Search

An alternative approach to acquire transmission travel time data is proposed, exploiting the geometry of devices commonly used in ultrasound computed tomography for medical imaging or non-destructive testing with ultrasonic waves. The intent is to (i) shorten acquisition time for devices with a large number of emitters, (ii) to eliminate the calibration step, and (iii) to suppress instrument noise. Inspired by seismic ambient field interferometry, the method rests on the active excitation of diffuse ultrasonic wavefields and the extraction of deterministic travel time information by inter-station correlation. To reduce stochastic errors and accelerate convergence, ensemble interferograms are obtained by phase-weighted stacking of observed and computed correlograms, generated with identical realizations of random sources. Mimicking an imaging setup, the accuracy of the travel time measurements as a function of the number of emitters and random realizations can be assessed both analytically and with spectral-element simulations for phantoms mimicking the model parameter distribution. The results warrant tomographic reconstructions with straight- or bent-ray approaches, where the effect of inherent stochastic fluctuations can be made significantly smaller than the effect of subjective choices on regularisation. This work constitutes a first conceptual study and a necessary prelude to future implementations.

SUMMARY Markov chain Monte Carlo (MCMC) sampling of solutions to large-scale inverse problems is, by many, regarded as being unfeasible due to the large number of model parameters. This statement, however, is only true if arbitrary, local proposal distributions are used. If we instead use a global proposal, informed by the physics of the problem, we may dramatically improve the performance of MCMC and even solve highly nonlinear inverse problems with vast model spaces. We illustrate this by a seismic full-waveform inverse problem in the acoustic approximation, involving close to 106 parameters. The improved performance is mainly seen as a dramatic shortening of the burn-in time (the time used to reach at least local equilibrium), but also the algorithm’s ability to explore high-probability regions (through more accepted perturbations) is potentially better. The sampling distribution of the algorithm asymptotically converges to the posterior probability distribution, but as with all other inverse methods used to solve highly nonlinear inverse problems we have no guarantee that we have seen all high-probability solutions in a finite number of iterations. On the other hand, with the proposed method it is possible to sample more high-probability solutions in a shorter time, without sacrificing asymptotic convergence. This may be a practical advantage for problems with many parameters and computer-intensive forward calculations.

Two-dimensional modeling of gravity and magnetic anomalies in terms of polygonal bodies is a popular approach to infer possible configurations of geological structures in the subsurface. Alternatively to the traditional trial-and-error manual fit of measured data, here we illustrate a probabilistic strategy to solve the inverse problem. First we derive a set of formulae for solving a 2.75-dimensional forward model, where the polygonal bodies have a given finite lateral extent, and then we devise a Hamiltonian Monte Carlo algorithm to jointly invert gravity and magnetic data for the geometry and properties of the polygonal bodies. This probabilistic approach fully addresses the nonlinearity of the forward model and provides uncertainty estimation. The result of the inversion is a collection of models which represent the posterior distribution, analysis of which provides estimates of sought properties and may reveal different scenarios., Geophysical Research Letters, 49 (20), ISSN:0094-8276, ISSN:1944-8007

We present preliminary results on a moment tensor inversion workflow for Distributed Acoustic Sensing (DAS). It makes use of a fast-marching Eikonal solver and synthetically modeled data. The study specifically focuses on borehole settings for geothermal sites. Distributed Acoustic Sensing measures the wavefield with high spatial and temporal resolution. In borehole settings, individual DAS traces generally prove to be noisier than co-located geophones, whereas the densely spaced DAS shot-gathers show features that would have otherwise been missed by the commonly more sparsely distributed geophone chains. For example, the coherency in the DAS records shows the polarity reversals of the arriving wavefield in great detail, which can help constrain the moment tensor of the seismic source. The synthetic tests encompass different source types and source positions relative to the deployed fiber to assess moment tensor resolvability. Further tests include the addition of a three-component seismometer at different positions to investigate an optimal network configuration, as well as various noise conditions to mimic real data. The synthetic tests are tailored to prepare for the data from future microseismicity monitoring with DAS in the conditions of the Utah FORGE geothermal test site, Utah, USA. The proposed method aims at improving amplitude-based moment tensor inversion for DAS deployed in downhole or underground lab contexts.

We present a probabilistic approach to constrain the density distribution in the Earth based on surface wave dispersion. Despite its outstanding importance in studies of the Earth’s thermo-chemical state and dynamics, 3D density variations remain poorly known, thereby posing one of the major challenges in geophysics. Since the sensitivity of most seismic data to density is small compared to sensitivity with respect to seismic velocities, regularisation in traditional deterministic inversion tends to bias the recovered density image significantly. To avoid this issue, we propose to solve a regularisation-free Bayesian inference problem using the Hamiltonian Monte Carlo Markov Chain algorithm. In the interest of simplicity, we consider anisotropic stratified media, where dispersion curves and corresponding sensitivity kernels can be computed semi-analytically. Exploiting derivative information for efficient sampling, Hamiltonian Monte Carlo approximates the posterior probability density of all model parameters, namely the P-wave velocities vPV and vPH , the S-wave velocities vSV and vSH , the anisotropy parameter η, and, of course, density ρ. The proposed method forms the foundation of an open-source tool box that can be used to assess the unbiased ability of surface wave dispersion data, characterised in terms of frequency and modal content, to constrain density variations and their trade-offs with other Earth model parameters.

We propose methods to efficiently explore the generalized nullspace of (non-linear) inverse problems, defined as the set of plausible models that explain observations within some misfit tolerance. Owing to the random nature of observational errors, the generalized nullspace is an inherently probabilistic entity, described by a joint probability density of tolerance values and model parameters. Our exploration methods rest on the construction of artificial Hamiltonian systems, where models are treated as high-dimensional particles moving along a trajectory through model space. In the special case where the distribution of misfit tolerances is Gaussian, the methods are identical to standard Hamiltonian Monte Carlo, revealing that its apparently meaningless momentum variable plays the intuitive role of a directional tolerance. Its direction points from the current towards a new acceptable model, and its magnitude is the corresponding misfit increase. We address the fundamental problem of producing independent plausible models within a high-dimensional generalized nullspace by autotuning the mass matrix of the Hamiltonian system. The approach rests on a factorized and sequentially preconditioned version of the L-BFGS method, which produces local Hessian approximations for use as a near-optimal mass matrix. An adaptive time stepping algorithm for the numerical solution of Hamilton's equations ensures both stability and reasonable acceptance rates of the generalized nullspace sampler. In addition to the basic method, we propose variations of it, where autotuning focuses either on the diagonal elements of the mass matrix or on the macroscopic (long-range) properties of the generalized nullspace distribution. We quantify the performance of our methods in a series of numerical experiments, involving analytical, high-dimensional, multimodal test functions. These are designed to mimic realistic inverse problems, where sensitivity to different model parameters varies widely, and where parameters tend to be correlated. The tests indicate that the effective sample size may increase by orders of magnitude when autotuning is used. Finally, we present a proof of principle of generalized nullspace exploration in viscoelastic full-waveform inversion. In this context, we demonstrate (1) the quantification of inter- and intraparameter trade-offs, (2) the flexibility to change model parametrization a posteriori, for instance, to adapt averaging length scales, (3) the ability to perform dehomogenization to retrieve plausible subwavelength models and (4) the extraction of a manageable number of alternative models, potentially located in distinct local minima of the misfit functional., Geophysical Journal International, 227 (2), ISSN:0956-540X, ISSN:1365-246X

Isothermal titration calorimetry (ITC) is a widely used method to determine binding affinities and thermodynamics in ligand-receptor interactions, but it also has the capability of providing detailed information on much more complex events. However, the lack of available methods to analyze ITC data is limiting the use of the technique in such multifaceted cases. Here, we present the software ANISPROU. Through a semi-empirical approach that allows for extraction of quantitative information from complex ITC data, ANISPROU solves an inverse problem where three parameters describing a set of predefined functions must be found. In analogy to strategies adopted in other scientific fields, such as geophysics, imaging, and many others, it employs an optimization algorithm which minimizes the difference between calculated and experimental data. In contrast to the existing methods, ANISPROU provides automated and objective analysis of ITC data on sodium dodecyl sulfate (SDS)-induced protein unfolding, and in addition, more information can be extracted from the data. Here, data series on SDS-mediated protein unfolding is analyzed, and binding isotherms and thermodynamic information on the unfolding events are extracted. The obtained binding isotherms as well as the enthalpy of different events are similar to those obtained using the existing manual methods, but our methodology ensures a more robust result, as the entire data set is used instead of single data points. We foresee that ANISPROU will be useful in other cases with complex enthalpograms, for example, in cases with coupled interactions in biomolecular, polymeric, and amphiphilic systems including cases where both structural changes and interactions occur simultaneously.

The Mt. Melbourne Volcanic Complex (MMVC) is located in Northern Victoria Land (Antarctica) along the western flank of the West Antarctic Rift System, at the boundary with the Transantarctic Mountains. It is constituted by two main volcanic areas, i.e. the Mt. Melbourne Edifice (MME) and the Cape Washington Shield (CWS), and some other minor centres.To date, the inner structure of this volcanic complex is still poorly known, being the direct geological information on site confined to either glacial erratics or a few rock outcrops not hidden by the ice sheet. Consequently, even the temporal building up and evolution of the MMVC as well as its primary magmatic source are still under investigation (debated).Recently, we attempted to define the geological structure of the MMVC by means of digital enhancement and forward modeling performed on a high-resolution aeromagnetic dataset (Ghirotto et al. 2020, EGU). Coupling both information derived from past geological/geophysical studies and unpublished magnetic susceptibility measurements from rock samples collected in the field, we proposed two models to explain the chronological evolution of the MME and CWS. These models involve either i) major magmatic events occurred in periods of both normal and reverse magnetic polarity or ii) only magmatic flows with normal polarity.To gain further insights into the geological structure and the geodynamic evolution of the MMVC in relation to the two proposed models, we develop here a Hamiltonian Monte Carlo (HMC) algorithm (Fichtner et al. 2018) based on the probabilistic approach to inverse problems. To date, this methodology has never been applied to aeromagnetic data for geological studies. In detail, the above proposed models provide some soft a priori information from which to start exploring potential solutions. The parameterization of the volcanic area is defined in terms of 2-D polygonal bodies, representing e.g. magmatic lava flows, where the unknown parameters are represented by both the position of the vertices and/or the magnetization (induced and/or remnant), resulting in a non-linear forward model. The HMC algorithm requires the computation of gradients of the posterior probability density (PPD), i.e., derivatives of the objective functional with respect to the position of vertices of the bodies and magnetization, in order to better move the inversion process toward high-probability areas in the model space manifold. We implement such calculations using automatic differentiation, a tool which is very accurate and fast compared to other approaches such as finite difference. The result of the inversion is then a collection of models representing the PPD, from which statistical analysis can provide measures of uncertainty and plausible geological scenarios.In this study we present some preliminary results of applying the above-mentioned methodology, which finally could help unravel the framework of the MMVC.

Uncertainty quantification is an essential part of many studies in Earth science. It allows us, for example, to assess the quality of tomographic reconstructions, quantify hypotheses and make physics-based risk assessments. In recent years there has been a surge in applications of uncertainty quantification in seismological inverse problems. This is mainly due to increasing computational power and the ‘discovery’ of optimal use cases for many algorithms (e.g., gradient-based Markov Chain Monte Carlo (MCMC). Performing Bayesian inference using these methods allows seismologists to perform advanced uncertainty quantification. However, oftentimes, Bayesian inference is still prohibitively expensive due to large parameter spaces and computationally expensive physics.Simultaneously, machine learning has found its way into parameter estimation in geosciences. Recent works show that machine learning both allows one to accelerate repetitive inferences [e.g. Shahraeeni & Curtis 2011, Cao et al. 2020] as well as speed up single-instance Monte Carlo algorithms using surrogate networks [Aleardi 2020]. These advances allow seismologists to use machine learning as a tool to bring accurate inference on the subsurface to scale.In this work, we propose the novel inclusion of adjoint modelling in machine learning accelerated inverse problems. The aforementioned references train machine learning models on observations of the misfit function. This is done with the aim of creating surrogate but accelerated models for the misfit computations, which in turn allows one to compute this function and its gradients much faster. This approach ignores that many physical models have an adjoint state, allowing one to compute gradients using only one additional simulation.The inclusion of this information within gradient-based sampling creates performance gains in both training the surrogate and the sampling of the true posterior. We show how machine learning models that approximate misfits and gradients specifically trained using adjoint methods accelerate various types of inversions and bring Bayesian inference to scale. Practically, the proposed method simply allows us to utilize information from previous MCMC samples in the algorithm proposal step.The application of the proposed machinery is in settings where models are extensively and repetitively run. Markov chain Monte Carlo algorithms, which may require millions of evaluations of the forward modelling equations, can be accelerated by off-loading these simulations to neural nets. This approach is also promising for tomographic monitoring, where experiments are repeatedly performed. Lastly, the efficiently trained neural nets can be used to learn a likelihood for a given dataset, to which subsequently different priors can be efficiently applied. We show examples of all these use cases. Lars Gebraad, Christian Boehm and Andreas Fichtner, 2020: Bayesian Elastic Full‐Waveform Inversion Using Hamiltonian Monte Carlo.Ruikun Cao, Stephanie Earp, Sjoerd A. L. de Ridder, Andrew Curtis, and Erica Galetti, 2020: Near-real-time near-surface 3D seismic velocity and uncertainty models by wavefield gradiometry and neural network inversion of ambient seismic noise.Mohammad S. Shahraeeni and Andrew Curtis, 2011: Fast probabilistic nonlinear petrophysical inversion.Mattia Aleardi, 2020: Combining discrete cosine transform and convolutional neural networks to speed up the Hamiltonian Monte Carlo inversion of pre‐stack seismic data.

The Transantarctic Mountains (TAM) separate the Mesozoic to recent West Antarctic rift system (WARS) from a wide and depressed triangular sector of East Antarctica spanning from 100° E to 160° E in longitude and from the Oates, George V and Adelie coastlines to 85° S in latitude. The sub-ice bedrock of this sector shows a basin and range style topography comprising two major basins of continental proportions -the Wilkes Basin and the Aurora Basin complex- and many smaller basins such as the Adventure, Concordia, Aurora and Vostok trenches. Most of these basins and trenches exhibit a triangular shape with the acutest angle pointing approximatively to a single pole towards the South, giving a fan shaped pattern of significant dimensions. We name here this region as the East Antarctic Fan shaped Basin Province (EAFBP). To the West, this province is limited by the intraplate Gamburtsev Mountains (GM).Origins and inter-relationships between these four fundamental Antarctic tectonic units (WARS, TAM, EAFBP, GM) are still poorly understood and strongly debated. In the EAFBP, very little is known about the mechanism generating the basins, their formation time, whether they are all coeval and if and how they relate to Australia basins before Antarctica-Australia rifting. Present genetic hypotheses for some of the basins span from continental rifting to a purely flexural origin or a combination of the two. Also, post-tectonic erosional and depositional processes may have had a significant impact on the present-day topographic configuration.Here we investigate the possibility that the EAFBP is the result of a single genetic mechanism: a wide fan-shaped intra-continental extension around a pivot point at about 135° E, 85° S that occurred at the Mesozoic-Cenozoic transition. We discuss evidence from the sub-ice topography and potential field airborne and satellite data.We have used international community-based Antarctic compilations in public domain, including BedMachine (Morlighem et al., 2020), AntGG (Scheinert et al., 2016) and ADMAP 2.0 (Golynsky et al., 2018). We have applied image segmentation techniques to the rebounded sub-ice topography to automatically trace the first order shape of the sub-ice basins. Then we have fitted the edges of the basins by maximum circles and we have estimated the best Euler pole identified by their intersection. Potential field anomalies have been taken into account in order to enlighten major discontinuities not revealed by the sub-ice topography.Software simulations of the EAFBP opening in the frame of global plate tectonics reconstructions indicate that it may be inserted in the frame of the later phase of the Antarctica-Australia rifting, giving constraints on timing that allow us to date the EAFBP opening at the Mesozoic-Cenozoic transition.The reconnaissance of the EAFBP as the result of a continental-scale fan-shaped extension may have deep implications on global and regional tectonics plate reconstructions, plate deformation assumptions and new tectonic evolutionary models of WARS, TAM and GM.

Ultrasound computed tomography (USCT) is a promising imaging modality for breast cancer screening. Two challenges commonly arising in time-of-flight USCT are (1) to efficiently deal with large data sets and (2) to effectively mitigate the ill-posedness for an adequate reconstruction of the model. In this contribution, we develop an optimization strategy based on a stochastic descent method that adaptively subsamples the data, and analyze its performance in combination with different sparsity-enforcing regularization techniques. The algorithms are tested on numerical as well as real data obtained from synthetic phantom scans of the previous USCT Data Challenges., Proceedings of SPIE, 11602, ISSN:0277-786X, Medical Imaging 2020: Ultrasonic Imaging and Tomography. Proceedings SPIE Medical Imaging | 15-20 February 2020, ISBN:978-1-5106-4033-7, ISBN:978-1-5106-4034-4

The concept of compressing deep Convolutional Neural Networks (CNNs) is essential to use limited computation, power, and memory resources on embedded devices. However, existing methods achieve this objective at the cost of a drop in inference accuracy in computer vision tasks. To address such a drawback, we propose a framework that leverages knowledge distillation along with customizable block-wise optimization to learn a lightweight CNN structure while preserving better control over the compression-performance tradeoff. Considering specific resource constraints, e.g., floating-point operations per inference (FLOPs) or model-parameters, our method results in a state of the art network compression while being capable of achieving better inference accuracy. In a comprehensive evaluation, we demonstrate that our method is effective, robust, and consistent with results over a variety of network architectures and datasets, at negligible training overhead. In particular, for the already compact network MobileNet_v2, our method offers up to 2x and 5.2x better model compression in terms of FLOPs and model-parameters, respectively, while getting 1.05% better model performance than the baseline network., Comment: This paper has been accepted to ICPR 2020

Since the '60s of the last century, the calculation of the magnetic anomalies caused by 2D uniformly polarized bodies with polygonal cross-section has been mainly performed using the popular algorithm of Talwani and Heirtzler (1962, 1964). Recently, Kravchinsky et al. (2019, https://doi.org/10.1029/2019GL082767) claimed errors in the above algorithm formulation, proposing new corrective formulas and questioning the effectiveness of almost 60 years of magnetic calculations. Here we show that the two approaches are equivalent and Kravchinsky et al.'s formulas simply represent an algebraic variant of those of Talwani and Heirtzler. Moreover, we analyze a large amount of random magnetic scenarios, involving both changing-shape polygons and a realistic geological model, showing a complete agreement among the magnetic responses of the two discussed algorithms and the one proposed by Won and Bevis (1987, https://doi.org/10.1190/1.1442298). We release the source code of the algorithms in Julia and Python languages.

The Hamiltonian Monte Carlo method (HMC) is gaining popularity in the geophysical community to fully address nonlinear inverse problems and related uncertainty quantification. We present here an application of HMC to invert seismic data in the acoustic approximation in the context of reflection seismology. We address a 2-D problem, in the form of a vertical cross section where both source and receivers are located near the surface of the model. To solve the forward problem we utilise the finite-difference method with PML absorbing boundary conditions. The observed data are represented by a set of shotgathers. The crucial aspect for a successful application of the HMC lies in the capability of performing gradient computations in an efficient manner. To this end, we use the adjont state method to compute the gradient of the misfit functional, which has a computational cost of only about twice that of the forward computation, a very efficient strategy. From the collection of samples characterising the posterior distribution obtained with the HMC, we can derive quantities of interest using statistical analysis and assess uncertainties. We illustrate an application of this methodology on a synthetic test mimicking the setup encountered in exploration problems., EGUsphere

Conventionally, AI models are thought to trade off explainability for lower accuracy. We develop a training strategy that not only leads to a more explainable AI system for object classification, but as a consequence, suffers no perceptible accuracy degradation. Explanations are defined as regions of visual evidence upon which a deep classification network makes a decision. This is represented in the form of a saliency map conveying how much each pixel contributed to the network's decision. Our training strategy enforces a periodic saliency-based feedback to encourage the model to focus on the image regions that directly correspond to the ground-truth object. We quantify explainability using an automated metric, and using human judgement. We propose explainability as a means for bridging the visual-semantic gap between different domains where model explanations are used as a means of disentagling domain specific information from otherwise relevant features. We demonstrate that this leads to improved generalization to new domains without hindering performance on the original domain.