Rainfall–Runoff Modelling Using Hydrological Modelling And Soft Computing Techniques

Abstract

This investigation offers an in-depth evaluation comparing conventional watershed modeling methodologies against computational intelligence approaches for predicting precipitation-discharge relationships. The analysis implemented several modeling frameworks including two conceptual representations (the SCS Curve Number approach and HEC-HMS) and distributed physics-based simulations, alongside machine learning techniques including neural network architectures, fuzzy-based systems, statistical learning algorithms (SVMs), and combined methodologies. Utilizing meteorological and hydrological measurements gathered across a five-year timeframe (spanning 2017-2022), each model underwent systematic calibration and validation procedures following established protocols. The findings reveal that integrated computational intelligence frameworks demonstrated superior performance compared to traditional hydrological simulations, with the neural network-SVM hybrid configuration achieving the highest performance metrics (Nash-Sutcliffe Efficiency of 0.89 and Root Mean Square Error of 8.2 cubic meters per second). Parameter sensitivity evaluation determined that pre-existing soil water content and precipitation rate were the most significant variables influencing predictions across all modeling approaches. This investigation highlights the enhanced predictive capabilities of computational intelligence methodologies for complex watershed response patterns while emphasizing the continued relevance of process-based understanding provided by conventional hydrological models