An isotope-spectral fingerprint coupling approach for dynamic tracer optimization in DOM source apportionment within highly anthropogenically disturbed watersheds

Accurate source tracking of dissolved organic matter (DOM) in anthropogenically disturbed watersheds is crucial for pollution control. This study took the Qujiang watershed, which experiences high-intensity human activities, as an example, systematically collecting pollution source samples including municipal wastewater, rural wastewater, agricultural wastewater, livestock effluent, aquaculture effluent, soil, road runoff, algae, and riparian plants. δ¹³C isotopes and fluorescence indices were used as tracers in the end-member mixing analysis (EMMA) model, and a tracer screening system was constructed based on the LOOic-UI90 index. The results show that a combination of 2-3 low-correlation tracers achieves the optimal balance between model accuracy (LOOic) and stability (UI₉₀), while excessive tracers lead to model performance degradation due to collinearity issues. Comparative analysis revealed that the isotope-spectral index combined model significantly outperformed the single spectral index model and the single δ¹³C model in terms of model performance and alignment with actual land use patterns. Optimized model analysis indicated significant spatiotemporal variations in DOM sources across the watershed: anthropogenic contributions (dry season: 64.41 ± 7.00 %, wet season: 70.21 ± 13.58 %) were significantly higher than natural and internal sources. Specifically, aquaculture wastewater (25.31 %) and rural domestic sewage (14.23 %) dominated in the dry season, while urban sewage (35.75 %) and agricultural wastewater (11.21 %) were the primary sources in the wet season. Notably, in dry-season observations, the DOC concentration in the urbanized downstream section of the Changshan River (2.96 mg/L) was significantly higher than that in the upstream forested area (1.48 mg/L). EMMA model analysis confirmed that this difference was mainly attributed to point-source pollution inputs from human activities (municipal sewage: 17.4 %, agricultural runoff: 16.1 %). This study establishes a dynamic DOM source apportionment framework for precise pollution management in heavily disturbed watersheds.