Multiple regression modelling approach for rainfall prediction using large-scale climate indices as potential predictors

Abstract

Some studies established the associations with different climate indices (Southern Oscillation Index, Indian Ocean Dipole and Southern Annular Mode) and seasonal rainfalls of different parts of Australia. Nevertheless, maximum predictability of South Australian rainfall was only 20% with individual effects of potential predictor. To establish a better relationship for South Australian spring rainfall prediction, this paper presents two further investigations: 1) relationship of lagged climate indices with rainfall; 2) combined influence of these lagged climate indicators on rainfall. Multiple linear regression (MLR) modelling was used to evaluate the influence of combined predictors. Three rainfall stations were selected from South Australia as a case study. It was revealed that significantly increased rainfall predictability has been achieved through MR models using the influences of combine-lagged climate predictors. The rainfall predictability ranging from 41% to 45% has been achieved using combined lagged-indices, whereas maximum 33% predictability can be achieved using individual climate index.