Capturing spatial patterns of environmental recovery after the Fukushima Daiichi Nuclear Power Plant accident based on the radiation monitoring network

More than eight years has passed since the Fukushima Daiichi Nuclear Power Plant accident. Radiation measurements and monitoring have been performed continuously in the region to protect the public health, guide decontamination efforts, and plan the return of evacuated residents. Radiation measurements have been conducted using various techniques and platforms such as walk, car and airborne surveys as well as more than 3000 monitoring posts. Currently, the monitoring program is expected to transition from the emergency response to long-term monitoring, which is critical to confirm the system stability, to provide assurance for the public as well as to detect anomaly if it occurs. To build a sustainable and effective long-term monitoring network, it is important to understand and quantify how quickly the radiation level (i.e., radiation air dose equivalent rates) is decreasing, and how heterogeneous the reduction rate is depending on various factors such as land use types and decontamination. In this study, we use machine learning methods to characterize the spatiotemporal variability of the reduction in the radiation air dose rate within the 80-km radius from the nuclear power plant. We first quantify the annual reduction rate between 2011 and 2017, and analyze the spatiotemporal variability in the reduction rates at more than 1000 monitoring locations. Results show that the reduction is larger in more contaminate areas, while the annual reduction rate is decreasing over time since the radiation level is getting closer to the background and fast-decaying Cs-134 is rapidly decreasing. We also apply unsupervised clustering to identify the locations and regions that have similar dynamics. The identified spatial patterns are helpful to plan the radiation measurements, as well as to find any anomalies in the reduction.