Nitrate source apportionment in a semiarid aquifer of Mexico using Bayesian dual-isotope mixing models (δ15N, δ18O, δ11B)

Across the semiarid landscapes of northern Mexico, groundwater systems that sustain both regional food production and population centers are approaching critical thresholds owing to sustained overextraction and widespread nitrate contamination. The Meoqui–Delicias aquifer (MDA) exemplifies this challenge, supplying water to approximately 265,000 people, one of the country’s largest irrigation districts, and more than 90,000 head of cattle while experiencing a severe groundwater deficit of ~165 Mm³·yr^-1. Moreover, 82% of the sampled wells exceeded the natural nitrate background levels (>3 mg L^-1 as N), raising concerns regarding drinking-water security and ecosystem health.

Identifying nitrate sources in such systems is inherently challenging because of the overlapping isotopic signatures of manure, sewage, synthetic fertilizers, and soil nitrogen, further complicated by active biogeochemical transformations. To address this complexity, groundwater samples were first classified using hydrochemical clustering based on self-organizing maps, which revealed two statistically coherent groups. Bayesian mixing models were then constrained using robust isotope end members from the literature. Each model was run with 300,000 MCMC iterations (200,000 burn-in, thinning interval of 100, three parallel chains), with convergence verified using Gelman–Rubin, Heidelberg–Welch, and Geweke diagnostics. A systematic sensitivity analysis (±10–20% perturbations of source signatures) demonstrated the greater robustness and stability of the δ¹⁵N vs δ¹¹B model compared to the conventional δ¹⁵N vs δ¹⁸O pairing model.

The results revealed a marked contrast between the isotopic approaches. The traditional δ¹⁵N vs δ¹⁸O model aggregates manure and sewage as the dominant combined source (65 ± 20%, ~4.5 mg L^-1 N), with secondary contributions from soil nitrogen (22 ± 19%) and fertilizers (13 ± 14%). In contrast, the incorporation of boron isotopes effectively resolved source overlap, identifying livestock manure as the primary contributor (52 ± 12%, ~3.5 mg L^-1 N), followed by soil nitrogen (37 ± 14%), with minor inputs from fertilizers (6 ± 8%) and negligible sewage contributions (5 ± 7%). Isotopic evidence further indicated that nitrification dominated nitrogen cycling in approximately 60% of the samples, whereas denitrification was restricted to riparian zones. Stable water isotopes confirm that meteoric recharge is modified by evaporation during irrigation return flows.

Overall, this study demonstrates that boron-enhanced Bayesian isotope mixing models provide a robust and transferable framework for nitrate source apportionment in complex semiarid aquifers, delivering quantitative discrimination where conventional isotope approaches remain ambiguous and offer direct relevance for targeted groundwater management and nutrient-mitigation strategies.