Radial Basis Function Neural Networks for Rainfall-Runoff Modeling

K. S Kasiviswanathan

doi:10.3759/jowrem.v1i2.1786

Abstract

Rainfall-runoff process is purely nonlinear and varies spatially as well as temporally. Any hydrological model requires many parameters which represent different components of the process. Availability of all the parameters is difficult for any catchment and probabilistic generation of such type of data is impossible. Under such circumstances, artificial neural networks (ANNs) have proven to be a better tool to model the rainfall-runoff process with minimum available data. The present study is to compare the performance of the model trained with K-means clustering algorithm and modified K-means clustering algorithm. The potential of these two algorithms was tested by developing rainfall runoff models for Vamsadhara river basin located in Andhrapradesh, India. Results of these two models were compared with observed data of Vamsadhara river basin. It is shown that modified K-means clustering algorithm results are more generalized than K-means clustering algorithm results.