Hourly Assimilation of Enhanced AMVs for Tropical Cyclone Initialization Using a Novel Ensemble Bogus Method

Initialization of numerical models for the purpose of tropical cyclone (TC) prediction remains an important and, thus far, largely unsolved problem. While advanced data assimilation techniques such as the ensemble Kalman filter (EnKF) have shown promise when applied with airborne Doppler radar data, they have performed somewhat less spectacularly with more readily available conventional and satellite data. One of the primary problems is unacceptably poor representation of the TC in the model background (i.e. prior), a deficiency that necessitates a lengthy period of cycling to correct. However, long-term cycling is itself problematic, since shortcomings in regional model physics often lead to the development of systematic biases that may negatively impact subsequent forecasts. To address these issues, we propose a means of obtaining a significantly improved background by replacing the model TC with a bogus created using a sophisticated new technique that explicitly accounts for the primary and secondary circulations as well as the boundary layer flow. We extend this method further by sampling the TC position and intensity about their estimated values and creating an ensemble of model states which is then used within a GSI Hybrid data assimilation framework. By assimilating enhanced atmospheric motion vectors (AMVs) using an hourly cycle over short windows (6-12hr) for hurricane Ike (2008), we demonstrate that the ensemble bogus technique produces excellent analyses and, when used to initialize forecasts, performs better than initializations obtained with either the straightforward GSI or single bogus approaches.