Please use this identifier to cite or link to this item:
Authors: Sick, Benjamin
Title: Temporal and spectral pattern recognition for detection and combined network and array waveform coherence analysis for location of seismic events
Issue Date: 2017 Dissertation 263
Abstract: The reliable automatic detection, location and classification of seismic events still poses great challenges if only few sensors record an event and/or the signal-to-noise ratio is very low. This study first examines, compares and evaluates the most widely used algorithms for automatic processing on a diverse set of seismic datasets (e.g. from induced seismicity and nuclear-test-ban verification experiments). A synthesis of state-of-the-art algorithms is given. Several single station event detection and phase picking algorithms are tested followed by a comparison of single station waveform cross-correlation and spectral pattern recognition. Coincidence analysis is investigated afterwards to demonstrate up to which level false alarms can be ruled out in sensor networks of multiple stations. It is then shown how the use of seismic (mini) arrays in diverse configurations can improve these results considerably through the use of waveform coherence. In a second step, two concepts are presented which combine the previously analysed algorithmic building blocks in a new way. The first concept is seismic event signal clustering by unsupervised learning which allows event identification with only one sensor. The study serves as a base level investigation to explore the limits of elementary seismic monitoring with only one single vertical-component seismic sensor and shows the level of information which can be extracted from a single station. It is investigated how single station event signal similarity clusters relate to geographic hypocenter regions and common source processes. Typical applications arise in local seismic networks where reliable ground truth by a dense temporal network precedes or follows a sparse (permanent) installation. The test dataset comprises a three-month subset from a field campaign to map subduction below northern Chile, project for the seismological investigation of the western cordillera (PISCO). Due to favourable ground noise conditions in the Atacama desert, the dataset contains an abundance of shallow and deep earthquakes, and many quarry explosions. Often event signatures overlap, posing a challenge to any signal processing scheme. Pattern recognition must work on reduced seismograms to restrict parameter space. Continuous parameter extraction based on noise-adapted spectrograms was chosen instead of discrete representation by, e.g. amplitudes, onset times, or spectral ratios to ensure consideration of potentially hidden features. Visualization of the derived feature vectors for human inspection and template matching algorithms was hereby possible. Because event classes shall comprise earthquake regions regardless of magnitude, signal clustering based on amplitudes is prevented by proper normalization of feature vectors. Principal component analysis (PCA) is applied to further reduce the number of features used to train a self-organizing map (SOM). The SOM arranges prototypes of each event class in a 2D map topologically. Overcoming the restrictions of this black-box approach, the arranged prototypes can be transformed back to spectrograms to allow for visualization and interpretation of event classes. The final step relates prototypes to ground-truth information, confirming the potential of automated, coarse-grain hypocenter clustering based on single station seismograms. The approach was tested by a two-fold cross-validation whereby multiple sets of feature vectors from half the events are compared by a one-nearest neighbour classifier in combination with an euclidean distance measure resulting in an overall correct geographic separation rate of 95.1% for coarse clusters and 80.5% for finer clusters (86.3% for a more central station). The second concept shows a new method to combine seismic networks of single stations and arrays for automatic seismic event location. After exploring capabilities of single station algorithms in the section before, this section explores capabilities of algorithms for small local seismic networks. Especially traffic light systems for induced seismicity monitoring rely on the real-time automated location of weak events. These events suffer from low signal-to-noise ratios and noise spikes due to the industrial setting. Conventional location methods rely on independent picking of first arrivals from seismic wave onsets at recordings of single stations. Picking is done separately and without feedback from the actual location algorithm. With low signal-to-noise ratios and local events, the association of onsets gets error prone, especially for S-phase onsets which are overlaid by coda from previous phases. If the recording network is small or only few phases can be associated, single wrong associations can lead to large errors in hypocenter locations and magnitude. Event location by source scanning which was established in the last two decades can provide more robust results. Source scanning uses maxima from a travel time corrected stack of a characteristic function of the full waveforms on a predefined location grid. This study investigates how source-scanning can be extended and improved by integrating information from seismic arrays, i.e. waveform stacking and Fisher ratio. These array methods rely on the coherency of the raw filtered waveforms while traditional source scanning uses a characteristic function to obtain coherency from otherwise incoherent waveforms between distant stations. The short term average to long term average ratio (STA/LTA) serves as the characteristic function and single station vertical-component traces for P-phases and radial and transverse components for S-phases are used. For array stations, the STA/LTA of the stacked vertical seismogram which is furthermore weighted by the STA/LTA of the Fisher ratio, dependent on back azimuth and slowness, is utilized for P-phases. In the chosen example, the extension by array-processing techniques can reduce the mean error in comparison to manually determined hypocenters by up to a factor of 2.9, resolve ambiguities and further restrain the location.
Appears in Collections:02 Fakultät Bau- und Umweltingenieurwissenschaften

Files in This Item:
File Description SizeFormat 
document.pdf49,74 MBAdobe PDFView/Open

Items in OPUS are protected by copyright, with all rights reserved, unless otherwise indicated.