Heuristic Optimization Recurrent Neural Network Model to Predict Rainfall Based on El-Nino Southern Oscillation (ENSO) Variable

Abstract

Prediction of rainfall in the agricultural sector has now become a major requirement, as well as the selection of seeds, fertilizer and pest eradication. Information about the rainfall is very useful to farmers in anticipation of extreme events such as droughts and floods. Therefore, the model prediction of rainfall needed a fast and accurate prediction. This research employ a recurrent optimized heuristic neural network approach using El-Nino Southern Oscilation (ENSO) variable, namely Wind, SST, SOI and OLR to forecast regional monthly rainfall. Three optimized heuristic learning algorithms are applied in recurrent Elman, i.e gradient descent adaptive learning rate with various parameter values, gradient descent adaptive learning rate & momentum with various parameter values, and resilient backpropagation with various parameter values. The best forecasting rainfall for leap 0 is resilient backpropagation algorithm application which is obtaining 77% maximum R2 with RMSE 138,52. While the best forecasting rainfall for leap 1 is resilient backpropagation algorithm application which is obtaining 84,8% maximum R2 with RMSE 125. Our result on leap 0 is better than previous approach which employing neural network standard backpropagation.