CWB Project
Welcome to the CWB Project page!
(Updated 16 November 2007)
3, 2007 progresses
CWB
Observation information over Taiwan area can be found from cwb_obs_info.
CWB-CAA-domains-settings: This
year the CWB and CAA NWP modeling system will be merged together with 3
nested domains:
Domain 1:
222x128x45, 45-km
Domain 2:
184x196x45, 15-km with the left-low corner at point (89,22) in Domain1.
Domain 3:
151x181x45, 5-km with the left-low corner at point (59,78) in Domain2.
Suggested CWB/CAA operational implementation:
Observations: Conventional
data including SYNOP, METAR, SHIP, BUOY, TEMP, AIREP, PILOT, and SATOB,
SATEM, QSCAT.
The BOGUS data are also
included. Currently the GPSPW and GPSRO data are not included, but no
problem
to assimilate these data with wrfvar system. The
time window is +/- 3-h centered at the analysis time.
First Guess :
For the Cold-Start run, the first guesses are derived with WPS and real
from CWB GFS datasets.
For the Cycling run, the
first guess at the beginning time is from CWB GFS. After that time, the
first guesses are from
the 6-h WRF model
forecast.
WRFVar system: use version 2.1 of
wrfvar with wrf_io.F90.2.2. At this point, the main features are
cv_options = 3, i.e. NCEP background error
statistics, is used.
cv_options_hum = 1, i.e.
specific humidity, is used as the moist control variable;
check_rh = 2, i. e. the final
moisture analysis is adjusted to remove the super-saturation and
maintain the water vapor
conserved in the vertical column.
sfc_assi_options = 2, i. e. the U10, V10, T2,
Q2 are directly assimilated based on the similarity theory.
WRF model: Version 2.2
with few bugs fixed. The 45 eta levels are recommended as
eta_levels
= 1.000, 0.985, 0.969, 0.953, 0.937, 0.921, 0.905, 0.888, 0.87 , 0.853,
0.835, 0.816, 0.798, 0.778, 0.759, 0.739, 0.718, 0.696, 0.674, 0.651,
0.628, 0.603,
0.577, 0.55 , 0.522, 0.492, 0.461, 0.426, 0.389, 0.348,
0.303, 0.273,
0.242, 0.212, 0.182, 0.152, 0.121, 0.091, 0.061, 0.03 , 0.000
and P_top = 3,000 Pa.
WSM 5-class microphysics
(mp_physics = 4);
rrtm long wave radiation scheme (ra_lw_physics = 1);
Goddard short wave scheme (ra_sw_physics = 2);
Monin-Obukhov surface scheme (sf_sfclay_physics = 1);
Noah LSM model (sf_surface_physics = 2);
YSU PBL scheme (bl_pbl_physics =
1);
Kain-Fritsch (new Eta) scheme
for 45 and 15-km domain (cu_physics = 1).
** The
namelist files can be downloaded from the
suggested namelist. which includes the namelist files to run WPS,
WRFV2/real, OBS_FGGE_PROC, 3DVAR_OBSPROC, WRFVar,
and WRF model for 3 nested domains.
Updated the
suggested namelist for reasons (Y.-R. Guo 20070726):
1)
namelist.3dvar.sample: a, turn on the switches for Use_GeoAMVObs,
Use_PolarAMVObs, and Use_SatemObs;
b, add the
switch for Bogus: Use_BogusObs = .TRUE.,
c, Change
RF_PASSES = 4, to be
RF_PASSES = 6,
2) namelist.input.sample was
adapted from Jim Bresch 20070625 updated namelist.input. But the
time_step is
still 180 seconds because 240s not working with
45-km domain for 2006 Shanshan Typhoon case.
** A running shell script: Run_CWB-WPS-WRFVAR-WRF.csh,
can also be downloaded for your reference. With this
shell script, the "pure" WRF model run (NOVAR), and
Cold-Start and Cycling runs are ver easy conducted.
Updated Run_CWB-WPS-WRFVAR-WRF.csh
(Y.-R. Guo 20070726):
1) using the
dependence: #BSUB ${WAIT} done(${JOBname}) instead of the "wait" and
"sleep";
2) using "bsub" link to
submit the jobs time period by time period.
As CWB HPC has the same sueue
system as NCAR IBM (blueice and bluevista), this shell script could be
ported to
CWB HPC. I used
this script to conduct one month (2006 Dec.) cycling runs in NCAR IBM
(blueice), which can
be completed
within a week. When I did samething in PC Linux using 6 CPUs
(leea.mmm.ucar.edu), it took about
one month
wall-clock time.
** All the code can be downloaded
from http://www.mmm.ucar.edu/individual/bresch/cwb/
.
Note
that the geogrid files for three domains ( geo_em.d01.nc,
geo_em.d02.nc,
geo_em.d03.nc)
should also be
downloaded from the above
web
site.
The description
of the namelist.3dvar for version 2.1 and
version2.2-beta of WRFVar.
CV5
Background Error Statistics (BES) data files available for CWB/CAA
configuration
Several BES files have been
derived based on the one-month CWB GFS
data from 0000 UTC 1 to 1200 UTC 31 December 2006:
1) NODA (No Data
assimilation WRF forecast with the configuration mentioned
above) 45-km, bin_type =5 by Hui Shao, DATC/NCAR.
The gen_be code is from WRFVar 2.2. The data
file can be obtained from CWB ftp site:
datwncr@163.29.179.171:/guo/be_45km_noda.cv5
2) DACV3 (WRFVar/WRF 6-h cycling run
with CV3 BES from 20061200Z to 2006123118Z, SATOB, SATEM and BOGUS data
not assimilated)
45/15/5km bin_type=5 and 1 by Yun-Tien Lin and Y.-R. Guo. gen_be code
is from WRFVar_2.1. The data files can be obtained from CWB
ftp site:
datwncr@163.29.179.171:/guo/be_45km_da1.cv5
datwncr@163.29.179.171:/guo/be_15km_da1.cv5
(bin_type=5)
datwncr@163.29.179.171:/guo/be_05km_da1.cv5
datwncr@163.29.179.171:/guo/be_45km_da1_bin1.cv5
datwncr@163.29.179.171:/guo/be_15km_da1_bin1.cv5
(bin_type=1)
datwncr@163.29.179.171:/guo/be_05km_da1_bin1.cv5
3) DACV5I (WRFVar/WRF 6-h cycling run
with Interpolated CV5 BES from Eric Chiang's L41 BES.
SATOB, SATEM and BOGUS not assimilated)
45/15/5km bin_type=5 and 1 by
Yun-Tien Lin and Y.-R. Guo. gen_be code is from WRFVar_2.1. The
data
files can be obtained from CWB ftp site:
datwncr@163.29.179.171:/guo/be_45km_da2.cv5
datwncr@163.29.179.171:/guo/be_15km_da2.cv5
(bin_type=5)
datwncr@163.29.179.171:/guo/be_05km_da2.cv5
datwncr@163.29.179.171:/guo/be_45km_da2_bin1.cv5
datwncr@163.29.179.171:/guo/be_15km_da2_bin1.cv5
(bin_type=1)
datwncr@163.29.179.171:/guo/be_05km_da2_bin1.cv5
A, Shell and code
Observation processing
GPSRO wetPrf decoder
program wetPrf_decoder.tar.gz
(20070323) is updated to make the LITTLE_R format consistent with the
GPSRO bfrPrf data.
GPSRO bfrPrf decoder program bfrPrf_decoder.tar.gz
(20070827) is updated for reasons:
1) make it to be able in
compiling and using in IBM machine;
2) introduce one more namelist
variable: one_lev_per_record .
one_lev_per_record = .FALSE. ---- one
GPSRO sounding per report in LITTLE_R file. This format can be used for
local operator (rafractivity) without
considering the latitude/longitude varying with impact_parameter
(height). Although the lat/lon information is included
in each of the level data records, they are not used in the
implementation of the current local obs operator. However,
those information will be used in future
non-local
obs operator (excess phase) implementation.
one_lev_per_record
= .TRUE. ---- one level
of GPSRO data per report in LITTLE_R file. Since the "halo" for MPP
code in WRFVar currently
are allocated as only 2-D arrays,
if users want to
consider the location (lat/lon) varys with the height
(impact_parameter),
the alternative way is to write out
one-level data per record. Now this just work for local obs operator
(refractivity)
assimilation.
If users don't want to consider the lat/lon
varying with height, keep everything in the shell script as before, no
changes needed even if using this new code.
Posted on 25 October 2007:
Since 20 September 2007, COSMIC/CDDAC updated
the bfrPrf data format. The refractivity and bending angle now are
available up to 60km (300 levels).
The bfrPrf decoder program bfrPrf_decoder.tar.gz
(20071025) is for the new bfrPrf data file. The
old bfrPrf data files with the size of 8487 should be decoded
by the previous
decoder program.
The
oservation pre-proccesor 3DVAR_OBSPROC
(20070821) has been updated for bug fix in Quik SCAT data thining
code. Eric Chiang found that the previous code
has
"Segmentation error" in running with the data: obs_2007081418.r. This
bug has the minor effect on the results in assimilation of thined the
QSCAT data.
CWB Observation data decoder OBS_FGGE_PROC
is updated by Jim Bresch on 24 October 2007 (see CWB blog by Jim
Bresch).
Data
assimilation
WRFVAR-based VERIFY
a, For U10, V10, T2 and Q2 verification (SYNOP,
METAR, SHIP, etc.), these parameters are directly read from WRF model
output, wrfout, or MM5 model
output, MMOUT, not re-calculated
in WRFVar (DA_transfer_wrftoxb.inc or DA_transfer_mm5toxb.inc). So we
suggest that for surface observation
verification, set the namelist variable:
sfc_assi_options = 2 in namelist.3dvar in running var_qc_obs.csh and
var_verify.csh. The updated WRFVar code
can be downloaded from wrfva_2.1.tar.gz.
b, The program
plot_var_stats was
modified i) for bug fix; ii) allowing any time, not necessay t=0, to
begin the verification. This is necessary for model
forecast U10, V10, T2, and Q2 verification
because those parameters are incorrect at t=0 in model output files.
So, for example, for SYNOP verification
with sfc_assi_options = 2, it could be
Start_fcst_time = 3 and End_fcst_time = 48, etc.
c, The shell scripts for running
VERIFY are also have minor modifications.
Background
error statistics
The details of the description for
CV_OPTIONS = 3 Background Error Statistics, which is current used in
CWB WRFVar operational implementation
can be found from CV3_BES.
The Stage0 for using
WRFV2.2 historic wrfout files in deriving the CV5 BES.
Because the netCDF format of the
historic forecast files (wrfout) from WRFV2.2 is something
different from WRFV2.1, the program gen_be_Stage0
must be modified, otherwise the results would be wrong even although
the previous gen_be_Stage0(v2.1) progam could be
running without error reported.
The updated Stage0 code for WRFV2.2 application can be downloded
from gen_be_Stage0 (v2.2).
This code have been tested in Linux machine with
the CWB
WRFV2.2 forecast data from 2006120100Z to 2006120300Z with 12h
inteval. The log file, gen_be_stage0.log,
can be downloaded here
for your reference.
To split your WRFV2.2
historic forecast files and extract the specific times, such as 12h and
24h forecasts, there is a updated shell script:
time_split.csh.
BE Tuning
utilities
Background and
observation error tuning utilities: daprog_diagnostics.f90
and daprog_ominusb.f90 (Hollingsworth and Lonnberg, 1986) and
tune.f90
(Desroziers and ivanov, 2001) have been modified and corrected.
These codes and shell scripts: daprog_ominusb.csh
and tune.csh
are
under the directory wrfvar_2.1/da_3dvar/utl
in wrfvar_2.1.tar.gz
(20070129),
Please read the changes file: wrfvar_2.1/da_3dvar/changes/change_20070109.yrg
after un-tarred wrfvar_2.1.tar.
Dale Barker applied these tuning techniques
to 2002 KMA project. The report
(doc file) can be downloaded here for your reference.
1, The Hollingsworth and Lonnberg approach
should have a
collection of fort.50 files from WRFVar runs as the input file
by using shell: daprog_ominusb.csh.
A sample dataset can
be downloaded from OMB_data.tar.gz
for your testing. There are two known problens in this version of
wrfvar_2.1: (1) to produce the
correct fort.50 file, the single
processor wrfvar run must be used currently; (ii) the vertical
binning is now only based on the pressure. for the observations
with the height, such as GPS refractivity,
the code need to be modified, but it is easy to do.
To
get the meaningful results, user must accumulate the fort.50 files as
much as possible, for example 3 months or more. The
Hollingsworth and Lonnberg
method is an off-line approach. They assume that the
background errors are correlated and observation errors between the
observation sites are uncorrelated.
And also the covariance matrix is homogeneous and
isotropic. Then the perceived short-rang forecast error variance can be
partitioned into the background
and observation errors.
2, The Desroziers and Ivanov
approach is based on the expectation of observation part of the cost
function Jo. Through the roandomized estimation of the trace
Tr(HK),
the expectation of Jo can be computed. Then, the observation tuning
factors can be obtained as (actual Jo / expected Jo)1/2.
The
input files to the Adaptive tuning program are
( i) Observation error vector R and the random
perturbation vector x (fort.45)
from the perturbed WRFVar run (omb_add_noise = .TRUE.);
( ii) Perturbed
observation yp = yo + dyo
(fort.46)
from the perturbed WRFVar
run (omb_add_noise = .TRUE.);
(iii) Original
observations y = yo (fort.47)
from the normal WRFVar run (omb_add_noise = .FALSE.);
( iv) Actual cost
function Jo (fort.48)
from the normal WRFVar run
(omb_add_noise = .FALSE.);
( v) WRFVar
printed log file: wrfvar.out (fort.49) from
the normal WRFVar run (omb_add_noise = .FALSE.).
The output file is errfac.dat. which will be used in the obs error
tuning WRFVar run (use_obs_errfac = .TRUE.).
In my understanding, if there is a large number of the obs
data available in a WRFVar run, the expectation of Jo for that
observation will be
estimated with high
confidence, and then the reliable tuning factors could be obtained. So
this obs error tuning technique could be online used in
a way of an
iterative fixed-point method. With the current code, tune.f90, a
collection of fort.45, 46, 47, 48, and wrfvar.out from a bunch of
WRFVar
runs over a epriod
could also be
used to get the reliable obs error tuning factors. This is an off-line
way, and the shell scripts can
be downloaded from Adaptive_tune.tar.gz.
In our experiences, it is questional with this
technique for Typhoon bogus data error tuning, especially for Sea Level
Pressure (SLP). Further work is
needed for Typhoon bogus data
error tuning.
B,
Project progresses
CWB (Dr.Terng provided the verification results (ppt file) from August 2006 to January
2007 for the WRFVar/WRF and OI/NFS parallel runs:
1) Height forecast: Anomaly Correlation (AC), S1 score
(S1),
Mean Error (ME), and Absolute Mean Error (AME);
2) Temperature: ME and AME;
3) U and V components: ME and AME.
Explanation of the experiments:
WOP1_0701_12 ---- WRF 12-h
operational forecast for Januaruy 2007 over domain1.
CWB1_0701_12 ---- WRF with small d3 and old
interpolation scheme) 12-h operational fprecast for January 2007 over
domain 1
NOP1_0701_12 ---- NFS 12-h operational
forecast for January 2007 over domain 1.
W_ASO_12 ---- WRF 12-h forecasts for August,
September, and October
N_ASO_12 ---- NFS 12-h forecasts for August, September, and October
W_ND_12 ---- WRF 12-h forecast for November
and December
N-ND_12 ------ NFS 12-h forecast for
November and December.
LARGE D3 means the expanded domain3 (see CWB/CAA domain settings)
CWB_Report (ppt file) on 27
April 2007 by Jing-San Hong.
C, CWB visitors to
NCAR
Jing-Shan Hong, CWB, visited NCAR from 6 to 17 August 2007
Cheryl Terng, CWB, visited NCAR
from 6 to 22 August 2007.
Yun-Tien Lin, CWB,
visits NCAR from 11 July to 29 December 2007
Schedule (doc file) for Drs. Terng and Hong visit.
Mid-term review (ppt file) of 2007 CWB project by Kuo, Guo, Liu, and Brayn, NCAR/MMM and UCAR/COSMIC.
Corporation for WRFVar (ppt file)
between CWB and NCAR for second half of year 2007.
D, Refrences
The WRFVar/WRF forecast system was steup in United Arab Emirates (UAE).
Almost all of the WRFVar techniques were implemented in their
system. The abstract and presentation
given in 2007 WRF user's workshop are posted here:
Abstract by
AJJAJI, et al.
First presentation by AJJAJI,
et al.
Second presentation by
Katheri, et al
E, WRF-DART -----
EnKF
As 2007 CWB project included TASK#2: Exploration of the WRF-based
Ensemble Kalman Filter (EnKF) data aasimilation, some of the related
link are listed below:
DART home page: http://www.image.ucar.edu/DAReS/DART
WRF-DART (EnKF) documentation: http://swiki.ucar.edu/wrf-enkf
Yun-Tien Lin, CWB, has
installed the WRF-DART (EnKF) system in CWB IBM system, and tried to
conduct experiments with Typhoon
Shanshan case study. The running procedure can be
found from WRF-DART_ENKF_run.pdf
(ppt file).
F, Annual report
a, The annual report presented
at CWB on November 28, 2007.
b,
Experimental results
(i)
Period of December 2006, Domain 1 (45km resolution): Verifying
against SOUND, Verifying
against SYNOP.
(ii) Period
of June 2007, Domain 1 (45km resolution): Verifying
against SOUND, Verifying
against SYNOP.
c, Single obs tests with CV5 BES
The rip-execution files for single obs (u, t, q, and p) plots with RIP4.
(i) CV5 BES
derived from one-month CV3 cycling run for 200612: Domain
1 (45km); Domain
2 (15km).
(ii) CV5 BES
derived from one-month cold-start run for 200706: Domain
1 (45km); Domain
2 (15km).
For comments, send email to mesouser@ucar.edu
Last Modified: 25 October 2007
