Abstract: Serving as the core water source for the Middle Route of China's South-to-North Water Diversion Project, the Danjiangkou Reservoir area confronts substantial risks to its water quality security from non-point source (NPS) pollution. Based on the "source-sink" landscape theory and the Minimum Cumulative Resistance (MCR) model, this study utilized multi-temporal land use data (2000-2020) to assess the spatiotemporal dynamics of NPS pollution risk in the Danjiangkou Reservoir area. Furthermore, it investigated the mechanisms by which landscape spatial patterns influence NPS pollution risk. Eight resistance factors, including relative elevation, the Normalized Difference Vegetation Index (NDVI), rainfall erosivity, and the topographic wetness index, were incorporated to develop a comprehensive resistance surface. Then, a random forest algorithm was applied to examine how various landscape pattern indices influenced the resistance values associated with NPS pollution. The results demonstrated that high-risk zones for NPS pollution were persistently clustered in southeastern areas near water bodies throughout the examined period, comprising approximately 22.6% of the reservoir's total area. Analysis of land-use transition dynamics revealed that both stable forestland and reforested areas exhibited a gradual increase in resistance values over time, demonstrating the sustained and stable high resistance characteristics of forestland in controlling NPS pollution. Notably, reforested areas presented lower resistance values than stable forestland, indicating that while reforestation improved interception capacity against NPS pollution, its efficacy remained secondary to that of long-established forests. Degraded forestland showed a significant decreasing trend in resistance value over time (p<0.05), and its resistance value was lower than that of the stable forestland and the reforested areas. According to the random forest algorithm, the patch cohesion index emerged as the most influential landscape pattern index affecting NPS pollution resistance throughout the study period, with relative importance values varying between 28.41% and 34.34%. This was followed by the aggregation index and the landscape shape index, which displayed relative importance ranges of 12.62%-16.18% and 11.15%-17.83%, respectively. Based on the above analysis, NPS pollution risk in the Danjiangkou Reservoir area is influenced not only by land use types but also, and more critically, by the spatial configuration of the landscape. Optimizing landscape patterns, particularly through enhancing the connectivity of "sink" landscapes, constitutes a strategic approach to intercept pollutant transport and ensure water quality safety for the Danjiangkou Reservoir area.