Evaluation of a Bayesian Data Assimilation Method in an Hourly Cycled Global Weather Prediction System

Current operational centers leverage ensemble Kalman filters and variational data assimilation methods to produce routine forecasts for global and regional weather. These strategies, however, can be restrictive for non-Gaussian forecast and observation errors, which narrows research opportunities on this topic. Localized particle filters provide an alternative methodology that relaxes many of the common assumptions made during operational data assimilation and consequently opens new pathways for research on non-Gaussian and non-linear problems such as the assimilation of all-sky satellite radiances to improve weather forecasts. This presentation discusses progress toward a rapidly updating (hourly) local particle filter for global weather prediction. We motivate these developments with idealized experiments performed using the 40-variable model of Lorenz (1996), which replications limitations in the current operational data assimilation methodology for global weather prediction.