Source-sink relationships of phosphorus between surface and groundwater in Honghu Lake area, China.

Managing phosphorus (P) pollution in hydrologically complex lake systems requires clarifying whether groundwater acts as a source or sink of P to surface water, a question that remains poorly constrained due to the difficulty of simultaneously tracing P origins and disentangling transformation processes. This study investigates the source-sink dynamics of P across the surface water-groundwater (SW-GW) interface in Honghu Lake, China, using an integrated framework where three methods play distinct, complementary roles: (1) phosphate oxygen isotopes (δ18O-PO43-) are primarily responsible for identifying P sources (anthropogenic vs natural); (2) stable isotopes (δD-δ18O) and hierarchical cluster analysis are used to interpret hydrological mixing zones and groundwater discharge pathways; and (3) Random Forest (RF) modeling identifies key hydrochemical drivers controlling P retention and mobilization, independent of source apportionment. Analysis of 21 water samples reveals a bidirectional source-sink relationship. GW functions as a source of P to the lake in specific hotspots: δ18O-PO43- signatures (7.14‰ to 12.07‰) confirm anthropogenic origins (sewage and agricultural waste) in GW, while stable isotopes (δD-δ18O) and hierarchical cluster analysis delineate active mixing zones where this contaminated GW discharges to SW. Simultaneously, GW acts as a transient sink: SW total phosphorus (TP) averages 0.09 mg/L consistently above eutrophication thresholds, while GW TP exhibits high spatial variability (0.05-0.19 mg/L), indicating localized retention. To interpret migration and transformation processes, an RF model was evaluated using leave-one-out cross-validation (LOOCV) given the modest sample size (n = 21); the model achieved a mean cross-validated R2 of 0.951 on held-out test folds, suggesting good predictive capability under the conditions tested. The model identified total dissolved nitrogen (TDN) and total nitrogen (TN) as the primary drivers of P distribution, revealing coupled nutrient dynamics where nitrogen loading appears to govern P mobility. Multivariate analyses further identify redox potential (Eh), dissolved oxygen (DO), and bicarbonate (HCO3-) as key hydrochemical controls on P retention or release. The study concludes that P dynamics in Honghu Lake operate as a three-stage process: anthropogenic loading establishes the P inventory (δ18O-PO43- identified sources), redox-sensitive transformations determine temporary storage or mobilization (RF-identified drivers), and SW-GW exchange controls final delivery to the lake (δD-δ18O/cluster-defined pathways). This framework, while hypothesis-generating and requiring further validation with flux-based measurements, offers a transferable approach for integrated nutrient management through hotspot intervention, redox-condition management, and isotope-informed, machine learning-enhanced monitoring in interconnected lake-aquifer systems worldwide.