Towards a Multi-Model Subseasonal Forecast of U.S. Weather

Based upon interactions with the forecast applications community, there is an increased desire for multi-model forecast solutions. However, it is still not clear whether multi-model subseasonal predictions are an improvement over calibrated single model forecasts. Further, there are important gaps in the current understanding of how to strategically construct multi-model ensembles. This study examines the weeks 3-4 forecast skill of surface temperature and precipitation using the CFSv2 and the ECMWF medium range forecasting systems. We are directly comparing the skill of each system, with skill scores stratified by teleconnection and weather regimes, season, and region. Additionally we are using the non-homogenous Gaussian regression approach to create a multi-model systems, and compare it's skill with a simple multi-model forecast where all ensemble members have equal weight. To aid decision support integration, an interactive web-based dashboard system designed to display and deliver the forecast information and real time prediction skill in a flexible and dynamic manner is presented.