Hydrological-state reorganization of eutrophication risk in temperate monsoon coastal rivers revealed by interpretable machine learning and structural path analysis

Hydrological-state variability strongly shapes eutrophication risk in small coastal catchments, but the transition from low-flow source accumulation to rainfall-driven transport remains poorly quantified. We analysed monthly eutrophication index (EI) patterns for 30 coastal rivers on the Shandong Peninsula, China, from 2020 to 2024 using an integrated interpretable machine-learning and observed-variable structural path-analysis framework. The EI, derived from TN, TP, DO, and COD Mn , was evaluated against terrain, hydroclimate, hydrology, human activity, point-source pollution, and anthropogenic-pressure proxies derived from annual wastewater and fertilizer inventories. EI was significantly higher under high-flow conditions than under low- and transition-flow conditions, accompanied by marked changes in the relative contributions of the individual EI components. Predictor associations also varied across hydrological states. Low-flow conditions were more closely associated with river discharge (Q), human pressure, and downstream catchment setting, whereas transition-flow conditions showed a stronger terrain-related association structure. Under high-flow conditions, human-activity, terrain, precipitation, and Q signals were more broadly represented. Structural path analysis further showed that precipitation was positively associated with Q and the TP- and COD Mn -related components under high-flow conditions, whereas temperature- and HFI-related associations were more evident under transition- and low-flow conditions. Overall, the results reveal a hydrological-state-dependent reorganization of eutrophication risk and its environmental associations, providing a basis for flow-state-specific water-quality monitoring.