A Novel Framework for Spatiotemporally Decoupled Source Reconstruction of Radionuclides Released Into the Atmosphere

Abstract In recent years, Europe has been confronted with a series of radionuclide leak incidents, emanating from sources that remain unidentified. Such incidents are typically assumed to originate from a single, static point source, allowing the source location to be pinpointed to a specific, time-unvarying coordinate. However, the unavailability of detailed information about these leaks, including crucial data like the operating status of involved nuclear plant, renders direct estimation of release details using a top-down approach unfeasible. Consequently, a bottom-up approach, relying on environmental measurements and atmospheric dispersion modeling, becomes the only feasible choice. This approach focuses on the spatiotemporal source reconstruction of atmospheric radionuclide release, aiming to optimize the source location and release rate by minimizing the discrepancy between simulations and measurements. Owing to the interdependence of these spatiotemporal source parameters, traditional source reconstruction methods often attempt to resolve both temporal and spatial aspects simultaneously. The resulting tremendous solution space may cause substantial estimation errors in both the source location and release rate. To tackle this issue, this paper proposes a novel framework for spatiotemporally decoupled source reconstruction. This framework splits the reconstruction into the source localization and release rate estimation stage, reducing the solution space at each stage. Initially, an artificial neural network model is employed to map a relationship between the source location and several simulation-measurement error metrics. The error metrics for each source location are obtained based on the release rates calculating by the Tikhonov regularization method. Subsequently, a likelihood function is constructed based on these error metrics, followed by assessing the posterior distribution of the source location using the Bayesian statistics. Finally, based on the estimated source location, the release rate is calculated using the Tikhonov regularization method. The proposed framework has been first validated using a European Tracer Experiment (ETEX) whose source is previously known. The source location error was less than 100 km and the estimated release rate profile match the truth well. Furthermore, this framework was applied to reconstruct the source of the 2017 106Ru leak incident, a case where the source remains unknown. The estimated source location and release amount are consistent with public literature. The successful application of this framework to both controlled experiment and real-world radionuclide leak incident underscores its potential as a key component of the nuclear emergency response system.