Rainfall Data Modeling with Artificial Neural Networks Approach

Abstract

Rainfall is one of the important information that widely used in various fields. Rainfall data involving location information is referred to as spatial rainfall data. Some of the model approaches to spatial rainfall data are the Vector Autoregressive (VAR) model, state-space, Markov chain stochastic model, and Geographically Weighted Regression (GWR). However, these models have not been able to produce predictions of the occurrence of no rain (zero value) or extreme values. Currently, theoretical modelling is mostly approached by artificial neural network (ANN) techniques. The purpose of this study is to model spatial rainfall data in East Java, Indonesia in 2020 with the ANN approach which is supported by several variables such as location and elevation information. The ANN used backpropagation and Rporp by combining the learning rate and layer which is then obtained the RMSE value. The results show that the best model has the smallest RMSE of 1.22 when the learning rate is 0.15 on 11 layers using Rprop algorithm.