Neural Network Based Approaches for Prediction of the Indian Summer Monsoon Rainfall

Abstract

This paper investigated the application of Single Layer Feed Forward Neural Network (SLFN), Extreme Learning Machine (ELM) and Regularized Online Sequential - Random Vector Functional Link (ROS-RVFL) neural network for prediction of the Indian Summer Monsoon Rainfall (ISMR) using Sea Surface Temperature (SST), Sea Level Pressure (SLP) and the combination of SST and SLP (SST+SLP) as input predictors. Nine sets of predictions are made using the SLFN, ELM, and ROS-RVFL techniques by applying SST, SLP, and the combination of SST and SLP (SST+SLP) predictors. It is found that when the combination of SST and SLP (SST+SLP) is chosen as the input predictor, the aforesaid neural networks based approaches produce minimal error scores and provide better accuracy. The ROS-RVFL technique outperformed in all the trials used in this study and the ELM technique produced more accurate outcomes than the SLFN technique.