Estimation of Unknown Source of Atmospheric Release

Our research is motivated by applications in radiation safety, particularly the estimation of unknown radionuclide emissions based on available measurements and the subsequent analysis of their dispersion using the inferred emissions. Specific examples include the estimation of releases during the Chernobyl accident, analyzing the iodine leak in Hungary in 2011, or the retrospective modeling of ruthenium releases from Siberia in 2017. Measurements and modeling of these events are often complicated by significant uncertainties and limitations in the available data, which are frequently incomplete and imprecise. We address these challenges by integrating all available information and uncertainties into inverse modeling frameworks based on Bayesian methods. This approach not only provides emission estimates but also quantifies their uncertainties, enabling an assessment of the reliability of the estimates derived from the available data.

An additional research direction focuses on complex spatiotemporal emissions, where releases may originate from any point within the computational domain. Examples include emissions of microplastics, ammonia, and wildfires around Chernobyl, which combine spatiotemporal sources with height-dependent releases and size-dependent particle distributions. These scenarios introduce new complexities in modeling and estimation, which can be addressed using Bayesian approaches or deep neural networks. Our results contribute to more accurate assessments of the environmental impacts of these emissions.