EKI-based inverse modeling for multi-radionuclide source term estimation from time-series gamma dose rates during radiological emergencies

Abstract

We developed an ensemble-based method for estimating multi-radionuclide source terms during nuclear accidents using gamma dose rate observations. By integrating a GPU-accelerated Gaussian puff model with advanced Ensemble Kalman Inversion (EKI) – including Standard, Localized, and Adaptive EKI – estimation accuracy and computational efficiency are enhanced. The method was tested using synthetic data simulating the release of radionuclides Kr-88, I-131, Te-132, Sr-91, and Sb-129 over a 12-hour period within a 30-kilometer radius. Gamma dose rates were measured at multiple locations, with relative errors introduced to simulate measurement uncertainties. The approach successfully estimated time-varying release rates of multiple radionuclides without prior knowledge of their ratios, especially for Kr-88, even under high-error conditions. The use of Localized and Adaptive EKI allowed for a significant reduction in the required ensemble size and iterations, optimizing computational load without compromising accuracy. This method enables dynamic estimation of transient multi-radionuclide source terms using time-series data, even in nonlinear systems, supporting improved emergency response strategies and environmental monitoring during nuclear incidents. The findings demonstrate that accurate inverse estimation is feasible based solely on gamma dose rate measurements, contributing to better preparedness and mitigation efforts in nuclear emergencies.