Can improved SSTA prediction be translated into better seasonal rainfall forecast?

Seasonal rainfall forecasts are predicted throughout Australia based on concurrent sea surface temperature anomalies (SSTAs) fields coming from the main dynamical Australian model, the Predictive Ocean-Atmosphere Model for Australia (POAMA). In this study, we derive SSTA fields using a multi-model combination approach by including information from five additional models. The combination takes into account the cross-dependence between all the models while combining individual SSTA fields. The resulting SSTA fields are then used to derive SSTA indices to issue seasonal rainfall forecasts rainfall over a 2.5 degree grid using a Bayesian Model Averaging approach currently in use operationally. These forecasts are compared with those derived from a single SSTA model in three settings: (i) 2.5 degree gridded rainfall over Australia; (ii) only grid cells where one of the models (multi-model or single model) had skill (measured in terms of mean squared error relative to climatology) ; and (iii) only grid cells where both of the models had skill. The results indicate that the forecasts derived using multi-model based SSTA indices offer clear improvements over the case where a single SSTA model is used in three seasons, ASO (August, September and October), FMA (February, March and April), and MJJ (May, June and July). A further assessment over grid locations where predictability is significantly better than climatology indicates consistent improvements when the multi-model SSTA fields are used.