Regional Weather Prediction Using the Local Particle Filter

Particle filters (PFs) are sequential Monte Carlo methods that can solve data assimilation problems characterized by non-Gaussian error distributions for prior variables or measurements. Recent efforts to apply PFs for high-dimensional geophysical models have resulted in “localized” PFs, which treat data assimilation problems as large sets of loosely coupled problems that can be solved independently -- much like localized ensemble Kalman filters. These methods significantly reduce the number of particles required for applications consisting of large spatial dimensions like weather prediction models. The current study provides a month-long test of a localized sequential importance resampling (SIR) PF technique for synoptic-scale weather forecasting using an experimental research-to-operations modeling testbed at the NOAA Atlantic Oceanographic and Meteorological Laboratory (AOML). The testbed is designed for hurricane research, model development, and satellite data assimilation activities at NOAA AOML. It is based on the operational Global Statistical Interpolation (GSI) data assimilation package and the Hurricane Weather Research and Forecasting (HWRF) model, and follows the same 6-h data assimilation schedule as the operational Global Forecast System (GFS), but with a regional domain that covers the Atlantic hurricane basin. Using the experimental modeling system, the local PF is compared with the GSI Ensemble Kalman filter over a period that includes the formation of multiple major hurricanes. This research investigates advantages of a localized PF for weather events known to pose challenges for Gaussian filters and smoothers, and provides new insight into how PFs may benefit future numerical weather prediction efforts.