Seasonal precipitation forecasts over China through calibration and combination of multiple CGCMs

To alleviate systematic deficiencies of two latest seasonal forecast systems in forecasting seasonal precipitation over China, a slightly modified Bayesian joint probability (BJP) modelling approach was employed to calibrate the ensemble means of the raw forecasts firstly. The calibrated forecasts were then merged through Bayesian model averaging (BMA) to combine strengths from different models. The results suggested that the BJP calibration models effectively removed biases and improved both reliability and overall accuracy of the raw forecasts. The calibrated ECMWF System4 (SYS4) forecasts exhibited some skill over broad regions of China in most seasons, whereas the calibrated Australian Bureau of Meteorology’s POAMA2.4 (P2.4) forecasts only showed weak skill over some regions in some seasons. Forecast skill of the merged forecasts from both sets of calibration models was improved greatly. Comparing with the SYS4 and P2.4 calibration forecast, the proportion of grid cells with positive RMSEP skill score was improved by 13.3% and 20.0%, respectively.