Prediction of anthropogenic 129I in the South China Sea based on machine learning

Abstract

With the rapid increase in the number of nuclear power plants along the China coast and the potential for releases of radioactive substances to marine ecosystems, it is important to investigate and predict the dispersion of radionuclides in the seas and assess their radiological risks. Due to iodine's high solubility in water, and the high fission yield and long half life of ¹²⁹I, it has been widely used for investigation of anthropogenic radioactive pollution dispersion in the marine environment. This work established a method to predict the dispersion of anthropogenic ¹²⁹I in the seas by machine learning. Two models: 1) a Random Forest model, and 2) a Support Vector Machine model, which were developed using measured ¹²⁹I and ¹²⁷I values from seawater in the northwestern South China Sea. Spearman analysis was employed to investigate the influence of various environmental parameters on ¹²⁹I levels, with water depth, temperature, and salinity identified as the main parameters affecting ¹²⁹I levels. The sensitivity of machine learning model outputs to different environmental parameters was determined; with salinity being the most significant parameter. Both models demonstrated good prediction performance as seen in comparisons of predicted data with measurement values (R² > 0.83). Based on a comprehensive evaluation of model metrics, the Random Forest model slightly outperformed the Support Vector Machine model. The model can be easily applied to predict the dispersion of soluble anthropogenic radionuclide in marginal seas, providing an effectively technical support for radiological risk assessment and emergency responses of nuclear pollution and accidents.