National-scale prediction of arsenic, fluoride and its co-occurrence in groundwater of Mexico: Implications for drinking-water.
Source: PubMed, NCBI / U.S. National Library of Medicine
Arsenic (As) and fluoride (F) in groundwater pose significant health risks and frequently co-occur in arid and semi-arid settings. We assembled a dataset comprising 3403 As measurements and 2084 Fmeasurements for Mexico (44% and 24% above the WHO guideline values of 10 µg/L and 1.5 mg/L, respectively) and trained Random Forest (RF), Extreme Gradient Boosting (XGBoost), and Artificial Neural Networks (ANN) classifiers using 69 predictor variables, including climate, hydrogeology, and anthropogenic pressures. Beyond separate single-contaminant models, we developed a co-occurrence classifier that explicitly predicts the joint exceedance of As and Fabove WHO drinking-water limits, capturing areas where concurrent exposure to both hazards is most likely. XGBoost performed best (test balanced accuracy/AUC: As 0.82/0.90, F0.83/0.94, co-occurrence 0.83/0.92) and generated 1-km probability of exceedance maps for each contaminant and their joint exceedance, providing a national-scale co-occurrence model for Mexico. Model interpretation showed that exceedance probabilities for As and Fare primarily controlled by elevation, climate (precipitation and temperature), and soil pH, with additional contributions from water‑table depth, population density, and volcanic-alluvial settings for co‑occurrence, highlighting coupled geogenic and anthropogenic controls. Risk mapping indicates at-risk areas covering 15.1% (As), 13.5% (F), and 8.9% (co-occurrence)
Abstract
Arsenic (As) and fluoride (F) in groundwater pose significant health risks and frequently co-occur in arid and semi-arid settings. We assembled a dataset comprising 3403 As measurements and 2084 Fmeasurements for Mexico (44% and 24% above the WHO guideline values of 10 µg/L and 1.5 mg/L, respectively) and trained Random Forest (RF), Extreme Gradient Boosting (XGBoost), and Artificial Neural Networks (ANN) classifiers using 69 predictor variables, including climate, hydrogeology, and anthropogenic pressures. Beyond separate single-contaminant models, we developed a co-occurrence classifier that explicitly predicts the joint exceedance of As and Fabove WHO drinking-water limits, capturing areas where concurrent exposure to both hazards is most likely. XGBoost performed best (test balanced accuracy/AUC: As 0.82/0.90, F0.83/0.94, co-occurrence 0.83/0.92) and generated 1-km probability of exceedance maps for each contaminant and their joint exceedance, providing a national-scale co-occurrence model for Mexico. Model interpretation showed that exceedance probabilities for As and Fare primarily controlled by elevation, climate (precipitation and temperature), and soil pH, with additional contributions from water‑table depth, population density, and volcanic-alluvial settings for co‑occurrence, highlighting coupled geogenic and anthropogenic controls. Risk mapping indicates at-risk areas covering 15.1% (As), 13.5% (F), and 8.9% (co-occurrence) of the national territory, with hotspots concentrated in north-central Mexico. Population overlays suggest ∼10.4 million people may be exposed to As (8.2%), ∼9.7 million to F(7.7%), and ∼7.2 million to concurrent exposure (5.7%). These quantitative products provide an operational baseline to prioritize monitoring wells, well management and treatment (well switching and point-of-use technologies), and risk communication in high-burden communities. SYNOPSIS: First national co-occurrence model identifies Mexico's As-Fgroundwater risk hotspots, enabling targeted monitoring, well management, and treatment to reduce population exposure.