Characterizing Isotropic Source Component of DPRK Nuclear Tests by Affine-invariant Bayesian Samplers with Uncertainty Estimate for Data Noise and Theory Error

A seismic moment tensor (MT, a 3x3 matrix) is a general source representation of various seismic events under the point source assumption, which is generally valid for small-to-medium size earthquakes. A full MT can be decomposed into isotropic (ISO), compensated linear vector dipole (CLVD), and double-couple (DC) components. The ISO represents the explosion/collapse source process that involves volumetric changes. Therefore, the relative significance of the ISO component, which can be learned from inverting seismic waveforms, is an essential indicator to discriminate between earthquakes and explosive events. However, an intrinsic ISO-CLVD tradeoff impedes resolving shallow explosive sources due to the high similarity of long-period waveforms at regional distances. Even though this tradeoff can be mitigated by extra constraints such as teleseismic P-waves, there is still an urgent need for advanced inversion algorithms to explore the solution space thoroughly. Apart from that, a rigorous uncertainty estimate is required to constrain the source better. Firstly, the inversion should consider the data noise. Secondly, the theory error primarily due to imperfect knowledge of Earth's structure is also significant but proven difficult to treat. Here, we propose a new Bayesian MT inversion scheme with affine-invariant ensemble samplers to explore the MT parameter space accounting for data and theory errors. Carefully designed synthetic experiments indicate the advantage of the newly developed method in resolving the isotropic components of a shallow seismic source. Our application to DPRK tests reveals a similar source mechanism dominated by a high ISO and significant CLVD components, including a small DC component. This study aims to characterize shallow explosive sources' physics better, thus helping verify compliance with the CTBT.