Enhancing infiltration rate predictions with hybrid machine learning and empirical models: addressing challenges in southern India

Abstract

Despite the success of machine learning (ML) in many disciplines, its application in hydrology, especially in water-scarce regions, faces challenges due to the lack of interpretability and physical consistency. This study addresses these challenges by integrating empirical hydrological models with ML techniques to predict infiltration rates in water-scarce regions of southern India. Using data from 199 observations across 11 sites, including soil characteristics and infiltration measurements, traditional models such as Philip’s, Horton’s, and Kostiakov’s were parameterized and combined with artificial neural networks (ANNs) and the MissForest (MF) algorithm to form hybrid models. The results demonstrate that the hybrid models, particularly those integrating Philip’s model with ANN and multiple predictors, achieved substantial improvements in prediction accuracy, with R2 values ranging from 0.803 to 0.918, root mean-square error (RMSE) from 0.083 to 0.118 cm/min, and Legates’ coefficient of efficiency (LCE) from 0.575 to 0.717 across the target sites. In contrast, empirical models alone at the test sites show lower performance, with R2 ranging from 0.499 to 0.902, RMSE from 0.091 to 0.152 cm/min, and LCE from 0.46 to 0.728, underscoring the limitations of traditional empirical models and the enhancement achieved through ML integration. By leveraging the strengths of empirical models and ML, the hybrid approach improves predictive accuracy and provides a more robust understanding of infiltration dynamics. The hybrid models enable accurate predictions using minimal, readily accessible data, offering a practical solution for water resource management and soil conservation in semi-arid, data-scarce regions. This study demonstrates that blending empirical knowledge with ML algorithms not only improves accuracy but also retains physical interpretability, presenting an innovative solution to hydrological modeling challenges in water-scarce environments. © The Author(s) under exclusive licence to Institute of Geophysics, Polish Academy of Sciences 2025.