Rainfall Forecasting Through ANN and SVM in Bolangir Watershed, India

Abstract

Simulating rainfall efficiently is a major complex problem owing to a high number of interrelated hydrological processes. Here, authors used various artificial neural network (ANN) methods to study rainfall forecast in Bolangir district, Odisha, India, by using past 48 years’ data. Different combinations of past rainfall values were formed as forecasting inputs for evaluating the efficiency of ANN approximation. RNN, SVM, and ANFIS techniques are employed for rainfall analysis. The performance of standardized ANN-based model and SVM has been accessed by taking a peak of observed and simulated floods and computation of coefficient of determination (R2) for the intermediate gauging stations on projected river basin. SVM gives better performance with the coefficient of determination 0.9526 and 0.9738 for both testing and training phase while in case ANFIS it gives 0.9342 and 0.9488. Simultaneously, RNN gives less value of R2 is 0.9039 and 0.9292 for training and testing phase.