Comparing variational, ensemble, and “hybrid” variational/ensemble data assimilation techniques for limited-area tropical cyclone forecasts

Three-dimensional variational (3DVAR), ensemble Kalman filter (EnKF), and “hybrid” variational/ensemble data assimilation (DA) methods were employed over a limited-area domain encompassing the Western Pacific Ocean to study three typhoons (Sinlaku, Hagupit, and Jangmi). Between 4-28 September 2008, parallel 3DVAR, EnKF, and hybrid experiments produced new analyses every six hours using a “cycling” DA approach. The hybrid and EnKF configurations used 32-member ensembles. Each 0000 and 1200 UTC analysis initialized a 72-hr Advanced Research Weather Research and Forecasting (WRF) model forecast (from the ensemble mean analysis in the EnKF experiment).

The EnKF and hybrid experiments produced comparable or better track forecasts than the 3DVAR experiment for Sinlaku and Hagupit, while 3DVAR yielded better track forecasts for Jangmi. Furthermore, the hybrid and EnKF techniques produced better intensity forecasts than those initialized from 3DVAR analyses. Along with a presentation of these results, the formidable challenges associated with developing and tuning a limited-area hybrid DA system will be discussed.