Abstract
Groundwater contamination by arsenic (As) poses a severe health risk to the public. Although the vadose zone plays a crucial role in controlling As transport, current regulatory frameworks conservatively assume an attenuation factor (AF) of 1 for unsaturated flow conditions (i.e., no attenuation). In this study, machine learning-based regression models were integrated with the mobile-immobile water (MIM) model for solute transport to develop a data-driven framework for deriving site-specific AF values of As in the vadose zone. The regression models were applied to rapidly estimate MIM modeling parameters related to the transport and fate of As based on readily available soil properties for the vadose zone. As transport was simulated by considering infiltrations controlled by both retardation and remobilization induced by wet-dry cycles. When applied to 21 sites across South Korea, the framework yielded AF values of >1 at all locations with substantial variability (3.56-24.38), which highlights the buffering capacity of the vadose zone and the necessity of site-specific assessment. This scalable framework offers a practical alternative to conventional reactive transport models that facilitates reliable risk assessments of As groundwater contamination by the infiltrates through the vadose zone.