Drought forecasting using adaptive stochastic models in New South Wales

Abstract

Australia is the driest continent, and drought is a serious threat. Prediction of droughts would help reduce their consequences especially for agriculture. The Standardized Precipitation Index (SPI) is a common measure of drought severity and can be calculated over multiple time scales. Rainfall data are available from three sites in NSW. Models for predicting the three-month SPI at lead times of one and two months are compared on the basis of root mean squared error (RMSE) and a RMSE restricted to drought periods. These are ARMA models for the SPI time series and the predictions based on predicting monthly rainfall. The benefits of including the Southern Oscillation Index (SOI) and the Multivariate ENSO Index (MEI) in the models are investigated. Similar models for predicting the twelve-month SPI at a lead-time of six months are compared.