Improving dynamical seasonal rainfall forecasts for hydrological applications

Predictability of seasonal streamflows relies on good forecasts of rainfall. Forecasting rainfall at the seasonal time scale remains highly challenging and it is therefore critical to extract the most skill out of available forecasting models using advanced techniques. In this study, we apply methods to improve seasonal rainfall forecasts from multiple general circulation models through application of forecast calibration and bridging techniques. We apply Bayesian model averaging to merge the resulting forecasts. Our application to forecasting seasonal rainfall across Australia shows the spatial and temporal coverage of positive forecast skill increases by adopting these approaches. We also find the merged forecasts reliably quantify forecast uncertainty.