Tracking the sources and fate of nitrate pollution by combining hydrochemical and isotopic data with a statistical approach

This study contributes to identifying and spatializing the different types of nitrate sources by combining hydrogeochemical and isotopic data with principal component analysis (PCA) and t-distributed stochastic neighbor embedding (t-SNE) multicriteria statistical methods. The methodology is applied to the strategic Mons Basin chalk aquifer (Belgium). The results are based on a whole dataset containing 72 water samples with analyses of the hydrogeochemical parameters (temperature, pH, electrical conductivity (EC), redox potential, dissolved O 2 ), alkalinity, total organic carbon (TOC), silica (SiO 2 ), major and minor ions (NO 3 – , NH 4+ , Ca 2+ , dissolved Fe and Mn, K + , Mg 2+ , Na + , Sr 2+ , Cl – , F – , SO 4 – , B) and multiple stable isotope ratios (δ 11 B, δ 15 N–NO 3 – , δ 18 O–NO 3 – ). Compared to classical PCA, the recently developed t-SNE method, which considers nonlinear relationships between variables and preserves local-scale similarities in a low-dimensional space, showed much better performance in discriminating different groups of samples and related zones in the aquifer. t-SNE results combined with isotope ratios highlighted four zones in the aquifer (grouped as A–D) and the presence of denitrification fronts. Group A presents a manure signature (δ 15 N–NO 3 – – mean ( μ ) +12.78‰, standard deviation ( σ ) 6.48‰; δ 11 B – μ 29.96‰, σ 6.91‰). Group B exhibits both manure and inorganic fertilizer signatures (δ 15 N–NO 3 – – μ 6.27‰, σ 2.55‰; δ 11 B – μ 15.86‰, σ 9.69‰). Group C shows a contamination by sewage (δ 15 N–NO 3 – – μ 12.67‰, σ 5.60‰; δ 11 B – μ 9.97‰, σ 7.08‰). Group D presents a mixed signature (δ 15 N–NO 3 – – μ 9.25‰, σ 2.94‰; δ 11 B – μ 20.00‰, σ 6.70‰).