The effectiveness of ARIMAX model for prediction of summer rainfall in northwest Western Australia

Abstract

Inherent climate mechanism plays a large role in Australian rainfall events, while most interestingly, such influence of climate drivers on rainfall generation depends largely on a place’s geographical location and the season it has been passing through. For Northwest Western Australia (NWWA), summer has been the main rainfall season; therefore, it became necessary to develop a seasonal forecast model several months ahead of the event. This study involves the selection of potential climate drivers for NWWA and using them to forecast and model long-term seasonal summer rainfall (Dec to Feb) several months advance. Two rainfall stations (Bidyadanga and Gogo) were selected considering the availability of continuous monthly rainfall data (from 1915 to 2015) with a minimal missing value. Simple multiple linear regression (MLR) and ARIMAX have been performed on acquired rainfall data and lagged climate indices. It has been observed that the Southern Oscillation Index (SOI) and the Western Indian Ocean Index (WIO) has a great influence on NWWA rainfall. As such, the WIO-SOI model (4 months advance) shows a significantly higher correlation in the ARIMAX model (0.68 and 0.65 for Bidyadanga and Gogo station, respectively) compared to the MLR model (0.35 and 0.36) and confirms its effectiveness to be used for rainfall prediction in the region. It is believed that the simple but effective nature of the developed models can produce optimum benefit on stakeholder’s decision making to tackle future socio-economic challenges.