EPA910-R-15-001a Alaska
United States Region 10 Idaho
Environmental Protection 1200 Sixth Avenue Oregon
Agency Seattle WA 98101 Washington
Office of Environmental Assessment October 2015
Combined WRF/MMIF/
AERCOARE/AERMOD
Overwater Modeling
Approach for Offshore
Emission Sources
Volume 1 - Project Report
-------
-------
Combined WRF/MMIF/
AERCOARE/AERMOD Overwater
Modeling Approach for Offshore
Emission Sources
Volume 1 - Project Report
EPA Contract No. EP-W-09-028
Work Assignment No. M12PG00033R
Prepared for:
U.S. Environmental Protection Agency
Region 10
1200 Sixth Avenue
Seattle, WA 98101
and
U.S. Department of the Interior
Bureau of Ocean Energy Management
45600 Woodland Road
Sterling, VA 20166
Prepared by:
Ramboll Environ US Corporation
773 San Marin Drive, Suite 2115
Novato, CA, 94998
and
Amec Foster Wheeler
Environmental & Infrastructure, Inc.
4021 Stirrup Creek Dr., Suite 100
Durham, NC 27703
October 2015
-------
-------
The Region 10 Project Officer for the Interagency Agreement No. M12PGT00033R and EPA
contract number EP-W-09-028 was Herman Wong with technical support provided by Robert
Kotchenruther, PhD and Robert Elleman, PhD. From BOEM, the Project Officer was Eric J.
Wolvovsky and the Technical Coordinator was Ronald Lai, PhD. The Project Lead for the prime
contractor Amec Foster Wheeler was James Paumier while Project Lead for subcontractor
RAM BOLL ENVIRON was Ken Richmond. Peer review of draft Volume 2 and/or draft Volume 3
was provided by Steven Hanna, PhD of Hanna Consultants, Robert Paine, CCM of AECOM and
Christopher Lindsey, Shell Exploration and Production. Their reviews and comments are greatly
appreciated by R10 and BOEM.
The collaboration study was funded in part by the U.S. Department of the Interior, Bureau of
Ocean Energy Management, Environmental Studies Program, Washington DC, and the U.S.
Environmental Protection Agency, Region 10, Seattle, WA.
-------
DISCLAIMER
The opinions, findings, conclusions, or recommendations expressed in this report are those of
the authors and do not necessarily reflect the view of the U.S. Environmental Protection Agency
or the U.S. Department of the Interior, Bureau of Ocean Energy Management, nor does the
mention of trade names or commercial products constitute endorsement or recommendation for
use by the Federal Government.
-------
PREFACE
The recommended American Meteorological Society/Environmental Protection Agency
Regulatory Model (AERMOD) dispersion program continues to be studied for assessing air
quality concentration impacts from emission sources located at overwater locations under an
Interagency Agreement (IA) Number M12PGT00033R dated 9 August 2012 between the U.S.
Environmental Protection Agency (EPA), Region 10 and the U.S. Department of the Interior
(DOI), Bureau of Safety and Environmental Enforcement (BSEE) on behalf of the Bureau of
Ocean Energy Management (BOEM). Specifically, the work scope under the IA calls for Region
10 and BOEM to (1) assess the use of AERMOD as a replacement for the Offshore and Coastal
Dispersion (OCD) model in a near-source (< 1,000 meters source-receptor distance) ambient air
quality impact analysis for sea surface based emission sources and (2) evaluate the use of
Weather Research and Forecasting (WRF) model predicted meteorology with AERMOD in lieu
of overwater meteorological measurements from platforms and buoys.
Results of the Region 10/BOEM collaboration study are described in a three volume report.
Volume 1 describes all six tasks completed under the IA. However, only a summary of the work
completed under Task 2 and Task 3 appears in Volume 1. Volume 2 and Volume 3 provides a
detailed description of the work in Task 2 and Task 3, respectively. The six tasks are:
Task 1. Evaluation of two Outer Continental Shelf Weather Research and Forecasting Model
Simulations
Task 2. Evaluation of Weather Research and Forecasting Model Simulations for Five Tracer
Gas Studies with AERMOD
Task 3. Analysis of AERMOD Performance Using Weather Research and Forecasting Model
Predicted Meteorology and Measured Meteorology in the Arctic
Task 4. Comparison of Predicted and Measured Mixing Heights
Task 5. Development of AERSCREEN for Arctic Outer Continental Shelf Application
Task 6. Collaboration Study Seminar
Prior to the collaboration study, Region 10 on 1 April 2011 approved the use of the Coupled
Ocean-Atmosphere Response Experiment (COARE) air-sea flux algorithm with AERMOD to
preprocess overwater measured meteorological data from platforms and buoys. Initially, the
preprocessing of the overwater measurements was done manually with COARE. Subsequently,
Region 10 funded a study that was completed in September 2012 that coded the COARE air-
sea flux procedure into a meteorological data preprocessor program called AERMOD-COARE
(AERCOARE). The AERCOARE program was uploaded to the EPA Support Center for
Regulatory Atmospheric Modeling (SCRAM) website on 23 May 2013 as a beta option for case-
by-case approval by EPA regional offices.
-------
[Blank]
-------
TABLE OF CONTENTS
LIST OF FIGURES IX
LIST OF TABLES XI
LIST OF ABBREVIATIONS AND ACRONYMS XIII
1 INTRODUCTION 1
2 TASK 1 - EVALUATION OF TWO OUTER CONTINENTAL SHELF WEATHER
RESEARCH AND FORECASTING MODEL SIMULATIONS 5
2.1 Introduction 5
2.2 Statement of Work 5
2.3 Analysis 7
2.4 Results 8
2.4.1 Updates to MM IF 8
2.4.2 Suitability of the two WRF datasets 9
3 TASK 2 - EVALUATION OF WEATHER RESEARCH AND FORECASTING
MODEL SIMULATIONS FOR FIVE TRACER GAS STUDIES WITH AERMOD 11
4 TASK 3 - ANALYSIS OF AERMOD PERFORMANCE USING WEATHER
RESARCH AND FORECASTING MODEL PREDICTED METEOROLOGY AND
MEASURED METEOROLOGY IN THE ARCTIC 17
5 TASK 4 - COMPARISON OF PREDICTED AND MEASURED MIXING
HEIGHTS 25
5.1 Overview and Objective 25
5.1.1 Upper-air and surface observations 25
5.1.2 WRF simulations 27
5.2 Analysis 27
5.3 Results 29
5.3.1 FNMOC WRF evaluated using Endeavor Island profiler 29
5.3.2 RTG WRF evaluated using JAMSTEC soundings 30
5.3.3 FNMOC WRF evaluated using Point Barrow soundings 31
5.4 Conclusions 31
-------
6 TASK 5 - DEVELOPMENT OF AERSCREEN FOR ARCTIC OUTER
CONTINENTAL SHELF APPLICATIONS 51
6.1 Introduction 51
6.2 Approach and Methodology 52
6.3 Overview 52
6.4 Methods 53
6.4.1 AERCOARE input requirements 53
6.4.2 AERCOARE overwater datasets 55
6.4.3 Emission sources 61
6.4.4 Receptor grid 62
6.4.6 AERSCREEN simulations 66
6.5 Results 66
6.6 Recommendations for Future Work 69
7 TASK 6 - COLLABORATION STUDY SEMINAR 85
7.1 I ntroduction 85
7.2 Session Overviews 85
7.2.1 Day 1, Morning 85
7.2.2 Day 1, Afternoon 86
7.2.3 Day 2, Morning 87
7.2.4 Day 2, Afternoon 87
8 REFERENCES 91
APPENDIX A: PROTOCOLS
APPENDIX B: PEER REVIEW COMMENTS FOR VOLUME 2 AND VOLUME 3
-------
LIST OF FIGURES
Figure 1. Alpine/CP 36-12-4 km domains, covering the Chukchi Sea 6
Figure 2. UAF/BOEM 10-km domain, covering the Chukchi and Beaufort Seas 6
Figure 3. Overwater meteorological measurement sites and corresponding WRF inner domain
extraction points 18
Figure 4. Source locations and structures and innermost receptor ring 20
Figure 5. Location of the profiler station on Endeavor Island 26
Figure 6. The K&Z profiler (left) and a view of the station looking North (right) 26
Figure 7. Profiler retrievals and WRF soundings for 2010-07-27 07:00 LSI 33
Figures. Profiler retrievals and WRF soundings for 2010-08-20 16:00 LSI 34
Figure9. Profiler retrievals and WRF soundings for 2010-08-16 17:00 LSI 35
Figure 10. WRF's PBLH vs hand-analyzed inversion base (ZiBase) 36
Figure 11. MMIF's Critical Bulk Richardson mixing height vs. hand-analyzed inversion base
(ZiBase) 37
Figure 12. WRF's PBLH vs. the Critical Bulk Richardson mixing height from the profiler 38
Figure 13. Critical Bulk Richardson mixing heights from MMIF's vs. from the profiler 39
Figure 14. MMIF's Critical Bulk Richardson mixing height vs. AERMOD's mechanical mixing
height 40
Figure 15. MMIF's Critical Bulk Richardson mixing height vs. AERMOD's convective mixing
height 41
Figure 16. Locations of the JAMSTEC 2009 soundings 42
Figure 17. RTG WRF and JAMSTEC profiles for 2009-09-11 06:00 UTC and 2009-09-28 12:00
UTC 43
Figure 18. RTG WRF and JAMSTEC profiles for 2009-09-30 18:00 UTC and 2009-10-03 18:00
UTC 44
Figure 19. RTG WRF and JAMSTEC profiles for 2009-10-07 18:00 UTC and 2009-10-09 09:00
UTC 45
Figure 20. RTG WRF and JAMSTEC profiles for 2009-10-10 06:00 UTC and 2009-10-10 18:00
UTC 46
Figure 21. JAMSTEC observed vs. WRF-MMIF mixing heights, both derived using the critical
bulk Richardson number technique (4km in red, 12km in blue) 47
-------
Figure 22. FNMOC WRF and Pt. Barrow profiles for 2010-08-14 12:00 UTC and 2010-08-28
12:00 UTC 48
Figure 23. FNMOC WRF and Pt. Barrow Profiles for 2010-09-20 12:00 UTC and 2011-06-27
12:00 UTC 49
Figure 24. Hypothetical Drill Ship Layout and Emission Source Locations 62
Figure 25. Screening Modeling Receptor Grid 63
Figure 26. Roster of Attendees 88
Figure 27. Agenda 89
-------
LIST OF TABLES
Table 1. Model configuration for UAF/BOEM and Alpine/CP WRF simulations 7
Table 2. WRF AERMOD Meteorology Extraction Methods 12
Table 3. Hypothetical Drill Ship Emission Sources 19
Table 4. Data Dependent AERCOARE Options 53
Tables. Task 3 AERCOARE Data Requirements 54
Table 6. Additional Required AERCOARE Input Parameters (Control File) 55
Table?. Review of Task 3 AERCOARE Overwater Data 57
Table 8. Ambient Air and Sea Surface Temperatures at Selected Arctic Buoys 59
Table 9. Initial Meteorological Screening Values (COARESCREEN1) 60
Table 10. Second Set of Meteorological Screening Values (COARESCREEN2) 61
Table 11. Third Set of Meteorological Screening Values (COARESCREEN3) 61
Table 12. AERMOD Simulations Identified By Meteorological Data Set 65
Table 13. AERSCREEN Input Values 66
Table 14. Percent of Model Runs Where Screening Results Are Conservative Compared to
Refined Modeling Results 67
Table 15. Ratio of COARESCREEN-to-AERSCREEN 1 -hour H1H Concentrations 68
Table 16. Comparison of Screening and Refined Modeling for Release Point S1P1 70
Table 17. Comparison of Screening and Refined Modeling for Release Point S1P2 71
Table 18. Comparison of Screening and Refined Modeling for Release Point S2P1 72
Table 19. Comparison of Screening and Refined Modeling for Release Point S2P2 73
Table 20. Comparison of Screening and Refined Modeling for Release Point S2P3 74
Table 21. Comparison of Screening and Refined Modeling for Release Point S3P1 75
Table 22. Comparison of Screening and Refined Modeling for Release Point S3P2 76
Table 23. Comparison of Screening and Refined Modeling for Release Point S3P3 77
Table 24. Comparison of Screening and Refined Modeling for Release Point S4P1 78
Table 25. Comparison of Screening and Refined Modeling for Release Point S4P2 79
Table 26. Comparison of Screening and Refined Modeling for Release Point S4P3 80
Table 27. Comparison of Screening and Refined Modeling for Release Point S5P1 81
Table 28. Comparison of Screening and Refined Modeling for Release Point S5P2 82
-------
Table 29. Comparison of Screening and Refined Modeling for Release Point S5P3 83
Table 30. Comparison of Screening and Refined Modeling for All Release Points Combined...84
-------
LIST OF ABBREVIATIONS AND ACRONYMS
AERC WRF meteorology extraction cases processed by AERCOARE
AERMIC American Meteorological Society/Environmental Protection Agency
Regulatory Model Improvement Committee
AERMOD American Meteorological Society/Environmental Protection Agency
Regulatory Model
AERCOARE AERMOD-COARE
ASTD Air-Sea Temperature Difference
AIDJEX Arctic Ice Dynamics Joint Experiment
6 Bowen ratio
BOEM Bureau of Ocean Energy Management
BSSE Bureau of Safety and Environmental Enforcement
c Model constant
c0 Observed concentration value
cp Predicted concentration value
c Average concentration value
cn nth highest concentration
°C Degrees centigrade
COARE Coupled Ocean-Atmospheric Response Experiment
DOI U.S. Department of the Interior
ECMWF European Center for Medium-Range Weather Forecasting
EPA U.S. Environmental Protection Agency
ERA ECMWF Reanalysis
ERA-40 ERA 45-year global atmospheric reanalysis
ERA-I ERA Interim
eta Vertical pressure coordinate in WRF
f Coriolis parameter
FF2 Fraction-factor-of-two
FNMOC Fleet Numerical Meteorology and Oceanography Center
g Grams
H Sensible heat flux
ISC3 Industrial Source Complex 3
K kelvin
kg Kilograms
km Kilometers
-------
L Monin-Obukhov length
LCC Lambert Conformal Conic
m Meters
METSTAT Meteorological Statistics
MG Geometric mean bias
MIXH PEL height or "mixing height"
MMIF Mesoscale Model Interface
MYJ Mellor-Yamada-Janjic
NAAQS National Ambient Air Quality Standards
NARR North American Regional Reanalysis
NBDC National Buoy Data Center
NCAR National Center for Atmospheric Research
NCEP National Center for Environmental Prediction
NOAA National Oceanic and Atmospheric Administration
NSR New Source Review
O Observed value
OBS, obs Label for observation-based AERMOD simulations
OCD Offshore and Coastal Dispersion
OCS Outer Continental Shelf
OLM Ozone Limited Method
P Sea Level Atmospheric Pressure (also used to indicate "predicted" value
in statistical calculations).
p Predicted value
PEL Planetary boundary layer
PFL Profile file input to AERMOD
PRIME Plume Rise Model Enhancements
PSD Prevention of Significant Deterioration
PVMRM Plume Volume Molar Ratio Method
Q-Q Quantile-Quantile
r Albedo
RCALF Label for WRF-MMIF AERMOD simulations where PEL height was not
recalculated by MMIF
RCALT Label for WRF-MMIF AERMOD simulations where PEL height was
recalculated MMIF
rg Geometric Correlation Coefficient
RH Relative humidity
RHC Robust High Concentration
-------
RMSE Root Mean Square Error
RPO Regional Planning Organization
RTG Real Time Global sea-surface temperature analysis (from NCEP)
s Seconds
SFC AERMOD surface meteorology input file
SST Sea Surface Temperature
T Temperature
TKE Turbulent Kinetic Energy, Thermal Kinetic Energy
TMS Total Model Score statistical measure
U Zonal wind component
UW-PBL University of Washington Shallow Convection PEL
i/* Friction Velocity
V Meridional wind component
VG Geometric Variance
VPTLR Virtual Potential Temperature Lapse Rate
w. Convective scaling velocity
W Watts
WD Wind Direction
WRF Weather Research and Forecasting
WS Wind Speed
YSU Yonsei University
z Height above the surface
z0 Roughness length
z,c Convective PEL height
z/m Mechanical PEL height
0 Degrees angular
/jg Geometric mean
OQ Standard deviation of wind direction
ow Standard deviation of vertical wind speed
(// Stability correction parameter
-------
[Blank]
-------
1 INTRODUCTION
Air quality modeling and impact assessment must be conducted for the New Source Review
(NSR) of significant sources of air pollutant emissions as promulgated by the U.S.
Environmental Protection Agency (EPA) and the U.S. Department of the Interior (DOI), Bureau
of Ocean Energy Management (BOEM). Given the recent and likely continued expansion of oil
and mineral exploration and extraction activities along the Outer Continental Shelf (OCS) off the
coast of Alaska along the Outer Continental Shelf (OCS) and other marine locations (e.g., mid-
latitudes and tropics), there will continue to be more demand for air quality permits and
exploratory/development plans related to such activities. The EPA and BOEM must therefore
provide modeling tools that can adequately assess air quality impacts over the OCS and other
overwater regions.
The American Meteorological Society/Environmental Protection Agency Regulatory Model
(AERMOD) modeling system ((USEPA, 2004c) is the preferred near-field (< 50 kilometers [km])
model used for the air quality assessment requirements of air emissions permitting1. However,
AERMOD's meteorological preprocessor, AERMET (USEPA, 2004a) was not designed to
process meteorological conditions over ocean waters and more extreme climates. Over land
energy fluxes are strongly driven by the diurnal cycle of heating and cooling. Over water, fluxes
are more dependent on air-sea temperature differences that are only slightly affected by diurnal
heating and cooling. In addition, the meteorological observations necessary to drive the
dispersion models are often not available, especially in the Arctic Ocean. For applications in the
Arctic, the remote location and seasonal sea-ice pose significant logistical problems for the
deployment of buoys and other offshore measurement platforms. AERMAP is not applicable in
overwater locations.
The dispersion model currently preferred by the EPA for offshore assessment of emission
sources is the Offshore and Coastal Dispersion (OCD) model (DiCristofaro & Hanna, 1989), as
promulgated under 40 CFR Part 51, Appendix W. However, OCD lacks the features required for
robust modern environmental assessment. OCD does not contain the PRIME downwash
algorithm (Schulman, et al., 2002), Plume Volume Molar Ratio Method (PVMRM) (Hanrahan,
1999), and Ozone Limiting Method (OLM) (Cole & Summerhays, 1979) and the capability to
calculate receptor averaged percentiles associated with sulfur dioxide (SO2), nitrogen dioxide
(NO2), and particulate matter less than or equal to 2.5 microns (PM2.5) concentrations for a
compliance demonstration.
State-of-the-art overwater parameterization schemes are used in the Coupled Ocean-
Atmosphere Response Experiment (COARE) air-sea flux algorithms. These algorithms have
1 As promulgated under 40 CFR Part 51, Appendix W. The AERMOD modeling system is available to the
public at the EPA modeling website: http://www.epa.gov/scram001/dispersion prefrec.htm
-------
been used to develop the AERMOD-COARE (AERCOARE)2 model (USEPA, 2012), a
counterpart to AERMET, to preprocess overwater observational data. AERCOARE takes air-sea
temperature difference and other features of marine influence into account to compute the
meteorological fields required for AERMOD modeling over the open water and coastal
environments. AERCOARE-AERMOD (using the current beta version of AERCOARE) has been
approved by EPA Region 10 and concurred by the EPA Model Clearinghouse as an acceptable
alternative approach for modeling emissions sources located in the Arctic, mid-latitude, and
tropic overwater environment. Use of the model still requires a procedural protocol in
accordance with Appendix W and review and acceptance by the appropriate EPA regional office
on a case-by-case basis (Tikvart, 1988) (Bridgers, 2011) (Wong, 2011).
Currently accepted overwater dispersion modeling methods require observational datasets.
These datasets are generally provided by meteorological buoys or instruments on platforms.
However, the observational coverage of the earth's oceans is sparse. It would be advantageous
if output from mesoscale meteorological models, such as the Weather Research and
Forecasting (WRF) model (NCAR, 2014) (Skamarock, et al., 2008), could be used to provide
hourly prognostic meteorological data for AERMOD in areas where observational data are
lacking.
To evaluate the use of prognostic meteorological data with AERMOD, the collaboration study
consists of six tasks in which Task 2 and Task 3 contains the evaluations and analyses. Prior to
carrying out the work, protocols for Tasks 1, 2, 3, and 5 were submitted to Region 10 and
BOEM for acceptance. The protocols appear in Appendix A. The Task 4 work followed the work
scope in the proposal. An agenda was used in lieu of a protocol for Task 6. The six tasks are:
Task 1. Evaluation of two Outer Continental Shelf Weather Research and Forecasting Model
Simulations
Task 2. Evaluation of Weather Research and Forecasting Model Simulations for Five Tracer
Gas Studies with AERMOD
Task 3. Analysis of AERMOD Performance Using Weather Research and Forecasting Model
Predicted Meteorology and Measured Meteorology in the Arctic
Task 4. Comparison of Predicted and Measured Mixing Heights
Task 5. Development of AERSCREEN for Arctic Outer Continental Shelf Application
Task 6. Collaboration Study Seminar
2 AERCOARE is publically available at the U.S. EPA at the website:
http://www.epa.gov/ttn/scram/dispersion related.htm
-------
In this Volume 1, the work performed for all six tasks are presented. However, only summaries
of the work completed under Task 2 and Task 3 are provided. Volume 2 provides detail
comparisons of WRF-driven AERMOD predictions against the concentrations measured during
five offshore tracer dispersion field experiments. Volume 3 summarizes the evaluation of
alternative methods for supplying meteorological variables to AERMOD for regulatory air quality
modeling of sources located over the ocean. Volume 2 and Volume 3 were submitted for peer
review. The response to the peer review comments for these two volumes is provided in
Appendix B.
-------
[Blank]
-------
2 TASK 1 - EVALUATION OF TWO OUTER CONTINENTAL SHELF WEATHER
RESEARCH AND FORECASTING MODEL SIMULATIONS
2.1 Introduction
Task 1 of the collaboration study examines two existing WRF datasets for the Arctic Ocean that
might be used to provide the necessary meteorological variables for dispersion model
simulations of OCS sources within their domains. The task objective is to examine the
differences between the two datasets, examine model performance with overwater
measurements, apply the Mesoscale Model Interface (MMIF) program and AERCOARE to the
datasets using several different options, and compare AERMOD model predictions from the
resulting datasets using simulations of typical OCS sources. The protocol for Task 1 is in
Appendix A.
2.2 Statement of Work
At the time, there were two existing WRF simulations of the North Slope of Alaska. The first was
generated by Alpine Geophysics, LLC (Alpine) under contract to ConocoPhillips (CP) which was
subsequently submitted to EPA Region 10 in support of a Prevention of Significant Deterioration
(PSD) application (McNally and Wilkinson, 2011). This PSD application covered lease blocks in
the Chukchi Sea. The second WRF simulation was developed by the University of Alaska,
Fairbanks (UAF) under contract to BOEM. The "Chukchi/Beaufort Seas Mesoscale Meteorology
Modeling Study" (MMM) produced a 31-year WRF simulation designed to support oil spill risk
assessments (Zhang, 2013). UAF completed an initial followed by a final 5 year simulation that
covered the period from 2005 to 2009.
The three subtasks described in the protocol, found in Appendix A, are:
• Generate AERMOD results from the two difference WRF datasets
• Compare AERMOD results from the two different WRF datasets
• Compare the WRF runs
The modeling domains for the Alpine/CP and the UAF/BOEM WRF simulations are shown in
Figure 1 and Figure 2, respectively. Table 1 summarizes the key WRF model options and input
data sources for the two simulations.
During the first subtask, MMIF would preprocess the WRF simulations to generate four
meteorological data sets. For the two direct meteorological data inputs into AERMOD, MMIF is
run separately to output a set of surface and profile files with and without rediagnosed planetary
boundary layer (PBL) heights. Similarly for the two indirect meteorological data input into
AERMOD, MMIF is run separately to produce a data file with and without rediagnosed PBL
heights that is read by AERCOARE (USEPA, 2012) to output two sets of surface and profile
files. (See Section 3 for more discussion.) Using hypothetical yet typical over water emission
sources at five (5) buoy locations, AERMOD would be run using the above four meteorological
files and the buoy observations processed with AERCOARE. The AERMOD results would be
examined, and the source of any discrepancies sought and explained.
-------
150°E 150°W
90°W
- 100°W
160°E
- 110°W
170°E
120°W
180°
165°W
150°W
135°W
1 25 75 200 500 1000 1500 2000 3000
Figure 1. Alpine/CP 36-12-4 km domains, covering the Chukchi Sea.
180'
180°
170°W
165°W 150°W 135°W
120°W
160°W
150°W
140°W
- 130°W
25 75 200 500 1000 1500 2000 3000
Figure 2. UAF/BOEM 10-km domain, covering the Chukchi and Beaufort Seas.
-------
Table 1. Model configuration for UAF/BOEM and Alpine/CP WRF simulations.
Parameter
Model Grid
Forcing Data
PEL
Micro physics
LW/SW Radiation
Surface Layer
Physics
Land-Surface
Model
Cumulus Physics
UAF/BOEM WRF
10 km with 49 vertical levels
ERA interim reanalysis (ERA-I)
MYJ
Morrison
RRTM/RRTMG
ETA Similarity
NOAH w/improved sea ice alb.
Kain-Fritch
Alpine/CP WRF
36-12-4 km with 37 vertical levels
GFS 1/4 degree dataset
YSU
Morrison
RRTM/RRTMG
MM5 similarity
NOAH
Kain-Fritch (36-1 2 only)
Data Assimilation
(obs. nudging)
In situ surface obs., radiosondes,
QuikSCAT sfc winds, MODIS profiles,
COSMIC profiles.
Nudging to MADIS data on 4-km domain
with a radius of influence of 50 km.
Analysis Nudging
Three-wavenumber spectral nudging
of all variables and all levels.
36 & 12 km for winds and temperature at
all model levels.
Lower Boundary
AMSR-E sea ice thickness and cone.
And CMC snow depth.
GFS initialized using static SST for each
5-day block. SST from NCEP RTG 1/12
degree analysis.
Vertical Velocity
Damping
Off
On
Advection
Positive-Definite
Monotonic
Simulation Period 2005-2009
2007-2009
(June 15-December 3rd only)
2.3 Analysis
After the first MMIF runs were completed using the UAF/BOEM initial 5-year dataset, peculiar
values were noticed every three hours for certain variables in the AERMET like output. For
example, the PEL height was a seemingly reasonable for two consecutive hours of simulated
data before it rose to an unreasonable value near 20 kilometers (km) for the third hour, then
-------
back to reasonable values for two hours, and so on. Precipitation fields in the WRF output had
similarly unexpected values every three hours.
Upon investigating these values, it was found that UAF was not performing traditional hindcast
runs with WRF, but was using the variational data assimilation version of WRF (WRF Data
Assimilation System [WRFDA]) to perform a reanalysis every three hours. WRFDA was being
run to generate analyses on a three hour interval, which in turn was being used to initialize a
two hour WRF run to fill in the hours till the next WRFDA reanalysis. The PEL height problem
stems from differences between WRFDA and WRF. WRFDA outputs prognostic variables such
as wind speed, pressure, temperature, specific humidity, etc. in the same way that WRF does,
but does not output diagnostic variables such as PEL height, or temperature at 2-meters (m), or
friction velocity (u*), in the same way as WRF.
The final 31-year version of the UAF/BOEM WRF simulation did not use this hybrid
WRFDA+WRF approach. Instead, UAF performed a WRFDA reanalysis for each hour, then ran
the output through WRF for a few time-steps-just enough for WRF to diagnose parameters
such as the PBL height, etc. Presumably, each WRFDA run covered some hours before the
analysis time, to allow the model to "spin-up" finer-scale fields such as potential vorticity. Typical
hindcast WRF runs discard the first 12-24 hours of a simulation to account for "spin-up".
The Alpine/CP WRF dataset includes output from each (nested) domain and was briefly
investigated for the differences in AERMOD output using the 12 km grid spacing data and the
4 km grid spacing data as input. The differences were unremarkable, and highlighted certain
deficiencies in MMIF noted in the next section.
2.4 Results
2.4.1 Updates to MMIF
During the course of Task 1, MMIF was modified in two ways. First, previous versions of MMIF
in AERCOARE mode would print an error statement when the extraction point was over a solid
surface as detected by the land use category for the point. This caused MMIF to stop when the
extraction point was covered with ice. During the early summer ice melt period, the sea ice
coverage fraction can vary at a particular point, causing havoc to a MMIF run. MMIF was
changed to print a warning instead of exiting with an error. Because the warning is printed for
each hour processed, the time stamp when the dominant land use category switches from ice to
ocean can easily be found.
Second, older versions of MMIF passed through all wind speeds when run in AERMOD
("direct") mode. Even wind speeds as low as 0.00001 meters/second (m/s) were passed to
AERMOD, which then calculated a correspondingly high concentration. AERCOARE has the
ability to set a minimum wind speed as an option, and on 8 March 2013 EPA issued a
clarification memo discussing the use of a 0.5 m/s minimum wind speed threshold for the
AERMOD modeling system (USEPA, 2013). An option setting the minimum wind speed for
AERMOD related processing was therefore added to MMIF, with a default value of 0.5 m/s.
-------
Two additional updates to MMIF occurred after the end of work on Task 1 which is noted here.
The minimum mixing height that MMIF would produce had been set to the middle of the lowest
model layer, consistent with CALPUFF requirements. Following the recommendations for
AERCOARE settings, an optional minimum mixing height of 25 m was added to MMIF. Similarly
and also following the AERCOARE recommendations, an option controlling the minimum
absolute value of the Monin-Obukhov length was added, with a default value of 5 (|L| > 5).
These were required to be able to make a fair comparison between WRF+MMIF+AERMOD
output and WRF+MMIF+AERCOARE+AERMOD output as described in the following Section 3.
More information on these additional changes to MMIF can be found in Volume 3, the full report
for Task 3 (Analysis ofAERMOD Performance Using Weather Research and Forecasting Model
Predicted and Measured Meteorology in the Arctic).
2.4.2 Suitability of the two WRF datasets
The Alpine/CP WRF 4 km domain dataset does not extend far enough east to cover the
Beaufort Sea where many potential lease blocks exist and several have already been leased.
Additionally, the Alpine/CP WRF dataset covers only the open water period of three years and
could not be used in dispersion modeling assessments of permanent sources. For these
reasons, it was deemed unsuitable for the long term needs of this project and future permitting.
Due to the problems in the UAF/BOEM initial 5 year WRF dataset noted above in Section 2.3, it
was also deemed unsuitable for the long term needs of this project and future permitting. The
decision was made to pursue new hindcast WRF modeling under Task 3, and not wait for the
final UAF/BOEM 31 year dataset to become available.
In addition, the smallest WRF mesh size was 10 km. Potential sources, especially in the
Beaufort Sea, would likely be within 2 - 3 grid points or closer to the shoreline. In order to better
resolve the open water transport between source and the shoreline, a simulation with a finer
mesh size is desirable. The inner domain of the WRF modeling described in Task 3 has a mesh
size of 4 km.
-------
[Blank]
10
-------
3 TASK 2 - EVALUATION OF WEATHER RESEARCH AND FORECASTING
MODEL SIMULATIONS FOR FIVE TRACER GAS STUDIES WITH AERMOD
This section summarizes the Task 2 tracer dispersion studies reported in Volume 2, presenting
the methodology used for each of the elements of the task, describing the results of the
performance evaluations, and analyzing how modeling options affect model performance. The
protocol for this task is in Appendix A.
The purpose of the task is to provide evidence to help answer some of the following questions:
• How well does WRF predict overwater surface meteorology?
• Are pollutant concentrations predicted by AERMOD driven by WRF meteorology as
conservative as those predicted by AERMOD driven by observations (processed through
AERCOARE)?
• What WRF modeling configurations and meteorology extraction methods provide the
best AERMOD inputs, based on the most accurate AERMOD predictions?
• How sensitive is AERMOD to differences between the WRF extracted meteorology and
observations for simulations of typical offshore sources?
To answer these questions five historical tracer dispersion field studies were selected for this
task:
• Ventura, CA: September 1980 and January 1981;
• Pismo Beach, CA: December 1981 and June 1982;
• Cameron, LA: July 1981 and February 1982;
• Oresund (between Denmark and Sweden): May/June 1984; and
• Carpinteria, CA: September 1985.
The four North American studies have been used for previous off-shore dispersion model
development. The tracer experiment datasets are well known to EPA and have a history of use
for model benchmark testing and development. The Cameron and Pismo Beach studies provide
tracer measurements for simple level terrain near the coastline and are useful for analyzing
model performance with marine influence only. The Ventura study also involves simple flat
terrain, but the receptors are located 500 m to 1 km inland from the shoreline. The Carpinteria
study involved short distance, low wind transport conditions and receptors located on tall bluffs
along the shoreline. The Oresund study involved longer transport distances (25 km to 40 km)
with tracer releases on both sides of the Oresund strait separating Denmark and Sweden.
Mesoscale model results generally have some bias and no single configuration provides the
best simulation in all circumstances. Since a single choice of model setup represents a single
deterministic solution, the accuracy and variability of the WRF model are critical to evaluate its
success as an input to downstream dispersion models. The modeling of each historical field
study was conducted as an ensemble of simulations using two reanalysis data sets and three
PEL schemes, resulting in six possible combinations of reanalysis and PEL scheme. The two
reanalysis input data sets were: 1) European Center for Medium Range Weather Forecasts
11
-------
(ECMWF) Reanalysis Project (ERA) reanalysis data and 2) North American Regional
Reanalysis (NARR). The PEL schemes were: 1) Yonsei University (YSU) (Hong et al., 2006); 2)
Mellor-Yamada-Janjic (MYJ) (Mellor& Yamada, 1982; Janjic,1994); and 3) University of
Washington Shallow Convection (Bretherton & Park, 2009).
AERMOD requires two input files of meteorology: a file of scalar values (SFC file) and a file of
multi-level values (PFL file). The files can be generated for AERMOD directly or indirectly by
MMIF using the fields available in the WRF output files. Four extraction methods were used to
generate the necessary files for AERMOD:
1. MMIF was applied to extract and prepare data sets for direct use by AERMOD (MMIF
produces the AERMOD SFC and PFL input files directly). The PEL height predicted by
WRF is used in the SFC.
2. Same as Method 1, but the PEL height was rediagnosed from the wind speed and
potential temperature profiles using the bulk Richardson algorithm within MMIF.
3. MMIF was applied to extract the key meteorological variables of overwater wind speed,
wind direction, temperature, humidity, and PEL height from WRF results. The MMIF
extracted data were used to build an AERCOARE input file. AERCOARE used these
variables to predict the surface energy fluxes, surface roughness length and other
variables needed for the AERMOD simulations. For this task, AERCOARE was applied
using the defaults recommended in the AERCOARE model evaluations study (Richmond
& Morris, 2012).
4. Same as for Method 3, but the PEL height was rediagnosed using the bulk Richardson
algorithm within MMIF.
The naming convention and description of the four extraction methods are listed in Table 2.
Note that "RCALT" refers to extractions with MMIF and rediagnosed PEL height and "RCALF"
refers to direct use of WRF PEL height (with the minimum 25 m PEL height applied). Also, note
that "AERC" refers to simulations with additional AERCOARE processing after MMIF extraction,
and "MMIF" refers to simulations using the WRF meteorology directly without further processing
by AERCOARE.
Table 2. WRF AERMOD Meteorology Extraction Methods.
WRF Extraction
Method
1) MMIF.RCALF
2) MMIF.RCALT
3) AERC.RCALF
4) AERC. RCALT
WRF-
Process Path
WRF -» MMIF -» AERMOD
» MMIF (with PEL diagnosis) -» AERMOD
WRF -» MMIF -» AERCOARE -» AERMOD
WRF -» MMIF
(with PEL diagnosis) -» AERCOARE -» AERMOD
12
-------
For the measurement based simulations, AERCOARE was applied using default options for
surface roughness, warm-layer heating, and cool skin effects; observed PEL heights for the
convective PEL height; mechanical PEL heights using the Venkatram option (Venkatram, 1980);
a minimum PEL height of 25 m, and a minimum |L| of 5 m. For these cases, AERMOD
simulations were performed with ("Case 13") and without ("Case 2") the measured standard
deviation of wind direction (oe), that is,
• Case 1: Require Abs(L) > 5 m, use oe measurements, and use the Venkatram equation
for zim and require zim > 25 m.
• Case 2: Require Abs(L) > 5 m, use AERMOD-predicted oe, and use the Venkatram
equation for zim and require zim > 25 m.
A total of 3,151 AERMOD simulations were conducted to account for the various combinations
consisting of:
• Five tracer experiments, each tracer release case simulated separately,
• Six WRF configurations,
• Four WRF-meteorology extraction methods,
• Measurement-based AERMOD simulations using Case 1 and Case 2 options.
WRF performance was assessed in two ways: quantitatively by computing statistics relating
WRF-predicted surface meteorology to observed values and qualitatively by graphical
comparison of extracted WRF meteorology to observed values. The quantitative analysis was
conducted using publically-available software, METSTAT (ENVIRON Int. Corp., 2014).
METSTAT calculates a suite of model performance statistics using wind speed and direction,
temperature, and moisture observations. WRF predictions are extracted from the nearest grid
cell for comparison to the observed values. METSTAT computes metrics for bias, error, and
correlation and compares them to a set of performance benchmarks set for ideal model
performance (Emery, et al., 2001). Graphical analysis includes Q-Q plots comparing predicted
versus observed concentration probability distributions. Log-log scatter plots are employed to
evaluate the temporal relationship between observed and predicted concentration. The
statistical and graphical results are presented in Volume 2.
The results of the task suggest that small differences in the key meteorological variables can
result in large differences in predicted tracer concentration for a given hour. Although many of
the WRF simulations perform quite well when compared to regional surface observation of
3 Note that only Case 2 was evaluated for Oresund because overwater oe data were not available for the
period of the study.
13
-------
winds and temperatures, small differences near the overwater point of release can result in
prediction of the opposite stability (stable vs. unstable or vice-versa).
Relatively small error in air temperature or sea surface temperature (SST) can have a large
effect on stability class because stability is a function of the "sign" of air-sea temperature
difference. Warm air advected over cool water results in stable conditions, while cool air
advected over warm water results in convective unstable conditions. Spatial gradients of SST
near the coast and wind direction thus play key roles in the simulation of the stability and
planetary boundary layer (PBL) heights over the water.
The modeling performance analysis of the five tracer experiments demonstrated WRF based
AERMOD simulations can result in estimates of concentration as good as or better than
AERMOD simulations using observations - but not in all cases.
For the Cameron, Pismo, Oresund, and Carpinteria studies, some of the AERMOD simulations
driven by WRF meteorology had better or similar performance statistics than simulations driven
by observed meteorology. The poorer performing AERMOD simulations, both WRF driven and
observation driven occurred when the meteorological inputs produced atmospheric stability
conditions that were not likely representative of the larger-scale stability at the time of the study.
This observation, however, is a fundamental flaw of dispersion modeling that relies on
meteorology at a single point. Models that use a 3-dimensional grid of meteorological variables
are likely more appropriate for dispersion modeling of heterogeneous conditions.
For the Oresund study, all AERMOD simulations performed poorly when compared to observed
tracer concentrations. This study was characterized by overwater and overland transport,
25-40 km transport distances, high elevated releases, and observed shoreline fumigation. Poor
model performance in this instance is likely the result of the limits of AERMOD's formulation, not
inaccurate characterization of surface conditions over water predicted by WRF.
It should also be noted that the limits suggested by Richmond & Morris (2012), namely limiting
the PBL height to be greater than 25 m and the absolute value of L to be greater than 5, were
only implemented in cases that applied AERCOARE and not the direct WRF-MMIF extraction
cases.
Based on the results of this study, the following conclusions can be made:
• ERA reanalysis datasets offered better WRF predictions of meteorology regionally, but
NARR reanalysis datasets performed better in some cases at the local tracer study
meteorology measurement sites. ERA-based runs resulted in better AERMOD results
overall. Both YSU and UW PBL schemes resulted in better predictions of meteorology
overall, leading to better AERMOD predictions.
• The METSTAT analyses suggested most WRF simulations met the performance criteria
goals for "complex terrain" conditions. The comparison of overwater measurements from
the archived buoy data suggested the METSTAT performance could be used as a
predictor at the site, despite a lack of overwater measurements in the METSTAT
analysis itself. However, the meteorological analysis suggests small errors in SST and
14
-------
air temperature can result in misdiagnosed stability conditions that can have profound
effects on the AERMOD results.
• The results suggest representative SST data are necessary to prevent misdiagnosis of
surface-layer heat flux and stability. The SST data from the periods of the tracer studies
integrated into the reanalysis data are not as representative or as resolved as today's
datasets. Today SST data are collected from sophisticated satellites at high resolution. It
is likely modern SST data are more accurate and air-sea temperature differences
estimated by WRF are less likely to result in a misdiagnosis of atmospheric stability
conditions.
• AERMOD results produced using meteorology extracted from the ERA-YSU WRF
simulations produced the simulations with the highest frequency of top performing Total
Model Score (TMS). This combination also tended to more closely match the upper-
range concentration predictions.
• Direct extraction by MMIF without AERCOARE produced more cases where
concentration predictions were conservative.
The MMIF rediagnosis of PEL height should be used to prevent excessively low PEL heights in
the SFC files.
15
-------
[Blank]
16
-------
4 TASK 3 - ANALYSIS OF AERMOD PERFORMANCE USING WEATHER
RESARCH AND FORECASTING MODEL PREDICTED METEOROLOGY AND
MEASURED METEOROLOGY IN THE ARCTIC
This section summarizes the Task 3 AERMOD performance evaluation reported in Volume 3.
The purpose of this task is to evaluate alternative methods for supplying meteorological
variables to AERMOD for regulatory air quality modeling of sources located over the water. The
protocol for this task is in Appendix A.
It is hypothesized given an appropriate overwater meteorological dataset that AERMOD can be
applied for NSR following the same procedures as used for sources over land. This task
evaluates a combined modeling approach where the meteorological variables are provided by
WRF, and then processed by a combination of MMIF and, optionally, AERCOARE. The
extracted meteorology is used to drive AERMOD for several test cases. The results are then
compared to results of AERMOD driven by observational datasets over the Chukchi and
Beaufort Seas along the Arctic coasts of Alaska.
The aim of this task is to provide evidence to help answer the following questions:
• How well does WRF predict overwater surface meteorology in the Arctic?
• Are pollutant concentrations predicted by AERMOD driven by WRF meteorology as
conservative as those predicted by AERMOD driven by observations (processed through
AERCOARE)?
• What WRF modeling configurations and meteorology extraction methods provide the
best AERMOD inputs?
• How sensitive is AERMOD to differences between the WRF extracted meteorology and
observations for simulations of typical OCS sources?
The first part of this task generated a WRF meteorological dataset suitable for dispersion
modeling in the Arctic, employed various combinations of MMIF and AERCOARE to extract
modeled and observational meteorology over water, and used these datasets to drive AERMOD
simulations for ice-free periods of 2009 - 2012, where overwater observational datasets were
available. Results from the buoy-based and WRF-based AERMOD simulations were compared
and contrasted to address the questions above.
Meteorological observation datasets from four overwater locations were obtained for this task.
Two of the locations were in the Beaufort Sea and two were in the Chukchi Sea. Data were
available at these locations for various time-spans during the ice-free summer and autumn
periods of 2010, 2011, and 2012. The sites B2, B3, C1 and C2 are shown in Figure 3 and
described in detail in Volume 3. The 2010 - 2012 period was selected to take advantage of the
vertical temperature profiler data collected at Endeavor Island during this period. The profiler is
a passive microwave radiometer operating from 2010 to 2012 at the offshore Endeavor Island
facility near Prudhoe Bay, Alaska. The profiler data were used to assist in the estimates of
atmospheric planetary boundary layer (PBL) height at each of the sites. The term "PEL height"
17
-------
is used to indicate the height or depth of the mixing layer and is synonymous with "mixing
height." These terms will be used interchangeably throughout this task.
Volume 3 summarizes the methodology and results for each element of the investigation,
including:
• The methodology used for the WRF simulations,
• Evaluation of the WRF performance,
• The methodologies used to prepare AERMOD meteorology from both the observational
datasets and the WRF simulations,
• The AERMOD modeling approach and methodology,
• Evaluation of the AERMOD results and comparisons of observation-based and
WRF-based AERMOD results, and
• Examination of the influence of the meteorological data on AERMOD performance.
300-
200-
100-
o-
-100-
-200-
: Chukchi-Burger, 2010, 2012
: Chukchi-Klondike,
.—."
B2:
Point Lay
$• Buoy Site
-fr WRF grid point
extracted
• WRF grid points
B3: Beaufort-Sivultiq, 2010-2012
3V~r-
ENDV: Endeavor-Island,
-500
-400
-300
-200
1
2100
2000
1900
1800
1700
1600
1500
1400
1300
1200
1100
1000
900
800
700
600
500
400
300
200
100
1
-100
"elevation (m)
-100
100
200
300
400
Figure 3. Overwater meteorological measurement sites and
corresponding WRF inner domain extraction points.
EPA provided five unique source group configurations for this study (Wong, 2012), as shown in
Table 3. Each group represented a hypothetical OCS source with stack characteristics typical of
drill ship sources that have operated on the OCS in the recent past or have been proposed in
recent permit applications for the Arctic. Fourteen stacks were divided among the five source
groups and located at the center of a hypothetical drill ship configuration (Figure 4). Each
18
-------
source group contained multiple vertical stacks with warm, buoyant plumes. Stack heights for all
14 sources ranged from 10 m to 39 m.
Table 3. Hypothetical Drill Ship Emission Sources.
Source Unit
Diesel engine
1
Incinerator
Diesel engine
2 Boiler
Incinerator
Propulsion engine
3 Generator
Boiler
Diesel engine
4 Winch
Heater
Diesel engine
5 Boiler
Incinerator
Source
ID
S1P1
S1P2
S2P1
S2P2
S2P3
S3P1
S3P2
S3P3
S4P1
S4P2
S4P3
S5P1
S5P2
S5P3
Stack
Height
(m)
16
14
18
17
10
25
20
15
39
25
23
18
17
10
Stack
Gas
Temp
(°K)
700
550
680
500
525
570
610
420
580
580
510
680
500
525
Stack Gas
Exit
Velocity
(mis)
30
20
28
10
17
30
22
2
21
14
42
28
10
17
Stack
Diameter
(m)
0.50
0.40
0.40
0.45
0.40
0.60
0.25
0.30
0.70
0.20
0.15
0.40
0.45
0.40
Downwash
No
No
No
No
No
No
No
No
No
No
No
Yes
Yes
Yes
For the first four source groups (totaling eleven stacks), building downwash was not applied.
AERMOD can account for the influence of the wakes of structures on downwind concentrations
using the Plume Rise Model Enhancements (PRIME) model. To examine the effects of building
downwash, Source Group #2 (with three stacks) was modeled as Source Group #5 with the drill
ship structure shown in Figure 4.
19
-------
200m
-200 -150 -100 -50
(m)
Figure 4. Source locations and structures and innermost receptor rings.
AERMOD predicts pollutant concentrations at locations based on their distance from a source.
For this task, a network of 50 receptor rings was used. Each ring contained 360 receptors at 1°
spacing. The rings were centered at the same origin as the sources with incremental radial
spacing of downwind distance based on a geometric series from 30 m to 10 km as shown in
Figure 4. Receptors were placed at a height of 0.0 m (no flagpole receptors). The total number
of receptors was 18,000. The vessel to the south had no sources nor downwash influence on
the drill rig sources. The vessel was considered part of the ambient air.
The meteorological data were extracted from WRF and processed by MMIF the same way as
described in Task 2. Similarly, the same limits on Monin-Obukov length and PEL height
presented in Task 2 apply here as well (i.e., a minimum PEL height of 25 m, and a minimum |L|
of 5 m).
In this task, AERMOD concentrations were calculated using meteorology from overwater
observations and meteorology extracted from WRF simulations. The maximum predicted
concentrations at each receptor ring were extracted and the observation-driven AERMOD
results were compared directly to the WRF-driven AERMOD results. This approach simplifies
20
-------
the investigation of the bias of the WRF simulations and removes some of the influence of wind
direction differences.
AERMOD version 14134 was used for this study, using all regulatory default options except the
"VECTORWS" flag (wind speeds are vector mean (or resultant) wind speeds, rather than scalar
means) was used for WRF wind speed. It was necessary to specify the "Beta" option in
AERMOD to use the MMIF extracted data, which allows for new features in AERMOD that are
in a draft BETA-test status. Five different block averaging periods were simulated for each
combination of site, year, and source type: 1-hour, 3-hour, 8-hour, 24-hour, and period-long
averaging periods. These averaging times were selected because they correspond to the
averaging periods applicable to the NAAQS. Comparison with statistical standards was not
performed. To ensure tracer emission rate independence, AERMOD simulations were
conducted using a stack unit emission rate of 1 g/s. The resulting AERMOD concentrations
were divided by the tracer release rates to provide normalized concentration with units of us/m3.
This study consisted of 1,125 AERMOD simulations and were conducted to satisfy all of the
possible scenarios:
• 3 sites per year made up of different combinations of sites:
o 2010: B2, B3, C2
o 2011: B2.B3.C1
o 2012: B3, C2, C1
• 3 years of WRF simulations (2010-2012),
• 5 source groups,
• 5 averaging periods,
• 5 meteorological datasets:
i) Observations
ii) MMIF.RCALF WRF extractions
iii) MMIF.RCALT WRF extractions
iv) AERC. RCALF WRF extractions
v) AERC. RCALT WRF extractions
The definitions the WRF extractions in ii) through v) are explained in Table 2 of Section 3.
A set of summary statistics, defined in Volume 3, were calculated for each simulation to
evaluate the performance of WRF-based simulations compared to the observation-based
simulations. The summary statistics can be found in Appendix B of Volume 3.
In summary, the analyses suggest WRF was able to produce hourly meteorological datasets
that compared favorably to overwater measurements. The regional METSTAT analyses found
temperature and wind speed were within simple-terrain criteria for the majority of periods. Wind
speed at overwater sites was biased high in October only. Overall, WRF meteorology at all four
sites agreed with measured data, but with some biases. Sea Surface temperatures were in
21
-------
agreement a majority of the time, but varied up to several degrees for short periods. WRF PEL
heights were biased low on average during stable periods, but were of similar magnitude to
measurements during unstable periods.
From a qualitative perspective, most of the WRF-based AERMOD simulations resulted in
concentration maxima that were favorable, being within a factor-of-two of the observation-based
AERMOD simulation results and producing RHC within 10-20% of the observation-driven
AERMOD results a majority of the time. Maximum concentrations at distances greater than
1,000 m from the source tended to be conservative. Maximum concentrations within 1,000 m of
the source were underpredicted for the tall stack simulations (i.e. Source Groups #3 and #4)
due to the persistence of overly stable conditions that prevented near-source mixing to the
surface. In general, maximum concentrations tended to occur from 100 m to 1,000 m during
unstable conditions characterized by higher PEL heights due to the increased rate of vertical
mixing. The WRF-based predictions for Source Group #5 (downwash cases) were consistently
the best performers of the five source groups when compared to the observation-based
predictions.
The main conclusions of the study are summarized as responses to the set of questions below:
• Is there a consistent bias across source type and/or location (e.g. Chukchi vs. Beaufort)?
In particular are there any instances where the WRF simulations result in a bias towards
underprediction compared to using the buoy observations?
For the sources considered in this task, the absolute maximum concentrations occurred
within 1000 m of source during unstable conditions. There was no consistent bias at the
Chukchi or Beaufort sites for maximum short term average concentrations. Prediction
accuracy (with respect to observation-based predictions) was better at the Chukchi sites
because the WRF-MMIF PEL heights were used for the observation-based simulations
(no PEL height measurements were available for these sites). With respect to averaging
time, the long-term period-averaged concentrations were underpredicted in some
instances. For example, the simulations using the MMIF recalculated PEL heights
underpredicted the long-term maximum concentrations at site C2.
The FF2 scores were persistently lower at site B2 than the other sites. Concentrations
were typically overpredicted at this site at distances greater than 1000 m and
underpredicted at distances less than 1000 m due to the high frequency of minimum
PEL height.
At distances less than 1000 m, WRF-based Source Group #4 simulations
underpredicted concentration with respect to observation-based simulations. It was
found that the taller stack groups (Source Groups #3 and #4) were more sensitive to
differences in meteorology than the other groups. At distances greater than 1000 m, tall-
stack maximum concentrations were underpredicted in cases where PEL height was
overpredicted. High PEL height corresponded to unstable conditions that promote
vertical mixing and concentration maxima for tall stacks in the near-source (< 1,000 m)
but promote lower concentrations in the far-source (> 1,000 m). On the other hand, tall-
22
-------
stack concentrations were underpredicted near the source when the PEL height was
underpredicted.
The MMIF-rediagnosis (RCALT) of PEL height tended to improve WRF-based AERMOD
performance by producing PEL heights that agreed better with the observation-based
PEL heights.
• For locations where WRF performed better, does that ultimately translate to different
dispersion model results?
The short-term maximum concentrations were less sensitive to bias in the WRF results.
This was likely because the concentration maxima occurred during the extreme
atmospheric stability conditions (either stable or unstable). The optional MMIF PEL
height and L limits result in observation- and WRF-based meteorological simulations that
are quite similar during the most extreme conditions.
The ice free period average maximum concentrations at distances exceeding 1000 m
were the most sensitive to the long-term meteorology bias. Underpredicted wind speeds
at Sites B2 and B3 favored conservative period-averaged concentrations greater than
1000 m. Site B2 period average far-source concentrations were highly conservative with
RHC values greater than a factor of two of the observation-based concentrations.
It is highly recommended that FNMOC SST analysis, or a similar high-resolution SST
dataset based on both remotely-sensed and in-situ measurements be used instead of
alternative datasets such as the NCEP RTG for simulations of open-water periods in the
Beaufort Sea. The Mackenzie River warm-water outflow plume is a prevalent feature on
the Beaufort Sea in summer. Low resolution SST analysis or excessive smoothing may
result in erroneous air-sea temperature difference estimates. The FNMOC SST analyses
gave a better spatial and temporal description of the SST distribution and gradient
across the Beaufort Sea over the 2010-2012 periods analyzed as discussed in Section
5.1.2.
• Did it make any difference when WRF predictions were processed by AERCOARE as
oppose to direct use for predictions of the surface energy fluxes?
Overall, there was little discernible advantage in using AERCOARE. Considering
average TMS, the "MMIF" runs (direct extraction from WRF without AERCOARE
processing) resulted in slightly higher scores. However, for the 1-hour averaging period
RHC scores, the AERC.RCALT simulations performed the best overall. The MMIF
recalculation of PEL height has a much greater influence (than AERCOARE processing)
on the accuracy of the simulations.
• Does it make any difference when PBL heights are rediagnosed by MMIF?
Overall, maximum concentration results were more accurate and more conservative
when the MMIF rediagnosis was applied.
23
-------
The concentration results from the shorter stack groups (Source Groups #1, #2, and #5)
and downwash-affected sources were less sensitive to differences in the PEL height. If
the plume is already near ground level, maximum concentrations at ground level will
occur in the near the source and are less sensitive to the height of the PEL.
Concentration maxima from taller stacks are much more sensitive to the PEL height.
Note that the height of the tall stacks used in this study is near to the minimum PEL
height (25 m). If the minimum PEL height was greater than the tallest stack, it is likely
that concentration estimates would be more comparable.
24
-------
5 TASK 4 - COMPARISON OF PREDICTED AND MEASURED MIXING
HEIGHTS
5.1 Overview and Objective
The goal of this Task is to compare upper-air observations with WRF simulations.
5.1.1 Upper-air and surface observations
In spring 2010, a Kipp & Zonen radiometric profiler and an instrumented meteorological tower
was installed on Endeavor Island (or Endicott) (Hoefler Consulting Group, 2003). The profiler
collected upper measurements from mid May 2010 and to late November 2012 while the tower
recorded surface data from mid May 2010 to the end of August 2013. Endeavor is a man-made
island located off the northern coast of Alaska in the Beaufort Sea and 15 miles northeast of
Prudhoe Bay. A three mile gravel road connects Endicott's two gravel islands; the Main
Production Island (MPI) and the Satellite Drilling Island (SDI). The location where the station
was installed is indicated in Figure 5. The profiler and a view of the co-located meteorological
station is shown in Figure 6.
The profiler is a passive microwave radiometer that measures radiances emitted from the
atmosphere with an optical path length of a few hundred meters. Every few minutes, the
radiometer performs a series of measurements at different angles ranging from horizontal to
vertical. From these radiances, a profile of air temperature can be inferred via a mathematical
technique. Values are reported every 10 m up to 100 m, every 25 m up to 200 m, and every 50
m up to 1000 m. No data is available above 1000 m. The soundings of temperature, when
compared with co-located radiosonde ascents, appear somewhat "smoothed" in the vertical due
to the nature of the retrieval. No retrieval of wind speed, wind direction, or humidity are available
from this profiler. The profiler was installed on the north end of SDI and pointed due north out
over the water.
Located next to the profiler was a 12 m meteorological tower, instrumented to measure wind
speed and wind direction at 10 m, temperature at 10 and 2 m, and solar radiation. The
temperature measurement is used to scale the retrieval because the profiler senses the change
of temperature in the vertical. The distance from the station to the water's edge was on the
order of 10 m which is short enough such that the internal thermal boundary layer growing from
the land surface would not affect the air measurements at the site. The measurements, both
radiometric and traditional, can therefore be assumed to be representative of the adjacent
ocean.
Periodically, the Japan Agency for Marine-Earth Science and Technology (JAMSTEC) sends
one of its research vessels into the Chukchi and/or Beaufort Seas for a research cruise. Such
cruises occurred in 2008, 2009, 2010, and 2012. Hourly near-surface measurements and twice-
daily upper-air measurements were collected in addition to the other scientific objectives of each
cruise. There is a delay in releasing these data to the public to allow JAMSTEC researchers to
publish journal articles based on the data. The upper-air soundings from 2008 and 2009 have
been released, but the soundings from 2010 and 2012 have not yet been made public.
25
-------
The twice-daily radiosondes released at Barrow (PABR) are not representative of the marine
boundary layer, and are not considered part of this analysis. However, PABR data (both
Monitoring Site Location
Figure 5. Location of the profiler station on Endeavor Island
Figure 6. The K&Z profiler (left) and a view of the station looking North (right)
26
-------
sounding and surface data) was used in the observational nudging of the 4 km domain of the
WRF simulations described below. WRF performance at PABR should therefore not be
considered an independent test unlike the profiler or JAMSTEC comparisons (which were not
used in the nudging dataset).
5.1.2 WRF simulations
As summarized in Section 4 and detailed in Volume 3, a three-year WRF run was produced,
spanning the years 2009-2011. This simulation used the NCEP RTG SST dataset (NOAA,
Environmental Modeling Center, no date [n.d.]) for the lower boundary condition and as
initialization data required by WRF. This dataset is a smoothed 0.083 degree (1/12th degree in
latitude and longitude or approximately 9 km) once-daily SST analysis. This WRF simulation
was later extended to 2012 and 2013, though only 2012 was completed before problems were
detected. It was discovered that the RTG SST data did not handle the Mackenzie River outflow
very well, leading to large (8 -12 °C) errors in WRF SST compared to the buoys in the Beaufort
Sea. An unpublished re-run of the RTG product for much of the problematic period on the NCEP
FTP server was found but that re-processed dataset only improved the WRF SST performance
at the Beaufort buoys slightly. This dataset is referred to as "RTG-WRF".
A second WRF simulation was run using the SST analysis from the FNMOC (USGODAE Data
Catalog, n.d.) in place of the RTG SST product. This analysis is produced four times per day at
a horizontal resolution of 9 km. Only the periods when buoys were deployed were re-run using
this SST dataset. This dataset was used in the Task 3 analysis comparing WRF-driven
AERMOD results to buoy-driven AERMOD results, so only the open-water period when buoy
data exists was required. This dataset covers August, September, and October of 2010, 2011,
and 2012. This dataset is referred to as "FNMOC-WRF".
The FNMOC-WRF output was used in the analysis involving the Endeavor Island profiler
because of its location to the Beaufort Sea buoys which had shown problems related to the use
of the RTG SST product. This limits the analysis to the three years of open-water periods - 2010
to 2012 - using the FNMOC-WRF.
Due to the constraints of the overlap in time between the JAMSTEC data and the two WRF
simulations, only the RTG-WRF was used in the analysis involving the JAMSTEC soundings -
the 2009 cruise (the FNMOC-WRF does not begin until 2010).
5.2 Analysis
The Kipp & Zonen radiometer's sub-hourly profiles were averaged to create a mean sounding
for each hour of operation. The 4 km domain WRF output for the grid cell containing the profiler
was extracted, with no horizontal interpolation between the nearest grid cells. Although the
profiler data can be plotted along with WRF data, the profiler's retrievals are smoothed as a
result of the averaging of the sub-hourly profiles. Consequently, a direct comparison was not
considered useful. Instead, this analysis focuses on mixing heights derived (analyzed) from the
profiler temperature data since concentrations estimated by AERMOD can be sensitive to this
value.
27
-------
The mixed layer height was analyzed for each observed sounding (profiler, JAMSTEC, PABR)
using two different approaches:
1. A plot of the sounding, visually identifying the base and top of any inversion layer. These
are labeled "ZiBase" and "ZiTop" in the plots that follow.
2. The profiler soundings were numerically analyzed using the Critical Bulk Richardson
Number (CBRN) calculation. These are labeled "ZiRib" in the plots that follow.
The Richardson Number (Ri) (Vogelezang and Holtslag, 1996) is named after Lewis Fry
Richardson (1881 -1953), and is essentially the ratio of the potential energy to the kinetic
energy of an air parcel. It is most often expressed using gradients, which are approximated
when using quantized data by the Bulk Richardson Number RiB.
Here "g"\s the acceleration due to gravity, "z"is the height above the surface, "9" is the
potential temperature, "U"\s the wind speed, "u." is the friction velocity, b is a constant taken to
be 100, and the subscript "s" refers to the near-surface observation (e.g. typical 10m wind
speed, 2 m temperature). Starting with the surface, the height "z"is increased until the Bulk
Richardson Number exceeds a critical value, called the CBRN. Experiments have identified a
CBRN of 0.25 over land (Vogelezang and Holtslag, 1996) and 0.05 over water (Gryning and
Batchvarova, 2003) when using this formulation.
Because the profiler does not provide a retrieval of wind speed, the Monin-Obukhov surface
layer similarity theory was used to extrapolate the wind speed measured at 10 m to the levels of
the profiler's temperature retrievals. The appropriate equation is:
Here "k"\s von Karman's constant, "z0" is the roughness length, "L"is the Monin-Obukhov
length, and "W" is the stability correction function.
In its output files, WRF includes the 2-dimensional variable "PBLH", listed as the "PEL HEIGHT"
in meters above ground level (AGL). Each PEL parameterization choice within WRF can have
its own definition of PEL height, and the same PBLH may not be diagnosed when given
identical profiles. Further, WRF's PBLH is not a continuous quantity, but is quantized to the
nearest layer's mid-point. This makes WRF's PBLH values dependent on the user's choice of
the vertical pressure coordinate (eta) levels in WRF. Because the conversion of the vertical
pressure coordinate levels to height above the surface varies in both time and space, this
estimate of the PEL height can be misleading. In the WRF simulations, the layers are
approximately 11m thick near the surface, -75 m thick near 500 m AGL, and -240 m thick near
1000 m AGL. This discrete set of WRF's PBLH therefore could lead to a 10 - 20% error. WRF's
PBLH is labeled as "ZiWRF" in the plots that follow.
28
-------
The WRF output was therefore processed using the MMIF twice: once using the "pass through"
of the WRF PBLH, and once using MMIF's numerical CBRN calculation of the PEL height.
MMIF's implementation of this numerical method is very similar to the one used to process the
profiler data, except that MMIF interpolates the WRF data vertically to provide a smoother (less
quantized) output. The numerical techniques used to analyze the profiler data did not attempt to
interpolate between the retrieval layers. They are labeled "ZiMMIF" in the plots that follow.
ZiMMIF is the result of the CBRN analysis of the WRF data, which ZiRib is the result of the
same type of analysis of the profiler's retrieval.
5.3 Results
5.3.1 FNMOC WRF evaluated using Endeavor Island profiler
Some typical sounding plots are shown in Figure 7 through Figure 9. The solid black line is the
profiler retrieval of temperature (left panels), converted to potential temperature (right panels)
using the hypsometric equation (AMS, 2012). The dashed black line is the WRF data profile.
The current conditions taken from an AERMET run using the 10 m tower data are printed near
the top-left of the right panels. The various estimates of the mixing height Z, are shown by
horizontal lines with labels. To summarize:
• ZiBase represents the typical American Meteorological Society (AMS) definition of an
inversion, the level where the temperature starts to increase with height. ZiTop is the
level where it starts to decrease again.
• ZiRis represents the CBRN technique applied to the profiler.
• ZiMMIF represents the CBRN technique applied to the WRF data.
• ZiWRF represents WRF's internal value for the PEL depth, i.e. the output variable
PBLH.
• Zim and Zic (when present) represent AERMET's mechanical and convective mixed
layer heights, respectively. AERMET is not always able to calculate these, due to
missing input data.
WRF is generally reproducing the shape of the profile, even if there is a constant offset in the
absolute value. Recall that the profiler uses the 10 meter air temperature measured
independently to "anchor" its profiles - it does not retrieve the absolute value of temperature,
just the change of temperature with height.
Note that the hand-analyzed inversion base is often at the surface (i.e. a "surface-based"
inversion) yet the wind speeds measured by the 10 meter tower are considerable. The shear-
driven turbulence alone should be creating a well-mixed layer, but presumably the strong
advection and surface fluxes are maintaining the profile of temperature. This quandary is what
the CBRN analysis is designed to solve. The RiB based mixed layer is a compromise between
the shear-driven turbulence mixing and the buoyancy-drive turbulence mixing (or suppression
thereof).
It is difficult to interpret the retrieval in Figure 9. A shallow super-adiabatic layer might exist near
the surface when there is large upward heat flux, but 200+ m deep layer aloft is not physically
29
-------
possible because the warmer, less dense air below (around 200 m) would rapidly exchange with
the cooler more dense air above (about 500 m) and the temperature gradient would be
moderated. The profiler may have been affected by the radiation from clouds, or other unknown
errors in its retrieval algorithm.
Figure 10 through Figure 15 show the various identifications of the mixed layer height, plotted
against to each other. Various statistics are shown in the upper-left corner, including the number
of points, the squared correlation coefficient, the root-mean-square error, the bias, and the
percentage of the data that are within a factor of 2 of the 1:1 "perfect fit" line. A least-squares fit
(the dashed line) that is forced through the origin is also plotted, and its slope given in the lower-
right corner.
As expected, ZiBase (the inversion base derived visually) has very little correlation with the RiB
based or WRF's identification of the mixed layer height. Even the profiler's CBRN mixing layer
has relatively little correlation with WRF's PBLH. In Figure 13, except for the lobe of points
where WRF + MMIF's diagnosed mixing height is high and the diagnosis of the profiler data is
low, there appears to be some correlation. Note that the profiler's highest retrieval is 1000 m
above the surface, but no similar constraint exists for the WRF + MMIF based Z,. The
quantization of the profiler data can also be seen, due to MMIF's ability to interpolate between
WRF layers when finding the RiB mixing height. Given the smoothed nature of the profiler's
retrievals, along with the other potential sources of error from the profiler and the somewhat
sensitive nature of the CBRN analysis, one might not expect a very high correlation.
AERMET's mechanical and convective mixed layer heights also do not compare favorably to the
profiler's mixed layer heights. The mechanical mixed layer height is generally too high, and
often would exceed AERMET's internal limit of 4000 m. Even the convective mixed layer height
is often much higher than the profiler's. This underscores why AERMET is not appropriate for
overwater conditions, as it assumes 90% of solar radiation reaching the surface goes into
heating (deepening) the mixed layer.
5.3.2 RTG WRF evaluated using JAMSTEC soundings
Evaluation of the FNMOC WRF run is not possible, because JAMSTEC has not yet released to
the public the soundings from cruises after 2009, and the FNMOC WRF run started in 2010.
Instead, a previous version of the WRF dataset developed under Task 3 will be used. This WRF
simulation for 2009 used the RTG SST product, which spread the influence of the warm
Mackenzie River outflow far into the Beaufort Sea. The JAMSTEC radiosondes were launched
far enough West (see Figure 16) to escape the most of the influence of this problem. Sample
sounding plots showing WRF and JAMSTEC soundings are shown in Figure 17 through Figure
20. There were a total of 47 atmospheric soundings taken within the 4 km domain during this
cruise. They were taken between September 10 and October 11, 2009. These data were not
used in the observational nudging of the WRF run; this is an independent evaluation of WRF's
performance.
30
-------
Figure 21 shows a comparison between the mixing heights derived via the CBRN technique
applied to the JAMSTEC soundings and the WRF soundings. This is the same technique that
MMIF uses to rediagnose the mixing height. Forty-seven of the points plotted were from the
4 km WRF domain (plotted as red plus signs) and 101 points were from the 12 km domain.
Although the R2 value is relatively low, near % of the data points fall within the factor-of-two
lines.
WRF captures the essence and variation of the JAMSTEC soundings remarkably well during
this short period of open water. Although the elevated inversions are sharper in the observations
than in the simulation, the WRF inversions are probably strong enough to set the mixed layer
depth at approximately the correct height. Because of the small sample size and the resulting
lack of statistical significance, no CBRN analysis has been completed on the JAMSTEC
soundings. Note also that no surface-based inversions were observed during the 2009
JAMSTEC cruise.
5.3.3 FNMOC WRF evaluated using Point Barrow soundings
Sample vertical profile plots showing 4 km WRF modeled data and observed upper-air sounding
data from Point Barrow, AK (PABR) are shown in Figure 22 and Figure 23. WRF handled
temperature and moisture well through the vertical atmosphere with an accurate representation
of observed conditions within the PEL. In Figure 22, WRF better represented the radiation
inversion from the surface in the left panel. In Figure 23, the left panel depicts WRF slightly
over-predicting the strong subsidence inversion beginning around 700 m. The right panel
displays a slight cold bias at the surface with modeled temperatures warming a bit too quickly in
the first 200 m shallow layer in the early morning hours. The summertime vertical profiles reflect
improved model performance with WRF predicting the height and depth of inversions in the
lower levels of the atmosphere.
It should be noted that the PABR upper-air data were used in the observational nudging of the
WRF run, so it's not surprising that WRF is similar to PABR.
5.4 Conclusions
After reviewing the many sounding plots of the Profiler vs. WRF and the JAMSTEC soundings
vs. WRF, the CBRN technique is difficult to apply. Quite often in the JAMSTEC plots, there is a
small "hack" in the trace of wind speed at the lowest three levels. The lowest level reported in
weather balloon ascents is typically from another (collocated) instrument, not from the weather
balloon itself. The second level is actually the first from the sonde, and is often less than the first
or third report. This can be seen in the bottom panel of both sides of Figure 17, and is very clear
in Figure 19. This "hack" in the trace can cause large changes in the calculated bulk Richardson
number, leading to too much sensitivity in identifying the mixed layer.
For example, although the mixing heights identified using the WRF profile (horizontal blue line)
and using the sonde profile (horizontal red line) are similar, both are driven by relatively small
wind speed changes near the surface and both miss what a trained meteorologist would call the
mixed layer (800m for 2009-09-11_06 and 1800m for 2009-09-28_12).
31
-------
In the "RTG WRF" Figure 17 through Figure 21, the WRF profiles match the JAMSTEC sonde
profiles well with only relatively small errors in the tops of the mixed layer that a meteorologist
would identify from the potential temperature traces. The wind speed profiles are simply too
"noisy" with small changes ("wiggles") over layers ~100m deep, triggering the CBRN falsely and
leading to mixing layers that do not agree with what a meteorologist might identify from viewing
the sounding.
Based on the above evaluations, more studies are warranted on a consistent method to identify
mixing layers, both from WRF and from observed profiles, before the WRF-AERCORE
methodology can be considered complete. The comparisons of WRF vs. JAMSTEC profiles,
which, unlike the Point Barrow profiles, were not used to nudge the WRF run and are
independent, show how accurate WRF can represent the profiles. The method used to identify
the mixing height from any profile (WRF or measured) is, however, too sensitive to small
changes in the profiles.
32
-------
Profiler Sounding 1400 - 2010-07-29 07:00 LSI
Profiler Sounding 1400 - 2010-07-29 07:00 LST
ZiTop = 450
\ '._ ZiMMIF'»891
•, \
\1
•I
ZIWF1F = 515
8 10
Temperature (C)
12
14
U = 8.7 ZO = 0.05 L = 8888 RH = NA
Profiler = Solid, WRF = Dashed
278 280 282 284 286 288
Potential Temperature (K)
290 292
Figure 7. Profiler retrievals and WRF soundings for 2010-07-27 07:00 LST.
33
-------
Profiler Sounding 1937 - 2010-08-20 16:00 LST
Profiler Sounding 1937 - 2010-08-20 16:00 LST
ZiTop = 1000.
'*••.
"'.'
» *
'"•••x ""V
\ •-
\
'-, \
ZiWRF\B12
\
ZiRib = f50
'•-ZiMMIF ='638
y
* s
x *,
567
Temperature (C)
U = 17.4ZO = 0 05 L = 88 SB RH = NA
Profiler = Solid, WRF = Dashed
' ZiMMIF = 638
I T
276 278 280 282 284 286 288 290
Potential Temperature (K)
Figure 8. Profiler retrievals and WRF soundings for 2010-08-20 16:00 LST.
34
-------
Profiler Sounding 1842 - 2010-08-16 17:00 LST
Profiler Sounding 1842 - 2010-08-16 17:00 LST
10 12 14
Temperature (C)
16
18
U = 7.4 ZO = 0.05 L = 8888 RH = NA
Profiler = Solid, WRF = Dashed
I
ZiMH/IIF = 674
ZiRib - 35Q
275 280 285 290 295 300
Potential Temperature (K)
Figure 9. Profiler retrievals and WRF soundings for 2010-08-16 17:00 LST.
35
-------
Endeavor Island WRF vs. Hand-Analyzed Mixing Heights
o
o
o
OJ
o
o
IL
DC
N
0)
I
_i
in
o_
LJ_
DC
o
o
in
N = 6951
Rz = 15%
RMSE = 300
Bias = -19
In [1/2,2] = 32%
IM.* ii * • * I I
| I I I I • I I » 1 *,*
Slope = 0.64 _
500 1000 1500
Hand-Analyzed Inversion Base (rn)
2000
Figure 10. WRF's PBLH vs hand-analyzed inversion base (ZiBase).
36
-------
Endeavor Island MMIF vs. Hand-Analyzed Mixing Heights
o
o
o
OJ
N
2:
i?
ID
X
Cj)
TJ
0)
|