Comparison of soft-computing and statistical techniques in simulating daily river flow: a case study in India

Abstract

This study compares the potential of soft-computing and statistical techniques to simulate the daily discharge at Tekra site on Pranhita River basin, India. The soft computing techniques include co-active neurofuzzy inference system (CANFIS) and multi-layer perceptron (MLP), and the statistical techniques include multiple non-linear and linear regressions (MNLR and MLR). The daily discharge data from June, 2000 to November, 2003 were used for this purpose. Before starting the simulation process, the appropriate combination of input variables for CANFIS, MLP, MNLR and MLR was selected using the gamma test (GT). The estimates provided by CANFIS, MLP, MNLR and MLR models were compared with the observed values of discharge (Qt) using the root mean squared errors (RMSE), Nash-Sutcliffe efficiency (NSE), and Pearson correlation coefficient(r). The results indicate that the performance of CANFIS models was better than MLP, MNLR and MLR models in simulating the daily discharge for the study location.