Approximation of cumulative infiltration of soils in arid and semi-arid regions

Effective management of water resources in arid and semi-arid regions relies on accurate infiltration models to support groundwater recharge, irrigation, and flood control projects. The study includes a comparative analysis of the four machine learning techniques such as Deep Neural Network (DNN), Distributed Random Forest (DRF), Multivariate Adaptive Regression Splines (MARS) and Gradient Boosting Machine (GBM), for forecasting the cumulative infiltration of soils. Field infiltration tests were carried out at 16 different sites in Iran (Davood Rashid, Kelat and Honam) and 17 different sites in Haryana, India and total of 372 observations were used for further analysis. Results suggested that GBM has an edge over MARS to forecast the cumulative infiltration values where coefficient of correlation (CC) and Nash Sutcliffe model efficiency (NSE) values for GBM were 0.9856 and 0.9704, respectively, during validation stage. Sensitivity analysis suggests that time is the most influencing input parameter for the measurement and forecasting of cumulative infiltration of arid and semi-arid regions of India and Iran. Taylor diagram analysis confirmed GBM's superior performance and DNN's lower accuracy among all applied models. These findings underscore the potential of GBM for precise infiltration modeling, enhancing water management strategies in arid and semi-arid environments.