Statistical Methods for Deriving Seasonal Climate Forecasts from GCM’S

Abstract

A method for developing a ‘practical’ climate forecast system from the output of a General Circulation Model (GCM) is described. This forecast system is compatible with input needs of agricultural simulation and decision analysis models. The method uses principal components and cluster analysis of GCM-generated forecasts of the time series of the Southern Oscillation Index (SOI) to create SOI ‘types’ or ‘phases’. The SOI predictions were derived from a long-term GCM simulation that was forced with historical sea-surface temperature (SST) data. The GCM-derived ‘SOI phases’ (g-phases) could thus be associated with historical analogue years in a manner similar to SOI phases that have been developed from historical SOI data. Rainfall probability distributions associated with g-phases were calculated from actual rainfall amounts in the analogue year sets associated with each phase. These rainfall probabilities were compared with the currently available distributions that have been derived using lag relationships between SOI phases and rainfall. In addition, the historical SST data was analysed in a similar way so that analogue year sets could be formed. Empirical orthogonal function (EOF) analysis and cluster analysis were used to derive SST ‘EOF-types’ that depended on the temporal dynamics of spatial patterns in the Pacific Ocean and Indian Ocean SST data. Lag relationships between the SST/EOF types and rainfall distributions could then also be derived for comparison. The results show that both the g-phases derived from the GCM forecast of SOI and the SST/EOF-types generally provide larger shifts in rainfall distributions than are currently available, especially at longer lead times. The methods outlined above may facilitate more practical output from General Circulation Models and analyses of SST data so that connections to agricultural simulation models and software packages are enhanced.