Bayesian reconstruction of atmospheric radionuclide releases from sparse observations using nonuniform continuous...

Renewed interest in nuclear energy generation globally makes Bayesian reconstruction increasingly important for interpreting atmospheric radionuclide releases. However, observations of such releases are usually sparse, leading to large reconstruction uncertainty when using common uniform priors. This study developed a novel Bayesian framework with nonuniform continuous priors for both spatial location and temporal release rates, enabling reliable reconstruction from sparse observations. The spatial prior assigns binary probabilities based on the Spearman's correlation between source-receptor sensitivities and observations to exclude false source locations, and interpolates a continuous probability distribution to ensure smooth convergence of the Bayesian reconstruction. The temporal prior constrains the release rate distribution, enabling time-varying release reconstruction. The framework, solved using the Maximum-A-Posteriori-Bayesian approach, was evaluated against six events featuring sparse observations, including two cases with monitoring stations surrounding the source and four cases with monitoring stations distributed on one side. Results revealed a much sharper a posteriori spatial distribution and up to 97% lower localization errors than achieved through uniform-prior Bayesian reconstruction. The reconstructed release profiles matched reported estimates and showed notably reduced uncertainty. Sensitivity analyses demonstrated robustness to meteorological inputs and temporal priors, an optimal temporal reconstruction resolution, and observation noises. The proposed method outperformed nonuniform-discrete-prior methods by enhancing convergence stability and reducing localization errors by an average of 66.0%, revealing the importance of nonuniform continuous priors in Bayesian source reconstruction with sparse observations.