Data-driven models for groundwater nitrate contamination prediction: A nation-wide approach for Mexico

Nitrate (NO₃⁻-N) stands as one of the prevalent chemical contaminants in groundwater, posing potential repercussions on both the environment and public health. However, the monitoring of this parameter on a national scale is notably limited, especially in developing regions. To address this gap, we applied distinct machine learning (ML) algorithms (Extreme Gradient Boosting, Boosted Regression Trees, Random Forest, and Support Vector Machines) capable of quantifying/predicting NO₃⁻-N concentrations in groundwater. These algorithms were validated through comprehensive application across Mexico. The models initially considered 68 covariates and identified significant predictors of NO₃⁻-N concentration spanning from climate, geomorphology, soil, hydrogeology, and human factors. We achieved an outstanding performance with about 10 times less availability of information compared to previous large-scale assessments, and thus efficiently countered the challenge of limited data availability/monitoring stations. Our success can be attributed mainly to the implementation of the 'Support Points-based Split Approach' during pre-processing, which effectively transformed the limited national groundwater quality database into spatial points suitable for appropriate train/test datasets. Areas exhibiting NO₃⁻-N concentrations exceeding the drinking water standard (>10 mg/L) were identified, notably in the north-central and northeast regions of the country, linked to agricultural and industrial activities. Individuals living in these regions face potential exposure to elevated NO₃⁻-N levels in groundwater. These NO₃⁻-N hotspots align with reported health implications such as gastric and colorectal cancer. This study not only showcases the potential of ML in data-scarce regions but also provides actionable insights for policy and management strategies.