Development and research of GSI based hybrid EnKF Var data assimilation for HWRF to improve hurricane prediction Xuguang Wang, Xu Lu, Yongzuo Li School of Meteorology University of Oklahoma, Norman, OK, USA Acknowledgement Mingjing Tong, Vijay Tallapragada, NCEP/EMC, College Park, MD Henry Winterbottom, Jeff Whitaker, NOAA/ESRL, Boulder, CO WWOSC, Aug. 204, Montreal, Canada
GSI based Var/EnKF/hybrid for global and regional modeling systems GSI based Var/EnKF/3 4DEnVar Hybrid GFS WRF NMMB WRF ARW Hurricane WRF (HWRF) SCI-PS26.04: Aaron Johnson talk A comparison of GSIbased multiscale EnKF and 3DVar for convective scale weather forecast 2
GSI based Hybrid EnKF Var DA system Wang, Parrish, Kleist, Whitaker 203, MWR member forecast member 2 forecast member k forecast EnKF Whitaker et al. 2008, MWR Ensemble covariance EnKF analysis EnKF analysis 2 EnKF analysis k Re-center EnKF analysis ensemble to control analysis member analysis member 2 analysis member k analysis member forecast member 2 forecast member k forecast control forecast GSI-ACV Wang 200, MWR control analysis control forecast data assimilation First guess forecast 3
GSI hybrid for GFS: GSI 3DVar vs. 3DEnsVar Hybrid vs. EnKF 3DEnsVar Hybrid was better than 3DVar due to use of flow dependent ensemble covariance 3DEnsVar was better than EnKF due to the use of tangent linear normal mode balance constraint (TLNMC) Wang, Parrish, Kleist and Whitaker, MWR, 203, 4, 4 4098 47
J ' t x ', α J x 2 J ' T 2 B e J static K e α k (x ) t k ' x x k GSI hybrid for GFS: 3DEnsVar vs. 4DEnsVar GSI-4DEnsVar: Naturally extended from and unified with GSIbased 3DEnsVar hybrid formula (Wang and Lei, 204, MWR, 42, 3303-3325). o x ' 2 α 2 T Add time dimension in 4DEnsVar C α 2 T o T - o t ( yt '-H t y tx ) R t ( t '-Ht x t ) B stat 3DVAR static covariance; R observation error covariance; K ensemble size; C correlation matrix for ensemble covariance localization; x kth ensemble perturbation; ' x total (hybrid) increment; ' x 3DVAR increment; H linearized observation operator; e k o' y innovation vector; weighting coefficient for static covariance; 2 weighting coefficient for ensemble covariance; α extended control variable. 5
GSI hybrid for GFS: 3DEnsVar vs. 4DEnsVar Results from Single Reso. Experiments 4DEnsVar improved general global forecasts 4DEnsVar improved the balance of the analysis Performance of 4DEnsVar degraded if less frequent ensemble perturbations used 4DEnsVar approximates nonlinear propagation better with more frequent ensemble perturbations TLNMC improved global forecasts See poster SCI POT040 and Wang, X. and T. Lei, 204: GSI based four dimensional ensemble variational (4DEnsVar) data assimilation: formulation and single resolution experiments with real data for NCEP Global Forecast System. Mon. Wea. Rev., 42, 3303 3325. 6
GSI hybrid for GFS: 3DEnsVar vs. 4DEnsVar 6 named storms in Atlantic and Pacific basins during 200 7
Approximation to nonlinear propagation 3h increment propagated by model integration 4DEnsVar (hrly pert.) 4DEnsVar (2hrly pert.) 3DEnsVar Hurricane Daniel 200 * time -3h 0 3h 8
Verification of hurricane track forecasts 3DEnsVar outperforms GSI3DVar. 4DEnsVar is more accurate than 3DEnsVar after the -day forecast lead time. Negative impact if using less number of time levels of ensemble perturbations. Negative impact of TLNMC on TC track forecasts. 9
Development and research of GSI based Var/EnKF/hybrid for HWRF GSI based Var/EnKF/3 4D EnVar Hybrid GFS WRF NMMB WRF ARW Hurricane WRF (HWRF) 0
GSI hybrid for HWRF Hurricane Sandy, Oct. 202 Complicated evolution Tremendous size 47 direct deaths across Atlantic Basin US damage $50 billion New York State before and after nhc.noaa.gov
Experiment Design Sandy 202 Model: HWRF Observations: radial velocity from Tail Doppler Radar (TDR) onboard NOAA P3 aircraft Initial and LBC ensemble: GFS global hybrid DA system Ensemble size: 40 2
Experiment Design Model: HWRF Oper. HWRF Observations: radial velocity from Tail Doppler Radar (TDR) onboard NOAA P3 aircraft Initial and LBC ensemble: GFS global hybrid DA system Ensemble size: 40 3
TDR data distribution (mission ) P3 Mission 4
Verification against SFMR wind speed Last Leg 5
Comparison with HRD radar wind analysis 6
Comparison with HRD radar wind analysis S N 7
Track forecast (RMSE for 7 missions) 8
Experiments for 202 203 seasons Correlation between HRD radar wind analysis and analyses from various DA methods 0.9 0.8 0.7 0.6 0.5 0.4 0.3 0.2 0. 0 2 3 4 5 6 7 8 9 0 2 3 4 5 6 7 8 9 20 2 22 Hybrid GSI3DVAR Hybrid-GFSENS Case# 9
ISSAC 202 (mission 7) 20
Verification against SFMR and flight level data
Experiments for 202 203 season Track MSLP 22
HWRF Hybrid DA with moving nests: () Dual resolution hybrid 9km 3km movable nest ingests 9km HWRF EnKF ensemble Two way coupling 3km Tests with IRENE 20 assimilating airborne radar data 23
Dual resolution hybrid IRENE 20
HWRF Hybrid DA with moving nests: (2) Self consistent system: 3km and 9km ingesting their own EnKF ensemble 3km 9km 27km 25
HWRF Hybrid DA with moving nests Near real time experiment during 204 season Track MSLP hpa Arthur 204 Newly developed self-consistent HWRF hybrid DA with moving nests show improvement on MSLP/Vmax than operational HWRF 26
Summary and ongoing work a. The GSI based hybrid EnKF Var data assimilation system was expanded to HWRF. b. Various diagnostics and verifications suggested this unified GSI hybrid DA system provided more skillful TC analysis and forecasts than GSI 3DVar and than HWRF GSI hybrid ingesting GFS ensemble. c. Airborne radar data improved TC structure analysis and forecast, TC track and intensity forecasts. Impact of the data depends on DA methods. d. Hybrid DA with movable nests were developed and showed promising results. e. Ongoing experiments with more cases. f. Ongoing research to investigate the difference among Var, EnKF, 3DEnsVar and 4DEnsVar hybrid for convective scales. 27
References Wang, X., 200: Incorporating ensemble covariance in the Gridpoint Statistical Interpolation (GSI) variational minimization: a mathematical framework. Mon. Wea. Rev., 38, 2990-2995. Wang, X., D. Parrish, D. Kleist and J. S. Whitaker, 203: GSI 3DVar-based Ensemble-Variational Hybrid Data Assimilation for NCEP Global Forecast System: Single Resolution Experiments. Mon. Wea. Rev., 4, 4098-47. Wang, X. and T. Lei, 204: GSI-based four dimensional ensemblevariational (4DEnsVar) data assimilation: formulation and single resolution experiments with real data for NCEP Global Forecast System. Mon. Wea. Rev., 42, 3303-3325. 28
29 ' ' ' ' 2 ' ' 2 ' 2 2 2, Hx y Hx y α C α x B x α x o T o T T J e J o J J R K k e k k ' ' x α x x 29 GSI-3DEnsVar: Extended control variable (ECV) method implemented within GSI variational minimization (Wang 200, MWR): Extra term associated with extended control variable Extra increment associated with ensemble (4D)EnKF: ensemble square root filter interfaced with GSI observation operator (Whitaker et al. 2008) GSI-based Hybrid EnKF-Var DA system
DA cycling configuration Cold Start OBS GSI3DVar Spin-up Forecast DA Cycle Deterministic Forecast OBS Hybrid Spin-up Forecast Deterministic Forecast Ensemble Perturbation HWRF EnKF OBS Ensemble Spin-up Forecast DA Cycle Deterministic Forecast 30
DA cycling configuration OBS Hybrid-GFSENS Spin-up Forecast Deterministic Forecast Ensemble Perturbation GFS ENS 3