Multi-model drought predictions using temporally aggregated climate indicators

Abstract

Long term skillful prediction of prolonged hydrologic anomalies is essential for proper planning to reduce the societal risk of extreme hydrological anomalies such as drought. Climate indices estimated from sea surface temperature anomalies (SSTA) of the Pacific and Indian Oceans are often used to predict monthly and seasonal rainfall in Australia and many other places around the world. This study investigates the merit of distorting the time aggregation of such indices before casting them in a predictive model. Aggregated climate indices are used to predict sustained drought and wet anomalies characterised here using a drought index (i.e. Standardize Precipitation Index, SPI) as response and the Australia as the study region of interest. The aim is to enhance the strength of relationships of drought index and climate indices (predictors) by tuning the frequency of climate predictors using an aggregation technique. Result shows that aggregated climate indices provide significant improvement in prediction of SPI over raw climate indices across Australia. As strong spatial variations in optimum aggregation window lengths are evident across Australia suggesting multiple candidate predictive models with similar accuracies, a model combination approach is also adopted. Model combination is found useful in reducing structural uncertainty and further improving the prediction efficiency. Given that the improved predictive accuracy for SSTA the current generation of climate models exhibit, the methodology developed in this study has significant implications for skillful prediction and projection of long term droughts.