Final Report for 2006 CWB project

Mesoscale and Microscale Meteorology Division

National Center for Atmospherics Research

Boulder, Colorado, USA

Submitted to Central Weather Bureau, Taiwan, ROC

2530 November 2006

1. Introduction

Following the 2005 CWB-UCAR collaborative project, the WRF 3DVAR system, now called as “wrfvar”, has been upgraded by NCAR staff. The accomplishments are: 1) the WRFVar 2.1 system including the auxiliary codes have been successfully migrated to the CWB IBM computer; 2) the background error statistics (BES) estimates (cv_options=5) for summer and winter months are derived based on the CWB NFS model forecasts, and the BES interpolation capability has been developed; 3) a WRFVar-based observation verification package was developed; 4) more types of observations, QuickSCAT, AWS, Dropsonde, GPSRO bufr data, and CWB GPSPW data, can now be ingested into the WRFVar assimilation system; 5) the WRFVar FGAT technique have been tested; and also 6) Mr. Eric Chiang from CWB has visited to NCAR for 6 months.

In order to exchange the information and the codes, a web page for CWB project (http://boxwww.mmm.ucar.edu/people/guo/individual_/guo/CWB/CWB_Project_2006.htm) is frequently updated. CWB also sets up a web site (http://cwb-iria.cwb.gov.tw/cwb2ncar/frame.htm) for “Data Exchange Interface”. 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.

2. Review of the project tasks and accomplishments

Task 1: Establishing a WRF-Var/NFS Operational System on CWB’s IBM Computer

A) Successfully compile all of the system codes, and improve the efficiency to meet the operational requirementsTechnical consultancy on retrospective experiments in CWB operational environment

The new machine, IBM p5-575 Cluster 1600, has been installed in CWB in 2006, which has 2496 CPUs (156 nodes, 13 Frames). The CPU speed is 1.5GHz and the machine has SMT Dual core, Superscalar Architecture, and the peak performance is 6GFlops. With the joint efforts, all the software of the WRF-Var system, including OBS_FGGE_PROC, 3DVAR_OBSPROC, NFS2WRF, WRFVar, WRF2NFS are running smoothly on the new machine.

CWB staff s have conducted many tests using the NFS forecast model, including the monthly pre-operational run from 15 August to 15 September 2003, Typhoon track forecast for Dujuan, Krovan, Maemir (2003), Haitang (2005), and Bilis, Kaemi, and Shanshan (2006), especially for assessing the impact of COSMIC GPS RO data.

Mr. Eric Chiang also carried out the WRF-Var+WRF forecast model tests in CWB IBM machine for GPSPW experiments and the BES tuning experiments. He also conducted a few experiments with 45/15/5-km 3 nests experiments for the precipitation forecasts with WRF model.

Note that for the NFS and WRF have the same domain size for 45-km (coarse) domain, but for the 15-km nested domain, NFS has the size of 181x193, and WRF has the size of 184x196 because NFS and WRF have different C-Grid staggering.

On the NCAR side, many experiments in cycling mode with WRF-Var+WRF nested domain (45/15-km) were also conducted for Typhoon Dujuan and Haitang. Also the coarse domain size is the same as CWB (222x128), but the 15-km nested domain size is 178x190.

Two main updates completed recently include:

1) A beta version of WRF-Var was posted on the NCAR-CWB web page:

http://www.mmm.ucar.edu/people/guo/individual_/guo/CWB/CWB_Project_2006.html

It should be noted that WRF and WRF-Var need to be compatible in terms of versions. Therefore, WRF-Var 2.2 can accept first guess fields from WRF2.2 or WRF2.1. However, the output from WRF-Var 2.2 can only be used to initialize WRF2.2 (not WRF2.1).

2) The 3DVAR_OBSPROC was modified to produce both ASCII and NCEP prebufr format observation files, and the GPSRO data can be used for both local and non-local operators.

B) Derivation of the new background error statistics (cv_options=5) based on the CWB NFS forecast data

The new background error statistics files are derived based on the NFS forecast datasets:

Winter: 2005010100Z to 2005013112Z 6-h cycling WRF-Var(cv=3)/NFS run

Summer: 2003081500Z to 2003091512Z 6-h WRF-Var(cv=3)/NFS cycling run

For CWB WRF domains, the BES files are generated by Eric Chiang and for the NCAR WRF domains, they are generated by Hui-Chuan Lin using the gen_be code in WRF-Var with bin_type=1.

The BES interpolation capability was developed for WRF-Var 2.1 code by NCAR staff. This allows a BES file derived from a different domain settings, i.e. different geographic location, different (horizontal and vertical) dimensions of the domains, etc., can be applied to the current WRF 3DVar experiments, which is extremely useful for the domain settingsmoving nests application in are changed as the tropical storm, Typhoon or Hurricane movingTyphoon forecast.

To obtain improved performance of the WRF-Var, some BES tuning experiments were conducted for Typhoon Haitang by NCAR staff and by CWB staff (Eric Chiang). We found that

tThe technique using multiple external loops with the different tuning factors to CV5 BES can give the correct analysis of typhoon location. With the cycling mode runs, this WRF-Var technique, in general, also gives the best forecast of the typhoon track and intensity (download from http://www.mmm.ucar.edu/events/2006wrfusers/agenda.php by click P4.2). This technique has not been released in the WRF-Var code downloadable from NCAR-CWB web page because some additional refinement is needed.

Mr. Chiang (CWB) has also done many experiments for scale-length tuning for the WRF-Var/WRF. He found that the tuning factor of 0.5 to scale-length improved the forecast scores in the summer. In winter, there are no significant differences between the tuning factors of 0.5 and 1.0. The results have been posted on the NCAR-CWB web page.

The objective BES tuning techniques based on the approaches from Hollingsworth and Lonnberg (1986) and Desroziers and Ivanov (2001) are under development. Those should be available in the official release version of WRF-Var 2.2 in the spring of 2007.

C) Develop the WRF-Var based observation verification package

Because WRF-Var calculated (O-B) for all types of the observations, based on these information it is easy to compute the bias error, RMS error, etc., forecast scores. The challenges are to get the exactly same QCed (Quality-Controlled) observations, against which all experiments are verified. The necessary code and the document have been posted on the NCAR-CWB web page

http://boxwww.mmm.ucar.edu/people/guo/individual_/guo/CWB/CWB_Project_2006.html

For the 2006 project, NCAR and CWB staff has used this package for many types of experiments, for example, Chiang’s GPSPW experiments below.

Task 2: Enhancement of the WRF-Var System

A) Assimilation of more observations, such as QuikSCAT, AWS, and GPSRO data in BUFR format

Several FGAT experiments have been conducted at NCAR with WRF-Var for Typhoon Haitang. The data assimilated include SYNOP, METAR, SHIP, BUOY, SOUND, AIREP PILOT, SATOB, GPSREF, QSCAT, and SATEM. The following setting is the best one for WRF-Var FGAT experiment. That is the cold-start with SI+WRF-Var at 2005071400Z, then for each of the 6-h cycles starting from 2005071406Z, using 3h to 9h forecast from the previous initial time as the first guess for WRF-Var. The schematic diagram is shown in Fig.1 below.

With the FGAT technique, much more non-conventional observations, such as SATOB, GPSREF, SATEM, QSCAT, and SYNOP (AWS), can be ingested by WRF-Var. From the Table.1, FGAT technique, in general, gave the improved Typhoon track forecast skill except the initial time at 2005071612Z. The details of the FGAT results can be found from NCAR CWB project web page: Hui-Chuan Lin’s report for FGAT and AWS assimilation.

Fig.1 Schematic diagram for WRF-Var FGAT experiments.

With the FGAT technique, much more non-conventional observations, such as SATOB, GPSREF, SATEM, QSCAT, and SYNOP(AWS), can be ingested by WRF-Var. From the Table.1, FGAT technique, in general, gave the improved Typhoon track forecast skill except the initial time at 2005071612Z.

Table.1 Typhoon Haitang 72-h averaged track forecast errors with FGAT technique

Exp.	1600Z	1612Z	1700Z	1712Z
NOFGAT	106.82	85.58	96.09	151.86
FGAT	104.01	98.44	70.38	114.72
FGAT(AWS+MESONET)	105.77	98.10	76.12	111.12

The QuikSCAT data have already been included in CWB operational database in FGGE format. Our OBS_FGGE_PROC can convert the QuikSCAT FGGE data to LITTLE_R format used by WRF-Var system. In the LITTLE_R file, the wind speed error is stored in the field of u-component, and the direction error is stored in the field of v-component. In WRF-Var, the minimum QuikSCAT wind errors are allowed to be 1 m/s because sometime the original QuikSCAT data give very small observation errors.

This year, CWB has more observation datasets available, such as AWS and mesonet data (Fig.2). The OBS_FGGE_PROC was updated to decode these data and assign them as the type of SYNOP in LITTLE_R file. This updated decoder program has been posted on the NCAR-CWB web page. Figure 2 showed the surface weather stations in Taiwan area. Because the AWS and mesonet data have high temporary resolution, with the FGAT technique, there are 587 SYNOP data assimilated within 15-km domain, but only 158 SYNOP data used in no FGAT assimilation. WRF-Var works well with CWB AWS and mesonet data, but as shown in Table.1, the added value with these data for Typhoon track forecast is rather neutral as expected. These high resolution data may have more impact on the weather forecast in Taiwan, not Typhoon track forecasts.

Figure 2. Mesoscale surface observation sites in Taiwan area

With the launch of FORMOSAT-3/COSMIC mission, the GPSRO soundings are available for operational use in BUFR format. NCAR staffs have developed the GPSRO BUFR data processing software, including a) data download shell script; b) gps_bufr_decoder program; and c) modified 3DVAR_OBSPROC program. All these programs are posted on the NCAR-CWB web page. The GPSRO data processing procedure is shown in Fig.3. NCAR staff has also used the GPSRO BUFR data assimilation for Hurricane Florence. The results can be seen from

http://box.mmm.ucar.edu/people/guo/individual_guo/CWB/CWB_Project_2006.html

under section “F, Typhoon Florence”.

CWB staff has already performed considerable testing on the FORMOSAT-3/COSMIC data assimilation for 2006 Typhoon forecast. They may still use the GPSRO wetPrf netCDF data. We suggest using the GPSRO BUFR data for future assimilation experiments.

Figure 3. GPSRO BUFR data processing procedure.

CWB staff has already performed considerable testing on the FORMOSAT-3/COSMIC data assimilation for 2006 Typhoon forecast. They may still use the GPSRO wetPrf netCDF data. We suggest using the GPSRO BUFR data for future assimilation experiments.

B) Test the WRF-Var FGAT technique with asynoptic observations

The preliminary results have been presented in the last subsection. We suggest that CWB can try the FGAT in future. The working code is already included in WRF-Var 2.1, no code modifications needed, just minor changes are needed in the running shell scripts.

C) Assimilation of ground-based GPS PW data

Figure 4, ground-based GPS PW observation sites in Taiwan at 1200 UTC 16 July 2005

C) Assimilation of ground-based GPS PW data

There are 107 ground-based GPS sites in Taiwan area now. In order to get the PW, the surface meteorological parameters, pressure and temperature, must be available.

Figure 4, ground-based GPS PW observation sites in Taiwan at 1200 UTC 16 July 2005

At an interval of 30-minute, there are about 50 PW data obtained over the Taiwan ground-based GPS network. Figure 4 shows the GPSPW assimilated at 1200 UTC 16 July 2005.

CWB provided one month GPS raw data from 1 to 31 July 2005. John Braun at COSMIC Office processed these data with the software Bernes to get the PW text file. The NCAR/MMM staff decoded these text files and convert it to the LITTLE_R format files as input to WRF-Var system. The GPS assimilation experiments have been performed for Typhoon Haitang case (2005071400Z to 2005072012Z) by Eric Chiang, during this visit to NCAR. In general, GPS assimilation has the positive impact on the PW forecasts. The RMS errors verifying against the observed PW were improved by 1.8%. For Typhoon Haitang track forecast initiated at 0600 UTC 17 July (before landfall in Taiwan) and 12 UTC 18 July (left from Taiwan west coast), the GPS PW assimilation also showed minor improvements. The precipitation forecast also looks improved by subjective evaluation, but we do need a verification package for rainfall to obtain an objective evaluation. The details of the results can be found from NCAR-CWB web page

http://box.mmm.ucar.edu/people/guo/individual_guo/CWB/CWB_Project_2006.html

under the section: “B, Project progresses, 2) Status of 2006 CWB project”.

Task 3: Continued interaction on WRF-Var

A) Update and improve the CWB project web pages on both CWB and NCAR/MMM sides

1) The NCAR-CWB web page was frequently updated to post all the bug fixes, new development, working progresses, etc. On this page, the standard versions of all the codes are provided to CWB staff.

2) Eric Chiang from CWB vested to NCAR for six months, and work successfully with NCAR staff. His work mainly focused on the Background error statistic tuning and GPSPW assimilation.

3. Future plan of the work

Based on the work in 2006, the following tasks for the year of 2007 are proposed:

0)Migrate the WRF-Var system from version 2.1 to version 2.2

Both of WRF model and WRF-Var at NCAR are upgraded to Version 2.2. There are many changes and new developments in version 2.2. For WRF-Var component, the verification package will be improved, the BES tuning code will be working properly, the radiance data can be assimilated directly, and also 4D-Var capability is included, etc. CWB should be prepared to obtain these new capabilities from NCAR. Because there are changes in the input files in Version 2.2, cares need to be paid to the WRF-Var/NFS configuration for NFS2WRF and WRF2NFS interface program. So we suggest that in WRF-Var/NFS configuration, CWB may still keep to use WRF-Var Version 2.1. But in the future, when NFS model is replaced by the WRF model in July 2007, upgrading to the WRF-Var 2.2 is necessary.

0)Enhance the WRF-Var system

)objective tuning the BES

)ground-based GPS ZTD observation operator

)non-local GPSRO observation operator

0)Improve the Typhoon bogus and relocation technique

It is very important to get the correct Typhoon analysis at the initial time, especially for the cycling mode forecast. Although we have got some success in 2006, the work needs to further refinement and improvement.

0)Develop and improve the verification package

)Improve the WRF-Var-based observation verification package to get the vertical profile of the bias and RMS errors;

)Develop the precipitation verification package for Taiwan area.

0)Test the convective scale WRF 4D-Var over Taiwan area.

This is one of the tasks for the CRIEPI project. CWB can take the advantages from WRF-Var

Version 2.2 andto apply it to Taiwan area.

6) Develop ground-based GPS processing capability to support operational use

a. Support for using WRF and WRF-Var 2.2 as an operational system at CWB

(we should be able to do everything NFSSF is providing now)

b. WRF-Var enhancement

c. Ground-based GPS water vapor processing and assimilation

d. Improve the assimilation of GPS radio occultation soundings using nonlocal observation operator