Interpreting, Improving, and Augmenting Multi-Model Ensembles

It is operational impossible to maintain a multi-model development, data assimilation, and forecasting system at a single NWP center. This motivates extracting as much information as possible from the analyses and forecasts currently available from different operational NWP centers. This collection of analyses and forecasts from different NWP centers is denoted the poor man’s multi-model (PM MM) ensemble. Because the existing PM MM has few members (there are only a handful of operational NWP centers around the world), methods for extracting as much information as possible from the ensemble are of interest. To increase the effective ensemble size without adding additional models, this project will explore implementing a lagged average forecasting technique where forecasts launched at different times are combined at common verification times. Because forecasts at longer leads lack the observational information available to short lead forecasts, the ensemble transform Kalman filter (ET KF) (Bishop et al, 2001) will be utilized to incorporate observations into existing