Modeling runoff in Bhima River catchment, India: A comparison of artificial neural networks and empirical models

Abstract

Effective water resource management in gauged catchments relies on accurate runoff prediction. For ungauged catchments, empirical models are used due to limited data availability. This study applied artificial neural networks (ANNs) and empirical models to predict runoff in the Bhima River basin. Among the tested models, the ANN-5 model, which utilized rainfall and one-day delayed rainfall as inputs, demonstrated superior performance with minimal error and high efficiency. Statistical results for the ANN-5 model showed excellent outcomes during both training (R = 0.95, NSE = 0.89, RMSE = 17.39, MAE = 0.12, d = 0.97, MBE = 0.12) and testing (R = 0.94, NSE = 0.88, RMSE = 11.47, MAE = 0.03, d = 0.97, MBE = 0.03). Among empirical models, the Coutagine model was the most accurate, with R = 0.82, MBE = 74.36, NSE = 0.94, d = 0.82, KGE = 0.76, MAE = 70.01, MAPE = 20.6%, NRMSE = 0.22, RMSE = 87.4, and DRV = −9.2. In contrast, Khosla's formula (KF) significantly overestimated runoff. The close correlation between observed and ANN-predicted runoff data underscores the model's utility for decision-makers in inflow forecasting, water resource planning, management, and flood forecasting.