Assessing Climate and Watershed Controls on Rain-on-Snow Runoff Using XGBoost-SHAP Explainable AI (XAI)

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 used daily discharge, precipitation, temperature, and snow depth data from 2000 to 2023, available from HYSETS, to identify ROS runoff. The XGBoost model’s performance for predicting ROS runoff was higher in winter (R2 = 0.65, Nash–Sutcliffe = 0.59) than in spring (R2 = 0.56, Nash–Sutcliffe = 0.49), indicating greater predictability in colder months. The results reveal that rainfall and temperature dominated ROS runoff generation, jointly explaining more than 60% of total model importance, while snow depth accounted for 8–12% depending on season. Winter runoff is predominantly governed by climatic factors—rainfall, air temperature, and their interactions—with soil permeability and slope orientation playing secondary roles. In contrast, spring runoff shows increased sensitivity to land cover characteristics, particularly agricultural and shrub cover, as vegetation-driven processes become more influential. Snow depth effects shift from predominantly negative in winter, where snow acts as storage, to positive contributions in spring at shallow to moderate depths. ROS runoff responded positively to air temperatures exceeding approximately 2.5 °C in both winter and spring. Land cover influences on ROS runoff differ by vegetation type and season. Agricultural areas consistently increase runoff in both seasons, likely due to limited infiltration, whereas shrub-dominated regions exhibit stronger runoff enhancement in spring. The seasonal shift in dominant controls underscores the importance of accounting for land–climate interactions in predicting ROS runoff under future climate scenarios. These insights are essential for improving flood forecasting, managing water resources, and developing adaptive strategies.