Multi-model integration for watershed nitrate source apportionment: Methods and applications

Overlapping source signatures and nonlinear hydrochemical relationships complicate nitrate source apportionment at the watershed scale. This letter presents an integrated framework linking hydrochemical source identification, quantitative apportionment, and data-driven prediction, with nitrate isotopes providing complementary constraints on sources and transformations. The framework’s core analytical components were demonstrated using Spearman correlation analysis, principal component analysis (PCA), positive matrix factorization (PMF), and XGBoost. Two principal components explained 75.9% of the total hydrochemical variance. PMF resolved mixed natural–anthropogenic, sewage-related, and mineral-dissolution factors, accounting for 54.4%, 23.9%, and 21.7% of the modeled hydrochemical contributions, respectively. XGBoost model achieved an overall classification accuracy of 0.833. K⁺, Ca²⁺, Cl⁻, and SO₄²⁻ collectively accounted for 59.7% of the feature importance, indicating the predictive relevance of major-ion signatures. Increasing the sample size could further improve predictive accuracy. These findings displayed the potential of integrating receptor modeling and machine learning for rapid source screening.