PSU HFIP 2010 Summary: Performance of the ARW-EnKF Real-time Cloud-resolving TC Ensemble Analysis and Forecasting System Fuqing Zhang Penn State University Contributors: Yonghui Weng, John Gamache and Frank Marks Acknowledgments: IFEX Team, JET and TACC Computing
Experimental Design for PSU ARW-EnKF w/ Assimilation of NOAA P3 Doppler Vr Ensemble Forecast Ensemble Forecast Ensemble Forecast Deterministic Forecast with fixed domains Deterministic Forecast with moving nested domains by vortex -18hr SOs from 1 st leg SOs from legs The last leg 00hr wrf-3dvar produce 30 members from GFS analysis with 3 domains GFS forecast as BCs EnKF with radar Vr EnKF with radar Vr EnKF with radar Vr Usually there are about 2 to 5 legs for one NOAA P3 mission during - 12hr and -5hr The forecast will be available and verified since this time.
PSU ARW-EnKF Real-time Configurations D1: 253x163x40.5km 35 vertical levels; WSM 6-class microphysics; YSU PBL; Grell-Devenyi CPS D2: 253x253x13.5km D3: 253x253x4.5km D4: 253x253x1.5km 30-member ensemble; Gaspairi&Cohn 99' covariance localization with varying RoI IC & BC: GFS using 3DVAR background uncertainty The inner domains are centered with the storm s center Assimilation is performed over D1-3. D4 is used only for high-res forecast.
AL2010: 23 Missions/Cases Total Storms P3 Missions Alex 1 Two 3 Earl 11 Karl 4 Richard 1 Tomas 3 Alex Karl Two Tomas Earl Richard NOAA P3 Airborne Doppler Missions for the 2010 Atlantic Hurricane Season
AL2009: 10 Missions/Cases Total Storms P3 Missions Ana 1 Bill 4 Danny 5 Danny Ana Bill NOAA P3 Airborne Doppler Missions for the 2009 Atlantic Hurricane Season
AL2008: 26 Missions/Cases Total Storms P3 Missions Dolly 6 Fay 6 Gustav 6 Ike 5 Paloma 3 Dolly Paloma Gustav Fay Ike NOAA P3 Airborne Doppler Missions for the 2008 Atlantic Hurricane Season
PSU ARW-EnKF Assimilating Airborne Radar OBS Mean Absolute Error and Ensemble Spread for all 59 cases 2008-2010 A1PS: PSU 1.5km single forecast initialized with EnKF analyses A4PS: PSU 4.5km single forecast initialized with EnKF analyses P400: ensemble forecast mean of 30 members in 4.5km resolution PSTD: averaged ensemble spread of P400
Consistency between Ensemble Spread and Error Mean Absolute Error and Ensemble Spread for all 59 cases 2008-2010
PSU ARW-EnKF Assimilating Airborne Radar OBS Mean Absolute Error and Mean Bias for all 59 cases 2008-2010
Performance with Simple Bias Correction Mean Absolute Error and Mean Bias for all 59 cases 2008-2010 Corrected WSP = WSP - 30h - t 30h Bias _ at _initial _ time
A Few Remarks on our ARW-EnKF All 59 cases with P3 Vr missions for 2008-2010; these are the cases that are usually near land with biggest impacts to US Forecasts is available at the 00h verification time (as in OFCL) Ensemble spread is broadly consistent with mean forecast error With some bias correction, 15-40% less error than OFCL to 96hr No ocean model is used; the same ARW setup is used; no tuning whatsoever yet; 4.5-km runs similar to 1.5-km runs Boundary conditions and large-scale environment beyond Vr influence are from -18hr GFS Analyses and Forecasts; further improvement is expected using updated GFS products Only 30 members are used in the EnKF analyses; further improvement is also expected with 60-90 members Assimilating more obs besides Vr and hybrid with 3DVar are planned for the next season; stream 1.5?
Hybrid Algorithms for Coupling 3DVar&4DVar w/ EnKF Necessary Variable Changes: i) EnKF provides ensemblebased background error covariance (P f ) for 3D/4DVar Coupler of EnKF-3DVar Hybrid (E3DVar): ii) iii) EnKF provides the prior f ensemble mean ( x ) as the first guess for 3D/4DVar 3D/4DVar provides a deterministic analysis ( x ) to replace the posterior ensemble mean for the next ensemble forecast Coupler of EnKF-4DVar Hybrid (E4DVar): (Zhang 2010; Zhang et al. 2009,2010)
Comparison of 3DVar, 4DVar, EnKF, E3DVar & E4DVar 0-72hr U, V, T & Q RMS forecast error over CONUS Jun 2003 (60 runs) (Zhang 2010; Zhang et al. 2010)
EXP 2.2 : Control Experiments for E4DVar Mean vertical profiles of month-averaged 12-h forecast RMSE
PSU Summary and Plans We performed realtime 40.5- and 13.5-km WRF-EnKF ensemble analyses and 32-members of 4.5-km ensemble forecasts down to 4.5-km grid spacing twice per day from Aug 1 to Sept 30. We also performed additional 4.5-km ensemble data assimilation in near realtime to assimilate airborne Doppler radar velocity if available, and produced additional subsequently ensemble forecast from these analyses and perturbations We performed systematic verifications on the performance of all 59 cases with P3 airborne Doppler missions during 2008-2010 We are in the development of a hybrid EnKF-3DVar data assimilation system for hurricane prediction; other non-radiance observations besides Doppler Vr will also be assimilated This new system is being tested for the 2010 Atlantic season and is planned for the 2011 hurricane season
AL212010 Tomas (3 cases)
AL212010 Tomas (3 cases)
Absolute error for 2008 (26), 2009 (10) and 2010 (23)
PSU ARW-EnKF Assimilating Airborne Radar OBS Mean Absolute Error and Ensemble Spread for all 59 cases 2008-2010 A1PS: PSU 1.5km single forecast initialized with EnKF analyses A4PS: PSU 4.5km single forecast initialized with EnKF analyses P400: ensemble forecast mean of 30 members in 4.5km resolution PSTD: averaged ensemble spread of P400