WRF Accuracy With GFS PDF

ARL-TR-7782 ● SEP 2016
US Army Research Laboratory
Change in Weather Research and Forecasting

(WRF) Model Accuracy with Age of Input Data
from the Global Forecast System (GFS)
by JL Cogan
Approved for public release; distribution unlimited.

NOTICES
Disclaimers
The findings in this report are not to be construed as an official Department of the
Army position unless so designated by other authorized documents.
Citation of manufacturer’s or trade names does not constitute an official

endorsement or approval of the use thereof.
Destroy this report when it is no longer needed. Do not return it to the originator.
ARL-TR-7782 ● SEP 2016
Change in Weather Research and Forecasting

(WRF) Model Accuracy with Age of Input Data
from the Global Forecast System (GFS)
by JL Cogan
Computational and Information Sciences Directorate, ARL

Form Approved
REPORT DOCUMENTATION PAGE OMB No. 0704-0188
Public reporting burden for this collection of information is estimated to average 1 hour per response, including the time for reviewing instructions, searching existing data sources, gathering and maintaining the
data needed, and completing and reviewing the collection information. Send comments regarding this burden estimate or any other aspect of this collection of information, including suggestions for reducing the
burden, to Department of Defense, Washington Headquarters Services, Directorate for Information Operations and Reports (0704-0188), 1215 Jefferson Davis Highway, Suite 1204, Arlington, VA 22202-4302.
Respondents should be aware that notwithstanding any other provision of law, no person shall be subject to any penalty for failing to comply with a collection of information if it does not display a currently valid
OMB control number.
PLEASE DO NOT RETURN YOUR FORM TO THE ABOVE ADDRESS.
1. REPORT DATE (DD-MM-YYYY) 2. REPORT TYPE 3. DATES COVERED (From - To)
September 2016 Technical Report April 2016–July 2016
4. TITLE AND SUBTITLE 5a. CONTRACT NUMBER
Change in Weather Research and Forecasting (WRF) Model Accuracy with Age
of Input Data from the Global Forecast System (GFS) 5b. GRANT NUMBER
5c. PROGRAM ELEMENT NUMBER
6. AUTHOR(S) 5d. PROJECT NUMBER

JL Cogan
5e. TASK NUMBER
5f. WORK UNIT NUMBER
7. PERFORMING ORGANIZATION NAME(S) AND ADDRESS(ES) 8. PERFORMING ORGANIZATION REPORT NUMBER

ATTN: RDRL-CIE ARL-TR-7782
2800 Powder Mill Road
Adelphi, MD 20783-1138
9. SPONSORING/MONITORING AGENCY NAME(S) AND ADDRESS(ES) 10. SPONSOR/MONITOR'S ACRONYM(S)
11. SPONSOR/MONITOR'S REPORT NUMBER(S)
12. DISTRIBUTION/AVAILABILITY STATEMENT

13. SUPPLEMENTARY NOTES
14. ABSTRACT
Mesoscale models generate output for many applications worldwide. Normally, initialization of a mesoscale model employs
the most recent output available from a larger-scale regional or global model. Such output also is used to adjust the lateral
boundary conditions as the larger-scale model evolves. However, some applications, especially in remote or minimally
connected areas, may require the use of large-scale model data that may be one or more days old. The analysis of this report
investigates the effect of using older large-scale data on mesoscale model accuracy. The report briefly describes the procedure
and presents the results of a preliminary analysis. As expected, accuracy generally tended to decline as the large-scale data
aged, but appeared to improve slightly as the age of the large-scale data increased from 3 days old to 4 days old. Also, there is
a wide variation between individual cases. The change in meteorological variables is explored, but the emphasis is on the
effect on simulated artillery trajectories. The use of trajectories with their major dependence on meteorological conditions
provides a means to check the net overall ability of a model to simulate the atmosphere over the space and time covered by the
trajectory.
15. SUBJECT TERMS
WRF, Weather Research and Forecasting, accuracy, aged GFS on WRF accuracy, Global Forecast System
17. LIMITATION 18. NUMBER 19a. NAME OF RESPONSIBLE PERSON
16. SECURITY CLASSIFICATION OF: OF OF
James L Cogan
ABSTRACT PAGES
a. REPORT b. ABSTRACT c. THIS PAGE 19b. TELEPHONE NUMBER (Include area code)
Unclassified Unclassified Unclassified UU 44 301-394-2304
Standard Form 298 (Rev. 8/98)
Prescribed by ANSI Std. Z39.18
ii
Contents
List of Figures iv
List of Tables v
1. Introduction 1
2. Data 3
3. Procedure 6
4. Initial Results: Meteorological 10
5. Initial Results: Trajectories 17
6. Conclusion 21
7. References 23
Appendix. Individual Mean and Median Radial Miss Distances (RMDs)

per Site 25
List of Symbols, Abbreviations, and Acronyms 35
Distribution List 36
iii
List of Figures
Fig. 1 Scheme employed to obtain GFS data for use with WRF. The circles at
the left end of the horizontal lines indicate the time of start of the GFS
data for each day. ...................................................................................7
Fig. 2 Graphs showing the 12-h statistics for differences in vector wind speed
(kn), virtual temperature (K), and density (gm–3) ................................12
Fig. 6 Graphs showing the 108-h statistics for differences in vector wind
speed (kn), virtual temperature (K), and density (gm–3) ......................14
Fig. 7 RMSD differences between the density, virtual temperature, and
pressure comparisons for the times shown on the charts. The statistics
for the METCMs derived from the 36-, 60-, 84-, and 108-h WRF runs
compared to those from the 12-h run are shown for each layer (zones
1–26) plus the surface (zone 0). The midpoints of the layers are in
meters AGL. .........................................................................................16
Fig. 8 RMSD differences between the vector wind speed, wind speed, and
wind direction comparisons for the times shown on the charts. The
statistics for the METCMs derived from the 36-, 60-, 84-, and 108-h
WRF runs compared to those from the 12-h run are shown for each
layer (zones 1–26) plus the surface (zone 0). The midpoints of the
layers are in meters AGL. ....................................................................16
iv
List of Tables
Table 1 RAOB sites and WMO identifiers, when available, are listed;
otherwise, the first 3 or 4 letters of the site’s name are used (e.g.,
MEIN for Meiningen). Each site has the respective dates and times for
a total of 50 cases. Latitude and longitude of the center of each WRF
domain used for the listed region are shown. Here Midwest applies to
several sites within the region often called the “southern plains”. East
Coast refers to locations along or near the center of the eastern coast of
the United States. US regions are listed first followed by several
international regions...............................................................................5
Table 2 Statistics for wind speed (kn) comparisons where 72- to 96-h GFS data
were used for the WRF calculations and compared to the respective
lines derived from the coincident RAOBs. WRF-based soundings were
obtained for 84 h after the start time of the GFS (0-h GFS forecast).
The heights shown have units of meters and are zone midpoints except
line 0, which is the surface. ..................................................................11
Table 3 Wind speed (kn) differences between comparisons using 96- to 120-h
GFS data (108-h WRF run) minus those using 0- to 24-h GFS data
(12-h WRF run). Comparisons using WRF-based soundings for 12 h
after the start time of the GFS (0-h GFS forecast) were subtracted from
comparisons for 108 h after the start time. The heights shown have
units of meters and are zone midpoints except line 0 which is the
surface. .................................................................................................15
Table 4 Means, medians, and standard deviations (Std Dev) over the 50 cases
of mean RMDs over the 4 firing directions in terms of meters and %
RD for the 5 WRF times from start of GFS (0-h GFS forecast) ..........18
Table 5 Means, medians, and Std Dev over the 50 cases of median RMDs over
the 4 firing directions in terms of meters and % RD for the 5 WRF
times from start of GFS (0-h GFS forecast).........................................18
Table 6 Number of occasions for each WRF time from start of GFS where
there was a minimum or maximum mean or median value of the RMD,
where a single mean or median is for the 4 directions of fire of each
case and WRF time ..............................................................................19
Table 7 Minimum and maximum mean RMDs for each WRF time (or GFS
data age) category. Minimum and maximum values shown on the top
and bottom lines, respectively, with values in meters on the left and %
RD on the right.....................................................................................19
Table 8 Minimum and maximum median RMDs for each WRF time (or GFS
data age) category. Minimum and maximum values shown on the top
and bottom lines, respectively, with values in meters on the left and %
RD on the right.....................................................................................19
Table 9 Number of cases where the mean or median RMD of the 4 azimuth
directions exceeded 3.5%.....................................................................20
v
Table A-1 Individual case values of mean RMDs for each WRF time from start of
GFS. The region (e.g., Alaska) and comparison RAOB (R) dates and
times are shown, and the site codes are listed in the left most column.
WMO notations are used when available, otherwise the first 3 or 4
letters of the site’s name are used (e.g., MEIN for Meiningen). The
values from Tables 7 and 4 are listed at the bottom. ...........................27
Table A-2 Individual case values of median RMDs for each WRF time from start
of GFS. The region (e.g., Alaska) and comparison RAOB (R) dates
and times are shown, and the site codes are listed in the left most
column. WMO notations are used when available, otherwise the first 3
or 4 letters of the site’s name are used (e.g., MEIN for Meiningen).
The values from Tables 8 and 5 are listed at the bottom. ....................31
vi
1. Introduction
Mesoscale models, such as the Weather Research and Forecasting (WRF) model
(http://wrf-model.org/wrfadmin/publications.php), generate output for many
applications worldwide that include, but are not limited to, regional forecasts for
civilian and military operations and planning; research such as for aerosol
distribution and transport; and other uses for Government, agriculture, and industry.
Normally, initialization of a mesoscale model employs the most recent output
available from a larger-scale regional or global model, such as the National Centers
for Environmental Prediction’s Global Forecast System (GFS; http://www.ncdc.
noaa.gov/data-access/model-data/model-datasets/global-forcast-system-gfs) or the
US Air Force’s Global Air Land Weather Exploitation Model, a version of the
United Kingdom’s Unified Model (http://www.metoffice.gov.uk/research/
modelling-systems/unified-model).
The GFS, or another large-scale model, is also used to adjust the lateral boundary
conditions during the forecast period of the mesoscale model as the larger-scale
model evolves over time. However, some applications, especially in remote or other
areas with minimal or no connectivity to sources of data, may require the use of
large-scale model data that may be one or more days old. First responders to an
emergency in a remote area may not have sufficient access to the Internet or other
means of obtaining large-scale model data. First-in military forces may not have
access to reachback for several days or severe limitations may result in
communication of only the very highest priority messages.
Some meteorological centers compare global and large-scale regional model

meteorological output versus radiosonde observation (RAOB) data more or less
continuously and some make intermodel comparisons. The National Center for
Environmental Prediction (NCEP) (http://www.emc.ncep.noaa.gov/gmb/
STATS_vsdb/) and the European Centre for Medium-Range Weather Forecasting
(ECMWF) (http://www.ecmwf.int/en/forecasts/charts/medium/monthly-wmo-
scores-against-radiosondes) have readily accessible websites. Worldwide
comparisons are available for deterministic forecasts at http://apps.ecmwf.int/
wmolcdnv/ and for ensemble forecasts at the Japan Meteorological Agency at
(http://epsv.kishou.go.jp/EPSv/). These sites provide comparisons of
meteorological variables at the standard World Meteorological Organization
(WMO) pressure levels (e.g., 850, 700, and 500 hPa), and the ECMWF site presents
graphs showing the change in global model accuracy with time starting at 24 h of
model time up to as much as 240 h for a subset of the models compared. Various
organizations display WRF output, especially in graphical formats. Examples
include the National Center for Atmospheric Research (NCAR) at http://wrf-
1
model.org/plots/realtime_main.php, NCEP at http://mag.ncep.noaa.gov/model-
guidance-model-area.php#, and the Earth System Research Laboratory (ESRL) of
the National Oceanic and Atmospheric Administration (NOAA) at http://
rapidrefresh.noaa.gov/. However, they do not have readily available sites showing
the change in accuracy of WRF output relative to observations as the large-scale
data age.
In addition, several investigators have compared WRF output against radiosonde
data as part of their experiments or evaluations. For example, Schroeder et al.
(2006) compared data from the older Mesoscale Model Fifth Generation over its
vertical extent and presented results for standard pressure levels from 850 to
100 hPa (some graphs to 150 hPa). They worked with data for 8 days in April 2002,
12 days in winter and summer 2003, and 18 days in August 2001. The former 2
periods were from the East Coast region of the United States and the latter from the
Great Plains region of the United States. Kilpelainen et al. (2012) and Dutsch
(2012) each evaluated WRF output for the boundary layer over Svalbard in the
Arctic in terms of height above ground compared to tower and tethered balloon
(tethersonde) data and radiosonde data, respectively. The tethersondes provided
data from the surface up to about 600, 800, and 1,250 m for each of 3 sites,
respectively. The radiosonde data reached as high as 2 km above the surface.
However, none of these investigations apparently considered the effect of the use
of older GFS or other large-scale model data on the output of WRF or other
mesoscale model.
The analysis in the present report investigated the effect on the accuracy of WRF
as a representative mesoscale model when using older large-scale data (i.e., from
GFS). Skamarock et al. (2008) describe the basic features of WRF (although the
model has evolved since). The present report briefly describes the analysis
procedure and presents preliminary results. As expected, accuracy generally tended
to decline as the large-scale data aged, but appeared to slightly increase from 3 days
old to 4 days old. Also, there was a wide variation in accuracy between individual
cases. The change in the meteorological variables was examined, but the primary
emphasis concerned the effect on the radial miss distances (RMDs) derived from
simulated artillery trajectories computed using the General Trajectory (GTRAJ)
model from the US Army Armaments Research Development and Engineering
Center (ARDEC). The use of trajectories, with their major dependence on
meteorological conditions, provides a means to check the net overall ability of a
model to simulate the atmosphere as an integrated whole over the space and time
covered by the trajectory. The Computer Meteorological Message (METCM),
commonly used for accounting for meteorological effects in artillery trajectory
calculations, provided the meteorological data required by GTRAJ.
2
2. Data
The METCM consists of meteorological data (i.e., pressure, virtual temperature,

wind speed, and wind direction) for the surface and a series of vertical layers,
extending up to 30 km above ground level (AGL), also known as METCM zones.
STANAG 4061 (2000) describes the METCM in greater detail. For the current
study, METCMs were derived from both WRF-based profiles and RAOB
soundings. The generation of vertical profiles of meteorological variables for the
METCM is based on algorithms implemented in software described in Cogan
(2015). That software package processes either model data or RAOB soundings
into a variety of message type including the METCM. Vertical profiles were
extracted from WRF Network Common Data Form output files via a NCAR
Command Language (NCL) script using a method developed by Reen (2015). The
profiles were placed into text files that included a header with information on the
location, time, model grid resolution, and method (if any) employed to interpolate
between model grid points to the selected location. These profiles contain data lines
for heights above mean sea level (MSL) of pressure levels. The user has the option
to use a “sounding” from the nearest grid point (i.e., no interpolation), or either
bilinear interpolation or inverse distance weight interpolation (via built-in NCL
capabilities). RAOB data were downloaded from the University of Wyoming’s
weather website (http://www.weather.uwyo.edu/upperair/sounding.html). An
alternative site is the RAOB archive hosted by the NOAA ESRL at their website
(http://www.esrl.noaa.gov/raobs/).
WRF v3.7.1 was run with 9-, 3-, and 1-km horizontal grid spacing nested domains.
The comparisons for this report used data from the 3-km domain. The initial and
boundary conditions were derived from GFS 0.5° horizontal grid spacing with a
3-h time interval. Where available, GFS snow fields were replaced with 1-km snow
fields from the National Weather Service’s National Operational Hydrologic
Remote Sensing Center (NOHRSC) (http://www.nohrsc.noaa.gov/technology/)
Snow Data Assimilation System, or if not available, with 4-km snow cover fields
from the National Ice Center’s Interactive Multi-sensor Snow and Ice Mapping
System (IMS) (http://nsidc.org/data/docs/noaa/ g02156_ims_snow_ice_analysis/).
Consequently, when available, NOHRSC fields were used for an area centered on
the United States, IMS data for the rest of the northern hemisphere, and GFS fields
for the southern hemisphere. A sea surface temperature product with higher
resolution than the GFS output is produced by NCEP’s Marine Modeling and
Analysis Branch, called the Real-Time Global Sea Surface Temperature (Gemmill
et al. 2007), which has 1/12th-degree horizontal grid spacing and was used to
specify sea surface temperatures. These WRF runs did not use data assimilation
(i.e., observation “nudging”) in order to simulate the conditions that could occur if
3
operating in a remote area without adequate connection to a source of external data.
If we assume observation data were available at the start of the initial period (e.g.,
prior to leaving a home location), but not afterwards, any noticeable effect on the
model calculations would have ended prior to the 12-h forecast of the first run. The
Mellor-Yamada-Janjić scheme is used to parameterize the atmospheric boundary
layer. As in Lee et al. (2012) and Reen et al. (2014), the background turbulent
kinetic energy is decreased to better simulate conditions with low turbulent kinetic
energy and the atmospheric boundary layer depth diagnosis is altered. The WRF
single-moment, 5-class microphysics parameterization and the Kain-Fritsch
cumulus parameterization (9-km domain only) are used. For radiation, the Rapid
Radiative Transfer Model is used for longwave and the Dudhia scheme for
shortwave. The Noah land surface model is used to represent land surface
processes.
The selected RAOB sites cover various regions and seasons in the northern and
southern hemispheres. Table 1 presents a list of regions with the included WMO
sites and the date and time of the WMO RAOBs. Analysis locations were selected
using locations from previous investigations and tests, which helped ensure
availability of data from WMO RAOB sites. To reduce the number of model runs
and reduce analysis time, groups of WMO sites were sought that fit within the
bounds of the middle WRF nest or domain. The horizontal grid resolution of the
middle nest (3 km) is similar to current and near-term planned resolutions used in
operational meteorological systems. For some regions, up to 5 sites fit within the
bounds of the 3-km domain (723 x 723 km), allowing for sufficient distance from
the lateral boundaries to avoid unwanted effects that potentially could arise from
the different grid resolutions of the outer and middle domains (9 and 3 km). That
distance is generally considered to be on the order of 20 to 30 (or more) grid points,
or, in this case, about 60 to 90 km. That leaves a central region no less than about
540 × 540 km (~180 × 180 grid points). The exception was San Diego, California,
which was only about 10 to 11 grid points from the edge of the domain. However,
the results seemed in line with, or even better, than from the other sites within the
“Southwest” region. In some regions, as in the southern hemisphere, only 2 sites fit
within the middle domain. Data were extracted for all seasons but mostly for winter
and spring and mostly for regions within the United States. A total of 50 cases were
examined where each case contains comparisons between a METCM derived from
a RAOB with 5 METCMs derived from the 5 runs of WRF (one for each GFS “age”
used for initialization).
4
Table 1 RAOB sites and WMO identifiers, when available, are listed; otherwise, the first
3 or 4 letters of the site’s name are used (e.g., MEIN for Meiningen). Each site has the
respective dates and times for a total of 50 cases. Latitude and longitude of the center of each
WRF domain used for the listed region are shown. Here Midwest applies to several sites within
the region often called the “southern plains”. East Coast refers to locations along or near the
center of the eastern coast of the United States. US regions are listed first followed by several
international regions.
Date and Time Center Lat, Lon

Region Site ID
(yyyy-mm-dd-hh) (degrees)
Alaska US Anchorage PANC 2016-01-11-00 59.46, –153.34

King Salmon PAKN and
Kodiak PADQ 2016-04-06-12
East Coast US Blacksburg, VA RNK 2016-03-23-00 36.88, –77.94

Dulles Int'l AP, VA IAD and
Greensboro, NC GSO 2016-05-19-12
Newport, NC MHX
Wallops Is., VA WAL
Midwest US Amarillo, TX AMA 2016-01-02-12 37.12, –98.66

Dodge City, KS DDC and
Lamont, OK LMN 2016-02-10-12
Norman, OK OUN
Topeka, KS TOP
Southwest US Flagstaff, AZ FGZ 2016-06-18-12 33.78, –113.57

Las Vegas, NV VEF
Phoenix, AZ PHX
San Diego, CA NKX
Tucson, AZ TUS
Australia Adelaide YPAD 2016-05-05-12 –36.05, 141.69

Melbourne YMML
Germany Bergan ETGB 2016-02-07-00 50.82, 9.62

Idar-Oberstein ETGI and
Kuemmersbruck ETGK 2016-03-07-00
Meiningen MEIN
Stuttgart STUT
5
Table 1 RAOB sites and WMO identifiers (continued)
Date and Time Center Lat, Lon

Region Site ID
(yyyy-mm-dd-hh) (degrees)
South Africa King Shaka FALE 2016-05-06-00 –27.76, 29.66
Pretoria FAIR
South Korea Cheju CHEJ 2015-11-07-12 35.19, 127.42

Heuksando HEUK
Kwangju RKJJ
Osan Air Base RKSO
Pohang POHA
Though a rare occurrence, the main NOAA archive used for the 0.5o data
(https://catalog.data.gov/dataset/global-forecast-system-gfs-0-5-deg) occasionally
omits some GFS output. Unfortunately, one such event meant that Day–4 GFS data
were not available for the first of 2 Midwest US groups listed in Table 1
(2016-01-02-12). However, the data were available via an alternate website that
contains GFS output for 0.5° and 1.0° grids, (http://www.ncdc.noaa.gov/
has/HAS.FileAppRouter?datasetname=GFS3&subqueryby=STATION&applnam
e=&outdest=FILE).
3. Procedure
The WRF was run to produce 12-h forecasts that ended at the time of the respective
coincident RAOB. It was initialized using 0.5° GFS data starting with the GFS 0-h
forecast and working backwards, and the GFS data were used to update the WRF
boundary conditions. Therefore, the WRF 0-h forecast time was the same as the
GFS time for a 0-, 24-, 48-, 72-, or 96-h forecast from the same day, 1, 2, 3, or 4
days earlier, respectively. For example, for a 12 Coordinated Universal Time
(UTC) RAOB on 25 March, GFS forecast data were obtained every 3 h from the
00 UTC, 25 March data set. In this example, the WRF-based sounding was
extracted from the 12-h WRF data so as to match the RAOB time. This procedure
was repeated using GFS data from progressively older GFS runs one day apart. For
example, the 2-day-old GFS data started on 00 UTC, 23 March. WRF would be run
again for 00 UTC, 25 March, but using data from the 2-day-old GFS data set to
produce the sounding 12 model hours later than was used to generate the respective
METCM. Consequently, WRF soundings were extracted to compare with the same
RAOB, but initialized using GFS data that began 0, 24, 48, 72, and 96 h earlier.
Figure 1 illustrates the scheme employed to obtain the “aged” GFS data. In the
figure, the circles at the left of each line indicate the GFS start time. The WRF start
6
times relative to the GFS start time (0-h forecast) are shown along the solid vertical
line (left), and the times of the WRF 12-h forecast and coincident RAOB are shown
along the vertical dashed line (right).
Fig. 1 Scheme employed to obtain GFS data for use with WRF. The circles at the left end
of the horizontal lines indicate the time of start of the GFS data for each day.
This method uses successive GFS data sets as input to the WRF model forecasts
used to produce soundings contemporaneous to an actual RAOB used as an
approximation to truth.
Use of this method to obtain the older GFS data eliminated the potential effect of
changes in meteorological conditions that would occur if a single GFS data set was
used along with sequential RAOBs. An example is a set of GFS files for
2016032500 with WRF soundings and RAOBs at 12 UTC on 25, 26, 27, 28, and
29 March. During that period of 5 days, the RAOBs would reflect the actual
changes in the weather and consequently could be confused with changes in the
WRF output arising from the use of older GFS data. Therefore, it would be difficult
to separate the effect of the aging GFS from the effect of the actual change in the
weather.
The next step was the preprocessing on a local computer of the GFS data for the
selected period and location using scripts developed by Reen (2015) that included
modifications that corrected some deficiencies in the original code. This
preprocessing step converts the GFS and other data (including observations when
available with data assimilation activated) into forms appropriate for input to the
WRF model itself. The resultant preprocessed files were transferred to a high-
performance computer (HPC) since the HPC could run multiple instances of WRF
in a much shorter time. The WRF output files for the 3-km grid domain were
7
transferred back to the local computer for generation of the meteorological profiles.
Vertical profiles were extracted from the WRF output files via an NCL script. The
RAOB and WRF profiles were placed into text files and transferred to a local
Windows desktop for conversion into text file “messages” that have the data in the
same units and structure as in the METCM. Cogan (2015) describe the program for
conversion of RAOB or WRF soundings into METCMs and various other types of
meteorological messages. This program also produces a form of the METCM that
is suitable for input to the GTRAJ trajectory simulation program. That format also
is convenient for use in spreadsheets employed for the meteorological comparisons
of this report.
The meteorological values of the METCM, that is, pressure, virtual temperature,
wind speed, and wind direction, plus some derived variables such as density, were
compared using spreadsheets similar to those in Cogan (2015) for METCM zones
up through zone 26 (19–20 km), which is the highest zone covered by the WRF-
based output, except where either the WRF- or RAOB-based sounding ended at a
height lower than 20 km AGL. In that situation the comparison ended at the
maximum level covered by both soundings. The mean difference (MD), the mean
absolute difference (MAD, standard deviation of the differences (SD), and the root
mean square difference (RMSD) between the WRF-generated METCMs and ones
from the corresponding RAOBs were computed for each zone for all cases for each
age of the GFS data, that is, each RAOB is compared to 5 WRF outputs generated
using current (day 0), 1-day-old, …, 4-day-old GFS data sets. These terms for the
several statistics replaced the more common mean, mean absolute error, standard
deviation, and root mean square error, because the WRF-based METCMs were
compared with those from RAOB data, not the actual atmosphere. While good
approximations, RAOBs are subject to errors arising from, for example, instrument
error, balloon drift, and differences from the nominal sounding time.
The differences in those variables for each “age” of GFS were compared in another
spreadsheet. Tables were constructed where the values for each variable and zone
were compared between various “ages” of GFS. For example, the differences were
computed between temperature statistics for WRF-based profiles that used
96–120 h GFS data (RAOB 108 h from start of the GFS data) and WRF-based
profiles that used current GFS data (0–24 h, RAOB 12 h from start of the GFS
data). Data produced from more recent GFS values were subtracted from the older
ones. As a result, positive numbers indicated better values for MAD, SD, and
RMSD from the newer GFS data. The MD values may or may not indicate better
or worse values. For example, subtracting a small positive or negative value from
a larger negative value would result in negative value that would suggest the older
GFS was better. The MAD normally is a better indicator of how close the WRF-
8
based METCMs are to the RAOB-based ones. Tabular and graphical depictions
were prepared and several examples are shown in Section 4.
The GTRAJ (v3.9.7) program was provided by ARDEC and is considered a

standard for trajectory calculations. As noted in the GTRAJ user’s guide, it is a
trajectory simulation program that may be run interactively or scripted. The
program uses the point mass or modified point mass equations of motion to simulate
the trajectory of a projectile in flight. The program uses a database to obtain
aerodynamics and ballistics for the selected projectile. There is also aerodynamic
data available for miscellaneous shapes such as fragments, cylinders, and cubes.
The user may elect to use a custom database instead of the standard one. Frehlich,
et al. (2008) describe an earlier version of GTRAJ they employed in their study of
the effect of turbulence on ballistic testing procedures.
For each case, the input to GTRAJ includes the METCM from the RAOB; the 5
METCMs from WRF; and the elevation (MSL) of the RAOB site or from the WRF
terrain database; latitude (RAOB and WRF have the same value in this study); and
firing information such as azimuth (direction) of fire, cannon type, projectile type,
and so on. The cannon was a standard 155-mm system with a commonly used
projectile fired at an elevation that would produce ranges on the order of 22–24 km
with apogees around 7.5 to 8.5 km AGL. The radial distance (RD) is then computed
for each instance (e.g., WRF at one GFS “age” for one comparison time for one site
for one azimuth). The RD is the distance from the gun to the target, which may be
computed from the range and deflection where RD is the square root of the sum of
the squares of the range and the deflection. In order to at least partially account for
the variation in RD with azimuth, the simulated firings were run for the 4 cardinal
directions (north, east, south, and west). For example, wind and density effects
could assist or counteract one another leading to larger or smaller RDs for a given
firing azimuth.
The output RDs, ranges, and deflections from GTRAJ for the RAOB and the 5
WRF runs (1 for each age of GFS) for the 4 directions of fire are copied into a
spreadsheet. Then the values from the runs that used RAOB-based input are
compared to those from the runs with WRF-based input to obtain the differences in
range and deflection, which are used to calculate RMD. RMD is the square root of
the squares of the differences (∆D) in range and deflection:
RMD = ((∆Dr)2 + (∆Dd)2)1/2, (1)
where the subscripts r and d refer to range and deflection, respectively. The mean
and median values of the RMDs over the 4 azimuths are computed for each of the
5 GFS ages in terms of meters and percent of the RD (also named % Range).
9
4. Initial Results: Meteorological
The first set of comparisons was made for the meteorological variables. The
emphasis was on the variables in the METCM as well as the derived variables of
density and vector wind speed. METCMs were computed from the RAOB and
WRF data for each WRF time from start of the GFS as described previously,
entered into spreadsheets where the differences were computed for each pair of
WRF and coincident RAOB-based soundings, and statistics computed for each of
the METCM layers and the surface (line or zone 0). One RAOB for each site and
time provided the comparison data for all of the WRF runs for that day and time.
For example, the METCM computed from the Wallops Island, Virginia, RAOB for
2016-05-19 at 12 UTC was compared to the WRF-based METCMs for that location
and time for all 5 WRF runs (WRF output for 12, 36, …, 108 h from start of the
respective GFS run). The sample size or number of pairs for each layer plus the
surface ranged from the maximum number of 50 to a lesser number at the highest
levels where either the WRF- or RAOB-based sounding failed to reach 20-km
AGL. Table 2 shows a sample for wind speed for WRF-based METCMs (i.e., from
the WRF simulations using 72- to 96-h GFS data). The WRF was run for 12 h after
the start of the selected period, that is, it began at 72 h of GFS model time from the
start of the GFS data for day –3.
10
Table 2 Statistics for wind speed (kn) comparisons where 72- to 96-h GFS data were used
for the WRF calculations and compared to the respective lines derived from the coincident
RAOBs. WRF-based soundings were obtained for 84 h after the start time of the GFS (0-h
GFS forecast). The heights shown have units of meters and are zone midpoints except line 0,
which is the surface.
Line Height Samples MD MAD SD RMSD

0 0 50 2.94 4.94 6.33 6.92
1 100 50 3.08 5.92 6.91 7.50
2 350 50 3.38 8.10 10.15 10.60
3 750 50 3.66 7.70 10.20 10.74
4 1,250 50 4.14 7.70 10.30 11.01
5 1,750 50 3.30 7.86 10.85 11.24
6 2,250 50 1.36 7.64 10.60 10.58
7 2,750 50 0.92 7.68 10.59 10.53
8 3,250 50 1.30 7.30 9.90 9.89
9 3,750 50 1.34 6.58 9.22 9.22
10 4,250 50 1.70 7.34 9.97 10.01
11 4,750 50 1.88 8.80 11.75 11.78
12 5,500 50 1.24 9.92 13.33 13.25
13 6,500 50 1.20 11.08 14.70 14.60
14 7,500 50 1.28 11.08 15.24 15.14
15 8,500 50 3.16 10.64 15.24 15.41
16 9,500 50 4.84 10.44 15.72 16.30
17 10,500 50 5.04 10.52 14.93 15.62
18 11,500 50 3.74 8.90 12.59 13.01
19 12,500 49 1.65 7.20 9.91 9.94
20 13,500 49 0.96 6.63 8.68 8.65
21 14,500 49 –0.27 5.24 6.80 6.73
22 15,500 46 –0.26 5.39 6.63 6.56
23 16,500 44 0.75 4.80 6.37 6.34
24 17,500 42 1.83 4.88 6.85 7.02
25 18,500 41 1.24 6.07 7.47 7.48
26 19,500 25 2.20 5.40 6.61 6.84
Tables similar to Table 2 were prepared for the other variables and for the other
WRF simulations or runs using GFS data with different forecast start times. WRF
simulations are referred to by the GFS forecast time at the verification time. For
example, the 108-h WRF run uses GFS data with a 0-h forecast 108 h prior to the
RAOB time (96-h prior to the WRF 0-h forecast). Figures 2–6 present graphical
representations of those WRF data using the 5 different GFS inputs for vector wind
speed, virtual temperature, and density.
11
Fig. 2 Graphs showing the 12-h statistics for differences in vector wind speed (kn), virtual
temperature (K), and density (gm–3)
12
13
Table 2 and Figures 2–6 present statistics for each METCM zone that show the
differences between METCMs derived from WRF using input from each of the
several “aged” GFS data sets and those derived from coincident RAOBs. The
following table and graphs have the same type of statistics that illustrate any
changes in forecast accuracy as compared to the initial forecast at 12 h from the
start of the GFS (0-h GFS forecast). The statistical values for the WRF runs were
compared for each variable for each METCM layer for each of the WRF model
runs. For example, virtual temperatures for the METCM layers plus the surface that
were derived from the 108-h WRF run were compared to the values for the
respective layers and surface from the 12-h run.
Table 3 presents a sample for wind speed where values from the 108-h WRF run
were compared to those from the 12-h run. The few negative MD values may or
may not indicate that the 108-h WRF occasionally led to better outcomes for the
reason given earlier, that is, subtracting a positive or negative value from a negative
value of a larger magnitude will lead to a negative number. Consequently, the MAD
is a better indicator as is the RMSD.
14
Table 3 Wind speed (kn) differences between comparisons using 96- to 120-h GFS data
(108-h WRF run) minus those using 0- to 24-h GFS data (12-h WRF run). Comparisons using
WRF-based soundings for 12 h after the start time of the GFS (0-h GFS forecast) were
subtracted from comparisons for 108 h after the start time. The heights shown have units of
meters and are zone midpoints except line 0 which is the surface.
Line Height Samples MD MAD SD RMSD

0 0 50 0.4 0.6 1.9 1.9
1 100 50 –1.1 0.7 2.1 1.5
2 350 50 –1.2 2.2 3.5 3.1
3 750 50 –0.6 3.3 4.5 4.4
4 1,250 50 0.4 3.1 2.9 2.9
5 1,750 50 2.2 3.3 3.6 3.9
6 2,250 50 1.6 3.6 6.0 6.1
7 2,750 50 1.1 3.9 6.2 6.2
8 3,250 50 1.5 4.1 5.6 5.7
9 3,750 50 1.8 3.7 5.4 5.6
10 4,250 50 1.8 4.0 5.3 5.5
11 4,750 50 1.8 4.0 5.4 5.5
12 5,500 50 1.7 4.3 5.7 5.8
13 6,500 50 1.3 5.1 7.2 7.2
14 7,500 50 0.3 5.8 8.1 8.0
15 8,500 50 1.1 6.8 10.5 10.5
16 9,500 50 2.0 8.0 9.5 9.6
17 10,500 50 2.3 7.5 8.5 8.8
18 11,500 50 0.6 5.7 7.5 7.2
19 12,500 49 –0.8 5.1 5.5 5.1
20 13,500 49 0.2 2.9 3.0 3.0
21 14,500 49 –0.2 1.8 2.1 2.0
22 15,500 46 –0.5 1.2 1.9 1.8
23 16,500 44 –1.1 0.9 1.2 1.1
24 17,500 42 1.3 0.7 0.9 1.1
25 18,500 41 1.1 1.8 2.4 2.5
26 19,500 25 0.7 1.8 1.2 1.4
As with the individual “GFS age” comparisons, the tables also may be presented in
graphical form. Figure 7 shows density and related RMSD values, and Fig. 8
presents wind RMSD values. In these figures, the forecast degrades if the values
for the older GFS data are larger than for the newer data. For example, the pressure
differences in Fig. 7 are greater for the 84-h WRF than for the 60-h WRF runs.
15
Fig. 7 RMSD differences between the density, virtual temperature, and pressure
comparisons for the times shown on the charts. The statistics for the METCMs derived from
the 36-, 60-, 84-, and 108-h WRF runs compared to those from the 12-h run are shown for
each layer (zones 1–26) plus the surface (zone 0). The midpoints of the layers are in meters
AGL.
Fig. 8 RMSD differences between the vector wind speed, wind speed, and wind direction
comparisons for the times shown on the charts. The statistics for the METCMs derived from
the 36-, 60-, 84-, and 108-h WRF runs compared to those from the 12-h run are shown for
each layer (zones 1–26) plus the surface (zone 0). The midpoints of the layers are in meters
AGL.
16
In general, the older the GFS input, the greater the RMSD “error” relative to the
12-h WRF-based METCMs. However, occasionally older GFS data appear to
provide values closer to the 12-h WRF-based METCMs. Most notably the vector
wind speed and the wind speed values for the 108-h WRF-based METCMs for more
than a few layers appear closer to the 12-h values than the 84-h values (10 and 15
of 27 zones, respectively). The wind direction differences vary widely though for
19 of 27 zones, including the surface, and the 84-h results are closer to the 12-h
values than the 108-h results. For several zones (4,750-, 5,500-, and 6,500-m AGL)
the 108-h results are closer than the 36-h wind directions.
5. Initial Results: Trajectories
GTRAJ was employed to compute trajectories using the METCMs from RAOBs
and the WRF runs with input from the 5 differently aged GFS data sets. Trajectories
computed with METCMs from WRF using GFS data for days 0 through minus 4
were compared with METCMs from coincident RAOBs, as discussed previously.
RMD values were computed independently for the 4 cardinal directions so as to
have an idea of likely variation with azimuth and consequently to mitigate possible
effects that could arise from selecting a single azimuth. A small RMD could occur
in one direction of fire if effects of a large density error compensated for a large
wind error, but the RMD would most likely be much larger for the opposite azimuth
where density and wind errors would reinforce their effects.
The RMDs computed using GTRAJ generally followed the accuracy (i.e.,
agreement to the coincident RAOBs) of the METCMs derived from the WRF
output. Nevertheless, there were major variations in RMD that did not seem to
follow the general trend. The following tables provide an indication of the overall
trends and the significant variation within those trends. The mean and median
values for each GFS age were computed for each site and time where one site and
time equals one case (e.g., Anchorage on 2016-04-06 at 12 UTC), where each case
has a mean and a median RMD computed from the RMDs for the 4 firing directions
for each GFS age. Then the means and medians of the mean and median RMDs of
each case were computed for all 50 cases for each GFS age. In addition, the standard
deviations of the mean and median values were computed for the 50 cases. The
standard deviations provide a measure of the variation within each GFS age
category.
Tables 4 and 5 present the means and standard deviations of the mean and median
RMDs, respectively, for all sites and times by age of GFS input in terms of meters
and percent of RD. The Appendix presents the case-by-case mean and median
RMDs. WRF was run for 12 h for the several GFS data sets as indicated in Fig. 1.
17
For example, WRF ran for 12 h after using the 0- to 12-h GFS forecast, 12 h using
the 24- to 36-h GFS forecast from the previous day, and so on through 12 h with
the 96- to 108-h GFS forecast from 4 days earlier. For each case, the 5 METCMs
derived from WRF were compared with the RAOB-based METCM at 12 h after
the 0-h WRF forecast (see Fig. 1).
Table 4 Means, medians, and standard deviations (Std Dev) over the 50 cases of mean
RMDs over the 4 firing directions in terms of meters and % RD for the 5 WRF times from
start of GFS (0-h GFS forecast)
WRF times from start of GFS data

12 h 36 h 60 h 84 h 108 h 12 h 36 h 60 h 84 h 108 h
Meters % RD
Mean 86.8 123.0 182.4 273.5 254.3 0.383 0.543 0.804 1.210 1.122
Median 76.1 103.2 155.7 219.6 184.5 0.334 0.443 0.690 0.971 0.815
Std Dev 53.6 75.3 110.3 207.6 200.7 0.239 0.339 0.492 0.921 0.898
Table 5 Means, medians, and Std Dev over the 50 cases of median RMDs over the 4 firing
directions in terms of meters and % RD for the 5 WRF times from start of GFS (0-h GFS
forecast)
WRF times from start of GFS data

12 h 36 h 60 h 84 h 108 h 12 h 36 h 60 h 84 h 108 h
Meters % RD
Mean 83.7 119.0 176.9 264.6 249.3 0.369 0.526 0.781 1.167 1.104
Median 72.7 98.7 153.9 202.4 179.2 0.311 0.418 0.681 0.907 0.785
Std Dev 52.4 74.2 107.4 200.5 197.1 0.232 0.335 0.482 0.889 0.889
As expected the mean and median values are similar, but not the same, and
increased as the GFS input aged from the 12-h through the 84-h computations.
However, the mean and median values decreased from the 84-h to the 108-h WRF
runs. Since the standard deviation gives a measure of the variation of the RMDs for
the 50 cases, the variation appears to peak around the 84-h output with a drop or at
least a leveling off afterward at around 108 h. This unexpected result appears to be
related to the wind speed and direction improvement noted in Fig. 8.
As suggested by the standard deviations, there is significant variation from case to

case in either mean or median RMD for the several WRF times. On a case-by-case
basis, the minimum mean and/or median RMD for an earlier time may exceed that
for a later time and vice versa for the maximum RMD. For example, the mean RMD
for the 36-h WRF-based METCM may be less than that for the 12-h METCM, or
the maximum mean RMD for all 5 times may occur for the 60-h METCM. Table 6
18
presents the number of cases where there was a minimum or maximum value of
mean or median RMD for each time (GFS period).
Table 6 Number of occasions for each WRF time from start of GFS where there was a
minimum or maximum mean or median value of the RMD, where a single mean or median is
for the 4 directions of fire of each case and WRF time
No. of minimums No. of maximums

12 h 36 h 60 h 84 h 108 h 12 h 36 h 60 h 84 h 108 h
Mean 25 14 8 3 0 0 3 7 23 17
Median 26 12 8 4 0 0 2 8 24 16
As shown in Table 6, the number of minimums or maximums of either mean or

median RMDs do not match for all times. That result is not unexpected since the
RDs for each simulated firing are different for different METCMs and elevations.
In addition, the minimum and maximum values for all 50 cases were extracted for
each WRF time category as presented in Table 7.
Table 7 Minimum and maximum mean RMDs for each WRF time (or GFS data age)
category. Minimum and maximum values shown on the top and bottom lines, respectively,
with values in meters on the left and % RD on the right.
Meters % RD
12 h 36 h 60 h 84 h 108 h 12 h 36 h 60 h 84 h 108 h
11.1 27.0 42.7 31.5 38.7 0.047 0.120 0.194 0.129 0.172
288.3 441.4 464.4 844.8 1135.7 1.282 2.006 2.110 3.719 5.088
As expected the maximum values increased with age of GFS data, but not entirely
for the minimum values. As shown in Table 7 the largest minimum value occurred
for the 60-h WRF run. On the other hand, the largest maximum value occurred for
the 108-h run. Table 8 presents the same numbers as Table 7 but for the median
RMDs.
Table 8 Minimum and maximum median RMDs for each WRF time (or GFS data age)
category. Minimum and maximum values shown on the top and bottom lines, respectively,
with values in meters on the left and % RD on the right.
Meters % RD
12 h 36 h 60 h 84 h 108 h 12 h 36 h 60 h 84 h 108 h
10.3 30.1 39.0 27.2 36.4 0.044 0.125 0.165 0.110 0.158
288.4 440.8 460.0 835.6 1106.0 1.268 1.989 2.116 3.748 5.033
For the median values, the second smallest minimum in meters and % RD was the
one at 84 h though the value in meters was close to that for 36 h. Again, the largest
minimum occurred for the 60-h category. The maximum values in Table 8 followed
19
the same trend as in Table 7 with increasing size as the WRF time from start of
GFS increased.
An additional consideration is whether or not the model produces better outcomes

than using standard MET. Standard MET is based on the International Civil
Aviation Organization 1976 Standard Atmosphere as described in STANAG 4061
(2000). It assumes a standard lapse rate for temperature and pressure, and wind
speed is set at 0. Earlier work using live-fire data suggested RMDs relative to RD
calculated from RAOB METCMs are on the order of 3% to 4%, occasionally much
more or much less. A comparison of 12 cases from this study led to similar results,
with averages of mean or median values of 3.9% and 3.5%, respectively. Individual
RMDs varied considerably, from as small as 0.68% (PANC on 2016-06-06 at
12 UTC) to as large as 6.47% (VEF on 2016-06-18 at 12 UTC). Both cases also
had much larger and smaller RMDs, respectively (3.74% and 2.80 %), suggesting
that the effects of wind and density errors augmented or counteracted one another.
Table 9 presents the number of times each of the 50 case’s mean or median RMD
(50 values of each) exceeded 3.5% for each GFS age. The number was the same
for both mean and median RMDs.
Table 9 Number of cases where the mean or median RMD of the 4 azimuth directions
exceeded 3.5%
12 h 36 h 60 h 84 h 108 h
Mean 0 0 0 1 1
Median 0 0 0 1 1
Even though there are not enough cases for a definitive statement, it appears that
the model-based results should be better than standard MET through the 60-h
category. Nevertheless, for one of the cases examined, the value at 60 h slightly
exceeded that for standard MET. On the other hand, the largest mean RMD for one
case from the use of model data (108-h WRF run) was slightly smaller than that
from the use of standard MET (both were slightly higher than 5% RD). As with all
similar data sets examined to date, there is wide variation from the mean and
median values of Tables 4 and 5, as also suggested in Tables 6–8. Occasionally,
standard MET can produce a better outcome than a model-based or RAOB-based
METCM for a specified direction of fire even when using relatively recent GFS or
other large-scale model input due to cancellation of otherwise large errors in density
and wind.
20
6. Conclusion
This study looked at the potential degradation of WRF-based METCMs relative to

RAOBs that served as an approximations to the real atmosphere using differences
of meteorological variables and impact points from simulated (GTRAJ)
trajectories. In general, the meteorological variables from the WRF-based profiles
deviated more from the RAOB-based values as the age of the GFS data increased.
However, it appears that the wind estimates slightly improved at many heights as
the GFS data aged from 72–84 h to 96–108 h (84- and 108-h WRF output,
respectively). While an investigation in depth is beyond the scope of this study, a
brief look showed that at least some of the situations leading to unexpected results
seemed to arise when the centerline of a trough or a low center lay not far from one
or more RAOB sites. A small displacement of the trough or low can lead to a GFS
grid point location being on the east or west side of the centerline or center,
respectively, or just being closer or further from the centerline or center. The
former, in turn, may lead to a southerly wind changing to a northerly wind or vice
versa sometimes with a fairly significant wind speed in each direction, and the latter
to very different wind speeds even if in a similar direction. Either one could result
in a greater difference from the observed RAOB wind. For example, the former
situation seems to have occurred over the Alaska sites (Table 1) around 6 April
2016 based on 700-, 500-, and 300-hPa weather charts (not shown, but available at
http://archive.atmos.colostate.edu via the included links).
As with the meteorological variables, the GTRAJ results suggested a general
degradation with increasing GFS age of the RMDs generated using the WRF-based
METCMs relative to those from the RAOB-based METCMs. However, a leveling
off or a decrease in the difference between the WRF- and RAOB-based RMDs
occurred from the 84- to 108-h comparisons, based on the data used in this report.
Also the mean or median values on a case-by-case basis varied widely from one
site to another or between the coincident RAOB times for a given site. In future,
the variation in results by site and between individual GFS data sets could be
investigated when a larger set of additional comparisons become available.
Overall, the data suggest that almost all the time model-based METCMs for
artillery simulations outperform standard MET up through a GFS age of 60 h.
While the mean differences in mean or median RMD for 84- and 108-h WRF are
noticeably smaller than those from the use of standard MET, occasionally standard
MET could lead to a better result. Nevertheless, the mean values and the other
statistics suggest the use of model-based METCMs is most often preferable than
the use of standard MET. If the use of standard MET was to be considered,
investigation of more complete models that include latitude, seasonal, and diurnal
21
variations would be suggested, but that would require additional study to evaluate
potential value.
The results of this report suggest that the use of METCMs from WRF-based
soundings using large-scale model input without data assimilation can lead to
simulated trajectories that are within 1.25% RD on average relative to trajectories
calculated using METCMs from RAOB for all WRF times of this report. For WRF
times of 60, 36, and 12 h, the mean RMD decreases to about 0.8%, 0.54%, and
0.38% RD, respectively, and a little less for the respective median RMDs. However,
the variation from one location and time to another is large and any one RMD may
be much smaller or larger. Future research should investigate the relation between
individual outcomes and the atmospheric situation, as well as differences by
location or region. The use of trajectory calculations from GTRAJ or a similar
simulation tool provides a means to assess the integrated or net accuracy of model
output spatially over the path of the simulated projectile. The simulated time of
flight is often on the order of a minute or 2, and consequently, can provide a nearly
“instantaneous” estimate. When possible, use of actual live-fire results can provide
an insight into the actual atmosphere not otherwise attainable.
22
7. References
Cogan J. A generalized method for vertical profiles of mean layer values of

meteorological variables. Adelphi (MD): Army Research Laboratory (US);
2015. Report No.: ARL-TR-7434. Available at http://www.arl.army.mil
/arlreports/2015/ARL-TR-7434.pdf.
Dutsch ML. Evaluation of the WRF model based on observations made by

controlled meteorological balloons in the atmospheric boundary layer of
Svalbard. Bergan (Norway): Meteorologisk Institutt; 2012.
Frehlich R, Sharman R, Clough C, Padovani M, Fling K, Boughers W, Walton WS.

Effects of Atmospheric Turbulence on Ballistic Testing. J Appl Met Climatol.
2006;47:1539–1549.
Gemmill W, Katz B, Li X. Daily real-time, global sea surface temperature–high

resolution analysis: RTG_SST_HR. Camp Springs (MD): The National Center
for Environmental Prediction; 2007 Oct. Report No.: 260.
Kilpelainen T, Vihma T, Mannienen M, Sjoblom A, Jakobson E, Palo T, Maturilli

M. Modelling the vertical structure of the atmospheric boundary layer over
Arctic fjords in Svalbard. Q. J R Met Soc. 2012;138:1867–1883.
Lee JA, Kolczynski WC, McCandless TC, Haupt SE. An objective methodology
for configuring and down-selecting an NWP ensemble for low-level wind
prediction. Mon Wea Rev. 2012;140:2270–2286.
Reen BP. Army Research Laboratory (US), Adelphi, MD. Personal

communication, 2015.
Reen BP, Stauffer DR, Davis KJ. Davis, 2014. Land-surface heterogeneity effects
in the planetary boundary layer. Boundary-Layer Met. 2014;150:1–31.
Skamarock WC, Klemp JB, Dudhia J, Gill DO, Barke DM, Duda MG, Huang X-
Y, Wang W, Powers JG. A description of the advanced research WRF Version
3. Boulder (CO): National Center for Atmospheric Sciences, Boulder; 2008
Jun. Report No.: NCAR/TN-475+STR.
STANAG 4061. Adoption of a Standard Ballistic Meteorological Message, MET

(Edition 4). Washington (DC): North Atlantic Treaty Organization, Military
Agency for Standardization; 2000.
23
INTENTIONALLY LEFT BLANK.
24
Appendix. Individual Mean and Median Radial Miss Distances
(RMDs) per Site
25
Tables A-1 and A-2 have the individual case mean and median radial miss distances
(RMDs) for each site, Weather Research and Forecast (WRF) model time relative
to Global Forecast System (GFS) start (0-h forecast), and radiosonde observation
(RAOB) (R) date and time used for the comparisons. An individual case value is
for the 4 azimuths/directions of fire. There are 50 cases, 5 GFS ages, and 4
trajectory azimuths used in the study. The overall minimum, maximum, mean,
median, and standard deviations of the case-by-case values are repeated from
Tables 4 and 7, and 5 and 8, respectively, at the bottom of Tables A-1 and A-2 for
convenience.
26
Table A-1 Individual case values of mean RMDs for each WRF time from start of GFS. The region (e.g., Alaska) and comparison RAOB (R) dates and
times are shown, and the site codes are listed in the left most column. WMO notations are used when available, otherwise the first 3 or 4 letters of the
site’s name are used (e.g., MEIN for Meiningen). The values from Tables 7 and 4 are listed at the bottom.
12 h 36 h 60 h 84 h 108 h 12 h 36 h 60 h 84 h 108 h
Meters % RD
Alaska R_2016011100
PADQ 108.8 152.4 434.0 421.8 553.1 0.487 0.686 1.943 1.901 2.466
PAKN 132.3 171.4 119.0 395.4 345.6 0.590 0.766 0.533 1.783 1.544
PANC 115.0 55.2 91.0 453.2 1135.7 0.519 0.248 0.410 2.029 5.088
R_2016040612
PADQ 145.6 441.4 464.4 492.8 542.1 0.662 2.006 2.110 2.240 2.463
PAKN 77.8 124.5 42.7 62.1 145.3 0.355 0.568 0.194 0.283 0.661
27
PANC 72.0 92.0 90.7 110.3 101.8 0.329 0.422 0.416 0.506 0.463
East Coast US R_2016032300
GSO 56.9 72.8 44.6 62.1 114.3 0.249 0.316 0.195 0.269 0.496
IAD 30.5 37.2 134.0 53.2 57.7 0.135 0.165 0.597 0.231 0.252
MHX 83.3 73.0 48.7 114.5 203.6 0.371 0.322 0.215 0.507 0.905
RNK 63.7 83.7 53.5 80.3 119.7 0.274 0.358 0.231 0.343 0.509
WAL 53.0 104.9 100.3 51.1 92.4 0.235 0.471 0.453 0.226 0.413
Table A-1 Individual case values of mean RMDs for each WRF time from start of GFS (continued).
12 h 36 h 60 h 84 h 108 h 12 h 36 h 60 h 84 h 108 h
Meters % RD
GSO 134.9 62.5 228.2 303.3 366.0 0.589 0.269 0.991 1.330 1.585
IAD 28.1 190.9 371.9 257.4 613.8 0.126 0.848 1.639 1.150 2.720
MHX 29.3 100.0 172.1 271.2 214.0 0.131 0.443 0.764 1.197 0.946
RNK 95.5 197.8 399.3 639.7 559.4 0.408 0.844 1.694 2.755 2.384
WAL 24.4 226.3 333.7 469.0 473.5 0.110 1.006 1.465 2.088 2.090
Midwest US R_2016010212
AMA 66.1 101.5 253.0 389.5 265.7 0.281 0.430 1.069 1.648 1.125
28
DDC 25.1 171.8 210.2 223.8 134.9 0.109 0.741 0.906 0.963 0.575
LMN 41.7 154.3 230.7 183.8 208.5 0.185 0.684 1.019 0.807 0.916
OUN 101.9 84.4 143.1 267.4 122.9 0.450 0.373 0.629 1.185 0.550
TOP 288.3 196.8 151.1 346.9 339.6 1.282 0.874 0.673 1.545 1.506
R_2016021012
AMA 45.3 40.0 85.4 126.0 117.6 0.191 0.168 0.360 0.534 0.494
DDC 97.8 120.8 66.6 130.3 139.3 0.421 0.519 0.281 0.576 0.601
LMN 25.5 94.9 77.1 157.6 38.7 0.116 0.423 0.347 0.716 0.172
OUN 88.0 89.8 93.4 122.2 128.1 0.391 0.402 0.415 0.551 0.567
TOP 58.0 27.0 174.4 177.5 50.8 0.253 0.120 0.787 0.818 0.229
12 h 36 h 60 h 84 h 108 h 12 h 36 h 60 h 84 h 108 h
Meters % RD
Southwest US R_2016061812
FGZ 79.6 127.1 126.9 141.7 152.0 0.310 0.496 0.495 0.554 0.594
NKX 52.2 50.3 134.6 149.3 150.4 0.227 0.218 0.586 0.650 0.655
PHX 11.1 57.4 87.3 138.8 123.4 0.047 0.246 0.374 0.594 0.528
TUS 62.4 105.7 183.5 212.7 185.6 0.262 0.444 0.771 0.893 0.779
VEF 83.6 116.1 75.6 48.8 103.2 0.353 0.491 0.320 0.205 0.436
Australia R_2016050512
YMML 61.6 61.5 123.8 73.6 98.7 0.272 0.272 0.547 0.325 0.438
YPAD 59.5 50.2 100.6 148.9 168.1 0.265 0.224 0.448 0.662 0.747
29
Germany R_2016020700
ETGB 199.0 140.6 169.8 718.1 356.4 0.891 0.632 0.762 3.210 1.611
ETGI 61.0 135.1 217.9 844.8 236.3 0.270 0.597 0.959 3.719 1.056
ETGK 90.9 150.5 151.5 471.0 367.5 0.402 0.664 0.668 2.073 1.626
MEIN 142.6 167.3 283.0 785.1 463.5 0.626 0.736 1.244 3.446 2.044
STUT 163.4 203.5 258.7 743.9 437.2 0.724 0.902 1.146 3.306 1.947
R_2016030700
ETGB 74.4 174.5 425.2 331.1 280.6 0.338 0.793 1.935 1.505 1.276
ETGI 110.9 47.8 199.0 454.6 65.7 0.495 0.213 0.889 2.036 0.295
ETGK 129.6 59.2 283.8 227.2 563.6 0.582 0.262 1.268 1.028 2.547
MEIN 91.4 240.7 295.3 496.9 374.7 0.412 1.075 1.314 2.224 1.664
STUT 61.9 88.6 168.9 215.5 181.4 0.281 0.401 0.757 0.979 0.818
12 h 36 h 60 h 84 h 108 h 12 h 36 h 60 h 84 h 108 h
Meters % RD
South Africa R_2016050600
FAIR 32.6 31.4 68.1 31.5 270.5 0.132 0.128 0.278 0.129 1.101
FALE 110.7 69.7 111.9 225.1 139.3 0.491 0.309 0.497 0.998 0.616
South Korea R_2015110712
CHEJ 231.3 240.0 282.6 87.8 112.5 1.017 1.051 1.239 0.381 0.486
HEUK 56.3 79.8 159.9 229.5 183.5 0.249 0.351 0.707 1.015 0.811
POHA 66.1 158.6 287.2 227.1 200.4 0.294 0.705 1.272 1.006 0.886
RKJJ 110.6 85.4 232.7 181.6 111.2 0.489 0.373 1.029 0.805 0.492
30
RKSO 108.1 241.6 77.4 128.7 206.7 0.476 1.070 0.345 0.571 0.918
Min 11.1 27.0 42.7 31.5 38.7 0.047 0.120 0.194 0.129 0.172
Max 288.3 441.4 464.4 844.8 1135.7 1.282 2.006 2.110 3.719 5.088
Mean 86.8 123.0 182.4 273.5 254.3 0.383 0.543 0.804 1.210 1.122
Median 76.1 103.2 155.7 219.6 184.5 0.334 0.443 0.690 0.971 0.815
Std Dev 53.6 75.3 110.3 207.6 200.7 0.239 0.339 0.492 0.921 0.898
Table A-2 Individual case values of median RMDs for each WRF time from start of GFS. The region (e.g., Alaska) and comparison RAOB (R) dates and
times are shown, and the site codes are listed in the left most column. WMO notations are used when available, otherwise the first 3 or 4 letters of the
site’s name are used (e.g., MEIN for Meiningen). The values from Tables 8 and 5 are listed at the bottom.
12 h 36 h 60 h 84 h 108 h 12 h 36 h 60 h 84 h 108 h
Meters % RD
Alaska R_2016011100
PADQ 115.6 142.9 407.5 424.4 538.0 0.499 0.616 1.896 1.903 2.438
PAKN 111.8 143.6 93.3 385.2 330.6 0.500 0.647 0.433 1.683 1.489
PANC 116.3 49.7 89.7 444.8 1106.0 0.516 0.234 0.419 2.013 5.033
R_2016040612
PADQ 146.0 440.8 460.0 491.2 538.6 0.656 1.989 2.116 2.205 2.469
PAKN 76.6 125.9 42.1 61.7 134.6 0.350 0.577 0.191 0.281 0.613
31
PANC 62.9 82.8 92.4 99.4 109.5 0.286 0.377 0.424 0.454 0.500
GSO 51.7 76.2 39.0 64.7 97.8 0.236 0.320 0.165 0.272 0.444
IAD 32.2 33.3 128.4 40.2 55.6 0.149 0.143 0.597 0.180 0.254
MHX 79.1 66.0 44.4 108.0 181.1 0.364 0.297 0.202 0.492 0.841
RNK 59.3 87.7 53.0 76.9 113.8 0.266 0.364 0.229 0.318 0.492
WAL 54.0 89.2 100.8 50.8 82.4 0.231 0.387 0.462 0.221 0.363
Table A-2 Individual case values of median RMDs for each WRF time from start of GFS (continued).
12 h 36 h 60 h 84 h 108 h 12 h 36 h 60 h 84 h 108 h
Meters % RD
GSO 127.2 63.7 191.1 307.6 351.4 0.581 0.269 0.810 1.302 1.565
IAD 29.7 180.2 356.4 251.1 619.0 0.127 0.845 1.619 1.168 2.675
MHX 28.7 99.9 154.8 261.8 177.2 0.126 0.432 0.694 1.192 0.787
RNK 87.4 198.1 400.2 647.2 536.8 0.394 0.849 1.658 2.676 2.343
WAL 23.0 219.0 317.8 458.5 462.5 0.098 1.019 1.437 2.097 1.993
Midwest US R_2016010212
AMA 65.7 101.4 244.8 369.9 235.3 0.283 0.428 1.065 1.573 1.011
32
DDC 27.2 170.7 213.4 219.6 128.9 0.114 0.738 0.903 0.973 0.561
LMN 39.6 142.7 227.6 186.9 211.7 0.176 0.645 1.027 0.804 0.925
OUN 99.1 79.3 136.3 255.5 122.9 0.451 0.356 0.613 1.147 0.536
TOP 288.4 191.3 140.5 314.8 297.8 1.268 0.845 0.631 1.384 1.335
R_2016021012
AMA 44.8 41.2 79.2 119.5 120.8 0.183 0.168 0.345 0.497 0.493
DDC 96.3 118.7 68.8 119.1 144.0 0.393 0.517 0.287 0.534 0.649
LMN 25.3 90.7 82.7 150.6 36.4 0.113 0.379 0.369 0.688 0.158
OUN 81.8 88.7 89.5 126.6 129.5 0.363 0.391 0.386 0.570 0.568
TOP 53.4 30.1 180.5 199.6 51.8 0.230 0.125 0.791 0.902 0.230
12 h 36 h 60 h 84 h 108 h 12 h 36 h 60 h 84 h 108 h
Meters % RD
Southwest US R_2016061812
FGZ 75.5 125.6 126.1 137.7 151.8 0.302 0.503 0.504 0.537 0.594
NKX 49.9 50.7 136.0 150.8 152.0 0.212 0.215 0.576 0.638 0.645
PHX 10.3 53.9 84.4 133.3 122.6 0.044 0.230 0.358 0.566 0.521
TUS 56.6 97.5 176.1 204.3 183.5 0.237 0.407 0.736 0.850 0.764
VEF 84.5 115.3 78.4 47.8 101.5 0.348 0.498 0.329 0.201 0.443
Australia R_2016050512
YMML 57.5 59.0 125.4 74.0 97.7 0.266 0.273 0.531 0.324 0.435
YPAD 57.1 48.8 100.7 143.1 164.8 0.257 0.219 0.462 0.656 0.749
33
Germany R_2016020700
ETGB 194.0 139.2 168.3 708.5 352.1 0.897 0.618 0.743 3.214 1.562
ETGI 66.1 137.6 214.6 835.6 238.2 0.302 0.606 0.962 3.748 1.064
ETGK 88.0 146.5 153.0 453.7 374.5 0.387 0.641 0.669 2.020 1.650
MEIN 133.7 165.2 286.4 766.3 453.2 0.575 0.730 1.267 3.341 2.018
STUT 150.8 191.8 258.5 611.4 438.4 0.652 0.843 1.138 2.649 1.942
R_2016030700
ETGB 69.9 164.3 402.2 294.9 281.1 0.319 0.745 1.818 1.332 1.282
ETGI 89.1 55.7 181.4 457.3 67.8 0.403 0.248 0.809 2.019 0.307
ETGK 127.1 53.2 265.5 220.7 576.1 0.568 0.241 1.143 1.000 2.582
MEIN 91.6 243.9 291.8 492.4 368.5 0.411 1.083 1.258 2.208 1.662
STUT 62.9 73.5 176.4 200.6 183.7 0.285 0.332 0.796 0.912 0.825
12 h 36 h 60 h 84 h 108 h 12 h 36 h 60 h 84 h 108 h
Meters % RD
South Africa R_2016050600
FAIR 33.8 37.4 60.7 27.2 255.9 0.138 0.155 0.246 0.110 1.059
FALE 98.7 72.8 101.7 178.4 127.8 0.437 0.321 0.451 0.789 0.553
South Korea R_2015110712
CHEJ 228.7 221.2 280.1 89.7 119.3 0.986 1.013 1.190 0.381 0.510
HEUK 58.1 59.7 162.0 220.1 175.0 0.250 0.270 0.725 0.944 0.783
POHA 59.2 157.8 282.7 227.8 238.3 0.256 0.697 1.281 1.008 1.043
RKJJ 111.9 83.6 230.7 179.7 113.8 0.486 0.373 1.002 0.774 0.507
34
RKSO 108.8 243.0 66.9 136.9 213.8 0.478 1.058 0.289 0.592 0.924
Min 10.3 30.1 39.0 27.2 36.4 0.044 0.125 0.165 0.110 0.158
Max 288.4 440.8 460.0 835.6 1106.0 1.268 1.989 2.116 3.748 5.033
Mean 83.738 119.023 176.881 264.553 249.313 0.526 0.781 1.167 1.104 Mean
Median 72.688 98.703 153.905 202.432 179.173 0.418 0.681 0.907 0.785 Median
Std Dev 52.445 74.236 107.367 200.528 197.122 0.335 0.482 0.889 0.889 Std Dev
List of Symbols, Abbreviations, and Acronyms
AGL above ground level
ARDEC Armaments Research Development and Engineering Center
ECMWF European Centre for Medium-Range Weather Forecasting
ESRL Earth System Research Laboratory
GFS Global Forecast System
GTRAJ General Trajectory
IMS Ice Mapping System
MAD mean absolute difference
MD mean difference
METCM computer meteorological message
MSL mean sea level
NCEP The National Center for Environmental Prediction
NCL NCAR Command Language
NCAR National Center for Atmospheric Research
NOAA National Oceanic and Atmospheric Administration
NOHRSC National Operational Hydrologic Remote Sensing Center
RAOB radiosonde observation
RD radial distance
RMD radial miss distance
RMSD root mean square difference
SD standard deviation
UTC Coordinated Universal Time
WRF Weather Research and Forecasting
WMO World Meteorological Organization
35
1 DEFENSE TECHNICAL
(PDF) INFORMATION CTR
DTIC OCA
2 DIRECTOR
(PDF) US ARMY RSRCH LAB
RDRL CIO LL
RDRL IMAL HRA RECORDS MGMT
1 DIRECTOR
(PDF) US ARMY RSRCH LAB
RDRL CIE
J L COGAN
36

WRF Accuracy With GFS PDF

Uploaded by

Copyright:

Available Formats

WRF Accuracy With GFS PDF

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

WRF Accuracy With GFS PDF

Uploaded by

Copyright:

Available Formats

ARL-TR-7782 ● SEP 2016

US Army Research Laboratory

Change in Weather Research and Forecasting

Approved for public release; distribution unlimited.

Citation of manufacturer’s or trade names does not constitute an official

US Army Research Laboratory

Change in Weather Research and Forecasting

Approved for public release; distribution unlimited.

5c. PROGRAM ELEMENT NUMBER

6. AUTHOR(S) 5d. PROJECT NUMBER

5f. WORK UNIT NUMBER

7. PERFORMING ORGANIZATION NAME(S) AND ADDRESS(ES) 8. PERFORMING ORGANIZATION REPORT NUMBER

11. SPONSOR/MONITOR'S REPORT NUMBER(S)

12. DISTRIBUTION/AVAILABILITY STATEMENT

4. Initial Results: Meteorological 10

5. Initial Results: Trajectories 17

Appendix. Individual Mean and Median Radial Miss Distances (RMDs)

List of Symbols, Abbreviations, and Acronyms 35

Approved for public release; distribution unlimited.

Approved for public release; distribution unlimited.

Approved for public release; distribution unlimited.

Approved for public release; distribution unlimited.

Some meteorological centers compare global and large-scale regional model

Approved for public release; distribution unlimited.

The METCM consists of meteorological data (i.e., pressure, virtual temperature,

Approved for public release; distribution unlimited.

Date and Time Center Lat, Lon

Alaska US Anchorage PANC 2016-01-11-00 59.46, –153.34

East Coast US Blacksburg, VA RNK 2016-03-23-00 36.88, –77.94

Midwest US Amarillo, TX AMA 2016-01-02-12 37.12, –98.66

Southwest US Flagstaff, AZ FGZ 2016-06-18-12 33.78, –113.57

Australia Adelaide YPAD 2016-05-05-12 –36.05, 141.69

Germany Bergan ETGB 2016-02-07-00 50.82, 9.62

Approved for public release; distribution unlimited.

Date and Time Center Lat, Lon

South Korea Cheju CHEJ 2015-11-07-12 35.19, 127.42

Approved for public release; distribution unlimited.

Approved for public release; distribution unlimited.

Approved for public release; distribution unlimited.

The GTRAJ (v3.9.7) program was provided by ARDEC and is considered a

RMD = ((∆Dr)2 + (∆Dd)2)1/2, (1)

Approved for public release; distribution unlimited.

Approved for public release; distribution unlimited.

Line Height Samples MD MAD SD RMSD

Approved for public release; distribution unlimited.

Approved for public release; distribution unlimited.

Approved for public release; distribution unlimited.

Approved for public release; distribution unlimited.

Line Height Samples MD MAD SD RMSD

Approved for public release; distribution unlimited.

Approved for public release; distribution unlimited.

5. Initial Results: Trajectories

Approved for public release; distribution unlimited.

WRF times from start of GFS data

WRF times from start of GFS data

As suggested by the standard deviations, there is significant variation from case to

Approved for public release; distribution unlimited.

No. of minimums No. of maximums

As shown in Table 6, the number of minimums or maximums of either mean or