Rainfall-Runoff Modeling Using Artificial Neural Network Technique: A Case Study of Semi-Arid Region of Rajasthan

Artificial neural networks (ANNs) have been used widely in problems related to water resources engineering. The ANNs are very much useful to quantify the complex hydrological processes in the catchment. In this research, the rainfall-runoff modeling has been carried out for the Bigod sub-catchment, a part of Banas basin, semi-arid region of Rajasthan. This study area covers the 18 in situ rain gage stations and one stream gaging station. The multiple ANN models have been used using Nntool in a MATLAB environment. It was observed that the results produced by Nntool are consistent and should be utilized to understand the catchment response, for the better management of available water resources in the watershed. The hydro-meteorological data like rainfall, runoff, temperature, specific humidity, and wind speed were utilized for rainfall-runoff modeling. The ANN regression plot using Levenberg–Marquardt method using 30, 40, and 50 number of neurons was generated. However, ANN regression plot using scaled conjugate gradient method using 30, 40, and 50 number of neurons was also generated. The comparative analysis of ANN models is discussed, based on correlative coefficient magnitude. The value of coefficient correlation revealed that the scaled conjugate gradient method with 30 number of neurons performs better in runoff predictions, while a further increase in number of neurons deteriorate the model performance in runoff prediction. Based on comparative analysis, the best-fit algorithm with a number of neurons was identified. These results can be implemented by water resource engineers, hydrologists, policymakers, and various govt. agencies to minimize the flood and drought consequences in the watershed.