Use of Teleconnections to Predict Western Australian Seasonal Rainfall Using ARIMAX Model

Increased demand for engineering propositions to forecast rainfall events in an area or region has resulted in developing different rainfall prediction models. Interestingly, rainfall is a very complicated natural system that requires consideration of various attributes. However, regardless of the predictability performance, easy to use models have always been welcomed over the complex and ambiguous alternatives. This study presents the development of Auto-Regressive Integrated Moving Average models with exogenous input (ARIMAX) to forecast autumn rainfall in the South West Division (SWD) of Western Australia (WA). Climate drivers such as Indian Ocean Dipole (IOD) and El Nino Southern Oscillation (ENSO) were used as predictors. Eight rainfall stations with 100 years of continuous data from two coastal regions (south coast and north coast) were selected. In the south coast region, Albany (0,1,1) with exogenous input DMIOct-Nino3(Nov), and Northampton (0,1,1) with exogenous input DMIJan-Nino3(Nov) were able to forecast autumn rainfall 4 months and 2 months in advance, respectively. Statistical performance of the ARIMAX model was compared with the multiple linear regression (MLR) model, where for calibration and validation periods, the ARIMAX model showed significantly higher correlations (0.60 and 0.80, respectively), compared to the MLR model (0.44 and 0.49, respectively). It was evident that the ARIMAX model can predict rainfall up to 4 months in advance, while the MLR has shown strict limitation of prediction up to 1 month in advance. For WA, the developed ARIMAX model can help to overcome the difficulty in seasonal rainfall prediction as well as its application can make an invaluable contribution to stakeholders' economic preparedness plans.