Drivers of Rain-on-Snow Runoff: An Explainable AI Approach

Rain-on-snow (ROS) events significantly impact hydrological processes in snowy regions, yet their seasonal drivers remain poorly understood, particularly in low-elevation and low-gradient catchments. This study uses an XGBoost-SHAP explainable artificial intelligence (XAI) model to analyze meteorological and watershed controls on ROS runoff in the Laurentian Great Lakes region. We compiled daily discharge, precipitation, temperature, and snow depth data from 2000 to 2023 from the HYSETS database to identify ROS events and their associated runoff. The XGBoost model was trained separately for winter and spring seasons to predict ROS runoff, while SHAP (SHapley Additive exPlanations) values were calculated to quantify the contribution and direction of influence of each predictor variable. Input features included climatic variables (rainfall amount, air temperature, snow depth), watershed characteristics (soil permeability, slope, aspect), and land cover types (agricultural, forest, shrub coverage). This approach enables systematic identification of dominant controls on ROS runoff and their seasonal variations across diverse catchment conditions.