Abstract
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 the environment and food around the Sanmen Nuclear Power Plant.
Introduction
Ensuring environmental safety around nuclear facilities is a critical public health priority, especially as nuclear power becomes a pillar for global carbon neutrality. While traditional monitoring is often reactive, this study, published in Frontiers in Public Health, proposes a proactive approach. Researchers developed and validated time-series models to predict radioactivity levels in food and the environment surrounding the Sanmen Nuclear Power Plant (SNPP)—the world’s first operational AP1000 unit. By moving from simple monitoring to predictive forecasting, the study aims to provide an "early warning" system for public health risk assessment.
Methods and Data
The study utilized a longitudinal dataset (2011–2022) collected by the Zhejiang and Taizhou Centers for Disease Control and Prevention. The researchers focused on several key variables:
- Strontium-90 ($^{90}Sr$): Measured in food items including rice, cabbage, crucian carp, and mullet.
- Cumulative Ambient Dose (CAD): Measured via thermoluminescent dosimeters (TLDs).
- Gross $\beta$ activity: Monitored in surface, tap, and well water.
The authors employed Autoregressive Integrated Moving Average (ARIMA) models. For $^{90}Sr$ in food, which exhibited periodic fluctuations, Seasonal ARIMA (SARIMA) was used to account for quarterly variations. Model selection was guided by the Bayesian Information Criterion (BIC), and accuracy was evaluated using Mean Absolute Percentage Error (MAPE).
Results and Discussion
The researchers found that radioactivity levels remained stable following the commencement of SNPP operations in 2018, indicating negligible environmental impact from the plant.
Predictive Performance
The study identified ARIMA(0, 0, 0) × (0, 1, 1)$_4$ as the optimal model for $^{90}Sr$ in food and ARIMA(4, 0, 1) for CAD. These models achieved high accuracy, with MAPEs of approximately 21.4% and 22.4%, respectively. Such results are considered strong for environmental field data.
Seasonality and Trends
A key finding was the clear seasonality of $^{90}Sr$ in food, likely linked to agricultural cycles and biological uptake patterns in species like crucian carp. In contrast, gross $\beta$ activity in water sources was identified as "white noise"—random fluctuations within a safe range that lack a predictable trend suitable for ARIMA modeling.
2024 Forecasts
The models predict that 2024 radioactivity levels will remain low (0.067–0.77 Bq/kg for $^{90}Sr$ and 0.067–0.129 mSv for CAD), well within Chinese national safety standards.
Conclusion
The application of ARIMA modeling provides a reliable tool for short-term environmental monitoring. By establishing a baseline of predicted values, any future observation significantly exceeding the forecast could serve as an immediate indicator of an anomalous release. While these models are limited by their short-term focus and dependency on stable conditions, they represent a significant step toward proactive environmental safety management.
Summary updated