Mapping radionuclide concentrations in the UAE using a Gaussian process Machine learning approach

This study introduces a machine learning approach for mapping soil radionuclide concentrations in the UAE using Gaussian Process (GP) regression. Aimed at enhancing environmental monitoring and public health, the approach utilizes soil samples from across the UAE to create a radiation map (covering Ra-226, Th-232, and K-40 isotopes). GP regression, known for its proficiency in spatial data interpolation, predicts radionuclide levels in areas without direct testing. This method is adept at managing the non-linear spatial complexities inherent in geographic data, offering both a qualitative and quantitative understanding beneficial for decision-making and further sampling strategies. The results reveal the GP model's capacity to accurately reflect geographic variances in isotopic concentrations, with RMSE values of 13%, 14%, and 22% for Ra-226, Th-232, and K-40, respectively. The model's success in learning the geographical variations of the concentrations showcases its potential to guide future research by identifying areas of increased radioactivity risk.