Final Report for 2005 CWB project
Mesoscale and Microscale Meteorology Division
National Center for Atmospherics Research
Boulder, Colorado, USA
Submitted to Central Weather Bureau, Taiwan, ROC
30 November 2005
1.
Introduction
Following the 2004 CWB-UCAR collaborative project, the WRF 3DVAR system, now called as ÒwrfvarÓ, has been upgraded by NCAR staff. The wrfvar has the following features: 1) the code is fully compatible with WRF V2.1 framework; 2) it has more options of the control variables ---- cv_options=4 (for global) and 5 (for regional) are added; 3) the code (gen_be) for calculating the background error statistics for cv_options 4 and 5 is included; and 4) especially for CWB, the capabilities of assimilating Typhoon and global bogus data are added; and 5) the new observation error specification and quality control procedure for GPS RO data are added. Moreover, the interface programs between the ÒwrfvarÓ and CWB NFS model and the 3DVAR OBS preprocessor are also modified to meet the CWB requirements.
In order to exchange the information and the codes, a web page for CWB project (http://www.mmm.ucar.edu/individual/guo/CWB/CWB_Project.htm) is established. CWB also sets up a web site (http://cwb-iria.cwb.gov.tw/cwb2ncar/frame.htm) for ÒData Exchange InterfaceÓ. On the UCAR side, an anonymous ftp account, ftp.ucar.edu:/pub/mmm/guo, is set for CWB staff to get the large data files from NCAR. These tools are effective and valuable for coordinating efforts by staff of CWB and NCAR.
In this report, the accomplishments for the CWB-NCAR collaborative project (through joint efforts between NCAR and CWB staff) are summarized task by task. The preliminary results from a series of the 3D-Var experiments with wrfvar/WRF V2.1.1 model for Typhoon Dujuan case are presented. Based on these results, the future tasks of the project are also suggested.
2. Review
of the project tasks and accomplishments
Task 1: Technical Support for the Operational Implementation of WRF
3D-Var/NFS at CWB
A) Technical consultancy on retrospective experiments in CWB operational environment
All interface programs and the latest version of WRF 3D-Var system (OBS
preprocessor code: 3DVAR_OBSPROC and 3DVAR code: wrfvar_2.1) are provided to
CWB through the CWB web page: http://www.mmm.ucar.edu/individual/guo/CWB/CWB_Project.htm set-up by NCAR. CWB staff has installed this
software on a PC Cluster: msccs1.msc.cwb at the Satellite Center and the
operational platform: VPP5000 at the Information Center in CWB.
Figure
1. Schematic diagram of WRF 3D-Var/NFS System.
The schematic diagram of WRF-3D-Var/NFS system is shown in Fig.1. The details on the system set-up can be found in appendix A.
CWB staff has completed retrospective experiments with WRF 3D-Var/NFS for two periods: one summer month from 15 August to 15 September 2003 and one winter month from 1 to 31 January 2005. To ensure the success of these experiments, NCAR staff continuously provided technical consultation to CWB staff, and solved the troubles in a timely manner.
For example, the OBS preprocessor: 3DVAR_OBSPROC had a Òdead loopÓ with the CWB pilot data at 2005081800Z because of the incoming data levels are not in proper order, i.e. from high pressure to low pressure. We modified the code to process this kind of data correctly. Another example is that there is difficulty in setting up the background of brightness temperature (Tb) with the CWB NFS forecast in wrfvar run. An Òif-statementÓ is added to skip over this Tb settings because CWB currently did not assimilate any SSMI brightness temperature.
Dr. Cheryl Terng from CWB had a one month visit to NCAR from 22 August to 24 September 2005. During her visit, NCAR staff and Dr. Terng discussed details on the implementation of the retrospective experiments, especially for the summer month (15 August to 15 September 2003). CWB needed to re-run the GFS (Global Forecast System) with 6-h interval output to provide the first guess and boundary conditions to 45-km CWB NFS model and WRF 3D-Var (wrfvar). As suggested by NCAR staff, the tuned NCEP global background error statistics (cv_options=3, see ÒFinal Report for 2004 CWB ProjectÓ submitted to CWB at the end of 2004) is used in WRF 3D-Var this year to ensure the system robust, and the results could be used as a benchmark for further development of the WRF 3D-Var/NFS system.
B) Ingest typhoon bogus data into WRF 3D-Var
To assimilate the Typhoon bogus data provided by CWB, NCAR staff
developed the capability of the WRF 3D-Var system. First, the decoder program
can convert the CWB Typhoon bogus data to LITTLE_R ASCII format and assign the
observation errors to wind and sea level pressure (see http://www.mmm.ucar.edu/individual/guo/CWB/CWB_Project.htm).
The total number of Typhoon bogus soundings provided by CWB is 40, including
the sea level pressure (SLP), and wind, height at 1000, 925, 850, 700, 500, and
400 hPa, and the temperature at 1000, 925, 850, and 700 hPa (Fig.3a).
For Typhoon bogus data assimilation, 1) we assume that all data have good
quality, and bypassed the quality control procedure (check_max_iv); 2) the observation errors at the surface Typhoon
center should be smaller (0.50 hPa for sea level pressure and 0.5 m/s for wind)
and gradually increased with height and distance from the Typhoon center; 3)
only SLP and wind are assimilated in WRF 3D-Var. Currently, the temperature is
not assimilated because we found that the bias for bogus temperature data is
too large (8o~10o), which may violate the assumption of variational
data assimilation approach (BLUE, Best Linear Unbias Estimate). Since WRF 3D-Var is a
multi-variant analysis system, we expected that assimilation the wind fields
should create balanced temperature fields over the Typhoon area. Figure 2 shows
the increment of the surface pressure (Fig.2a), and the cross sections of the
increments for wind components (U/V, Fig.2b), temperature and pressure (T/P,
Fig.2c), and specific humidity (q, Fig.2d). Although only SLP and wind were
assimilated, at the Typhoon center, the warm core (positive 3.3o of
T increment, Fig.2c) below 700 hPa was produced from 3D-Var, and the moist air
(positive 3.25 g/kg of q increment, Fig. 2d) was obtained by moisture
adjustment in 3D-Var.
(d) (c) (b) (a)
Figure 2. Increments form assimilation of Typhoon bogus data only from CWB at 0000 UTC 31 August 2003. a) Surface pressure, contour level 1.0 hPa, the central value is -12 hPa. The thick line indicates the location of the cross section; b) Wind components U (red line and V (green line), the contour level is 3 m/s; c) Temperature (red line) and pressure (green line), the contour levels are 0.5o for T and 1 hPa for P; and d) Specific humidity, the contour level is 0.25 g/kg.
C) Ingest of global model bogus data into WRF 3D-Var system
CWB
provided 134 bogus soundings (FM-135 BOGUS) from their global analysis and 40
typhoon bogus soundings (FM-135 TCBOG) at each of analysis time as shown in
Fig. 3a. To assimilate the global bogus soundings (wind, temperature, and
moisture) in WRF 3D-Var, the observation errors same as the conventional
soundings were assigned to the global bogus data in OBS preprocessor
(3DVAR_OBSPROC) but the quality control procedure (check_max_iv) in wrfvar was bypassed.
Figure 3. Assimilation
of CWB bogus data at 0000 UTC
31 August 2003. a) Data distribution; b) Differences of SLP field between
the assimilations with and without CWB bogus data. (b) (a)
D) Technical consultancy on background error statistics (BES) derivation based on WRF 3D-Var/NFS
During the year of 2005, a new set of control variables (cv_options=5), which are the stream function (y), unbalanced potential velocity (c_u), unbalanced temperature (T_u), pseudo relative humidity (r) (Fisher 2000), and unbalanced surface pressure (Ps_u) in eigenvector space, is included in WRF 3D-Var (wrfvar) (see the right column and bottom row in Table 1).
Table 1. The Background Error Statistics (BES) in WRF 3D-Var (wrfvar) for the different options of the control variables.
The code for derivation of the corresponding BES (gen_be) was also developed. The technical consultancy work from NCAR includes the following items: (i) NCAR delivered the first version of the BES generation code to CWB in March, 2005. (ii) Dr. Barker gave the training course in CWB, Taiwan in May, 2005. (iii) NCAR included the latest version of the gen_be code now in wrfvar code and posted on the CWB web page. (iv) NCAR provided the explanation of the background error statistics in WRF 3D-Var (wrfvar), which has been posted on the CWB project web page, http://www.mmm.ucar.edu/individual/guo/CWB/2005/WRFVAR_BE_Guo.htm, (v) the preliminary BES files based on variety of NWP forecast datasets, such as WRF model forecast initiated from NCEP analysis, NCEP AVN forecast, CWB/OI-NFS forecast, and CWB/var-NFS forecast, etc., are created for winter (January 2005) and summer (15 August to 15 September 2003) seasons over the CWB 45-km and 15-km domains.
The README file, source code and shell scripts for running the gen_be code are setup in CWB machine for reference: msccs1.msc.cwb:/users4/hclin/GEN_BE.
Moreover a utility program under wrfvar/da_3dvar/utl is also provided to plot the resulted BES (see http://www.mmm.ucar.edu/individual/guo/CWB/2005/WRFVAR_BE_Guo.htm on page 27). After completing the BES calculation, users just need to edit the shell script, wrfvar/da_3dvar/utl/plot_gen_be_nmc.csh and run it to obtain the gmeta file. An example of the figures of the eigenvectors for the different control variables is shown on Fig. 4.
(d) (c) (b) (a)
Figure 4. The first five eigenvectors for a) y, b) c_u, c) T_u, and d) rh derived from one summer month WRF forecast (15 August to 15 September
2003) over the CWB 45-km domain.
Task 2: Assimilation of GPS
Radio Occultation (RO) Data for Typhoon Prediction Using WRF 3D-Var System
A) Improvement of GPS RO observation error specification in WRF 3D-Var
In addition to the old GPS RO observation error estimate (Huang et al 2004), two new observation error specifications are tested in assimilation of CHAMP GPS RO data with the 3D-Var system: 1) Kuo et al. (2004) gave an GPS RO observation error estimate based on the winter month data (December 2001), called ÒKuo et al 2004Ó; 2) during the year of 2005,
(b)
Figure 5. The GPS RO observation error specifications. a) Chen-Kuo 2005. The red line indicates the percentage of GPS refractivity errors with height at North/South poles (f=90o), and the blue line indicates the error at equator (f=0o); b) Kuo et al 2004. The vertical coordinate is pressure; c) Huang et al 2004, the old one used in CWB 2004 project. The black thin curve in a) represents error percentages directly obtained from Chen-Kuo for CWB domain.
COSMIC/CDAAC re-processed the CHAMP GPS RO data and Shu-Ya Chen and Bill Kuo used one month (15 August to 15 September 2003) CHAMP GPS RO data from COSMIC/CDAAC (/cdaac/login/champ/level2/atmPrf) over the domain of the 45-km NFS model to derive the new observation errors, called ÒChen-Kuo 2005Ó. NCAR staff modified the observation preprocessor program to use these improved error specifications. Figure 5 showed three types of GPS RO observation error specifications. The newest GPS RO observation error specification, Chen-Kuo 2005, used the percentages, i.e. the ratio of error to the observed data, as a function of height and latitude at the data points. The observation error specification with the ÒKuo et al 2004Ó is the same as Chen-Kuo 2005 but as a function of pressure and latitude at the data points while ÒHuang et all 2004Ó defined the errors with the absolute values in N-unit and only as a function of pressure at the data points. The details on the observation error specifications can be found from http://www.mmm.ucar.edu/individual/guo/CWB/2005/Report_CWB_Oct2005.htm on page 8-9.
B) Development of GPS RO data quality control procedure in preprocessor
Because the quality control procedure requires the use of the Background fields (B), it is implemented in WRF 3D-Var (wrfvar) code, not in the observation preprocessor (3DVAR_OBSPROC). The procedure is adapted from NCEP GSI (Cucurull 2005, COSMIC) with minor modifications. There are three steps in this procedure:
1) innovation (O-B) check; the data will be discarded when (O-B) > 5so (observation error);
2)
relative error check: the relative error is defined as
, when 
or 
or 
the data will be
discarded;
3) Level data check: if the data at certain level below 7 km fails to pass the relative error check, all data below that level will be discarded. Usually the data at the low levels have lower confidence (see Fig. 5a) due to tracking errors, negative biases, and super refraction.
Figure 6a showed the total of 53 GPS RO soundings available during the Typhoon Dujuan period from 0900 UTC 28 to 1500 UTC 31 August 2003 from CHAMP downloaded from COSMIC/CDAAC. Figure 6b gave the innovations (O-B) (red lines) and the observation errors from Chen-Kuo 2005 (green lines) for all these GPS RO soundings. The background fields (B) are obtained from 6-h interval NCEP AVN analyses.
(b) (a)
Figure 6. GPS RO soundings from CHAMP and the innovations. a) the data distribution during the 78 hours period from 0900 UTC 28 to 1500 UTC 31 August 2003; b) innovations (O-B) from the 6-h interval NCEP AVN analysis and the observation errors [Kuo: Which one? Kuo et al. 2004? Chen-Kuo 2005?].
When the quality control procedure was applied with the different observation error specifications, different data rejection rates are obtained (as shown in Table 2).
|
Error specification |
Huang et al (2004) |
Kuo et al (2004) |
Chen-Kuo (2005) |
Chen-Kuo (2005) Perigee variable |
|
Ingested |
8084 |
8084 |
8084 |
7925 |
|
Qc step-1 |
0 (0.0%) |
1444 (17.862%) |
1048 (12.964%) |
927 (5.198%) |
|
Qc step-2 low level |
81 (1.002%) |
2 (0.025%) |
69 (0.854%) |
53 (0.508%) |
|
Qc step-2 middle level |
26 (0.322%) |
0 (0.000%) |
10 (0.124%) |
8 (0.076%) |
|
Data Assimilated |
7977 (98.68%) |
6638 (82.11%) |
6957 (86.06%) |
6937 (87.53%) |
Table 2. The number of the data points assimilated after quality Control
procedure. The percentages of the points relative to ingested points are shown
in parenthesis.
(c) (b) (a)
Figure 7. Innovations (O-B) after quality control for the different
observation error specifications. a) Huang et al 2004; b) Kuo et al 2004; and
c) Chen-Kuo 2005. The green curves are the observation errors.
With the old
observation error specification (Huang et al 2004), nearly all the ingested
data are assimilated (98.68%). The ÒChen-Kuo 2005Ó specification gave a
reasonable percentage (86.06%) for the amount of data points being assimilated,
and ÒKuo et al 2004Ó rejected a little bit more data than that of ÒChen-Kuo 2005Ó.
Figure 7 shows
the innovations of (O-B) after the quality control procedure is applied with
the different observation error specifications. Table 3 gave the results after
assimilation of the GPS RO data with 3D-Var for the different observation error
specifications. Still the ÒChen-Kuo 2005Ó error specifications produced the
most reasonable reduction of the analysis error 
= 36.07%, close
to a value of e-1.
|
Error specification |
Huang et al. 2004 |
Kuo et al. 2004 |
Chen-Kuo 2005 |
Chen-Kuo, 2005 Variable Perigee |
|
(O-B) bias |
0.076 |
0.013 |
-0.035 |
0.013 |
|
(O-A) bias |
0.136 |
0.008 |
-0.003 |
0.000 |
|
(O-B) RMS |
1.488 |
1.297 |
1.547 |
1.558 |
|
(O-A) RMS |
1.038 (69.76%) |
0.369 (28.45%) |
0.558 (36.07%) |
0.557 (35.75%) |
Table 3. Averaged (O-B) and (O-A) of GPS refractivity from thirteen 3D-Var experiments with 6-h assimilation time window for the different observation error specifications. The RMS reduction percentages (O-A)/(O-B) are shown in parenthesis. Here the letter ÒAÓ represents the analysis after 3D-Var assimilation.
Figure 8 shows the (O-A), the difference between observation and analysis after assimilation of the CHAMP GPS RO data with the different observation error specifications. With the old specification (Huang et al 2004), the (O-A) is very close to the (O-B) above the 7 km. ÒKuo et al 2004Ó and ÒChen-Kuo 2005Ó specifications obtained the reduction of (O-A) from (O-B) at all levels, the differences between these two specification are mainly at the low levels. Considering the relatively low accuracy of the GPS RO data at the lower levels, the ÒChen-Kuo 2005Ó observation error specification will be used in the numerical experiments and implemented in CWB semi-operational system.
(c) (a)
Figure 8. The differences between the observations and analysis (O-A) for
the different observation error specifications. a)
Huang et al 2004; b) Kuo et al 2004; and c) Chen-Kuo 2005. The green curves are
the observation errors.
C) Improvement of GPS RO observation operator
Because the ray-path perigee points from GPS measurements may vary in the vertical, the GPS RO soundings should not be considered as an upright profile, i.e. the location of the data points in a GPS RO soundings should be a function of height, {f(h), l(h)}. To consider the variable perigee points in the vertical, logically the observation operator should be modified in the WRF 3D-Var (wrfvar) code, but it may cause some difficulties in parallelization of the code. At moment, an alternative way is used to account for the variable perigee points in GPS RO soundings, i.e. each of soundings is broken to be multiple reports with a single level data and its location {f(h), l(h), h}. This work was completed in the wetPrf_decoder program before running WRF 3D-Var system, and the WRF 3D-Var system (3DVAR_OBSPROC and wrfvar code) do not need to be modified. When the variable perigee points in the vertical are considered, the number of ingested data points and the values of (O-B) are changed (see Column 5 in Table 2 and Table 3). Figure 9 shows the (O-A) after assimilation of the GPS RO soundings with variable perigee points in the vertical. By comparing Fig. 9 with Fig.8c, and Columns 5 with Column 4 in Table 2 and 3, considering the variable perigee points in soundings only gave very minor impacts on the results in this 45-km resolution CWB domain. Whether its impacts will be increased for a high resolution domain or not, and need to be further investigated.
Figure 9. The differences between the observations and analysis (O-A) from assimilation of the GPS RO soundings with the variable perigee points in the vertical. ÒChen-Kuo 2005Ó observation error specification is used; the green curves are the observation errors.
D) Assimilation of GPS RO data for Typhoon cases
With the improvements stated above, including the capability to assimilate the Typhoon and global bogus data fro CWB, the new GPS RO observation error specification and quality control procedure, etc., to mimic the operational environment, we conducted a series of numerical experiments for Typhoon Dujuan case in a 6-h cycling mode starting from 1200 UTC 28 to 1200 UTC 31 August 2003, using WRF_V2.1.1. It is impossible to run CWB NFS forecast model at NCAR (note: CWB staff has successfully carried out the retrospective experiments with wrfvar/NFS on CWB computer system). The 72-h forecasts initiated at 0000 and 1200 UTC are performed while only 6-h forecasts were made for model initiated at 0600 and 1800 UTC to provide the first guess to WRF 3D-Var (wrfvar). The experiment design is listed in Table 4.
Table 4. Experiment design for Typhoon Dujuan case with WRF 3D-Var.
|
Exp |
Initial condition |
Data assimilated |
First guess |
|
1 |
NCEP AVN |
Non |
Non |
|
2 |
6-h cycle WRF 3D-Var analysis |
CWB conventional data: SOUND, SYNOP, SATOB, AIREP,
PILOT, METAR, SHIPS, SATEM, and BUOY) |
6-h
forecasts except for the cycling starting time, 1200 UTC 28, which is from
NCEP AVN analysis. |
|
3 |
Same as Exp. 2 |
Same as Exp.2 plus CWB Typhoon and global bogus
data |
Same as Exp.2 |
|
4 |
Same as Exp. 2 |
Same as Exp.3 plus GPS RO data from CHAMP |
Same as Exp.2 |
|
5 |
WRF
3D-Var analysis at 0000 UTC 31 August (cold-start) |
Same as
Exp.4 |
NCEP AVN
analysis |
Figure 10 gave an example of 72-h track forecast initiated at 0000 UTC 31
August 2003 for experiment 2 (assimilation of CWB conventional data only), 3
(assimilation of CWB conventional data plus bogus data), and 4 (assimilation of
CWB conventional and bogus data plus CHAMP GPS RO data). Fig.10b shows the
track forecast errors for Exp.2, 3, and 4, and the averaged 72-h forecast
errors are 140.3 km for Exp.2, 64.3 km for Exp.3, and 57.4 km for Exp.4. This
shows that assimilation of CHAMP GPS RO data gave the highest accuracy for
Typhoon Dujuan track forecast, and without assimilation of CWB bogus data and
CHAMP GPS data (e.g., using the NCEP AVN initial condition) the largest
forecast errors are 
obtained.
0 9 18 27 36 45
54 63 72 Forecast time Figure 10.
72-h track forecast for Exp.2, 3, and 4 initiated at 0000 UTC 31 August
2003. a) Tracks: observed (black), Exp.2 (red), Exp.3 (green), and Exp4
(blue); b) 72-h Track errors for Exp.2, 3, and 4 (the ticks represent 3h
interval).
To obtain the track forecast accuracies for all five initial times at 1200 UTC 29, 0000 and 1200 UTC 30, and 0000 and 1200 UTC 31 August 2003, we defined a 24-h averaged increment of track errors InER:
where N = 8 with the track forecast in a 3-h interval for a period of 24 hours.
The Error(Exp.#) is defined as
Table 5 and 6 gave the comparisons between Exp.4 and 3 (with/without assimilation of CHAMP GPS RO data) and between Exp.3 and 2 (with and without assimilation of CWB bogus data) for track errors. Based on the definition of InER(A,B) above, a positive value means Exp.B is better than Exp.A, and a negative value indicates Exp.A better than Exp.B.
Table 5. The 24 hours Average Increment of
Track error InER( 3,4 ) in km
|
Init. Time |
29_12 |
30_00 |
30_12 |
31_00 |
31_12 |
Ave |
|
00-24h |
11.885 |
8.935 |
-5.07 |
7.863 |
19.173 |
8.557 |
|
24-48h |
10.15 |
4.305 |
-18.823 |
11.138 |
4.95 |
2.344 |
|
48-72h |
13.803 |
10.075 |
-11.255 |
5.21 |
6.22 |
4.811 |
|
Ave |
11.946 |
7.772 |
-11.712 |
8.07 |
10.114 |
5.237 |
Table 6. The 24 hours Average Increment of
Track error InER(2,3) in km
|
Init. Time |
29_12 |
30_00 |
30_12 |
31_00 |
31_12 |
Ave |
|
00-24h |
0.1975 |
162.59 |
196.75 |
107.84 |
-12.19 |
91.039 |
|
24-48h |
-30.143 |
205.28 |
31.003 |
65.19 |
86.443 |
71.554 |
|
48-72h |
23.233 |
225.238 |
23.238 |
34.225 |
142.887 |
89.764 |
|
Ave |
-2.236 |
197.702 |
83.663 |
69.086 |
72.381 |
84.119 |
From Table 5, except for the forecast initiated at 1200 UTC 30 August, the Typhoon track forecast with assimilation of CHAMP GPS RO data in addition to the CWB conventional and bogus data gave better track than those without assimilation of the GPS RO data. The averaged improvement over 5 initial times and three 24-h periods (00-24h, 24-48h, and 48-72h) is 5.237 km. Since the mean track error for Exp. 3 (which assimilated conventional data and bogus data that CWB routinely uses) is 64 km for a 72-h forecast and about 50 km for a 36-h forecast, this implies that the assimilation of GPS RO data from CHAMP produced approximately 10% improvement in terms of track forecast. Keeping in mind that only very few CHAMP GPS RO soundings are available (53 soundings over a period of 78 hours) over the CWB 45-km regional domain, this improvement is very encouraging to show the positive impact of GPS RO data.
From Table 6, except for two cases: 24-48h period initiated at 1200 UTC 29 and 00-24h period initiated at 1200 UTC 31 August, as expected, assimilation of CWB bogus data in addition to the conventional data gave significant improvement of the track forecast. The averaged improvement is 84.119 km. This was due to the fact that the initial typhoon vortex was not properly described in the NCEP AVN global analysis.
Overall, through the numerical experiments for Typhoon Dujuan on CWB 45-km domain, the new capabilities developed in WRF 3D-Var system are very effective. Increasing the model resolution and fine tuning the BES in WRF 3D-Var may lead the further improvements of Typhoon forecasts.
Task 3: Continued
Interaction on WRF 3D-Var System and GPS RO Data Assimilation
1) As mentioned above, the CWB project web sites at both UCAR and CWB were setup this year: http://www.mmm.ucar.edu/individual/guo/CWB/CWB_Project.htm and http://cwb-iria.cwb.gov.tw/cwb2ncar/frame.htm. This is much more helpful for timely interaction between UCAR and CWB staff, for up-to-date code development, and for sharing the results from experiments, the data files, and the documents, reports and paper, etc. UCAR staff updated the CWB_Project web page frequently, and informed the changes to CWB staffs.
2) UCAR informed CWB on 2005 Annual WRF/MM5 workshop, WRF-ARW summer tutorial, and Second GPS RO Data UserÕs workshop. Dr. Terng from CWB attended the Second GPS RO Data UserÕs workshop held in Washington, D.C., 22-24 August 2005, and visited UCAR for a month followed the workshop. Mr. Shen, Director of IT Division, CWB, visited UCAR in October 2005, and UCAR staff reported the project status to him.
3. Future
plan of the work
Based on
discussion with Dr. Terng, we identified the following tasks for year 2006. These proposed tasks will
have to be prioritized based on availability of funding.
Task-1. Improvement of the WRF 3D-Var/WRF
semi-operational system on CWBÕs newly procured computer
n
Compile all the system codes
successfully, and improve the efficiency to meet the operational requirement;
n
For Typhoon prediction,
increase of model resolution (15-km, 5-km, etc.) is the first priority for
improving the skill of forecast.
n
For the cycling mode, the
frequent cycles, such as 3-h or 1-h, should be explored to account for the
asynoptic observations, especially for GPS RO data
n
Develop the WRF
3D-Var(wrfvar)-based observation verification package
.25
FTE NCAR
Task-2.
Improvement of the WRF 3D-Var system
n
Assimilation of more
observations, such as QuikScat, GPS ZTD, AWS, BUFR GPS RO data, etc.
n
Test the WRF 3D-Var FGAT
technique with CWB cases;
n
New type (cv_options=5) of
the background error statistics (BES) need to be derived based on CWB NFS data;
Develop
the code for application of BES over a different domain;
Conduct
the subjective tuning of the BES based on the single OBS tests and a selected
period of analysis/forecast experiments;
Derive
the BES season by season and does the impact study of BES changes on analysis
and forecast;
Develop
the objective BES and observation error statistics (OES) tuning techniques
1.0
FTE NCAR
Task-3.
Continued interaction on WRF3D-Var system
n
Keep CWB staff up-to-date on
current developments;
n
Update and improve the NCAR
CWB project web page timely.
n
.25 FTE NCAR
Task-4.
Improvement of the GPS RO assimilation
n
When other approaches of
assimilating GPS RO data, such as non-local observation operator, bending angel
assimilation, etc., are ready, more comparison study should be conducted
between the local observation operator and other operators. More GPS RO will be
available in 2006 following the launch of FORMOSAT-3/COSMIC.
n
Conduct data assimilation
experiments in support of IOP campaign to be led by Dr. Frank Cheng at NSPO and
TaiwanÕs science community.
n
1.5 FTE NCAR
These
tasks will all require substantial involvement of CWB staff. We anticipate that
CWB will participate at the level of 2 FTE, in addition to Ms. Hui-Chuan Lin.
References
Appendix A. System set-up for WRF
3D-Var/NFS
初步NCAR/WRFVAR與氣象局觀測資料及NFS模式資料結合等功能已於去年（2004
計晝建置完成。
NCAR於2005年8月5日發佈新版WRF系統，目前版本為V2.1.1。WRF系統中有關觀測資料處理部份由去年（2004年）的WRF-3DVAR更名為WRF Var，因為觀測資料變分分析系統不僅有3D-Var，還包含背景場誤差統計萛程式，未來還將包含4D-Var。
WRF-Var目前版本為V2.1。WRF-Var V2.1主要新增特點與功能包括：
（1）使用WRF model V2.1架構
（2）背景場誤差統計計萛程式
（3）可處理Global bogus data
（4）可處理typhoon bogus data
目前本計畫所建置的WRF-Var在氣象局的作業平台為資訊中心VPP5000（vpp5000.mic.cwb）及衛星中心PC-cluster（msccs1.msc.cwb）。
在使用者帳號hclin之下，建立一套WRFVAR系統，包含說明檔、原始程式、執行檔、資料檔及shell
scripts等，提供氣象局人員建立作業流程的依據。
氣象局人員已完成2003年8月15日至2003年9月15日（夏季月份）以及2005年1月份（冬季月份）各一個月的WRF-Var結合NFS的cycling預報實驗，並提供該預報結果以計算NFS背景場誤差值（計算過程與結果於本報告說明）。
主要目錄架構如下：
|
|
|
|
~hclin/WRFVAR/README |
|
|
/README.changes |
檔案：記錄針對目前版本增修的原因與日期。 |
|
/README.upgrade |
檔案：說明目前版本（WRFVAR V2.1）與前版本（WRF-3DVAR）主要差異。 |
|
/bin |
目錄：存放執行檔與shell scripts |
|
/dat |
目錄：存放資料檔 |
|
/etc |
目錄：存放namelists範本檔 |
|
/lib |
目錄：存放環境變數設定檔 |
|
/log |
目錄：存放執行過程記錄檔 |
|
/src/3DVAR_OBSPROC_v2.1 |
目錄：存放觀測資料前處理程式 |
|
/OBS_FGGE_PROC |
目錄：存放氣象局觀測資料格式轉換程式 |
|
/nfs2wrfvar |
目錄：存放氣象局NFS模式資料轉換為WRF-Var格式的程式 |
|
/wrfvar_2.1 |
目錄：存放WRF-Var V2.1程式 |
|
/wrfvar_v2.1 |
目錄：存放WRF-Var V2.1程式 |
|
/wrfvar2nfs |
目錄：存放WRF-Var分析資料轉換為氣象局NFS模式資料格式的程式 （註：由於該程式無法以msccs1.msc.cwb的PGF
5.0.1版本編譯，目前慬能在vpp5000.mic.cwb執行） |
|
/workdir |
目錄：執行工作目錄 |
WRFVAR/README.upgrade內容
|
Notice on upgrading WRF3DVAR to WRFVAR_V2.1 (My WRF3DVAR system is in ~hclin/3DVAR
WRFVAR system is in
~hclin/WRFVAR) Use updated executables: cwbobs.exe OBS_FGGE_PROC/src/cwbobs.exe maintained by Hui-Chuan Lin
3dvar_obs.exe
3DVAR_OBSPROC_v2.1/src/3dvar_obs.exe from NCAR downloads
nfs2wrfvar.exe
nfs2wrfvar/nfs2wrfvar.exe
maintained by Hui-Chuan Lin
wrfvar.exe wrfvar_v2.1/main/wrfvar.exe from NCAR downloads NOTE:
previous nfs2wrf3dvar.exe is
replaced by nfs2wrfvar.exe previous
da_3dvar.exe
is replaced by wrfvar.exe Use updated namelists: Changes
have been made to namelist.cwbobs
namelist.nfs
namelist.3dvar
namelist.input Please see ~hclin/WRFVAR/etc/nml for complete namelists
templates |
WRFVAR/README.changes內容
|
20051110: (1) added
src/README to clarify the source of each
program (2)
downloaded 3DVAR_OBSPROC.tar.gz (version 20051025) from
http://www.mmm.ucar.edu/individual/guo/CWB/CWB_Project.htm The code is untarred and compiled
in ~hclin/WRFVAR/src/3DVAR_OBSPROC (3)
downloaded wrfvar_2.1.tar.gz (version 20051027) from
http://www.mmm.ucar.edu/individual/guo/CWB/CWB_Project.htm The code is untarred and compiled
in ~hclin/WRFVAR/src/wrfvar_2.1 Read
wrfvar_2.1/da_3dvar/changes/change_20051027.yrg for the changes. 20051007: (1) added
src/nfs2wrfvar/README (2) added
src/plotobs/README 20051006: (1)
modified src/nfs2wrfvar for reading 34-key DMS data affected files:
src/nfs2wrfvar/Makefile
src/nfs2wrfvar/module_NFS_DMS.F
src/nfs2wrfvar/nfs2wrfvar.exe (2)
corrected bin/nfs2wrfvar.csh for "for_bes" application 20050929: (1)
modified src/OBS_FGGE_PROC/src/cwbobs.F to change the qscat file_prefix from "gtdf"
to "rtdf". src/OBS_FGGE_PROC/src/cwbobs.exe
is updated. (2) added
src/OBS_FGGE_PROC/README (3)
downloaded wrfvar_2.1.tar.gz from
http://www.mmm.ucar.edu/individual/guo/CWB/CWB_Project.htm The code is untarred and compiled
in ~hclin/WRFVAR/src/wrfvar_2.1 Read wrfvar_2.1/da_3dvar/changes/change_20050921.yrg
and
wrfvar_2.1/da_3dvar/changes/change_20050923.yrg for the changes that have been
made to the official released code
(~hclin/WRFVAR/src/wrfvar_v2.1) (4) added
~hclin/bin/julianday.f90, a program to get julian_day from input ccyymmdd. The executable is
~hclin/bin/julianday (5) added
script ~hclin/WRFVAR/bin/get_wetPrf.csh for getting CHAMP data from CDAAC/COSMIC web site. Read
~hclin/WRFVAR/bin/get_wetPrf.csh for more details. *** NOTE *** get_wetPrf.csh can
not be run on msccs1.msc.cwb
because of internet restriction. |
上述的README.upgrade與README.changes主要目的為記錄在氣象局所建置WRFVAR系統的變動。
另外在WRFVAR/src/每個程式主目錄下，均包含README檔案，主要說明該程式目的、架構及執行方法等。
WRFVAR/src/3DVAR_OBSPROC及WRFVAR/src/wrfvar_v2.1為NCAR提供程式，除了參考該目錄下的README檔案外，也可參考網頁http://www.mmm.ucar.edu/wrf/WG4/wrfvar.htm。
WRFVAR/src/OBS_FGGE_PROC、WRFVAR/src/plotobs、WRFVAR/src/nfs2wrfvar及WRFVAR/src/wrfvar2nfs為依據氣象局需求另撰寫的程式
OBS_FGGE_PROC/README內容如下：
|
Hui-Chuan Lin, Sept 29, 2005 1. Part of this program is adapted from cwbobs used in
AOAWS MM5 system. 2. This program includes
OBS_FGGE_PROC/gts_sttnid_final
STATION.METAR
namelist.cwbobs
src/Makefile
cwbobs.F
module_airep.F
module_bathy.F
module_bogus_gfs.F
module_bogus_ty.F
module_buoy.F
module_drop.F
module_hisatw.F
module_metar.F
module_misc.F
module_output.F
module_pilot.F
module_qscat.F
module_satem.F
module_satob.F
module_ship.F
module_synop.F
module_temp.F module_tempship.F
params.inc 3. To compile: cd OBS_FGGE_PROC/src vi Makefile (if
necessary) make
(OBS_FGGE_PROC/src/cwbobs.exe will be generated) 4. To run: cd your-working-dir ln -s -f
OBS_FGGE_PROC/gts_sttnid_final . ln -s -f
OBS_FGGE_PROC/STATION.METAR . vi namelist.cwbobs OBS_FGGE_PROC/src/cwbobs.exe 5. file description: STATION.METAR and gts-sttnid-final
are two 'constant' files used by program
OBS_FGGE_PROC. They are maintained by Jim Bresch
at NCAR. gts_sttnid_final: is a global WMO
station table and contains the
WMO numbers for sounding and surface sites.
It is used for getting WMO station elevation
when elevation info is missing in CWB observation record
or when there is a big difference between elevation from
CWB and from gts-sttnid-final. STATION.METAR: is a subset of a
global METAR station table. It covers CAA
Domain 1. It is used for getting lat and lon for METAR
stations, because sometimes there are lat/lon mismatches
between lat/lon from CWB
and from STATION.METAR. namelist.cwbobs:
&record1
obs_rootdir = '/users4/hclin/3DVAR/data/obs/cwbfgge',
; the directory where FGGE observations are.
obs_date = '2005-02-27_12:00:00',
gfsbogus_as_bogus = .true.
; if .true. (default), global bogus data are labled as
; "FM-135 BOGUS" and will be treated as bogus data in
WRFVAR.
; if .false., global bogus data are labled as "FM-35 TEMP"
; and will be treated as sounding data in WRFVAR. /
&record2
use_airep = .TRUE., ; default is .FALSE.
use_bogus_gfs =
.TRUE., ; default is .FALSE.
use_bogus_ty = .FALSE.,
; default is .FALSE.
use_metar = .TRUE., ; default is .FALSE.
use_ship = .TRUE., ; default is .FALSE.
use_bathy = .TRUE., ; default is .FALSE.
use_buoy = .TRUE., ; default is .FALSE.
use_satob = .TRUE., ; default is .FALSE.
use_satem = .TRUE., ; default is .FALSE.
use_pilot = .TRUE., ; default is .FALSE.
use_temp = .TRUE., ; default is .FALSE.
use_tempship =
.TRUE., ; default is .FALSE.
use_drop = .TRUE., ; default is .FALSE.
use_hisatw
= .TRUE., ; default is
.FALSE.
use_synop = .TRUE., ; default is .FALSE.
use_qscat = .TRUE. ; default is .FALSE. / |
plotobs/README內容如下：
|
Hui-Chuan Lin, 2005 1. This program is adapted from 3DVAR_OBSPROC/MAP_plot
in WRFVAR system and
plot_level utility in LITTLE_R/MM5 system. 2. This program plots the observations contained in
obs_gts.3dvar
(the 3DVAR_OBSPROC output). 3. This program includes
plotobs/src/Makefile
DA_Constants.F
DA_Define_Structures.F
module_map_stuff.F
module_ncarg.F
module_obs.F
plotobs.F
DA_Read_Obs.inc
DA_Read_Obs_Info.inc
finddtg.inc
plotloc.inc
plotval.inc
plotval2.inc
plotval3.inc 4. To compile: cd plotobs/src vi Makefile (if necessary) make
(plotobs/src/plotobs.exe will be generated) 5. To run: cd your-working-dir ln -s -f obs_gts.3dvar fort.99 plotobs/src/plotobs.exe |
nfs2wrfvar/README內容如下：
|
Hui-Chuan Lin, 2005 1. This program is used to convert DMS-format CWB NFS
sigma-level data to
be in WRF-like-format eta-level data.
With "if_for_var_anal" choice in namelist.nfs (described
below), the
outputs can be used in either wrfvar (for single-time analysis) or gen_be (for mulitple-time
statistics). 2. This program includes
nfs2wrfvar/Makefile
define_cons_types.F
module_NFS_DMS.F
module_WRF_fields.F
module_allocate.F
module_vinterp.F nfs2wrfvar.F 3. To compile: cd nfs2wrfvar vi Makefile (choose proper options
for VPP5000 and msccs1, also
see notes described below) make
(nfs2wrfvar/nfs2wrfvar.exe will be generated) *NOTE*
on nfs2wrfvar/Makefile -Ddms24key: for
reading 24-key DMS data -DLINUX: for
running nfs2wrfvar.exe on msccs1, where the
DMS data is in single precision. -Dluindexextfile: for reading
Land-Use-Index from a file.
Environment variable LUINDEX_D0? must be set
before running nfs2wrfvar.exe. -Dterrainextfile: for reading
terrain height data from a file.
Environment variable TERRAIN_D0? must be set
before running nfs2wrfvar.exe. -Dmapextfile: for reading mapping
factors, coriolis parameters,
latitude, and longitude from files. Environment
variable MAPPINGDAT_D0? and LATLON_D0? must be
set before running nfs2wrfvar.exe. 4. To run: cd your-working-dir setenv CWBNFSL
CWBNFSL@/nwp/nwpnfs/.DMSDATA/NWPDB@nwpnfs@vpp5000.mic.cwb setenv
MAPBKG CWBNFSL@/nwp/nwpnfs/.DMSDATA/NWPDB@nwpnfs@vpp5000.mic.cwb setenv
LUINDEX_D0[1|2] LU_INDEX_[45|15]_LD setenv TERRAIN_D0? (if compiled
with -Dterrainextfile) setenv MAPPINGDAT_D0 and
LATLON_D0? (if compiled with -Dmapextfile) vi namelist.nfs nfs2wrfvar/nfs2wrfvar.exe 5. file description: LU_INDEX_[45|15|05]_LD are in
ascii format and are extracted from
wrfsi output using "read_wrf_nc -w LU_INDEX"
read_wrf_nc is a WRF utility available on
http://www.mmm.ucar.edu/wrf/users/download/get_source2.html 6. namelist.nfs &run_type if_for_var_anal
= .TRUE. ; if .TRUE., the output (wrf3dvarinput_d0?)
;
is used to run wrfvar ;
if .FALSE., the outputs
;
(initial_date/nfsout_d0?_ccyy-mm-dd_hh:00:00)
; are
used to run gen_be / &record0 initial_date = '2005100500', ; used when if_for_var_anal=.FALSE. forecast_hours = 12, 24 ; used when
if_for_var_anal=.FALSE.
; the maximum dimension
for forecast_hours is 20
; in this example,
; 2005100500/nfsout_d0?_2005-10-05_12:00:00
and
;
2005100500/nfsout_d0?_2005-10-06_00:00:00
; will be generated.
; *NOTE* the directory
2005100500 must be
;
created first in the script. / &record1 analysis_date = '2005100500', ; used when
if_for_var_anal=.TRUE. itau_prev_forecast = 6, ; used when
if_for_var_anal=.TRUE.
; in this example, 6-hour
forecast from
; initial time 2005100418
will be the
; first guess for the
analysis_date 2005100500 / &record2 domain_id
= '2', dsm =
15000., nx_nfs =
181, ny_nfs =
193, nz_nfs =
30,
; number of half sigma levels dt = 40., deltz_nfs
= 0.0070, 0.0088, 0.0106, 0.0124, 0.0141, 0.0159, 0.0237, 0.0306, 0.0367,
0.0419, 0.0462, 0.0497, 0.0522, 0.0539, 0.0547, 0.0546, 0.0520, 0.0493,
0.0467, 0.0440, 0.0414, 0.0387, 0.0361, 0.0335, 0.0308, 0.0282, 0.0255,
0.0208, 0.0200, 0.0200, 30*0.0, ; d (sigma) values between full levels p_top =
0., / &record3 map_projection = 1, clat =
22.86559, ; current domain
center latitude clon =
122.2716, ; current domain
center longitude stand_lon
= 120., truelat1 =
10., truelat2 =
40., / &record4 nz_wrf =
30,
; number of half eta levels deltz_wrf
= 0.0070, 0.0088, 0.0106, 0.0124, 0.0141, 0.0159, 0.0237, 0.0306, 0.0367,
0.0419, 0.0462, 0.0497, 0.0522, 0.0539, 0.0547, 0.0546, 0.0520, 0.0493,
0.0467, 0.0440, 0.0414, 0.0387, 0.0361, 0.0335, 0.0308, 0.0282, 0.0255,
0.0208, 0.0200, 0.0200, 30*0.0, ; d (eta) values between full levels / |