Improvements of Hurricane Forecast with Vortex Initialization using WRF Variational (WRF-Var) Data Assimilation

Abstract

Although the forecasts of hurricane track have steadily improved during the past two decades, intensity forecasts remain unsatisfactory (Elsberry, 2005). Previous studies have shown that vortex structures can significantly affect the behavior of hurricane intensity (Ross & Kurihara, 1995; Willoughby & Black, 1996; Xiao et al., 2000). The initialization of hurricane vortex as well as its environment using an advanced data assimilation technique is a key procedure to improve the accuracy of hurricane forecasts and to extend lead-time for hurricane forecasts with increased certainty. Particularly, assimilation of the data in the vortex region deserves more scientific research and technical development. Assimilation of Doppler radar observations from coastal and aircraft Doppler radars in hurricane vortex initialization is of great interest to the weather service and research communities but with a lot of challenges (Liu et al., 1997; Marks et al., 1998; Xiao et al., 2000; 2006). The complexity comes from how we should establish the data assimilation system for hurricanes that are usually in the low latitude, compounded by a lack of data over the ocean and inadequate computer resources to resolve the inner core. From the several-year realtime hurricane forecasts using the Advanced-research Hurricane WRF (AHW) model (Davis et al., 2008; Xiao et al., 2009a; b), however, one conclusion is that an advanced analysis scheme has to be implemented for improved vortex initialization. Data assimilation is a process that incorporates observations into numerical model with consideration of both observational data and model background information. There are several data assimilation techniques that can be used for hurricane initialization. The fourdimensional variational (4D-Var) data assimilation (Courtier et al., 1994) and Ensemble Kalman filters (EnKF, Evensen, 1994) are two of the most advanced in algorithm formulation and technique design. 4D-Var employs a forecast model as a strong constraint in a least-squares fit problem (Lewis & Derber, 1985, Le Dimet & Talagrant, 1986, Thepaut & Courtier, 1991, Navon et al., 1992). It has an implicit update of the flow-dependent background field and the capability to assimilate data at the exact observation time. The 4DVar adjoint approach is an attractive assimilation technique. As a retrospective assimilation algorithm, it can derive the optimal time-trajectory fit to observational data, including nonsynoptic data (Xiao et al., 2002, Simmons & Hollingsworth, 2002). 4D-Var has been