Statistical calibration and bridging of ECMWF System4 outputs for forecasting seasonal precipitation over China

This study evaluates seasonal precipitation forecasts over China produced by statistically postprocessing multiple-output fields from the European Centre for Medium-Range Weather Forecasts' System4 (SYS4) coupled ocean-atmosphere general circulation model (CGCM). To ameliorate systematic deficiencies in the SYS4 precipitation forecasts, we apply a Bayesian joint probability (BJP) modeling approach to calibrate the raw forecasts. To improve the skill of the calibration forecasts, we use six large-scale climate indices, calculated from SYS4 sea surface temperature forecasts, to establish a set of BJP statistical bridging models to forecast precipitation. The calibration forecasts and bridging forecasts are merged through Bayesian model averaging to combine strengths of the different models. The BJP calibration effectively removes bias and improves statistical reliability of the raw forecasts. The calibration forecasts are skillful at a 0 month lead in most seasons, but skill decreases sharply at a 1 month lead. The skill of the bridging forecasts is more stable at different lead times. Consequently, the merged calibration and bridging forecasts at a 1 month lead are clearly more skillful than the calibration forecasts, and the skill is maintained out to a 4 month lead. The forecast framework used in this study can help to better realize the potential of CGCM ensemble forecasts. The increased reliability as well as improved skill of seasonal precipitation forecasts suggests that the system proposed here could be a useful operational forecasting tool.