Runoff Prediction Using Hybrid Neural Networks in Semi-Arid Watershed, India: A Case Study

Abstract

Predicting runoff is a nonlinear and intricate procedure that is very much important to design canals, managing and arrangement of water usability, controlling flood, and prediction of soil erosion. Many methods to predict runoff on basis of hydro-meteorological and geomorphologic condition is available and being readily used. But recently, a number of soft computing methods have evolved for predicting runoff. The proposed study uses a crossbreed smart model which is an amalgamation of data pre-processing techniques, Genetic Algorithm (GA), and Support Vector Machine (SVM) algorithm for Neural Networks (NN). This manuscript explores the efficiency of SVM evolved NN to forecast rainfall-runoff and usability of this for predicting runoff in Balangir watershed. For an affluent organization of water resources in semi-arid province, predicting one-day lead runoff is very much essential. At a specified instant for predicting runoff, input variables taken into consideration are rainfall and runoff that are witnessed on earlier time period. Genetic operator is cautiously considered for optimizing the NN to avoid untimely meeting and variation problems.