Predicting and assessing the distribution of high-fluoride groundwater in China using multiple machine learning models: A big data-driven geospatial analysis.

Drinking-water fluorosis remains a major public health concern in China due to sustained reliance on high-fluoride groundwater in rural regions. Using the largest harmonized national dataset to date-138,180 site records and 38 environmental predictors spanning climatic, topographic, soil, hydrological, geological, and volcanic domains-artificial neural network (ANN), random forest (RF), and logistic regression (LR) models were developed and evaluated via spatial block 5-fold cross-validation to mitigate spatial autocorrelation and improve geographic generalization. All models demonstrated strong spatial generalizability, with precision-recall area under the curve (PR_AUC) values of 0.985 (ANN), 0.992 (RF), and 0.956 (LR). The resulting probabilistic risk maps revealed both well-established and previously under-recognized high-fluoride zones, including the Northeast Plain, North China Plain, Inner Mongolia, and parts of northwest China, as well as newly highlighted candidate areas such as the Jianghan Plain, Sanjiang Plain, and Junggar and Tarim basin rims. An estimated 40.78-54.77 million rural residents reside in predicted high-risk zones. Meteorological conditions, topographic-soil properties, and volcanic-geologic factors emerged as dominant controls on fluoride enrichment. This three-model comparative framework provides a validated, scalable tool for national groundwater fluorosis risk assessment and offers actionable insights to support targeted surveillance, mitigation strategies, and groundwater safety management, offering a framework applicable to environmentally similar regions such as Mongolia and Kazakhstan.