Abstract: Full three-dimensional tomography (F3DT) is a computationally intensive methodology for imaging the Earth structure through the iterated assimilation of waveform data into 3D models of seismic wave propagation. F3DT performance is strongly determined by data resolution, often limited by the ability to identifying coherent seismic phases, measuring the phase and amplitude spectra, and modeling the effects of interference from other signals. For seismograms computed for a 1D Earth model, the decomposition into standard seismic phases is relatively straightforward, e.g., numerical synthesis of traveling modes or generalized rays, but these classical techniques are of limited utility for structures with strong 3D heterogeneity. We present a new technique for systematically separating seismic phases using time-frequency spectra computed by the S-transform (Stockwell et al., 1996), which is a linear transformation based on Gaussian wavelets. Iterative waveform stripping decomposes seismograms in the spectral domain into a finite set of wave packets using time-frequency filters centered in the spectrum local maxima. Subsets of waveforms are then combined into waveforms that can be used in seismic source and structural inversion. Weights for the combinations are estimated by the Sum-of-Wavelets Theorem (Gee & Jordan, 1992). This algorithm allows the identification of seismic phases that are not predicted by a 1D structural model, such as basin-edge conversions, increasing the structural information that can be derived from a single seismogram. We demonstrate the flexibility of the algorithm to analyze large data sets by showing examples of seismogram decomposition and tomographic kernels computed from real data from earthquakes in Southern California.
Media has not been submitted for this Presentation
Abstract: The purpose of this work is to determine the quality of the information obtained from the seismological stations that constitute the Broadband Network of the Mexican Servicio Sismológico Nacional (National Seismological Service, SSN). We estimate the orientation of the seismometer installed at each site of the Broadband Network for different epochs; each epoch is defined by changes in digitizer or the seismometer. The correct orientation of the sensors is relevant to many seismological studies that involve rotation or require exact information regarding the orientation of the components. However, orienting a triaxial sensor is not an easy task, even for an experienced field engineer. Several factors can affect it during the installation of a seismometer. Between 2015 and 2017, all the seismometers were reoriented using a gyroscope, 62% of them were misoriented by more than 5°; however, the majority of those were less than 15°. This misorientation corresponds to the last instrumental epoch. To estimate the seismometer orientation during previous periods, we use two methods. The first one is based on the polarization of Rayleigh surface waves (Chael, 1997; Selby, 2001). The second uses a principal component analysis, a linear technique widely used in different branches of physics (Walck and Chael, 1991). The estimations obtained with both methods for the last epoch show high correlation with the measurements with the electronic gyroscope, validating the orientation estimation for the earlier instrumental epochs. Furthermore, we show the effect of sensor misorientation in earthquake location and estimation of receiver functions.
Media has not been submitted for this Presentation
Abstract: The 2016 Kaikoura, New Zealand, M7.8 earthquake involved complex ruptures on more than a dozen faults, and the first-order pictures depicted by regional and teleseismic data are still in discrepancy. Most models dominated by far-field data show mainly slip on the megathrust interface, while models dominated by near-field data prefer mostly crustal faulting with secondary interpolate contribution. One challenge for perhaps all finite-fault inversions of this earthquake is that assumptions must be made about where and when rupture starts on a particular fault. The large number of free parameters makes it difficult to assess the effects of these assumptions on the finite-fault models. Here we characterize the Kaikoura earthquake as a sequence of several subevents, each of which is a Haskell source with unilateral and constant rupture speed. With relative few free parameters, we conduct a non-linear search of the locations and timings of subevents to allow different rupture scenarios and to assess the model uncertainties. In the preliminary result, we find that the large energy release between 50s and 80s during the event consists of a deep thrust subevent and a shallow strike-slip subevent. As for these two major subevents, the thrust subevent is systematically earlier than the strike-slip subevent. This appears to favor that the megathrust slip is driving the shallow crustal faulting, instead of ruptures jumping from one crustal fault to another.
Media has not been submitted for this Presentation
Abstract: Recently published studies of the seismic attenuation (Q) boundary and stress parameter estimates for the Gulf Coast and southeastern U.S. (SEUS) improve our understanding of seismic hazard for the SEUS. First, the improved Gulf Coast Q boundary location indicates a more restricted region of high attenuation in the SEUS than previous studies that were based on limited Q observations, seismotectonic regionalization, and crustal structure regionalization. Second, for earthquakes outside the Gulf Coast region and at long periods, the high attenuation region appears to be more restricted to close to the Gulf Coast. Third, most of Florida has less attenuation than the Gulf Coast region and is more like the central and eastern U.S. (CEUS) mid-continental attenuation, in keeping with the differing geological history of the Florida Peninsula from the Gulf Coast. Fourth, stress parameter, which is related to the high frequency energy and ground motion expected for CEUS earthquakes, appears to be lower in the south central U.S. than the east coast, northeastern U.S., and eastern Canada, but still significantly higher than for the western U.S. And fifth, induced earthquakes in the CEUS appear to have similar stress parameter levels as shallow natural events. Thus seismic hazard from induced earthquakes is the same as for shallow natural earthquakes in a given region. In general seismic ground motions and hence hazard from large CEUS earthquakes is not reduced in Alabama, Georgia, and Florida due to Gulf Coast Q. At long periods for large earthquakes outside the Gulf Coast Q region, lower hazard from Gulf Coast Q is restricted to near the Gulf Coast in Texas, Louisiana, and Mississippi. This pattern of seismic hazard is reflected in the intensity observations from the 1886 Charleston, South Carolina, M7.0 earthquake.
Media has not been submitted for this Presentation
Abstract: The method widely used for the simulation of seismic wave propagation in three dimensional sedimentary basin is the finite element method combining with viscous-elastic artificial boundary. However, in the finite element method, the boundary surfaces are often limited in regular cases, such as a horizontal free surface, to determine the input wave motion on the artificial boundary. A degenerate model method is proposed in this study for determining the input wave motion on artificial boundary in three dimensional finite element model, in which the solution of the free fields outside and on the lateral artificial boundaries in three dimensional model is given based on four two-dimensional models, and the solution of the free fields outside and on the lateral artificial boundaries for each two-dimensional model is given based on two one-dimensional models. Furthermore, an explicit finite element method with viscous-elastic artificial boundary is established for the simulation of seismic wave propagation in three dimensional site. The numerical results obtained from a free half space model indicates that the proposed input wave motion method is reasonable. In the Mw 7.9 Wenchuan earthquake on 12 May 2008, the Wudu township which is located in a sedimentary basin suffered serious damage even if it is far away from the fault rupture (the distance >100km). In order to interpret the abnormality of damage phenomenon, the elastic wave field simulation in the Wudu Basin is conducted by the proposed method based on the ABAQUS software. The simulation results reveals the basin edge and the basin focusing effect. The topography scatters the body waves and the surface wave generated is trapped at the shallow part of the basin, and most of the energy is reflected from the interfaces of soil strata and focused back into the basin. It results in extraordinary strong shaking patterns in Wudu basin area.
Media has not been submitted for this Presentation
Abstract: Cotopaxi is an active stratovolcano located in Ecuador, which has previously presented 5 eruptive periods since 1533. The aim of this work is to develop an automatic recognition system based on machine learning techniques for both detection and classification of seismic events from Cotopaxi volcano. Our primary goal is an automatic recognition system that may help experts to make more appropriate decisions towards a real time system, which may allow authorities to launch effective early warnings. This work is divided in two stages: in the detection stage, a non-overlapping segmentation was used, with 15-sec windows; in each window several features were extracted in the time, frequency, and scale domains by using methods such as moving average, power spectral density, and wavelet transforms, 84 features were extracted: 13 in the time domain, 21 in the frequency domain and 50 in the scale domain; this stage allows to identify the starting and ending point of the microseism; by using the Moving Average method, a threshold equal to 0.2 of the normalized signal was identified for the detector to allow the identification of the existence of a microseisms. The best result was obtained by using Support Vector Machine techniques, it presents a performance metrics with a Nu = 0.5 obtaining 99.72% accuracy with a BER of 0.0014.The following Machine Learning techniques were used for classification of seismic events: Decision Trees (DT), Linear and Non-linear Support Vector Machines (SVM). An analysis of the BER was performed, varying the percentage of the training and test matrices for the different Machine Learning techniques. This variation improves the performance measures. The system was trained to have the ability to identify Long Period (LP) and Volcano-Tectonic (VT) events; the best result (BER equal to 0.1125 with a 94.29% accuracy) were obtained with non-linear SVM.
Media has not been submitted for this Presentation
