Characterization of seismic waveforms and classification of seismic events using chirplet atomic decomposition. Example from the Lacq gas field (Western Pyrenees, France)

Abstract

S U M M A R YIn the present paper, we present a generalization of the wavelet transform, known as chirplet transform, specially designed to quantify the morphological attributes of individual seismic sections (packets) constituting the seismic waveforms. The proposed transform relies on an atomic decomposition of individual seismograms based on local multiscale chirps (swept frequency wave packets) of various shape and duration. We developed an algorithm that provides an optimal representation of the waveform packets in terms of (i) arrival time, (ii) central frequency, (iii) modulus, (iv) phase, (v) duration, (vi) envelope shape and (vii) frequency modulation compacting the information contained in each seismogram into a reduced set of parameters particularly well suited to describe seismic waveforms. In the present work, we focus on the ability of atomic decomposition to classify seismic events. We illustrate the developed methodology and resulting hierarchical classification scheme (agglomerative clustering displayed as a dendrogram) to seismograms of the induced seismicity recorded in the Lacq gas field between 1989 and 1997 by a local seismic network. For the present case-study, the resulting classification reveals different levels of similarity between seismic events of a same swarm. Accurate analysis of the subsequences of seismic events associated to an injection well shows temporal changes in the morphological attributes of the recorded seismic waveforms. These changes are highly correlated with water over-pressure records of the reservoir demonstrating the capability of the method to guide investigation of the underlying processes (properties of propagation media, source, rupture processes), and in a general manner the physical properties of the reservoir. Two major difficulties in earthquake studies are the lack of (1) controlled direct and near-field observations (essential for the validation of models and concepts) and (2) signal processing tools and analysis methods closely connected to the physics of wave propagation in a heterogeneous and dispersive medium (scattering). In this way, seismologists attempt to bridge the gap between laboratory experiments and tectonic earthquakes in the crust by investigating intermediate-scale systems where the cause of seismicity is more or less understood, as it is controlled by anthropogenic activities or by visible volcano activities (Ruiz et al. 1998). Providing that seismologists are able to relate the cause of seismicity to the observed seismic activity, systems associated to subsidence caused by fossil fuel extraction, local pressure changes over large water reservoirs, alteration of the local stress field of significant volume of rock around mining area, extraction of geothermal energy, or volcano activity can be seen as natural laboratories useful for investigating seismogenic processes.The physics of earthquakes (i.e. including nucleation and rupture initiation, rupture propagation) is complex and its understanding requires to efficiently monitor small pe...

Autonomous detection of calving-related seismicity at Kronebreen, Svalbard

“…Joswig, 1990;Dowla et al, 1990). On the other hand, unsupervised pattern recognition may be used to generate an initial understanding of the unknown data properties without utilizing existing class or event labels as done for supervised learning (Bardainne et al, 2006;Köhler et al, 2009). Clustering is a well-known unsupervised learning method which describes the task to find a meaningful grouping of unlabeled data into respective categories (e.g.…”

Noise Corruption of Empirical Mode Decomposition and Its Effect on Instantaneous Frequency

“…The signal was constructed using elementary chirplet wave packets. Such chirplet packets were proposed by Bardainne et al [2006] to decompose seismograms. Details of the construction are given in Appendix A.…”

“…5 is based on the work of Bardainne et al [2006] and is constructed as follows: The parameter values used are shown in …”

The April 2007 Swarm in Trichonis Lake Using Data From a Microseismic Network

“…In order to avoid the limitations of well known clustering algorithms such as the hierarchical methods, the k -means, the fuzzy means, and the Expectation Maximization algorithm (Everitt et al, 2001, Bardaine et al, 2006, Becker et al, 2006, we propose a fast, sequential, graph based algorithm that exploits the structure of the similarity graph and produces a single cluster in each iteration. The proposed algorithm (Pikoulis et al, 2006) emphasizes on the quality of the produced clusters by introducing a suitable measure to evaluate the participation of each object to a cluster and by expressing the overall quality of the cluster as a function of the participations of the individuals that comprise it.…”