* — \
*1 PROt^
2016 Alaska State-wide Weather Research
Forecast (WRF) Meteorological Model
Performance Evaluation
-------
-------
EPA-454/R-20-003
May 2020
2016 Alaska State-wide Weather Research Forecast (WRF) Meteorological Model Performance
Evaluation
U.S. Environmental Protection Agency
Office of Air Quality Planning and Standards
Air Quality Assessment Division
Research Triangle Park, NC
-------
Executive Summary
In 2018, the EPA Office of Air Quality Planning and Standards proposed to develop a set of one-
year WRF datasets for regions not included in the prior CONUS WRF modeling domains
produced annually by the EPA. The Alaska, Hawaii, and Puerto Rico regions were selected for
this modeling effort. The modeling was intended as a preliminary testbed to expand the extent of
the national annual WRF runs provided by the EPA. Given the limited resources initially
assigned to the testbed effort, efforts were not made to determine optimized settings and
parameterization schemes for simulation of the regions modeled. For Alaska, the effort was
intended to provide a state-wide dataset at relatively low resolution, analogous to the EPA
CONUS WRF datasets. Like the prior annual EPA CONUS WRF runs, the modeling was not
intended to simulate flows in complex terrains that may require high resolution modeling to
resolve.
The 9 km 2016 Alaska WRF dataset was generated using Version 3.9.1.1 of the Advanced
Research WRF (ARW) core. The 9 km domain contains the entire state of Alaska except for a
few of the extreme western Aleutian Islands. This report provides a model performance
evaluation (MPE) of the dataset. Tools such as bias/error soccer-plots, parameter time series, and
wind roses were used to assess the performance of surface and upper-air meteorological
parameters such as temperature, wind speed, wind direction, and humidity. Also, state-wide
monthly precipitation maps were produced for qualitative comparison to observation-based
regional precipitation maps.
The analysis plots and other products contained in this report can be used by air permitting and
other authorities for project-specific model performance evaluations to determine if the dataset is
appropriate for regulatory use. Notably, this report itself does not represent an EPA endorsement
or validation of the dataset. Any use of the dataset for regulatory purposes (such as those
specified under 40 CFR Part 51, Appendix W) will require a project-specific performance
evaluation and approval of the evaluation by the regulatory authorities. However, the plots,
statistics, and other evaluation products included in this report can be adopted or referenced to as
part of any project-specific MPE.
Questions or requests concerning the 2016 EPA Alaska WRF dataset and the MPE can be sent to
Jay McAlpine of the EPA Region 10, mcalpine.iav@epa.gov or Kirk Baker of the EPA OAQPS
at baker.ldrk@epa.gov.
iv
-------
V
-------
List of Tables
Table 1. WRF domain configurations 2
Table 2. WRF vertical domain configuration 2
Table 3. Physics parameterization schemes used in the 2016 Alaska WRF model 4
Table 4. Surface meteorology performance benchmarks 7
vi
-------
List of Figures
Figure 1. Alaska 27-km and 9-km horizontal resolution WRF modeling domains 2
Figure 2. North Slope domain 8
Figure 3. Fairbanks domain 8
Figure 4. Cook Inlet domain 9
Figure 5. Juneau domain 10
Figure 6. "Alaska Peninsula" domain 11
Figure 7. Upper-air stations selected for the model performance study 12
Figure 8. Monthly 2-m temperature error and bias over the 9 km Alaska domain for 2016 14
Figure 9. Monthly mean temperature bias, January and February 2016 14
Figure 10. Monthly mean temperature bias, March and April 2016 15
Figure 11. Monthly mean temperature bias, May and June 2016 15
Figure 12. Monthly mean temperature bias, July and August 2016 15
Figure 13. Monthly mean temperature bias, September and October 2016 16
Figure 14. Monthly mean temperature bias, November and December 2016 16
Figure 15. Monthly 10-m wind speed error and bias over the 9 km Alaska domain for 2016.... 17
Figure 16. Monthly mean wind speed bias, January and February 2016 17
Figure 17. Monthly mean wind speed bias, March and April 2016 18
Figure 18. Monthly mean wind speed bias, May and June 2016 18
Figure 19. Monthly mean wind speed bias, July and August 2016 18
Figure 20. Monthly mean wind speed bias, September and October 2016 19
Figure 21. Monthly mean wind speed bias, November and December 2016 19
Figure 22. Monthly 10m wind direction error and bias over the 9 km Alaska domain for 2016.20
Figure 23. Monthly mean wind direction error, January and February 2016 20
Figure 24. Monthly mean wind direction error, March and April 2016 21
Figure 25. Monthly mean wind direction error, May and June 2016 21
Figure 26. Monthly mean wind direction error, July and August 2016 21
Figure 27. Monthly mean wind direction error, September and October 2016 22
Figure 28. Monthly mean wind direction error, November and December 2016 22
Figure 29. Monthly absolute humidity error and bias over the 9 km Alaska domain for 2016... 23
Figure 30. Monthly mean humidity bias, January and February 2016 23
Figure 31. Monthly mean humidity bias, March and April 2016 24
Figure 32. Monthly mean humidity bias, May and June 2016 24
Figure 33. Monthly mean humidity bias, July and August 2016 24
Figure 34. Monthly mean humidity bias, September and October 2016 25
Figure 35. Monthly mean humidity bias, November and December 2016 25
Figure 36. Anchorage seasonal, hour-of-day temperature, ASOS vs. Alaska 9-km WRF 26
Figure 37. Monthly 2-m temperature error and bias over the Cook Inlet subdomain for 2016... 27
Figure 38. Wind rose comparison, Anchorage, ASOS observed, Alaska 2016 WRF 28
Figure 39. Anchorage hour-of-day wind speed distributions, ASOS vs. Alaska 9-km WRF 28
Vll
-------
Figure 40. Monthly wind speed error and bias over the 9 km Cook Inlet domain for 2016 29
Figure 41. Monthly wind direction error and bias over the Cook Inlet domain for 2016 30
Figure 42. Anchorage hour-of-day absolute humidity, ASOS vs. Alaska 9-km WRF 31
Figure 43. Monthly absolute humidity error and bias over the Cook Inlet subdomain 32
Figure 44. Fairbanks hour-of-day temperature distributions, ASOS vs. Alaska 9-km WRF 33
Figure 45. Monthly 2-m temperature error and bias over the Fairbanks subdomain for 2016.... 34
Figure 46. Wind rose comparison, Fairbanks, ASOS observed, Alaska 2016 WRF 35
Figure 47. Fairbanks hour-of-day wind speed distributions, ASOS vs. Alaska 9-km WRF 35
Figure 48. Monthly wind speed error and bias averaged over the Fairbanks subdomain 36
Figure 49. Monthly wind direction error and bias over the Fairbanks subdomain for 2016 37
Figure 50. Fairbanks hour-of-day absolute humidity, ASOS vs. Alaska 9-km WRF (red). 38
Figure 51. Monthly absolute humidity error and bias over the Fairbanks subdomain 39
Figure 52. Deadhorse hour-of-day temperature distributions, ASOS vs. Alaska 9-km WRF 40
Figure 53. Monthly 2-m temperature error and bias over the North Slope subdomain for 2016.41
Figure 54. Wind rose comparison, Deadhorse, ASOS, Alaska 2016 WRF 42
Figure 55. Deadhorse hour-of-day wind speed distributions, ASOS vs. Alaska 9-km WRF 42
Figure 56. Monthly wind speed error and bias over the North Slope subdomain 43
Figure 57. Monthly wind direction error and bias over the North Slope subdomain 44
Figure 58. Deadhorse hour-of-day absolute humidity, ASOS vs. Alaska 9-km WRF 45
Figure 59. Monthly absolute humidity error and bias over the North Slope subdomain 46
Figure 60. Juneau airport seasonal, hour-of-day temperature, ASOS vs. Alaska 9-km WRF 47
Figure 61. Temperature error and bias averaged over the Juneau subdomain 48
Figure 62. Wind rose comparison, Juneau ASOS and Alaska 2016 WRF 49
Figure 63. Juneau hour-of-day wind speed distributions, ASOS vs. Alaska 9-km WRF 49
Figure 64. Monthly wind speed error and bias averaged over the Juneau subdomain 50
Figure 65. Monthly wind direction error and bias averaged over the Juneau subdomain 51
Figure 66. Juneau hour-of-day absolute humidity distributions, ASOS vs. Alaska WRF 52
Figure 67. Monthly absolute humidity error and bias averaged over the Juneau subdomain 53
Figure 68. Unalaska hour-of-day temperature distributions, ASOS vs. Alaska 9-km WRF 54
Figure 69. Monthly 2-m temperature error and bias over the Alaska Peninsula for 2016 55
Figure 70. Wind rose comparison, Unalaska ASOS vs. Alaska 2016 WRF 56
Figure 71. Unalaska hour-of-day wind speed distributions, ASOS vs. Alaska WRF 56
Figure 72. Monthly wind speed error and bias over the Alaska Peninsula subdomain 57
Figure 73. Monthly wind direction error and bias averaged over the Alaska Peninsula 58
Figure 74. Unalaska hour-of-day absolute humidity distributions, ASOS vs. WRF 59
Figure 75. Absolute humidity error and bias over the Alaska Peninsula subdomain 60
Figure 76. Inuvik, Canada upper-air distribution of temperature error 61
Figure 77. Inuvik, Canada upper-air distribution of wind speed error 62
Figure 78. Inuvik, Canada upper-air distribution of wind direction error 63
Figure 79. Inuvik, Canada upper-air distribution of relative humidity error 64
Figure 80. PAFC (Anchorage) upper-air distribution of temperature error 65
Figure 81. PAFC (Anchorage) upper-air distribution of wind speed error 66
viii
-------
Figure 82. PAFC (Anchorage) upper-air distribution of wind direction error 67
Figure 83. PAFC (Anchorage) upper-air distribution of relative humidity error 68
Figure 84. PAFA (Fairbanks) upper-air distribution of temperature error 69
Figure 85. PAFA (Fairbanks) upper-air distribution of wind speed error 70
Figure 86. PAFA (Fairbanks) upper-air distribution of wind direction error 71
Figure 87. PAFA (Fairbanks) upper-air distribution of relative humidity error 72
Figure 88. PABR (Utqiagvik) upper-air distribution of temperature error 73
Figure 89. PABR (Utqiagvik) upper-air distribution of wind speed error 74
Figure 90. PABR (Utqiagvik) upper-air distribution of wind direction error 75
Figure 91. PABR (Utqiagvik) upper-air distribution of relative humidity error 76
Figure 92. PASN (St. Paul Island) upper-air distribution of temperature error 77
Figure 93. PASN (St. Paul Island) upper-air distribution of wind speed error 78
Figure 94. PASN (St. Paul Island) upper-air distribution of wind direction error 79
Figure 95. PASN (St. Paul Island) upper-air distribution of relative humidity error 80
Figure 96. PANT (Annette Island) upper-air distribution of temperature error 81
Figure 97. PANT (Annette Island) upper-air distribution of wind speed error 82
Figure 98. PANT (Annette Island) upper-air distribution of wind direction error 83
Figure 99. PANT (Annette Island) upper-air distribution of relative humidity error 84
Figure 100. January 2016 Monthly total precipitation 85
Figure 101. February 2016 monthly total precipitation 86
Figure 102. March 2016 monthly total precipitation 86
Figure 103. April 2016 monthly total precipitation 87
Figure 104. May 2016 monthly total precipitation 87
Figure 105. June 2016 monthly total precipitation 88
Figure 106. July 2016 monthly total precipitation 88
Figure 107. August 2016 monthly total precipitation 89
Figure 108. September 2016 monthly total precipitation 89
Figure 109. October 2016 monthly total precipitation 90
Figure 110. November 2016 monthly total precipitation 90
Figure 111. December 2016 monthly total precipitation 91
IX
-------
List of Acronyms
AMET
Atmospheric Model Evaluation Tool
ARW
Advanced Research WRF
ASOS
Automated Surface Observing System
CFR
Code of Federal Regulations
COOP
Cooperative Observer Program (NWS)
EPA
Environmental Protection Agency
ESRL
Earth System Research Laboratory
GFS
Global Forecast System
MADIS
Meteorological Assimilation Data Ingest System
MM5
Fifth-generation Penn State/NCAR Mesoscale Model
MPE
Model Performance Evaluation
NCAR
National Center for Atmospheric Research
NCEP
National Centers for Environmental Prediction
NLCD
National Land Cover Database
NO A A
National Oceanic and Atmospheric Administration
NWS
National Weather Service
PBL
Planetary Boundary Layer
RAWS
Remote Automatic Weather Stations
RMSE
Root Mean Square Error
SNOTEL
Snow Telemetry Station
SST
Sea Surface Temperature
WRAP
Western Regional Air Partnership
WRF
Weather Research and Forecasting (Model)
X
-------
Table of Contents
Executive Summary iv
List of Tables vi
List of Figures vii
List of Acronyms x
Table of Contents xi
1. Introduction 1
2. WRF modeling configuration 1
2.1 Domain configuration 1
2.2 Inputs 3
2.3 WRF options and parameterization schemes 4
3. Model performance evaluation methodology 4
3.1 Surface meteorological parameters and performance criteria 5
3.1.1 Performance benchmarks 6
3.2 Selected sub-regions for analysis 7
3.3 Upper-air evaluation 11
3.4 Precipitation 12
4. Model performance evaluation results 13
4.1 Domain-wide surface parameters 13
4.1.1 Temperature 13
4.1.2 Wind speed 16
4.1.3 Wind direction 19
4.1.4 Humidity 22
4.2 Cook Inlet region performance 25
4.2.1 Temperature 25
4.2.2 Wind speed 27
4.2.3 Wind direction 29
4.2.4 Humidity 30
4.3 Fairbanks region performance 32
xi
-------
4.3.1 T emperature 32
4.3.2 Wind speed 34
4.3.3 Wind direction 36
4.3.4 Humidity 37
4.4 North slope region performance 39
4.4.1 T emperature 39
4.4.2 Wind speed 41
4.4.3 Wind direction 43
4.4.4 Humidity 44
4.5 Juneau region performance 46
4.5.1 T emperature 46
4.5.2 Wind speed 48
4.5.3 Wind direction 50
4.5.4 Humidity 51
4.6 Alaska Peninsula region performance 53
4.6.1 T emperature 53
4.6.2 Wind speed 55
4.6.3 Wind direction 57
4.6.4 Humidity 58
4.7 Upper-air analysis 60
4.8 Precipitation 85
5. Conclusions 91
References 93
xii
-------
1. Introduction
The United States Environmental Protection Agency (EPA) developed a one-year (2016)
meteorological dataset for the state of Alaska, using the Weather Research and Forecasting
model (WRF). This report provides a summary of the modeling methodology used to develop the
dataset and provides a model performance evaluation (MPE) of the dataset. Like EPA's annual
12-km United States CONUS WRF models, the purpose of this dataset is to provide a resource to
support air pollution and atmospheric photochemical modeling applications for regulatory and
planning purposes.
The WRF model represents the state-of-the-art tool for modern mesoscale numerical
meteorological simulation, designed for research and operational forecasting purposes (NCAR,
2019). WRF has largely displaced its predecessor, the MM5 model, which was widely used to
provide meteorological datasets for regulatory air quality analyses in the past. WRF is now used
widely by universities and government agencies, including the National Weather Service (NWS),
to provide simulations and downscaling of past weather fields and forecasts of future conditions.
The 2016 Alaska WRF dataset contains gridded hourly meteorological fields across two
domains, at 27-km and 9-km horizontal resolutions. The modeling was conducted using
meteorological inputs from a global reanalysis dataset, applying parameterization and physics
schemes selected based on a survey of other recent WRF modeling efforts focused on Alaska and
other arctic domains.
The MPE contained in this report evaluates the performance of the 9-km domain modeling
results compared to surface and upper-air measurements of wind, humidity, and temperature.
Regional precipitation estimates are also evaluated qualitatively by comparison to measurement-
based regional precipitation maps.
2. WRF modeling configuration
WRF Version 3.9.1.1 of the Advanced Research WRF (ARW) core (Skamarock, 2008) was used
for this work. The WRF-ARW is developed and maintained by the National Center for
Atmospheric Research's (NCAR) Mesoscale and Microscale Meteorology Lab (NCAR, 2019).
The modeling was conducted on the EPA High Performance Computing Center's scientific
cluster located at Research Triangle Park in North Carolina. The 2016 Alaska dataset is
maintained and archived by the EPA and may be obtained upon request (contact information
provided in Executive Summary).
2.1 Domain configuration
Modeling was conducted using a set of nested grids. The inner grid, at a horizontal resolution of
9 km, is nested within an outer domain at 27 km horizontal grid spacing. The grids are shown in
Figure 1. The modeling domains were defined with a goal of capturing all parts of the state of
1
-------
Alaska within the inner 9-km domain. The sizing and location of the domains also were
configured to fit well within the global GEOS-CHEM grid domain, to account for derivative
photochemical modeling purposes. The grids were defined using a Lambert Conformal Conic
(LCC) projection centered at 155° W, 63° N and true latitudes at 60° and 70°. Table 1 provides a
summary of the configuration of the two domains.
Figure 1, Alaska 27-km (left) and 9-km (right) horizontal resolution WRF modeling domains.
Table 1. WRF domain configurations.
Grid
Resolution
Nx
Ny
Outer
27 km
156
139
Inner
9 km
325
265
The domain was configured using 36 vertical layers with a first (surface) layer approximately 20
meters deep. Vertical resolution was greatest near the surface to resolve boundary layer
processes. The vertical domain configuration is outlined in Table 2 by sigma level and
approximate height and pressure coordinates.
Table 2. WRF vertical domain configuration.
Layer
Sigma
Level
Approximate
Pressure
(mb)
Approximate
Height (mb)
Approximate
Layer
Thickness (m)
36
0.0000
50.00
19313
3423
35
0.0500
98.15
15890
2243
34
0.1000
146.30
13648
1706
2
-------
Layer
Sigma
Level
Approximate
Pressure
(mb)
Approximate
Height (mb)
Approximate
Layer
Thickness (m)
33
0.1500
194.45
11942
1392
32
0.2000
242.60
10551
1183
31
0.2500
290.75
9367
1034
30
0.3000
338.90
8333
921
29
0.3500
387.05
7412
832
28
0.4000
435.20
6580
761
27
0.4500
483.35
5820
702
26
0.5000
531.50
5117
652
25
0.5500
579.65
4465
610
24
0.6000
627.80
3856
573
23
0.6500
675.95
3283
541
22
0.7000
724.10
2742
412
21
0.7400
762.62
2330
298
20
0.7700
791.51
2032
289
19
0.8000
820.40
1742
188
18
0.8200
839.66
1554
185
17
0.8400
858.92
1369
182
16
0.8600
878.18
1188
178
15
0.8800
897.44
1010
175
14
0.9000
916.70
834
87
13
0.9100
926.33
748
86
12
0.9200
935.96
662
85
11
0.9300
945.59
577
84
10
0.9400
955.22
492
84
9
0.9500
964.85
409
83
8
0.9600
974.48
325
83
7
0.9700
984.11
243
82
6
0.9800
993.74
162
41
5
0.9850
998.56
121
40
4
0.9900
1003.37
80
40
3
0.9950
1008.19
40
20
2
0.9975
1010.59
20
20
1
1.0000
1013.00
0.0
--
2.2 Inputs
WRF modeling requires inputs from databases to define the static and dynamic features of the
land and water interfaces. Also, WRF requires inputs from a global-scale meteorological model
or reanalysis dataset to provide the initial and boundary conditions of the atmosphere.
Land use and vegetation information were obtained from the recent National Land Cover
Database (NLCD) provided with WRF. Topographic information for WRF was developed using
the standard 30 arc-second (-900 m) resolution WRF terrain database.
The WRF model was initialized using the 0.25-degree National Centers for Environmental
Prediction (NCEP) Global Forecast System (GFS) analysis and 3-hour forecast from the 00Z,
3
-------
06Z, 12Z, and 18Z simulations. This dataset was also used to provide boundary conditions and
analysis nudging throughout the model runs. Analysis nudging was applied aloft only; Planetary
Boundary Layer (PBL) nudging of winds, temperature, and humidity was turned off. No
observational nudging was used in the modeling. Sea surface temperatures (SST) and snow cover
were obtained from the GFS analysis also.
2.3 WRF options and parameterization schemes
WRF contains a suite of state-of-the-art atmospheric physics parameterization schemes. The
options and models used in the Alaska WRF modeling are listed in Table 3. A local-closure PBL
scheme was selected to more accurately simulate the PBL in highly stable conditions, based on
the understanding non-local schemes may result in excessive mixing in highly stable conditions
near the surface.
Table 3. Physics parameterization schemes used in the 2016 Alaska WRF model.
Physics Scheme
WRF scheme
variable name
Model Selected
Description
Land Surface Model
sf_surface_physics
NOAH
Surface flux parameterization scheme
based on a four-layer soil temperature and
moisture model. Accounts for fractional
snow cover and frozen soil physics.
PBL parameterization
bl_pbl_physics
MYNN PBL model
Local 2.5-order closure K-theory based
model. Model set to compute each time
step.
Atmospheric surface
layer parameterization
sf_sfclay_physics
MYNN surface
layer model
Local 2.5-order closure surface layer model.
Cumulus/convective
physics
cu_physics
Kain-Fritsch
Deep and shallow convection sub-grid
scheme that applies a mass-flux approach
for downdrafts. Moisture-advection
modulation function used for this modeling.
Cloud/precipitation
microphysics
mp_physics
Morrison double-
moment scheme
Double-moment ice, snow, rain, and
graupel model.
Longwave/shortwave
radiation
ra_lw_physics,
ra_sw_physics
RRTMG scheme
Rapid Radiative Transfer Model, including
the MCICA method for random cloud
overlap. Radiation physics calculation
frequency set to 20 minutes.
3. Model performance evaluation methodology
The purpose of this evaluation is to determine whether the simulated meteorological outputs
sufficiently represent a reasonable approximation of actual meteorological conditions that
occurred over Alaska in 2016. The evaluation is conducted using both quantitative and
qualitative analyses using archived surface and upper-air meteorological measurements. Since
4
-------
the intended purpose of this WRF dataset is for use in air quality models, performance in the
PBL and at the surface is of particular concern. The quantitative assessment focuses on WRF
performance at specific locations where wind, temperature, and humidity were measured at
surface and upper-air radiosonde stations. A qualitative evaluation of precipitation across Alaska
is also provided in this report.
Quantitative analysis is conducted using common statistical measures such as mean prediction
error, root mean square error (RMSE), and mean prediction bias. The equations for these
measures are given below where P is the predicted variable at a site and 0 is the observed
variable at the site, determined over an incremental timeframe:
1 N
Mean Bias = — ^(P£ — 0£)
i=1
N
RMSE= \^(Pi~°i)2
4 i=i
i N
Mean Error = — ^JP£ — 0£|
i=1
The predicted values obtained from the WRF dataset are extracted from the grid points located
nearest to the locations of the meteorological stations. However, the WRF grid point selected for
any given station may not be representative of the meteorological conditions at that station,
especially in complex terrain. For example, the meteorological station could be located in a
valley, whereas the grid point could be located on an adjacent mountain peak. Given the 9 km
horizontal resolution of the WRF dataset, the nearest grid point to any given station may be some
distance from that station. Also, the WRF grid point is meant to represent the average conditions
within the 9-km wide grid cell, rather than the conditions at a single point within that cell. These
factors may contribute to an inherent degree of bias and error that is not necessarily an indication
of poor model performance.
3.1 Surface meteorological parameters and performance criteria
WRF model performance was assessed by comparing modeled surface-layer meteorological
parameters to measured values obtained from the Earth System Research Laboratory (ESRL)
Meteorological Assimilation Data Ingest System (MADIS). The MADIS surface hourly dataset
consists of hourly-averaged meteorological values collected mainly at airport ASOS and other
official meteorological stations operated by government agencies. The assessment compared
modeled hourly-averaged values of 10-m wind speed and wind direction and 2-m temperature
and absolute humidity (in units of grams of water vapor per kilogram of dry air) to measured
5
-------
values. In total, 216 surface meteorological stations from the MADIS database were located
within the 9 km domain and selected for the analysis. The selection included three stations in
Russia and about 40 stations in Canada. Not every meteorological dataset contained enough data
for analysis over all the periods examined.
3.1.1 Performance benchmarks
Several sets of benchmarks reported in the scientific literature have been developed to evaluate
the performance of meteorological model datasets used for air quality modeling applications.
EPA's Atmospheric Model Evaluation Tool (AMET) (EPA, 2019) uses a set of statistical
performance goals in its "soccer plot" tool, based on common benchmarks for normalized bias
and error (Appel, et al., 2011). Several sets of benchmarks, namely Emery et al. (2001) and
Kemball-Cook et al. (2005), have been widely used in recent meteorological performance
evaluations (Bowden, et al., 2015) (Bowden, et al., 2016) (Brashers, et al., 2015) (Brown, 2014)
(Ramboll-Environ, 2015) and were adopted for use in this evaluation.
Emery et al. (2001) developed performance benchmarks for meteorological inputs to
photochemical models drawing on the (Tesche, et al., 2001) evaluation of statistics from over 30
regional modeling datasets within the continental United States. Emery et al. selected a set of
error and bias thresholds based on the 80th percentile performance values for 29 of the datasets.
The majority of the datasets were developed at a horizontal resolution of 12 km using the MM5
or RAMS meteorological models.
Kemball-Cook et al. developed a model performance analysis of the Western Regional Air
Partnership (WRAP) 2002 Alaska MM5 model dataset. It was noted, in their report, that the
Emery et al. performance benchmarks were excessively stringent for model performance in
regions of complex terrain or regions comprised of highly heterogeneous microclimates, such as
most of Alaska. Kemball-Cook et al. adopted a less stringent set of benchmarks based on the
previous performance of the 2002 WRAP Rocky Mountains and Sierra Nevada MM5 datasets.
These benchmarks have been adopted for "complex conditions," as opposed to the "simple
conditions" benchmarks of Emery et al.
The benchmarks used in this study are listed in Table 4. For complex conditions, Kemball-Cook
et al. did not provide a benchmark for wind direction bias. Given the benchmark for wind
direction error is roughly twice the value for complex conditions as the value for simple
conditions, the EPA adopted a wind direction complex-conditions bias benchmark double the
value for simple conditions.
Note, the EPA does not recommend using these benchmarks as a "pass/fail" indicator of dataset
performance (EPA, 2018). The benchmarks are intended to be used to assess the general
confidence in the representativeness of the model outputs. The benchmarks have been developed
considering average bias and error over wide regions that include a number of surface stations.
Therefore, the benchmarks are most useful for assessing performance on a regional basis. They
can be used to evaluate performance on a single-station basis, but the modeler must use more
caution in assuming the criteria are applicable to single-station performance.
6
-------
Table 4. Surface meteorology performance benchmarks.
Simple Conditions benchmarks
Complex Conditions benchmarks
based on Emery et al. (2001)
based on Kemball-Cook et al. (2005)
Bias
Error3
Bias
Error3
Wind speed
0.5 ms1
2.0 ms1
1.5 ms1
2.5 ms1
Wind direction
10°
30°
20°b
55°
Temperature
0
LO
o
2.0° K
2.0° K
3.5° K
Absolute Humidity
1 g kg1
2 g kg1
1 g kg1
2 g kg 1
a Wind speed benchmarks are based on RMSE, while others are mean absolute error.
b Kemball-Cook et al. does not provide a recommendation for wind direction bias. A value of 20° was assumed for this study, which is twice the
simple conditions benchmark.
3.1.2 Qualitative analysis
In addition to quantitative evaluation of WRF performance using measures of error and bias,
qualitative evaluation tools such as wind roses and time series of meteorological variables were
developed for this assessment. Hour-of-day time series of temperature, wind speed, and humidity
were developed on a seasonal basis for both the measured and modeled parameters at each
meteorological station location.
3.2 Selected sub-regions for analysis
The State of Alaska is a large landmass roughly 20% the size of the contiguous United States,
that encompasses a significant range of climates. Five subdomains were selected to facilitate the
performance evaluation based on the division of climates and regions of strategic importance
with regards to air quality regulation. Subregion performance is judged by comparison of a
subdomain-wide average error and bias on a monthly basis against the benchmarks identified in
Section 3.1.
The first domain, referred to as the "North Slope" domain, encompasses surface weather stations
located at 12 sites along the Arctic coast of Alaska, spanning from Point Hope (PAPO) to Barter
Island (PABA). Included in the domain are Alaskan village sites such as Nuiqsut (PATQ) and
Utqiagvik (PABR). The North Slope domain is of particular importance from an air quality
perspective due to the large number of existing and planned oil and gas facilities at Prudhoe Bay
and locations within the National Petroleum Preserve.
A plot of the domain, including positions of the meteorological stations, is shown in Figure 2.
7
-------
< 2 < <
n,
Figure 2. North Slope domain.
The second domain selected for regional analysis is the "Fairbanks" domain, which encompasses
five surface meteorological stations in the vicinity of the city of Fairbanks. This area is of
particular interest from an air quality perspective due to persistently high PM2.5 concentrations in
the Fairbanks area. In 2009, the EPA designated parts of the Fairbanks North Star Borough as
nonattainment for the 2006 24-hour PM2.5 National Ambient Air Quality Standards, and in 2017,
the EPA reclassified the area from a "moderate to a "serious" nonattainment area. The domain
contains the Fairbanks International airport (PAFA) and Nenana airport (PANN) airport
meteorological stations, as well as Eielson Airforce Base (PAEI), Allen Army Airfield (PABI),
and Wainwright AAF airport (PAFB) stations.
A plot of the Fairbanks domain and selected meteorological stations is shown in Figure 3.
¥af¥afb
Vaei
Vann
^ABI
Figure 3. Fairbanks domain.
8
-------
The third sub-region selected for analysis is the "Cook Inlet" domain. This domain consists of 12
meteorological stations located within the Anchorage metropolitan area and along the coast of
the Cook Inlet. This area of Alaska is of particular concern because most of the state's
population resides in the Anchorage metropolitan area and because there is a significant
concentration of industry across the Cook Inlet area. Also, offshore oil and gas facilities are
located in the Cook Inlet. Development of additional offshore facilities within the Cook Inlet are
possible in the future. The domain is shown in Figure 4.
The fourth subregion selected for the analysis is the "Juneau" domain. Although this area is
currently not a significant concern from an air quality perspective, the area encompasses the state
capitol and a significant portion of the state population. The region is also subject to the shipping
emission impacts from heavy industrial and cruise ship traffic. The region contains highly
complex terrain and a variety of microclimates. The domain is centered on the Juneau
International airport station (PAJN) and contains stations as far south as Wrangell airport
(PAQG) and far north as Haines (PAHN). The domain is shown in Figure 5.
%Nsx
HMSA2
Figure 4. Cook Inlet domain.
9
-------
<5i
IRNA2
'AGN
'AS I
'AFE
'APG
'AWG
Figure 5. Juneau domain.
The fifth subregion selected for the analysis is the "Alaska Peninsula" domain. This domain
encompasses meteorological stations located on Kodiak Island, along the Alaska Peninsula, and
on several of the Aleutian Islands. This domain also contains several areas of concentrated
industrial activity including Dutch Harbor and Kodiak. The domain and selected meteorological
stations are plotted in Figure 6.
10
-------
~apn
~adq
~aph
Vakh
^AOU
¥ajc
~acd
Vakf
¥asd
V7 VAUT
\PADU
Figure 6. "Alaska Peninsula" domain.
3.3 Upper-air evaluation
A set of 16 upper-air meteorological station datasets were obtained for the model performance
evaluation. The set consists of the 13 stations operated by the National Weather Service in
Alaska as well as three stations located in Canada. Upper-air stations deploy radiosonde
instruments on weather balloons, released twice daily just prior to hours 0 and 12 UTC, each
day. The radiosondes measure wind, temperature, and humidity through the atmospheric column.
Profiles of hourly-averaged wind, temperature, and humidity were obtained from the VVRF grid
cell nearest to the location of each upper-air station at times corresponding with the radiosonde
measurements. The station identifiers and locations are plotted in Figure 7.
Performance was assessed using a set of boxplots to describe the distribution of residuals of wind
speed, wind direction, temperature, and humidity for each of the four seasons. Seasons were
defined as winter (December, January, February), spring (March, April, May), summer (June,
July, August), and autumn (September, October, November). The residuals (the difference
between the modeled and measured values) were calculated for pressure levels 1000, 925, 850,
700, 500, 400, and 300 mb.
11
-------
A
Figure 7. Upper-air stations selected for the model performance study.
3.4 Precipitation
A qualitative evaluation of precipitation was conducted using monthly-averaged precipitation
maps for the state of Alaska. Monthly total precipitation maps for the state of Alaska were
obtained from the National Centers for Environmental Prediction (NCEP) Alaska climate
monitoring database (NOAA-NCEP, 2019). These maps were developed using the gridded 5-km
NOAA "nClimGrid" dataset, developed from the Global Historical Climatology Network
(GHCN). The GHCN dataset is a daily 5-km resolution grid of meteorological variables,
including precipitation, determined using measured surface data. Climatologically aided spatial
interpolation is used to assign daily average values to each grid point from the available
measurements (Vose, et a!.. 2014). Precipitation datasets used to develop the grids are obtained
from the COOP, ASOS, RAWS, and SNOTEL networks.
Monthly-averaged precipitation maps provided by the NOAA-NCEP tool are compared to plots
12
-------
of state-wide monthly average precipitation values determined from the 2016 Alaska WRF
model (in inches of liquid-equivalent precipitation).
4. Model performance evaluation results
Bias and error were determined for each hourly-average wind speed, wind direction,
temperature, and humidity record available by comparison of WRF outputs to measurements.
The average bias and error results were evaluated on a domain-wide scale (bias and error
averaged from all stations in the 9-km domain) and also on a regional scale in this section (bias
and error averaged from meteorological stations located in each subdomain). The plots for a
selection of individual stations of interest are also analyzed in this section. The plots of all other
individual surface and upper-air stations that are not reviewed in the body of this report are
available in electronic format from the EPA upon request (Appendix A).
4.1 Domain-wide surface parameters
Domain-wide performance is evaluated using error/bias "soccer plots" that compare monthly
averaged error and bias against the adopted benchmarks presented in Section 3.1.1. This
evaluation is used to assess the likelihood of systematic error driven by selection of
parameterization schemes or other factors that could impact domain-wide performance.
4.1.1 Temperature
The domain-wide soccer plot for surface temperature is shown in Figure 8. The plot
demonstrates the domain-wide surface temperature tends to be biased cold, particularly in winter
and spring. However, the bias is within the criteria for complex conditions, except in March,
which is slightly outside of the bounds. Temperature error is also within the complex criteria,
except for December, which slightly exceeds the benchmark.
Domain-wide temperature bias per month is illustrated in Figure 9 through Figure 14 through a
plot of all stations in the domain shaded by the magnitude of bias. The regional bias maps in
these figures demonstrate the domain-wide average cold bias, evident in the soccer plot, is
mainly driven by bias along the North Slope and coastal areas of the Seward Peninsula during
winter and spring months.
Surface temperatures are biased warm in the early winter months in the eastern Alaska interior
and Yukon interior. These stations are generally located within steep valleys in mountainous
areas. Further examination of precipitation and snow cover would be needed to investigate the
bias, but it is assumed the bias is due to incorrect parameterization of the surface energy bias
possibly due to error in snow cover.
The temperature bias over summer months is very low, and within the bounds of simple
condition benchmarks.
13
-------
AK Region: Temperature (K) Performance
g 4-
a
L.
3
4*
(0
k
® .
a j
E
o
H
of
§ 2 -
cc
Hi
1 -
Complex Conditions
m Simple Conditions
-3
-2
Month (Completeness)
X Jan (97.8%)
Feb (97.8%)
Mar (98.1%)
Apr (98.5%)
May (98.8%)
Jun (98.6%)
Jul (97.1%)
Aug (97.2%)
Sep (97.2%)
Oct (97.2%)
Nov (95.6%)
Dec (95.9%)
-10 12 3
BIAS: Temperature (K)
Figure 8. Soccer plot of monthly 2-m temperature error and bias averaged over the 9 km Alaska domain for 2016.
I
Feb. 2016 Temperature Mean Bias
° O
tog
3 £
09 °
o o °
_ o0 o
°o°o
o o
o o °
oo o n •
o j3.Co, eo o Q o
• o
Figure 9. Monthly mean temperature bias, January (left) and February (right) 2016.
Jan. 2016 Temperature Mean Bias
14
-------
Mar. 2016 Temperature Mean Bias
t
Apr. 2016 Temperature Mean Bias
ODOO
O ©
o
\
~ . * • *•
^ ° °x»•
*o c. o
Figure 10. Monthly mean temperature bias, March (left) and April (right) 2016.
May. 2016 Temperature Mean Bias
J <3
°° 'C35G
f
°0 P
Jun. 2016 Temperature Mean Bias
i
%°° o
°0
0°° _ c53)
°o 9
I
Figure 11. Montlily mean temperature bias. May (left) and June (right) 2016.
Jul. 2016 Temperature Mean Bias
0°3 _ C®.
Aug. 2016 Temperature Mean Bias
o
Ooo ^ c®
Oo 8
«¦
Figure 12. Montlily mean temperature bias, July (left) and August (right) 2016.
15
-------
Sep. 2016 Temperature Mean Bias
' V\ ^
1. ' ,,o
w
%
o§
OBOO r
^—TZ
is>-
...JS0
o o
Qt» O o
°®ofo °°o ° 0 Q°
00 o o o O o
o °o o o °
o °° *• °
\ ^ A
- ©
jt K
Oct. 2016 Temperature Mean Bias
Q» o o o
o o
U&°g>o °oo »0o^
> _ ° o ° o
r)<^- °o o. q o
"a o o q
V? *0 ® o
I-
Figure 13. Monthly mean temperature bias, September (left) and October (right) 2016.
Nov. 2016 Temperature Mean Bias
O o °
I
Dec. 2016 Temperature Mean Bias
Is
' 0. i&.
0 S!
Figure 14. Montlily mean temperature bias, November (left) and December (right) 2016.
4.1.2 Wind speed
The domain-wide soccer plot for surface wind speed is shown in Figure 15. The plot
demonstrates the domain-wide surface wind speed tends to be just slightly biased low, but all
within the criteria for simple conditions. However, wind speed error tends to slightly exceed
complex criteria for the winter months. The error and bias appear to be driven mainly by a low
wind speed bias along the coastal regions of the North Slope and Seward peninsula during winter
months, as shown in Figure 16 through Figure 21.
16
-------
AK Region: Wind Speed (m/s) Performance
T>
0)
u
a.
tn
¦D
C
4.0 -
3.5 ¦
3.0 -
2.5 -
2.0-
1.5 -
1.0 ¦
0.5 -
0.0 ¦
Complex Conditions
Simple Conditions
-2
-10 1
BIAS: Wind Speed (m/s)
Month (Completeness)
X Jan (79,4%)
Feb (79.7%)
Mar (81.3%)
Apr (84.2%)
May (83.4%)
Jun (81.9%)
Jul (81.1%)
Aug (81.5%)
Sep (81.1%)
Oct (79.9%)
Nov (76.1%)
Dec (77.0%)
Figure 15. Soccer plot of monthly 10-m wind speed error and bias averaged over the 9 km Alaska domain for 2016.
Jan. 2016 Wind Speed Mean Bias
Feb. 2016 Wind Speed Mean Bias
Figure 16. Monthly mean wind speed bias, January (left) and February (right) 2016.
17
-------
Mar. 2016 Wind Speed Mean Bias
&
° o
o
^ °° ° O _. H
°
rk
a6P o °
•a ,o„ o
1
I
Apr. 2016 Wind Speed Mean Bias
0 1
Figure 17. Monthly mean wind speed bias, March (left) and April (right) 2016.
May. 2016 Wind Speed Mean Bias
o<»?c2>„ °o
O o °
% e°'i°
^ ®
09 °6A0 °
Oq 0_ o
Aug. 2016 Wind Speed Mean Bias
Figure 19. Monthly mean wind speed bias, July (left) and August (right) 2016.
18
-------
Oct. 2016 Wind Speed Mean Bias
Figure 20. Monthly mean wind speed bias, September (left) and October (right) 2016
Nov. 2016 Wind Speed Mean Bias
Dec. 2016 Wind Speed Mean Bias
Figure 21. Monthly mean wind speed bias, November (left) and December (right) 2016.
4.1.3 Wind direction
The domain-wide soccer plot for surface wind direction is shown in Figure 22. Wind direction
bias is very low on average. Wind direction error exceeds simple criteria but falls within the
complex benchmark criteria. The distribution of wind direction error per month is shown in
Figure 23 through Figure 28. Wind direction error is greatest in the inland mountainous regions
of eastern Alaska and the Yukon. High wind direction error in mountainous areas is not
necessarily an indication of poor WRF performance. Observations are typically representative of
mountain valleys where airports are located and where the wind climate is generally aligned with
the terrain. At 9 km resolution, WRF provides an average wind vector across an area that can
easily contain several mountain peaks and valleys in a single grid cell.
19
-------
AK Region: Wind Direction (deg) Performance
80
O)
-------
Mar. 2016 Wind Direction Mean Error
Apr. 2016 Wind Direction Mean Error
Figure 24. Monthly mean wind direction error, March (left) and April (right) 2016.
_-60
V
r 40
¦30 |
F
™-o
Jun. 2016 Wind Direction Mean Error
c
a oo n
Figure 25. Monthly mean wind direction error, May (left) and June (right) 2016.
Jul. 2016 Wind Direction Mean Error
•o o
I
I -2<
o
Aug, 2016 Wind Direction Mean Error
6
O o °
Figure 26. Monthly mean wind direction error, July (left) and August (right) 2016.
May. 2016 Wind Direction Mean Error
21
-------
Figure 27. Monthly mean wind direction error, September (left) and October (right) 2016.
Nov, 2016 Wind Direction Mean Error
CO °
Dec. 2016 Wind Direction Mean Error
oa a
f
Figure 28. Monthly mean wind direction error, November (left) and December (right) 2016.
4.1.4 Humidity
The domain-wide soccer plot for surface absolute humidity is shown in Figure 29. WRF
performance is within the benchmark criteria all months. There are no regions with significant
absolute humidity bias through any season, as seen in Figure 30 through Figure 35.
22
-------
AK Region: Mixing Ratio (g/kg) Performance
Simple and Complex Conditions
Month (Completeness)
X Jan (73.9%)
Feb (74.9%)
Mar (75.1%)
Apr (75.7%)
May (75.5%)
Jun (75.1%)
Jul (75.2%)
Aug (75.6%)
Sep (75.8%)
Oct (75.6%)
Nov (74.8%)
Dec (74.2%)
-0.5 0.0 0.5
BIAS: Mixing Ratio (g/kg)
Figure 29. Soccer plot of monthly absolute humidity error and bias averaged over the 9 lan Alaska domain for 2016.
Jan. 2016 Humidity Mean Bias
0°° " °o o O
O o °
a °o q.q °
Feb. 2016 Humidity Mean Bias
°°= ° 5»o °o o
Figure 30. Monthly mean humidity bias, January (left) and February (right) 2016.
23
-------
nsr-
Mar. 2016 Humidity Mean Bias
<£
cr^^OBoo,,
!, o °
P O o
° °
i aocY o
o ° a© 0
v W|
*TT^r
o
r o
*'o &
Apr. 2016 Humidity Mean Bias
s'h
Figure 31. Monthly mean humidity bias, March (left) and April (right) 2016.
May. 2016 Humidity Mean Bias
Jun. 2016 Humidity Mean Bias
Figure 32. Monthly mean humidity bias. May (left) and June (right) 2016.
Aug. 2016 Humidity Mean Bias
Figure 33. Monthly mean humidity bias, July (left) and August (right) 2016.
Jul. 2016 Humidity Mean Bias
24
-------
Figure 34. Monthly mean humidity bias, September (left) and October (right) 2016.
I
-0 if
I
Figure 35. Monthly mean humidity bias. November (left) and December (right) 2016.
4.2 Cook Inlet region performance
Cook Inlet regional performance was assessed using soccer plots to analyze average error and
bias across all stations in the subdomain. Also, Anchorage International airport (PANC) was
selected as the individual station of interest for the analysi s because it is located in the largest
metropolitan area in the region. Temperature, wind, and humidity performance is examined over
the subregion and at the individual station of interest.
4.2.1 Temperature
A plot of temperature distribution at PANC, examined by hour per season is shown in Figure 36.
The distributions match well except VVRF is biased low during summer nighttime and early
morning hours. This could be partially explained by the local urban heat island effect at the
monitor, which would be muted across a 9 km grid cell.
Nov. 2016 Humidity Mean Bias
Dec. 2016 Humidity Mean Bias
25
-------
Winter (99.6% Complete)
250
~~~~
295
290
285
ro 280
275
270
265
Spring (98.6% Complete)
-~*
ill if11 1
OBS
¦¦
WRF
IN
~ ~
~ ~ ~
~ ~ ~ ~ ~
1 2 3 4 5 6 7 8 9 101112131415161718192021222324
Hour of Day
Summer (100.0% Complete)
297.5
295.0
292.5
5 290.0
285.0
282.5
280.0
12 3 4 5 6 7 8 9 101112131415161718192021222324
Hour of Day
Autumn (100.0% Complete)
iii Ul i Miiiiiiii VI
~~~
~ ~
~ ~~
1 2 3 4 5 6 7 8 9 101112131415161718192021222324
Hour of Day
12 3 4 5 6 7 8 9 101112131415161718192021222324
Hour of Day
Figure 36. PANC (Anchorage) seasonal, hour-of-day temperature distributions, ASOS (blue) vs. Alaska 9-kin WRF
(red).
The regional average temperature error and bias soccer plot is shown in Figure 37. All results
are within the complex conditions benchmark criteria. Winter-time temperatures are biased high,
but still within the criteria.
26
-------
Cook Inlet Region: Temperature (K) Performance
4 -
Complex Conditions
0
o.
E
a
l-
ai
o
ec
ce
LU
2 -
1 -
Simple Conditions
~
.
J
-3
-2
-10 1
BIAS: Temperature (K)
Month (Completeness)
X Jan (99.5%)
Feb (97.7%)
Mar (99.9%)
Apr (99.0%)
May (99.8%)
Jun (96.1%)
Jul (90.5%)
Aug (92.8%)
Sep (93.4%)
Oct (99.0%)
Nov (97.4%)
Dec (98.6%)
Figure 37. Soccer plot of monthly 2-m temperature error and bias averaged over the Cook Inlet subdomain for
2016.
4.2.2 Wind speed
Wind rose plots developed from both the observed (PANC) and modeled datasets are included in
Figure 38. WRF simulates the modes of predominant wind well at PANC but appears to
overpredict wind speed on average.
A plot of wind speed distribution at PANC, examined by hour per season is shown in Figure 39.
WRF consistently overpredicts wind speed all hours of the day every season. The overprediction
could be a result of the local surface roughness in the area of the PANC ASOS station. The
ASOS station is likely subject to higher roughness than what is "seen" by the 9-km wide WRF
grid cell. The grid cell encompasses areas over the water west and south of Anchorage and
therefore represents a region of lower surface roughness than what is representative in the
immediate vicinity of PANC.
The regional average wind speed error and bias soccer plot is plotted in Figure 40. All results are
within the complex conditions criteria. Wind speed on average is unbiased and winter-time error
is higher than error during the other seasons.
27
-------
Figure 38. Wind rose comparison. PANC (Anchorage), ASOS observed (left), Alaska 2016 WRF (right).
Wind speed (m/s)
¦¦ 0.0 to 2.0 m/s
2.0 to 4.0 m/s
r 1 4.0 to 6.0 m/s
IZZJ 6.0 to 8.0 m/s
C~1 B.0 to 10.0 m/s
[=1 10.0 to 12.0 m/s
Ml 12.0 m/s +
Wind speed (m/s)
H 6.0 to 2.0 m/s
' 2.0 to 4.0 m/s
=3 4.0 to 6.0 m/s
Z3 6.0 to 8.0 m/s
=~ 8.0 to 10.0 m/s
^ 10.0 to 12.0 m/s
B 12.0 m/s +
17.5
15.0
12.5
10.0
¦ 7.5
5.0
2.5
0.0
Winter (74.5% Complete)
~ ~
~
~ ~ ~
t,; ' «
: h:ijit! •''•V2
* ~Tilt t IttI v * * ~ll'Vt* It " 7
¦ Mil iiiiUHi< ¦
Spring (75.3% Complete)
L
OBS
WRF
1 2 3 4 5 6 7 8 9 101112131415161718192021222324
Hour of Day
Summer (78.4% Complete)
1 2 3 4 5 6 7 8 9 101112131415161718192021222324
Hour of Day
Autumn (76.9% Complete)
1
1 2 3 4 5 6 7 8 9 101112131415161718192021222324
Hour of Day
1 2 3 4 5 6 7 8 9 101112131415161718192021222324
Hour of Day
Figure 39. PANC (Anchorage) seasonal, hour-of-day wind speed distributions, ASOS (blue) vs. Alaska 9-km WRF
(red).
28
-------
Cook Inlet Region: Wind Speed (m/s) Performance
Complex Conditions
tions
Month (Completeness)
X Jan (69.5%)
Feb (74.5%)
Mar (67.9%)
Apr (74.7%)
May (75.8%)
Jun (75.7%)
Jul (70.0%)
Aug (68.6%)
Sep (65.8%)
Oct (60.9%)
Nov (67.8%)
Dec (64.7%)
-10 1
BIAS: Wind Speed (m/s)
Figure 40. Soccer plot of monthly wind speed error and bias averaged over the 9 km Cook Inlet domain for 2016.
4.2.3 Wind direction
The wind direction monthly bias and error plot of the Cook Inlet subdomain is shown in Figure
41. The subdomain monthly average error exceeds the simple conditions criteria but generally
falls within the complex condition benchmarks. The subregion bias is higher during the winter
and spring months, with only April slightly exceeding the criteria.
29
-------
Cook Inlet Region: Wind Direction (deg) Performance
80 ¦
si
u
- 60-
Complex Conditions
______
K.UIII|/ieA \.UIIUILIUIIJ
X
~
•a
c
en
o
CC
CC
40 -
20-
S |mp]e_C o ndjtjo n s_
-30
-20
-10 0 10
BIAS: Wind Direction (deg)
20
30
Month (Completeness)
X Jan (69.5%)
Feb (74.5%)
Mar (67.9%)
Apr (74.7%)
May (75.8%)
Jun (75.7%)
Jul (70.0%)
Aug (68.6%)
Sep (65.8%)
Oct (60.9%)
Nov (67.8%)
Dec (64.7%)
Figure 41. Soccer plot of monthly wind direction error and bias averaged over the Cook Inlet domain for 2016.
4.2.4 Humidity
A plot of the PANC station distributions of absolute humidity by hour of day per season is
shown in Figure 42. The results demonstrate the WRF model tended to slightly overpredict
daytime humidity during winter months on average and underpredict morning humidity during
summer months.
The Cook Inlet subdomain humidity soccer plot is shown in Figure 43. The plot demonstrates
WRF humidity performance has low bias and error and within the criteria.
30
-------
Winter (99.6% Complete)
Spring (98.6% Complete)
~
OBS
WRF
1 2 3 4 5 6 7 8 9 101112131415161718192021222324
Hour of Day
Summer (99.9% Complete)
1 2 3 4 5 6 7 8 9 101112131415161718192021222324
Hour of Day
Autumn (100.0% Complete)
1 2 3 4 5 6 7 8 9 101112131415161718192021222324
Hour of Day
1 2 3 4 5 6 7 8 9 101112131415161718192021222324
Hour of Day
Figure 42. PANC (Anchorage) seasonal, hour-of-dav absolute humidity distributions, ASOS (blue) vs. Alaska 9-km
WRF (red).
31
-------
Cook Inlet Region: Mixing Ratio (g/kg) Performance
rc
ce
u)
c
OC
o
cc
cc
Simple and Complex Conditions
-1.5
-0.5 0.0 0.5
BIAS: Mixing Ratio (g/kg)
Month (Completeness)
X Jan (99.5%)
Feb (97.7%)
Mar (99.4%)
Apr (98.9%)
May (99.8%)
Jun (96.1%)
Jul (90.4%)
Aug (92.5%)
Sep (93.3%)
Oct (90.7%)
Nov (97.4%)
Dec (98.2%)
Figure 43. Soccer plot of monthly absolute humidity error and bias averaged over the Cook Inlet subdomain for
' 2016.
4.3 Fairbanks region performance
The Fairbanks subregion performance was assessed using soccer plots to analyze average error
and bias across all stations in the subdomain. Also, Fairbanks International airport (PAFA) was
selected as the individual station of interest for the analysis. Temperature, wind, and humidity
performance is examined in this section for the subregion and at the individual station of interest.
4.3.1 Temperature
A plot of temperature distribution at PAFA, examined by hour per season is shown in Figure 44.
The distributions match well in spring and autumn. The WRF model is biased warm all hours of
the day in winter on average. Also, WRF is biased slightly cold all hours of the day over the
summer.
The soccer plot of bias and error across the Fairbanks subregion is shown in Figure 45. The
regional bias matches that of PAFA, with WRF highly overpredicting temperature in winter
months and slightly underpredicting temperature in the spring and summer. WRF winter
temperature bias and error falls outside the complex conditions criteria.
32
-------
280
270
260
250
240
Winter (99.6% Complete)
~
Spring (98.6% Complete)
300
8. 270
E
OBS
WRF
1 2 3 4 5 6 7 8 9 101112131415161718192021222324
Hour of Day
Summer (100.0% Complete)
1 2 3 4 5 6 7 8 9 101112131415161718192021222324
Hour of Day
Autumn (100.0% Complete)
1 2 3 4 5 6 7 8 9 101112131415161718192021222324
Hour of Day
1 2 3 4 5 6 7 8 9 101112131415161718192021222324
Hour of Day
Figure 44. PAFA (Fairbanks) seasonal, hour-of-day temperature distributions, ASOS (blue) vs. Alaska 9-km WRF
(red).
33
-------
Fairbanks Region: Temperature (K) Performance
6-
*
-------
Figure 46. Wind rose comparison, PAFA (Fairbanks), ASOS observed (left), Alaska 2016 WRF (right).
Wind speed (m/s)
¦ 0.0 to 2.0 m/s
2.0 to 4.0 m/s
4,0 to 6.0 m/s
CZ1 6,0 to 8.0 m/s
[=] 8.0 to 10.0 m/s
[ZZI 10.0 to 12.0 m/s
E^l 12.0 m/s +
Wind speed (m/s)
0.0 to 2.0 m/s
¦¦ 2.0 to 4.0 m/s
Hi 4.0 to 6.0 m/s
i' i 6.0 to 8.0 m/s
!ZZI 8.0 to 10.0 m/s
IZZ1 10.0 to 12.0 m/s
¦¦ 12.0 m/s +
Winter (44.3% Complete) Spring (69.6% Complete)
1 2 3 4 5 6 7 8 9 101112131415161718192021222324 1 2 3 4 5 6 7 8 9 101112131415161718192021222324
Hour of Day Hour of Day
Summer (68.2% Complete) Autumn (56.2% Complete)
10
10
~ ~
~ ~ ~
~ ~ ~ *
~ ~ ~ ~
~ ~ ~
~ ~ ~ ~ H ~ 4 A ~
~ ~~~~~~~ ~~~
~~~~~~~ «« ~~~ . *~~ ~ ~ ~.\ 4 ~
~ ~~~~ ~
1
1 2 3 4 5 6 7 8 9 101112131415161718192021222324
Hour of Day
1 2 3 4 5 6 7 8 9 1011 12131415161718192021222324
Hour of Day
Figure 47. PAFA (Fairbanks) seasonal, hour-of-day wind speed distributions, ASOS (blue) vs. Alaska 9-kin WRF
(red).
35
-------
Fairbanks Region: Wind Speed (m/s) Performance
4.0 -
3.5
3.0 -
-a 2.5
t>
41
a
in
¦a 2.0
S 1.5 H
s
a.
1.0 -
0.5 -
0.0
Complex Conditions
%imple Conditions
~
Month (Completeness)
X Jan (57.2%)
Feb (53.9%)
Mar (60.3%)
Apr (73.7%)
May (74.1%)
Jun (66.1%)
Jul (66.1%)
Aug (68.8%)
Sep (66.0%)
Oct (58.0%)
Nov (41.2%)
Dec (41.7%)
-2-10 1 2
BIAS: Wind Speed (m/s)
Figure 48. Soccer plot of monthly wind speed error and bias averaged over the Fairbanks subdomain for 2016.
4.3.3 Wind direction
The Fairbanks subdomain wind direction monthly bias and error is plotted in Figure 49. Average
wind direction error falls within the complex conditions criteria and average bias generally falls
within the simple conditions criteria.
36
-------
Fairbanks Region: Wind Direction (deg) Performance
80 -
01
HI
- 60 H
c
o
¦o
.£ 40 H
ce
o
cc
DC
111
20 -
Complex Conditions
~ +
<
~ •
~
X ~ D#
x • ~
Y
Simple Conditions
-30
-20
20
30
Month (Completeness)
X Jan (57.2%)
Feb (53.9%)
Mar (60.3%)
Apr (73.7%)
May (74.1%)
Jun (66.1%)
Jul (66.1%)
Aug (68.8%)
Sep (66.0%)
Oct (58.0%)
Nov (41.2%)
Dec (41.7%)
-10 0 10
BIAS: Wind Direction (deg)
Figure 49. Soccer plot of monthly wind direction error and bias averaged over the Fairbanks subdomain for 2016.
4.3.4 Humidity
A plot of the PAFA station distributions of absolute humidity by hour of day per season is shown
in Figure 50. WRF tended to overpredict humidity at PAFA in the winter and midday in spring
and underpredict humidity in the early morning summer hours.
The Fairbanks subdomain humidity soccer plot is shown in Figure 51. Despite the bias shown at
PAFA in the winter, the average bias and error across the subregion falls within the simple and
complex criteria.
37
-------
Winter (98.6% Complete)
~
Spring (98.6% Complete)
~ ~ **********
OBS
WRF
1 2 3 4 5 6 7 8 9 101112131415161718192021222324
Hour of Day
Summer (100.0% Complete)
1 2 3 4 5 6 7 8 9 101112131415161718192021222324
Hour of Day
Autumn (99.7% Complete)
1 2 3 4 5 6 7 8 9 101112131415161718192021222324
Hour of Day
1 2 3 4 5 6 7 8 9 101112131415161718192021222324
Hour of Day
Figure 50. PAFA (Fairbanks) seasonal, hour-of-day absolute humidity distributions, ASOS (blue) vs. Alaska 9-km
WRF (red).
38
-------
Fairbanks Region: Mixing Ratio (g/kg) Performance
Simple and Complex Conditions
Month (Completeness)
X Jan (99.9%)
Feb (100.0%)
Mar (100.0%)
Apr (100.0%)
May (100.0%)
Jun (99.9%)
Jul (99.9%)
Aug (100.0%)
Sep (99.9%)
Oct (99.7%)
Nov (99.5%)
Dec (97.5%)
-0.5 0.0 0.5
BIAS: Mixing Ratio (g/kg)
Figure 51. Soccer plot of monthly absolute humidity error and bias averaged over the Fairbanks subdomain for
'2016.
4.4 North Slope region performance
The North Slope subregion performance was assessed using soccer plots to analyze average error
and bias across all stations in the subdomain. Also, Deadhorse airport (PASC) was selected as
the individual station of interest for the analysis. PASC was selected based on the proximity of
the station to the Prudhoe Bay oil developments. Temperature, wind, and humidity performance
for the subregion and at the individual station of interest is reviewed in this section.
4.4.1 Temperature
A plot of temperature distribution at PASC, examined by hour per season is shown in Figure
5 2Figure36. WRF tends to underpredict temperature all hours of the day in the winter at PASC.
WRF temperature distributions compare well to PASC observations over the other seasons.
The soccer plot of bias and error across the North Slope subregion is shown in Figure 53. The
results demonstrate poor WRF performance across the subregion in winter months. WRF
predicts much cooler temperatures than observed, on average.
39
-------
Winter (99.6% Complete)
Spring (98.6% Complete)
I
* 255
5 260
OBS
WRF
1 2 3 4 5 6 7 8 9 101112131415161718192021222324
Hour of Day
Summer (99.8% Complete)
~~~~~
1 2 3 4 5 6 7 8 9 101112131415161718192021222324
Hour of Day
Autumn (100.0% Complete)
280
270
260
250
240
1 2 3 4 5 6 7 8 9 101112131415161718192021222324
Hour of Day
1 2 3 4 5 6 7 8 9 101112131415161718192021222324
Hour of Day
Figure 52. PASC (Deadliorse) seasonal, hour-of-day temperature distributions. ASOS (blue) vs. Alaska 9-kin WRF
(red).
40
-------
North Slope Region: Temperature (K) Performance
6-
5 -
«
k.
3
n 4-
L.
0>
a
E
v
*7. 3'
oc
o
DC
0C
111
Complex^ Conditions
2 -
Month (Completeness)
X
Jan (99.9%)
A
Feb (93.2%)
*
Mar (99.9%)
+
Apr (99.8%)
•
May (97.4%)
~
Jun (100.0%)
~
Jul (100.0%)
*
Aug (99.9%)
¦
Sep (99.9%)
Y
Oct (99.9%)
<
Nov (99.4%)
•
Dec (99.9%)
le Conditio(i
1-
-6 -4 -2 0 2 4 6
BIAS: Temperature (K)
Figure 53. Soccer plot of montlily 2-m temperature error and bias averaged over the North Slope subdomain for
2016.
4.4.2 Wind speed
Wind roses developed from both the observed (PASC) and modeled datasets are plotted in
Figure 54. Although the modes of predominant northeast and southwest winds appear to be well
predicted by WRF, it is evident from the plot that WRF underpredicts wind speed. This is clear
also in the plot of wind speed distribution at PASC, shown in Figure 55. Wind speed predictions
are biased low all hours and seasons of the year at PASC.
The regional average wind speed error and bias soccer plot for the North Slope subregion is
plotted in Figure 56. It is clear WRF is not optimized in this case to adequately predict surface
winds along the North Slope, given the significant low wind speed bias. Bias and error exceed
complex conditions criteria all seasons except summer.
41
-------
Wind speed (m/s)
¦ 0.6 to 2.6 m/s
=1 2.0 to 4.0 m/s
!~ 4.0 to 6.0 m/s
ZD 6.0 to 8.0 m/s
!~ 8.0 to 10.0 m/s
=~ 10.0 to 12.0 m/s
m 12.0 m/s +
SUMMARY:
UNITS:
RECORDS TOTAL:
TOTAL CALM:
% CALM:
HISSING:
% MISSING:
AVG. WIND SPEED
MIN. WIND SPEED
MAX. WIND SPEED
4.0
6.0
1.5
20.6
Wind speed (m/s)
H 0.0 to 2.0 m/s
—2 2.0 to 4.0 m/s
si 4.0 to 6.0 m/s
ZD 6.0 to 8.0 m/s
8.0 to 10.0 m/s
=~ 10.0 to 12.0 m/s
!~ 12.0 m/s +
SUMMARY:
UNITS:
RECORDS TOTAL:
TOTAL CALM:
% CALM:
MISSING:
% MISSING:
AVG. WIND SPEED:
MIN. WIND SPEED:
MAX. WIND SPEED:
Figure 54. Wind rose comparison. PASC (Deadhorse), ASOS observed (left), Alaska 2016 WRF (right).
Winter (95.4% Complete)
Spring (96.8% Complete)
~ * ~!~~~~ ~
17.5
15.0
E 12.5
g 10.0
V)
~ S. 10
OBS
WRF
1 2 3 4 5 6 7 8 9 101112131415161718192021222324
Hour of Day
Summer (96.3% Complete)
~
~ ~
1 2 3 4 5 6 7 8 9 101112131415161718192021222324
Hour of Day
Autumn (95.6% Complete)
~
•:,1•: iNi'iii
~
~ ~
1 2 3 4 5 6 7 8 9 101112131415161718192021222324
Hour of Day
1 2 3 4 5 6 7 8 9 101112131415161718192021222324
Hour of Day
Figure 55 PASC (Deadhorse) seasonal, liour-of-day wind speed distributions, ASOS (blue) vs. Alaska 9-km WRF
(red).
42
-------
North Slope Region: Wind Speed (m/s) Performance
Complex Conditions
Simple Conditions
Month (Completeness)
X Jan (94.2%)
Feb (87.6%)
Mar (90.2%)
Apr (93.1%)
May (92.9%)
Jun (94.8%)
Jul (95.8%)
Aug (97.0%)
Sep (96.0%)
Oct (95.2%)
Nov (89.4%)
Dec (93.2%)
-10 1
BIAS: Wind Speed (m/s)
Figure 56. Soccer plot of monthly wind speed error and bias averaged over the North Slope subdomain for 2016.
4.4.3 Wind direction
The wind direction soccer-plot for the North Slope subdomain is shown in Figure 57. The plot
demonstrates WRF predicts wind direction well across the subregion. Wind direction bias and
error is within the simple conditions criteria all months of the year except November, which is
within the complex conditions criteria.
43
-------
North Slope Region: Wind Direction (deg) Performance
80-
o
fl>
~ 60 -
c
o
¦C
.E 40 -
5
ec
O
cc
cc
Complex Conditions
20 -
S impjeC o ndjtjo n s
-30
-20
20
30
Month (Completeness)
X Jan (94.2%)
Feb (87.6%)
Mar (90.2%)
Apr (93.1%)
May (92.9%)
Jun (94.8%)
Jul (95.8%)
Aug (97.0%)
Sep (96.0%)
Oct (95.2%)
Nov (89.4%)
Dec (93.2%)
-10 0 10
BIAS: Wind Direction (deg)
Figure 57. Soccer plot of monthly wind direction error and bias averaged over the North Slope subdomain for 2016.
4.4.4 Humidity
A plot of the PASC distributions of absolute humidity by hour of day per season is shown in
Figure 58Figure 42. WRF tends to underpredict humidity during the winter and summer seasons.
The North Slope subdomain humidity soccer plot is shown in Figure 59. The average subdomain
bias and error are within the criteria for all months.
44
-------
Winter (99.1% Complete) Spring (97.6% Complete)
123456789 101112131415161718192021222324 123456789 101112131415161718192021222324
Hour of Day Hour of Day
Summer (99.8% Complete) Autumn (100.0% Complete)
123456789 101112131415161718192021222324 123456789 101112131415161718192021222324
Hour of Day Hour of Day
Figure 58. PASC (Deadhorse) seasonal, hour-of-day absolute humidity distributions, ASOS (blue) vs. Alaska 9-km
WRF (red).
45
-------
North Slope Region: Mixing Ratio (g/kg) Performance
Simple and Complex Conditions
.2 2.0
Month (Completeness)
X Jan (99.5%)
Feb (92.4%)
Mar (97.3%)
Apr (98.9%)
May (97.1%)
Jun (99.3%)
Jul (99.2%)
Aug (99.7%)
Sep (99.8%)
Oct (99.8%)
Nov (99.3%)
Dec (99.6%)
-0.5 0.0 0.5
BIAS: Mixing Ratio (g/kg)
Figure 59. Soccer plot of monthly absolute humidity error and bias averaged over the North Slope subdomain for
2016.
4.5 Juneau region performance
The Juneau subregion performance was assessed using soccer plots to analyze average error and
bias across all stations in the subdomain. Also, the Juneau airport (PAJN) was selected as the
individual station of interest for the analysis, due to its location near the center of the subdomain
and its relative distance from significant high terrain features compared to other stations in the
domain. Temperature, wind, and humidity performance is examined for the subregion and at the
individual station of interest in this section.
4.5.1 Temperature
A plot of temperature distribution at PAJN, examined by hour per season, is shown in Figure 60.
Generally, the temperature daily trends predicted by WRF match the observed value trends,
except WRF is consistently cooler than the observations on average.
The soccer plot of bias and error across the Juneau subregion is shown in Figure 61. The results
fall within the complex criteria all months of the year. Temperature bias and error fall within the
simple criteria during the autumn months. The year-round cold biases evident in the PAJN
comparisons are evident in the performance statistics for the entire subregion.
46
-------
Winter (99.5% Complete)
300
Spring (98.6% Complete)
OBS
WRF
P 270
01 280
1 2 3 4 5 6 7 8 9 1011 12131415161718192021222324
Hour of Day
Summer (99.9% Complete)
1 2 3 4 5 6 7 8 9 101112131415161718192021222324
Hour of Day
Autumn (100.0% Complete)
a 290
290
285
280
S 275
270
265
~.ii-
y v ty * *
~ M
1 2 3 4 5 6 7 8 9 101112131415161718192021222324
Hour of Day
1 2 3 4 5 6 7 8 9 101112131415161718192021222324
Hour of Day
Figure 60. PAJN (Juneau airport) seasonal, hour-of-day temperature distributions, ASOS (blue) vs. Alaska 9-km
WRF (red).
47
-------
Juneau Region: Temperature (K) Performance
cc
o
cc
cc
Complex Conditions
®mpie Conditions
t *
~V-
-10 1
BIAS: Temperature (K)
Month (Completeness)
X Jan (99.9%)
Feb (100.0%)
Mar (100.0%)
Apr (99.0%)
May (99.9%)
Jun (99.7%)
Jul (100.0%)
Aug (100.0%)
Sep (99.9%)
Oct (99.9%)
Nov (100.0%)
Dec (90.9%)
Figure 61. Soccer plot of monthly 2-m temperature error and bias averaged over the Juneau subdomain for 2016.
4.5.2 Wind speed
A plot of wind roses developed from both the observed (PAJN) and modeled datasets is included
in Figure 62. WRF appears to predict the magnitude and easterly modes of wind well, given
some error in wind direction is expected at 9 km resolution in a region of complex terrain.
The distribution of wind speed per season by hour of day is plotted in Figure 63. Wind speed
distributions are predicted well in spring and autumn months. WRF tends to overpredict wind
speed in winter and autumn early morning periods and highly underpredicts windspeed
continuously through the summer months.
The regional average wind speed error and bias soccer plot for the Juneau subregion is plotted in
Figure 64. Regional wind speed error exceeds complex criteria over the winter months.
48
-------
Wind speed (m/s)
M D.0 to 2.0 m/s
B 2.0 to 4.9 m/s
ZZ 4.6 to 6.9 m/s
ZZi 6.0 to 8.9 m/s
ZZI 8.6 to 10.6 m/s
ZD 19.9 to 12.9 m/s
H 12.9 m/s ~
SUMMARY:
UNITS:
RECORDS TOTAL:
TOTAL CALM:
% CALM:
MISSING:
% MISSING:
AVG. WIND SPEED
MIN. WIND SPEED
MAX. WIND SPEXD
Wind speed (m/s)
H 0,0 to 2.0 ro/s
¦I 2.0 to 4.0 m/s
ZZI 4.0 to 6.0 m/s
ZZI 6.0 to 8.0 m/s
ZD 8.0 to 10.0 m/s
ZZI 1G.0 to 12.0 m/s
IZ1 12.0 m/s +
SUMMARY:
UNITS:
RECORDS TOTAL:
TOTAL CALM:
% CALM:
MISSING:
% MISSING:
AVG. WIND SPEED:
MIN. V/IND SPEED:
MAX. WIND SPEED:
Figure 62. Wind rose comparison, PAJN (Juneau), ASOS observed (left). Alaska 2016 WRF (right).
Winter (66.8% Complete)
~
~ ~ *
Spring (70.9% Complete)
~~~ ~
r
~ ~
~ ~~»~~~~~* t *
9 10.0
if) 6
OBS
WRF
1 2 3 4 5 6 7 8 9 101112131415161718192021222324
Hour of Day
Summer (73.7% Complete)
$ ^ 100
1 2 3 4 5 6 7 8 9 101112131415161718192021222324
Hour of Day
Autumn (63.3% Complete)
—~—
itri ~ »!~~~ t ~ ~ ~ ~ t
1 2 3 4 5 6 7 8 9 101112131415161718192021222324
Hour of Day
1 2 3 4 5 6 7 8 9 101112131415161718192021222324
Hour of Day
Figure 63. PAJN (Juneau) seasonal, liour-of-dav wind speed distributions, ASOS (blue) vs. Alaska 9-km WRF
(red).
49
-------
Juneau Region: Wind Speed (m/s) Performance
4.0 -
3.5 -
3.0
¦a 2.5 -
ai
•I
a.
t/i
¦a 2.0 H
c
1.5 -
1.0
0.5 -
0.0
Complex Conditions
Simple Conditions
-2
Month (Completeness)
X Jan (70.0%)
Feb (71.5%)
Mar (73.8%)
Apr (72.3%)
May (70.1%)
Jun (68.0%)
Jul (63.0%)
Aug (61.1%)
Sep (64.1%)
Oct (65.3%)
Nov (79.2%)
Dec (70.8%)
-10 12
BIAS: Wind Speed (m/s)
Figure 64. Soccer plot of monthly wind speed error and bias averaged over the Juneau subdomain for 2016.
4.5.3 Wind direction
The Juneau subdomain wind direction monthly bias and error is plotted in Figure 65.
Performance is within the bias complex criteria but exceeds the error benchmark in summer
months.
50
-------
Juneau Region: Wind Direction (deg) Performance
80-
Ol
V
- 60
c
o
Q
¦D
.E 40
5
oc
O
cc
cc
20-
* .
Complex Conditions ^
~
' %
«
v x ~
A +
*
Simple Conditions
-30
-20
20
30
Month (Completeness)
X Jan (70.0%)
Feb (71.5%)
Mar (73.8%)
Apr (72.3%)
May (70.1%)
Jun (68.0%)
Jul (63.0%)
Aug (61.1%)
Sep (64.1%)
Oct (65.3%)
Nov (79.2%)
Dec (70.8%)
-10 0 10
BIAS: Wind Direction (deg)
Figure 65. Soccer plot of monthly wind direction error and bias averaged over the Juneau subdomain for 2016.
4.5.4 Humidity
A plot of the PAJN station distributions of absolute humidity by hour of day per season is shown
in Figure 66. Overall, WRF tends to be biased low, most evidently in summer and spring
morning hours.
The Juneau subdomain humidity soccer plot is shown in Figure 67. The average subdomain bias
and error are within the benchmark criteria for all months.
51
-------
Winter (99.5% Complete)
Spring (98.6% Complete)
OBS
WRF
12
11
10
AC
3 9
0
1 8
U)
c
:* 7
1 2 3 4 5 6 7 8 9 101112131415161718192021222324
Hour of Day
Summer (99.9% Complete)
~ ~
1 2 3 4 5 6 7 8 9 101112131415161718192021222324
Hour of Day
Autumn (100.0% Complete)
10
~ ~ ~
6
5 ~
4
liHftflifl
p-¦* ~ JI ~~ 2 ' II
1 2 3 4 5 6 7 8 9 101112131415161718192021222324
Hour of Day
1 2 3 4 5 6 7 8 9 101112131415161718192021222324
Hour of Day
Figure 66. PAJN (Juneau) seasonal, hour-of-day absolute humidity distributions, ASOS (blue) vs. Alaska 9-km
WRF (red).'
52
-------
Juneau Region: Mixing Ratio (g/kg) Performance
5imple and Complex Conditions
.2 2.0
Month (Completeness)
X Jan (90.8%)
Feb (90.9%)
Mar (90.9%)
Apr (89.8%)
May (90.7%)
Jun (90.1%)
Jul (90.6%)
Aug (90.8%)
Sep (90.8%)
Oct (90.8%)
Nov (90.5%)
Dec (81.8%)
-0.5 0.0 0.5
BIAS: Mixing Ratio (g/kg)
Figure 67. Soccer plot of monthly absolute humidity error and bias averaged over the Juneau subdomain for 2016.
4.6 Alaska Peninsula region performance
The Alaska Peninsula subregion performance was assessed using soccer plots to analyze average
error and bias across all stations in the subdomain. Also, the Unalaska airport dataset (PADU)
was selected as the individual station of interest for the analysis, due to its location near
industrial developments at Unalaska and Dutch Harbor.
4.6.1 Temperature
A plot of temperature distribution at PADU, examined by hour per season is shown in Figure 68.
WRF results are consistently cooler than the observations all of the year. The cool bias is evident
regionally in the soccer plot of bias and error across the Alaska Peninsula subregion, shown in
Figure 69. However, the bias and error are within the complex conditions criteria.
53
-------
a. 274
Winter (94.9% Complete)
~
~ ~ 9 ~
~~
~Ml
Spring (73.1% Complete)
~~~MM
285.0
282.5
2/5.0
272.5
270.0
267.5
265.0
OBS
WRF
1 2 3 4 5 6 7 8 9 101112131415161718192021222324
Hour of Day
Summer (83.9% Complete)
1 2 3 4 5 6 7 8 9 101112131415161718192021222324
Hour of Day
Autumn (98.1% Complete)
295
290 +
285
275
290.0
287.5
285.0
282.5
§ 280.0
E
277.5
275.0
272.5
1 2 3 4 5 6 7 8 9 101112131415161718192021222324
Hour of Day
12 3 4 5 6 7 8 9 101112131415161718192021222324
Hour of Day
Figure 68. PADU (Unalaska) seasonal, hour-of-day temperature distributions, ASOS (blue) vs. Alaska 9-km WRF
(red).
54
-------
Alaska Peninsula Region: Temperature (K) Performance
g 4
-------
Wind speed (m/s)
H 0.0 to 2.0 ra/s
I 8.0 to 10.0 m/s
I 10.0 to 12.0 m/s
I 12.0 m/s <¦
Wind speed {m/s)
H 0.0 to 2.0 m/s
SUMMARY:
UNITS:
RFCORDS TOTAI :
TOTAL CALM:
% CALM:
MISSING:
% MISSING:
AVG. WIND SPEED
MIN. WIND SPEED
MAX. WIND SPEED
9 m/s
0 m/s
0 m/s
8.0 to 10.0 m/s
10.0 to 12.0 m/s
12.0 m/s +•
I 2.0 to 4
I 4.0 to 6
1 6.0 to 8
SUMMARY:
UNITS:
RECORDS TOTAL:
TOTAL CALM:
% CALM:
MTSSING:
% MISSING:
AVG. WIND SPEED:
MIN. WIND SPEED:
MAX. WIND SPEED:
Figure 70. Wind rose comparison, PADU (Unalaska), ASOS observed (left), Alaska 2016 WRF (right).
Winter (77.5% Complete)
Spring (68.4% Complete)
E 12.5
I
i
1 2 3 4 5 6 7 8 9 101112131415161718192021222324
Hour of Day
Summer (68.6% Complete)
~
1 2 3 4 5 6 7 8 9 101112131415161718192021222324
Hour of Day
Autumn (84.5% Complete)
m v
£10
1 2 3 4 5 6 7 8 9 101112131415161718192021222324
Hour of Day
1 2 3 4 5 6 7 8 9 101112131415161718192021222324
Hour of Day
Figure 71. PADU (Unalaska) seasonal, hour-of-dav wind speed distributions, ASOS (blue) vs. Alaska 9-krn WRF
(red).
56
-------
Alaska Peninsula Region: Wind Speed (m/s) Performance
4.0
3.5
3.0
-o 2.5 H
v
v
a
i/i
¦D 2.0
1.5 -
1.0 -
0.5
0.0
«
Complex Conditions
A
£imn|fe Conditions
~ *"
•
-2
-10 1
BIAS: Wind 5peed (m/s)
Month (Completeness)
X Jan (91.6%)
Feb (88.4%)
Mar (84.6%)
Apr (88.7%)
May (91.1%)
Jun (76.7%)
Jul (78.5%)
Aug (86.2%)
Sep (85.9%)
Oct (83.7%)
Nov (90.5%)
Dec (90.6%)
Figure 72. Soccer plot of monthly wind speed error and bias averaged over the Alaska Peninsula subdomain for
2016.
4.6.3 Wind direction
The Alaska Peninsula subdomain wind direction monthly bias and error is plotted in Figure 73.
Wind direction bias and error fall within the complex criteria all months of the year.
57
-------
Alaska Peninsula Region: Wind Direction (deg) Performance
80 -
"O
c
EC
o
a.
cc
40 -
20 -
Complex Conditions
*
S Lm_Pl_e_Co ndjtjo n _
1
1
1
1
1
1
t
*»
< 1
A
X r 1
1
1
1
1
l
1
1
1
1
1
1
1
-30 -20 -10 0 10
BIAS: Wind Direction (deg)
20
30
Month (Completeness)
X Jan (91.6%)
Feb (88.4%)
Mar (84.6%)
Apr (88.7%)
May (91.1%)
Jun (76.7%)
Jul (78.5%)
Aug (86.2%)
Sep (85.9%)
Oct (83.7%)
Nov (90.5%)
Dec (90.6%)
Figure 73. Soccer plot of monthly wind direction error and bias averaged over the Alaska Peninsula subdomain for
2016.
4.6.4 Humidity
A plot of the PADU distributions of absolute humidity by hour of day per season is shown in
Figure 74. The Alaska Peninsula subdomain humidity soccer plot is shown in Figure 75. The
magnitude of bias and error are low, falling within the benchmark criteria.
58
-------
Winter (94.4% Complete)
Spring (71.3% Complete)
a> 4 .0
K 3.5
OBS
WRF
1 2 3 4 5 6 7 8 9 101112131415161718192021222324
Hour of Day
Summer (80.3% Complete)
1 2 3 4 5 6 7 8 9 101112131415161718192021222324
Hour of Day
Autumn (97.3% Complete)
1 2 3 4 5 6 7 8 9 101112131415161718192021222324
Hour of Day
1 2 3 4 5 6 7 8 9 101112131415161718192021222324
Hour of Day
Figure 74. PADU (Unalaska) seasonal, hour-of-day absolute humidity distributions, ASOS (blue) vs. Alaska 9-kin
WRF (red).
59
-------
Alaska Peninsula Region: Mixing Ratio (g/kg) Performance
w
o
to
ec
x
'£
a
o
cc
en
LU
3.0-
2.5-
2.0-
1.5 -
1.0
0.5 -
0.0
Simple and Complex Conditions
•
~
-1.5 -1.0 -0.5 0.0 0.5
BIAS: Mixing Ratio (g/kg)
1.0
1.5
Month (Completeness)
X Jan (96.1%)
Feb (99.4%)
Mar (96.7%)
Apr (95.7%)
May (92.9%)
Jun (88.1%)
Jul (84.3%)
Aug (98.9%)
Sep (99.6%)
Oct (98.0%)
Nov (96.4%)
Dec (99.3%)
Figure 75. Soccer plot of monthly absolute humidity error and bias averaged over the Alaska Peninsula subdomain
'for 2016.
4.7 Upper-air analysis
A subset of upper-air station datasets was selected to illustrate general model performance at
different locations across the 9 km domain. More specifically, five stations were selected to
evaluate a variety of locations in Alaska, one in each of the subregions analyzed in the previous
sections. The upper air stations were selected because they were representative of the regions in
which they were located, with preference given to upper air stations located near the surface
stations that were analyzed previously. Performance plots of all the upper-air station datasets are
available in Appendix A.
The CYEV (Inuvik, Canada) upper air station was selected to analyze performance in the
northeast region of the WRF domain. The Oz and 12z sounding distributions of temperature,
wind speed, wind direction, and relative humidity error per season are plotted in Figure 76,
Figure 77, Figure 78, and Figure 79, respectively. Temperature near the surface tends to be
biased slightly cool in the morning (12z soundings) and biased warm in the afternoons (Oz
soundings) in the spring and summer. Relative humidity is shown to be biased a bit dry in the
winter and summer and a bit wet in the spring and autumn.
60
-------
Winter 12z
•HIH*
•KF
• h|H
|-HH
Winter Oz
Spring 12z
Spring Oz
925.0
1000.0 ~
-8
Summer 12 z
300.0
«|-[
]->»
1^0 %
400.0
H-
b
100%
500.0
'Kh*
100%
700.0
~
100%
850.0
i-HH •
100%
925.0
»~ ~ h
¦
1000.0
~ ~
I
\
HIH-
i ~
98.9%
HIH*
98.9 %
hQH*
98.9%
~
~HIH *
~
98.9 %
~
KIH
> ~~
98.9 %
~
b
-0}-
-|*» ~
98|%
1—
—\
II-
1 ~
86.8%
I
HI
+•
iHh*
•i B i"
Summer Oz
—I
H
h
1 ~
2 4 6 8
KIH"
•
HH>
h|H'
~ ~
~ #
^ 1M 'A
^ 1M Vi
J# ~ ipo^
-2 0
Autumn 12z
i-QH"
•HIH
HB-i'
• i—|—i*
- HIH
^90.9
-6 -4 -2
| 1 81«%
0 2 4
4+
4
¦4+
h
-b<
-10.0 -7.5 -5.0 -2.5 0.0 2.5 5.0
Autumn Oz
h ~~
~ ~
HI
> I •
-[EH*
| "X)
2 4 6
Temperature Error (K) Temperature Error (K) Temperature Error (K) Temperature Error (K)
Figure 76. CYEV (Inuvik. Canada) upper-air distribution of temperature error by pressure-level and season (dataset seasonal completeness indicated in blue).
61
-------
Spring 12z
Spring Oz
H
10
0f}%
~ *\
-\*
10
0.0%
-{I
]—
+ ~
10
0.0%
~
~~H
i—
"fflH
—I
10
10
0.0%
0.0%
~ #~
~ H
I—
I
10
s
0.0%
.3%
-7.5 -5.0 -2.5 0.0 2.5 5.0 7.5
Summer 12z
I
~
—I
f 98.9%
Hi
D-
—I
~
98.9%
98.9%
h-
—IUU-
H
98.9 %
HS-
—
~
98.9 %
~ ~ «
H
~
98,9%
i—
Hl I
—I
~
45^%
£~
~ h
~ h
i**
H ~#
-m-
~ ~
| 1 | |—
| |00 ¦*.
4ni—
~
+ f llWUVi
| 100.0%
~
•m-CH—
—f-B—
•J-1 ¦
| 100.0%
I %
~ ~~
••i—HH
-CI—
~«(-
~~~
h
i—DH
-m
~ ~
h
~ 71#«
-6^-2 0 2 4
8 -10.0 -7.5 -5.0 -2.5 0.0 2.5 5.0 7.5-10.0 -7.5 -5.0 -2.5 0.0 2.5 5.0
Summer Oz
Autumn 12z
Autumn Oz
~ ~b
~ ~
m
~ ~
£~—>
m-n
-m-
~ *i-
~ #
~ ~
~ h
H ~
KTH-
' M '•
-ffl-
^9B.g %
~ ~ s
•if-Bn
{EH
50^%
~ ~
HI—1
98.9 %
-•KIM
98.9%
•4-flH *
97# %
~
i-EH*
97.8 %
•HU—<•
97.B %
~
97f %
* JlHB—1 *•
^48.4%
-6 -4 -2 0 2 4 6
Wind Speed Error (m/s)
8 -7.5 -5.0 -2.5 0.0 2.5 5.0 7.5
Wind Speed Error (m/s)
-7.5 -5.0 -2.5 0.0 2.5 5.0 7.5 -10.0 -7.5 -5.0 -2.5 0.0 2.5 5.0 7.5
Wind Speed Error (m/s)
Wind Speed Error (m/s)
Figure 77 CYEV (Inuvik, Canada) upper-air distribution of wind speed error by pressure-level and season (dataset seasonal completeness indicated in blue).
62
-------
300.0
~
~
100.9%
400.0
~ ~
100.0%
1" 500.0
~
•fH'
~ ~
100.0 *4
"55
>
0)
-J 700.0
a)
3
~ ~ '
100.0%
V)
| 850.0
—Ql—!**+++ *
100.0%
925.0
~ ~
ioqjp%
1000.0
~
~~ ~
59.3%
-100 -50 0 50 100 150
Summer 12z
~ «H
~ ~
850.0 ~ ~
1000.0 ~
+f*
b
h|+»
<4+
•t-QH
•HI
#
H ~~
98^%
-150 -100 -50 0 50 100 150
Wind Direction Error (Degs.)
~ M
~ ~
A
H 4|>*
+>
hQH
~ ~
~
>~ ~
~ ~
Spring 12z
+1+
+4+4
~ I—[[]—I ~ ~
1,1 I 1 M
Spring Oz
^10 0 %
10Q^) %
• ~ 58#% ~ ~ ~~
~ ~
-150 -100 -50 0 50 100 150
Summer Oz
• \
•h
~~
#~
~~ \
!+~
H[H#*
I—¦—I'
OH
10^%
-150 -100
-50 0
Autumn 12
50
Z
100
150
~4^.
~
98.9 %
98.9%
•hj+«<
98.9 %
•4+
96|%
~
•hih'
~ ~
98.9 %
~
KIH ¦
~ ~#
98.9%
~ ~ # I-
— | —1
~
~
50.5%
97 8 %
~
~
#•
97.8 'X.
A 4
~ *h{[>
iaI—TILi <
L
9«.7%
95.«%
~~
~ ~
~ 1
~
lfMJJ 1 i
* 111*
f
^95.6 %
~
HIH
«*#
95f %
~
~~~h
-{Jh
U
-150 -100 -50 0 50 100 150
Autumn Oz
*iJbm
T P
"4t»
"#•
•4+
-i—^
KD
~~ ~~
~ MHIH* *
-I*
~~ ~ I—
—~ «
H ~~
-150 -100 -50 0 50 100 150
Wind Direction Error (Degs.)
-150 -100 -50 0 50 100 150
Wind Direction Error (Degs.)
-150 -100 -50 0 50 100 150
Wind Direction Error (Degs.)
Figure 78. CYEV (Inuvik, Canada) upper-air distribution of wind direction error by pressure-level and season (dataset seasonal completeness indicated in blue).
63
-------
Spring 12z
Spring Oz
~ 1
j|| 100.0 «
• HI
~ w \
~ ~ ~ hi
y—~ ,000*4
JJ 1 ?f>
~ i—i
• • H
1 l#f w 101.01
J l» ~~ '™»*
KD
J 01.2 %
-60 -40 -20 0 20 40 60
Summer 12z
1
* »*\
J 1 ~ ~ 11*
# ~ | 1~
j |
m* \ j 1 ~ »>»'¦
i—uj—i
• 1 II
1—III
i
| 54 ,6 %
~ ~
* *
•i—CD—
huh
HI]—i
* 4
~ #|—UJ—I m ~~
J 1 ~
J f 86.8
-60 -40 -20 0 20 40 60
Summer Oz
KU
~ ~ h
~ » ~
~ ~
-| ~~ '(P-
I yo0
~~ I-
~ h
H ~
IH
-40 -20 0 20 40 60 -60 -40 -20 0 20 40
Relative Humidity Error (%) Relative Humidity Error (%)
~ h
-| ~~ ~"
-I ~ "
—|* ~ "-*%
-40 -20 0 20 40
Autumn 12z
I
98 » %
~ I 1 | | 1
i—m—1 * "ri
«»~ #«i—^
J 1 m m
~ ~
-------
Figure 81, Figure 82, and Figure 83, respectively. The results demonstrate WRF is biased warm in winter, autumn, and summer
months near the surface but relatively unbiased aloft. Wind performance does not appear to be significantly biased aloft (few 1000 mb
records of wind speed were available). Humidity is biased a bit dry nearer the surface and biased wet in the mid-levels most of the
year.
£ 500.0 ~
HrH"
H[H* *
*
Winter 12z
~
HH
#
925.0 ~ ~
HEH
Winter Oz
. nip •
« ~~~)- —|w
> *1—1|—I •
I—[]]—b ~~
HH—!~ •
Spring 12 z
Spring Oz
-2.5 0.0 2.5 5.0 7.5 10.0 -€-4-2 0
Summer 12z
300.0
KEH
100 %
400.0
. i [Qh *
100 %
| 500.0
100%
"5
>
-J 700.0
ID
•H-f-H. •
100%
o
0)
£ 850.0
•HUH* •
1 100 %
925.0
.... i-[j
1
^ 100%
1000.0
~
»i—|
1 h-
—110**
03
Summer Oz
H * ~
IH"
~ ~
KB-h
K&-
H.I
~ ~ ~
~ ~
-I ~
h
I—
~ b
}>
H
~
100%
100%
»\-
~
11X1%
~
.|-§H .
100%
i-OH*
~~
^100%
~
~
I-
KI
—\
y-
i b
—I
10^%
t 1M*
-6
4
2 0
Autumn 12z
6
~ •
~~~~ im%
HD
•HIH
• h|H
"• 'i-QH
••h-UH
•i—on-
~ ~
~ ~
~ ~
~~~~
_| t 10|%
~
—!~
00%
'HI~
H
00%
~~i—[f-
>
00%
~
i-fl-
¦\* ~ ~
00%
~
—[]]—1 ~~~
00%
~ [
I—I
-tEhf -
H1 H
00%
0^%
-4 -2
0
2
Autumn Oz
6
IH
LI—
>
00%
~
HIH
00%
~
HIH"
00 %
HDh •
00%
hQH-
~
00%
~
~—l#'
w
00 %
1
0|%
-2 0 2
Temperature Error (K)
0 2 4
Temperature Error (K)
-2 0 2
Temperature Error (K)
Temperature Error (K)
Figure 80. PAFC (Anchorage) upper-air distribution of temperature error by pressure-level and season (dataset seasonal completeness indicated in blue).
65
-------
Winter 12z
Winter Oz
Spring 12z
Spring Oz
~
[¦
H
~
100 %
100%
1
E-
—1
-h
fiOO %
100 %
~
4
—1
|
Eb
H *
100%
m*
T ^
1
H
11.0%
10|'% ~ # (-
H ~
~~
~ «
•HIH
~~
~ ~ ~
'—EE—1<
Summer 12z
H
Summer Oz
Autumn 12z
~ mi-
lt
« ~
£ 850.0
1 ~ 1M"
-| | | 1* * ~ ,M*
-| | j 1 # t 1[V'
925.0 ~ ~
~
i—
¦
—1
100%
~
i—
100%
1
h-
:
IH
#
1
100%
r
~
h
-L
Lh
1
—1
~
10^%
~ ~
L
—1
~
100%
~ ~
~ ~
' *—II
{~
-(~~ ~ ~
~ ~
H ~
am
*• KE
>—II
~ 1f*
7.5
-7.5
-5.0 -2.5 0.0 2.5
Autumn Oz
5.0
7.5
100%
~
i—LLI
H ~
100%
100%
~
HEH- t
100%
100%
~
¦1 m
100%
f)Q%
##
^ 10|%
n%
KD-m*
#
100%
4.4%
2.2%
-7.5 -5.0 -2.5 0.0 2.5 5.0 7.5
6 8 -7.5 -5.0 -2.5 0.0 2.5 5.0 7.5 10.0
Wind Speed Error (m/s) Wind Speed Error (m/s) Wind Speed Error (m/s) Wind Speed Error (m/s)
Figure 81. PAFC (Anchorage) upper-air distribution of wind speed error by pressure-level and season (dataset seasonal completeness indicated in blue).
66
-------
Winter 12z Winter Oz Spring 12z Spring Oz
300.0
~
~
100 %
~
100%
11X1%
•• +
># ~
100%
400.0
4
b
~
100%
~ ~
100%
~ ~
•Mb
~
~
100%
• 4-
»~
100%
^ 500.0
*h|
b
100%
~
+fl-
~
~
100%
H+
~
100%
H
100%
0)
-• 700.0
jD
~I-IP—\*
~ ~
100%
~
¦
H ~
100%
+1*
100%
H-fl-
»~ ~ ~
100%
V}
10
£ 850.0
~
H
Ib
+ •
100%
~
•Hth*
1<^%
~ ~
#|
10|%
~
~
•i—QH •
100%
925.0
~ «M
* ~~
b
b
~ ~
11X1%
~
H
~
~
100%
~
H
—CD-
~
100%
~
~
I—Q]—
iqj%
1000.0
~
I—
—~
~ ~
~ b
¦
Hh
»
~~
>-m-
H ~
~~~
^00%
~~
<
—{E
H :
100%
-150 -100 -50 0 50 100 150 -150 -100 -50 0 50 100 150 -150 -100 -50 0 50 100 150 -150 -100 -50 0 50 100 150
Summer 12z Summer Oz Autumn 12z Autumn Oz
300.0
~
100%
¦Hb f
10C
%
~
4f»*
100%
«4j*
100%
400.0
~
•H}+.
f)00%
~~
<+§+•
10
%
~
4+
100%
~
H
100%
| 500.0
~
100%
~
•4+.
10
%
i
'•4+ *
100%
~f[H«
*
100%
u
£
-i 700.0
~
.hflH.
~
100%
~
~
»h|h» •
10G
%
~
~
•4b •
100%
~
m
^ 100%
in
i/)
0)
& 850.0
mm
h-OH*
100%
~
KD-" •
10C
%
~
~ ~ ~
i-HH*
100%
~
~
•HD-f"
^ 100%
925.0
~
~ ~
t-QH
• ~
~
100%
~
h
-[
EM
m ~'*
%
~
H
i-db-
~
100%
~
~ ~
~
HI
^ ^ 100 'S.
1000.0
~
~ ~
i—
H «
~ #
10^%
H ~
~ b
H
I—
—1
%
~
~
(—
IH
~
90^ •/,
~ ~
b
HI
I—I
~ MM1*
-150 -100 -50 0 50 100 150 -150 -100 -50 0 50 100 150 -150 -100 -50 0 50 100 150 -150 -100 -50 0 50 100 150
Wind Direction Error (Dogs.) Wind Direction Error (Degs.) Wind Direction Error (Degs.) Wind Direction Error (Degs.)
Figure 82. PAFC (Anchorage) upper-air distribution of wind direction error by pressure-level and season (dataset seasonal completeness indicated in blue).
67
-------
Winter 12z
¦HEH
•HDH
Spring 12z
Spring Oz
| 500.0 ~ ~ ~
~ ,0°%
_| , 10,%
" 700.0 ~ ~ ~ ~!-
~
~
H
1 H
LLJl—"•
#
~
~i-l—<4*
00%
~
1
1 ' *
00%
~
h
—c
h
00 %
~
HJJ-
—1
00%
-60
40
20 0 20
Autumn 12z
40
60
h
HI
|—
H ~
^ 100 %
~
~h
H ~
~
H
n-
H#*
fpo %
~ ~
H3-
^100 %
~
*
1
1
H*
100 %
\~
-\
h
~
^ 100 %
1—
H
100-a
100%
1—II
]—1
100%
100%
H]
1
1
, ^ 100%
1
—~
1
100%
100% ^
—0:
H
100%
100%
—1*
-40 -20
0 20 40 60
Autumn Oz
~
t
~ i
II—"
|rH»*
100%
, ^100 %
~
~ h
~
—II I ^
—LLH *
| 10|%
( 100%
~
~ f
1
100%
#
H 1
H
A 100 %
f-
-03
—1|
100%
-60 -40 -20 0 20 40
Relative Humidity Error (%)
-60 -40 -20 0 20 40 60
Relative Humidity Error (%)
-40 -20 0 20 40 60 -60 -40 -20 0 20 40 60
Relative Humidity Enror (%) Relative Humidity Error (%)
Figure 83. PAFC (Anchorage) upper-air distribution of relative humidity error by pressure-level and season (dataset seasonal completeness indicated in blue).
The PAFA (Fairbanks) upper air station was selected to analyze performance in the central inland part of the state, in the region of the
Fairbanks PM2.5 non-attainment area. The Oz and 12z soundi ng error distributions of temperature, wind speed, wind direction, and
relative humidity per season are plotted in Figure 84, Figure 85, Figure 86, and Figure 87, respectively. Notably, the plots show a
bias in temperature, wind speed, and relative humidity at the surface in winter, but less bias at higher pressure levels.
68
-------
Spring 12 z
Spring Oz
925.0 ~ |-
H
i]-" i
100%
HIH
^ 100%
~I—
]-\*
100%
H
Jh «
100%
I-
100%
~~~b)
]-~
100%
~ (-
—m—iw #
100%
10|%
• 1
•»f—rn
h
—1
100%
^ 100%
i* ~ ~***
1 LU
1
1 1 «
—| m*
b
|l3 2%
11.0%
H
~—1
100%
~~ h
100%
>~ H
b> ~
100%
~
'I uj i
I 1 1 * ~ ~
100%
(—
—1
10^%
-4-2 0 2
Summer 12z
~ ~~
~ ~
£ 850.0
~ *
~
I—[J]—l» I—[J]—h
I—[[]—~,M% «4-Q]H
i—(EH— - KIH
HUH ~ ,M% I—[J]—I
I [ I "" ~ ~ I—]J] 1
I QJ 1»~ |
-4 -2 0 2 4 6 -2 0 2
Summer Oz Autumn 12z
I#
~ ~
#~
ll-H
~
00%
»h
{[H' •
00%
~
•HIH
•|—CD—i
M
00%
00%
I—
~ ~ ~ ~
I
—11
| 1
100%
4.4%
KD-
100%
~ h
P "
100%
~~h
u**
~
100%
~
|—
¦KIH
-in—i
# ~
1|0%
100%
I
I A A
I
2.2 %
-4 -2 0 2 4
Autumn Oz
~
~ ~
~ *
~ ~~ ~
i-OH
KTH
HEM
HM
~ ~
~
-4 -2 0
Temperature Error (K)
2 -3 -2 -1 0 1 2 3 -4 -3 -2 -1 0 1 2 3 -3 -2 -1 0 1 2
Temperature Error (K) Temperature Error (K) Temperature Error (K)
Figure 84. PAFA (Fairbanks) upper-air distribution of temperature error by pressure-level and season (dataset seasonal completeness indicated in blue).
69
-------
H
D-i
m
100%
M0-
H
I
~
100 %
•
I
100%
~
#
—m-
i—I I
H *
100%
m*
h
H
11.0 "A
10|% ~ 4 L
Spring 12z
O
Spring Oz
H ~
~ *
44
KEH
m
~~
1Q|%
~
—rn-
»\— —
H ~
100%
100%
I—l~l—
" i-B—
Ml h
H
—I
100%
100%
100%
I—r-
—I
~ 1-[n-
)~ *
100%
y H~
~-
t
100%
•H
11-
i
100%
~
H
I]-
-I*
~
100%
~
KD-
~
~
100%
I-
I
co-
—I
—
|po%
~1
1—
—l
H ~
#1
#
^ 10|%
100%
4.4
2.2%
-6 -4 -2 0
-7.5 -5.0 -2.5 0.0 2.5 5.0 7.5
-6 -4 -2 0
-7.5 -5.0 -2.5 0.0 2.5 5.0 7.5 10.0
Wind Speed Error (m/s) Wind Speed Error (m/s) Wind Speed Error (m/s) Wind Speed Error (m/s)
Figure 85. PAFA (Fairbanks) upper-air distribution of wind speed error by pressure-level and season (dataset seasonal completeness indicated in blue).
70
-------
850.0
925.0
1000.0
100%
~
~
10^%
~
••HDh ..
100%
~
MIH*
100%
~ ~
4IH
~
100%
••I—hh
100%
~
D
11.0%
-100 -50 0 50 100
Summer 12z
300.0
~
~ # 4|>
•
100%
400.0
LA»*
100%
| 500.0
~
~ ~
100%
•3 700.0
~
#~4—UH"
~
100%
£ 850.0
"*HIH
~ H
10^%
925.0
—HD-"
M ~
~ ~
100%
1000.0
Winter Oz
~ ~
# ~
~ ~ ~
*
•Hi*
• -i-|>«
«»hjj-t
hOH
HIH
I
~ ~ ~
M
~
Spring 12z
00% ~ #
Spring Oz
~ ~ ~
auILl
;'l ¦'
HI
00%
00%
#
~ ~ ~
00% # f 4
0|% 4 f
HIH
.hfn
l-i-H
~ ~ ~ ~
~ ~
0||%
» h|+
~ ~
~«|-Jj-|# ~ ~ ~
\
A 1 4 L
> i—jjjn Hi*
1 A 4 1MW
9 V f p
# t ~ ~ 1
I I * ' *
\—{J] 1 - *•*
J.2%
150 -100 -50 0 50 100 150 -150 -100 -50 0 50 100 150 -150 -100 -50 0 50 100 150
Summer Oz Autumn 12z Autumn Oz
~ #
«b|> -
«• (-|h»
~i-IH
#
i-fflH"
•h
~I-
-150 -100 -50 0 50 100 150
Wind Direction Error (Degs.)
•I-—[J]—H ~ ~
-150 -100 -50 0 50 100 150
Wind Direction Error (Degs.)
~ ~ HIH ~~ ~
4 ~ ~!
~ ~~
1 Uu
jj rw
1H
•4*
HDh
~ ~ ~
HD-I
1^) V,
~ ~ ~
~ ~ ~ ~
EH
-150 -100 -50 0 50 100
Wind Direction Error (Degs.)
-150 -100 -50 0 50 100
Wind Direction Error (Degs.)
Figure 86. PAFA (Fairbanks) upper-air distribution of wind direction error by pressure-level and season (dataset seasonal completeness indicated in blue).
71
-------
i—Q]—i
Spring 12z
Spring Oz
400.0 ~
* * —ED
~ # h
H ~
HIH
10|%
-40 -20 0 20 40
Summer 12z
~ *h
~ ~h
~ ~ H
+ + + 100%
ID-
¦ ¦
-+~
100%
~ ~ ~
HTb
—|
100%
~ 1—
H
t
~ h
-Eh
H ~
w
~ J—
H ~
100%
~
~ 4
1 "
—1 ~
100%
~
1
110%
-40
-21
0 0
Summer O2
20
40
^>00 %
~
~
f 10|%
~
~
kn—1
~ <
~ ~
•hdOj-i'
100%
* 1—m—k
^ 100%
100%
-20 0 20 40
Relative Humidity Error (%}
-40 -20 0 20
Relative Humidity Error (%)
H ~ 11
» h
~~ h
~ h
| t|0%
Juki %
0 20 40
Autumn 12z
HI
1—1
100%
~ ~~
i—[EH"
I III l .
100%
~ ~
1 II 1 1 '
•HlbH •
100%
~ ~
HFH •
J 1
100%
100 ¦».
4.4%
1
iH-
J 1
^ 100%
^ 100%
n
~
~ 1—|
1
1 1
m 10^%
, ^ 100%
—1
JJ—
^ 100%
1—T
| |
2.2%
0 20 40 60
Autumn Oz
1—1
—I*
, 100%
~ ~~ l 1
4 V—UK
— ~ ~
H ~ «
100%
\*
f100%
~ »h
¦
—1
#
1—
u
—1
100%
2.2%
-20 0 20 40
Relative Humidity Error (%}
-20 0 20
Relative Humidity Error (%}
Figure 87. PAFA (Fairbanks) upper-air distribution of relative humidity error by pressure-level and season (dataset seasonal completeness indicated in blue).
The PABR (Utqiagvik/Barrow) upper air station was selected to analyze performance at the most northern upper-air station in the
domain. The Oz and 12z sounding distributions of temperature, wind speed, wind direction, and relative humidity error per season are
plotted in Figure 88, Figure 89, Figure 90, and Figure 91, respectively. There is a wide distribution of temperature error near the
surface, but predicted temperatures aloft are generally unbiased and accurate. Wind speed and direction error and bias are also
72
-------
relatively low at all heights. Relative humidity is biased low near the surface, especially in winter.
Spring 12z
300.0
Hh*
100%
400.0
~f
}>~
100%
500.0
H
b
100%
15
>
(U
-J 700.0
g)
3
•HF-
100%
(O
fi!
a 8500
100%
925.0
~
¦HIH
H
-------
Spring 12z
Spring Oz
300.0 ~
too % 4 ^ ^
-)~ 10,%
i-m-
700.0 ~
£ 850.0
~ I-
~ ~ f
H
~ ~ i—[J]—I
~ ~~ I—
~
> ~
^0 %
-7.5 -5.0 -2.5 0.0 2.5 5.0 7.5 -6
Summer 12z
-4-2 0 2
Summer Oz
300.0 ~
~ ~ ~~ ~
~ ~~~ (-
~ »h
~ ~
_j , 10,%
H
-m-
-m-
¦+
H ~
H"
^M%
-8 -6 -4 -2 0 2 4
Wind Speed Error (m/s)
-6 -4 -2 0 2 4
Wind Speed Error (m/s)
~
~ ~
I
H 1
XI %
H
I—
i—[
n-
H 1
1KI %
HI %
••HTH
~ ~ 1
1)0%
•¦-in-
I
H ~'
i -
Ml %
~ ~
I
HI I-
j^%
-6-4-2 0
Autumn 12z
~
~
1—
~
1
H
100%
100%
~
ns~
H
H
100%
100%
~
~
H
h
¦
H ~
H ~
100%
10^%
~
—|
H
1
100%
-6 -4 -2 0 2 4
Wind Speed Error (m/s)
~b
~ I—
• i-B-i
\
^ 100%
j~| ] 1 ~ ~ 1™%
.j ^ ^ 100%
_J 100%
-4 -2 0 2 4
Autumn Oz
~ ~
~ b
I 1
|P0%
~
I—
H
IH
|0|%
~ ~h
n
H
—\*
100%
I—
-r
r
—i
100%
~
I—
I
I h •
100%
~
h
—h
100%
~
~ I—
-I
|—I
100%
-4 -2 0 2 4
Wind Speed Error (m/s)
Figure 89. PABR (Utqiagvik) upper-air distribution of wind speed error by pressure-level and season (dataset seasonal completeness indicated in blue).
74
-------
#
~ »H
Winter 12z
~
*
Winter Oz
~ ~
Spring 12z
Spring Oz
H •»
700.0 ~
*h|>M
h
~~~h
~ ~
~ ~
•HH"
HJh
-HD-"'
HIH
* »"¦
~~ hfl"
~
100%
100%
~
•4+
100%
••HIH
100%
~ ~
~
~h-IJ-k*
~
100%
~ ~
•HIH*
~ ~
10|%
~
~
~i—|h ~
~ 1nf*
~ 1,»*
-|«* ~
~ ~ ~ ~~
• «*
~~f
HH
• mm! III lit
^iir
al 01 Ut
•rjrr
~ ~
~ ~
-100 -50 0 50
Summer 12z
100 150
-150
-100 -50
Summer Oz
0
50
100
-50
0 50
Autumn 12z
100
150
-10C
-50
0 50
Autumn Oz
10C
150
300.0
~
100%
~
100%
10|%
100%
400.0
100%
100%
•||
I>
100%
•4* ¦
100%
£* 500.0
* HIH*
100%
~
•Hh
,0»"
>~
1M%
H
H
~
««
I
-J 700.0
Z3
100%
~~h|-i -
100%
~
»h|h ~~
100%
••"DH*
~
100%
£ 850.0
100%
I I
I-
100%
100%
~
H-|-h
~
~
100%
925.0
~ ~ ~
~ 4]
0-i"
w*
~
••H
100 %
~ ~
—I* ~
~
100 %
~~hj
100%
1000.0
~ ~ ~
~~4
Dh-
100%
~
~ ~ ~
~ ~ ~
h;
H •
^ 100%
~
~
HI
H
iyi%
~
~
HI
~ ~
10^%
-100 -50 0
Wind Direction Error (Degs.)
-150 -100 -50 0
Wind Direction Error (Degs.)
-150 -100 -50 0 50 100 150 -150 -100 -50 0 50 100
Wind Direction Error (Degs.) Wind Direction Error (Degs.)
Figure 90. PABR (Utqiagvik) upper-air distribution of wind direction error by pressure-level and season (dataset seasonal completeness indicated in blue).
75
-------
Spring 12z
~ 4
~
100%
~ ~ ~
1
K
~-
>
1 M
100%
~
1
~~~
-0
H-
1 *
10|%
#«~
HD—
m ~
100 'i
~
«~ b-
—|
1—
100%
1—
"1 1
I
100%
-40 -20 0 20 40
Summer 12z
~ ~ ~ ~
i—CD-
HI}
H *~
*« ~~
~ ~ ~
—CD—4
¦
Hfr-
HH
H*
~ ~
« #
H—Q]—I
• HIH
HIH
•HDh
HO
~ m
~~
10^%
m ~ ~
M ~
-60 -40 -20 0 20 40 60
Summer Oz
HTH
H'
0|%
~ ~n
~ ~ ~ ~~ l
£~—
ID-h
~ 4M
-60 -40 -20 0 20 40 60
Relative Humidity Error (%)
~ ~~ i—f]|—i ~
~ I—[[]—I ~» *
~ ~ I—[[]—I ~
-60 -40 -20 0 20 40 60
Relative Humidity Error (%)
~ ~ ~
¦ •
m—
1—i ~
-Q
~~ h
~ ••~f—Q]—i
* <—CD—
-40 -20 0 20
Autumn 12z
~ W*
H ~10l)%
1 ~
~ ~
{IH
-OH
~ ~
~ ~~
~
~
»HIH
HIH
HHh
HIH'
HD-i
~
I
-60 -40 -20 0 20 40 60
Relative Humidity Error (%)
~ # ~
Spring Oz
HEH'
HW
* HCH
-u ~ ~~
###~
• i—EH
Hit
~ ~
-40
-20
0 20
Autumn Oz
40
6C
~ ~
"h
I
H*
—I*
100 %
10^%
HOh
#~
~
H
—I *
yo%
—| ~~
H ~
^100%
~ h
-DE
H
~
100%
-40 -20 0 20 40
Relative Humidity Error (%)
Figure 91. PABR (Utqiagvik) upper-air distribution of relative humidity error by pressure-level and season (dataset seasonal completeness indicated in blue).
The PASN (St. Paul Island) upper air station was selected to analyze performance in the western marine portions of the domain. The
Oz and 12z sounding error distributions of temperature, wind speed, wind direction, and relative humidity per season are plotted in
Figure 92, Figure 93, Figure 94, and Figure 95, respectively. VVRF near-surface temperatures are biased cool in spring and autumn
months, but generally unbiased aloft. Wind performance appears good, with low wind speed and direction error across all layers.
76
-------
Relative humidity is biased a bit high (wetter) in lower layers across all seasons.
Spring 12 z
Spring Oz
|
£ 850.0
700.0
8?
w
I 850.0
925.0
100%
HH
100%
i—[EH**
100 'Si
~~I—\Q—&*
100%
HIH
|0|%
~ ~ ~ I
100%
HEH
|>7.1 %
-4
2 o ;
Summer 12z
>
HIH
4-§|.
#~ ~
HD-
KDh
_| 10^%
-6-4-2 0
Temperature Error {K}
~ I—[J] 1
• KIH
f 100 %
~ i—u—i ~~
• HH
100%
~
i—m—1 f
•HIH-
100%
~
~ i—QJ—i ~ •»*
•HIH
100%
~ i—| —i ~ ~ *»»
~
"HH
100%
~ H
—| 11—i ~ ~'"i" ~
H
HH
t°|»
HJJ—i
HIH *
101*%
Summer Oz
•HH
100%
• HIH*
100%
HH"
^100%
*i—EH"
100%
HIH
w-
~ 1 [
IH
~ -
M <
~ ~! f
1
| 100%
-6 -4
-2 0
Autumn 12z
2 4
i U "
~ 10|%
~
100%
~
HIH* •
100%
HIH -
100%
i—|j[]—i * *
#
HIH
100%
HIH
~ itrA
Temperature Error {K}
Temperature Error (K)
~
HE—<
100%
•HIH
f 100%
• HH
1100 '!h
HH
f 100%
HIH
10^%
#
H
I
H I
1
100%
100%
Autumn Oz
~
I 1
—I
£100%
~
H
11—
H ~
~
~
• [HIH
H ~
100%
^10|%
~
H
i—[_
i—|
—i
1
H
100%
100%
100 %
Temperature Error (K)
Figure 92. PASN (St. Paul Island) upper-air distribution of temperature error by pressure-level and season (dataset seasonal completeness indicated in blue).
77
-------
Spring 12z
Spring Oz
£ 850.0
~~ I—
H
1 lOOT*
~
~H
—I
1N%
~ w—
H
~ 1B»*
h
|—
H
100 %
~ I—
—I*
^ 100A
i
~
I—
CT
h
->~
10|%
~ I—
-m
-
H #
100*
I—
d
h
H
100%
I—
a I
1
10014
~
~
~ ~
HID-'
~ ~
•h
10^%
1|0*
>
-7.5 -5.0 -2.5 0.0 2.5 5.0 7.5 -10.0 -7.5 -5.0 -2.5 0.0 2.5 5.0 7.5
Summer 12z Summer Oz
-6
-4
-2 0
Autumn 122
2
4
~ ~
I—
H i
.10^%
~
I
—h
100%
~
I-
H ~~
100%
*h
-OH
~
100%
~~h
-0}-
100%
~
I—
h
—I
~ ~
10^%
~
H
H
:
M.7%
-7.5 -5.0 -2.5 0.0 2.5 5.0
Autumn Oz
H
i°»*
• i— ~—i
HH—'
•i—EH •
D-h
-10.0 -7.5 -5.0 -2.5 0.0
Wind Speed Error (m/s)
-4 -2 0 2 4
Wind Speed Error (m/s)
-2.5 0.0 2.5 5.0
Wind Speed Error (m/s)
-7.5 -5.0 -2.5 0.0 2.5 5.0 7.5
Wind Speed Error (mfe)
Figure 93. PASN (St. Paul Island) upper-air distribution of wind speed error by pressure-level and season (dataset seasonal completeness indicated in blue).
78
-------
Winter 12z
~~I—li I—~
4DQ.0 ~
~ ~
~ ~ ~
~ ~ ~
~ « ~
~
HIH
HIH
-150 -100 -50 0 50 100 150
Summer 12z
) ~ |—Qj—| ~ # i—Qj—i ~ 110,1
"t-tth
¦ • 1+
> • • 11 i
1 • * HID—1
> ~ ~ ~!—J]h « *~
~
I
10
K)%
~ i
' 'fn
~
10
K)%
f
• t
^ 10
K)%
..hfH.
10
i
«
« 4-||*
~ ~ ~
f
K)%
~
•HIH"
I 10
6%
-150 -100 -50 0 50 100 150
Summer Oz
io|%
~#
~
•HIH*
100%
~
10014
1MJH
HIH
f 100%
~
HEH*
~
~
~ ~HIH ^
10|*
~
—'***
-50 0 50
Wind Direction Error (Degs.)
-50 0 50 100
Wind Direction Error (Degs.)
Spring 12z
]~ ~ r01*
-|m ~
—~ , I
MEH'
~~~ ~ \-
• 41
H
~ ~
#
~ ~
~ ~
~ ~
-100 -50 (
Autumn 12z
'HIH*
[hN
HIH"'
i—H—i ••
h-OH- •
HDi*
HIH *
~~~ 1
50
10f%
~ ~
Spring Oz
~~ ~
~ » H
~ ~
1H-
HIH
¦HIH
** I J] h-*
h-|h
~ ~ ~
t ~
~ ~
~ ~
0 50 100
Autumn Oz
IH
IH
~ |-_
~ ~!—j]
• • hQH*
~ H
~ I—
—I ~ ~
¦4 -
10^%
150
-50 -25 0 25 50 75 100 125 -60 -40 -20 0 20 40 60
Wind Direction Error (Degs.) Wind Direction Error (Degs.)
Figure 94. PASN (St. Paul Island) upper-air distribution of wind direction error by pressure-level and season (dataset seasonal completeness indicated in blue).
79
-------
~ #~
i—CD—'•
i-kD-'
~ ~
f 500.0 ~~ ~ *| || |
-------
99, respectively. Overall, wind speed and direction error aloft is low. The moist bias seen in the other upper-air datasets is also
prevalent in the PANT dataset, especially in spring.
1000.0 |-
300.0
~
i—i
I—
| 100 %
400.0
~
•HIH*
~ 1I0%
| 500.0
•HIH
f 100%
" 700.0
0)
HEH
10^%
n
6
£ 850.0
~h
—ffl—<
I 100%
925.0
~
~
f 100%
1000.0
H-
|-
—i
SO.5%
Summer 12z
• ni •«
.h[i]—1»
-OH
HH
-EH
~ ~ ~
~ ~ ~ *
~ ~ ~ 10*%
i—in—
~ \—jj]—\» !~
HIH
^ 100 %
HUH
~ 1V%
•HIH"
100 %
I
~
H
HIH"
§ 1
f 100 %
54.9%
Summer Oz
•HIH
•HIH
•HIH
• i—Eh
~ ~
~~~
~ ~
~ ~"
Spring 12z
Spring Oz
4
•HIH"
Hi°l%
HIH
100%
~
•HH'
^ 100%
~
~ HQ—l"'
I 100%
l-§H •
100%
~ ~
HIH •
100%
~ ~ ~
• HM—1 ••
65.9 %
-6
4-202
Autumn 12z
>
• H
HIH
—i
^ 100 %
~ H
KEH
100%
h
-EH
H *~
100%
~
HI
H
->~
^ 100%
~
~I—
HIH
H ~
^ 100 %
~ ~«
h
H
1
f 100 %
74 J" %
Autumn Oz
KEH
I ~
10|%
~
-HIH* •
~ ~
100%
—Q]—i-
ioo%
i •
100%
•~HIH*'
100%
4
y=i
-HIH
liEd
~
100%
11.8 %
-!~
• i—El—1
~ 4—[J]—I* ~
~ ~~ ~(¦—m—i ~~
•* <—[0—
~ ~ ~
_ja|%
3 -4 -3
Temperature Error (K) Temperature Error (K) Temperature Error (K) Temperature Error (K)
Figure 96 PANT (Annette Island) upper-air distribution of temperature error by pressure-level and season (dataset seasonal completeness indicated in blue).
81
-------
Spring 12z
Spring Oz
300.0 ~ ~ ~
H* ~'
400.0 ~ ~ ~
~~ h
925.0 ~ ~!-
-10.0 -7.5 -5.0 -2.5 0.0 2.5 5.0 7.5
Summer 12z
300.0 t
-i 700.0
0)
3
n
| 850.0
925.0 ~
iHSH1
<1—
i—DO-1 *
«~ \—~ [~ ~
HOO—i •
•
-2 0 2 4
Wind Speed Error (mis)
^%
~ ~ #
~
-10
~ ~
H ~
EH
ffl-
HE
f ^ + 100%
^ 100 %
j ,0»*
f4J
~ I
i—m—"
100%
~
[EH
f 100%
• HIH'
^ 100%
~
D-i
100%
•HE—'•
f 100%
hth]
^ 65.9 %
-5 0 5
Summer Oz
^ 100 %
10 -e
0
Autumn 12z
10
~ ~ I—f
~~~!—[J
1
H-
100%
^ 100%
~ •
100%
~
> ¦ <
f 100%
1
1
D-h
^ ^ 100%
1
H.
EH
39.6 %
~f
~ ~ ~
I
—i
—i
100%
100%
H
~
I—
E&
—H
—\*
100%
^ 100%
~
i-4
# i-
hiih
—I I I 1
^ 100%
~<#%
H
rr
H—
74.7 %
Autumn Oz
~
H ~
^ 10|%
~ ~ ~
^100 %
~ ~~ h
¦b-f-H.
i B !•
-M
-4-2 0 2
Wind Speed Error (m/s)
-5 0 5
Wind Speed Error (m/s)
10 -7.5 -5.0 -2.5 0.0 2.5 5.0 7.5 10.0
Wind Speed Error (m/s)
Figure 97. PANT (Annette Island) upper-air distribution of wind speed error by pressure-level and season (dataset seasonal completeness indicated in blue).
82
-------
Winter 12z Winter Oz Spring 12z Spring Oz
300.0
400.0
\
> b§H> ~
S" 500.0
» h[[]H
"55
700.0
=j
~
~ J i™%
tn
£ 850.0
~
925.0
~ ~
~ |—jJJ—|« ~ 1M%
1000.0
h
—m—i j
•ij^
100%
100%
.*|h .
100%
~
hjjH"
100%
~
* <
¦i—EH
100%
~
~
10|%
~
h-
H II—i
54.9%
-150 -
100
c
-50 0 50
Summer 12z
100
150
-150
-100
>0 0
Summer Oz
50
100
300.0
*!-§>~
~
100%
~
~ ~
HIH •*
100%
400.0
~
100%
»H[]h
~
^ 100%
| 500.0
#•
100%
~
~
HH
100%
J 700.0
0)
~~
100%
~~
h|h»
100%
to
(O
(1)
£ 850.0
~
HIH
# <
~ ~
100%
~
•HIH
~ ~
f 100 %
925.0
~ I
—[
IH
W
~
100%
~~
HIH
~~~ ~
10|%
1000.0
~
~ b
-OH
~
f0|%
H
\
H
100%
-150 -100
Autumn 12z
Spring 12z
^ 100%
~ ~
100%
f 100%
100%
100%
~h[]]-i«
~ ~
-50 0 5
~~ 65#%
0 100
~
4
H*
100%
100%
~
4H
~
100%
~ ~
~ ~
•hflH'"
10|%
~
~ ~ <
¦HIH'
~
1C^%
<
~ ~
100%
I—
—|
I
h
39.6 %
~ ~
~ ~
#
~4|>h
*
# *
HQh -
-150
-100
-50 0 i
Autumn Oz
50 100 15C
»4
100 %
~
¦HIH"
~
^ 100 %
I
\
100%
•-B- •
100 %
~ ~
' 1 ' *
~
^ 100%
KD-
H ~~
10|%
I—
h
1
39.6 %
-100 -50 0 50 100 150 -100 -50 0 50 100 -100 -50 0 50 100 150 -150 -100 -50 0 50 100 150
Wind Direction Error (Degs.) Wind Direction Error (Degs.) Wind Direction Error (Degs.) Wind Direction Error (Degs.)
Figure 98 PANT (Annette Island) upper-air distribution of wind direction error by pressure-level and season (dataset seasonal completeness indicated in blue).
83
-------
~ ~
700.0
~ ~ h
~ ~ h
H ~
H
850.0 #
925.0 ~ ~ ~
HD-<
Hffl-
m ~
1000.0
KI
-60 -40 -20 0 20 40
Summer 12z
m
~ ~~
~
ii-
-EH
HDh
-60 -40 -20 0 20 40
Relative Humidity Error (%)
~ ~ I
HI
—i
100%
~
~
4
-C
H-
^0%
~ ~ ~
~H
TT
I—>
^ 100%
i
>
~ h
¦
~
~ ~
~
««
'HEH
0 100 %
# ~ ~ I-
-n
H
100%
~ b
H IH
54.9 %
-60 -40 -20 0 20 40
Summer Oz
M—Of1 ¦
H ~
100%
. »Hhh ¦
»\ | | 1
>~ ~
~
100%
100%
~h
~ <
—m—
•'—UJi1
#
100%
10|%
"HIH
100%
HfflH
100%
~ ~ ~ ~~
Spring 12z
HIH
MEH
~ ~~
HBi1
T^PT
HD-i*
| ^ ^ 100%
1 OB %
~ I
~ h
-60 -40 -20 0 20 40 60
Autumn 12z
~ ~
~ ~ I-
HEH
H
¦
» 1
KH ¦
HQ
_|^ ^ 100 %
—I
• ~I
~ ~ 4
Spring Oz
-CD-
0}H
~ ~ ~'r1''
_J ^ | 100%
~ ~~
~ h
~
Oh
0 20
Autumn Oz
~ ~
I—
-m-
H ~ ~
—I ~
H ~
~ »b
"HH
¦OH
-40 -20 0 20 40
Relative Humidity Error (%)
-60 -40 -20 0 20 40
Relative Humidity Error {%)
-40 -20 0 20 40
Relative Humidity Error (%)
Figure 99 PANT (Annette Island) upper-air distribution of relative humidity error by pressure-level and season (dataset seasonal completeness indicated in
blue).
84
-------
4.8 Precipitation
Plots of monthly-averaged precipitation (in liquid-equivalent inches) determined from the 2016
WRF Alaska run are compared to NOAA-NCEP measured precipitation maps in this section.
First quarter precipitation comparisons for January, February, and March are plotted in Figure
100, Figure 101, and Figure 102, respectively. The precipitation patterns across the state
predicted by WRF match the NOAA-NCEP maps well. One notable difference is WRF predicts
considerably more precipitation along the east slopes of the Alaska range and western Cook Inlet
in January and February. WRF also tends to be a bit wetter in northwestern Alaska, more
specifically in the Brooks Range northeast of Kotzebue Sound.
Total Precipitation
January 2016
Figure 100. January 2016 monthly total precipitation, Alaska 9-km WRF (right) compared to NOAA-NCEP GHCN
dataset map (left). The area of significant difference along west of Cook Inlet is highlighted by the red box.
January 2016 WRF Total Precipitation
85
-------
Total Precipitation
February 2016
February 2016 WRF Total Precipitation
Figure 101. February 2016 monthly total precipitation, Alaska 9-km WRF (right) compared to NOAA-NCEP
GHCN dataset map (left). The area of significant difference west of Cook Inlet is highlighted by the red box.
Total Precipitation
March 2016
Figure 102. March 2016 monthly total precipitation, Alaska 9-km WRF (right) compared to NOAA-NCEP GHCN
dataset map (left).
Second quarter precipitation comparisons for April, May, and June are shown in Figure 103,
Figure 104, and Figure 105, respectively. Again, the overall precipitation patterns appear to be
well simulated by the WRF model. Notable differences include excessive precipitation over the
Alaska Range west of Cook Inlet in April, too little precipitation in the Chugach Range of
southeast Alaska and Togiak River valley of southwest Alaska in May, and too much
precipitation in the Wrangell-St. Elias mountains of southeast Alaska in June.
86
-------
Total Precipitation
April 2016
April 2016 WRF Total Precipitation
Figure 103. April 2016 monthly total precipitation, Alaska 9-km WRF (right) compared to NOAA-NCEP GHCN
dataset map (left). The area of significant difference west of Cook Inlet is higlilighted by the red box.
May 2016 WRF Total Precipitation
Total Precipitation
May 2016
Figure 104. May 2016 monthly total precipitation. Alaska 9-km WRF (right) compared to NOAA-NCEP GHCN
dataset map (left). Underprediction in Chugach Range and Toliiak Valley higlilighted by the red and blue boxes,
respectively.
87
-------
Figure 105. June 2016 monthly total precipitation, Alaska 9-kni WRF (right) compared to NOAA-NCEP GHCN
dataset map (left). Area of overprediction in the Wrangell-St. Elias National Park region highlighted by red box.
June 2016 WRF Total Precipitation
Total Precipitation
June 2016
Third quarter precipitation comparisons for July, August, and September are shown in Figure
106, Figure 107, and Figure 108, respectively. The summer patterns of precipitation appear to
be simulated well by WRF, A few notable differences include some excessive precipitation
across west-central Alaska in July and along the central Brooks Range in August. WRF also
appears to overpredict precipitation in Wrangell-St. Elias mountains of southwest Alaska in
September.
Total Precipitation
July 2016
July 2016 WRF Total Precipitation
Figure 106. July 2016 monthly total precipitation. Alaska 9-km WRF (right) compared to NOAA-NCEP GHCN
dataset map (left). Overpredicted precipitation in Wrangell-St. Elias range highlighted by red box.
88
-------
August 2016 WRF Total Precipitation
Total Precipitation
August 2016
0*
Figure 107. August 2016 monthly total precipitation, Alaska 9-km WRF (right) compared to NOAA-NCEP GHCN
dataset map (left). Region of overprediction highlighted by red box in the central Brooks Range.
September 2016 WRF Total Precipitation
Total Precipitation
September 2016
Figure 108. September 2016 monthly total precipitation. Alaska 9-km WRF (right) compared to NOAA-NCEP
GHCN dataset map (left).
Fourth quarter precipitation, represented by the October, November, and December plots, is
shown in Figure 109, Figure 110, and Figure 111, respectively. Again, the precipitation
patterns predicted by WRF appear to match the observation-based maps well. It appears WRF
may underpredict precipitation over the Chugach mountains in October. November and
December patterns match the observation-based maps very well with no notable regional
differences.
89
-------
Total Precipitation
October 2016
Figure 109. October 2016 monthly total precipitation, Alaska 9-km WRF (right) compared to NOAA-NCEP GHCN
dataset map (left). Area of underprediction, in the Cliugach Mountains area, highlighted by red box.
Total Precipitation
November 2016
Figure 110. November 2016 monthly total precipitation, Alaska 9-km WRF (right) compared to NOAA-NCEP
GHCN dataset map (left).
October 2016 WRF Total Precipitation
November 2016 WRF Total Precipitation
90
-------
Total Precipitation
December 2016
4.0 6.0 8.0 10.0 12.0 15.0 20.0
Inches
Figure 111. December 2016 monthly total precipitation, Alaska 9-km WRF (right) compared to NOAA-NCEP
GHCN dataset map (left).
5. Conclusions
The EPA's 2016 Alaska WRF model 9 km dataset was evaluated in this report. WRF dataset
hourly meteorological parameters were extracted at grid points nearest to the locations of surface
and upper-air meteorological measurement stations, for comparison to measurements. Tools such
as bias/error soccer-plots, parameter time series, and wind roses were used to assess the
performance of surface meteorological parameters such as temperature, wind speed, wind
direction, and humidity. Also, state-wide monthly precipitation maps were developed, using the
WRF dataset, for qualitative comparison to observation-based precipitation maps.
The results demonstrate WRF error and bias vary by location, and the error and bias of the WRF
model will need to be critically evaluated on a project-specific basis. This work was not intended
as an investigative study, so no efforts were made to examine the causes of model error and bias
in depth, although this could be a part of future project-specific evaluations.
Significant temperature biases along the north and northwest coasts of Alaska are likely due to
the limits of the standard WRF model regarding the simulation of complex surface energy
balances in the vicinity of sea-ice and tundra. These biases could potentially be corrected by use
of parameterization schemes optimized for arctic conditions. The Polar WRF model, developed
and maintained by the Byrd Polar Climate Research Center at Ohio State University (Byrd Polar
and Climate Research Center, 2019), provides an alternative model parameterization that has
been shown to improve WRF performance in Arctic regions (Brashers, et al., 2015).
Application. Observational nudging may also serve to improve model predictions. Use of more
layers in the vertical grid in some regions may also effectively improve performance, especially
December 2016 WRF Total Precipitation
%
91
-------
in the central and northern portions of Alaska impacted by strong winter inversions at the
surface.
Future modeling efforts may focus on adoption of optimized schemes, methods, and physics
modules to produce annual WRF datasets with less winter-time biases over the North Slope and
interior of Alaska. The EPA may apply optimized schemes, such as those used in Polar WRF, as
part of the optimization strategy.
The analysis products and plots contained in this report can be used by air permitting and other
authorities in project-specific model performance evaluations to determine if the dataset is
appropriate for regulatory use. Notably, this report itself does not represent an EPA endorsement
or validation of the dataset. Any use of the dataset for regulatory purposes (such as those
specified under 40 CFR Part 51, Appendix W) will require a project-specific performance
evaluation. However, results from this report can be used as part of any project-specific MPE.
92
-------
References
Appel, K., et al. 2011. Overview of the atmospheric model evaluation tool (AMET) v. 1.1 for
evaluating meteorological and air quality models, s.l. : Environmental Modelling & Software,
Vol. 26, Iss. 4, pp. 434-443, 2011.
Bowden, J., Talgo, K. and Adelman, Z. 2015. Western Air Quality Modeling Study WRF
Meteorological Model Winter Modeling Application/Evaluation. Chapel Hill, NC : UNC
Institute for the Environment, 2015.
—. 2016. Western-States Air Quality Modeling Study — WRF 2014 Meteorological Model
Evaluation. Chapel Hill, NC : Prepared for Wester Regional Air Partnership by UNC Chapel Hill
Institute for the Environment, 2016.
Brashers, B., et al. 2015. Arctic Air Quality Modeling Study — Meteorological Model
Performance Evaluation: 2009-2013 BOEMArctic WRFDataset. s.l. : Prepared by Ramboll
Environ for U.S. Dept. Interior, BOEM. OCS Study BOEM 2015-049., 2015.
Brown, D. 2014. WRF Meteorological Modeling in Support of Regional Air Quality Modeling
for the 2011 Base Year. s.l. : Iowa Dept. of Natural Resources, 2014.
Byrd Polar and Climate Research Center. 2019. Byrd Polar and Climate Research Center.
[Online] Ohio State University, 2019. http://polarmet.osu.edu/PWRF/.
Emery, C., Tai, E. and Yarwood, G. 2001. Enhanced Meteorological Modeling and
Performance Evaluation for Two Texas Ozone Episodes. Novato, CA : ENVIRON Int. Corp.,
2001.
EPA, U.S. 2019. EPA CMAQ Atmospheric Model Evaluation Tool homepage. U.S. EPA.
[Online] 2019. https://www.epa.gov/cmaq/atmospheric-model-evaluation-tool.
—. 2018. Modeling Guidance for Demonstrating Air Quality Goals for Ozone, PM2.5, and
Regional Haze. Research Triangle Park, NC : U.S. EPA Office of Air Quality Planning and
Standards, Air Quality Assessment Division, 2018.
Kemball-Cook, S., et al. 2005. Alaska MM5 Modeling for the 2002 Annual Period to Support
Visibility Modeling. Novato, CA : ENVIRON Int. Corp., 2005.
NCAR. 2019. Weather Reseach and Forecasting Model. National Center for Atmospheric
Research. [Online] August 2019. https://www.mmm.ucar.edu/weather-research-and-forecasting-
model.
—. 2019. WRF ARW Model Users Page. WRF Users Page. [Online] 2019.
http://www2.mmm.ucar.edu/wrf/users/.
93
-------
NOAA-NCEP. 2019. Climate Monitoring National Temperature and Precipitation Maps.
[Online] 2019. https://www.ncdc.noaa.gov/temp-and-precip/us-maps/.
Ramboll-Environ. 2015. Allegheny County Health Dept. S02 SIP WRFModel Performance
Evaluation. Novato, CA : Prepared for Allegheny County Health Dept., 2015.
Skamarock, W.C., Klemp, J.B., Dudhia, J., Gill, D.O., Barker, D.M., Duda, M.G., Huang,
X., Wang, W., Powers, J.G. 2008. A Description of the Advanced Research WRF Version 3.
2008.
Tesche, T. and McNally, D. 2001. Evaluation of CAMx andModels-3/CMAQ over the Lower
Lake Michigan Region with Inputs from the RAMS3c andMM5 Models. Ft. Wright, KY :
Prepared for the Coordinating Research Council by Alpine Geophysics, LLC, 2001.
Vose, R.S., et al. 2014. Improved historical temperature and precipitation time series for U.S.
climate divisions, s.l. : Journal of Applied Meteorology and Climatology, 2014.
94
-------
Appendix A: Station Performance Plots
[Available in electronic format, upon request]
[Please contact: Jav McAlpine at MCALPINI ftEPA.GOV!
95
-------
United States Office of Air Quality Planning and Standards Publication No. EPA-454/R-20-003
Environmental Protection Air Quality Assessment Division May 2020
Agency Research Triangle Park, NC
------- |