arXiv subjects

Daniel Peter

Publications and source records attributed to Daniel Peter.

7 recordsLinked to original sources

arXiv2026

Crustal and upper mantle model of the Middle East based on full-waveform inversion

We present MEAD-M20, a new tomographic model of the Middle East and its surrounding regions, including Anatolia, Iran, and the Caucasus. The model is developed within a full-waveform inversion framework, based on 3D wavefield simulations and the adjoint method, after 20 iterations, utilizing an extensive dataset from permanent and temporary stations available from EarthScope and regional networks. Starting from the global FWI model GLAD-M25 on a 60{\deg}x 60{\deg} regional mesh, we invert 210 regional earthquakes recorded by 1,215 stations to obtain the P- and S-wave model with transverse isotropy in the upper mantle. For the first 12 iterations, we combine multitaper traveltime measurements of 15-50 s body waves and 50-100 s body and surface waves on three components. We use a refined crustal mesh to better sample the crust after the 12th iteration and gradually decrease the minimum surface-wave period to 30 s. MEAD-M20 provides a self-consistent P- and S-wave model ready for seismic wave simulations, which is essential for accurate earthquake location, source parameter estimation, and seismic hazard assessment in the geologically and tectonically complex region. MEAD-M20 reveals several important geodynamical and tectonic features, including local mantle plumes beneath the Arabian Plate, Jordan, and the Levant, characterized by low-velocity anomalies and likely associated with volcanism in the Harrats, Jordan, and the Karacadag regions. In addition to the active subduction and rifting in the area, the model clearly identifies remnants of the Tethys Ocean beneath Eastern Anatolia, which become progressively shallower toward the Makran region in the south, consistent with the subduction history along the Bitlis-Zagros suture zone. We also observe lithospheric-scale low-velocity anomalies associated with the North and East Anatolian faults, extending to depths of approximately 200 km.

physics.geo-ph↗

arXiv2025

Deep learning enhanced initial model prediction in elastic FWI: application to marine streamer data

Low-frequency data are essential to constrain the low-wavenumber model components in seismic full-waveform inversion (FWI). However, due to acquisition limitations and ambient noise it is often unavailable. Deep learning (DL) can learn to map from high frequency model updates of elastic FWI to a low-wavenumber model update, producing an initial model estimation as if it was available from low-frequency data. We train a FusionNET-based convolutional neural network (CNN) on a synthetic dataset to produce an initial low-wavenumber model from a set of model updates produced by FWI on the data with missing low frequencies. We validate this DL-fused approach using a synthetic benchmark with data generated in an unrelated model to the training dataset. Finally, applying our trained network to estimate an initial low-wavenumber model based on field data, we see that elastic FWI starting from such a 'DL-fused' model update shows improved convergence on real-world marine streamer data.

physics.geo-ph↗

arXiv2022

Deconvolutional double-difference misfit measurements and the application for full-waveform inversion

It is challenging for full-waveform inversion to determine geologically informative models from field data. An inaccurate wavelet can make it more complicated. We develop a novel misfit function, entitled deconvolutional double-difference misfit measurement to cancel the influence of wavelet inaccuracy on inversion results. Unlike the popular double-difference misfit measurement in which the first difference is evaluated by cross-correlation, the proposed one employs deconvolution to do this step. Numerical examples demonstrate that full-waveform inversion with the new misfit function is resilient to the wavelet inaccuracy. It can also converge to plausible local minima even from rough initial models.

physics.geo-ph↗

arXiv2021

Cycle-skipping mitigation using misfit measurements based on differentiable dynamic time warping

The dynamic time warping (DTW) distance has been used as a misfit function for wave-equation inversion to mitigate the local minima issue. However, the original DTW distance is not smooth; therefore it can yield a strong discontinuity in the adjoint source. Such a weakness does not help nonlinear inverse problems converge to a plausible minimum by any means. We therefore introduce for the first time in geophysics the smooth DTW distance, which has demonstrated its performance in time series classification, clustering, and prediction as the loss function. The fundamental idea of such a distance is to replace the $\min$ operator with its smooth relaxation. Then it becomes possible to define the analytic derivative of DTW distance. The new misfit function is entitled to the differentiable DTW distance. Moreover, considering that the warping path is an indicator of the traveltime difference between the observed and synthetic trace, a penalization term is constructed based on the warping path such that the misfit accumulation by the penalized differentiable DTW distance is weighted in favor of the traveltime difference. Numerical examples demonstrate the advantage of the penalized differentiable DTW misfit function over the conventional non-differentiable one.

physics.geo-ph↗

arXiv2020

Preconditioned BFGS-based Uncertainty Quantification in elastic Full Waveform Inversion

Full Waveform Inversion (FWI) plays a vital role in reconstructing geophysical structures. The Uncertainty Quantification regarding the inversion results is equally important but has been missing out in most of the current geophysical inversions. Mathematically, uncertainty quantification is involved with the inverse Hessian (or the posterior covariance matrix), which is prohibitive in computation and storage for practical geophysical FWI problems. L-BFGS populates as the most efficient Gauss-Newton method; however, in this study, we empower it with the new possibility of accessing the inverse Hessian for uncertainty quantification in FWI. To facilitate the inverse-Hessian retrieval, we put together BFGS (essentially, full-history L-BFGS) with randomized singular value decomposition towards a low-rank approximation of the Hessian inverse. That the rank number equals the number of iterations makes this solution efficient and memory-affordable even for large-scale inversions. Also, based on the adjoint method, we formulate different diagonal Hessian initials as preconditioners and compare their performances in elastic FWI. We highlight our methods with the elastic Marmousi benchmark, demonstrating the applicability of preconditioned BFGS in large-scale FWI and uncertainty quantification.

physics.comp-ph↗

arXiv2019

A Modular Benchmarking Infrastructure for High-Performance and Reproducible Deep Learning

We introduce Deep500: the first customizable benchmarking infrastructure that enables fair comparison of the plethora of deep learning frameworks, algorithms, libraries, and techniques. The key idea behind Deep500 is its modular design, where deep learning is factorized into four distinct levels: operators, network processing, training, and distributed training. Our evaluation illustrates that Deep500 is customizable (enables combining and benchmarking different deep learning codes) and fair (uses carefully selected metrics). Moreover, Deep500 is fast (incurs negligible overheads), verifiable (offers infrastructure to analyze correctness), and reproducible. Finally, as the first distributed and reproducible benchmarking system for deep learning, Deep500 provides software infrastructure to utilize the most powerful supercomputers for extreme-scale workloads.

cs.DC↗

arXiv2016

Anelastic sensitivity kernels with parsimonious storage for adjoint tomography and full waveform inversion

We introduce a technique to compute exact anelastic sensitivity kernels in the time domain using parsimonious disk storage. The method is based on a reordering of the time loop of time-domain forward/adjoint wave propagation solvers combined with the use of a memory buffer. It avoids instabilities that occur when time-reversing dissipative wave propagation simulations. The total number of required time steps is unchanged compared to usual acoustic or elastic approaches. The cost is reduced by a factor of 4/3 compared to the case in which anelasticity is partially accounted for by accommodating the effects of physical dispersion. We validate our technique by performing a test in which we compare the $K_\alpha$ sensitivity kernel to the exact kernel obtained by saving the entire forward calculation. This benchmark confirms that our approach is also exact. We illustrate the importance of including full attenuation in the calculation of sensitivity kernels by showing significant differences with physical-dispersion-only kernels.

physics.comp-ph↗