Application of Machine Learning Prediction Model in Spatial Distribution of High Nitrate and Fluoride in the Groundwater of a Typical Rift Zone (Republic of Djibouti)

Groundwater in the East African Rift System (EARS) is often characterized by high levels of dissolved fluoride and nitrate, which pose inherent risks to public health. Furthermore, there is limited data available for EARS, as is the case for most developing countries. Within the EARS, the aquifer systems of the Republic of Djibouti were over-exploited and subjected to anthropogenic and/or geogenic pollution with high NO3− (up to 256 mg L−1) and F− (up to 9.20 mg L−1) concentration. This study uses geochemical and thermodynamic tools, along with stable isotope ratios, such as δ2H(HO2), δ18O(HO2), δ15N(NO3−), and δ18O(NO3−), to decipher the enrichment mechanisms of F− and NO3−. Geochemical data indicate that the high F− concentrations in groundwater from the study area were primarily controlled by dissolution of fluorite, cation exchange, and desorption. Furthermore, stable nitrate isotopes were used to identify anthropogenic and geogenic sources of NO3− in the groundwater. This study also applies the CatBoost machine learning algorithm to generate 30 m resolution for spatial predictions of NO₃− and F− using satellite-derived environmental predictors. Despite limited sampling and environmental heterogeneity, the models effectively captured spatial patterns, consistent with both anthropogenic and geogenic sources. The resulting maps offer a practical decision-support tool for groundwater monitoring and targeted intervention, particularly in communities relying on untreated water supplies. These findings underscore the value of machine learning for environmental prediction in complex hydrogeological settings and highlight the need for enhanced sampling, hydrogeochemical integration, and improved uncertainty quantification to support informed groundwater management.