RESUMEN
Objectives: Model prediction of radioactivity levels around nuclear facilities is a useful tool for assessing human health risks and environmental impacts. We aim to develop a model for forecasting radioactivity levels in the environment and food around the world's first AP 1000 nuclear power unit. Methods: In this work, we report a pilot study using time-series radioactivity monitoring data to establish Autoregressive Integrated Moving Average (ARIMA) models for predicting radioactivity levels. The models were screened by Bayesian Information Criterion (BIC), and the model accuracy was evaluated by mean absolute percentage error (MAPE). Results: The optimal models, ARIMA (0, 0, 0) × (0, 1, 1)4, and ARIMA (4, 0, 1) were used to predict activity concentrations of 90Sr in food and cumulative ambient dose (CAD), respectively. From the first quarter (Q1) to the fourth quarter (Q4) of 2023, the predicted values of 90Sr in food and CAD were 0.067-0.77 Bq/kg, and 0.055-0.133 mSv, respectively. The model prediction results were in good agreement with the observation values, with MAPEs of 21.4 and 22.4%, respectively. From Q1 to Q4 of 2024, the predicted values of 90Sr in food and CAD were 0.067-0.77 Bq/kg and 0.067-0.129 mSv, respectively, which were comparable to values reported elsewhere. Conclusion: The ARIMA models developed in this study showed good short-term predictability, and can be used for dynamic analysis and prediction of radioactivity levels in environment and food around Sanmen Nuclear Power Plant.