Spatiotemporal Dynamics and Driving Mechanisms of Water Ecosystem Service Flows in the Yangtze River Basin Based on SWAT and Machine Learning

Water ecosystem service flows (WESFs) help address spatial mismatches in water resources and support basin resilience. However, their dynamic evolution and nonlinear drivers under climate change and intensive human activities remain poorly understood. This study evaluates the spatiotemporal dynamics of WESFs in the Yangtze River Basin (YRB) from 2005 to 2022 by integrating dynamic flow analysis with mechanism interpretation. We developed an integrated framework coupling SWAT hydrological simulations with a proxy-based spatial allocation approach for social water demand. Using the Water Stress Index (WSI) and river topology, dynamic inter-regional WESFs were simulated. Furthermore, an interpretable machine learning approach was employed to identify the nonlinear effects of multiple driving factors. Results reveal a persistent supply–demand mismatch: supply exhibited a northwest–southeast gradient (averaging 567.21 mm annually), while demand concentrated in mid-lower plains and urban corridors. The flow network, which accounts for accumulated upstream inflow, demonstrated a stable “upstream supply, mid-reach transmission, and downstream benefit” pattern, highlighting downstream reliance on upstream inputs. Driving analysis identified land surface and vegetation as the largest associated driver category, while climate–hydrology and human activity were not cleanly separable. Climate provided the hydro-climatic conditions for redistribution. Nonlinear responses and blue–green interactions were also identified, informing transboundary ecological compensation and regional water-resilience management.