Mid-Term review of 2007 CWB project P.I.: Y-H. Kuo 1,2 Y.-R. Guo 1 (WRF-Var lead) H. Liu 3 (WRF-EnKF lead) J. Braun 2 (Ground-based GPS lead) CWB Visitor: Y.-T Lin 1 NCAR/MMM, 2 UCAR/COSMIC, and 3 iMage 7 August 2007
CWB Project Tasks: Task 1: Support and enhancement of the WRF-Var system for CWB operation Task 2: Exploration of the WRF-based Ensemble Kalman Filter (EnKF) data assimilation Task 3: Training on ground-based GPS PW data processing Task 4: Continued interaction on WRF data assimilation systems
Task 1: Support and enhancement of the WRF-Var system for CWB operation Performance Period: a. develop the namelist files and shell scripts for pre- operational configuration 1/1/07 – 06/30/07 b. end-to-end tests on UCAR IBM for short period of time04/01/ /30/07 c. post results of WRFVar/WRF tests for short period of time on web page 04/01/ /30/07 d. timely responses to solve WRFVar related problems 04/01/07 – 12/31/07 Deliverables: 1. namelist files and shell scripts for pre-operational configuration 06/30/07 2. Brief report on experiment results for short period of time in UCAR 12/31/07
Task 2: Exploration of the WRF-based Ensemble Kalman Filter (EnKF) data assimilation Performance Period: a) perform WRF-based EnKF assimilation of COSMIC GPSRO data on Typhoon ShanShan 1/1/07 – 12/31/07 b) perform WRFVar assimilation of COSMIC GPSRO data on Typhoon ShanShan 1/1/07 – 12/31/07 c) conduct comparison of WRFVar and EnKF data assimilation1/1/07 – 12/31/07 d) training of CWB staff on WRF-based EnKF system1/1/07 – 12/31/07 Deliverables: 1. EnKF code and run shell scripts12/31/07 2. Report of the comparison study between WRFVar and EnKF 12/31/07
Task 3: Training on ground-based GPS PW data processing Performance Period: a) Training of GPS PW retrieval (not to exceed one week)1/1/07 – 12/31/07 Deliverables: 1. UCAR software on GPS PW retrieval12/31/07 Fig. 1. (a) The distribution of the 57 GPS receiver stations over Taiwan, and (b) the eight GPS stations that are equipped with collocated surface meteorological observations.
Task 4: Continued interaction on WRF data assimilation systems Performance Period: 1. Updated CWB project web pages on both CWB and UCAR sides 1/1/07 – 12/31/07 2. Site visit to CWB 4/1/ /31/07 Deliverables: 1. Updated web page for project12/31/07 2. Site visits12/31/07 NCAR CWB Web Page: CWB Blog:
Task 1: Support and enhancement of the WRF-Var system for CWB operation Yong-Run Guo
Task#1 Support and enhancement of the WRFVar system for CWB operation Design the operational configuration To answer the questions from Eric Chiang about WRFVar a, wrf_io.F related things for WRFV2.2 b, Stage0 in gen_be with WRFV2.2 data c, GPSRO wetPrf decoder d, obs error statistic tuning code e, ….. To update the WRFVar code (bug fix and development) a, introduce the TSK increment by using the lowest level T increment b, corrected the sfc_assi_option=2 code: the height above the sea level should be the height above the ground c, Relative humidity check: check_rh = 2 d, WRFVar-based VERIFY: use U10, V10, T2, Q2 read from WRF model output To deliver the namelist files and running shell script on web page on 8 May 2007 and updated on 26 July 2007.
Major problems encountered in operational testing by Dr. Hong (CWB) are: 1) WRFVar2.1/WRFV2.1 cycling runs blew up; 2) Warm bias drift in late December NCAR recommended: WRF model: a) use WPS and WRFV2.2; b) use 45 vertical levels with Ptop=3000Pa, and re-define levels; c) use Noah LSM to replace the thermal diffusion scheme. WRFVar: a) sfc_assi_options = 2; b) cv_option_hum = 1; c) current use cv_options = 3, but in future use cv_options = 5. The suggested namelist files and running shell scripts are post on the web:
WRFVar/WRF testing at NCAR End-to-end 3 domains (45/15/5km) 6-h cycling run with CWB FGGE observation data and GFS intermediate data files from 0000 UTC 1 to 1800 UTC 31 December End-to-end testing of the following programs: WPS (metgrid) real.exe FGGE decoder OBSPROC WRFVar Update_BC WRFV2.2 Archive (to NCAR MSS) Loop domains Loop time
a) Two Exps completed on PC Linux cluster (leea) with 6 CPUs: ExpA: cv_options = 3 ExpB: cv_options = 5 with interpolate_stats=.TRUE. By use 41-level BES from Eric Chiang (CWB) It takes about 1 day wall-clock time to advance 1 day (four 6-h cycles) initial times, i.e. it took one month to complete one Exp. b) One Exp (cv_options =3) completed in NCAR IBM (blueice). NCAR IBM has a queuing system to submit (bsub) the jobs. The maximum time limit for one job is 6 hours. Some of WRFVar/WRF system are single CPU code, some are MPP code. The shell script of running on NCAR IBM is more complicated than that on a local machine. The job dependence “#bsub –w ${previous_job}” must be used. “Implicit” submitting the ${next job} need to be used to avoid a pile of jobs (for a month,11 x4 x31 =1364 jobs) listed in the job list.
On NCAR IBM blueice, normally 4 days (16 6-h cycles) took one-day wall-clock time. According to Jim Bresch CWB HPC has the same queuing system as NCAR machine, this running shell may be implemented in CWB. Results have the minor differences between PPPC Linux cluster and IBM. For example, for Z domain1 (45km) WRFVar: IBM (blueice): Diagnostics of OI for Sound var u (m/s) n k v (m/s) n k t (K) n k q (kg/kg) n k Number: Minimum(n,k): E Maximum(n,k): E Average : E-03 RMSE : E-03 Linux (leea): Diagnostics of OI for Sound var u (m/s) n k v (m/s) n k t (K) n k q (kg/kg) n k Number: Minimum(n,k): E Maximum(n,k): E Average : E-03 RMSE : E-03
CWB pre-operational testing Yun-Tien Lin (CWB) arrived at NCAR in early July a) The CV5 BES have been derived based on the ExpA (on leea) one month results for 3 domains: 45km, 15km, and 5km with 45-L. Now check the correctness…… b) WRFVar-based VERIFY: generating the 6 hourly CWB QCed observation data by using NCEP AVN analysis…… c) When the new CV5 BES is ready, will conduct CV5 BES Exp on IBM for DEC2006 data; d) Conduct the JUL2007 (summer time) Exps.
Task 2: Exploration of the WRF- based Ensemble Kalman Filter (EnKF) data assimilation Hui Liu and Yong-Run Guo
Task#2 Exploration of the WRF-based Ensemble Kalman Filter (EnKF) data assimilation Conduct WRF-based EnKF assimilation of COSMIC GPSRO data on a typhoon case - What have been done for WRF/ENKF 1. Impact of COSMIC data on forecast of Shanshan using WRFv2.1/EnKF. Positive impact was found in the presence of satellite winds and radiosondes. 2. Upgraded the system to WRFv2.2/EnKF. 3. Tested assimilation of CWB observations and COSMIC data in WRFv2.2/EnKF with CWB WRF options. 4. Initial results show that COSMIC data has positive impact on forecast of Shanshan in the presence of satellite winds and radiosondes.
An assimilation experiment of Shanshan with WRF2.1/WRFSI with NCAR options 1 hour assimilation window Assimilation continuously done for Sep 6-12 Control run: Assimilate radiosonde and satellite winds GPS run: Control run + COSMIC data Forecast from 12Z Sep 12, 2006.
Best track – black Control run – blue GPS run – red Forecast from 12Z Sep 12 (dots for every 12 hour)
An initial study with CWB configuration 1. Assimilation of CWB radisondes, satellite cloud winds, and COSMIC refractivity for 12 hour from Sep 13 12Z to Sep 14 00Z over CWB domain 1 (45km). CNTL: radiosonde + satellite winds GPS : CNTL + COSMIC refractivity 2. 3-day forecast from the analyses at Sep 14 00Z 3. No typhoon bogus
Track forecasts Black: OBS Green: NoDA Blue: CNTL Red: GPS
Track and intensity errors
For 2008 of WRF/ENKF 1.Setup WRF/EnKF for CWB domain and tune observation errors, filter localization, ensemble sizes, ensemble inflations etc. 2. Develop QC of CWB observations for use in WRF/EnKF, especially surface observations 3. Performance 2-week assimilation of CWB data in operational setting. Validate the analyses and forecasts against observations. Compare with WRF/3dvar. 4. Evaluate impact of the local and non_local RO refractivity operators in WRF/EnKF on forecasts.
Comparison between WRFVar and WRF-based EnKF Six experiments have been done with WRFVar for Typhoon Shanshan: 1, NODA --- Initiated at Z with NCEP AVN analysis; 2, COLDNB --- FG at Z is the 24-h forecast from NODA, OBS data are SOUND, SYNOP, SATOB, AIREP, PILOT, METAR, SHIPs, SATEM, QuikScat, and BUOY within 1 h time window from 2330 UTC 13 to 0030 UTC 14 September , COLDALL --- Same as COLDNBNG, but GPSREF and BOGUS (global and TC Bogus) included too. 4, CYCLNBNG h cycles starting from Z with WRFVar/WRF, but no GPSREF and BOGUS data assimilated. 5, CYCLNB --- Same as CYCLNBNG, but GPSREF assimilated. 6, CYCLALL --- Same as CYCLNB, but BOGUS data assimilated too. All Exps were conducted over a domain of 222x128x45 with grid size of 45- km. The BES is interpolated from Eric Chiang’s (CWB) 41-L CV5 BES based on the 3 months forecast data from June to August 2006.
1300Z 1400Z 1306Z1312Z1318Z NODA 96h forecast Cold start 72h forecast COLDNBNG COLDALL Hourly Cycling: CYCLNBNG CYCNB CYCLALL 72h forecast Schematic diagram of experiments NCEP AVN Analysis CWB OBS data
Results Track forecast Intensity forecast
Track forecast errors averaged over the different periods (km) Exp3-24h27-48h51-72h3-72h NODA COLDNBNG COLDALL CYCLNBNG CYCLNB CYCLALL
summary Assimilation of the CWB observation data, no matter Cold-start or cycling mode, improved the Typhoon Shanshan track forecast. However, only when the BOGUS data are assimilated the intensity forecast is improved. Assimilation with the cycling mode gave better track forecast than the cold-start runs because more observation information was injested into the initial condition at Z. GPSREF data assimilation showed the positive impact on the track forecast (compare the CYCLNB with CYCLNBNG). BOGUS data assimilation improved both track and intensity forecast (COLDALL, CYCLALL).
Task 3: Training on ground- based GPS PW data processing John Braun
Task #3: Training of GPS PW Retrieval Visit to CWB scheduled last week of August (August 25 - September 1) Goals of Trip: Install and test B50 processing scripts on CWB machines Teach class on using Bernese software for PW retrieval. Reprocess July 2005 data set for training and verification Process current data set for Give two public seminars
Scientific Seminars Tropical Cyclone Intensity and Precipitable Water Vapor Estimates from GPS Recent Improvements and Results in Water Vapor Estimates from GPS
Topics in Training Course Bernese Software Description Overview Bernese Processing Engine External data sources, orbits, reference stations, and ancillary information. Important components of PWV analysis strategy. GPS observation equation
Topics for Discussions Date of project final review Deliverables check-off Contract payments (not yet been paid) Tasks to be performed for the remainder of 2007 both at NCAR and CWB