Concentrations of naturally occurring fluoride in groundwater exceeding the WHO guideline of 1.5 mg/L have been detected in many parts of Pakistan. This may lead to dental or skeletal fluorosis and thereby poses a potential threat to public health. Utilizing a total of 5483 fluoride concentrations, comprising 2160 new measurements as well as those from other sources, we have applied machine learning techniques to predict the probability of fluoride in groundwater in Pakistan exceeding 1.5 mg/L at a 250 m spatial resolution. Climate, soil, lithology, topography, and land cover parameters were identified as effective predictors of high fluoride concentrations in groundwater. Excellent model performance was observed in a random forest model that achieved an Area Under the Curve (AUC) of 0.92 on test data that were not used in modeling. The highest probabilities of high fluoride concentrations in groundwater are predicted in the Thar Desert, Sargodha Division, and scattered along the Sulaiman Mountains. Applying the model predictions to the population density and accounting for groundwater usage in both rural and urban areas, we estimate that about 13 million people may be at risk of fluorosis due to consuming groundwater with fluoride concentrations >1.5 mg/L in Pakistan, which corresponds to ~6% of the total population. Both the fluoride prediction map and the health risk map can be used as important decision-making tools for authorities and water resource managers in the identification and mitigation of groundwater fluoride contamination.
Keywords: Aquifers; Drinking water quality; Fluorosis; Geogenic groundwater pollution; Human health threat; Random forest modeling.
Copyright © 2022 The Authors. Published by Elsevier B.V. All rights reserved.