Optimizing radiation monitoring networks to improve emergency response strategies during nuclear power plant accidents.

This paper presents a new strategy to optimize radiation monitoring networks for effectively predicting contaminated areas and radiation levels during nuclear power plant accidents in order to improve emergency response efforts. Our strategy addresses variable metrological fields by generating ensemble simulations of wind fields and radionuclide migration in the atmosphere using the WSPEEDI (Worldwide version of System for Prediction of Environmental Emergency Dose Information) simulator. GPCAM (Gaussian Process for Continuous-time Acquisition of Measurements) is then used to capture the heterogeneity of radiation levels by sparse monitoring points, and to optimize their locations. We consider three different scenarios: (a) a single static spatial distribution of the radiation levels, (b) the temporal evolution of the distribution within a single release scenario for mobile sensor deployment, and (c) ensemble optimization with variable metrological conditions for assessing risks and emergency responses at a particular site a priori. The results are compared with the homogeneously-distributed network. Our results show that GPCAM is able to identify effective monitoring locations for each of these scenarios, except that a prevailing wind direction is required for the ensemble case. In addition, we compare the effect of different acquisition functions, kernel functions, and hyperparameters in GPCAM on the sensor locations.