Soft Computing For Hydrological Modelling: A Proportional Reading Of ANN And MLR Regarding Runoff Prediction

Hydrological modelling is essential for water resource management, particularly in forecasting runoff to mitigate flood and drought risks. This study applies Artificial Neural Networks (ANN) and Multi-Linear Regression (MLR) to predict a day advanceexcessrainfall in the Krishna Catchment, focusing on the Hiranyakeshi, Ghataprabha, and Tambraparni watersheds in Maharashtra, India. The study developed hydrological models using daily rainfall, antecedent rainfall, runoff, antecedent runoff with one to three-day time lags, and daily evaporation as inputs. A Multilayer Perceptron (MLP) ANN with a feed-forward backpropagation network is employed, and model performance is assessed using statistical indicators. The results show that ANN models significantly outperform MLR in predicting runoff, particularly in capturing peak flows. The Rainfall-Runoff Model demonstrates the highest predictive accuracy, achieving an R-value above 0.90, while the best MLR model achieves approximately 0.85. ANN models exhibit lower Mean Square Error (MSE) and Root Mean Square Error(RMSE) values, confirming their superior predictive capabilities, though minor prediction of peak flows is noted, remaining within 20% of observed values. MLR models provide reasonable estimates but struggle with nonlinearity, limiting their accuracy in extreme conditions. The findings highlight ANN's effectiveness in handling complex hydrological processes, making it a suitable tool for real-time runoff forecasting and flood risk management. Future research should explore additional hydrological parameters like soil moisture and temperature variations to enhance prediction accuracy and consider integrating ANN with optimization. The study establishes a data-driven approach for improved runoff forecasting, offering a framework applicable to other river basins for efficient water resource planning and disaster management.