Landscape Fragmentation Driven by Land Use Change Dominates the Spatiotemporal Heterogeneity of Non‐Point Source Pollution Loads in Human‐Intensive Basins: A Case Study of the Dongjiang River Basin, China

Landscape fragmentation may influence the spatial distribution and temporal variation of non‐point source (NPS) pollution by altering hydrological processes, pollutant transport pathways, and ecosystem service functions. However, under the combined effects of natural factors (e.g., precipitation, topography) and anthropogenic factors (e.g., livestock density, fertilizer application), the degree and quantifiable contribution of landscape fragmentation to basin‐scale NPS pollution remain inadequately understood. In this study, we constructed a comprehensive Landscape Fragmentation Index (LFI) by integrating 10 landscape pattern metrics using Principal Component Analysis (PCA). Subsequently, we developed an improved ECM‐SWAT model to simulate the spatiotemporal variation of total nitrogen (TN) and total phosphorus (TP) loads at the sub‐basin scale in Dongjiang River Basin (DRB) from 2000 to 2022. To reveal how landscape fragmentation shapes the spatial and temporal heterogeneity of pollution loads within the basin, we proposed a multi‐model analytical framework that incorporates Redundancy Analysis (RDA), bivariate Moran's I, interpretable machine learning methods, and the Geographically and Temporally Weighted Regression (GTWR) model. The key findings are as follows: (1) The enhanced model significantly improved simulation accuracy ( R 2 > 0.9), with R 2 increasing by approximately 14% compared to the traditional SWAT model. It also identified the urbanization‐driven midstream and downstream areas of DRB as pollution hotspots. (2) Based on RDA and bivariate Moran's I, the LFI has emerged as a dominant factor influencing NPS pollution in the basin since 2010 ( p ‐value = 0.001***), with hotspots concentrated in the downstream regions with intensive human activities, and coldspots distributed in the midstream and upstream areas. (3) Finally, interpretable machine learning and GTWR analyses indicated that LFI contributed 18.6% and 4.9% to TN and TP pollution, respectively, while Population, LS, LFI, and Slope were all identified as key drivers of pollution risk variation, LFI exhibited the highest regression coefficient. Its influence demonstrated a distinct spatial gradient, shifting from a negative effect in the southwestern region (C LFI = −7.3) to a positive effect in the eastern areas (C LFI = 9.16). These results underscore the importance of implementing landscape‐scale management in basin‐wide NPS pollution control. Incorporating landscape fragmentation as a key indicator into the optimization of Best Management Practices can greatly enhance the precision and effectiveness of future pollution mitigation strategies.