Refine
Year of publication
Document Type
- Article (48) (remove)
Keywords
- Site effects (4)
- Surface waves and free oscillations (3)
- Wave propagation (3)
- Early warning (2)
- Interferometry (2)
- Inverse theory (2)
- Inversion (2)
- MASW (2)
- Rayleigh waves (2)
- Seismic attenuation (2)
Earthquake localization is both a necessity within the field of seismology, and a prerequisite for further analysis such as source studies and hazard assessment. Traditional localization methods often rely on manually picked phases. We present an alternative approach using deep learning that once trained can predict hypocenter locations efficiently. In seismology, neural networks have typically been trained with either single-station records or based on features that have been extracted previously from the waveforms. We use three-component full-waveform records of multiple stations directly. This means no information is lost during preprocessing and preparation of the data does not require expert knowledge. The first convolutional layer of our deep convolutional neural network (CNN) becomes sensitive to features that characterize the waveforms it is trained on. We show that this layer can therefore additionally be used as an event detector. As a test case, we trained our CNN using more than 2000 earthquake swarm events from West Bohemia, recorded by nine local three-component stations. The CNN successfully located 908 validation events with standard deviations of 56.4 m in east-west, 123.8 m in north-south, and 136.3 m in vertical direction compared to a double-difference relocated reference catalog. The detector is sensitive to events with magnitudes down to M-L = -0.8 with 3.5% false positive detections.
Advances in the field of seismic interferometry have provided a basic theoretical interpretation to the full spectrum of the microtremor horizontal-to-vertical spectral ratio [H/V(f)]. The interpretation has been applied to ambient seismic noise data recorded both at the surface and at depth. The new algorithm, based on the diffuse wavefield assumption, has been used in inversion schemes to estimate seismic wave velocity profiles that are useful input information for engineering and exploration seismology both for earthquake hazard estimation and to characterize surficial sediments. However, until now, the developed algorithms are only suitable for on land environments with no offshore consideration. Here, the microtremor H/V(z, f) modelling is extended for applications to marine sedimentary environments for a 1-D layered medium. The layer propagator matrix formulation is used for the computation of the required Green’s functions. Therefore, in the presence of a water layer on top, the propagator matrix for the uppermost layer is defined to account for the properties of the water column. As an application example we analyse eight simple canonical layered earth models. Frequencies ranging from 0.2 to 50 Hz are considered as they cover a broad wavelength interval and aid in practice to investigate subsurface structures in the depth range from a few meters to a few hundreds of meters. Results show a marginal variation of 8 per cent at most for the fundamental frequency when a water layer is present. The water layer leads to variations in H/V peak amplitude of up to 50 per cent atop the solid layers.
We apply and evaluate a recent machine learning method for the automatic classification of seismic waveforms. The method relies on Dynamic Bayesian Networks (DBN) and supervised learning to improve the detection capabilities at 3C seismic stations. A time-frequency decomposition provides the basis for the required signal characteristics we need in order to derive the features defining typical "signal" and "noise" patterns. Each pattern class is modeled by a DBN, specifying the interrelationships of the derived features in the time-frequency plane. Subsequently, the models are trained using previously labeled segments of seismic data. The DBN models can now be compared against in order to determine the likelihood of new incoming seismic waveform segments to be either signal or noise. As the noise characteristics of seismic stations varies smoothly in time (seasonal variation as well as anthropogenic influence), we accommodate in our approach for a continuous adaptation of the DBN model that is associated with the noise class. Given the difficulty for obtaining a golden standard for real data (ground truth) the proof of concept and evaluation is shown by conducting experiments based on 3C seismic data from the International Monitoring Stations, BOSA and LPAZ.
Volcanic eruptions are often preceded by seismic activity that can be used to quantify the volcanic activity. In order to allow consistent inference of the volcanic activity state from the observed seismicity patterns, objective and time-invariant classification results achievable by automatic systems should be preferred. Most automatic classification approaches need a large preclassified data set for training the system. However, in case of a volcanic crisis, we are often confronted with a lack of training data due to insufficient prior observations. In the worst case (e. g., volcanic crisis related reconfiguration of stations), there are even no prior observations available. Finally, due to the imminent crisis there might be no time for the time-consuming process of preparing a training data set. For this reason, we have developed a novel seismic-event spotting technique in order to be less dependent on previously acquired data bases and classification schemes. We are using a learning-while-recording approach based on a minimum number of reference waveforms, thus allowing for the build-up of a classification scheme as early as interesting events have been identified. First, short-term wave-field parameters (here, polarization and spectral attributes) are extracted from a continuous seismic data stream. The sequence of multidimensional feature vectors is then used to identify a fixed number of clusters in the feature space. Based on this general description of the overall wave field by a mixture of multivariate Gaussians, we are able to learn particular event classifiers (here, hidden Markov models) from a single waveform example. To show the capabilities of this new approach we apply the algorithm to a data set recorded at Soufriere Hills volcano, Montserrat. Supported by very high classification rates, we conclude that the suggested approach provides a valuable tool for volcano monitoring systems.
Other than commonly assumed in seismology, the phase velocity of Rayleigh waves is not necessarily a single-valued function of frequency. In fact, a single Rayleigh mode can exist with three different values of phase velocity at one frequency. We demonstrate this for the first higher mode on a realistic shallow seismic structure of a homogeneous layer of unconsolidated sediments on top of a half-space of solid rock (LOH). In the case of LOH a significant contrast to the half-space is required to produce the phenomenon. In a simpler structure of a homogeneous layer with fixed (rigid) bottom (LFB) the phenomenon exists for values of Poisson's ratio between 0.19 and 0.5 and is most pronounced for P-wave velocity being three times S-wave velocity (Poisson's ratio of 0.4375). A pavement-like structure (PAV) of two layers on top of a half-space produces the multivaluedness for the fundamental mode. Programs for the computation of synthetic dispersion curves are prone to trouble in such cases. Many of them use mode-follower algorithms which loose track of the dispersion curve and miss the multivalued section. We show results for well established programs. Their inability to properly handle these cases might be one reason why the phenomenon of multivaluedness went unnoticed in seismological Rayleigh wave research for so long. For the very same reason methods of dispersion analysis must fail if they imply wave number k(l)(omega) for the lth Rayleigh mode to be a single-valued function of frequency.. This applies in particular to deconvolution methods like phase-matched filters. We demonstrate that a slant-stack analysis fails in the multivalued section, while a Fourier-Bessel transformation captures the complete Rayleigh-wave signal. Waves of finite bandwidth in the multivalued section propagate with positive group-velocity and negative phase-velocity. Their eigenfunctions appear conventional and contain no conspicuous feature.
This study presents results of ambient noise measurements from temporary single station and small-scale array deployments in the northeast of Basle. H/V spectral ratios were determined along various profiles crossing the eastern masterfault of the Rhine Rift Valley and the adjacent sedimentary rift fills. The fundamental H/V peak frequencies are decreasing along the profile towards the eastern direction being consistent with the dip of the tertiary sediments within the rift. Using existing empirical relationships between H/V frequency peaks and the depth of the dominant seismic contrast, derived on basis of the lambda/4-resonance hypothesis and a power law depth dependence of the S-wave velocity, we obtain thicknesses of the rift fill from about 155 m in the west to 280 in in the east. This is in agreement with previous studies. The array analysis of the ambient noise wavefield yielded a stable dispersion relation consistent with Rayleigh wave propagation velocities. We conclude that a significant amount of surface waves is contained in the observed wavefield. The computed ellipticity for fundamental mode Rayleigh waves for the velocity depth models used for the estimation of the sediment thicknesses is in agreement with the observed H/V spectra over a large frequency band
The resonance frequency of the transmission response in layered half-space model is important in the study of site effect because it is the frequency where the shake-ability of the ground is enhanced significantly. In practice, it is often determined by the H/V ratio technique in which the peak frequency of recorded H/V spectral ratio is interpreted as the resonance frequency. Despite of its importance, there has not been any formula of the resonance frequency of the layered half-space structure. In this paper, a simple approximate formula of the fundamental resonance frequency is presented after an exact formula in explicit form of the response function of vertically SH incident wave is obtained. The formula is in similar form with the one used in H/V ratio technique but it reflects several major effects of the model to the resonance frequency such as the arrangement of layers, the impedance contrast between layers and the half-space. Therefore, it could be considered as an improved formula used in H/V ratio technique. The formula also reflects the consistency between two approaches of the H/V ratio technique based on SH body waves or Rayleigh surface waves on the peak frequency under high impedance contrast condition. This formula is in explicit form and, therefore, may be used in the direct and inverse problem efficiently. A numerical illustration of the improved formula for an actual layered half-space model already investigated by H/V ratio technique is presented to demonstrate its new features and its improvement to the currently used formula.
Dispersion-curve inversion of Rayleigh waves to infer subsurface shear-wave velocity is a long-standing problem in seismology. Due to nonlinearity and ill-posedness, sophisticated regularization techniques are required to solve the problem for a stable velocity model. We have formulated the problem as a minimization problem with nonlinear operator constraint and then solve it by using an inexact augmented Lagrangian method, taking advantage of the Haney-Tsai Dix-type relation (a global linear approximation of the nonlinear forward operator). This replaces the original regularized nonlinear problem with iterative minimization of a more tractable regularized linear problem followed by a nonlinear update of the phase velocity (data) in which the update can be performed accurately with any forward modeling engine, for example, the finite-element method. The algorithm allows discretizing the medium with thin layers (for the finite-element method) and thus omitting the layer thicknesses from the unknowns and also allows incorporating arbitrary regularizations to shape the desired velocity model. In this research, we use total variation regularization to retrieve the shear-wave velocity model. We use two synthetic and two real data examples to illustrate the performance of the inversion algorithm with total variation regularization. We find that the method is fast and stable, and it converges to the solution of the original nonlinear problem.
Avalanche activity is an important factor when estimating the regional avalanche danger. Moreover, a complete and detailed picture of avalanche activity is needed to understand the processes that lead to natural avalanche release. Currently, information on avalanche activity is mainly obtained through visual observations. However, this involves large uncertainties in the number and release times, influencing the subsequent analysis. Therefore, alternative methods for the remote detection of snow avalanches in particular in non-observed areas are highly desirable. In this study, we use the excited ground vibration to identify avalanches automatically. The specific seismic signature of avalanches facilitates the objective detection by a recently developed classification procedure. A probabilistic description of the signals, called hidden Markov models, allows the robust identification of corresponding signals in the continuous data stream. The procedure is based upon learning a general background model from continuous seismic data. Then, a single reference waveform is used to update an event-specific classifier. Thus, a minimum amount of training data is required by constructing such a classifier on the fly. In this study, we processed five days of continuous data recorded in the Swiss Alps during the avalanche winter 1999. With the restriction of testing large wet-snow avalanches only, the presented approach achieved very convincing results. We successfully detect avalanches over a large volume and distance range. Ninety-two percentage of all detections (43 out of 47) could be confirmed as avalanche events; only four false alarms are reported. We see a clear dependence of recognition capability on run-out distance and source-receiver distance of the observed events: Avalanches are detectable up to a source-receiver distance of eight times the avalanche length. Implications for analyzing a more comprehensive data set (smaller events and different flow regimes) are discussed in detail.
In this paper, we propose a method of surface waves characterization based on the deformation of the wavelet transform of the analysed signal. An estimate of the phase velocity (the group velocity) and the attenuation coefficient is carried out using a model-based approach to determine the propagation operator in the wavelet domain, which depends nonlinearly on a set of unknown parameters. These parameters explicitly define the phase velocity, the group velocity and the attenuation. Under the assumption that the difference between waveforms observed at a couple of stations is solely due to the dispersion characteristics and the intrinsic attenuation of the medium, we then seek to find the set of unknown parameters of this model. Finding the model parameters turns out to be that of an optimization problem, which is solved through the minimization of an appropriately defined cost function. We show that, unlike time-frequency methods that exploit only the square modulus of the transform, we can achieve a complete characterization of surface waves in a dispersive and attenuating medium. Using both synthetic examples and experimental data, we also show that it is in principle possible to separate different modes in both the time domain and the frequency domain
Nowadays, an increasing amount of seismic data is collected by daily observatory routines. The basic step for successfully analyzing those data is the correct detection of various event types. However, the visually scanning process is a time-consuming task. Applying standard techniques for detection like the STA/LTAtrigger still requires the manual control for classification. Here, we present a useful alternative. The incoming data stream is scanned automatically for events of interest. A stochastic classifier, called hidden Markov model, is learned for each class of interest enabling the recognition of highly variable waveforms. In contrast to other automatic techniques as neural networks or support vector machines the algorithm allows to start the classification from scratch as soon as interesting events are identified. Neither the tedious process of collecting training samples nor a time-consuming configuration of the classifier is required. An approach originally introduced for the volcanic task force action allows to learn classifier properties from a single waveform example and some hours of background recording. Besides a reduction of required workload this also enables to detect very rare events. Especially the latter feature provides a milestone point for the use of seismic devices in alpine warning systems. Furthermore, the system offers the opportunity to flag new signal classes that have not been defined before. We demonstrate the application of the classification system using a data set from the Swiss Seismological Survey achieving very high recognition rates. In detail we document all refinements of the classifier providing a step-by-step guide for the fast set up of a well-working classification system.
Shallowly situated evaporites in built-up areas are of relevance for urban and cultural development and hydrological regulation. The hazard of sinkholes, subrosion depressions and gypsum karst is often difficult to evaluate and may quickly change with anthropogenic influence. The geophysical exploration of evaporites in metropolitan areas is often not feasible with active industrial techniques. We collect and combine different passive geophysical data as microgravity, ambient vibrations, deformation and hydrological information to study the roof morphology of shallow evaporites beneath Hamburg, Northern Germany. The application of a novel gravity inversion technique leads to a 3-D depth model of the salt diapir under study. We compare the gravity-based depth model to pseudo-depths from H/V measurements and depth estimates from small-scale seismological array data. While the general range and trend of the diapir roof is consistent, a few anomalous regions are identified where H/V pseudo-depths indicate shallower structures not observed in gravity or array data. These are interpreted by shallow residual caprock floaters and zones of increased porosity. The shallow salt structure clearly correlates with a relative subsidence in the order of 2 mm yr(-1). The combined interpretation of roof morphology, yearly subsidence rates, chemical analyses of groundwater and of hydraulic head in aquifers indicates that the salt diapir beneath Hamburg is subject to significant ongoing dissolution that may possibly affect subrosion depressions, sinkhole distribution and land usage. The combined analysis of passive geophysical data may be exemplary for the study of shallow evaporites beneath other urban areas.
We compute seismic velocity profiles by a combined inversion of surface-wave phase-velocity dispersion curves together with the full spectrum of the microtremor horizontal-to-vertical (H/V) spectral ratio at two sediment-covered sites in Germany. The sediment deposits are approximately 100 m thick at the first test site and approximately 400 m thick at the second test site. We have used an extended physical model based on the diffuse wavefield assumption for the interpretation of the observed microtremor H/V spectral ratio. The extension includes the interpretation of the microtremor H/V spectral ratio observed at depth (in boreholes). This full-wavefield approach accounts for the energy contribution from the body and surface waves, and thus it allows for inverting the properties of the shallow subsurface. We have obtained the multimode phase velocity dispersion curves from an independent study, and a description of the extracted branches and their interpretation was developed. The inversion results indicate that the combined approach using seismic ambient noise and actively generated surface-wave data will improve the accuracy of the reconstructed near-surface velocity model, a key step in microzonation, geotechnical engineering, seismic statics corrections, and reservoir imaging.
Constructing a hidden Markov Model based earthquake detector: application to induced seismicity
(2012)
The triggering or detection of seismic events out of a continuous seismic data stream is one of the key issues of an automatic or semi-automatic seismic monitoring system. In the case of dense networks, either local or global, most of the implemented trigger algorithms are based on a large number of active stations. However, in the case of only few available stations or small events, for example, like in monitoring volcanoes or hydrothermal power plants, common triggers often show high false alarms. In such cases detection algorithms are of interest, which show reasonable performance when operating even on a single station. In this context, we apply Hidden Markov Models (HMM) which are algorithms borrowed from speech recognition. However, many pitfalls need to be avoided to apply speech recognition technology directly to earthquake detection. We show the fit of the model parameters in an innovative way. State clustering is introduced to refine the intrinsically assumed time dependency of the HMMs and we explain the effect coda has on the recognition results. The methodology is then used for the detection of anthropogenicly induced earthquakes for which we demonstrate for a period of 3.9 months of continuous data that the single station HMM earthquake detector can achieve similar detection rates as a common trigger in combination with coincidence sums over two stations. To show the general applicability of state clustering we apply the proposed method also to earthquake classification at Mt. Merapi volcano, Indonesia.
In this paper we present densely sampled fumarole temperature data, recorded continuously at a high-temperature fumarole of Mt. Merapi volcano (Indonesia). These temperature time series are correlated with continuous records of rainfall and seismic waveform data collected at the Indonesian - German multi-parameter monitoring network. The correlation analysis of fumarole temperature and precipitation data shows a clear influence of tropical rain events on fumarole temperature. In addition, there is some evidence that rainfall may influence seismicity rates, indicating interaction of meteoric water with the volcanic system. Knowledge about such interactions is important, as lava dome instabilities caused by heavy-precipitation events may result in pyroclastic flows. Apart from the strong external influences on fumarole temperature and seismicity rate, which may conceal smaller signals caused by volcanic degassing processes, the analysis of fumarole temperature and seismic data indicates a statistically significant correlation between a certain type of seismic activity and an increase in fumarole temperature. This certain type of seismic activity consists of a seismic cluster of several high-frequency transients and an ultra-long-period signal (< 0.002 Hz), which are best observed using a broadband seismometer deployed at a distance of 600 m from the active lava dome. The corresponding change in fumarole temperature starts a few minutes after the ultra-long-period signal and simultaneously with the high-frequency seismic cluster. The change in fumarole temperature, an increase of 5 degreesC on average, resembles a smoothed step. Fifty-four occurrences of simultaneous high-frequency seismic cluster, ultra-long period signal and increase of fumarole temperature have been identified in the data set from August 2000 to January 2001. The observed signals appear to correspond to degassing processes in the summit region of Mt. Merapi. (C) 2004 Elsevier B.V. All rights reserved
Ambient vibration techniques are promising methods for assessing the subsurface structure, in particular the shear-wave velocity profile (V-s). They are based on the dispersion property of surface waves in layered media. Therefore, the penetration depth is intrinsically linked to the energy content of the sources. For ambient vibrations, the spectral content extends in general to lower frequency when compared to classical artificial sources. Among available methods for processing recorded signals, we focus here on the spatial autocorrelation method. For stationary wavefields, the spatial autocorrelation is mathematically related to the frequency-dependent wave velocity c(omega). This allows the determination of the dispersion curve of traveling surface waves, which, in turn, is linked to the V-s profile. Here, we propose a direct inversion scheme for the observed autocorrelation curves to retrieve, in a single step, the V-s profile. The powerful neighborhood algorithm is used to efficiently search for all solutions in an n- dimensional parameter space. This approach has the advantage of taking into account the existing uncertainty over the measured curves, thus generating all V-s profiles that fit the data within their experimental errors. A preprocessing tool is also developed to estimate the validity of the autocorrelation curves and to reject parts of them if necessary before starting the inversion itself. We present two synthetic cases to test the potential of the method: one with ideal autocorrelation curves and another with autocorrelation curves computed from simulated ambient vibrations. The latter case is more realistic and makes it possible to figure out the problems that may be encountered in real experiments. The V-s profiles are correctly retrieved up to the depth of the first major velocity contrast unless low-velocity zones are accepted. We demonstrate that accepting low-velocity zones in the parameterization has a dramatic influence on the result of the inversion, with a considerable increase in the nonuniqueness of the problem. Finally, a real data set is processed with the same method
We present an approach for rapidly estimating full moment tensors of earthquakes and their parameter uncertainties based on short time windows of recorded seismic waveform data by considering deep learning of Bayesian Neural Networks (BNNs). The individual neural networks are trained on synthetic seismic waveform data and corresponding known earthquake moment-tensor parameters. A monitoring volume has been predefined to form a three-dimensional grid of locations and to train a BNN for each grid point. Variational inference on several of these networks allows us to consider several sources of error and how they affect the estimated full moment-tensor parameters and their uncertainties. In particular, we demonstrate how estimated parameter distributions are affected by uncertainties in the earthquake centroid location in space and time as well as in the assumed Earth structure model. We apply our approach as a proof of concept on seismic waveform recordings of aftershocks of the Ridgecrest 2019 earthquake with moment magnitudes ranging from Mw 2.7 to Mw 5.5. Overall, good agreement has been achieved between inferred parameter ensembles and independently estimated parameters using classical methods. Our developed approach is fast and robust, and therefore, suitable for down-stream analyses that need rapid estimates of the source mechanism for a large number of earthquakes.
We develop an amplitude spectral ratio method for event couples from clustered earthquakes to estimate seismic wave attenuation (Q-1) in the source volume. The method allows to study attenuation within the source region of earthquake swarms or aftershocks at depth, independent of wave path and attenuation between source region and surface station. We exploit the high-frequency slope of phase spectra using multitaper spectral estimates. The method is tested using simulated full wave-field seismograms affected by recorded noise and finite source rupture. The synthetic tests verify the approach and show that solutions are independent of focal mechanisms but also show that seismic noise may broaden the scatter of results. We apply the event couple spectral ratio method to northwest Bohemia, Czech Republic, a region characterized by the persistent occurrence of earthquake swarms in a confined source region at mid-crustal depth. Our method indicates a strong anomaly of high attenuation in the source region of the swarm with an averaged attenuation factor of Qp < 100. The application to S phases fails due to scattered P-phase energy interfering with S phases. The Qp anomaly supports the common hypothesis of highly fractured and fluid saturated rocks in the source region of the swarms in northwest Bohemia. However, high temperatures in a small volume around the swarms cannot be excluded to explain our observations.
In this article, we address the question of how observed ground-motion data can most effectively be modeled for engineering seismological purposes. Toward this goal, we use a data-driven method, based on a deep-learning autoencoder with a variable number of nodes in the bottleneck layer, to determine how many parameters are needed to reconstruct synthetic and observed ground-motion data in terms of their median values and scatter. The reconstruction error as a function of the number of nodes in the bottleneck is used as an indicator of the underlying dimensionality of ground-motion data, that is, the minimum number of predictor variables needed in a ground-motion model. Two synthetic and one observed datasets are studied to prove the performance of the proposed method. We find that mapping ground-motion data to a 2D manifold primarily captures magnitude and distance information and is suited for an approximate data reconstruction. The data reconstruction improves with an increasing number of bottleneck nodes of up to three and four, but it saturates if more nodes are added to the bottleneck.