Swapnil Mache

We consider the problem of seismic reflectivity inversion, which pertains to the high-resolution recovery of interface locations and reflection coefficients from seismic measurements, which are vital for estimating the subsurface structure. We develop two model-based neural networks within the framework of deep-unfolding. First, we propose a nonuniform minimax concave penalty regularized formulation for reflectivity inversion and unfold the resulting iterative algorithm into a network. Second, we propose a nonuniform sparse model that relies on a combination of regularizers (composite regularization) and develop the nonuniform sparse proximal averaging network (NuSPAN). We demonstrate the efficacy of the proposed approaches over the benchmark techniques through numerical experiments on synthetic 1-D seismic traces and 2-D wedge models. We also report validations on the 2-D Marmousi2 simulated model and 3-D real field measurements from the Penobscot 3D survey off the coast of Nova Scotia, Canada. The accuracy of the proposed approaches is higher than the state-of-the-art techniques in terms of amplitude and support recovery. Further, for Marmousi2, the proposed deep-unfolding networks result in 600× faster inference than the fast iterative shrinkage-thresholding algorithm (FISTA). In terms of combined training and inference times, the learned iterative shrinkage-thresholding algorithm (LISTA) is the fastest. The inference speed-up is significant given that the volume of data is typically large in seismic signal processing.

The analyses of cores retrieved from three sites near Port Blair (South Andaman) revealed out-of-sequence deposits at various depths. They are identified as previous episodes of tsunami by their sediment characteristics and microfossil content, using the 2004 event deposition as a template. These deposits have median ages of 601 cal. yr BP, 837 cal. yr BP, 1440 cal. yr BP, 3018 cal. yr BP, 3591 cal. yr BP, 4712 cal. yr BP, 5607 cal. yr BP, and 6357 cal. yr BP and are chronologically equivalent of those from the far-field locations in the Indian Ocean region. The distant deposits that are correlated with the South Andaman sites most likely owe their origin to the 2004-type events, as indicated by tsunami simulations in the study region. The long-term record presented here is characterized by an early phase of a quasiperiodic recurrence regime that transitions into a distinct interval of temporally clustered events. The quasiperiodic regime that appears around the mid-Holocene with an inter-event interval of 980 ± 385 years is followed by a sizable quiescent period of 1605 ± 245 years, before being succeeded by a regime of temporally clustered events. The chronology of nine tsunami events in the last 6500 years from the Indian Ocean region, thus implies a nonlinear pattern for the causative earthquakes. As demonstrated in the subduction zones elsewhere, the temporal variability of tsunamigenic great earthquakes is supported by the theoretical models espousing the characteristics of long-term stress recycling processes active within the subduction zones and transfer processes between the lower viscoelastic layer and the upper seismogenic crust.

The Hilbert–Huang Transform (HHT) has been sparsely applied to problems in seismology, although previous studies have pointed to its broad scope. In this maiden attempt, we use the HHT to represent earthquake energy release duration and frequency content and compare the results with two conventional inversion methods. By selecting examples from interplate, intraplate, and intraslab settings, we demonstrate that the HHT has the power to discriminate energy release of earthquakes with different tectonic affiliations. We observe that the dominant frequencies for energy release are higher for intraslab earthquakes than for interplate and intraplate events. We use the empirical mode decomposition‐based HHT and introduce a new parameter, which we name the energy rate function (ERF), to quantify the energy release. By employing empirical Green’s functions to remove the path and site effects and using a linear combination of a select set of intrinsic mode functions, we generate the station‐specific relative measure of energy that we refer to as relative ERFs (RERFs). Averaged over RERFs from multiple stations, the ERF represents a measure of the total relative energy release, comparable to the moment rate functions (MRFs) and SCARDEC source time functions (STFs). Results for six of the seven earthquakes we analyzed show high cross correlation with the STFs (0.84 ± 0.03) and MRFs (0.79 ± 0.06), but there are mismatches between ERFs and MRFs or STFs when the energy release is complex and involves multisegment or bilateral ruptures. The proposed method is computationally efficient, requiring only 3.46 ± 2.62 s on average, compared to  20 min ( 1200 s) for the teleseismic inversion method we employ. With its ability to represent the seismic source in terms of energy release, the ERF method has the potential to evolve not as an alternative to waveform inversion but as a rapid time–frequency analysis tool, useful for earthquake hazard assessment.