Long-term seasonal rainfall forecasting using linear and non-linear modelling approaches: a case study for Western Australia

Abstract

This paper presents the comparison of the performances between the linear and non-linear modelling techniques in re-generating the patterns of long-term seasonal rainfall in Western Australia. To construct the linear and non-linear models, commonly adopted multiple linear regression (MLR) and artificial neural network (ANN) modelling approaches were applied. Lagged (past) values of the oceanic climate drivers, El Nino Southern Oscillation (ENSO) and Indian Ocean Dipole (IOD) were considered to be the prospective forecasters of seasonal rainfall. The MLR models which were statistically significant and not susceptible to multicollinearity problems were considered as the potential models. The Lavenberg–Marquardt algorithm with Multilayer Perceptron training rule was adopted to construct the non-linear ANN models. The capability of both the MLR and ANN models were evaluated through commonly used statistical parameters. Since rainfall vary not only temporally but also spatially, the analysis were performed on regional scale. The methods were applied to three Western Australian rainfall stations. As expected, in estimating Western Australian spring rainfalls, non-linear ANN models performed much better compared to MLR models in regards to Pearson correlation as well as statistical errors.