Improved sub-seasonal meteorological forecast skill using weighted multi-model ensemble simulations

Sub-seasonal to seasonal weather and hydrological forecasts have the potential to provide vital information for a variety of water-related decision makers. Here, we investigate the skill of four sub-seasonal forecast models from phase-2 of the North American Multi-Model Ensemble using reforecasts for the period 1982-2012. Two weighted multi-model ensemble means from the models have been developed for predictions of both sub-seasonal precipitation and temperature. By combining models through optimal weights, the multi-model forecast skill is significantly improved compared to a 'standard' equally weighted multi-model forecast mean. We show that optimal model weights are robust and the forecast skill is maintained for increased length of time and regions with a low initial forecast skill show significant skill after optimal weighting of the individual model forecast. The sub-seasonal model forecasts models show high skill over the tropics, approximating their skill at monthly resolution. Using the weighted approach, a significant increase is found in the forecast skill for dry, wet, cold and warm extreme events. The weighted mean approach brings significant advances to sub-seasonal forecasting due to its reduced uncertainty in the forecasts with a gain in forecast skill. This significantly improves their value for end-user applications and our ability to use them to prepare for upcoming extreme conditions, like floods and droughts.