A Bayesian Maximum Entropy Fusion model for enhanced prediction and risk assessment of fluoride and arsenic contamination in groundwater.

In the central and western regions of Jilin Province, excessive groundwater extraction has resulted in elevated levels of fluoride (F) and arsenic (As) in drinking water. Prolonged exposure to these contaminants is linked to endemic health issues, including dental and skeletal fluorosis as well as chronic arsenic poisoning, posing significant risks to both environmental integrity and public health. Since traditional models lack prediction accuracy, this study aimed to construct a Bayesian Maximum Entropy Fusion (BMEF) integrating hard and soft data for improved prediction of F and As concentrations and probabilistic estimations of contamination risk. This was accomplished by combining the spatial heterogeneity modeling of Geographically Weighted Regression (GWR), the probabilistic forecasting strength of Gaussian Process Regression (GPR), and the uncertainty quantification framework of Bayesian Maximum Entropy (BME). The BMEF model provided predictions of F with improvements in accuracy (R) over GWR, GPR, and BME of 60.8 %, 112 %, and 26.9 %, respectively; improvements in As prediction were 456 %, 84.4 %, and 58.5 %, respectively. The model was able to identify high-risk F zones spanning 3208 km, 3650 km, and 5240 km at 95 %, 90 %, and 75 % confidence levels, respectively; those for As were 3884 km, 4573 km, and 6490 km. The model effectively addressed the limitations of individual models in prediction accuracy and uncertainty quantification by integrating water quality sampling data with regression-based predictions. However, it faced challenges such as high computational complexity, sensitivity to data quality, and a lack of consideration for temporal variability.