Paper Title
Application of Non-Linear Modelling Technique in Predicting Long-Term Seasonal Rainfall

Abstract
This paper presents the efficiency of non-linear modelling technique in predicting long-term seasonal rainfall of Western Australia. The emphasis was given to evaluate non-linear correlations between long-term seasonal rainfall and numerous significant climatic variables. One of the commonly sued non- linear modelling approaches, Artificial Neural Network (ANN) was adopted for the construction of the non-linear models. The models were developed considering the past values of El Nino Southern Oscillation (ENSO) and Indian Ocean Dipole (IOD) as the probable influential variables of rainfall. In this research, the non-linear ANN models were developed adopting the algorithm proposed by Lavenberg-Marquardt. The training of the ANN models were performed by implementing the Multilayer Perceptron (MLP) training algorithm. The developed non-linear models were verified with the independent data sets that were not applied during the training of the ANN analysis. The potential ANN models’ performances were assessed with the commonly used statistical parameters. Since rainfall vary not only temporally but also spatially, the non-linear ANN models were developed and tested on a provincial scale. In this paper, the non-linear analysis was performed for three Western Australian rainfall stations. The developed ANN models showed good generalisation capability of Western Australian spring rainfall with Pearson correlations varying from 0.73 to 0.96 during the training phase and 0.35 to 0.93 during the testing phase. The errors of the IOD-ENSO based non-linear models were also acceptable to be applied for rainfall forecasting. The index of the agreement further suggested that the non-linear ANN models could be used to predict seasonal rainfall. Keywords - Seasonal Rainfall, Rainfall Forecasting, ANN, ENSO, IOD