SUMMARY Seismic source classification often relies on a single best-fitting moment tensor and deterministic rules on source-type diagrams. This study formulates classification directly in terms of the posterior distribution of the moment tensor on the fundamental lune. Moment tensors from waveform inversion are mapped into a uniform $(v,w)$ source-type domain and used to build empirical source-type probability density functions $p(v,w)$. We evaluate five hypotheses ($\pm \mathrm{ISO}_{\mathrm{CAL}}$, DC, $\pm$CLVD) plus a residual baseline, where $\mathrm{ISO}_{\mathrm{CAL}}$ denotes a calibrated isotropic (ISO) kernel relative to the canonical ISO endmember. Classification probabilities $P(H_i)$ are obtained by integrating $p(v,w)$ against the hypothesis kernels; a residual probability $P_{\mathrm{residual}}$ is included as a sixth baseline (uniform) model that absorbs probability when the posterior is diffuse or poorly matched by the prototypes. The framework is applied to published regional moment tensor ensembles for 32 events, including 23 nuclear tests at the Nevada National Security Site and the Punggye-Ri, North Korea, test site, plus nearby earthquakes and collapses. Well-recorded underground nuclear tests yield tightly localized +$\mathrm{ISO}_{\mathrm{CAL}}$ probabilities with small $P_{\mathrm{residual}}$, while events recorded by sparse, complex source-receiver paths show broader probability vectors and larger $P_{\mathrm{residual}}$. Tectonic earthquakes map primarily to double-couple (DC) and compensated linear vector dipole (CLVD) hypotheses, while collapses map primarily to negative ISO (–$\mathrm{ISO}_{\mathrm{CAL}}$), with secondary CLVD in some events. The method is applicable wherever source-type posteriors are available, given a declared hypothesis family, and produces explicit hypothesis probabilities with uncertainty and a residual diagnostic that flags diffuse or weakly matched posteriors, and can feed directly into monitoring and decision pipelines.
High-speed trains can serve as highly repeatable, low-cost seismic sources for subsurface imaging and monitoring along railway corridors. In viaduct settings, however, a passing train continuously excites multiple bridge piers, causing strong superposition of left- and right-going wavefields and severe crosstalk. As a result, train-passage records, despite their high signal-to-noise ratio, cannot be directly used. To address this limitation, we propose a wavefield-separation method based on randomized Cadzow filtering. The workflow integrates Radon-based apparent-velocity estimation, time flattening, local-block random permutation, and low-rank block Hankel reconstruction to separate left- and right-going wavefields in HST seismic data. Synthetic and field examples demonstrate that, compared with traditional methods, the proposed approach more effectively suppresses backward-propagating energy and wavefield crosstalk, while recovering unidirectional wavefields with improved continuity and greater amplitude and phase stability, leading to higher-quality virtual shot gathers and clearer dispersion images.
Summary High-accuracy marine gravity fields are essential for resolving fine-scale seafloor structures and supporting marine geophysical studies. However, the recovery of short-wavelength gravity signals from conventional satellite altimetry is limited by track spacing, measurement noise and filtering. The Surface Water and Ocean Topography (SWOT) mission provides dense wide-swath observations and improves the recovery of shorter-wavelength sea-surface slopes. Long-term multi-mission altimetry models, however, generally provide more stable gravity information at longer wavelengths. We therefore developed an iterative frequency-domain method that combines the medium- and long-wavelength component of DTU21 with the medium- and short-wavelength component of SWOT_04. A Gaussian low-pass filter and its complementary high-pass filter were used to combine the two models. Candidate cutoff wavelengths were tested at 1 km intervals. Quality-controlled National Centers for Environmental Information (NCEI) shipborne gravity observations were used only to select the cutoff wavelength. The minimum residual standard deviation (STD) of 4.25 mGal was obtained at 22 km. The final model was independently validated using Japan Agency for Marine-Earth Science and Technology (JAMSTEC) shipborne gravity data. The combined model achieved an STD of 3.83 mGal, compared with 4.11 mGal for SWOT_04 and 4.21 mGal for DTU21. This represents reductions of 6.8 per cent and 9.0 per cent, respectively. The combined model also had the lowest STD across all water-depth and distance-to-coast intervals. These results show that the proposed method combines the complementary spectral information of the two altimetry-derived gravity models and improves the accuracy of regional marine gravity fields.
Summary Seismic coda-correlation methods have revealed numerous body-wave-like features in globally stacked correlograms, offering new opportunities to constrain the structure of Earth’s deep interior. Unlike the common assumptions about the diffuse ambient noise field that enable reconstruction of surface waves, earthquake coda-correlation features arise from cross-terms in reverberating body waves of common slowness that share a subset of propagation legs. However, quantitative interpretation of these features for tomographic imaging has been hindered by the lack of distributed sensitivity kernels relating observed correlation signals to Earth heterogeneities. Here, we develop a forward modeling framework for the earthquake coda-correlation wavefield that bypasses conventional cross-correlation of late-coda waveforms. Our approach simulates correlation features directly by treating the time-reversed generating wavefield as a distributed source throughout Earth’s interior. The forward-modeled correlograms reproduce features obtained through conventional stacking with a significantly improved signal-to-noise ratio. Building on this framework, we derive finite-frequency traveltime banana-doughnut sensitivity kernels using adjoint methods, which quantify how correlation time measurements of correlation features depend on perturbations in P-wave velocity, S-wave velocity, and density. We compute sensitivity kernels for prominent correlation features (P*, ScS*, K*, and others) at station separations ranging from 30° to 180°. The kernels reveal extensive sampling of Earth’s mantle, outer core, and inner core, with spatial-sensitivity patterns fundamentally different from those of classical body-wave tomography. For antipodal station pairs, correlation features exhibit strong sensitivity throughout the deep interior, including regions poorly sampled by traditional seismic phases. Our results confirm that correlation features form through constructive interference of body-wave pairs with similar slowness, rather than representing Green’s functions between station pairs. This work establishes the theoretical foundation for coda-correlation tomography, enabling future three-dimensional imaging of Earth’s internal structure with unprecedented sampling of the deep interior. The sensitivity kernels provide a pathway to exploit the wealth of information contained in earthquake coda for high-resolution mantle and core tomography.
Summary Earthquake focal mechanisms provide important constraints on the crustal stress field. In stress inversion based on focal mechanism data, each mechanism contains two nodal planes, whereas the actual seismogenic fault plane is often difficult to identify. Consequently, Bootstrap resampling with random nodal plane selection is widely used to estimate uncertainties in stress inversion results. However, the reliability of these uncertainty bounds remains poorly understood. In this study, we evaluate the reliability of this approach using synthetic experiments and real fault-slip datasets. The results show that the reliability of Bootstrap-derived uncertainty bounds is jointly controlled by fault geometry and the true stress shape ratio (R). When seismogenic faults are distributed on both sides of the maximum compressive stress axis, the uncertainty bounds of principal stress orientations generally encompass the true solutions, whereas the uncertainty bounds of R may exhibit systematic biases depending on the true R value. By contrast, when faults occupy only one side of the fault space, the uncertainty bounds of both stress orientations and R may deviate from the true values. Further analysis indicates that this behavior reflects variations in the kinematic consistency between auxiliary-plane slip and the stress field under different R conditions, causing the influence of misselected nodal planes on inversion results to vary systematically. These findings highlight the importance of correctly identifying seismogenic fault planes and provide important implications for the interpretation of stress inversion uncertainties in practical applications.
