In this report, we describe how to estimate the time-variable components of the seismic moment tensor and compare these estimates to the more conventional analysis that incorporates an assumption of the source time function (STF) across all components of the seismic moment tensor. The advantage of our method is that we are able to independently estimate the time-evolution of each component of the seismic moment tensor, which may help to resolve the complex source phenomena associated with buried explosions. By performing an eigen decomposition of the time-evolving seismic moment tensor components, we are able to plot the seismic mechanism as a trajectory on a lune diagram. This technique enables interpretation of the seismic mechanism as a function of time, as opposed to the more conventional analysis which assumes that the seismic mechanism is time invariant. Finally, we describe the differences between the seismic moment and the seismic moment rate STFs, how to implement each one in inversion schemes, and the relative strengths/weaknesses of each. Our key take-away is that we are able to distinguish nearly-overlapping sources with highly different mechanisms, such as an explosion immediately following an earthquake, by estimating moment rate from seismic data through a STF-invariant inversion for the full time-variable moment tensor.
In this report, we assess the data recorded by a Distributed Acoustic Sensing (DAS) cable deployed during the Source Physics Experiment, Phase II (DAG) in comparison with the data recorded by nearby 4.5-Hz geophones. DAS is a novel recording method with unprecedented spatial resolution, but there are significant concerns around the data fidelity as the technology is ramped up to more common usage. Here we run a series of tests to quantify the similarity between DAS data and more conventional data and investigate cases where the higher spatial resolution of the DAS can provide new insights into the wavefield. These tests include 1D modeling with seismic refraction and bootstrap uncertainties, assessing the amplitude spectra with distance from the source, measuring the frequency dependent inter-station coherency, estimating time-dependent phase velocity with beamforming and semblance, and measuring the cross-correlation between the geophone and the particle velocity inferred from the DAS. In most cases, we find high similarity between the two datasets, but the higher spatial resolution of the DAS provides increased details and methods of estimating uncertainty.
This report details a method to estimate the energy content of various types of seismic body waves. The method is based on the strain energy of an elastic wavefield and Hooke’s Law. We present a detailed derivation of a set of equations that explicitly partition the seismic strain energy into two parts: one for compressional (P) waves and one for shear (S) waves. We posit that the ratio of these two quantities can be used to determine the relative contribution of seismic P and S waves, possibly as a method to discriminate between earthquakes and buried explosions. We demonstrate the efficacy of our method by using it to compute the strain energy of synthetic seismograms with differing source characteristics. Specifically, we find that explosion-generated seismograms contain a preponderance of P wave strain energy when compared to earthquake-generated synthetic seismograms. Conversely, earthquake-generated synthetic seismograms contain a much greater degree of S wave strain energy when compared to explosion-generated seismograms.
In this report, we assess the data recorded by a Distributed Acoustic Sensing (DAS) cable deployed during the Source Physics Experiment, Phase II (DAG) in comparison with the data recorded by nearby 4.5-Hz geophones. DAS is a novel recording method with unprecedented spatial resolution, but there are significant concerns around the data fidelity as the technology is ramped up to more common usage. Here we run a series of tests to quantify the similarity between DAS data and more conventional data and investigate cases where the higher spatial resolution of the DAS can provide new insights into the wavefield. These tests include 1D modeling with seismic refraction and bootstrap uncertainties, assessing the amplitude spectra with distance from the source, measuring the frequency dependent inter-station coherency, estimating time-dependent phase velocity with beamforming and semblance, and measuring the cross-correlation between the geophone and the particle velocity inferred from the DAS. In most cases, we find high similarity between the two datasets, but the higher spatial resolution of the DAS provides increased details and methods of estimating uncertainty.
Seismic source modeling allows researchers both to simulate how a source that induces seismic waves interacts with the Earth to produce observed seismograms and, inversely, to infer what the time histories, sizes, and force distributions were for a seismic source given observed seismograms. In this report, we discuss improvements made in FY21 to our software as applies to both the forward and inverse seismic source modeling problems. For the forward portion of the problem, we have added the ability to use full 3-D nonlinear simulations by implementing 3-D time varying boundary conditions within Sandia’s linear seismic code Parelasti. Secondly, on the inverse source modeling side, we have developed software that allows us to invert seismic gradiometer-derived observations in conjunction with standard translational motion seismic data to infer properties of the source that may improve characterization in certain circumstances. First, we describe the basic theory behind each software enhancement and then demonstrate the software in action with some simple examples.
This report summarizes work completed under the Laboratory Directed Research and Development (LDRD) project "Uncertainty Quantification of Geophysical Inversion Using Stochastic Differential Equations." Geophysical inversions often require computationally expensive algorithms to find even one solution, let alone propagating uncertainties through to the solution domain. The primary purpose of this project was to find more computationally efficient means to approximate solution uncertainty in geophysical inversions. We found multiple computationally efficient methods of propagating Earth model uncertainty into uncertainties in solutions of full waveform seismic moment tensor inversions. However, the optimum method of approximating the uncertainty in these seismic source solutions was to use the Karhunen-Love theorem with data misfit residuals. This method was orders of magnitude more computationally efficient than traditional Monte Carlo methods and yielded estimates of uncertainty that closely approximated those of Monte Carlo. We will summarize the various methods we evaluated for estimating uncertainty in seismic source inversions as well as work toward this goal in the realm of 3-D seismic tomographic inversion uncertainty.
This final report summarizes the work completed under the Laboratory Directed Research and Development (LDRD) project “Seismic Spatial Gradients as a Machine Learning-Based Classifier for Explosion Monitoring.” The overarching goal of the project was to explore the efficacy of using machine learning-based classification algorithms where the input data are the spatial gradient of the seismic wavefield collected at a single point on the Earth’s surface. The methods that I describe here are in direct contrast to conventional methods of seismic discrimination which typically rely on a spatially extended network of instruments and physics-based wavefield attributes such as, for example, the ratio between $\textit{P}$ and $\textit{S}$ waves. Rather, we use the spatial gradient of the seismic wavefield observed at a single point on the Earth’s surface and data processing approaches inspired by the machine learning community. We tested two algorithms, a neural network and a modified version of principal component analysis termed Spectrally Filtered Principal Component Analysis (SFPCA). To test these algorithms, we first conducted a series of numerical tests using synthetic data and then conducted a small-scale controlled field experiment. The tests using synthetic data showed that both algorithms had high success rates on gradiometric data, even when simulated noise was added to the signal. Furthermore, we found that using seismic spatial gradients increased the performance of our discrimination algorithms when compared to using just the traditional translational motion seismic data. The tests with field data also showed a high degree of discriminative success.
Underground explosions nonlinearly deform the surrounding earth material and can interact with the free surface to produce spall. However, at typical seismological observation distances the seismic wavefield can be accurately modeled using linear approximations. Although nonlinear algorithms can accurately simulate very near field ground motions, they are computationally expensive and potentially unnecessary for far field wave simulations. Conversely, linearized seismic wave propagation codes are orders of magnitude faster computationally and can accurately simulate the wavefield out to typical observational distances. Thus, devising a means of approximating a nonlinear source in terms of a linear equivalent source would be advantageous both for scenario modeling and for interpretation of seismic source models that are based on linear, far-field approximations. This allows fast linear seismic modeling that still incorporates many features of the nonlinear source mechanics built into the simulation results so that one can have many of the advantages of both types of simulations without the computational cost of the nonlinear computation. In this report we first show the computational advantage of using linear equivalent models, and then discuss how the near-source (within the nonlinear wavefield regime) environment affects linear source equivalents and how well we can fit seismic wavefields derived from nonlinear sources.
We present a computationally efficient method to approximate the probability distribution of seismic Green's functions given the uncertainty of an Earth model. The method is based on the Karhunen-Loève (KL) theorem and an approximation of the Green's function (or seismogram) covariance. Using Monte Carlo (MC) simulations as a control case, we demonstrate that our KL-based method can accurately reproduce a probability distribution of seismograms that results from an uncertain Earth model for a MC-derived seismogram covariance. We then describe a method to estimate the covariance of the seismograms resulting from those Earth models that is not based on MC simulations. We use the estimated Green's function covariance in conjunction with our KL-based method to produce a Green's function probability distribution, and compare that distribution to a Green's function probability distribution produced using a MC finite difference method. We find that the Green's function probability distribution approximated using our KL-based method generally mimics that produced using the MC simulations, especially for direct-arriving body waves. However the accuracy of the KL-based method generally decreases for later times in the simulated Green's function distribution.
We present preliminary work on propagating model uncertainty into the estimation of the time domain source time functions of the seismic source. Our method is based on an estimated model covariance function, which we estimate from the data. The model covariance function is then used to construct a suite of surrogate Greens functions which we use in a Monte Carlo type inversion scheme. The result is a probability density function of the six independent source time functions, each of which corresponds to an individual component of the seismic moment tensor. We compare the results of our method with those obtained using a computationally expensive finite difference Monte Carlo method and find that our new method produces results that are deficient in low frequencies. The advantage of our new method, which we term the Karhunen-Loeve Monte Carlo (KLMC) method, is that is several orders of magnitude faster than our current method, which uses a finite difference scheme to produce the suite of forward models.
We present a new method to discriminate between earthquakes and buried explosions using observed seismic data. The method is different from previous seismic discrimination algorithms in two main ways. First, we use seismic spatial gradients, as well as the wave attributes estimated from them (referred to as gradiometric attributes), rather than the conventional three-component seismograms recorded on a distributed array. The primary advantage of this is that a gradiometer is only a fraction of a wavelength in aperture com¬pared with a conventional seismic array or network. Second, we use the gradiometric attributes as input data into a machine learning algorithm. The resulting discrimination algorithm uses the norms of truncated principal components obtained from the gradio- metric data to distinguish the two classes of seismic events. Using high-fidelity synthetic data, we show that the data and gradiometric attributes recorded by a single seismic gra¬diometer performs as well as a conventional distributed array at the event type discrimi¬nation task.
This report describes a proof-of-concept method of seismic source discrimination using seismic gradiometry and a common machine learning technique. The tests described here are purely numerical, using synthetic seismic data and well understood mathematical techniques. The primary innovation described here is the application of a richer seismic data set derived from seismic gradiometry. Seismic gradiometry is a method to estimate the time variable spatial gradient of the wavefield to compute various wavefield attributes such as slowness, dynamic strain, and rotational motions. With the addition of these wavefield attributes, we are afforded up to twenty "compo- nents" of time series data measured at a single point on, or in, the Earth. This is in direct contrast to conventional three-component seismic data collected at several locations using a seismic network. Using the gradiometrically-derived wavefield components directly in a single-layer neural network, I show that it is possible to discriminate between three common seismic source types (earthquakes, explosions, and opening fractures) for various noise conditions and gradiometry configurations.
As a part of the series of Source Physics Experiments (SPE) conducted on the Nevada National Security Site in southern Nevada, we have developed a local-to-regional scale seismic velocity model of the site and surrounding area. Accurate earth models are critical for modeling sources like the SPE to investigate the role of earth structure on the propagation and scattering of seismic waves. We combine seismic body waves, surface waves, and gravity data in a joint inversion procedure to solve for the optimal 3D seismic compres-sional and shear-wave velocity structures and earthquake locations subject to model smoothness constraints. Earthquakes, which are relocated as part of the inversion, provide P-and S-body-wave absolute and differential travel times. Active source experiments in the region augment this dataset with P-body-wave absolute times and surface-wave dispersion data. Dense ground-based gravity observations and surface-wave dispersion derived from ambient noise in the region fill in many areas where body-wave data are sparse. In general, the top 1–2 km of the surface is relatively poorly sampled by the body waves alone. However, the addition of gravity and surface waves to the body-wave data-set greatly enhances structural resolvability in the near surface. We discuss the method-ology we developed for simultaneous inversion of these disparate data types and briefly describe results of the inversion in the context of previous work in the region.
We document azimuthally dependent seismic scattering at the Source Physics Experiment (SPE) using the large-N array. The large-N array recorded the seismic wavefield produced by the SPE-5 buried chemical explosion, which occurred in April 2016 at the Nevada National Security Site, U.S.A. By selecting a subset of vertical-component geophones from the large-N array, we formed 10 linear arrays, with different nominal source-receiver azimuths as well as six 2D arrays. For each linear array, we evaluate wavefield coherency as a function of frequency and interstation distance. For both the P arrival and post-P arrivals, the coherency is higher in the northeast propagation direction, which is consistent with the strike of the steeply dipping Boundary fault adjacent to the northwest side of the large-N array. Conventional array analysis using a suite of 2D arrays suggests that the presence of the fault may help explain the azimuthal dependence of the seismic-wave coherency for all wave types. This fault, which separates granite from alluvium, may be acting as a vertically oriented refractor and/or waveguide.
We invert far-field infrasound data for the equivalent seismoacoustic timedomain moment tensor to assess the effects of variable atmospheric models and source phenomena. The infrasound data were produced by a series of underground chemical explosions that were conducted during the Source Physics Experiment (SPE), which was originally designed to study seismoacoustic signal phenomena. The first goal of this work is to investigate the sensitivity of the inversion to the variability of the estimated atmospheric model. The second goal is to determine the relative contribution of two presumed source mechanisms to the observed infrasonic wavefield. Rather than using actual atmospheric observations to estimate the necessary atmospheric Green’s functions, we build a series of atmospheric models that rely on publicly available, regional-scale atmospheric observations. The atmospheric observations are summarized and interpolated onto a 3D grid to produce a model of sound speed at the time of the experiment. For each of four SPE acoustic datasets that we invert, we produced a suite of three atmospheric models for each chemical explosion event, based on 10 yrs of meteorological data: an average model, which averages the atmospheric conditions for 10 yrs prior to each SPE event, as well as two extrema models. To parameterize the inversion, we assume that the source of infrasonic energy results from the linear combination of explosion-induced surface spall and linear seismic-to-elastic mode conversion at the Earth’s free surface. We find that the inversion yields relatively repeatable results for the estimated spall source. Conversely, the estimated isotropic explosion source is highly variable. This suggests that 1) the majority of the observed acoustic energy is produced by the spall and/or 2) our modeling of the elastic energy, and the subsequent conversion to acoustic energy, is too simplistic.
This document serves to guide a researcher through the process of running the Weather Research and Forecasting (WRF) model and incorporating observations into coarse resolution reanalysis products to model atmospheric conditions at high (50 m) resolution. This documentation is specific to WRF and the WRF Preprocessing System (WPS) version 3.8.1 and the Objective Analysis (OBSGRID) code released on April 8, 2016. Output from WRF serves as an input into the Time-Domain Atmospheric Acoustic Propagation Suite (TDAAPS) which performs staggered-grid finite difference modeling of the acoustic velocity pressure system to produce Green's functions through these atmospheric models.
We invert far field infrasound data for the equivalent seismo-acoustic time domain moment tensor to assess the effects of variable atmospheric models as well as to quantify the relative contributions of two presumed source phenomena. The infrasound data was produced by a series of underground chemical explosions that were conducted during the Source Physics Experiment, (SPE) which was originally designed to study explosion-generated seismo-acoustic signal phenomena. The goal of the work presented herein is two-fold: the first goal is to investigate the sensitivity of the estimated time domain moment tensors to variability of the estimated atmospheric model. The second goal is to determine the relative contribution of two possible source mechanisms to the observed in- frasonic wave field. Rather than using actual atmospheric observations to estimate the necessary atmospheric Green's functions, we build a series of atmospheric models that rely on publicly avail- able, regional atmospheric observations and the assumption that the acoustic energy results from a linear combination of an underground isotropic explosion and surface spall. The atmospheric observations are summarized and interpolated onto a 3D grid to produce a model of sound speed at the time of the experiment. For each of four SPE acoustic datasets that we invert, we produced a suite of three atmospheric models, based on ten years of regional meteorological observations: an average model, which averages the atmospheric conditions for ten years prior to each SPE event, as well as two extrema models. We find that the inversion yields relatively repeatable results for the estimated spall source. Conversely, the estimated isotropic explosion source is highly variable. This suggests that the majority of the observed acoustic energy is produced by the spall source and/or our modeling of the elastic energy propagation, and it's subsequent conversion to acoustic energy via linear elastic-to-acoustic coupling at the free surface, is too simplistic.
Resolving the time dependent terms in the seismic moment tensor provides important informa- tion that can be used to interpret the source process of an explosion, including the separation of isotropic explosion terms from shear forces and potentially isolated force couples. In this report, we detail our method of inverting three component seismic data for the seismic moment tensor. We review possible seismic source models from the simplest isotropic explosion type source to those incorporating the six independent moment tensor terms. The inversion we describe is formulated in the frequency domain, and results in estimates of time dependent moment tensor components. The inversion relies on an accurate estimate of the Green's functions of the Earth. However, given the complexity of the Earth, we explore the effects of inaccuracies in the presumed Earth model used to estimate the Green's functions needed for the inversion. Specifically, we explore the effects of stochastic variations in the Earth models on the inversion results. These tests are syn- thetic throughout, and show that adding stochastic density/velocity heterogeneity in the presumed Earth model results in reduced amplitude seismic moment tensor estimates, as well as degrading the data misfit. We suggest two mitigation strategies. First, produce a suite of Green's functions using different realizations of the stochastic field within the Earth Model. Secondly, perform the in- version in the power spectral domain, eliminating all phase information. Finally, we analyze actual seismic data collected in winter 2017/2018. The seismic data was collected at in active geothermal well site outside of Winnimucca, NV, and was produced during well stimulation operations. In general, the inversion results were poor, with a high degree of data misfit. We hypothesize that the poor results are a function of a poorly constrained Earth model as well as noisy, high-frequency data being used in the inversion.
This report shows the results of constructing predictive atmospheric models for the Source Physics Experiments 1-6. Historic atmospheric data are combined with topography to construct an atmo- spheric model that corresponds to the predicted (or actual) time of a given SPE event. The models are ultimately used to construct atmospheric Green's functions to be used for subsequent analysis. We present three atmospheric models for each SPE event: an average model based on ten one- hour snap shots of the atmosphere and two extrema models corresponding to the warmest, coolest, windiest, etc. atmospheric snap shots. The atmospheric snap shots consist of wind, temperature, and pressure profiles of the atmosphere for a one-hour time window centered at the time of the predicted SPE event, as well as nine additional snap shots for each of the nine preceding years, centered at the time and day of the SPE event.
This report describes a new single-station analysis method to estimate the dispersion and uncer- tainty of seismic surface waves using the multiwavelet transform. Typically, when estimating the dispersion of a surface wave using only a single seismic station, the seismogram is decomposed into a series of narrow-band realizations using a bank of narrow-band filters. By then enveloping and normalizing the filtered seismograms and identifying the maximum power as a function of frequency, the group velocity can be estimated if the source-receiver distance is known. However, using the filter bank method, there is no robust way to estimate uncertainty. In this report, I in- troduce a new method of estimating the group velocity that includes an estimate of uncertainty. The method is similar to the conventional filter bank method, but uses a class of functions, called Slepian wavelets, to compute a series of wavelet transforms of the data. Each wavelet transform is mathematically similar to a filter bank, however, the time-frequency tradeoff is optimized. By taking multiple wavelet transforms, I form a population of dispersion estimates from which stan- dard statistical methods can be used to estimate uncertainty. I demonstrate the utility of this new method by applying it to synthetic data as well as ambient-noise surface-wave cross-correlelograms recorded by the University of Nevada Seismic Network.