Modeling gamma radiation intensity in monazite rich coastal environments using neural network

Meaipe beach, in Espirito Santo, Brazil, is a high background radiation area due to monazite-rich sands containing thorium, whose decay releases thoron (²²⁰Rn). This coastal region has been linked to both physiological benefits and adverse effects, but the lack of conclusive evidence underscores its relevance as a site for environmental radioactivity research. This study analyzes gamma radiation intensity and its relationship with meteorological and soil variables, with continuous field measurements revealing temporal and vertical variations related to thoron exhalation. A multilayer perceptron model was developed to predict hourly gamma radiation using field data from 2018 to 2022. Initially trained with normalized data and adapted to the original scale for physical interpretability, the model with two hidden layers of 15 neurons each and ReLU activation, optimized by the Adadelta algorithm, achieved excellent performance (R² = 0.99). Height was the most influential variable (41%) for prediction, while surface longwave radiation and day of the year contributed 12% and 9%. These results indicate the relevance of surface radiation processes and show that day of year acts as a proxy for temporal variability. The model enabled the evaluation of multiple hypothetical scenarios, demonstrating the sensitivity of gamma radiation to meteorological and soil variations. A simultaneous increase of 10 °C in air and soil temperatures was associated with an increase in intensity of up to 18%. Gamma radiation was slightly modulated by high humidity and reduced solar radiation, with minor nocturnal changes in specific humidity affecting its diurnal pattern. The model successfully captured the observed pattern, reinforcing its predictive capability for assessing radiation in tropical environments.