Integrating Landsat-Derived Surface Water Occurrence Frequency and XGBoost-SHAP to Reveal Nonlinear Drivers of Surface Water Dynamics in the Lixiahe Plain, China

Hydroclimatic variability and intensive human regulation have reshaped surface water dynamics in plain river network regions (PRNs), altering both water extent and stability. Unraveling the frequency structure of surface water and its nonlinear associations is critical for sustainable water resource management. Based on Landsat-derived surface water dynamics from 2015 to 2025, this study constructed water occurrence frequency (WOF) indicators for Lixiahe Plain (LP) and employed XGBoost-SHAP to identify key natural and human factors, nonlinear responses, and interaction effects. The results showed that: (1) the annual water surface ratio (WSR) exhibited a phased inverted-U pattern, increasing from 5.22% in 2015 to 6.04% in 2018, remaining high during 2018–2021, and decreasing to 4.57% in 2025; (2) WOF revealed marked stability restructuring, with low-frequency water accounting for 31.0–53.7% of annual water and becoming more dominant after 2023; (3) the XGBoost-SHAP models showed reliable performance, with R2 of 0.714–0.813 and RMSE below 0.018. Natural factors contributed more than human factors, and wind speed (WS), normalized difference vegetation index (NDVI) related features, and cropland ratio (CropR) were the influential factors, with approximate response transitions of 2.5–3.0 m s−1, 0.50–0.52, and 65–68%. These findings support refined water resource management and sustainable development in PRNs.