Assimilation of T-TREC-retrieved wind data with WRF 3DVAR for the short-Term forecasting of Typhoon Meranti (2010) at landfall Xin Li1, Yuan Wang1, Jie.
Download ReportTranscript Assimilation of T-TREC-retrieved wind data with WRF 3DVAR for the short-Term forecasting of Typhoon Meranti (2010) at landfall Xin Li1, Yuan Wang1, Jie.
Assimilation of T-TREC-retrieved wind data with WRF 3DVAR for the short-Term forecasting of Typhoon Meranti (2010) at landfall Xin Li1, Yuan Wang1, Jie Ming1, Kun Zhao1, Ming Xue2 1The Key Laboratory of Mesoscale Severe Weather, School of Atmospheric Sciences, Nanjing University, China 2Center for Analysis and Prediction of Storms and School of Meteorology, University of Oklahoma Background • Doppler radar is the only platform that observes the 3D structure of Typhoons at high enough temporal and spatial resolutions. • Significant progress has been made in the TC forecasting using Radar data direct assimilation (Vr and Reflectivity). • Wind field is crucial in Typhoon assimilation and the importance of full coverage by Multi-Doppler Radar and cycling assimilation. (Xiao et al. 2005,2007; Zhao and Jin 2008; Zhang et al. 2009,2011; Zhao and Xue,2009,2011,2012). Motivation • Single Radar provides full information of wind field in Typhoon inner core. • T-TREC (an extended TREC retrieving method) uses the information of both Reflectivity and Vr to retrieve wind field. Make full use of the large coverage of Reflectivity data. • Provide full circle of vortex circulation in the inner-core region. T-TREC Retrieving wind(T-TREC VS. TREC) Target cell Searching distance T-TREC wind vector Initial cell T2 T1 R Z V 1) Polar coordinates centered on the TC center 2)Anti-clock wise searching 3)Velocity correlation matrix 4)Objective center finding and searching area determining Saomai(0608) Z=1km 1hour before landfall Wang and Zhao, 2010 TREC T-TREC Meranti(2010) Radar data information and coverage 3-km wind Vr T-TREC Experiment WRF Forecast from GFS Reanalysis CTL 1200 UTC/09 Vr 1800 UTC/09 0000 UTC 0600UTC/10 WRF Forecast with Radar Vr DA ExpVr 1200 UTC/09 T-TREC 1800 UTC/09 0000 UTC 0600UTC/10 WRF Forecast with Radar T-TREC wind DA ExpTrec 1200 UTC/09 1800 UTC/09 0000 UTC 0600UTC/10 Model Grid CTL ExpVr ExpTrec Domain 3 nested 257*237 12km 462*462 4km 615*615 1.33km 3 nested 257*237 12km 462*462 4km 615*615 1.33km 3 nested 257*237 12km 462*462 4km 615*615 1.33km Observation None Radial velocity (Vr) T-TREC wind Assimilation window None Only once at initial time Only once at initial time Physics Lin microphysics YSU boundary-layer Kain-Fritsch (Domain 1) Lin microphysics YSU boundary-layer Kain-Fritsch (Domain 1) Lin microphysics YSU boundary-layer Kain-Fritsch (Domain 1) Radar data impact at initial time CTL ExpVr ExpTrec Impact on Typhoon structure Forecast D03 1.33km 06h 12h 18h OBS CTL ExpVr ExpTrec Impact on Track and Intensity Forecast D03 1.33km Impact on 6-h accumulated Precipitation Forecast D03 1.33km OBS 06-12h 12-18h CTL ExpVr ExpTrec Conclusion • The impact of T-TREC retrieving wind has been recognized in Typhoon forecast at landfall • The assimilation only need once due to the large coverage and full vortex circulation of T-TREC retrieving data • The improved Typhoon initial condition by T-TREC wind data leads to not only the better track, intensity and structure prediction, but also the precipitation forecast even no Reflectivity data is assimilated Recent research • The climatological (static) background error covariance matrix (B matrix) of 3DVAR only reflect the constraint of large scale balance and the flow-dependent covariance through ensemble is needed. • The ensemble-based flow dependent background error covariance matrix could reflect the current flow pattern and correct multivariate covariance for Typhoon structure • WRF Hybrid En-3DVAR assimilation system(Wang et al.,2007,2008,2011) incorporates ensemble flow dependent background covariance in the 3DVAR by extending the control variables in variational framework, combining climatological and flow-dependent background error covariance WRF Hybrid En-3DVAR WHY Hybrid? Advantage? • Flow-dependent B matrix is important and can be adapted to the existing 3D-VAR system easily through an extended control variable • The physics constraint could be added easily to the variational framework of Hybrid En-3DVAR • Hybrid can be robust for small size ensembles. • While, similar with EnKF, the horizontal and vertical covariance localization are applied. The hybrid formulation…. Ensemble covariance is implemented into the 3D-VAR cost function via extended control variables: (Wang et. al. 2008) 1 'T 1 ' 1 T 1 1 o' J(x , ) 1 x1 B x1 2 C (y Hx' )T R1 (yo' Hx' ) 2 2 2 ' 1 K x x ( k ox ) ' ' 1 e k Conserving total variance requires: β1+β2=1 k1 C: correlation matrix for ensemble covariance localization x ' 1 x' 3D-VAR increment Total increment including hybrid Extended control variable 1 Weighting coefficient for static 3D-VAR covariance 2 Weighting coefficient for ensemble covariance Hybrid data assimilation -6h 30 members 0h Initial Ensemble Forecast 18h Deterministic Forecast 0600 UTC/09 1200 UTC/09 Generate Ensemble perturbations use “RANDOMCV” in WRF-3DVAR Hybrid DA T-TREC wind 0600UTC/10 Be matrix : Ensemble flow-dependent & 3DVAR static Spread of 6-h pre-ensemble forecast 3-km V-wind Ens-Mean 3-km V-wind Ens-Spread Experiment configuration Exp3DVAR ExpHybrid0.5 ExpHybrid1.0 Domain 3 nested 257*237 12km 462*462 4km 615*615 1.33km 3 nested 257*237 12km 462*462 4km 615*615 1.33km 3 nested 257*237 12km 462*462 4km 615*615 1.33km Observation T-TREC wind T-TREC wind T-TREC wind Assimilation window Only once at 1200 UTC/09 Only once at 1200 UTC/09 Only once at 1200 UTC/09 Background error covariance matrix Only 3DVAR static (β1=1.0,β2=0) Hybrid 3DVAR static and Ensemble flowdependent (β1=0.5,β2=0.5) Only Ensemble flow-dependent (β1=0,β2=1.0) Flow-dependent B matrix impact 3-km V-wind Single point Test 3DVAR Hybrid0.5 Hybrid1.0 Empirical Vertical Covariance Localization Spurious sampling error are not only confined to horizontal error correlations, it affects vertical too. So vertical localization is needed. Old: Grid-Dependent Localization Scale L : 10 grids K : vertical grids Apply Gaussian Vertical Covariance Localization function: 2 (k kc ) exp k kc / L2c New: Distance-Dependent Localization Scale L : 3000 m K : vertical distance Impact of Vertical Covariance Localization 3-km wind Single point Test No vertical localization Vertical cross-section increment With old vertical localization With new vertical localization Flow-dependent B matrix impact 3DVAR 3-km Wind analysis and increment by T-TREC wind Hybrid0.5 Hybrid1.0 Flow-dependent B matrix impact 3DVAR 1-km T increment Vertical cross-section Hybrid0.5 Hybrid1.0 Impact on Track and Intensity Forecast D03 1.33km Impact on Typhoon structure Forecast D03 1.33km 06h 12h 18h OBS Exp3DVAR ExpHybrid0.5 ExpHybrid1.0 Summary • The 3DVAR performs well in vortex circulation initialization while the mass fields are adjusted during the model’s spinning up mostly • The Hybrid En-3DVAR provides more balance analysis due the use of flow-dependent B matrix even it only from the cold start pre-ensemble forecast. The enhanced thermal structure leads to better intensity and structure prediction • Based on three Typhoon case (chanthu,megi,2010,not shown). Ensemble-based flow-dependent B matrix is important for Typhoon structure assimilation. • The cycling use of T-TREC wind or the so-called Multi-scale assimilation (T-TREC combining Vr) are being tested ongoing for more balanced initial condition. Thanks