Assessment of Calibration and Combination of Multi-Model Ensembles for Subseasonal Prediction

The NOAA Climate Prediction Center (CPC) began operational subseasonal forecasts in October of 2015. CPC is evaluating retrospective and real-time ensemble prediction system (EPS) subseasonal forecasts from the European Centre for Medium-range Weather Forecasts (ECMWF), the Japan Meteorological Agency (JMA), and the NCEP Climate Forecast System (CFS). The North American Multi-Model Ensemble (NMME) models are now providing data at daily resolution for research on subseasonal prediction. The Canadian Meteorological Centre EPS and the NCEP Global Ensemble Forecast System (GEFS), which comprise the North American Ensemble Forecast System (NAEFS), are to provide model data on this time scale in the near term. To ensure that ensemble probability forecasts are unbiased and provide reliable probabilities representing the likelihood of events, a method of calibration of ensemble forecasts, or ensemble regression, is applied individually to each model (Unger et al., 2009) before combination into a multiple model ensemble (MME). Bias, dispersion, correlation and mean square error of direct and post-processed ensemble model forecasts are evaluated individually and combined as an MME, assessing the value of the MME to subseasonal predictions.