Robust source reconstruction of atmospheric radionuclides from observations of different sparsity with spatial preselection and non-smooth constraints

Abstract

As the global-nuclear-capacity-tripling plan is implemented, reconstruction of the source locations and release rates of atmospheric radionuclides becomes increasingly important for the environment and human health. However, such reconstruction is vulnerable to unrealistic solutions because it is ill-posed. This study proposed a spatiotemporally constrained reconstruction method that excludes false estimates and achieves high accuracy. It uses the Spearman’s correlation coefficient to constrain the spatial search range and applies the L2 cost-function within this range to retrieve the source location. Using this location, time-varying release rates are estimated with non-smooth constraints, which simultaneously reconstructs the peak releases and removes unrealistic oscillations. Validation against both field experiments and real-world events demonstrated that this method effectively excludes false source locations. The estimated location is up to 96.27%, 98.31%, and 96.48% closer to the reported sources than those of the L2 cost-function, Pearson-correlation-constrained L2 cost-function, and Bayesian methods, respectively. The estimated release rates matched the reported time windows and total amounts, avoiding the unrealistic oscillations in other estimation methods. The proposed method exhibited superior performance and speed over the L2 cost-function method under different station layouts and numbers. Furthermore, it could improve other methods using different cost functions, indicating its potential for various applications.

Environmental Implication

Source reconstruction of atmospheric radionuclide releases is becoming increasingly important for the environment and human health because of the persistent leakage risk and increased demands following the global-nuclear-capacity-tripling announcement. We propose a spatiotemporally constrained reconstruction method that adaptively excludes false source locations and unrealistic temporal variations, both of which are long-standing problems in reconstruction of real events, and achieves robust performance under scenarios with different observation sparsity and station layouts. Additionally, the proposed approach can improve existing methods, especially with limited observations, thereby representing a versatile tool both for global radioactivity surveillance and environmental process investigation based on radioactivity tracers.