Talks and presentations

Sensitivity analysis of acoustic waveform inversion to velocity anomalies in the near-surface using the Fréchet derivative

September 01, 2022

Talk, Society of Exploration Geophysicists, Houston, Texas

Seismic waveform inversion (WI) has been applied to high- resolution velocity model building at all scales, but how truthful are the amplitudes of velocity anomalies recovered in the WI models? Previous sensitivity analysis of WI has focused on estimating model uncertainties, understanding the radiation patterns of different geophysical parameters, and separating the tomography and migration mode of WI in the wavenumber-domain. Few of the studies explore if the WI is equally sensitivity to high and low velocity anomalies, whereas nonlinear traveltime tomography (TT) is known to be more sensitive to high velocity anomalies than low velocity anomalies because of wavefront healing. In this paper, we analyze the sensitivity of WI to velocity anomalies by first comparing the amplitudes of wavefield perturbations generated by injected high and low velocity anomalies. We provide theoretical calculations for a point scatter and numerical simulations for a velocity anomaly with a size equivalent to a fraction of the dominant wavelength. Then, we perform WI tests to recover high and low velocity anomalies of equal velocity or slowness perturbations. Acoustic waveform modeling and inversion are used for simplicity. The preliminary results show acoustic WI is potentially more sensitive to low velocity anomalies than high velocity anomalies. The joint inversion of traveltime and waveform data is advocated for near-surface velocity estimation, since the different sensitivity of TT and acoustic WI can be complementary to each other.

Early-arrival waveform inversion using data uncertainties and matching filters with application to near-surface seismic refraction data.

March 01, 2021

Talk, Seismological Society of America, Online (due to pandemic)

We develop an early arrival waveform inversion (EAWI) technique for high-resolution near-surface velocity estimation by iteratively updating the P-wave velocity model to minimize the difference between the observed and calculated seismic refraction data. Traditional EAWI uses a least-squares penalty function and an acoustic forward-modeling engine. Conventional least-squares error is sensitive to data with low signal-to-noise ratio (S/N) and iterations of EAWI stop at a local-minimum data misfit or at the preassigned maximum number of iterations. These stopping criteria can result in overfitting the data. In addition, fitting the elastic field data with an acoustic modeling engine can introduce artifacts in velocity estimation, especially in land data with significant elastic effects. To overcome these challenges, we develop a robust EAWI (REAWI) method by (1) incorporating the data uncertainties into the penalty function and (2) mitigating the elastic effects using a matching filter workflow. The data uncertainties are estimated from waveform reciprocal errors. When full-waveform reciprocity is not available, trace interpolation is applied. The proposed method prevents closely fitting data with low S/N, avoids overall overfitting by stopping the iterations when a normalized chi-square (⁠χ2 ⁠) waveform misfit of one is achieved, and is less affected by elastic effects. Numerical examples and application to near-surface refraction data at a groundwater contamination site suggest that the final REAWI models are more accurate than the corresponding EAWI models, at the same level of misfit. This is the first known application of a matching filter workflow to real land data. The final REAWI models satisfy an appropriate misfit between the real data and predicted elastic P-wave data, making this approach in this respect equivalent to elastic waveform inversion. We also develop a method to analyze model constraint by examining the energy of the wavefield Fréchet derivative thereby avoiding the influence of the data residuals in traditional Fréchet kernels.

Semi-supervised Data-driven Surface Wave Tomography using Wasserstein Cycle-consistent GAN: Application on Southern California Plate Boundary Region

December 01, 2020

Talk, American Geophysical Union Fall Meeting, Online (due to pandemic)

Current machine learning based shear wave velocity (Vs) inversion using surface wave dispersion measurements utilizes synthetic dispersion curves calculated from existing 3-D velocity models as training datasets. It is shown in the previous studies that the performances of the resulting networks are dependent on the diversity of the training data. We present an improved semi-supervised deep learning algorithm-based method that incorporates both observed and synthetic surface wave dispersion curves in the network training process. The algorithm is termed Wasserstein cycle-consistent generative adversarial networks (Wcycle-GAN), which combines the architecture of cycle-consistent GAN with Wasserstein loss metrics in optimization. Different from conventional supervised deep learning approaches, the GAN architecture also extracts structural information from the observed surface wave dispersion data in the training process that may improve generalization of the resulting network. The cycle-consistent loss addresses soft constraints on the trained neural networks to be reversible and thus reduces the variance of the trained networks. The Wasserstein metric provides weaker topology for convergence and improves spatial continuity of the predicted shear velocity (Vs) models. We demonstrate these improvements by applying the Wcycle-GAN to 4066 fundamental mode Rayleigh wave phase and group dispersion curves obtained in Southern California (SC). In general, the 3-D Vs model predicted by the best training Wcycle-GAN is consistent with previous surface wave tomography studies of SC in the overlapping area, but with smaller data misfit, yields better spatial smoothing, and provides improved images of structures near faults and in the top 5 km. Our results indicate that the proposed Wcycle-GAN algorithm has strong training stability and generalization abilities.

Wasserstein cycle-consistent generative adversarial network for improved seismic impedance inversion: Example on 3D SEAM model

September 01, 2020

Talk, Society of Exploration Geophysicists, Online (due to pandemic)

The convolutional neural networks (CNNs) have attracted great attentions in seismic exploration applications by their capability of learning the representations of data with multiple level of abstractions, given an adequate amount of labeled data. In seismic impedance inversion, however, the availability of labeled data-impedance pairs is often limited. The cycle-consistent generative adversarial networks (Cycle-GAN) is proven as a powerful semi-supervised learning solution by incorporating these unpaired data into its training, but it suffers from the training instability like most GAN algorithms. This study, starting from the recent progress in tackling such a challenge by using the Wasserstein GAN, proposes improving the Cycle-GAN algorithm by integrating the Wasserstein loss with gradient penalty loss as the target loss function, here denoted as the Wasserstein cycle-consistent GAN (WCycle-GAN). Correspondingly, the new algorithm benefits from both weaker topology in Wasserstein distance and better data regularizations in cycle-consistent loss, leading to enhanced training robustness and generalization abilities. We validate its performance through an example of impedance inversion on a subset of the 3D Seismic Advanced Modeling (SEAM) data. The results not only demonstrate the outperformance of the new algorithm over the conventional Cycle-GAN but also suggest its great potential for assisting other semi-supervised learning applications.

Data weighted full-waveform inversion with adaptive moment estimation for near-surface seismic refraction data

September 01, 2019

Talk, Society of Exploration Geophysicists, San Antonio, TX

Full-waveform inversion (FWI) is a technique for high resolution subsurface velocity estimation. Normally a least-squares error between observed and calculated waveform is used as a penalty function, which is sensitive to outliers in the data. We propose a noise-resistant inverse strategy by estimating and incorporating the data covariance matrix into time-domain seismic waveform inversion, which is termed Data weighted full-waveform inversion (DWFWI). The data covariance matrix is built from waveform reciprocal errors or from pre-first-arrival signal-to-noise ratio. There are two modifications in DWFWI compared to conventional FWI: (1) Waveform gradient and misfit function are weighted by data covariance matrix to suppress artifacts from data with low signal-to-noise ratio and to determine stopping criteria for the inversion; (2) FWI penalty function optimization using adaptive moment estimation method. Synthetic examples show the estimated DWFWI models are more accurate than the corresponding FWI models, at the same level of misfit. In addition, Adam outperforms the conjugate gradient method in reaching an ideal fit of the data, resulting in better amplitude recovery. We present preliminary results of applying DWFWI to a 2D near-surface seismic refraction data.

Variable Grid Traveltime Tomography for Near-surface Seismic Imaging

December 01, 2017

Talk, American Geophysical Union, New Orleans, LA

We present a new algorithm of traveltime tomography for imaging the subsurface with automated variable grids upon geological structures. The nonlinear traveltime tomography along with Tikhonov regularization using conjugate gradient method is a conventional method for near surface imaging. However, model regularization for any regular and even grids assumes uniform resolution. From geophysical point of view, long-wavelength and large scale structures can be reliably resolved, the details along geological boundaries are difficult to resolve. Therefore, we solve a traveltime tomography problem that automatically identifies large scale structures and aggregates grids within the structures for inversion. As a result, the number of velocity unknowns is reduced significantly, and inversion intends to resolve small-scale structures or the boundaries of large-scale structures. The approach is demonstrated by tests on both synthetic and field data. One synthetic model is a buried basalt model with one horizontal layer. Using the variable grid traveltime tomography, the resulted model is more accurate in top layer velocity, and basalt blocks, and leading to a less number of grids. The field data was collected in an oil field in China. The survey was performed in an area where the subsurface structures were predominantly layered. The data set includes 476 shots with a 10 meter spacing and 1735 receivers with a 10 meter spacing. The first-arrival traveltime of the seismogram is picked for tomography. The reciprocal errors of most shots are between 2ms and 6ms. The normal tomography results in fluctuations in layers and some artifacts in the velocity model. In comparison, the implementation of new method with proper threshold provides blocky model with resolved flat layer and less artifacts. Besides, the number of grids reduces from 205,656 to 4,930 and the inversion produces higher resolution due to less unknowns and relatively fine grids in small structures. The variable grid traveltime tomography provides an alternative imaging solution for blocky structures in the subsurface and builds a good starting model for waveform inversion and statics.

Ao Cai