GSI‐based ensemble‐variational hybrid data assimilation for HWRF for hurricane initialization and prediction: impact of various error covariances for airborne radar observation assimilation

Abstract

The hybrid Ensemble Kalman Filter ‐ Variational (EnKF‐Var) data assimilation (DA) system based on Grid‐point Statistical Interpolation (GSI) is extended for the Hurricane‐WRF model (HWRF). Background ensemble forecasts initialized by the EnKF are used to provide the flow‐dependent error covariance to be ingested by GSI using the extended control variable method. The hybrid system is then applied to assimilate airborne radar data.In this article, the newly developed HWRF hybrid system capable of assimilating airborne radar observations is introduced. The impact of using variously estimated background error covariances on tropical cyclone (TC) core analyses and subsequent forecasts is explored by a detailed study of Hurricane Sandy (2012) and by systematic comparison of various sensitivity experiments for multiple cases during the 2012–2013 seasons. The hybrid system using the HWRF EnKF ensemble covariance (Hybrid‐HENS) is able to correct both the wind and mass fields in a dynamically and thermodynamically coherent fashion. In contrast, the wind and pressure adjustments by GSI three‐dimensional variation (GSI3DVar) using the static covariance are inconsistent. The wind and pressure relation in the covariances derived from the GFS ensemble (Hybrid‐GENS) improves upon the static covariance, but is still inconsistent compared to that of HWRF. Verifications against independent flight‐level and Stepped Frequency Microwave Radiometer (SFMR) wind data, and Hurricane Research Division (HRD) radar wind composite reveal that the Hybrid‐HENS system improves the analysed TC structure upon both GSI3DVar and Hybrid‐GENS. Hybrid‐HENS and Hybrid‐GENS improve the track, minimum sea‐level pressure (MSLP) and Vmax forecast relative to GSI3DVar. Hybrid‐HENS further improves track forecasts compared to Hybrid‐GENS. Hybrid‐HENS provides the largest positive impact of the airborne radar data. In comparison, GSI3DVar shows consistently negative impact of the data when analysing the structure and verifying track forecasts. Blending the static background error covariance in the hybrid system improves the maximum wind forecast while little benefit is found in the analysed structures and the MSLP and track forecasts.