On the optimization of grand multi-model probabilistic performance and the independence of the contributing seasonal prediction systems

To optimize the performance of seasonal climate forecasts we used a Grand Multi-Model Ensemble (MME) approach. The Grand MME consists of five Seasonal Prediction Systems (SPSs) provided by the European Copernicus Climate Change Service (C3S) and of other six SPSs independently developed by centres outside Europe, five by the North American (NMME) plus the SPS by the Japan Meteorological Agency (JMA).

All the possible Grand MME combinations have been evaluated for temperature and precipitation, for different geographical regions. Results show that, in general, only a limited number of SPSs is required to maximize the skill. Although the selection of models that optimize performance is usually different depending on the region, variable and season, it is shown that the performance of the Grand-MME seasonal predictions is enhanced with the increase of the independence of the contributing SPSs.

Independence is measured by using  a novel metric developed here, named the Brier score covariance (BScov), which estimates the relative independence of the SPSs. Together with probabilistic skill metrics, BScov is used to develop a strategy for an effective identification of the combinations of SPSs that optimize the probabilistic performance of the predictions, thus avoiding the inefficient and ineffective use of all SPSs available.