Objective This study aims to elucidate the mechanisms through which landscape pattern dynamics affect watershed ecohydrological processes, thereby providing a scientific basis for optimizing landscape configuration, improving water resource management, and mitigating soil erosion.

Method Using the Daning River Basin in the Three Gorges Reservoir Area as a case study, landscape composition and configuration metrics were derived from land-use data for 2000, 2010, and 2020. A calibrated and validated SWAT model was used to simulate annual runoff depth and specific sediment yield at the sub-basin scale, while XGBoost-SHAP was applied to quantify the relative importance and effects of precipitation, topography, land use, and landscape-pattern variables.

Result (1) Forest and cropland remained dominant, together accounting for more than 84% of the basin. Forest area increased slightly, whereas cropland and grassland declined; built-up land and water bodies expanded by 449.06% and 161.55%, respectively. (2) The landscape shifted overall from fragmentation toward aggregation, as indicated by increases in AI, COHESION, and CONTAG and decreases in PD, ED, and LSI, suggesting enhanced aggregation and structural connectivity. (3) Annual runoff depth was generally higher in the northwest and lower in the southeast, whereas specific sediment yield was higher in the northern and central regions and lower in the southeast. High-sediment-yield areas largely coincided with fragmented landscapes and intensive human disturbance. (4) Runoff depth was controlled primarily by precipitation and topography, whereas sediment yield was more sensitive to land-use composition and landscape fragmentation.

Conclusion unoff depth and sediment yield in the Daning River Basin respond to different dominant controls. Maintaining the spatial integrity of forests and other ecological patches, optimizing the distribution of sloping cropland, and reducing fragmentation in high-risk areas are essential for mitigating soil-erosion risk.