference to observed winds
at nearby grid-points.
Results of applying these procedures to the terrain data for the various grids used in this
evaluation are shown in Figures 2-7 to 2-17 for the surface and heights of 50 m, 100 m, and
200m.
The weighting function (W0) is computed for each cell in the 80-km grid used to define
the MM4 winds that are treated as "observations" in these evaluations. Bilinear interpolation is
then used to obtain W0 at the center of each cell in the diagnostic modeling grids (5 km, 18 km,
and 54 km). One set of weights is presented for each layer in the vertical. These weights are to
be applied in addition to the inverse-distance-squared weighting used to interpolate winds to the
diagnostic grids.
2.4.2 Sensitivity to the Method of use of MM4-FDDA Data
The CALMET model contains three options for incorporating coarse-grid wind fields
such as MM4-FDDA fields as input:
• as the initial guess field,
• as the Step 1 wind field, or
• as "observations."
When used as the initial guess field, the coarse-grid winds are first interpolated to the
fine-scale CALMET grid. The normal diagnostic adjustments for the fine-scale terrain are then
made. This produces a "Step 1" field which is then subject to an objective analysis procedure
fc\«314\a314rav\Md2.«ph 2-36
-------
Weights (5 km) at Surface
1 4 710131619222528313437404346495255586164677073767982858891
55
52
49
46
43
40
37
34
31
28
25
22
19
16
13
W
7
Mt
=.0.
-.04
;/
—
—
~
—
—
•^
—
^•^H
"Tl
Lss
\ ^—
$
\
I I
/
/
TT
— «
/
^-
1 Jll
TT
•*=
•^
M
ft
\
X
1
•W.
"•^
^
s
— M/
II
9^*1
**s
It
Q
\
{
rff
1 1
"••^
^
o
\
~v,
II
Ir
\
\
\
x"
II
TF
***~»
•^
\
J
X
X
II
TT
•— '
^-~
/
^
y
ll'
TT
-OJ
-O.4
S
9
!
•?
Tf
TT
• —
/
/
/
II
1 J
x^
/
/
I
1
/
II
1 1
^
'
-
,•*
II
1 1
/
3
/
/
II
1 1
/
(*
^/
II
1 1
( — *
9
^
(
J
II
1 1
^•^'
X*
9
•I
* '
)
U
TT
S
^
r
/
}
•R:
N,
N
/
(X
*^
x^
V
II
IF
V
•
e
'
^
^
II
1 1
N,
/
II
W.
\
f
'
/
e
V
1
1 1
J
/
O*
.
^
Jl
1 1
X
r
;
,
^
i,
ii
/
V
,,
II
/
*
II
^1
%
<^
*
il
1 1
1
1
11
J-
—
.^
—
-
^
—
—
r
55
52
»Q
ns
46
43
40
37
34
31
28
25
22
19
16
13
10
7
4
1
1 4 710131619222528313437404346495255586164677073767982858891
Figure 2-7. Terrain weighting factor, W0, for the surface level on the 5-km grid.
l:\»314\«314f«v\«ett.wph
2-37
-------
Weights (5 km) at 50 m
1 4 710131619222528313437404346495255586164677073767982858891
55
52
49
46
43
40
37
34
31
28
25
22
19
16
13
10
7
4
i
^rfl
lEas
•.0.1
p/
—
z
—
Z
—
•*- <
—
—
—
z
—
—
—
Zv
"vl
,1 1
\ —
X
,/
/
1 1
1 1
y
/
-/
^-6
I/
1 1
™""N
(
/
t>
/
^
\ \
H.
Xj
S
/
t
/
f
/
1
/
J
s.
I
{
\
y-
/
/
'
/I
| |
r
^
»
\
\^
0
*
y
9
y
^
^
1 1
\
N^
\
s
k
-^
/
S
$
/
/
VL
\
\
0
N
^
X
y
^
/
Ill/l
1 1
'•x
s
r
N
r
^
/
y
o^
— /
1 1
1 1
L
\
\
\
\
)
f
•
/^
X
.*»
| |
1 1
s
^
\
\
1
/
9
J
/
1 1
1 1
^
o
\
\
e*
1
(
/
1 1
1 r>
i
/
ff
I,
/
1
H
1 1
v-
/>
a*
r
|,
1 1
^
/
||
1 1
l^-
/
,,
i 1
^
/
||
1 1
||
1 1
||
1 1
M
1 1
n
1 1
,i
1 1
i,
1 1
ii
1 1
i,
1 1
i,
1 1
,i
1 1
,i
j_
z
—
—
z
z
z
_
z
—
_
—
z
—
z
55
52
49
46
43
40
37
34
31
28
25
22
19
16
13
10
4
1
1 4 710131619222528313437404346495255586164677073767982858891
Figure 2-8. Terrain weighting factor, W0, for the 50 m level on the 5-km grid.
2-38
-------
Weights (5 km) at 100 m
1 4 710131619222528313437404346495255586164677073767982858891
55
52
49
46
43
40
37
34
31
28
25
22
19
16
13
10
4
1
4 I
»OM
-OJ
I0-1
_
-
Z
-/
^
j.
\S
~x
^£
2—"
?f
1 1
i-^m
t ••_.
»
4
/
/
X
^
_^~
/
II
II 1 | PS
M^
— i"—
y
/
•B.^_
J
/
/ i /
1^,
/
c
^^"
.
/-
II
V
y
b'
^
s
M
Ji
N,
/
?
/
/
/
/
(
II
V>j,
s,
V
&
I
t
/
/
/
/
\v
1 1
^
)
f
/
y
1
jj 1
^
.
^
|
/
/
/
1
J,
3
\
V
\
J
1
1 1
^
\
/
^
f
II
1 1
<-
II
1 1
^.
(
/
II
1 1
3
s
(
/•
ff
e
II
1 1
II
1 1
II
1 1
II
1 1
II
1 1
II
1 1
II
1 1
II
1 1
II
1 1
II
1 1
II
1 1
II
1 1
II
1 1
II
1 1
II
1 1
II
1 1
II
1 1
II
4-
_
H
Z
_
.
_
r
55
52
49
46
43
40
37
34
31
28
25
22
19
16
13
10
7
4
1
1 4 7 10 13 16 19 22 25 28 31 34 37 40 43 46 49 52 55 58 61 64 67 70 73 76 79 82 85 88 91
Figure 2-9. Terrain weighting factor, W0, for the 100 m level on the 5-km grid.
l:\l314\t314KV\MC*2.»pb
2-39
-------
Weights (5 km) at 200 m
1 4 710131619222528313437404346495255586164677073767982858891
4 1
—o
™
-
1
-*
-
™
£~
-/
"
1^.
~
-
-
~
-/
-
11
1 1
.
^
I
^
/
/
/
1!
1 1
^.
f
/
£>
ntf
II
1 1
-\_
f
•»
-P--
J
)
_/
/
II
1 1
/
II
1 1
/
II
\\
"4
/*
II
1 1
v^
\
^
>
r
ii
1 1
\
,
J
;
n
1 1
n
1 1
n
1 1
ii
1 1
n
1 1
n
1 1
ii
1 1
n
1 1
n
1 1
n
1 1
ji
1 1
n
1 1
n
1 1
M
1 1
II
1 1
II
1 1
II
1 1
II
1 1
II
1 1
II
1 1
II
1 1
II
i.
—
—
—
—
—
H
_
—
—
_
_
—
M
-
_
_
±
55
52
49
46
43
40
37
34
31
28
25
22
19
16
13
10
7
1 4 710131619222528313437404346495255586164677073767982858891
Figure 2-10. Terrain weighting factor, W0, for the 200 m level on the 5-km grid.
I:\i314\i3Mwv\ted2 .«pb
2-40
-------
Weights (18 km) at Surface
1 3 5 7 9 11 13 15 17 19 21 23 25 27 29 31 33 35 37
"35
33
31
29
27
-«*
t
\
.**
/
27
\
>
21
19
17
15
13
11
9
7
5
3
1
A
*
/
19
17
15
11
9
7
5
3
1
-OJ
-*
OJO •
1—OJ
- /
\
*
<*.
1 3 5 7 9 11 13 15 17 19 21 23 25 27 29 31 33 35 37
Figure 2-11. Terrain weighting faaor, W0, for the surface level on the 18-km grid.
I:\tJ14\i314iw\ieet2-Wpb
2-41
-------
Weights (18 km) at 50 m
i 3 5 7 9 11 13 15 17 19 21 23 25 27 29 31 33 35 37
35
33
31
27
T^ I
I I
i I
31
27
'•> '
-«no
n
5
19
17
15
13
e
4
L/
/
u
r
ii
1,
1
9
7
*
e
\
s
^L
1
9
7
5
3
OJ
_i%<3
V i
'•Jo
1 3 5
11 13 15 17 19 21 23 25 27 29 31 33 35 37
Figure 2-12. Terrain weighting factor, W0, for the 50 m level on the 18-km grid.
2-42
-------
Weights (18 km) at 100 m
13 5 7 9 11 13 15 17 19 21 23 25 27 29 31 33 35 37
35
33
31
29
27
25
23
21
19
17
15
13
11
1
i
-
-
-
-
-
-
-
- x
x
~ *>"
>s —
\l
I
/
_^
-x
*/
I
^~
s
I
I
/
/ /
I
\
\
s*
/
/
I/
I
/
A
y.
^
\
\
°\
^
/
!
1
^^
N
S y
S f
/
y
^
J
\
1
— --
/'
i
/
0
/
1
1
"**N
/
'
/
y
i
i
N
\
)
a'
1
1
1
1
1
1
1
1
1
1
1
1
1
1
I
1
!
i
-
-
-
-
-
-
-
-
-
-
-
-
35
33
31
29
27
25
23
21
19
17
15
13
11
9
7
5
3
1
1 3 5 7 9 11 13 15 17 19 21 23 25 27 29 31 33 35 37
Figure 2-13. Terrain weighting factor, Wc, for the 100 m level on the 18-km grid.
2-43
-------
Weights (18 km) at 200 m
1 3 5 7 9 11 13 15 17 19 21 23 25 27 29 31 33 35 37
35
33
31
29
27
25
23
21
19
17
15
13
11
35
33
31
29
27
25
23
21
i:
1!
1.
1
9
7
r7
vt
f
\
(S i
I !
1 3 5 7 9 11 13 15 17 19 21 23 25 27 29 31 33 35 37
Figure 2-14. Terrain weighting factor, W0, for the 200 m level on the 18-km grid.
I:\«3M\«314«v\ieett.«|>h
2-44
-------
Weights (54 km) at Surface
30
12345678 910111213141516171819202122232425262728293031323334
12345678 910111213141516171819202122232425262728293031323334
Figure 2-15. Terrain weighting factor, W0, for the surface level on the 54-km grid.
LA«314\«314rev\iect2.wph
2-45
-------
Weights (54 km) at 50 m
123456789 1011 12 1314 151617 18192021 22232425262728293031 323334
29
28
19
18
17
16
15
14
13
12
11
10
9
8
6
5
4
3
2
1
30
28
27
23
21
20
18
17
16
14
12
11
10
9
8
1
1234567 8-910111213141516171819202122232425262728293031323334
Figure 2-16. Terrain weighting factor, W0, for the 50 m level on the 54-km grid.
l:\«314\«314rev\MCt2.wpb
2-46
-------
Weights (54 km) at 100 m
30
29
28
27
26
25
24
23
22
21
20
19
18
17
16
15
14
13
12
1 1
10
9
8
7
6
5
4
3
2
i
r//,
//
y
(
^
? I N
/
006
y
\
\
r-
/
/
y
—^
^^
N
I
/
9
,
1
s
/
^
>
-~_
^0;
T
o
>y
/
>
^ '
\ ^
\
)
"•x
f
k
V
V
ft
l(
\\
s
MI
S
\\
\\i
ill
\
8
I
L—
I
~v
/
\
r^
\
)
\
*-**>
1
JU
29
28
27
26
25
24
23
22
21
20
19
18
17
16
15
14
13
12
n
10
9
8
7
6
5
4
3
2
1
12345678 910111213141516171819202122232425262728293031323334
Figure 2-17. Terrain weighting factor, W0, for the 100 m level on the 54-km grid.
I:\i314\i3M«v\ied2.wph
2-47
-------
using the observed wind data. Thus, in this mode, the coarse-grid winds are adjusted for the
fine-scale terrain effects and observations.
In the second option, the coarse-grid winds are interpolated to the CALMET grid and
then are used as the "Step 1" field. Thus, the coarse-grid winds are not adjusted for the fine-
scale terrain effects, but rather they are assumed to already contain the most significant terrain
effects. The "Step 1" winds are combined with observations using an objective analysis
procedure to produce the final "Step 2" winds.
In the third case, the coarse-grid winds are treated in exactly the same manner as the
observations. If the diagnostic wind option is used in CALMET, a "Step 1" wind field is
produced by adjusting the domain-scale wind for the fine-scale terrain effects. The actual
observations and MM4-FDDA "pseudo-observations" are then used to modify the Step 1 fields
using the objective analysis procedure. If the "objective-analysis-only" option is selected in
CALMET, the computation of the Step 1 wind field is eliminated, and the final winds are based
on the objective analysis of the MM4-FDDA winds and the actual observational data. Note that
in this case, both the observations and MM4-FDDA winds are given a high weight in the
analysis procedure.
In the sensitivity analysis, all three methods of incorporating the MM4-FDDA field into
CALMET were examined. The weighing factor, W0, discussed above, was applied as follows:
• MM4-FDDA wind as initial guess wind
- no weighting by W0
• MM4-FDDA used as Step 1 winds
- W0 is used to weight observations
- Step 1 winds are weighted by factor (1.0 - W0)
• MM4-FDDA used as "observations"
- W0 is used to weight actual observed data
- MM4-FDDA data are weighted by factor (1.0 - W0)
In the first case, the terrain-weighting factor is not used because the MM4-FDDA
coarse-grid winds are subject to the full adjustment for the fine-scale terrain data by the
diagnostic model, whereas in the other two cases, the MM4-FDDA winds are not adjusted for
the effects of the fine-scale terrain.
l\i3M\»314rev\«ect2.wpb 2-48
-------
In NUATMOS, an initial guess field is produced by an objective analysis of observational
data. MM4-FDDA winds wfll be incorporated as additional observations in the computation of
the initial guess field. Two sets of sensitivity tests were conducted:
• MM4-FDDA data and observations given equal weight at all levels.
• MM4-FDDA data weighted by factor (1.0 - W0), observations weighted by W0.
25 Statistical Measures
Two sets of statistical measures were used to assess the similarities between a reference
or base wind field (defined as the 18-km or 54-km grid fields produced by the nested grid runs
of MM4-FDDA) and the wind fields produced by the CALMET or NUATMOS diagnostic wind
field models. As described in Section 23 and summarized in Table 2-3, some of the diagnostic
model runs used 80-km MM4-FDDA winds as input "data" to supplement the winds observations
and other runs used only observational data as input. None of the simulations used the nested
grid MM4-FDDA winds at 18-km or 54-km to drive the diagnostic models. These data were
reserved for use as the reference fields.
In Section 2.5.1, a trajectory analysis technique is described which was used to evaluate
the effects of differences between the reference nested grid MM4-FDDA wind fields and the
diagnostic model wind fields. This scheme provides an indication of the effect of the differences
in the wind fields on plume trajectories. In Section 2.5.2, statistical measures are described
which are direct measures of the wind fields on point-by-point basis.
2.5.1 Trajectory Analysis
Trajectories were computed from four release locations at three levels for each of the
model simulations. A statistical analysis was conducted on the trajectories to assess the effects
the type and amount of input data and method in which the data was used had on the quality of
the wind fields produced by the diagnostic models. The statistical analysis was performed only
for the 18 km and 54 km grids, so that the MM4-FDDA nested grid data could be used to
provide benchmark winds against which to compare the performance of the CALMET and
NUATMOS techniques.
The trajectory release locations were as follows:
I:\t3\4\l3l4m\ttd2.vpl 2-49
-------
5km Grid
Richmond, VA
Baltimore, MD
Norfolk, VA
Luray, VA
18 km Grid
Baltimore, MD
Pittsburgh, PA
Richmond, VA
Charleston, WV
54 km Grid
Chicago, IL
Pittsburgh, PA
Baltimore, MD
Knoxville, TN
Trajectories were released at 10 m, 200 m, and 400 m every 4 hours, 6 hours, and 12
hours from the beginning of the simulation up to 24 hours before the end of the simulation for
the 5 km, 18 km, and 54 km grids, respectively.
Three statistical expressions were used to evaluate the diagnostic model simulations. A
root-mean-square rate of separation was computed to express a verification score in terms of an
average absolute error. The RMSR was computed using the approach of Haagenson et al.
(1990),
RMSR
Tt
1/2
(2-77)
where, N
i
and
M' *M
> Y
number of time steps,
current time step,
coordinates of model trajectory (derived from CALMET or
NUATMOS winds),
coordinates of reference trajectory (derived from MM4 winds),
elapsed time.
The second statistic used was the mean rate of separation (MRS) of the trajectories.
The following equation was used to calculate the MRS:
MRS
AT,
(2-78)
L\»314\«314rev\iectt.wph
2-50
-------
where, AX, = (XJ, -
AYf -Oto-
f * final time step,
AT, - elapsed time at final time step,
XM»YM • coordinates of reference model trajectory (derived
from CALMET or NUATMOS winds), and
XR,YR = coordinates of reference trajectory (derived from MM4 winds).
The mean fractional separation (MFS) was also examined to represent a measure of the
separation of the trajectories at the final time step relative to the distance to the trajectory
release location. This analysis is independent of trajectory behavior at intermediate time steps,
therefore its value will not be biased due to high wind speed or light wind speed conditions.
The MFS is represented as:
MFS,
I |PH - *J * <(H - r.fl* + [((H - *.f * (w, - r.ff]
(2-79)
where, a) the numerator represents the distance between the trajectories (e.g.,
CALMET and MM4) at the final time step
b) the denominator represents the average of the distance from the final
trajectory location to the release location
and
AX, = (XM)f - (XR)f
AY, = (YM), - (YR),
f = final time step,
XM>YM = coordinates of model trajectories (derived from CALMET or
NUATMOS winds),
XR
-------
2.5.2 Point-by-Point Comparisons
Although the trajectory analysis discussed above provides a useful method of relating the
effects of differences in wind fields on a modeling result (Le., the plume trajectory), it only
allows the sampling of winds at a very small fraction of the total number of grid points in the
simulations. Therefore, the trajectory statistics can be quite sensitive to the necessarily arbitrary
selection of the trajectory release points and release times. Also, the trajectories accumulate
errors in such a way that small deviations in the winds near the release points can be amplified
into large deviations in the final trajectory end points.
A useful extension to the trajectory analysis which provides a direct and robust method
for quantifying the differences in the reference and diagnostic winds is to perform a statistical
analysis of the grid point-by-grid point winds for the entire domain and simulation period. Thus,
the cumulative statistics are direct comparisons of the diagnostic winds with the reference winds
at nearly 2 million data points.
The statistics listed below were computed for the u component, the v component, and
the set of u and v components combined. These statistics are discussed by Pielke (1984), who
used them as a quantitative measure of model skill. Since each model (CALMET, NUATMOS,
and MM4-FDDA) has a different grid structure, the diagnostic winds were first interpolated to
the MM4-FDDA model layers. This interpolation was limited to those MM4-FDDA layers at or
below the highest CALMET layer (approximately 1240 m).
1) x^ - Mean value of the MM4-FDDA wind field variable x
(i.e., data at an 18 km or 54 km grid resolution).
2) o^ - Standard deviation of the MM4-FDDA wind field variable x.
3) x^ Mean value of the diagnostic wind field variable x (i.e.,
CALMET or NUATMOS predictions for the 18 km or 54 km grids).
4) 0,4 - Standard deviation of the diagnostic wind field variable x.
5) r(Xp, xj - correlation coefficient
6) RMSE (Xp, x,,) - root mean square error .
I:\«314\«314rev\ied2.wpb 2-52
-------
7) RMSE*, (Xp, Xj) - RMSE with mean bias removed
where x can be u, v, or the set of u and v combined,
p refers to the MM4-FDDA model, and
d refers to the diagnostic model (CALMET or NUATMOS).
The u component of the wind is the east-west component, v is the north-south component, and
the combination of u and v refers to the concatenation of the u and v time series (each of length
N) into one time series of length 2N.
The following subsets of data were considered:
1) surface layer only,
2) upper layers combined (usually ~ 7 levels),
3) all layers
The 7 statistics above were calculated for each hour. The same statistics for all the
hours in the simulation (120 hours in all) were derived from the hourly values. In addition to
the statistical measures, scatter plots of selected variables were also produced at aid in the
analysis.
2.6 Meteorological Sensitivity Tests
2.6.1 CALMET
The initial set of simulations with CALMET (Runs 1-3) focused on the sensitivity of the
mesoscale wind fields produced by the model to the use of a single domain mean wind as the
initial guess field. As described in Section 2.1, the original version of the DWM in CALMET
used a domain mean wind as a spatially constant initial guess field, then modified the winds for
local terrain influences to produce a "Step 1" wind field. The Step 1 field is modified within the
radius of influence of observational stations to reflect the winds at the stations. However, in
areas of the grid outside of the region of influence of observational stations, the final wind field
is determined primarily by the modified initial guess (Step 1) field.
I:\«3M\«314rev\«eet2.wph 2-53
-------
In order to test the ability of the domain scale wind to represent flow fields over regional
domains, CALMET was run using relatively small values of the radius of influence parameters
(Rt = 50 km, R2 « 100 km, RMAX1 = 100 km, RMAX2 = 200 km). R, and R2 are the
distances at which the Step 1 wind and observations have equal weight in the surface layer and
upper layers, respectively. RMAX1 and RMAX2 define the maximum radius of influence of
observations in the surface and upper levels, respectively. In all the simulations, the overwater
radius of influence, RMAX3, was set to 1000 km. The CALMET vector plot for 54-km grid at
the 200 m level is shown in Figure 2-18(a).
The strong effect of the observational data in modifying the Step 1 wind field is apparent
as circular regions of distinctly different winds surrounding several of the meteorological
stations. Outside the region of influence of the stations, the winds approach the southerly
domain scale wind. In this run, the Wallops Island upper air station was selected to define the
domain scale flow field.
A second set of runs of CALMET with larger values of the radius parameters were
conducted to reduce the influence of the domain mean wind and increase the influence of the
observations. Values of Rt = 500 km, R2 = 1000 km, RMAX1 = 500 km, and RMAX2 - 1000
km were used. The results for the 54 km grid at the 200 m level shown in Figure 2-18(b), are
significantly different, especially in the northern half of the grid. The sharp differences around
meteorological stations have been eliminated. The southerly domain scale flow no longer
dominates the northern portion of the grid, but instead, the flow is primarily from the west or
southwest.
For comparison, the 80-km MM4-FDDA winds are shown in Figure 2-18(c). The second
CALMET run with the reduced influence of the domain scale wind, provides a better match to
the MM4-FDDA fields.
A second set of vector plots are presented in Figures 2-19 (a,b, and c) for August 6, 1988
at the 400 m level. Similar conclusions can be drawn regarding the nonrepresentativeness of the
domain scale wind. The radius of influence can be clearly seen in Figure 2-19(a) around many
of the meteorological stations, and the regions influenced by the relatively weak southerly
domain scale flow are apparent. In Figure 2-19(b), the fields derived using the larger radius of
influence parameters provide a better description of the regional flow than the terrain-adjusted
domain mean field which dominates in Figure 2-19(a). Strong westerly flow in Ohio, Michigan,
and north into Ontario and strong southwesterly flow in New England, New York, and Quebec
can be seen in the MM4-FDDA winds (Figure 2-19(c)) and the CALMET fields in Figure
2-19(b), but not in the CALMET "small" radius of influence run (Figure 2-19(a)).
L\*314\a314rav\wel2.«pb 2-54
-------
30 -
28
\ \ N\ V «. ^ ^>«_ >.•.».>> •» ^.
! ! I \ I ! I • \ . I I I I I
Figure 2-18(b).
Vector plot of CALMET winds. Level 4 (200 m), 54-km grid. August 4,
1988 (6:00 am EST). Uses "large" radius of influence parameters.
Observations only.
I:\«314\«314«»\ieel2.»ph
2-56
-------
\ t t i \ \ \ r r
I 1 t t 1 \ \ f T
r t J t 1 t
x / 1 t M
\iiit;
^Figure 2-18(a).
Vector plot of CALMET winds. Level 4 (200 m), 54-km grid. August 4,
1988 (6:00 am EST). Uses "small* radius of influence parameters.
Observations only.
I:\«314\i3M«v\iee«2.wpb
2-55
-------
Figure 2-18(c). Vector plot of 80-km grid MM4-FDDA winds. Interpolated to 200 m level
August 4, 1988 (6:00 am EST).
I:\«314\»314i«v\i«awph
2-57
-------
\\A\lltI1\
\\\\\\\\\\\\
\\\1 \ttt\\\\
1t////f\
/ \ \ \ \ \ 1 1 f I
r\ i i t-^. i i \ \ v ^
\ > i . . N •
Figure 2-19(a).
Vector plot of CALMET winds. Level 5 (400 m), 54-km grid. August 6,
1988 (6:00 am EST). Uses "small" radius of influence parameters.
Observations only.
I:\i3M\«3M«v\ieel2.wpb
2-58
-------
\ I
\ \ I
\ \ \
\ \
\ \ \ \ \ I
\ \ \ \ \ \ .
\ \ \ \ N \ \ \ 1
\\\\\\\\\1t.
\\\\\\\\ \ 1.
\\\N\\\\\\.
7 / X / / /• L.
. i i \ i i i i .\ . i i i i i
Figure 2-19(b).
Vector plot of CALMET winds. Level 5 (400 m), 54-km grid. August 6.
1988 (6:00 am EST). Uses "large" radius of influence parameters.
Observations only.
I:\«3M\i314rev\§eel2.wph
2-59
-------
80KM GRID-flUG
8821S06
0
Figure 2-19(c). Vector plot of 80-km grid MM4-FDDA winds. Interpolated to 400 m level.
August 6, 1988 (6:00 am EST).
IA«314\*314icv\iecl2.«ph
2-60
-------
Figure 2-20 is a vector plot of the CALMET winds on August 6 at the 400 m level on
the smaller 18-km grid domain. The "small" radius of influence parameters were used in this
run. The 80-km grid MM4-FDDA winds for the same period are shown in Figure 2-19(c).
Even over the smaller domain, the terrain-adjusted domain mean winds are not able to
reproduce the spatial variability of the flow field seen in the MM4-FDDA winds. For example,
the CALMET simulation produced southerly flow in western Virginia and North Carolina,
whereas the MM4 fields indicate westerly flow.
These results led to the introduction of an option within CALMET to allow spatially-
variable winds as the initial guess field. This feature is considered important for the application
of the model to regional scale modeling problems. The spatially-variable initial guess field in the
enhanced version of CALMET can be determined from a 1/r2 weighting of observational data,
or externally-specified gridded wind fields (e.g., from an MM4-FDDA simulation).
In the second group of CALMET simulations (Runs 4-6), the 80 km MM4-FDDA wind
data were input to the model as the initial guess field. The 80-km MM4-FDDA winds were
interpolated to the fine-scale CALMET grid and diagnostic adjustments due to the fine-scale
terrain were made. A resulting vector plot for the 54 km grid at 200 m is presented in Figure 2-
21(a). Figure 2-21(b) contains a 54 km grid wind field derived by CALMET (Run 12) using the
weighted 80 km MM4-FDDA data as a step 1 field. Figure 2-2 l(c) presents a vector plot of the
54 km grid wind field derived from using weighted 80 km MM4-FDDA wind data as
observations in addition to the meteorological station observations to drive CALMET (Run 15).
For comparison purposes, the nested 54 km grid MM4-FDDA wind field is presented in Figure
2-21(d). It can be seen that the wind fields derived by CALMET using the 80 km MM4-FDDA
data as either a step 1 field or as observation produce similar flow fields to the nested 54 km
grid MM4-FDDA wind field. The CALMET simulation which used the 80 km MM4-FDDA as
the initial guess field produced westerly winds over Wisconsin, where the other simulations
produced southwesterly winds as did the MM4 model.
The CALMET wind fields generated for the 18 km grid at 400 m with the 80 km MM4-
FDDA data used as a model input are presented in Figures 2-22(a-c). The nested 18 km grid
MM4-FDDA wind vectors interpolated to 400 m are plotted in Figure 2-22(d). As was true on
the larger grid scale, the 18 km grid results also show that the use of the 80 km MM4-FDDA as
the step 1 field or as observations reproduces the flow field of the nested grid MM4-FDDA data
more closely than the simulation which used the 80 km MM4-FDDA data as the initial guess
field. The flow field generated through central Virginia using the MM4 data as the initial guess
field in CALMET differs noticeably from the flow fields of the nested MM4 data and the other
CALMET simulations in that region of the grid.
I:\t314\i314rev\tect2.wph 2-61
-------
lllltV\\V\\\\\\\\\\
I
ilt1\\\\\\\\\\\\A\\
I I I , t . I . I I , t IM
0
Figure 2-20. Vector plot of CALMET winds. Level 5 (400 m), 18-km grid. August 6,1988
(6:00 am EST). Uses "small" radius of influence parameters. Observations only.
I:\i314\«314rev\teeawph
2-62
-------
, V «V V V \ X I ^. X. .. _V— _ _ _
,.'-.. \ , I . , \ , ' .
Figure 2-21(a).
Vector plot of CALMET winds. Level 4 (200 m), 54 km grid. August 4.
1988 (6:00 am EST). Uses 80 km MM4-FDDA data as the initial guess
field.
2-63
-------
2 p-
Figure 2-21(b).
Vector plot of CALMET winds. Level 4 (200 m), 54 km grid. August 4,
1988 (6:00 am EST). Uses 80 km MM4-FDDA data as the Step 1 field.
I:\*314\a314icv\iect2.wph
2-64
-------
Figure 2-21(c). Vector plot of CALMET winds. Level 4 (200 m), 54 km grid. August 4, 1988
(6:00 am EST). Uses 80 km MM4-FDDA data as observations.
l:\«314\i314iw\iecU.wpb
2-65
-------
\ \ \ sl«. s v vs^ — » «» x v s
i , ,\ . i , , \ . i , . , ,
Figure 2-21(d).
Vector plot of nested 54 km grid MM4-FDDA winds. Interpolated to the
200 m level. August 4, 1988 (6:00 am EST).
2-66
-------
35
30
25
20
15 -
10
5 r-'
'/SS///
xx/mm ///>?/
Y/////:mw:L
. t
f ft M r M r
/ / M f I M f
' ' / M I M M
M r i t t i
II M M \
\LM M M \
M \
\ \ \ \ \ \ \ \ \ \ \ \ \ \
\\\\\\\\\\\\\\\\
\\\\\\\\\\\\\\\
\\\\\\\\\\\\\\\\\\\
\ \\\\\\\\\\\\\\\s\\\\\
\ \ \ \ \\\\\\\\\\\\\\\\\\\\
\\\\\\\\\\\\\\s\ \\vk^v*rrrir\ \ \ \ \ \
\\\\\\\\\\s \j^w-»r-rrCC\\ \\N\\\\\
S\\\\\\\\\\\\N\
MM
\
.SVVvs.vVXV.XNN
VXN
/////,
f f f f t.
i i t r t.
t M r,
\ M t i.
M t t I,
\ \ \ \ \.
\ \ \ \ \ \,
\\\\\\.
\ \ \\\ \ \.
\\\\\\\.
\\\\\\\,
\\\\\\\\.
\N\N\\\
\ \ \ \ \ \ \.
\ \ \ \ \ \
\ \ \ \ \ \
\ \ \ \ \ ,
\ \ N \ .
\ \ \ \.
\\N\\
\ \ \ \ \
\ \ \ \ \ \
N\\\\\\\
0
15
25
35
Figure 2-22(a).
Vector plot of CALMET winds. Level 5 (400 m), 18 km grid. August 4,
1988 (6:00 am EST). Uses "large" radius of influence parameters and 80
km MM4-FDDA data as the initial guess field.
2-67
-------
war/././/
i t
! t M
t ! It \
t t M \ \
M \ \
\ \ \
\ \ \ \ \ .
\ \ \ \ \A \ \ \ \ \ \ \ \ \ \ \ \ \
\\\\Y\\\N\\\\\\\\\\
\\\ \X\\N\\\\\\\\\\\\\
\\N\\\\\\\\\\\\
\\\\\\\\\\\\\\\\
\\\\\\\\\\\\\\\\
\\\\\\\\\\\\\\NSX\V\
\\N\\\VNVNV
\ \ \ \ \ \ .
\\\\\\ .
\\\\\\_
\\\\\\ .
\\N\\NN .
10
5 -'
0
0
35
Figure 2-22(b).
Vector plot of CALMET winds. Level 5 (400 m), 18 km grid. August 4,
1988 (6:00 am EST). Uses "large" radius of influence parameters and
80 km MM4-FDDA data as the Step 1 field.
I:\»314\«314t«v\ieel2.wph
2-68
-------
i
i t
\ \ \ \ \
\\\\
t M \\
M
M I \ \
\ \ \ \
\\\\\\\\ . .
\\ \\\ \ \\\ \ .
\\\\\\\\\\\\
\\\\\\\\\\\\\\ \XAJ\\
\\\\\\\\\\\\\\\\X\
\\N\\\\\\N\\\\\\
\\\\\\\ \ \ \N\\\\\\
\N\\\\v. vsvs
35
Figure 2-22(c). Vector plot of CALMET winds. Level 5 (400 m), 18 km grid. August 4, 1988
(6:00 am EST). Uses "large" radius of influence parameters and 80 km MM4-
FDDA data as observations.
2-69
-------
'77777
fit ft
tit ft
Jill I
if
T7
\ \ \ \ \.
. k .\\Mt\M\
\ \ \ \ \ \ \ \ \ \ \ \ \
%. \\\\\\\\\\\\\\
\\\\\\\\N\\\\\\
\\\\\\\\\\\\\\\
\\N\N\\\\S\\\\S\
\\\s
\\\\v
25
Figure 2-22(d).
Vector plot of nested 18 km grid MM4-FDDA winds. Interpolated to the
400 m level. August 4, 1988 (6:00 am EST).
I:\aM4\a314rev\Md2.wph
2-70
-------
Trajectories were computed from the wind fields generated by CALMET and the wind
fields derived by MM4-FDDA for the fine scale grids. Figures 2-23(a-d) show the trajectories at
an elevation of 10 m on the 54 km grid for each of the four scenarios of CALMET runs
(observations only, 80 km MM4 data as initial guess field, as Step 1 field, and as "observations")
and the nested 54 km grid MM4-FDDA trajectories released from four locations on the grid.
The CALMET run made using station observations only (Figure 2-23(a)) dearly does not
represent the flow of the nested MM4 data in the area of Knoxville, TN. Introduction of the 80
km MM4-FDDA data as model input improves the trajectory results.
Other Meteorological Parameters
In addition to wind fields, CALMET produces gridded fields of other meteorological
variables such as mixing heights (zj), surface friction velocities (u.), Monin-Obukhov lengths (L),
convective velocity scales (w.), and stability classes. These variables are used by the CALPUFF
dispersion model to characterize turbulence and dispersion.
Time series plots of selected variables were prepared to determine the reasonableness of
the CALMET predictions for these fields. One such set of plots for the Zj, u., w., and stability
class fields are presented in Figures 2-24 to 2-27, respectively. These plots contain the values of
the parameters at grid point (10,12) on the 54-km grid for the period from August 1
through August 6, 1988. This grid point was selected because it corresponds to Dayton, Ohio
which has an upper air station with a fairly complete record of data during the episode.
The predicted mixing height pattern shows a strong diurnal cycle, with peak daytime
(convective) mixing heights in the range from 1600-1800 m, except for August 1 fe ~ 1300 m).
The estimated nighttime mixing heights are typically much lower, from one to several hundred
meters. The higher nighttime mixing height on August 4-5 corresponds to a period with higher
winds, and therefore higher mechanical mixing heights.
The surface friction velocity is strongly related to the surface wind speeds. Typical
estimated values of u. in the range of 0.5-1.0 m/s are produced by CALMET (see Figure 2-25).
Lower nighttime values of u. are estimated, corresponding to periods of lighter winds.
The convective velocity scale (w. « (H^)1^, where H0 is the surface sensible heat flux),
is plotted in Figure 2-26. Typical daytime values of order 2 m/s are estimated by the model.
These values are reasonable for a summer period with substantial solar radiation.
I:\t3U\i314rev\iect2.wph 2-71
-------
99954
NORTH
810
1242
1674
459-
ui
u
-81-
-621
999
-54
-459
--81
378
810
SOUTH
1242
1674
-62
Figure 2-23(a).
Trajectories at 10 m on 54 km grid produced by CALMET (large, using
observations only) and 54 km MM4 winds. Starting date August 1, 1988,
7 am. Small numbers indicate hours after release.
I:\«314\i314«v\iee«2.*ph
2-72
-------
999
378
NORTH
810
1242
1674
459-
oo
LJ
s I i i i i i i i j i i i i
999
-81-
-459
to
<
LJ
--81
378
810
SOUTH
1242
1674
-621
Figure 2-23(b).
Trajectories at 10 m on 54 km grid produced by CALMET (large, using
80 km MM4 winds as initial guess) and 54 km MM4 winds. Starting date
August 1, 1988, 7 am. Small numbers indicate hours after release.
I:\«314\«314rev\tect2.wpfc
2-73
-------
NORTH
810
1242
1674
u
' '.ij .U '''i ill J-621
-81-
-621
-54
378
810
SOUTH
1242
Figure 2-23(c). Trajectories at 10 m on 54 km grid produced by CALMET (large, using 80 km
MM4 winds as step 1 field) and 54 km MM4 winds. Starting date August 1,
1988, 7 am. Small numbers indicate hours after release.
I:\t314\«314icv\Kd2.«pb
2-74
-------
378
NORTH
810
1242
1674
oo
LJ
999
-459
to
<
UJ
--81
378
810
SOUTH
1242
1674
-621
Figure 2-23(d).
Trajectories at 10 m on 54 km grid produced by CALMET (large, using
80 km MM4 winds as observations) and 54 km MM4 winds. Starting date
August 1, 1988, 7 am. Small numbers indicate hours after release.
l:\i3M\»314rev\i«et2.wph
2-75
-------
AUGUST 1-6 1988 GRID POINT (10.121
12
36
I 72 84
TIME (HOURS)
96
108
12B
Figure 2-24. Time series plot of CALMET mixing heights for grid point (10,12) on the 54-km
IWAQM grid (Dayton, OH). Time period covered is August 1-6,1988.
I:\«314\«314rev\ied2.*pb
2-76
-------
z. we
RUGUST 1-6 1988 GRID POINT 110.123
-i 1 1—' P
5
i
72 84
TIME (HOURS)
120 132 144
Figure 2-25. Time series plot of CALMET friction velocity (u*) for grid point (10,12) on the
54-km IWAQM grid (Dayton, OH). Time period covered is August 1-6,1988.
. I:\«314\a314rev\nd2.wph
2-77
-------
flUGUST 1-6 1988 GRID POINT (10.12]
3.1
2.000
1.500
l.i
0.500
12 24
48 68 72 84
TIIC (HOURS)
96
IBB 120 132 144
Figure 2-26. Time series plot of CALMET convective velocity scale (w*) for grid point (10,12)
on the 54-km IWAQM grid (Dayton, OH). Time period covered is August 1-6,
1988.
I:\a314\t314mr\ieelZwpb
2-78
-------
RUGUST 1-6 1988 GRID POINT (10.12)
Ill I I I I I I I I
36
68 72 84
TDC (HOURS!
96
108 120 132 144
Figure 2-27. Time series plot of CALMET stability dass (1 « A stability, 2 « B stability,
3 = C stability, 4 « D stability, 5 « E stability, 6 - F stability) for grid point
(10,12) on the 54-km IWAQM grid (Dayton, OH). Time period covered is
August 1-6, 1988.
I:\i314\i314nv\iect2.wph
2-79
-------
Finally, the stability class time series is plotted in Figure 2-27. The diurnal variation is
consistent with a summer period with generally light winds and strong solar radiation (F stability
at night, A stability near noon). The period of stronger winds during August 4-5 is evident by
the sequence of neutral stability classes during this period, which is consistent with the behavior
of u. and z, shown in the plots.
2.6.2 NUATMOS
The NUATMOS model was applied using three model input scenarios. First, the model
was run using surface and upper air station observations available on each grid. Then the 80 km
MM4-FDDA wind data was introduced as additional model input. The MM4-FDDA winds
were either used directly by NUATMOS or they were weighted by terrain factors (see
Section 2.4.1) when used by NUATMOS. Specifics on the NUATMOS modeling exercise
performed by the National Forest Service can be found in Appendix A of this volume.
Figures 2-28 (a-c) show vector plots of the wind fields generated by NUATMOS at 200 m on the
54 km grid for each of the modeling scenarios. The introduction of the 80 km MM4-FDDA
wind data affects the flow field produced, especially in the southeast region of the grid. The
resultant flow fields, Figures 2-28(b and c), represent the winds in this region as is seen in the
nested 54 km grid MM4-FDDA wind field (Figure 2-21(d)). The same relationship can be seen
in the 18 km grid results (Figures 2-29(a-c)).
2.6.3 Statistical Results
Trajectory statistics were computed from each release time for each site and three levels.
Final statistics were computed for each trajectory including all release times, all four sites and
all three levels. The final statistics for comparing CALMET trajectories and the trajectories
derived from the nested MM4-FDDA data for both the 18 km grid and 54 km grid are
presented in Table 2-4. The three statistics agree qualitatively, that is, a general trend is
observed. The radius of influence parameters are shown to be important only when running
CALMET with surface and upper air observations only and then only for the 54-km grid
comparisons. When the MM4-FDDA data are used as model inputs, the choice of the radius of
influence appears to statistically have no effect on the model results. Significant improvement in
model performance occurs when the MM4-FDDA data are used as either a step 1 field or as
observations in CALMET.
Table 2-5 represents the summary statistics for comparing NUATMOS trajectories and
the nested grid MM4 data for the 18 km and 54 km grids. The use of the MM4-FDDA data as
I:\i314\«314rev\wet2.wph 2-80
-------
„ )X t \ \ \ I \ V \ s
"~T ' * V \ < \ N «. _\
1
Figure 2-28(a). Vector plot of NUATMOS winds. Level 4 (200 m), 54 km grid. August
4, 1988 (6:00 am EST). Observations only.
I:\*314\*3Miev\Mct2.*pfe
2-81
-------
//'XXXX.XX'X'X'
S/SSS/////S/
/ X X X" X /
\ \ \ \ \
\ \ \ \
\ \ \ \ \ \ \ V V
V \ \ \ \ \ \ \ V x
\ \ \ \ \ \ \ \ v
\\VW\\\ \,l\*N *^^ "" **"
t I • , \ , t . . \ , '
Figure 2-28(b).
Vector plot of NUATMOS winds. Level 4 (200 m), 54 km grid. August
4, 1988 (6:00 am EST). Uses 80 km MM4-FDDA data as input.
I:\«314\i314RV\MCt2.*ph
2-82
-------
Figure 2-28(c). Vector plot of NUATMOS winds. Level 4 (200 m), 54 km grid. August 4,1988
(6:00 am EST). Uses terrain weighted 80 km MM4-FDDA data as input.
2-83
-------
-77T/7V77/7•7/77771
• 777//T/7777777771 /
• 777/777777777/7'/ ' '
''
\ \ \ \ \
\N\\S
\N\\\
\ \ \ \ \
\ \ \ \ \
\\N\\
\ \ \ \ \
N \ \ \ \
\ \ \ \ \
\ \ \ \
\\
\\\\\\\\\
\\^\\\^\\\
\\\\\\\\\sss\\
\\\\\\\\
\v\\\\\xv
\\ \ \
S \ \ \ \ \
\ \ \ \
0
35
Figure 2-29(a). Vector plot of NUATMOS winds. Level 5 (400 m), 54 km grid. August
4, 1988 (6:00 am EST). Observations only.
l:\»3M\»3Mi«Aied2.wj*
2-84
-------
35
Figure 2-29(b).
l\i3M\i3Mi*v\ieca.wph
Vector plot of NUATMOS winds. Level 5 (400 m), 54 km grid. August
4, 1988 (6:00 am EST). Uses 80 km MM4-FDDA data as input.
2-85
-------
35
30 -
m
in
\
\
\
\
\
\ \ \ \\\ .
\\N\\\ .
\\\\\\\_
\\\\\\\ .
\\\\\\\ .
\\NNN\\\ .
\N\N\\\ .
\x,\NXS\_
XXNXXXXXXXXXXX
V VXXVXXXXVXXXN
25
30
35
Figure 2-29(c). Vector plot of CALMET winds. Level 5 (400 m), 54 km grid. August 4, 1988
(6:00 am EST). Uses terrain weighted 80 km MM4-FDDA data as input.
2-86
-------
Table 2-4
Statistical Summary of Summer Episode
CALMET vs. Nested Grid MM4-FDDA Trajectories Combined Over
Three Levels and All Sites
18 km Grid
Observations Only
(CALMET Run 2)
With 80 km MM4-FDDA as
Initial Guess Field
(CALMET Run 5)
With 80 km MM4-FDDA as
Step 1 Field
(CALMET Run 11)
With 80 km MM4-FDDA as
"Observation"
(CALMET Run 14)
54 km Grid
Observations Only
(CALMET Run 3)
With 80 km MM4-FDDA as
Initial Guess Field
(CALMET Run 6)
With 80 km MM4-FDDA as
Step 1 Field
(CALMET Run 12)
With 80 km MM4-FDDA as
"Observations"
(CALMET Run 15)
Root-Mean-Square Rate
of Separation (km/day)
Small R Lf ryg R
164 175
126 122
91 91
98 97
Small R Large R
260 159
116 119
110 111
113 113
Mean Rate of
Separation (km/hr)
Small R IitITt R
63 73
5.0 4.9
3.9 3.9
43 43
• Small R Large R
10.9 6.5
43 4.4
4.9 4.9
5.0 5.0
Mean Fractional
Separation
smaii B Irirrr F
0.61 0.73
0.45 0.46
037 037
0.43 0.42
Small R Large R
0.68 0.49
029 030
028 028
029 029
IA«314\a314rev\KCt2.wph
2-87
-------
Table 2-5
Statistical Summary of Summer Episode
NUATMOS vs. Nested Grid MM4-FDDA Trajectories Combined
Over Three Levels and All Sites
18 ton Grid
Observations Only
(NUATMOS Run 2)
With 80 km MM4-FDDA
(NUATMOS Run 5)
With a Weighted 80 km
MM4-FDDA (NUATMOS
Run 8)
54 km Grid
Observations Only
(NUATMOS Run 3)
With 80 km MM4-FDDA
(NUATMOS Run 6)
With a Weighted 80 km
MM4-FDDA (NUATMOS
Run 9)
Root-Mean-Square Rate
of Separation fkm/dav')
128
95
97
153
127
127
Mean Rate of
Separation fkm/hrt
5.6
43
6.6
5.4
S3
Mean Fractional
Separation
0.58
0.42
0.42
0.43
032
031
I:\t314\»314rw\i«t2.wph
2-88
-------
model input improves the performance of NUATMOS, however the terrain weighting of the
MM4 data does not have a significant effect on the model results.
Both the CALMET and NUATMOS diagnostic models perform similarly. Each shows
improvements in the model results when MM4-FDDA data are incorporated as input into the
models. CALMET performed best on the 18 km grid when the MM4-FDDA data were input as
a Step 1 field with a RMSR of 91 and similarly, the best RMSR for NUATMOS, 95, occurred
when the MM4-FDDA data were used as model input These two values are sufficiently close
to suggest similar model performance. On the 54 km grid, the same scenarios produced the best
performance, RMSR - 110 for CALMET and 127 for NUATMOS.
Both CALMET and NUATMOS also performed similarly for the 54 km grid when only
station observations were input, Le., CALMET (large R) RMSR « 159 and NUATMOS
RMSR = 153. On the 18 km grid, the RMSR differences are slightly larger with the CALMET
(small R) RMSR « 164 and NUATMOS RMSR - 128.
Overall, these trajectory comparison results suggests that the performance of both
CALMET and NUATMOS may be improved by the use of 80 km MM4-FDDA data with its
improved spatial and temporal resoluted over typical observational networks.
Scatter Plots
Straight scatter plots of u (and v) predicted by the diagnostic model versus those
predicted by the MM4-FDDA for (1) the surface level and (2) all upper air levels combined are
presented in Appendix B for Julian Day 217, Hour 6 (EST). In addition to the various
simulations with CALMET and NUATMOS described in Table 2-3, one interesting test which
was performed was to evaluate how well the 80 km MM4-FDDA data could represent the fine
resolution MM4-FDDA results by simply linearly interpolating the 80 km data to the fine
resolution grid. The linear interpolation was applied both vertically and horizontally.
Observations were not used in any way (except insofar as they influence the MM4-80 km winds),
and no additional terrain adjustments were made. This new field is called "MM4-80 km data
interpolated". (Due to the large number of data points involved, only about 10% of the points
were actually used in plotting).
Some general conclusions suggested by the scatter plots include:
• Similar results were found for upper levels between MM4-80 km and MM4-18
km and between MM4-80 km and MM4-54 km.
I:\«3M\a314Rv\Kd2.«pb 2-89
-------
• Some deviations were found for the surface level between MM4-80 km and
MM4-18 km due to the fact that some fine terrain features were not captured in
the 80-km grid simulation. However, the fields are still quite similar.
• Better agreement between CALMET and MM4-FDDA was achieved when the
80-km MM4-FDDA results were brought in after the diagnostic terrain
adjustment procedure (Le, as the Step 1 field or as "observations"). This is
probably due to the fact that (1) MM4-80 km grid results are already dose to
MM4-54 km and MM4-18 km grid results, and (2) the CALMET diagnostic
adjustments may duplicate or "double count" the effects of the terrain, since the
MM4-FDDA has the effects of the 80 km terrain. The similarity of the
interpolated MM4-FDDA 80 km data (which has 80 km terrain included) to the
54 km and 18 km MM4-FDDA fields (which have the fine scale terrain
represented) suggests that there might not be significant new terrain effects
between these scales.
• In general, CALMET performs better than NUATMOS in reproducing the
nested grid MM4-FDDA gridded wind fields.
Point-by-Point Statistics
Tables 2-6 and 2-7 contains a summary of the statistics for each run when combining all
hours and u and v for the three subsets of data surface layer, upper layers, and all layers. The
results suggest the following:
• The use of the hourly MM4-FDDA fields on an 80-km grid improves the ability
of both CALMET and NUATMOS to reproduce the reference field results.
• CALMET is able to reproduce the reference field is better than NUATMOS.
• For the fine grid sizes tested (18-km and 54-km), the MM4-FDDA data is best
used as the Step 1 field in the CALMET/DWM scheme.
• A simple interpolation of the 80-km MM4-FDDA data to 18-km and 54-km grids
produced better statistics than all simulations except the best CALMET runs.
l:\«3l4\»314rev\Ka2.wpb 2-90
-------
Table 2-6
Summary of Statistics for all hours and all levels
54-km Grid Results
RUN1 Corr. Coef. (r) RMSE
MM4 - 80 km data interpolated2 .961 0.992
CM3L - obs. only .789 2.211
CM3S - obs. only 342 3.220
CM6L - MM4 data as initial guess field .926 1378
CM6S - MM4 data as initial guess field .932 1316
CM12L - MM4 data as Step 1 field .962 0.992
CM12S - MM4 data as Step 1 field .962 0.993
CM15L - MM4 data as "observations" .960 1.020
CM15S - MM4 data as "observations" .959 1.021
NA3 - obs. only .481 3.430
NA6 - using MM4 data .515 3.500
NA9 - using MM4 data and o factor .514 3303
Dotation: CM = CALMET,-NA - NUATMOS, L * large radius of influence, S * small radius
of influence, Numerical values (3, 6, 9, 12, 15) « CALMET or NUATMOS run numbers.
In this table, the 54-km MM4-FDDA gridded winds are used as the reference field.
2MM4 80 km data linearly interpolated to the 54-km grid.
L\i314\«314rev\wd2.wph 2-91
-------
Table 2-7
Summary of Statistics for all hours and all levels
18-km Grid Results
RUN1 Corr. Coef. frt RMSE
MM4 - 80 km data interpolated2 .958 0.949
CM2L - obs. only .858 1.774
CM2S - obs. only .814 2.039
CM5L - MM4 data as initial guess field .920 1328
CM5S - MM4 data as initial guess field .922 1300
CM11L - MM4 data as Step 1 field .959 0.932
CM US - MM4 data as Step 1 field .959 0.938
CM14L - MM4 data as "observations" .952 1.011
CM14S - MM4 data as "observations" .952 1.010
NA2 - obs. only .767 2.262
NA5 - using MM4 data .797 2.127
NA8 - using MM4 data and o factor .795 2.135
Dotation: CM « CALMET, NA = NUATMOS, L « large radius of influence, S * small radius
of influence, Numerical values (3, 6, 9, 12, 15) = CALMET or NUATMOS run numbers.
In this table, the 18-km MM4-FDDA gridded winds are used as the reference field.
2MM4 80 km data linearly interpolated to the 18-km grid.
l:\«314\«314rev\ieea.wph 2-92
-------
There is an indication that the lack of fine scale terrain in the 80 km MM4-FDDA data
will begin to degrade the performance of the interpolated field as the grid size gets smaller.
This is expected since the terrain unresoh'ed by the 80 km MM4-FDDA grid is likely to become
increasingly important at finer grid scales. Interpolating the 80 km MM4-FDDA data to an 18
km grid works slightly less well than to the 54 km grid, possible due to the effect of unresolved
terrain starting to become important. The effect is most significant for the surface layer
(statistics shown below), although even down to 18 km, the 80 km MM4-FDDA winds are still
very highly correlated with the fine scale MM4-FDDA winds.
The statistics provided in this summary offer a direct method to compare two gridded
wind fields. By using brute force, we can compare each and every observed and predicted value,
which eliminates some of the arbitrary assumptions required in the trajectory analysis (i.e.,
where and how often to release particles, etc.).
I:\a314\a314Rv\Het2.«ph 2-93
-------
3. DISPERSION MODELS
Several non-steady-state puff models were considered by IWAQM as candidates for the
recommended Phase n model for assessing pollutant impacts at Class I areas. These models
included MESOPUFF-II (the interim (Phase I) recommended model), CTTPUFF (Ross and
Fox, 1991), CALPUFF (Scire et aL, 1990), and the ARMS model in the National Park Service
Air Quality Modeling System (NPSAQMS) (Morris et aL, 1988).
A review of the capabilities and features of each of these models was conducted and
compared to the capabilities of the Industrial Source Complex (ISC2) regulatory plume model,
(US EPA, 1992). Table 3-1 contains a summary of the model comparison. The capabilities of
the models were compared in the areas of dispersion coefficients, plume rise, wet deposition,
dry deposition, chemical transformation, complex terrain treatment, building downwash, the puff
sampling method, and other features of the modeling system. Since MESOPUFF-n was
designed to be optimized for mesoscale applications, it does not include the effects of near-field
processes such as momentum plume rise and building downwash. The MESOPUFF-n sampling
algorithm is most efficient for large source-receptor distances, although it can be applied at
short distances if suitably high puff sampling rates and puff release rates are used. One of the
major disadvantages of MESOPUFF-n is that it is a flat terrain model which does not include
the effects of complex terrain. Its advantages includes the ability to treat wet deposition, dry
removal, chemical transformation, its use of PG or Heftier dispersion coefficients, and the use
of an efficient integrated puff sampling algorithm.
CALPUFF was designed to be suitable for use on scaling ranging from tens of meters
from a source to hundreds of kilometers. It includes virtually all of the point source capabilities
of the ISC2 model (PG dispersion in rural areas, McElroy-Pooler dispersion in urban areas,
momentum and buoyant plume rise, terrain adjustments, and building downwash effects). In
addition, it has the mesoscale capabilities of MESOPUFF-n, including wet and dry deposition,
chemical transformation. The elongated puff algorithm in CALPUFF allows the near-field
impacts of sources to be evaluated in a computational efficient manner. CALPUFF is the only
model with the generalized capability of reproducing the plume results of ISC2 under steady-
state conditions. Under non-steady-state conditions, its puff formulation will allow time and
space variability in the meteorological fields and emission parameters. Thus, under those
conditions when the plume approach is appropriate (i.e., steady-state conditions), CALPUFF is
consistent with the ISC2 modeling results, but it will incorporate non-steady-state effects when
appropriate.
i:\i314\a314fev\iect3.wpb 3-1
-------
Table 3-1
Summary of Puff/Plume Model Features
Model
Algorithm
MESOPUFF-D CALPUFF OTPUFF NPSAOMS
ISC2
Dispersion Coefficients
PaaquuMjifford (rural Y (a)
McEboy-Poaler (urban
Hefftereqns. Y
Computed from observed af ow
Computed from <«^"""^ e,, 0V
MacReady eqns.
Buoyancy-induced dispersion
Wind direction shear (<7y)
22J* sector dispersion
Plume rise
Momentum rise
Buoyant rise Y
Stack tip effects
Building downwash effects
Vertical wind shear effects
Partial plume penetration
Wet Deposition
Scavenging coefficient approach Y
Dry Deposition
Oases Y (c)
Panicles Y (c)
Chemical Transformation
SO, chemistry (SO* SO4) Y (0
NO, chemistry (NOP HNO3, NO,) Y (f)
Complex Terrain
Terrain adjustments (constant elevation plumes)
Plume pith coefficients
Dividing streamline concept (CTDM-type model)
Y(b)
Y
Y
Y
Y
Y
Y
Y
Y
Y
Y
Y
Y
Y(d)
Y(d)
Y(Q
Y(f)
Y
Y
Y
Y
Y
Y(a)
Y
Y
Y
Y
Y
Y
Y(c)
Y(c)
Y(g)
Y(g)
Y(k)
Y
Y
Y
Y
Y(e)
Y(h)
Notes:
(a) MESOPUFF-n uses a fit of the PC curves of the form ax*, with stability-dependent, but distance-independent values
of a and b. NPSAQMS contains an option to use the MESOPUFF-Q fits of the PC curves.
(b) CALPUFF contains options for either the piece-wise ISC fit of the PC curves or the MESOPUFF-n fit
(c) MESOPUFF-n and NPSAQMS use a look-up table for the canopy resistance for gases. In MESOPUFF-n, panicle
deposition layer resistances are specified for SO4 and NO3 only. In NPSAQMS, particle deposition layer resistances
are computed internally based on pollutant, surface, and meteorological conditions.
(d) CALPUFF computes canopy resistances for gases and the deposition layer resistance for particles are computed
internally based on pollutant, surface, and meteorological conditions.
(e) ISC2 uses a reflection coefficient scheme which is appropriate only for large particles dominated by gravitational
settling.
(0 MESOPUFF-n and CALPUFF use the same highly-parameterized chemical transformation equations.
(g) NPSAQMS contains two different chemical transformation mechanisms: (1) the RIVAD nvrhanism (best for
relatively dean non-urban areas), and (2) the MESOPUFF-n nwhanttm.
(h) ISC2 contains a constant half-life decay factor to represent depletion of the primary pollutant due to chemical
conversion effects.
(i) CALPUFF applies a minimum miss distance for elevated plume approach terrain obstacles of 10 m.
0) CALPUFF uses a CTDM-type complex terrain algorithm for subgrid-scale terrain obstacles.
(k) ISC2 chops off terrain at stack height on a source-by-source and receptor-by-receptor basis.
c\*314\a314rev\HO(3.wph
3-2
-------
Table 3-1 (Continued)
Summary of Puff/Plume Model Features
Model
MESOPUFF-n CALPUFF OTPUFF NPSAOMS ISO
Building Do»n»ath
NeaMwke (cavity)
Far-wake (H-S, S-S scheme)
fvff yilPPti"* Method
Snapshot sampling approach
ZuLDetti method
Integrated puff approach
EkMgttcd tluf approach
Other Features
Valley-cource algorithm
Terrain data base and processor
Land use data base and processor
Meteorological data processors
Y
Y
Y
Y
Y
Y
C\«314\i314rev\ieet3.vph
3-3
-------
CTTPUFF is a basic puff model using a snapshot sampling algorithm. It contains the
option to use PG dispersion coefficients or derived dispersion rates based on observed ov and
ow values, buoyant and momentum plume rise, partial plume penetration, and terrain
adjustments to the plume height. However, it does not contain algorithms important for AQRV
determination studies, such as wet and dry removal, and chemical transformation. Its sampling
algorithm is flexible, bit more computationally costly than either CALPUFF or MESOPUFF-H.
NPSAQMS was designed primarily for mesoscale transport applications. It does not
contain algorithms for building downwash, urban dispersion, or momentum plume rise. It has
parameterizations of wet and dry deposition, and chemical transformation. It contains a
convenient set of processor programs and terrain and land use data files which facilitate the
model's use. Also, it is the only model which has a valley-source algorithm and terrain-adjusted
(MacReady) dispersion coefficients.
3.1 Sensitivity Tests of Model Components
Based on the capabilities of the various puff models, it was decided by IWAQM to use
the CALPUFF model as the framework for the Phase n puff modeling system. Sensitivity tests
of various algorithms from CALPUFF and other models were tested to determine their
behavior in response to various model inputs.
3.1.1 Dispersion Parameters
CALPUFF has three options of calculating oy and oz for small to medium transport
distances:
1) ISC curves: function of discrete PG class
2) MESOPUFF curves: function of discrete PG class
3) Draxler's formulas: continuous function of u., w., L, h, and z.
The ISC and MESOPUFF-II curves are different only in the "rural" mode. When in the
"urban" mode, both models use the McElroy and Pooler curves.
The CALPUFF model also has the option of switching to the Heffter formulas for
longer transport distances (e.g., x > 10 km). The Heffter time-dependent puff growth equations
are defined as:
t\i3M\*314nv\iccl3.vph 3-4
-------
M, ' MM * 0"5
w, -
(3-2)
where, At is the incremental time (s),
t is the total travel time(s), and
a,, is a stability-dependent factor (a,, « 5.0, 3.873, 2.739, 1.871, 1225, and 0.707 for
stability classes A-F, respectively)
Equations (3-1) and (3-2) are currently implemented as:
(o,) M - (o,) lft . * 0.5 Ar
V Wjc. > 10 *» V "x • 10 b»
(3-3)
v '
\
«/«. > 10 fen
10
ag Af
~
(M)
where (oy), . 10km and (ojx » 10ta are the values of oy and oz at x = 10 km calculated by one of
the three options mentioned in the beginning, At is the time required to travel from x = 10 km
to x, and t is the time required to travel from x = 0 to x.
The comparison of the ISC and MESOPUFF-n curves is straightforward since they both
depend on the stability class only. The comparison of the oy and oz values based on the
Draxler's formulas with those predicted by either ISC or MESOPUFF-n is more involved since
for each stability class, there are infinite number of corresponding combinations of u., w., L, and
h. Each of these combinations yield a different set of oy and oz curves.
In order to evaluate the Draxler's formulas, the following primary variables were varied:
Month and Temperature:
Hour:
Bowen Ratio:
Albedo:
z0:
Wind speed:
Cloud Cover:
(Jan., 283°K; April, 293°K; July, 303°K)
(1-24)
03, 1.0
0.15, 0.3
0.05, 0.25, 1.0 m
2, 5 m/s
2, 5, 10 tenths
i:\t3M\i314icv\iect3.wph
3-5
-------
In addition, the following variables were fixed:
Reference height * 10 m
Anthropogenic heat flux = 0
Soil heat flux parameter = 0.15
Pressure « 1000 mb
Latitude « 40° N
Longitude » 80° W
Time Zone * 5
The following parameters were derived:
Sensible Heat Flux
Short-wave Solar Radiation
u.
L
h (3000 u., neutral formula, for daytime only)
w.
°v
°w
PG Class (based on Golder's nomogram)
After all the possible combinations of the "primary" variables are covered, the results are
then subset according to the value of the corresponding PG class. Instead of having a unique
curve for each PG class, we now have a "whole family" of curves for each PG class. These
curves are then shown in the form of box plots.
Figure 3-1 and 3-2 show the horizontal and vertical dispersion parameters, oy and or
respectively, as a function of downwind distance, x, for the ISC and MESOPUFF-E models for
rural areas. Figures 3-3 and 3-4 are the same as Figures 3-1 and 3-2 but for urban areas where
both models use the McElroy and Pooler curves. Note that or in Figures 3-2 and 3-4 is capped
at 5000 m. Since the formulas for oy and ot used in the two models depend only on PG class
and downwind distance, the curves presented in Figures 3-1 through 3-4 do not have any
scattering.
Figures 3-5 and 3-6 show the horizontal and vertical dispersion parameters, oy and or
respectively, as a function of downwind distance, x, for Draxler's formulas for each equivalent
PG class, described later. Draxler's formulations of ay and oz depend on micrometeorological
l\«314\«3M«v\ieel3.wpb 3-6
-------
Rural — Solid: ISC; Dashed: MESOPUFF II
20000
2000
200
20
0.1
1.0 10.0
x (km)
100.0
Figure 34. o, vs. x for ISC and MESOPUFF, rural.
3-7
-------
Rural — Solid: ISC; Dashed: MESOPUFF II
20000
2000 -
N
200 -
20 -
I
0.1
1.0 10.0
x (km)
100.0
Figure 3-2. oz vs. x for ISC and MESOPUFF, rural.
oz is clipped at SOOO m.
3-8
-------
20000
2000 -
200 -
20 b>
0.1
Urban — McElroy and Pooler
1.0 10.0
x (km)
100.0
Figure 3-3. oy vs. x for urban areas.
-------
20000
2000 -
N
200 -
20 <
0.1
Urban — McElroy and Pooler
1.0 10.0
x (km)
100.0
Figure 3-4. or vs. x for urban areas. or is clipped at 5000 m.
t\t3U\a314RV\MCt3.wpb
3-10
-------
Draxler's Formulas, Class A Equivalent
20000
2000 -
200 -
20 -
0.1
1.0 10.0
x (km)
100.0
Figure 3-5(a). oy vs. x based on Draxler's formula, one frame for each PG class.
Each box represents variations due to temperature, time of day, albedo, Bowen
ratio, z0, wind speed, and cloud cover.
3-11
-------
Draxler's Formulas, Class B Equivalent
20000
2000 -
200 -
20
0.1
1.0 10.0
x (km)
100.0
Figure 3-5(b). oy vs. x based on Draxler's formula, one frame for each PG class.
3-12
-------
Draxler's Formulas, Class C Equivalent
20000
2000 -
200 -
20 -
0.1
1.0 10.0
x (km)
100.0
Figure 3-5(c). oy vs. x based on Draxler's formula, one frame for each PG class.
3-13
-------
Draxler's Formulas, Class D Equivalent
20000
2000 -
200 -
20
0.1
1.0 10.0
x (km)
100.0
Figure 3-5(d). oy vs. x based on Draxler's formula, one frame for each PG class.
3-14
-------
Draxler's Formulas, Class E Equivalent
20000
2000 -
200 -
20
0.1
1.0 10.0
x (km)
100.0
Figure 3-5(e). oy vs. x based on Draxler's formula, one frame for each PG class.
3-15
-------
Draxler's Formulas, Class F Equivalent
20000
2000 -
200 -
20
0.1
1.0 10.0
x (km)
100.0
Figure 3-5(f). oy vs. x based on Draxler's formula, one frame for each PG class.
3-16
-------
Draxler's Formulas, Class A Equivalent
20000
2000 -
200 -
N
20 V-
0.1
1.0 10.0
x (km)
100.0
Figure 3-6(a). oz vs. x based on Draxler's formula, one frame for each PG class.
Each box represents variations due to temperature, time of day, albedo, Bowen
ratio, z0, wind speed, and cloud cover.
3-17
-------
Draxler's Formulas, Class B Equivalent
20000
2000
200
N
20
0.1
t i t t t i I i i
1.0 10.0
x (km)
100.0
Figure 3-6(b). oz vs. x based on Draxler's formula, one frame for each PG class.
3-18
-------
Draxler's Formulas, Class C Equivalent
20000
2000 -
200 -
20
0.1
1.0 10.0
x (km)
100.0
Figure 3-6(c). oz vs. x based on Draxler's formula, one frame for each PG class.
3-19
-------
Draxler's Formulas, Class D Equivalent
20000
2000
200
20
0.1
1.0 10.0
x (km)
100.0
Figure 3-6(d). oz vs. x based on Draxler's formula, one frame for each PG dass.
3-20
-------
Draxler's Formulas, Class E Equivalent
20000 p
2000 -
200 -
20 ,r
0.1
1.0 10.0
x (km)
100.0
Figure 3-6(e). oz vs. x based on Draxler's formula, one frame for each PG dass.
3-21
-------
Draxler's Formulas, Class F Equivalent
20000
2000 -
200 -
20 -
0.1
1.0 10.0
x (km)
100.0
Figure 3-6(f). or vs. x based on Draxler's formula, one frame for each PG dass.
3-22
-------
parameters such as friction velocity, Monin-Obukhov length, and convective velocity scale that
are in turn dependent on ambient meteorological conditions and site characteristics. Although
Draxler's formulas do not depend on PG class explicitly, an equivalent PG class can be
determined from Golder's nomograms given values of Monin-Obukhov length, L, and surface
roughness, z0. Since different combinations of L and z<, may yield the same PG class, box plots
are used in Figures 3-5 and 3-6 to represent the scattering associated with each PG class due to
different ambient and site conditions. The significant points in each box are median, ± one
standard deviation, and ± two standard deviations, respectively.
3.1.2 Dry Deposition
At the reference height, z,, the deposition velocity for gases can be expressed (Wesely
and Hicks, 1977; Hicks, 1982) as the inverse of a sum of three resistances.
(3-5)
where, r. is the atmospheric resistance (s/m) through the surface layer,
rd is the deposition layer resistance (s/m), and,
r is the canopy (vegetation layer) resistance (s/m).
c
Atmospheric Resistance
The atmospheric resistance is obtained by integration of the micrometeorological flux-
gradient relationships (Wesely and Hicks, 1977):
«
* **•
where, z, is the reference height (m),
z0 is the surface roughness length (m),
k is the von Karman constant (~ 0.4),
u. is the friction velocity (m/s),
4>H is a stability correction term, and,
L is the Monin-Obukhov length (m).
The stability correction term accounts for the effects of buoyancy on the eddy diffusrvity
of the pollutant. It is assumed that the pollutant transfer is similar to that for heat (Wesely and
Hicks, 1977). A gridded field of surface roughness lengths is passed to the model in the output
file of the meteorological model, CALMET. In CALMET, the surface roughness length is
i:\»314\a314icv\iect3wph 3-23
-------
either estimated from the predominant land use of each grid cell, or, if available, based on
actual values entered by the user. Over water, due to the effect of the wind on wave height, the
surface roughness length varies as a function of wind speed, and is computed internally within
CALPUFF using the parameterization of Hosker (1974).
Zo « 2.0 x 10* u2* (3-7)
where u is the wind speed (m/s) at 10 m.
Deposition Layer Resistance
Due to the importance of molecular diffusion to the transport through the laminar
deposition layer, the deposition layer resistance for gaseous pollutants is parameterized in terms
of the Schmidt number:
r,-^/(fcu.) (3-8)
where, Sc is the Schmidt number (w/D),
u is the kinematic viscosity of air (m2/s),
D is the molecular diffusivity of the pollutant (m2/s), and,
dj, d2 are empirical parameters.
Experimental studies summarized by Hicks (1982) suggest a range of values for the
empirical variables of 1.6 to 16.7 for dj and 0.4 to 0.8 for d2. Intermediate values of dj = 2, and
d2 = 2/3 are recommended based on Shepherd (1974), Slinn et al. (1978), and Hicks (1982).
Canopy Resistance
The canopy resistance is the resistance for gases in the vegetation layer. There are three
main pathways for uptake/reaction of the pollutant within the vegetation or surface:
(1) Transfer through the stomatal pore and dissolution or reaction in the mesophyll
cells.
(2) Reaction with or transfer through the leaf cuticle.
(3) Transfer into the ground/water surface.
fc\«314\«314rev\ieet3.wph 3.24
-------
In the resistance model, these pathways are treated as three resistances in parallel
re * [LAI IT, + LU/r^ + 1/rJ'1 (3-9)
where, rf is the internal foliage resistance (s/m) (Pathway 1),
r^ is the cuticle resistance (s/m), (Pathway 2),
rf is the ground or water surface resistance (s/m), (Pathway 3), and,
LAI is the leaf area index (ratio of leaf surface area divided by ground surface area).
The LAI is specified in the model as a function of land use type.
The first pathway is usually the most important for uptake of soluble pollutants in
vegetated areas. The internal foliage resistance consists of two components:
where, r, is the resistance (s/m) to transport through the stomatal pore, and,
rm is the resistance (s/m) to dissolution or reaction of the pollutant in the mesophyll
(spongy parenchyma) cells.
Stomatal action imposes a strong diurnal cycle on the stomatal resistance, and, due to its
important role in determining deposition rates for gaseous soluble pollutants such as SO2, on the
deposition velocity, as well. Stomatal opening/closing is a response to the plant's competing
needs for uptake of CO2 and prevention of water loss from the leaves. The stomatal resistance
can be written (O'Dell et al., 1977) as:
(3-11)
where, p is a stomatal constant (- 2.3 x 10"* m2),
b is the width of the stomatal opening (m), and,
D is the molecular diffusivity of the pollutant (m2/s).
The width of the stomatal opening is a function of the radiation intensity, moisture
availability, and temperature. The variation of b during periods when vegetation is active can be
represented (Pleim et al., 1984) as:
- b^ (3-12)
where, b,,^ is the maximum width (m) of the stomatal opening (- 25 x 10"6 m),
b,,^ is the minimum width (m) of the stomatal opening (- 0.1 x 10"* m),
S is the solar radiation (W/m2) received at the ground, and,
SB,, is the solar radiation (W/m2) at which full opening of the stomata occur.
t\i3M\«314rev\«ed3.wpb 3-25
-------
However, during periods of moisture stress, the need to prevent moisture loss becomes
critical, and the stomata close. It can be assumed that b « b^, for unirrigated vegetation under
moisture stress conditions. When vegetation is inactive (e.g., during the seasonal dry periods in
much of California), the internal foliage resistance becomes very large, essentially cutting off
Pathway 1. In CALPUFF, the state of the unirrigated vegetation is specified as one of these
states (A) active and unstressed, (B) active and stressed, or (C) inactive.
The effect of temperature on stomatal activity has been reviewed by Pleim et al. (1984).
The most significant effects are due to temperature extremes. During cold periods (T < 10°
C), metabolic activity slows, and b is set equal to b^,. During hot weather conditions (T > -
35° C), the stomata are fully open (b « b^J to allow evaporative cooling of the plant (assuming
the vegetation is in state A - active and unstressed). These temperature effects provide
additional bounds on the value of r. given by Eqn. (3-11).
Mesophyil Resistance
The mesophyll resistance depends on the solubility and reactivity of the pollutant. It is
an input parameter supplied to the deposition model for each gaseous species. O'Dell et al.
(1977) estimate the mesophyll resistance for several pollutants. For soluble pollutants such as
HF, SO2, C12 and NH3, rm ~ 0.0. The mesophyll resistance can be large for less soluble
pollutants such as NO2 (~ 500 s/cm) and NO (9400 s/cm). For other pollutants, rm can be
estimated based on the solubility and reactivity characteristics of the pollutant.
Cuticle Resistance
The second pathway for deposition of gases in the vegetation layer is via the leaf cuticle.
This includes potential direct passage through the cuticle or reaction of the pollutant on the
cuticle surface. Hicks (1982) notes that measurements of SO2 deposition to wheat (Fowler and
Unsworth, 1979) show significant cuticle deposition. However, Hosker and Lindberg (1982)
suggest that passage of gases through the cuticle is negligible. Therefore, the cuticle deposition
is likely to be controlled by the pollutant reactivity. Pleim et al. (1984) parameterize r^, as a
function of the pollutant reactivity of the depositing gas relative to the reference values for SO2.
where, A is the reactivity parameter for the depositing gas,
A^, is the reference reactivity of SO2 (- 8.0), and,
reut(ref) is the empirically determined reference cuticle resistance (s/m) of SO2.
t\t314\*314rev\MCt3.vph 3-26
-------
Pleim et aL (1984) suggest re-(ref) is about 30 s/cm. Reactivity values for other
pollutants are estimated at 8.0 (NO2), 15.0 (O3), 18.0 (HNO3), and 4.0 (PAN).
Ground/Water Resistance
The third pathway through the "vegetation layer" involves deposition directly to the
ground or water surface. In moderately or heavily vegetated areas, the internal foliage and
cuticle resistances usually control the total canopy resistance. However, in sparsely vegetated
areas, deposition directly to the surface may be an important pathway. Over water, deposition
of soluble pollutants can be quite rapid.
The ground resistance, rr over land surfaces can be expressed (Pleim et aL, 1984)
relative to a reference value for SO2:
where rg(ref) is the reference ground resistance of SO2 (~ 10 s/cm).
Slinn et al. (1978) parameterize the liquid phase resistance of the depositing pollutant as
a function of its solubility and reactivity characteristics. Their results can be expressed as:
rf = Hl(a. d, «.) (3-15)
where, H is the Henry's law constant, which is the ratio of gas to liquid phase concentration
of the pollutant, (H ~ 4 x 10'2 (SO2), 4 x 10'7 (H2O2), 8 x 10* (HNO3), 2 x 10°
(03), 3.5 x 10° (N02), 1 x 10 "2 (PAN), and 4 x 10"6 (HCHO)),
a. is a solubility enhancement factor due to the aqueous phase dissociation of the
pollutant(a. - 103 for SO2> - 1 for CO2), and
d3 is a constant (~ 4.8 x 10"4).
Resistances for Partkulate Matter
Because paniculate matter does not interact with vegetation in the same way as gaseous
pollutants, particle deposition velocities are commonly expressed only in terms of r., rd and a
gravitational settling term. The atmospheric resistance, r,, for a particle is the same as for a gas
(Eqn. 3-6) The resistance in the vegetation layer (rc) is not a factor because once penetrating
the deposition layer, particles are usually assumed to stick to the surface (e.g., Voldner et aL,
1986). Therefore, their behavior is similar to highly soluble/reactive gases with rc ~ 0. Based on
an assumption of steady-state conditions, the deposition velocity for particles can be expressed
(Slinn and Slinn, 1980; Pleim et al., 1984) as:
i:\t314\t314rrv\MCt3.wph 3-27
-------
(r«
* v
where \f is the gravitational settling speed (m/s) of the particle.
The deposition layer resistance is parameterized in terms of the Schmidt number (Sc -
v/D, where u is the viscosity of air, and, for particles, D is the Brownian diffusivity of the
pollutant in air) and the Stokes number (St - (v,/g)(u.2/u), where v( is the gravitational settling
velocity and g is the acceleration due to gravity).
(3-17)
Sensitivity tests were conducted with the CALPUFF deposition velocity model for 5
species:
SO2, SCV, NOr HNO3, and NO/
The following parameters were varied in the sensitivity analysis:
Month and temperature:
Hour:
Bowen ratio:
Albedo:
z0:
Vegetation state:
Leaf area index:
Wind speed:
Cloud cover:
January - 283°K,
April - 293TC,
July • 303°K
1-24
0.5, 1.0
0.15, 030
0.05, 0.25, 1.0 m
1 = active, unstressed
2 - active, stressed
3 - inactive vegetation
0.5, 3.0, 7.0
2.0, 5.0 m/s
2, 5, 10 (tenths)
which produces a total of 46,656 combinations of variables. The CALPUFF default input
parameter for the diffusivity, d., reactivity, mesophyll resistance, and H were used in the
analysis.
fc\«314\i314rev\*ect3.wpb
3-28
-------
The results of the sensitivity tests are shown in Figures 3-7 through 3-11. These results
show increased diurnal variability for gases (SO* HNO,, NOJ as compared to particles (SO/,
NO3"), an increase in vd as a function of temperature, Bowen ratio, and roughness. The state of
the vegetation is important for gases, but not particle deposition. The leaf area index is
predicted by the model to be an important parameter in the gas deposition calculation. In
general, the deposition velocities predicted by the model are consistent with typical observed
values for each of the pollutants under similar conditions.
3.13 Chemical Transformations
CALPUFF and MESOPUFF-n employ the same algorithm for computing chemical
transformation rates. The transformation pathways for the five active pollutants (SO* SOJ,
NOr HNO3, and NO;) are included in the MESOPUFF-n scheme. Transformation rate
expressions were developed by statistically analyzing hourly transformation rates produced by a
photochemical model. The photochemical model employed the RHC/NO^SO, chemical
mechanism of Atkinson et al. (1982). Plume SOj/NO, dispersing into background air containing
ozone and reactive hydrocarbons was simulated over a wide range of conditions representing
different solar radiation intensities, temperatures, dispersion conditions, background ozone and
RHC concentrations, plume NOX concentrations and emission times. The following
transformation rate expressions, representing curve fits to the daytime hourly conversion rates
predicted by the photochemical model, were determined:
- 36 ^'[O,]0-71 S~l» * ^ (3-18)
= 1206 [03]u S'W1 [N0t]-** (3-19)
1261 [03]M5 5'IJ4 [NO,]-012 (3-20)
where, k, is the SO2 to SO4 transformation rate (percent/hour),
k2 is the NO, to HNO3 + RNO3 transformation rate (percent/hour),
k3 is the NO, to HNO3 (only) transformation rate (percent/hour),
R is the total solar radiation intensity (kw/m2),
S is a stability index ranging from 2 to 6 (PGT class A and B=2, O3, D-4, E«5,
F-6),
RH is the relative humidity (percent),
[O3] is the background ozone concentration (ppm),
[NOJ is the plume NO, concentration (ppm), and,
is the aqueous phase SO3 oxidation term (percent/hour).
i:\i314\i314rev\MKt3.wph 3-29
-------
101
ai
O
CO
10°
in
If
IQ-Z
101
10°
ID-'
10-2
273 283 293 303
Temp. (K)
i...i... i ... i ... i
0 4 8 12 16 20 24
Hour
CO
I,
o
10°
10-1
10~2
313
0.0 0.5 1.0
Bowen Ratio
1.5
Figure 3-7(a). Deposition velocities for SO2 as a function of hour, Bowen ratio, and
temperature.
3-30
-------
101
N
O
CO
h 10
o
•o
101
u 1°°
o
c
» ID'1
10-2
0.00 0.15 0.30
Albedo
N
O
CO
o
to
10'
1C0
10-2
0.00 0.25 0.50 0.75 1.00 1.25
0.45
1 2 3
Vegetation State
Figure 3-7(b). Deposition velocities for SO2 as a function of Z0, vegetation state, and albedo.
3-31
-------
m 10o
fe
i
10-8
I
012345678
Leaf Area Index
1C1
CM
O
u 10°
o
CO
1 10-1
•o
10-2
' II I "
i M .
u>'
M 10°
h
b«
•2- io-»
m-E
i i i i i i i i i :
"I I i
S S 5^
i i i i i i i i i
2 5
Wind Speed (m/s)
25 10
Cloud Cover (tenths)
Figure 3-7(c). Deposition velocities for SO2 as a function of wind speed, cloud cover, and LAI.
3-32
-------
10°
w 10-1
o
10-3
TTT1I
10°
O
CO
10-1
ID"3
I ! I
11111
0 4 8 12 16 20 24
Hour
10°
O
CO
10-1
lio-2
10-3
273 283 293 303 313
Temp. (K)
0.0 0.5 1.0
Bowen Ratio
1.5
Figure 3-8(a). Deposition velocities for 5O4" as a function of hour, Bowen ratio, and
temperature.
3-33
-------
10°
CO
CO
h
o
1 ID'2
10-3
ID'3
0.00
0.15 0.30
Albedo
10°
CQ
It
1Q-2
•o
10
-3
0.00 0.25 0.50 0.75 1.00 1.25
z0 (m)
.45
1 2 3
Vegetation State
Figure 3-8(b). Deposition velocities for 5O4" as a function of Z0, vegetation state, and albedo.
3-34
-------
10°
o*
w 10-1
fe
10-2
ID'3
01234567B
Leaf Area Index
10°
o"
W
*E
•o
ID'3
2 5
Wind Speed (m/s)
° 10-1
i.
o
I I I I I I I I
25 10
Cloud Cover (tenths)
Figure 3-8(c). Deposition velocities for 5O4" as a function of wind speed, cloud cover, and
LAI.
3-35
-------
O
t,
O
101
10°
» 10-1
10-2
273 283 293 303 313
Temp. (K)
O
t.
O
10°
10-1
10-2
0 4 B 12 16 20 24
Hour
I*
O
6
" ID"1
10~z
0.0 0.5 1.0
Bowen Ratio
1.5
Figure 3-9(a). Deposition velocities for NOX as a function of hour, Bowen ratio, and
temperature.
3-36
-------
o
S5
In
O
10°
" 10'1
•o
10
-2
1-
o
101
10°
1 ID'1
ID'8
0.00 0.15 0.30
Albedo
101
i .... i
0.00 0.25 0.50 0.75 1.00 1.25
z0 (m)
M
O
u, 10°
o
" 10-1
•o
>
10
-2
0.45
1 2 3
Vegetation State
Figure 3-9(b). Deposition velocities for NOX as a function of Z0, vegetation state, and albedo.
3-37
-------
101
O
w
"e
1C
1
•o
ID
"2
OH
Z 10°
o
10-8
I
012345678
Leaf Area Index
2 5
Wind Speed (m/s)
101
OH
2 10"
I I I I
I 1 1
25 10
Cloud Cover (tenths)
Figure 3-9(c). Deposition velocities for NOX as a function of wind speed, cloud cover, and LAI.
3-38
-------
o"
2
u
o
i
101
10°
10
-2
273 283 293 303
Temp. (K)
313
o
£
n
1
10° -
0 4 B 13 16 30 34
Hour
n
O
o
10°
"V
I 10-1
•o
10
-2
0.0 0.5 1.0
Bowen Ratio
1.5
Figure 3-10(a). Deposition velocities for HNO3 as a function of hour, Bowen ratio, and
temperature.
3-39
-------
n
O
t-
o
101
10°
10
-2
1U
n
O
« 100
Ji
I 10-1
v— •*
10-2
O.C
. . ,- , ...j— , — , — . — ^-, — , — _ — ,^
' N 1
" "
1 .... |
)0 0.15 0.30 0.45
u r
n °
. ... i .... i .1 i
Albedo
101
cT
a 10o
O
VH
'm'
10-8
III'
J fl ft 1
H 8 .
i i i
0.00 0.25 0.50 0.75 1.00 1.25
1 2 3
Vegetation State
Figure 3-10(b). Deposition velocities for HNO3 as a function of Z0, vegetation state, and albedo.
3-40
-------
IO1
10°
n
It
io-2
I
012345876
Leaf Area Index
1 nl
C5
O
i 100
O
»*— i
'w
1 io-1
v^--
•O
1 " " I ' " ;
1 1
1 fl
H o
tf
io1
o"
K 10"
fe
n
A io-1
10-8
I 1 1 1 1 1 1 1 1 1 ;
1 1 1
! ! ft
DO o
.1 I 1 .
2 5
Wind Speed (m/s)
25 10
Cloud Cover (tenths)
Figure 3-10(c). Deposition velocities for HNO3 as a function of wind speed, cloud cover, and LAI.
3-41
-------
10°
10-1 .
10-2 .
ID'3
WWii
iU
n
0
fe 1°"1
J.10-2
ID'3
• ' i i i
fl ft ft •
-HI;
l.i. 1 , .-.i
273 283 293 303
Temp. (K)
I " ' I • • • I . I... I... I... r
iVVVVV
I... I... I... I... I... I... I
0 4 B IB 16 20 24
Hour
10°
O
b,
o
10
lMr <
» ID'2
10~3
313
0.0 0.5 1.0
Bowen Ratio
1.5
Figure 3-ll(a). Deposition velocities for NO3 as a function of hour, Bowen ratio, and
temperature.
3-42
-------
10°
10~3
0.00 0.15 0.30
Albedo
0.45
10°
n
O
2
u
O
.0
-
ID"2
•o
10°
o"
1Q-3 I . . , , I , , I . , , , I , , . I 1Q-3
0.00 0.25 0.50 0.75 1.00 1.25
z0 (m)
1 2 3
Vegetation State
Figure 3-1 l(b). Deposition velocities for NO3 as a function of Z0, vegetation state, and albedo.
3-43
-------
10°
n
O
a 10-1
— 10-8
I?
I I I t 1
T
012345678
Le&t Area Index
1U"
C3
0
S5 _
L*
o
**-H
'w'
If 10-8
•o
ID-3
1 i
-
: 8
1 i
IV
o"
* ID'1
LI
o
IM
n
x^^
I10-a
ID"3
1 1 1 1 1 1 1 1 1
-
1*1 1*1 1*1
"0 0 r
2 5
Wind Speed (m/s)
25 10
Cloud Cover (tenths)
Figure 3-1 l(c). Deposition velocities for NO3 as a function of wind speed, cloud cover, and LAI.
3-44
-------
The aqueous phase component of the SO2 conversion rate was parameterized as:
*!(„) = 3 x ID'* KH4 (3-21)
Equations (3-18) to (3-20) apply only during daytime periods when gas phase free radical
chemistry is active. The use of the ozone concentration and the radiation intensity as surrogates
for the OH concentration, as in the above equations, is appropriate only during the day. At
night, SO2 and NO, oxidation rates resulting from heterogeneous reactions, are generally much
lower than typical daytime rates (Wilson, 1981; Forrest et aL, 1981). Nighttime oxidation rates
of 0.2% and 2.0% for SO2 and NOr respectively, are used as default values in the model.
NPSAQMS contains two options for chemical transformation. The first is the
MESOPUFF-n/CALPUFF scheme described above. The second mechanism is the RTVAD
condensed pseudo-first-order chemical scheme. Morris et al. (1988) note that the RIVAD
mechanism is based on low background concentrations of VOCs, and is best suited for relatively
clear non-urban areas.
In the RTVAD scheme, the rate of sulfate and nitrate production is estimated by
calculating the steady-state concentration of OH" with a plume, the steady-state equation for
0('D) and OH' are:
(0('D)]= , (3-22)
1 J
'D)] [H20] - K?, [OH-] [SO2] - K» [OH-] [NO2] , (3-23)
1 J [NQJ
With this steady-state concentration, plume pseudo-first-order SO2-to-SO4 and NO2-to-HNO3 +
NO3" conversion rates can be calculated as:
(3-25)
(3-26)
i:\a314\a314rev\iect3.wph
-------
Heterogeneous Chemistry
The RIVAD chemical mechanism does not explicitly calculate the aqueous-phase
oxidation of SO2 to sulfates. Instead the RIVAD chemistry assumes a constant heterogeneous
oxidation rate of SO2 of 0.2 %/h that is added to the homogeneous rate.
where,
1.3 jc 1(T3 (cos Zf4 ppm-1 min"1
*35 = 4.45 x 101
mm
fcjg = 3.4 x 10s ppm~l min"1
ky, = 2.0 x 103 ppm'1 min"1
kyg = 1.4 x 10* ppm'1 min"1
Sensitivity tests were conducted with the MESOPUFF-H and RIVAD chemical
mechanisms with the following range of primary variables:
(3-27)
(3-28)
(3-29)
(3-30)
(3-31)
Date and temperature:
Time:
4 O3 concentrations:
3 NO, concentrations:
3 relative humidities:
2 wind speeds:
3 cloud covers:
Location:
January, 283°K
April 293°K
July, 303°K
Hours 1-24
20, 50, 80, 200 ppb
7, 350, 1400 ppb
20, 50, 80%
3, 10 knots
2, 5, 10 (tenths)
40° N latitude, 80° W longitude
i:\a314\n314rev\iect3.wph
3-46
-------
Hie following parameters were derived from primary parameters: solar elevation angle
from latitude/longitude and date/time, solar radiation from solar elevation and doud cover,
stability class from solar elevation angle, doud cover, wind speed, and ceiling height (Turner's
method).
The results are presented in Figures 3-12 through 3-19 using diurnal box plots for each
combination of O3 and NO, values. As a result, each box represents variations in inversion rates
resulting from variation in season (temperature), relative humidity, wind speed, and doud cover.
The significance points for each box are the 2nd, 16th, 50th, 84th, and 98th percentiles.
In general the SO2/SO4 transformation rates predicted by the MESOPUFF-E and
RTVAD schemes are similar and consistent with typical observed values of transformation rates.
Under low-moderate O3 concentrations, SO2 transformations of a few tenths of a percent/hour
are predicted. At high O3 concentrations, peak transformation rates reach a few percent/hour.
A typical diurnal cycle of transformation rate was computed with both the MESOPUFF-
II and RTVAD chemical mechanism. The following assumptions were made:
Latitude: 40° N O3 concentration: SOppb
Longitude 80° W NOX concentration: 7ppb
Pressure 1000 mb Relative humidity: 50%
Date: July 15 Wind speed: 2.5 m/s
Temperature: 303°K Cloud cover: 0
Figures 3-20 and 3-21 show the conversion rates for MESOPUFF-II/CALPUFF and
RTVAD, respectively. Note that the reason that the conversion rates for NO, and TNO3 for
CALPUFF do not display as distinct diurnal cycles on RIVAD is because the equation defining
the rate constants depend on stability class, which is only allowed to assume discrete values
between 2 (class B) and 6 (class F).
On the NO, and TNO3 side, the MESOPUFF-II and RTVAD schemes show larger
differences than with SO2. MESOPUFF-II predicts significantly higher NO, loss rates than
ARM3. The TNO3 peak formation rates of 10 to 20%/hour are similar at moderate O3
concentrations, but under higher ozone concentrations, the MESOPUFF-II scheme produces
significantly faster rates of TNO3 formation than ARM3.
t\«314\a314rev\«ed3.wph 3-47
-------
CALPUFF
102
I
5 10>
10«>
S 10-'
o
O
en
03=20 ppb. NOX=7 ppb
ID"2
12 16
Hour
20 24
O
io2
101
10°
03=50 ppb. NOX=7 ppb
10-z
i IT ' i ' i ' r i ~*n ™ • i • i
I . I . I . I . I . I . I . I . I . I . I . I
12 16
Hour
20 24
03=80 ppb. NOX=7 ppb
03=200 ppb, NOX=7 ppb
iU
u.
1
3 10'
o
u
3
O
\ 10°
it
01
to
L.
8 10-'
o
N
O
ID-2
c
-
-
.1.1.
) 4
1 1 {
in
IP
1 1 '
mill
. i .
6 12 16
' 1 ' 1 ' 1 ' :
-
ll '
I
20 24
1U*
|
^ 10'
o
L.
3
O
< 10°
fr?
v
IB
« 10-'
_o
lO-2
Hour
1U~
10'
10°
10-'
in-2
C
1 1 ' 1 ' 1 ' 1
- |i
.1.1.1.1.1
) 4 B
1
' 1 ' 1 ' 1 ' 1 ' 1 ' 1 ' :
111.
"II
.1.1.1.1.1.
2 16 20 24
Hour
Figure 3-12(a).
SO2 loss rate for CALPUFF for NOX = 7 ppb.
Note that each box represents variations due to season (temperature,
relative humidity, wind speed, and cloud cover).
3-48
-------
CALPUFF
03=20 ppb. NOX=350 ppb
10Z
I
i
* 10'
u
o
3
O
JC
10°
\ I I T I 1 I It I i i
8 10-'
O
1Q-2 I . t . I . I . I . I . I . I . I . I . I . I . I ,
0 4 8 12 16 20 24
Hour
I
102
10'
10°
03=50 ppb. N0)[=350 ppb
I • I • I ' I • I • i ' I ' I ' I ' I ' I ' I
S 10-1
JQ-Z I . t . I . I . I . I . I . I . I . I . I . I . I .
0 4 8 12 16 20 24
Hour
03=80 ppb, NOX=350 ppb
u.
D
102
10'
10°
S 10-1
g
io-2
i • i"n' i ' i ' i• i • i ' i • i •i ' i
8 12 16 20 24
Hour
IO2
03=200 ppb. NOX=350 ppb
b.
b.
D
° 10'
u
O
3
O
J3
10° -
a
L,
» 10-'
i i^ i • ii i \ i ^ii" i
. i . i . i . i . i . i . i . i . i . i . i .
4 8 12 16 20 24
Hour
Figure 3-12(b).
SO2 loss rate for CALPUFF for NOX = 350 ppb.
Note that each box represents variations due to season (temperature,
relative humidity, wind speed, and cloud cover).
3-49
-------
CALPUFF
03=20 ppb, NOX=1400 ppb
iu-
b.
b.
P
23
^ 10'
(%/hour) !
»-*
°0
0>
e
i'o-
(V
O
01
in-z
• i • i • i •
-
.1
•
-
.1.1.
• i ' i
I f( ft f
V ¥ ' <
L
I
' 1
1
*
' 1 ' 1 ' 1 ' 1 '
-
Ill -
y
.1.1.1.1.
8 12 16 20 24
Hour
fc
§
3
102
101
10°
03=50 ppb. NOX=1400 ppb
10-'
ID'2
1 ^T I ' I I I I ' I ' i ' I I I
0 4 8 12 16 20 24
Hour
102
03=80 ppb, NOX=1400 ppb
u.
b.
u 101
L.
O
o
tn
10° -
10-2
12 16 20 24
Hour
108
u.
u.
10'
10°
l>
a
u
I 10-'
o
CO
03=200 ppb. NOX=1400 ppb
10-
i ' i i ' i i r T • i \
12 16 20 24
Hour
Figure 3-12(c). SO2 loss rate for CALPUFF for NOX = 1400 ppb.
Note that each box represents variations due to season (temperature,
relative humidity, wind speed, and cloud cover).
3-50
-------
CALPUFF
03=20 ppb. NOX=7 ppb
£ iu-
|
u
•2 10'
3
O
10°
o
10-
O
n
ID"2
T T
12 16 20 24
Hour
t io2
1
u
o
ts°
-' 10°
o
| IO-1
03=50 ppb. NOX=7 ppb
w
10
-2
12 16 20 24
Hour
03=80 ppb, NOX=7 ppb
2j
o
•2 io1
10°
10-'
o
CO
1^ I 1^ I I I n I I I t 1
8 12 16 20 24
Hour
fc 102
CJ
•2 10'
3
O
•
03=200 ppb, NOX=7 ppb
10°
C
o
"5 10-'
o
in
10-
8 12 16 20 24
Hour
Figure 3-13(a).
SO4 formation rate for CALPUFF for NOX = 7 ppb.
Note that each box represents variations due to season (temperature,
relative humidity, wind speed, and cloud cover).
3-51
-------
CALPUFF
03=20 ppb, NOX=350 ppb
03=50 ppb, NOX=350 ppb
fc. 102
fc. lu
D
£3
u
^
^ 10'
IT
3
0
.e
*~ 10°
^
G
e
o
| 10-'
t.
8*
1Q-2
C
; ' i ' 1 ' I '
-
flf
_
•
-
.1.1.
) 4
i '
M
•
Y*
¥
8
1 i '
11 1 1
• • •
r T ¥ V
12
Hour
1 ' ' 1 ' 1 ' 1 ' :
-;
ill
ml
-
1
r ' n
v . u :
-
.1.1.1.
16 20 24
&1 n2
10
p
a
<
u
u
£ 10'
t.
o
j:
\
b 10°
V
"e
^
c
0
1 10-'
|
V
1O*2
1U
c
-
_
-
.1.1.
4
• il
nOn
nnVi
U
u
.1.1
8
fifi
U
VV1]
12
Houi
-
11.
1
rl n
Ml
I] LJ
_
.1.1.1.1.1.
16 20 24
li.
u.
I
L)
102
03=80 ppb. NOX=350 ppb
03=200 ppb, NOX=350 ppb
£ 10'
O
tn
10°
10-
1-2 I . ! . I . I . t . I . 1 . I . I . I . | . | . | .
0 4 8 12 16 20 24
Hour
b.
a
10'
10°
c
o
I 10-'
. i . i . i . i . i . i . i . i . i . i . i . i .
0 4 8 12 16 20 24
Hour
Figure 3-13(b).
SO4 formation rate for CALPUFF for NOX = 350 ppb.
Note that each box represents variations due to season (temperature,
relative humidity, wind speed, and cloud cover).
3-52
-------
CALPUFF
03=20 ppb. NOX=1400 ppb
03=50 ppb. NOX=1400 ppb
£ 10*
|
L)
t.
fi 10'
^~.
o
\
^ 10°
01
"o
u
c
0
| 10-'
u
_£
O
in-2
' 1 ' '
-
nfl
-
•
•
-
..i.i.i.
ijifti)/
• • '
y v»i
r
. i .
1 A 1 l
• • .
' > T /
' 1 ' 1 ' 1 ' 1 '
-
(Idl
;
n :
v U ,
-
.1.1.1.
b. tu
1
U
k
£ 10'
t.
3
O
\
^ 10°
0)
"o
o
| 10-'
u
•2
^
W io-2
8 12 16 20 24
Hour
w~
10'
10°
10-'
in-2
lu
(
1 1 ' 1 '
_
-
.1.1.
) 4
1 l ' f
• 11
Aftn .
[Jll
|p
. i .
8
ill (III:
-!
iiiii i :
H1 yfini
v uuBul
"DUD •
-.
.1.1.1.1.1.1.
12 16 20 24
Hour
10'
[t.
I
~ 10°
11
•^
c
10-
c
to
10-
03=80 ppb. NOX=1400 ppb
i ^ j . i . i . i . i . i . ]
8 12 16 20 24
Hour
03=200 ppb. NOX=1400 ppb
b.
b.
|
U
£ 10'
-" 10°
u
c
o
I io-1
o
V)
10
-2
1 ' 1 • 1 • 1 ' 1 ' 1 ' 1 ' 1 1 1 1 1
12 16 20 24
Hour
Figure 3-13(c).
SO4 formation rate for CALPUFF for NO, = 1400 ppb.
Note that each box represents variations due to season (temperature,
relative humidity, wind speed, and cloud cover).
3-53
-------
CALPUFF
03=20 ppb, NOX=7 ppb
03=50 ppb, NOX=7 ppb
for CALPUFF
3 c
M
•3
O
x. 10°
V
e
i
io-2
(
' i ' i '
-
-
.1.1.
• i ' i • i • i • i • i
.1.1.1.1.1.1
• i • i • i • i
-
-
.1.1.1.
) 4 8 12 16 20 24
for CALPUFF
»-• t-
° c
L.
3
O
x 10°
41
e
o"
Z
ID"2
C
' i ' i ' i
_
-
.1.1.1
• i ' i • r • i • i
.i.i.i.i.i.
• i • i •
-
-
.1.1.1.
) 4 8 12 16 20 24
Hour Hour
IO2
b.
IL.
3
U 10'
o
u
3
O
S 10°
o>
e
j,o"
K
O
z
io-2
03=80 ppb. NOX=7 ppb
• i • i • i
-
-
_
.1.1.
' l ' i ' i ' i • i •
HBBDQODD
.i.i.i.i.i.
• i • i • i •
.1.1.1.
3 4 B 12 16 20 24
102
fe
D
D.
" 10'
£
L.
3
O
x 10°
JS
V
a
K
O
z
io-2
(
03=200 ppb. NOX=7 ppb
: ' 1 ' 1 '
-
-
,1.1,
JOB0B0B8BGD
• 1 • 1 • 1 • :
L \
-
;
-
. i . i . i . i . i . i . i . i . i .
D 4 8 12 16 20 24
Hour Hour
Figure 3-14(a).
NOX loss rate for CALPUFF.
Note that each box represents variations due to season (temperature,
relative humidity, wind speed, and cloud cover).
3-54
-------
CALPUFF
3
O
Si
K
10'
10°
o
10-
03=20 ppb, NOX=350 ppb
1I ' I ^T II ' I I I I
12 16 20 24
Hour
03=50 ppb, NOX=350 ppb
3
O
.C
10'
10°
UT'
O
io-2
i • i i ' i • i • ~ i i i ' i * i • i
4 8
12 16
Hour
20 24
03=80 ppb, NOX=350 ppb
u.
u.
D
0.
J
<
U
3
O
£
10*
0
10°
o
z
11 T I I ~ 1 1 ™ ~ 1
12 16
Hour
20 24
10*
u.
b.
10'
10°
i 10-'
03=200 ppb, NOX=350 ppb
O
JQ-2 I . I . I . I . I . I . I . I . I . I . I .
0 4 8 12 16 20
Hour
24
Figure 3-14(b).
NOX loss rate for CALPUFF.
Note that each box represents variations due to season (temperature,
relative humidity, wind speed, and cloud cover).
3-55
-------
CALPUFF
03=20 ppb. NOX=1400 ppb
03=50 ppb. NOX=1400 ppb
iu-
b.
0_
J
o 10i
u
o
L.
3
0
S lo°
0>
o
J10-i
M
O
io-2
c
: 1 t 1
r .
r-
.1.1.
) 4
' I ' I ' I ' i ' I •
mmm
UUUUU UUUULI
.i.i.i.i.i.i
8 12 16
Hour
• i • I • i •
-.
-.
i i i
20 24
iir-
u.
r »
10
o
a
\ 10°
j?
t>
u
n ,
m 1Q-1
o
K
0
in-2
1U
C
i i i
-
-
.1.1.
> 4
i i i ' i ' i ' i
mi
.1.1.1.1.1.1
8 12 16
Hour
' 1 ' 1 ' 1 • :
-
;
-.
•
i.i i
20 24
IO2
10'
10° -
£
D
i
i 10- t-
03=80 ppb, NOX=1400 ppb
1 1 ' 1 ' 1 ' 1 ' 1 ' 1 ' 1 ' 1 ' 1
8 12 16 20 24
Hour
102
03=200 ppb. NO =1400 ppb
b.
b.
3
O
10°
O
io-z
1 1 1 ' 1 ' 1 " 1 ' i ' 1 * 1 ' 1
8 12 16 20 24
Hour
Figure 3-14(c).
NO, loss rate for CALPUFF.
Note that each box represents variations due to season (temperature,
relative humidity, wind speed, and cloud cover).
3-56
-------
CALPUFF
03=20 ppb. NOX=7 ppb
03=50 ppb. NOX=7 ppb
B; 1U~
o
o
- 10'
•s
0
A
\
7 10°
e
TN03 formation
*- t—
0 0
1 1
M i-
' ' 1 ' 1 1 ' 1 ' 1 ' 1 ' 1 1 ' 1 ' ! ' 1 ' I '
-
-
-
. i , i i
BBBB8BBBDD
.i.i.i.i.i.i
-
-
.1.1.1.
) 4 8 12 16 20 24
O
o
- 10'
o
\
7 10°
a
TN03 formation
0 0
i i
M *•
1 1 1 1 ' 1 ' I ' 1 ' 1 " 1 ' 1 ' 1 ' 1 "
-
-
-
.1.1.
DBBBBBBBDI
.i.i.i.i.i.
.
_
_
.it i
) 4 8 12 16 20 24
Hour Hour
CALPUFF
o
(o
u
o
I 10'
3
O
.C
\
7 10°
o
c
0
1 10-'
L.
0
n
O
03=80 ppb, NOX=7 ppb
, ' i ' i ' t
_
Dnnnnnnnnn
udDdDduUU
' 1 ' 1 ' 1 ' :
-
-
.1.1. . .i.i.i.i. . i.i
CALPUFF
%
u
o
- 10'
o
X
7. 10°
-------
I
3
&. sr o
Ul
oo
TN03 formation rate (%/hour) for CALPUFF
TN03 formation rate (%/hour) for CALPUFF
00
X w
o
5
to
0
M
i - -» »-*
O
TN03 formation rate (?!/hour) for CALPUFF Cl
O O i— i— ^ t-ft
O
u
II
ro
o
o
T3
•a
cr
o
en
o
•o
X)
er
f
O
U
II
en
o
•o
n
tr
Z
o
CO
en
o
•o
•a
cr
-------
CALPUFF
03=50 ppb. NOX=1400 ppb
fc, „-
fc 10Z
! 10.
u
3
O
x:
7 10°
j->
(0
c
0
|io-'
1,
o
c
a
1 i • i ' i
_
„
-
I ' •
) 4
X
' 1 ' 1 ' 1 1 1 1
PflpOQflpOflp
UUUUuUUuUU
8 12 16
Hour
III.
_
-
-
.1.1.1.
20 24
Du 10^
D
u
£
3
O
X
7 10°
"a
c
o
1 10-'
-£
1
{
o
k.
f 10'
0
7 10°
a
c
O
c 10"'
L.
O
n
O
H
(
J
' 1 ' 1 ' 1
•
.1.1.
) 4
r f - • - - x
' 1 ' [ ' 1 ' 1 1 1
•nnnnnnnnnn
uutitiHtjtjyuu
j
.1.1.1.1.1.
8 12 16
Hour
* *
III:
-
_
.1.1.1.
20 24
£*•
D
a
u
o
I 10'
3
O
7 10°
o
c
Q
1 10-'
C
L.
0
O
H _g
- C
«J
' 1 ' 1 ' 1
n
-
_
.1.1.1
) 4
• - *
' 1 ' 1 ' 1 ' 1 ' 1 ' 1
IJPPPPPPPIJII
.i.l.l.l.l.
8 12 16
Hour
' 1 ' 1 ' 1 :
-:
-
.1.1.1.
20 24
Figure 3-15(c).
TNO3 formation rate for CALPUFF.
Note that each box represents variations due to season (temperature,
relative humidity, wind speed, and cloud cover).
3-59
-------
ARM3
102
CO
1
3
o
f.
t? 10°
o 10-1
o1
w
ID"2
No Dependence on 03 and N0_
1 • 1 • 1 • 1 • 1 • 1 ' 1 ' 1 ' 1 ' 1 ' 1
_L
I . I . I • I . I
12 16
Hour
20
24
Figure 3-16.
SO2 loss rate for ARMS.
Note that each box represents variations due to season (temperature, relative
humidity, wind speed, and cloud cover).
3-60
-------
ARMS
No Dependence on 03 and NOX
10"
I
"£
O
-*-
£
0 10°
O
I*
c
O
c
E 10-'
O
d'
en
i r»— Z
10 "
(
' | ' | ' | ' 1 ' 1 ' 1 ' 1 ' 1 ' 1 ' 1 ' 1 ' 1 ' ;
nfjfjfjfjn
i nfl()U Uljjg '•
_ —
. I . I . i . I . i . I . I , 1 . I . I . 1 . 1 .
D 4 8 12 16 ZQ 24
Hour
Figure 3-17. SO4 formation rate for ARM3.
Note that each box represents variations due to season (temperature, relative
humidity, wind speed, and cloud cover).
3-61
-------
ARM3
03=20 ppb
10Z
K
fe 101
10°
° 10-'
o*
io-2
I \ 1 [ ' I ' I It I I I I
. 1
12 16
Hour
D
O
.c
102
10'
10°
03=50 ppb
.2 10-'
o"
10
-2
8 12 16 20 24
Hour
102
03=80 ppb
o
SI
10' -
10°
° 10-'
O*
I • I I ^
I ' i ' i • i I ' I
JQ-2 I . I . I . I . I . I . I . I . I . I . I . I . I .
0 4 8 12 16 20 24
Hour
102
03=200 ppb
O
j:
10°
° 10-'
o
z
10
-2
i ' i i ' i ill l i II l
12 16
Hour
20 24
Figure 3-18. NOX loss rate for ARMS.
Note that each box represents variations due to season (temperature, relative
humidity, wind speed, and cloud cover).
3-62
-------
ARM3
03=20 ppb
102 r-
3
O
ft
5 100
! 10-'
ID'
12 16
Hour
10Z
• 10'
o
03=50 ppb
o
10-'
I I I 111 I T I ' I ' I ' I
8 12 16 20 24
Hour
102
Z
K
•D 10'
3
O
.C
•x
je
5 10°
(0
o
10-
03=80 ppb
8 12 16 20 24
Hour
102 c
03=200 ppb
10
24
Figure 3-19. TNO3 formation rate for ARMS.
Note that each box represents variations due to season (temperature, relative
humidity, wind speed, and cloud cover).
3-63
-------
102
CO
OS
10
10
-2
Conversion Rates (%/hour) for CALPUFF
101 -
10° -
-1 _
24
Figure 3-20(a).
Diurnal distribution of conversion rates for SO2, SO4, NOr and TNO3
predicted by the CALPUFF model assuming (1) concentrations of O3 and
NOX equal 50 ppb and 7 ppb, respectively, (2) ambient temperature =
303° K, and (3) relative humidity = 50%.
3-64
-------
102
101
o
.fl
(D
-4_)
CO
10°
10
-1
10
-2
Conversion Rates (%/hour) for CALPUFF
0
8
12 16
Hour
20
24
Figure 3-20(b).
Diurnal distribution of conversion rates for SO2, SO4, NOr and TNO3
predicted by the CALPUFF model assuming (1) concentrations of O3 and
NOX equal 50 ppb and 7 ppb, respectively, (2) ambient temperature =
303° K, (3) relative humidity = 50%, and (4) the nighttime conversion
rate for NOX is set to 0.093% hour as the ARMS model.
3-65
-------
102
101
10°
CO
10
-i _
10
-2
0
Conversion Rates (%/hour) for ARM3
1 I ' I ' I ' I ' I '
so2
S04
N0
TNO
8
12 16
Hour
20
24
Figure 3-21. Diurnal distribution of conversion rates for SO2, SO4, NOr and TNO3 predicted
by the ARM3 model assuming (1) concentrations of O3 and NOX equal 50 ppb
and 7 ppb, respectively, (2) ambient temperature = 303° K, and (3) relative
humidity = 50%.
3-66
-------
4. MODEL EVALUATION WITH CAPTEX TRACER DATA
The Cross-Appalachian Tracer Experiment (CAPTEX-83) provides a database for
assessing the combined effectiveness of using various wind-field modeling approaches with a puff
model to simulate the transport and diffusion of a tracer-cloud over distances of hundreds of
kilometers. CAPTEX is comprised of seven individual experiments in which tracer was released
from either Dayton, Ohio or Sudbury, Ontario (Ferber et al., 1986). In this section, the results
obtained when simulating one release from Dayton (CAPTEX #3), and two releases from
Sudbury (CAPTEX #5 and #7), using MESOPUFF n and CALPUFF, are described.
4.1 Model Application to CAPTEX
PMCH tracer was released from ground-level for a 3-hour period, and subsequent
ambient concentrations were obtained at up to 86 sampling locations spread across Ohio,
Pennsylvania, New Jersey, New York, New England, and southeastern Canada. Each sampler
collected six sequential samples, for either 3-hour or 6-hour periods (the samplers nearer the
release used the 3-hour sampling period). Consequently, the 3-hour samplers monitored an 18-
hour period, and the 6-hour samplers monitored a 36-hour period. Concentrations of the
PMCH tracer are reported in fL/L (1.56 x 10'" g/m3).
Meteorological data for the three periods used in this evaluation were obtained from 122
NWS surface locations reporting hourly and 13 "upper-air" stations reporting twice-daily
throughout the region. Furthermore, hourly profiles of meteorological data from the MM4-
FDDA model were available on an 80-km grid. Three wind field models were used to obtain a
gridded field of meteorological data with a horizontal resolution of 18 km: MESOPAC II,
CALMET, and CALMET using the MM4-FDDA fields as its "STEP-1" field. In the discussion
to follow, the latter modeled wind field is referred to as "CALMET/MM4".
MESOPUFF II is driven by each of the modeled wind fields to produce a set of three
simulations of ground-level concentrations for each of the three trials (9 simulations in all).
MESOPUFF II produced hourly concentration values in g/m3, at each of the CAPTEX
monitoring locations. Significant MESOPUFF II options were set as follows:
- Puffs/hour 8
- Minimum samples/hour 8
- Variable sampling rate True
I:\a314\report\a314rev\iect4.wph 4-1
-------
- Vertical distribution (puff) Gaussian
- Dispersion coefficients PG coefficients ( < 10 km travel distance)
Heffter eqns. ( > 10 km travel distance)
- Wind-field 2-layers (MESOPACII: mixed-layer average
wind (lower), mixing ht. to 700 mb ave.
(upper); CALMET: mixed-layer ave.
(lower), mixing ht. to 3000 m ave. (upper))
MESOFILE n was used to obtain 3-hour and 6-hour average concentrations in fL/L
from the hourly concentrations, and a specific post-processing program was developed to pair
the MESOPUFF H concentrations with valid measured concentrations.
CALPUFF was driven by two of the modeled wind fields, CALMET and
CALMET/MM4. The resulting hourly concentrations were processed in the same way as those
from MESOPUFF n. The model options used in the CALPUFF runs included:
- Maximum puff travel distance 0.10 met. grid units
during one sampling step (1.8 km)
(controls the sampling rate)
- Sampling function Integrated puff algorithm
- Vertical distribution (puff) Gaussian
- Dispersion coefficients PG coefficients ( < 10 km travel distance)
Heffter eqns. ( > 10 km travel distance)
- Wind-field 1 Mayers (cell face heights: 0, 20, 80, 120,
280, 520, 1080, 1400, 1800, 2200, 3000,
4000 m)
In both the MESOPUFF n and CALPUFF runs, chemical transformation, dry
deposition, and wet removal were turned off (not modeled) because the CAPTEX tracer gas
(PMCH) is inert and non-depositing. The models were run in "hands off mode, with most
model variables assigned to their default values. No attempt was made to optimize the choice of
model options, such as the dispersion coefficients, mode of incorporation of MM4-FDDA data
I:\a314\repott\a314rev\Kct4.wph 4-2
-------
into the CALMET model (e.g., as an initial guess field, Step 1 field, or observations), and
meteorological model assumptions (e.g., grid size, vertical layer structure). Also, CALPUFF was
run in a mode designed to make it most like MESOPUFF n in order that the effects of
different wind fields and data inputs could be identified.
4.2 Model Comparison Measures
The five wind field/puff modeling approaches are evaluated by means of statistical
measures, and graphical displays. The measures include:
O mean observed concentration (fL/L)
P mean predicted concentration (fL/L)
SD standard deviation (of either O or P) (fL/L)
d average residual = O - P (fL/L)
AAR average absolute residual (fL/L)
RMSE square root of the mean square error (fL/L)
R Pearson correlation coefficient
FB fractional bias of the means = 2 (O - P)/(O + P)
FS fractional bias of SD's = 2 (SD0 - SDp)/(SD0 + SDp)
These measures are applied to two groups of concentrations within each trial:
(1) time and space-paired concentrations (all periods)
(2) peak concentrations at each monitor
All occurrences in which both the modeled and observed concentrations are zero are retained in
the analysis.
When applied to all data from a trial (Group 1), information about how each simulation
matches each observed value is obtained. This comparison is "strict," in that spatial or temporal
"errors" in the simulation will degrade its score. A less strict measure of performance is
obtained from analysis of peak concentrations paired in space (Group 2). This is equivalent to
comparing the observed and modeled "footprint" of peak concentrations during a trial.
Graphical displays include the "footprints" of peak concentrations, and profile-plots of
the peak concentrations as a function of distance. These plots are readily used to interpret the
statistics of the peak concentrations (paired in space).
I:\a314\report\t314rev\tect4.wph 4-3
-------
4.3 Results
"Footprints" of peak concentrations from each valid monitoring location are shown in
Figures 4-1 through 4-18. The first group of 6 figures show the observed and modeled
concentrations for CAPTEX #3; the other two experiments (#5, #7) follow in order. Numbers
displayed along the borders label the 18 km grid used in all the simulations. Pertinent
information about these figures includes;
Grid-cell dimension 18 km
Sudbury location (x,y) (41.19, 76.71)
Dayton location (x,y) (30.06, 35.36)
Concentration units fL/L
Isopleths 10 and 100 fL/L
For CAPTEX #3 (Dayton release, indicated with a "D" in the figures), the following can
be observed.
1) MESOPAC n drives the plume directly across the sampling network, toward the
NE initially and then the E, which differs substantially from the observed pattern.
2) The CALMET wind fields using NWS data only results in plume transport more
to the NE across the grid. Although the location of the high observed
concentrations in southern Vermont is better characterized than with MESOPAC
13, the low observed concentrations in eastern Ohio, Pennsylvania, and western
New York are not reproduced.
3) The use of MM4-FDDA data in CALMET substantially improved the
concentration pattern with both MESOPUFF n and CALPUFF over runs using
NWS data only.
4) The overall concentration patterns produced by MESOPUFF n and CALPUFF
are similar when driven by the same wind field model, although CALPUFF
produces some differences in the near field. The reasons for this are discussed
below.
I:\a314\report\a314rev\iert4.wpb 4-4
-------
90-
CAPTEX #3
OBSERVED (fl/f)
9
o:
o
00
25
I j 1 i I I i ! I I I j I 1 I I I I ) 1 ! |
85
18 km GRID UNITS
Figure 4-1. Distribution of peak observed concentrations (fL/L) during CAPTEX #3. The
location of the CAPTEX tracer release points, Sudbury (S) and Dayton (D), are
also shown.
I:\«314\rrport\«314rev\ieet4.wph
4-5
-------
90
CAPTEX #3
MESOPUFF (MESOPAC)
30
45 65
18 km GRID UNITS
85
Figure 4-2. Distribution of peak concentrations (fL/L) produced by MESOPUFF n using
MESOPAC n wind fields for CAPTEX #3. The location of the CAPTEX tracer
release points, Sudbury (S) and Dayton (D), are also shown.
I:\«314\report\«314rev\iecU,wph
4-6
-------
90-i
CAPTEX #3
MESOPUFF (CALMET)
30
45 65
18 km GRID UNITS
Figure 4-3. Distribution of peak concentrations (fL/L) produced by MESOPUFF n using
CALMET wind fields for CAPTEX #3. The location of the CAPTEX tracer
release points, Sudbury (S) and Dayton (D), are also shown.
I:\a314\report\a314rrv\Kd4.wph
4-7
-------
90^-
CAPTEX #3
CALPUFF (CALMET)
30
25
45 65
18 km GRID UNITS
Figure 4-4. Distribution of peak concentrations (fL/L) produced by CALPUFF using
CALMET wind fields for CAPTEX #3. The location of the CAPTEX tracer
release points, Sudbury (S) and Dayton (D), are also shown.
I:\«314\report\a314rev\iect4.wph
4-8
-------
90-i
CAPTEX #3
MESOPUFF (CALMET/MM4)
30
25
18 km GRID UNITS
Figure 4-5. Distribution of peak concentrations (fL/L) produced by MESOPUFF n using
CALMET/MM4 wind fields for CAPTEX #3. The location of the CAPTEX
tracer release points, Sudbury (S) and Dayton (D), are also shown.
J:\a3H\report\»314trv\»ect4.wph
4-9
-------
90
CAPTEX #3
CAi_PUFF (CALMET/MM4)
t
00
tz
9
o
|
oo
30
45 65
18 km GRID UNITS
85
Figure 4-6. Distribution of peak concentrations (fL/L) produced by CALPUFF using
CALMET/MM4 wind fields for CAPTEX #3. The location of the CAPTEX
tracer release points, Sudbury (S) and Dayton (D), are also shown.
I:\«314\rejxHt\«314rev\teel4.wpb
4-10
-------
90-i
ID
Q
O
00
50-
CAPTEX #5
OBSERVED (fl/f)
loo
30
45 65
18 km GRID UNITS
85
Figure 4-7. Distribution of peak observed concentrations (fL/L) during CAPTEX #5. The
location of the CAPTEX tracer release points, Sudbury (S) and Dayton (D), are
also shown.
l:\«314\reportVt314rev\Kd4.wpb
4-11
-------
90 -i
CAPTEX #5
MESOPUFF (MESOPAC)
E
j*:
00
30
45 65
18 km GRID UNITS
85
Figure 4-8. Distribution of peak concentrations (fL/L) produced by MESOPUFF n using
MESOPAC H wind fields for CAPTEX #5. The location of the CAPTEX tracer
release points, Sudbury (S) and Dayton (D), are also shown.
I:\«314\report\«314rev\iect4.wph
4-12
-------
90-n
CAPTEX #5
MESOPUFF (CALMET)
//
",/
00
t;
z:
ID
Q
o:
o
00
45 65
18 km GRID UNITS
85
Figure 4-9. Distribution of peak concentrations (fL/L) produced by MESOPUFF n using
CALMET wind fields for CAPTEX #5. The location of the CAPTEX tracer
release points, Sudbury (S) and Dayton (D), are also shown.
I:\i3W\report\»31«rcv\«KJ4,wph
4-13
-------
90-i
CAPTEX #5
CALPUFF (CALMET)
Z)
Q
cr
o
oo
30
45 65
18 km GRID UNITS
Figure 4-10. Distribution of peak concentrations (fL/L) produced by CALPUFF using
CALMET wind fields for CAPTEX #5. The location of the CAPTEX tracer
release points, Sudbury (S) and Dayton (D), are also shown.
I:\a314\report\a314rrv\iect4.vph
4-14
-------
90-1
CO
t
z:
Z)
O
ct:
o
co
CAPTEX #5
MESOPUFF (CALMET/MM4)
30
45 65
18 km GRID UNITS
Figure 4-11. Distribution of peak concentrations (fL/L) produced by MESOPUFF n using
CALMET/MM4 wind fields for CAPTEX #5. The location of the CAPTEX
tracer release points, Sudbury (S) and Dayton (D), are also shown.
I:\«3M\i*port\«31*rev\ieei4.wph
4-15
-------
90-!
CAPTEX #5
CALPUFF (CALMET/MM4)
45 65
18 km GRID UNITS
85
Figure 4-12. Distribution of peak concentrations (fL/L) produced by CALPUFF using
CALMET/MM4 wind fields for CAPTEX #5. The location of the CAPTEX
tracer release points, Sudbury (S) and Dayton (D), are also shown.
I:\«3M\report\«314rev\«ee«4.wph
4-16
-------
CAPTEX #7
OBSERVED (fl/f)
30
45 65
18 km GRID UNITS
Figure 4-13. Distribution of peak observed concentrations (fL/L) during CAPTEX #7. The
location of the CAPTEX tracer release points, Sudbury (S) and Dayton (D), are
also shown.
I:\a314\report\i314rev\iecl4.wph
4-17
-------
90-
CAPTEX #7
..MESOPUFF (MESOPAC)
in
in
z:
ID
Q
C£
O
00
30
45 65
18 km GRID UNITS
Figure 4-14. Distribution of peak concentrations (fL/L) produced by MESOPUFF H using
MESOPAC H wind fields for CAPTEX #7. The location of the CAPTEX tracer
release points, Sudbury (S) and Dayton (D), are also shown.
I:\a314\report\s314rcv\Kxa4.wpb
4-18
-------
90-i
CAPTEX #7
MESOPUFF (CALMET)
25
45 65
18 km GRID UNITS
85
Figure 4-15. Distribution of peak concentrations (fL/L) produced by MESOPUFF n using
CALMET wind fields for CAPTEX #7. The location of the CAPTEX tracer
release points, Sudbury (S) and Dayton (D), are also shown.
I:\a314\report\a314rcv\iect4.wpb
4-19
-------
CO
^
2
ID
O
E
00
70-
CAPTEX #7
CALPUFF (CALMET)
45 65
18 km GRID UNITS
85
Figure 4-16. Distribution of peak concentrations (fL/L) produced by CALPUFF using
CALMET wind fields for CAPTEX #7. The location of the CAPTEX tracer
release points, Sudbury (S) and Dayton (D), are also shown.
I:\a314\report\a314rev\sect4.wph
4-20
-------
90-i
(f)
70-
cr
o
E
oo
CAPTEX #7
MESOPUFF (CALMET/MM4)
tg&v'ify
25
45 65
18 km GRID UNITS
Figure 4-17. Distribution of peak concentrations (fL/L) produced by MESOPUFF n using
CALMET/MM4 wind fields for CAPTEX #7. The location of the CAPTEX
tracer release points, Sudbury (S) and Dayton (D), are also shown.
I:\a314\report\a314rev\KCt4.wph
4-21
-------
90 -i
30
25
CAPTEX #7
PALPUFF (CALMET/MMA)
45 65
18 km GRID UNITS
85
Figure 4-18. Distribution of peak concentrations (fL/L) produced by CALPUFF using
CALMET/MM4 wind fields for CAPTEX #7. The location of the CAPTEX
tracer release points, Sudbury (S) and Dayton (D), are also shown.
I:\a314\report\a314rev\Mct4.wpb
4-22
-------
One of the differences between MESOPUFF H and CALPUFF is in the way puffs are
advected. MESOPUFF n uses a two-layer wind field, the lower layer for transport of puffs
within the mixed layer, and an upper layer field for winds above the mixing height. For a
surface release, as in CAPTEX, MESOPUFF n will always use the lower layer (mixed-layer
averaged) wind field. For situations with substantial wind shear such in CAPTEX #3, the use of
the mixed layer averaged field errors in the transport direction, even if the meteorological
model's winds are representing the actual vertically averaged winds well. CALPUFF uses a
different technique to determine the advective wind which is better suited for treating the near-
field transport direction. CALPUFF internally computes each sampling step, a transport wind
averaged over the depth of the plume from the multi-layer winds provided to it from CALMET.
Thus, the transport wind for a surface release will be determined initially from surface wind
field. As the plume grows in the vertical, the depth through which the wind is averaged to
provide the transport wind is increased, until eventually the plume becomes well mixed in the
vertical and the mixed-layer averaged wind is used (as in MESOPUFF II). A comparison
CALMET/MM4 runs in Figures 4-5 and 4-6 with Figure 4-1 at the closest receptors to the
source suggests that CALPUFF is transporting the plume in a more northerly direction initially,
closer to the observed pattern. The MESOPUFF n plume travels slightly more to the NE.
For CAPTEX #5 (Sudbury release indicated with an "S" in the figures), the following
observations are made from the figures.
1) MESOPAC H/MESOPUFF D results are generally of the correct magnitude, but
the transport pattern is clearly wrong (toward the SSE over the entire trajectory
rather than shifting to the E and ESE).
2) The CALMET winds shifted more to the SE, which is an improvement over the
MESOPAC II pattern, but does not reproduce the observed shift to mostly
easterly transport.
3) CALMET/MM4 results match the magnitude of the concentrations better, and
the overall pattern of the footprint is more to the E as seen in the observed
concentration field, but the apparent sharp bend in the observed footprint is still
missing.
4) All simulated footprints appear broader than the observed.
I:\«314\report\»314rev\«ect4.wph 4-23
-------
For CAPTEX #7 (Sudbury release), the following can be seen in the plots.
1) The peak concentrations are observed along a string of receptors located to the
SSE of the source.
2) The MESOPAC n produced winds drive the plume to the E of the observed
plume, resulting in little overlap with the observed footprint.
3) CALMET-driven simulations with CALPUFF and MESOPUFF H correctly
produces peak concentrations along the string of receptors where they are
observed.
4) The use of MM4-FDDA data in CALMET pushes the centerline of the plume
too far to the east, although the observed and predicted plumes still shown
significant overlap of their footprints.
One conclusion which can be drawn from the simulations is that the selection of the wind
field model and the data driving it have a larger influence on model performance that the choice
of the dispersion model. The differences between CALPUFF and MESOPUFF n are most
significant in the near-field of a source, which was not the focus of the CAPTEX experiments.
Also, the options used in CALPUFF for these simulations were selected to be "MESOPUFF II-
like" so that the effects of different wind field models and data inputs could be isolated and
identified.
Generally, the performance of the CALMET-driven simulations is better than those
derived using MESOPAC n winds, and the use of MM4-FDDA data in CALMET improves the
performance over simulations using observations only. This is especially apparent in CAPTEX
#3, where the MM4-FDDA data, significantly altered (and improved) the concentration
predictions. An exception is CAPTEX #7, where CALMET without MM4-FDDA data
performed the best.
The statistical performance measures for the data paired in time and space are listed in
Table 4-1. For CAPTEX #3, the performance statistics improve as one goes from using
MESOPAC H to CALMET, and then CALMET/MM4. The average residual for the
MESOPUFF n simulations with winds from MESOPAC H, CALMET, and CALMET/MM4
improve from -35, -31, to -10 fL/L, respectively. Similar improvement is indicated in the
average absolute residual (AAR) and root mean square error (RMSE) statistics. A dramatic
improvement in the correlation coefficient occurs when MM4-FDDA data are used (from -.02
I:\a314\rcport\a314rev\iea4.wph 4-24
-------
Table 4-1
Performance Measures for all CAPTEX Concentrations Paired in Time and Space
MESOPUFF n Simulations
Winds:
Trial:
N
O
SD0
P
SDp
d
AAR
RMSE
R
FB
FS
MESOPAC D
#3
394
4.0
20.0
39.4
235
-35.4
42.1
239
-.010
-1.63
-1.69
#5 #7
320 271
9.7 13.8
54.7 60.5
12.7 153
60.6 88.9
-3.0 -1.5
22.0 28.9
823 109
-.013 -.038
-.27 -.10
-.10 -38
#3
394
4.0
20.0
34.9
151
-30.8
37.6
156
-.019
-1.59
-153
CALMET
#5
320
9.7
54.7
252
972
-15.6
322
110
.064
-.89
-.56
CALMET/MM4
#7
271
13.8
605
14.9
94.0
-1.1
12.7
70.7
.659
-.08
-.43
#3
394
4.0
20.0
14.0
109
-10.0
16.1
963
.693
-1.11
-138
#5
320
9.7
54.7
19.7
85.8
-10.0
23.6
885
279
-.68
-.44
#7
271
13.8
605
31.4
206
-17.6
37.9
208
.134
-.78
-1.09
I:\a314\fcpoit\i314nv\Kd4.«pb
4-25
-------
Table 4-1 (Continued)
Performance Measures for all CAPTEX Concentrations Paired in Time and Space
CALPUFF Simulations
Winds:
CALMET
CALMET/MM4
Trial:
#3 #5 #7
#3 #5 #7
N
393 320 270
343 320 270
O
SDn
4.0 9.7 13.8 4.0 9.7 13.8
20.1 54.7 60.6 20.1 54.7 60.6
37.9 21.7 11.1 11.4 17.4 22.4
238 98.4 99.0 89.7 79.4 155
-33.9 -12.0 2.7
-7.4 -7.8 -8.6
AAR
40.9 29.7 16.3
13.6 21.1 29.0
RMSE
242 112 93.2
78.1 85.5 157
-.021 .031 399
.669 236 .155
FB
FS
-1.62 -.77 .22 -.96 -.57 -.47
-1.69 -.57 -.48 -127 -37 -.87
l.\»3H\report\»3Urev\»ect4.wph
4-26
-------
without MM4-FDDA data to .69 with MM4-FDDA data). The CAPTEX #3 simulation with
CALPUFF shows similar improvement when MM4-FDDA data are used with CALMET.
CALPUFF tends to perform slightly better than MESOPUFF H in the CALMET/MM4
simulations, based on the average bias, average absolute bias, and RMSE statistics. The values
of d, AAR, and RMSE for CAPTEX #3 using CALMET/MM4 winds for MESOPUFF H and
CALPUFF, respectively, are (in fL/L): d « -10.0, -7.4, AAR = 16.1, 13.6, and RMSE = 96, 78.
The correlation coefficient are similar in magnitude (0.69 for MESOPUFF n and 0.67 for
CALPUFF).
The use of MM4-FDDA data in CALMET improves the performance of both
MESOPUFF n and CALPUFF in CAPTEX #5 as well. CALPUFF performs slightly better
than MESOPUFF n in the d, AAR, and RMSE statistics when the same wind field is used to
drive the models. The correlation coefficient is improved when the MM4-FDDA data are used,
but not as dramatically as in CAPTEX #3.
In CAPTEX #7, the CALMET wind field performs the best overall. As indicated in the
contour plots of CAPTEX #7 (Figures 4-13 through 4-18), the peak concentrations are
predicted at the same line of receptors as observed in the CALMET simulations. The plume is
displaced too far to the east in the CALMET/MM4 simulations, which affects the statistical
performance measures.
In general, the model predictions show a negative bias (indicating overprediction of the
observed concentrations). In CAPTEX #3, this is due primarily to the MESOPAC n and
CALMET winds transporting the plume directly across the monitoring network, rather than the
complex northerly transport and later southeasterly transport that is observed. The result of this
displacement of the plume is a large set of receptors with high to moderate predicted
concentrations directly NE of the source where the plume was not detected, and therefore a
large negative bias. With the improved prediction of the plume transport with CALMET/MM4
in CAPTEX #3, the average residual drops to -7.4 fL/L with CALPUFF-CALMET/MM4 vs. -
35.4 fL/L with MESOPAC II-MESOPUFF II.
In CAPTEX #5, the average residuals are much lower, even though the average
observed concentrations are higher. For example, d ranges from -3.0 to -15.6 fL/L. In
CAPTEX #7, the CALMET/MESOPUFF H simulation produces the lowest average residual (-
1.1 fL/L). The average observed concentration for this experiment was 13.8 fL/L.
The corresponding performance measures for peak concentrations (paired in space,
unpaired in time) are presented in Table 4-2. It is immediately apparent that this less strict
I:\«3M\report\«314rev\iecM.wph 4-27
-------
Table 4-2
Performance Measures for Peak CAPTEX Concentrations Paired in Space
Winds:
Trial:
N
O
SD0
P
SDp
d
AAR
RMSE
R
FB
FS
MESOPAC D
#3
70
14.1
42.5
153
50.2
-139
160
525
-.072
-1.66
-1.69
#5 #7
56 57
45.9 482
120 116
57.4 64.1
127 181
-11.6 -15.8
89.4 108
177 227
-.017 -.122
-22 -.28
-.06 -.43
MESOPUFF
El Simulations
CALMET
#3
70
14.1
42.5
130
302
-116
124
325
.035
-1.61
-LSI
#5
56
45.9
120
113
198
-66.8
118
211
292
-.84
-.49
#7
57
482
116
57.1
193
-8.9
37.7
142
.684
-.17
-.50
CALMET/MM4
#3
70
14.1
42.5
61.1
242
-47.0
62.0
217
.767
-125
-1.40
#5
56
45.9
120
91.5
181
-45.6
663
133
.722
-.66
-.40
#7
57
482
116
114
422
-65.6
124
412
264
-.81
-1.14
I:\a314\rcpoit\a314nv\iect4.wph
4-28
-------
Table 4-2 (Continued)
Performance Measures for Peak CAPTEX Concentrations Paired in Space
CALPUFF Simulations
Winds: CALMET CALMET/MM4
Trial: #3 #5 #7 #3 #5 #7
N 70 56 57 70 56 57
O 14.1 45.9 482 14.1 45.9 482
SD0 42.5 120 116 42.5 120 116
P 148 103 37.8 51.6 82.6 79.0
SDP 513 211 198 202 168 316
d -133 -56.8 10.4 -37.5 -36.8 -30.8
AAR 147 121 54.0 52.7 59.0 90.1
RMSE 533 231 185 177 131 308
R -.027 .167 .406 .746 .663 271
FB -1.65 -.77 24 -1.14 -57 -.48
FS -1.69 -55 -52 -131 -33 -.93
I:\a314\report\314«ect4 4-29
-------
measure generally improves the correlation between the simulated and observed concentration
patterns. In CAPTEX #3 and #5, the best performing model combination in terms of the d,
AAR, and RMSE statistics is the CALMET/MM4 and CALPUFF combination. Overall, the
correlation coefficients range from small negative values (with MESOPAC n) to values in the
.68 to .77 range for the best simulations with CALMET/MM4 and CALMET. The choice of
the wind field model and its inputs data had the most significant effect in the matrix of tests
performed, rather than the selection of the dispersion model. This is not unexpected, given the
large effect a spatial displacement of a plume has on such short-term, paired statistical
measures. For example, a perfectly predicted plume in terms of concentrations as a function of
distance from the source will yield very poor spatially-paired statistics if it is displaced by the
wind field so as not to overlap the observed plume.
Figures 4-19 through 4-21 show the peak modeled and observed concentrations as a
function of distance from the release. The concentration values are taken from those displayed
in Figures 4-1 through 4-18. These plots highlight two features of the "footprints":
1) The spatial resolution of the observational monitoring network appears too
coarse at times (particularly CAPTEX #5 and #7) to resolve the peak plume
concentrations at distances of up to 400-500 km, since concentrations do not
decrease monotonically or smoothly with distance.
2) The large predicted concentrations in the near-field, which are missed by the
observational network, contributes to the bias in the modeled fields of
concentration. The average bias may be reduced if a more dense sampling
network were used that could detect the near-field peak concentrations.
3) In CAPTEX #5 and #7 at distances greater than 500 km, the predicted and
observed concentrations are generally within a factor of two, even though the
plume may be displaced spatially.
In conclusion, this evaluation indicates that CALMET, with MM4-FDDA fields used as
its Step 1 wind field, improves the ability of CALPUFF and MESOPUFF U to predict the tracer
patterns observed during the CAPTEX Experiments #3 and #5. The improvement in CAPTEX
#3 is especially dramatic in terms of the contour plots and statistical measures. During
CAPTEX #7, CALMET (without MM4-FDDA data) produced the best simulations of predicted
concentrations. In general, the differences in model performance for the short-term, paired
measured considered here, are more strongly influenced by the choice of the wind field model
I:\a314\report\314wct4 4-30
-------
and input data, rather than the dispersion model. In most statistics, however, the CALMET-
MM4/CALPUFF combination performed better than the others (except in CAPTEX #7).
I:\a314\report\314ied4 4-31
-------
O
CAPTEX #3
•*— -• _
O -
LD -
O ~
^_^OI
1 ^-
— 1 ^r -
\-
-
i
i
Ol
0-
.° =
'-!-'
Do-
L- 01
-+-> 0-
CCN-
(D -
(J :
^— - o ~
°s-
o2:
— -
n —
MOD
Disp
* *— * *-* MES
» j r~ ^™**
i *— * -*-*-* y ps
. 1 VI L. O
i *«-«-»-« MES
i v i u_ *^^
/*** A i r
i «-«-«• M^ L/ALr
1 y^\ A I r
i **-*-*-* CALF
1 /^ r-\ ^k
1 v w ^ ^/ rj^ ^
I \
i i
\
; \
i \
\
^ ' \
\ \
\ \ i
V!\ \
%< ^
*\ x
Meteorological
MESOPAC
CALMET
CALMET/MM4
CALMET
CALMET/MM4
0 200 400 600 " 800
000 1200
p\ *
1 I
* \r~\ s^- /~ N / X
i 1 ^ c 1 K
Figure 4-19. Peak modeled and observed concentration as a function of distance for
CAPTEX #3.
I:\a314\repoit\314wct4
4-32
-------
o
o.
o
o.
o
C
D
L_
C§:
O
<§8 =
H
0
CAPTEX #5
MODELS:
Dispersion
~ MESOPUFF
•*~ MESOPUFF
, *^^ MESOPUFF
\ > *,*-*** CALPUFF
\\ ++-++-* CALPUFF
/< \\o^^-o OBSERVED
"\
'' \
/
Meteorological
MESOPAC
CALMET
CALMET/MM4
CALMET
CALMET/MM4
200 400 600
• Distance (''
-------
CAPTEX #7
MODELS:
Dispersion Meteorological
MESOPUFF MESOPAC
MESOPUFF CALMET
MESOPUFF CALMET/MM4
CALPUFF CALMET
CALPUFF CALMET/MM4
i i i i i i i i i i i i i i , i i
200 400 600 800 1000
• Distance (km)
Figure 4-21. Peak modeled and observed concentration as a function of distance for
CAPTEX #7.
I:\«314\rrport\3M«ect4 4-34
-------
5. REFERENCES
Allwine, KJ. and C.D. Whiteman, 1985: MELSAR: A mesoscale air quality model for complex
terrain: Volume l~Overview, technical description and user's guide. Pacific Northwest
Laboratory, Richland, WA
Anthes, R A, E. Hsie, Y Kuo, 1987: Description of the Penn State/NCAR Mesoscale Model
Version 4 (MM4). Technical Note NCAR/TN-282+STR, National Center for
Atmospheric Research (NCAR), Boulder, CO.
Atkinson, R., A.C. Lloyd and L. Winges, 1982: An updated chemical mechanism for
hydrocarbon/NOj/SO, photo oxidation suitable for inclusion in atmospheric simulation
models. Atmospheric Environ., 16, 1341.
Berkowicz, R. and L.P. Prahm, 1982: Evaluation of the profile method for estimation of surface
fluxes of momentum and heat. Atmospheric Environment, 16, 2809-2819.
Briggs, G A., 1982: Simple substitutes for the Obukhov length. Proceeding, 3rd Joint Conference
onApplic. ofAirPolL Meteor., American Meteorological Society, Boston, MA, pp. 68-71.
Briggs, GA, 1985: Analytical parameterizations of diffusion: The convective boundary layer. J.
Ctim. andAppL Meteor., 24, 1167-1186.
Carson, DJ., 1973: The development of a dry, inversion-capped, convectively unstable boundary
layer. Quart. J.Roy. Meteor. Soc., 99, 450-467.
Douglas, S. and R. Kessler, 1988: User's guide to the diagnostic wind field model (Version 1.0).
Systems Applications, Inc., San Rafael, CA, 48 pp.
Dyer, AJ. and B.B. Hicks, 1970: Flux-gradient relationships in the constant flux layer. Quart. J.
Roy. Meteor. Soc., 96, 715-721.
Ferber, GJ., J.L. Heffter, R.R. Drexler, RJ. Lagomarsino, F.L. Thomas, R.N. Dietz, and
C.M. Benkovitz, 1986: Cross-Appalachian tracer experiment (CAPTEX '83) final
report. NOAA Technical Memorandum ERL ARL-142, Air Resources Laboratory,
Silver Spring, MD.
I:\a314\a314rev\iec*5.wpb 5-1
-------
Forrest, J., R.W. Garber and L. Newman, 1981: Conversion rates in power plant plumes based
on filter pack data: The coal-fired Cumberland plume. Atmospheric Environ., 15, 2273.
Fowler, D. and M.H. Unsworth, 1979: Turbulent transfer of sulphur dioxide to a wheat crop.
Quart J.R. Meteor. Soc., 105, 767-784.
Garratt, J.R., 1977: Review of drag coefficients over oceans and continents. Mon. Wea. Rev.,
105, 915-929.
Godden, D. and F. Lurmann, 1983: Development of the PLMSTAR model and its application
to ozone episode conditions in the South Coast Air Basin. Environmental Research and
Technology, Inc., Westlake Village, CA,
Goodin, W.R., GJ. McRae and J.H. Seinfeld, 1980: An objective analysis technique for
constructing three-dimensional urban scale wind fields. /. AppL MeteoroL, 19, 98-108.
Haagenson, P.L., K Gao and Y-H Kuo, 1990: Evaluation of meteorological analyses,
simulations, and long-range transport calculations using ANATEX surface tracer data. /.
AppL MeteoroL, 29, 1268-1283.
Hanna, S.R., L.L. Schulman, RJ. Paine, I.E. Pleim and M. Baer, 1985: Development and
evaluation of the Offshore and Coastal Dispersion Model. JAPCA, 35, 1039-1047.
Hanna, S.R., J.C. Weil and RJ. Paine, 1986: Plume model development and evaluation. Report
Number D034-500. Electric Power Research Institute, Palo Alto, CA.
Hicks, B.B., 1982: In: Critical Assessment Document on Acid Deposition (Chapter VII-Dry
Deposition). ATDL Contribution File No. 81/24. Atmospheric Turbulence and
Diffusion Laboratory, NOAA, Oak Ridge, TN.
Holtslag, AA.M. and A.P. van Ulden, 1982: Simple estimates of nighttime surface fluxes from
routine weather data. KNMI Scientific Report, W.R. 82-4, 11 pp.
Holtslag, AA.M. and A.P. van Ulden, 1983: A simple scheme for daytime estimates of the
surface fluxes from routine weather data. /. Clim. and AppL Meteor., 22, 517-529.
I:\a314\a314rev\ieetS.wph 5-2
-------
Hosker, R.P., 1974: A comparison of estimation procedures for ovenvater plume dispersion.
Proceedings, Symposium on Atmospheric Diffusion and Air Pollution. American
Meteorological Society, Boston, MA.
Hosker, R.P., Jr. and S.E. LJndberg, 1982: Review: Atmospheric deposition and plant
assimilation of gases and particles. Atmospheric Environ., 16, 889-910.
Kitaigorodskii, SA, 1973: The physics of air-sea interaction. Israel Program for Scientific
Translations. Jerusalem.
Landsberg, H.E., 1981: The Urban Heat Island. Academic Press, New York, NY.
Liu, M.K. and MA. Yocke, 1980: Siting of wind turbine generators in complex terrain. /.
Energy, 4, 10:16.
Maul, P.R., 1980: Atmospheric transport of sulfur compound pollutants. Central Electricity
Generating Bureau MID/SSD/80/0026/R. Nottingham, England.
Morris, R.E., R.C. Kessler, S.G. Douglas, K.R. Styles, and G.E. Moore, 1988: Rocky Mountain
Acid Deposition Model Assessment. Acid Rain Mountain Mesoscale Model (ARM3).
U.S. Environmental Protection Agency, Research Triangle Park, NC 27711.
O'Brien, J J., 1970: A note on the vertical structure of the eddy exchange coefficient in the
boundary layer. J. Atmos. Scl, 27, 1213-1215.
O'Dell, RA., M. Taheri and R.L. Kabel, 1977: A model for uptake of pollutants by vegetation.
JAPCA, 27, 1104-1109.
Oke, T.R., 1978: Boundary Layer Climates. John Wiley & Sons, New York, NY.
Oke, T.R., 1982: The energetic basis of the urban heat island. Quart. J.R. Met Soc., 108, 1-24.
Pielke, RA., 1984: Mesoscale Meteorological Modeling, Academic Press, Inc. Orlando, FL 32887,
612 pp.
Pleim, J., A. Venkatram and R.Y. Yamartino, 1984: ADOM/TADAP model development
program. Volume 4. The dry deposition model. Ontario Ministry of the Environment,
Rexdale, Ontario, Canada.
I:\t314\a314rev\KCt5.wph 5-3
-------
Ross, D.G., I.N. Smith, P.C. Manins and D.G. Fox, 1988: Diagnostic wind field modeling for
complex terrain: Model development and testing. Journal of Applied Meteorology, 27,
785-7%.
Ross, D.G. and D.G. Fox, 1991: Evaluation of an air pollution analysis system for complex
terrain. Journal of Applied Meteorology, 30, 909-923.
Scire, J.S., E.M. Insley and RJ. Yamartino, 1990: Model evaluation and user's guide for the
CALMET meteorological model. California Air Resources Board, Sacramento, CA.
Scire, J.S., D.G. Strimaitis, and RJ. Yamartino, 1990: Model formulation and user's guide for
the CALPUFF dispersion model. California Air Resources Board, Sacramento, CA.
Scire, J.S., F.W. Lurmann, A. Bass and S.R. Hanna, 1984: User's guide to the MESOPUFF H
model and related processor programs. EPA-600/8-84-013. U.S. Environmental
Protection Agency, Research Triangle Park, NC.
Shepherd, J.G., 1974: Measurements of the direct deposition of sulphur dioxide onto grass and
water by the profile method. Atmospheric Environ., 8, 69-74.
Slinn, W.G.N., L. Hasse, B.B. Hicks, A.W. Hogan, D. Lai, P.S. Liss, K.O. Munnich, GA. Sehmel
and O. Vittori, 1978: Some aspects of the transfer of atmospheric trace constituents past
the air-sea interface. Atmospheric Environ., 12, 2055-2087.
Slinn, S A. and W.G.N. Sunn, 1980: Predictions for particle deposition on natural waters.
Atmospheric Environ., 14, 1013-1016.
Smith, I.N. and D.G. Ross, 1988: Diagnostic wind field studies. Clean Air (Aust.)., 22, 141-143.
Stauffer, D.R. and N.L. Seamon, 1990: Use of Four-Dimensional Data Assimilation in a
Limited-Area Mesoscale Model. Part I: Experiments with Synoptic-Scale Data. Man.
Wea. Rev., 18, 1250-1277.
Steyn, D.G. and T.R. Oke, 1982: The depth of the daytime mixed layer at two coastal locations:
A model and its validation. Bound. Layer Meteor., 24, 161-180.
I:\a314\a314rev\iect5.wph 5-4
-------
Tesche, T.W., J.G. Wilkinson, D.E. McNally, R. Kapahi and W.R. Oliver, 1988: Photochemical
modeling of two SCCCAMP-1984 oxidant episodes. Volume IE-Modeling procedures
and evaluation results. Prepared for the U.S. Environmental Protection Agency, Region
IX by Radian Corporation, Sacramento, CA.
U.S. Environmental Protection Agency, 1992: User's Guide for the Industrial Source Complex
(ISC2) Dispersion Models. Volume II • Description of Model Algorithms.
EPA-450/4-92-008b. U.S. Environmental Protection Agency, Research Triangle
Park, NC.
van Ulden, A.P. and AA.M. Holtslag, 1985: Estimation of atmospheric boundary layer
parameters for diffusion applications. / Clim, andApp. Meteor., 24, 1196-1207.
Venkatram, A., 1980a: Estimating the Monin-Obukhov length in the stable boundary layer for
dispersion calculations. Boundary Layer Meteorology, 19, 481-485.
Venkatram, A., 1980b: Estimation of turbulence velocity scales in the stable and the unstable
boundary layer for dispersion applications. In: Eleventh NATO-CCSMInternational
Technical Meeting on Air Pollution Modeling and its Application. 54-56.
Voldner, E.G., LA. Barrie, and A. Sirois, 1986: A literature review of dry deposition of oxides
of sulfur and nitrogen with emphasis on long-range transport modelling in North
America. Atmos. Environ., 20, 2101-2123.
Weil, J.C., 1985: Updating applied diffusion models. /. Clim. Appl Meteor., 24, 1111-1130.
Weil, J.C. and R.P. Brower, 1983: Estimating convective boundary layer parameters for
diffusion application. Draft Report Prepared by Environmental Center, Martin Marietta
Corp. for Maryland Dept. of Natural Resources.
Wesely, M.L. and B.B. Hicks, 1977: Some factors that affect the deposition rates of sulfur
dioxide and similar gases on vegetation. J. Air Poll Control Assoc., 27, 1110-1116.
Wheeler, N., 1990: Modeling of mixing depths during a southern California air quality study
ozone episode. Proceedings oftheAWMA International Specialty Conference on
Tropospheric Ozone and the Environment. March 19-22, Los Angeles, CA.
I:\«3M\«314rev\iec*5.wph 5-5
-------
Wilson, W.E., 1981: Sulfate formation in point source plumes: a review of recent field studies.
Atmospheric Environ., 15, 2573.
Zilitinkevich, S.S., 1972: On the determination of the height of the Ekman boundary layer.
Boundary Layer Meteorology, 3, 141-145.
I:\a314\a314rev\iedS.wpb 5-6
-------
Appendix
Forest Service Notes on
NUATMOS Modeling Exerdse
fc\a314\»314rev\«pp_a.wph
-------
Forest Service Notes on
NUATMOS Modeling Exercise
Data used for the modeling scenarios consisted of different combinations of actual
observations from surface (WBAN data), upper air measurements (NAMER-WINDTEMP
(North American wind and temperature)), and gridded data produced by the Penn State/NCAR
mesoscale model with four-dimensional data assimilation (MM4-FDDA).
Programs were written to put the data into the appropriate NUATMOS format (see the
NUATMOS Version 6 User Manual (CAMM, 1993) for more details) and to ensure that all data
used were within the same time frame and employed the same units (m/s for the wind). A
Lambert conformal projection was used for the terrain grid, which necessitated the conversion
of the latitude/longitude coordinates and a rotational adjustment to the wind direction
component.
The NUATMOS model is a diagnostic model, which takes available wind information
from a discrete time period and produces an interpolated, divergence-free wind output. The
model runs were executed on personal computers (386 and 486 IBM-compatible machines) with
the program compiled under the Microsoft FORTRAN Version 5.1 compiler. The desired
model output for a particular run was a file containing unformatted output for all hours. Due to
this stipulation and some limitations imposed by the DOS operating system, the NUATMOS
program was modified to "loop" through many hour runs. Each hour of meteorological input
data were kept in separate files. The NUATMOS program was modified slightly so that the log
of each run could be written to a file.
Specifically, NUATMOS checks for input data to be within the bounds of the grid and
rejects the data if the coordinates of the observations are outside the range: (UTMX,
UTMX+(NX+2)*DX) for the X direction and (UTMY, UTMY+(NY+2)*DY) for the Y
direction, where DX and DY define the grid resolution, NX and NY denote the number of grid
points in the X and Y directions, respectively, and UTMX and UTMY denote the origin as
input by the user and modified for the extended grid: UTMX=UTMXO-DX* 1.5,
UTMY=UTMYO-DY» 1.5. Figure A-l, reproduced here from the NUATMOS Version 6 User's
Manual is helpful in depicting this. The extended grid is necessary because the equations to be
solved include terrain slopes and other derivatives on the lateral boundaries.
t\a314\a314rev\app_a.wph A-l
-------
North]
j=Ny+1
j=Ny
j=3
1
1
1
1
r
i
i
i
»_ _ _ _
i
1
i
T
1
i
1
i
1
1
1
i
• ,
f
i i
i '
1
r
i
•
-*
•
•
X
1
•
•* *».
•
•
•
1
I
1
1
1
•
•» «—
•
•
•
J
1
r
J
^ ^
f
J
'
L_ _t. _J
|
1
1
r
•
•* ^
•
•
•
L 1
r 1
f
i
•
^
•
•
•
L -1
r »
- - _
— — -j
i
J
1
_
1
"1
i
o
II
II
I
X
II
i
East
Figure A-l. Schematic plan view of the computational grid used by NUATMOS. The user
must specify the average terrain elevation for each grid cell shown with a central
dot, and NUATMOS provides wind velocity estimates for the same cells. The
coordinates (UTMXO,UTMYO), which the user must enter, correspond to the
center of the most southwesterly of these cells, shown with a cross. Internally,
NUATMOS augments the user's specified grid with an additional row of cells
outside each boundary, as shown by dashed lines (from CAMM, 1993).
i:\a314\a314rev\appjLWph
A-2
-------
Data were sorted from each original set, i.e., surface data, upper air data, and MM4
data, for the coarsest grid (54-km resolution) and then combined according to scenario. Data
from the finer grid resolutions were sorted from the data selected for the coarsest grid.
Processing of the Upper Air (NAMER-WINDTEMP) Data
Data were originally extracted from the National Climatic Center Data Base and were of
the TD-9743 format. The Forest Service received this data from EPA via John Vimont,
National Park Service (NPS). Station latitude and longitude were checked to ensure that the
location was within the grid. Latitude/longitude were converted to Lambert conformal
projection coordinates, and the necessary correction was applied to the wind direction.
Greenwich mean time (GMT) time was converted to local standard time (LST) Zone 5. Data
were sorted by station and then interpolated linearly between the OZ and 12Z reporting times.
Where there was a gap in information larger than 12 hours, no data were interpolated for that
particular station. Following interpolation, data were sorted into separate files.
Processing of the Surface WBAN Data
These data were referenced to LST and were of National Weather Service (NWS)
CD 144 format. These data were extracted from the TD-3280 tapes by John Vimont and sent to
the Forest Service. As with the upper air data, latitude/longitude coordinates were converted to
Lambert conformal grid values and the wind direction correction was computed. Most of the
data were within time zone 5; time was adjusted to correspond to zone 5 for stations outside
this zone. Data were decoded; if there were missing values, the data were not interpolated.
After all stations were processed, the data were sorted into appropriate hour files.
Processing of the MM4-FDDA Data
The data supplied by Russell Bullock (EPA) through John Vimont (NPS) contained
surface level values, NWS mandatory level values, and the 15 levels of values obtained directly
from the MM4 output file. Surface level values and the 15 levels of MM4 data were used as
input to the NUATMOS model. The wind speed and wind direction, in NWS format (dddss),
were decoded and put in NUATMOS format. The I and J indexed gridded data were
referenced to latitude/longitude values, and corrections were applied to both the coordinates
and wind direction to conform to the Lambert conformal terrain projection. Data were
originally referenced to GMT time and were adjusted to conform to LST Zone 5.
fc\«314\«314rev\app_«.wph A-3
-------
Minor Adaptations to the NUATMOS Code
NUATMOS Version 6 came about as a result of contractual agreements between Rocky
Flats Emergency Preparedness Group (Department of Energy) and the Center for Applied
Mathematical Modeling (CAMM), Australia. The NUATMOS code uses multigrid techniques
in the divergence elimination calculation phase. The number of grids utilized is calculated
internally and depends on whether the next-coarser grid has odd dimensions; i.e., if M, N
represent the dimensions of the input or finest grid, then M2=(M-l)/2, N2 = (N-l)/2 must be
odd for the coarser grid to be utilized. In view of the relatively flat elevation to the east of the
Rocky Flats site, a switch to the next-coarser grid was determined by checking for odd grid
dimensions in a north-south direction. Elevation data were extrapolated to the east as
warranted. To avoid any confusion in the future use of the NUATMOS code when the terrain
to the eastern edge of the grid is not flat, the code was modified to check for odd grid
dimensions in both the north-south and west-east directions.
Terrain elevation data were provided by EARTH TECH/Sigma Research. The grid
dimensions chosen by EARTH TECH/Sigma Research for the 54-km, 18-km, and 5-km grid
resolutions were 34 x 30, 38 x 36, and 93 x 57, respectively. In order to efficiently run
NUATMOS utilizing the maximum number of multigrids, different grid dimensions of 35 x 31,
39 x 39, and 95 x 59 were chosen for the 54-km, 18-km, and 5-km grid resolutions, respectively,
for the actual model runs. In order that the output be comparable for the same area and
minimize extra work for EARTH TECH/Sigma Research in doing the model output
comparisons (even though the runs were done on a larger grid), the output was truncated to the
appropriate grid. This constituted another minor modification to the NUATMOS code.
NUATMOS Handling of Input Information
For explanatory purposes, station data in the following description will refer to both real
surface and upper air data and MM4-FDDA data as input to the NUATMOS model. After all
station information has been read into the program, the individual observations are mapped
onto the grid. Any station with only one level of data is processed by translating its observation
to the nearest sigma level center. Data from stations with multiple levels of information are
interpolated linearly onto sigma level centers between each pair of data levels. Horizontal
interpolation of the data then proceeds for each level from the surface up. The method of
horizontal interpolation is chosen by the user and proceeds in one of three ways: (1) mean
velocity everywhere, (2) conventional inverse-distance squared weighting of observations, or (3)
incorporating a radius of influence scaled by the velocity variance for a particular level over the
inverse-distance squared weighting. If a level is encountered in which there are no
t\a314\a314rev\app_a.wph A-4
-------
observations, information is copied from the layer beneath. An exception to this is the case in
which no observations are found for the surface level, in which case, the program is terminated.
If an observation falls directly on a grid point, that observation is preserved and the influence of
surrounding observations are ignored.
The conceptual idea behind the use of a weighting scheme for incorporation of real
versus MM4-FDDA generated data based on terrain and grid resolution characteristics is
explained (elsewhere). EARTH TECH/Sigma Research supplied gridded data files of the
weights to be applied (see Section 2 of this report). Since NUATMOS employs a variable
vertical grid based on levels of constant sigma, the weights were linearly interpolated to
correspond to correct elevation levels. The top and bottom levels were left unchanged. The
NUATMOS code and input data files were modified to incorporate the weighting scheme. On
the heading line of the input file, an extra parameter was added to indicate whether the
observation was real (1) or MM4 (2). The three-dimensional weighting information was read
into the program. When the observations were horizontally interpolated onto the grid, the
weighting factor was applied.
Initially, a variable radius of influence interpolation method was run on the three grids
using length scales of 5 and 10 km. It soon became apparent that the choice of a length scale
must take into account the grid resolution. Ross (1992) does not recommend the use of the
variable radius of influence scheme unless sufficient data are available to undertake a detailed
evaluation of the terrain and desired grid resolution. The recommended interpolation method is
the inverse-distance squared weighting scheme. Sample runs were performed to determine the
influence of different length scales. Taking into account the spacing of the input observations
and the grid resolutions, the length scale that seemed appropriate for interpolating the
observations produced output that correspond closely to the output obtained by using the
inverse-distance squared interpolation method. It was then decided to use the inverse-distance
squared method for all runs.
Another area of choice for input to the NUATMOS model concerns the parameter alpha
as it relates to stability. Ross (1992) suggests alpha equal 1 in "data-rich" regions, where the
influence of stability can reasonably be expected to be represented in the data. In "data-sparse"
regions, Ross suggests that an appropriate value for alpha be related to the characteristic
Froude number in the manner outlined in a previous paper (Ross et al., 1988).
The choice of an appropriate value for alpha is not straightforward. For most of the
runs being performed here, the input of the MM4-FDDA constitutes "data-rich" regions. It
could be argued that surface and upper air observations for the coarser grid resolutions (18 km
fc\«314\a314rev\app_«.wph A-5
-------
and 54 km) constitute a "data-rich" region when comparing the grid density with the station
density. A counter argument would contain reservations that the resolution of the upper air
data is not sufficient to resolve the influence of stability.
The choice of the alpha parameter is related to the bulk Froude number. Only one
value of alpha is input for the entire grid. For the areas under consideration, there are many
terrain features that need to be evaluated with reference to the bulk Froude number. It is
apparent that stability over such area! extents is not uniform everywhere. As a result of many of
these concerns, the basic model runs were executed with the value of alpha set to 1.
Hourly input data were scrutinized to determine periods of stability for the summer and
winter periods. This was relatively straightforward to do with the MM4-FDDA hourly data, but
did not yield adequate information for the 12-hour upper air data. In these cases, stable periods
were chosen by considering time of sunrise and sunset, as well as the season. Stable periods
were modeled with alpha set to 0.1; all other periods were modeled with alpha set to 1.
References to Appendix
Centre for Applied Mathematical Modeling (CAMM), 1993: NUATMOS (Version 6) User
Manual. Centre for Applied Mathematical Modeling, Monash University, Caulfield East,
Australia.
Ross, D.G., I.N. Smith, P.C. Manins and D.G. Fox, 1988: Diagnostic wind field modeling for
complex terrain: Model development and testing. Journal of Applied Meteorology, 27,
785-796.
Ross, D.G., 1992: Personal Communication to Bernie Connell, USDA Forest Service.
fc\a314\a314rev\«pp_a.wph A-6
-------
I REPORT NO. 2.
EPA-454/R-95-005
4 TITLE AND SUBTITLE
Testing of Meteorological and Dispersion Models
for Use in Regional Air Quality Modeling
7. AUTHOR(S)
9 PERFORMING ORGANIZATION NAME AND ADDRESS
The CADMUS Group
Waltham, MA 02154
ft
12 SPONSORING AGENCY NAME AND ADDRESS
U.S. Environmental Protection Agency
Office of Air Quality Planning and Standards
Emissions, Monitoring and Analysis Division
Research Triangle Park, NC 27711
3. RECIPIENT'S ACCESSION NO
5. REPORT DATE
March 1995
«. PERFORMING ORGANIZATION CODE
8. PERFORMING ORGANIZATION REPORT NO.
10. PROGRAM ELEMENT NO.
11. CONTRACT/GRANT NO.
Contract 68 -DO -0095
Work Assignment 218
13. TYPE OF REPORT AND PERIOD COVERED
Final Report
14. SPONSORING AGENCY CODE
TECHNICAL REPORT DATA
(Please read Instructions on reverse before completing)
15. SUPPLEMENTARY NOTES
EPA Work Assignment Manager:
John S. Irwin
16. ABSTRACT
This report summarizes a series of sensitivity analyses of several meteorological
models and puff dispersion models suitable for application for air pollution
assessment involving long-range transport. The analysis was conducted in three
phases: 1) sensitivity testing of diagnostic meteorological models; 2)
sensitivity tests of dispersion model components; and 3) evaluation of the
recommended meteorological and dispersion models with tracer data collected during
the Cross Appalachian Tracer Experiment (CAPTEX) field study. The meteorological
sensitivity tests included simulations with CALMET and NUATMOS diagnostic wind field
models. These models were applied in two ways: 1) using observed surface and upper
air wind data as input, and 2) using gridded wind fields produced by a mesoscale
meteorological model (MM4) that employed four-dimensional data assimilation on an
80-km grid in addition to the actual observed wind data. One of the primary
objectives of the CAPTEX comparisons was to assess whether the introduction of the
hourly, gridded 80-km MM4 data with its improved spatial and temporal resolution
over typical observational networks would improve the quality of the
characterization of the transport and dispersion. Based on the results obtained
there was a noticeable improvement in the trajectory of the transport when MM4 data
were employed in comparison to results obtained using diagnostic winds developed
using available wind observations.
KEY WORDS AND DOCUMENT ANALYSIS
2 DESCRIPTORS
Air Pollution
Long Range Transport
Air Quality Dispersion Modeling
l"8. DISTRIBUTION STATEMENT
Release Unlimited
b IDENTIFIERS«>PEN ENDED TERMS
Dispersion Modeling
Meteorology
Air Pollution Control
19. SECURITY CLASS (Kcport)
Unclassified
20. SECURTTY CLASS (fatc)
Unclassified
c. COSATl Field/Group
21. NO. OF PAGES
223
22. PRICE
EFA Form 2220-1 (Rev. 4-77) PREVIOUS EDITION IS OBSOLETE >
------- |