Hydrological connectivity and landscape composition shape nutrient export: Evidence from SWAT and machine learning analysis in the Longxi River Basin

Understanding how hydrological connectivity and landscape patterns influence nutrient export is critical for clarifying the mechanisms driving non-point source pollution in river basins, especially where agriculture is largely practice. This study employed the Soil and Water Assessment Tool (SWAT) to simulate total nitrogen (TN) and total phosphorus (TP) export loads in the Longxi River Basin (LRB, in China) during 2021–2022. Using Random Forest (RF) and partial least squares structural equation modeling (PLS-SEM), the research explored the influences of terrain, landscape configuration, landscape composition, and aggregated hydrological connectivity (AIC) on TN and TP exports. Results indicated that average summer TN and TP exports accounted for 51.11% and 52.55% of the annual totals, respectively. Liangping and Dianjiang Countiy, along with the adjacent Changshou Lake reservoir, exhibited the highest hydrological connectivity indices within the watershed, especially during summer. RF model analysis identified landscape composition and hydrological connectivity as the primary factors governing TN and TP exports. Findings highlight the importance of managing hydrological connectivity and optimizing landscape composition as strategies for reducing nutrient losses and mitigating non-point source pollution in watershed management.