Machine learning reveals eutrophication mechanisms and governing thresholds in an engineered throughflow lake

Engineered throughflow lakes serve as critical nodes in inter-basin water-transfer megaprojects; however, accelerating eutrophication increasingly threatens water-conveyance security. Persistent uncertainties in eutrophication mechanisms under coupled anthropogenic pressures and hydrological variability further compromise downstream water quality and the reliability of water supply. Luoma Lake exemplifies this challenge as the terminal receiving lake of the Eastern Route of the South-to-North Water Diversion Project (SNWDP) in Jiangsu Province. By integrating machine learning methods, including self-organizing maps (SOM), generalized additive models (GAM), and change-point analysis, this study elucidates the mechanisms driving eutrophication and derives spatially adaptive control thresholds from a 2-year observational dataset. The lake exhibited moderate to mild eutrophication with pronounced spatial autocorrelation, with SOM delineating four hydro-biogeochemical zones. Nonlinear interactions among exogenous nutrient loading, sediment phosphorus feedback, algal metabolic cascades, and flow-mediated perturbations drove eutrophication dynamics. Critical whole-lake thresholds, defined at a 90% probability of exceeding the trophic level index (TLI), were derived from a two-year post-SNWDP dataset under the prevailing hydrological regime. Spatially explicit thresholds expose divergent regulatory requirements across hydro-biogeochemical zones. External loading zones (Clusters 1 and 4) require prioritized control of permanganate index (COD Mn ) and chlorophyll- a (Chl a ), whereas the hydraulic stagnation zone (Cluster 3) necessitates stringent regulation of total phosphorus (TP) to suppress internal nutrient recycling. In contrast, the hydrodynamically disturbed zone (Cluster 2) requires targeted hydrodynamic monitoring to pre-empt bloom development. These findings provide mechanistic insights and spatially explicit, quantitative-derived management thresholds for engineered throughflow lakes. • Hydrodynamic and pollutant transport complex drive lake eutrophication. • SOM identifies coherent water-quality zones along inflow–outflow corridors. • GAM quantifies nonlinear thresholds linking key drivers to TLI responses. • Spatially explicit thresholds enable targeted control beyond lake-wide averages.