SENSITIVITY OF RECHARGE ESTIMATION TO INPUT PARAMETERS

Abstract

Groundwater recharge estimation is highly dependent on a range of climatic, hydrological, and land surface parameters, making it essential to understand the sensitivity of recharge predictions to input variability. This study investigates the sensitivity of groundwater recharge estimates to key input parameters—precipitation, evapotranspiration, soil moisture, land use, and hydraulic conductivity—using a combination of physically based water balance models and data-driven machine learning approaches. The Rainfall Infiltration Breakthrough (RIB) model was employed for vertical percolation simulation, while sensitivity analyses were performed using both One-at-a-Time (OAT) and global Sobol methods. Additionally, feature importance was evaluated using XGBoost and Random Forest models to validate parameter rankings. Results indicate that precipitation and evapotranspiration exert the most significant influence on recharge outcomes, with sensitivity indices exceeding 0.7 in all methods. Land use scenarios showed that forested areas promote higher recharge, while urban areas significantly reduce infiltration capacity. Machine learning models reinforced the dominance of climate-related features, further validating the model outputs. These findings underscore the critical importance of accurate input parameterization in groundwater modeling and highlight the value of integrated modeling approaches for sustainable water resource management.