Aristotle University Of Thessaloniki (AUTH)
Modeling of earthquake occurrence times through Markov Processes: a contribution to seismic hazard assessment in Greece
Abstract
dc:descriptionStrong earthquakes exhibit the largest fatality among natural hazards, posing a unique threat to the society, and causing serious damage and loss of life. In recent years an increased emphasis is given on the development of stochastic models for earthquake forecasting and the quantification of their predictive skills, which provide information that help to reveal aspects of seismogenesis and contribute to the seismic hazard assessment. This is a part of the statistical seismology research field that is focused on the statistical modeling of earthquake occurrences for the better understanding of their distribution in time, space and magnitude. The main goal of this dissertation is to propose new stochastic models and advanced statistics for the study of the short and long-term properties of seismicity in time. Towards this direction, we introduce the use of the Markovian Arrival Process for modeling the temporal distribution of seismicity, which can be seen as a stochastic point process with intensity rate driven by a hidden Markovmodel. It shows large flexibility that has been emerged to be useful for capturing a large variety of behaviors and under appropriate parameterizationcan approximate a wide class of counting processes like the Poisson process, renewal models and more bursty ones. However, the increased flexibility of the model is linked to the large parameter set necessary for the approximation of the observed behavior sufficiently close. For the fitting of the parameters we use the Expectation-Maximization algorithm which is an appropriate approach in problems with unobserved data. We introduce a grid-based method for the choice of the initial parameter set which we implement in a parallel-framework for reducing the required computational time. One basic issue when a hidden process is applied, is to estimate the most probable sequence of latent states. For this problem, we propose the use of a local decoding algorithm that considers the forward-backward equations, and we verify its stability on simulated catalogs. The evaluated transitions among the hidden states of the MAP model indicate changes in seismicity rate, therefore, we propose the use of the model as an off-line tool for change point detection. The identification of seismicity rate changes is important as they can be associated with seismic clusters triggered either from stress changes or fluid intrusion. We establish a two-step clustering procedure that comprises the MAP model, for an initial separation of the background seismicity from potential seismic excitations, using the revealed changes in the seismicity rate, and a density-based clustering algorithm, DBSCAN, for the detection of elevated density areas in space. We evaluate its performance on a simulated earthquake catalog where the structure of the clusters is known a-priori. Earthquake clustering is an essential aspect of short-term seismicity that can provide crucial information for the determination of faulting geometry as well as to extract useful information on the aftershock productivity of the study area and the behavior of the foreshock activity, whereas background seismicity is also essential to probabilistic seismic hazard analysis for the production of hazard maps. We detect the seismic clusters of three major seismic zones in Greece and provide their clustering properties with the use of the Epidemic Type Aftershock Sequence model, that incorporates the well established Utsu productivity law and Omori-Utsu law, respectively. Concerning the long-term properties of seismicity, we assume that large earthquakes temporal distribution is characterized by non-stationarity, between extended periods of seismic quiescence with long inter-event times that characterize the tail of their distribution and periods of moderate seismic activity. The short-term concentration of seismicity often obscures long-term features that may characterize the earthquakes temporal distribution, therefore we consider the long inter-event times as extreme events due to their rarity and propose a two-step estimation procedure of the model, where the extremeevents are estimated separately from the short-time values. Statistical analysis and forecasting in problems that incorporate extreme events is known to be highly complex as the short times do not conform well with the rare large values, and the extremes are estimated separately ignoring the potential effects of the short-time data. We provide pseudo-prospective experiments based on simulations of the earthquake temporal distribution to demonstrate the contribution of the MAP model to the forecasting of large earthquakes number and for the comparison against the Poisson, non-Poisson renewal models and the temporal ETAS model.
Degree
thesis:*- Grantor dc:publisher
- Aristotle University Of Thessaloniki (AUTH)
- Year dc:date
- 2022
Author and committee
dc:creator, dc:contributor.*- Authors dc:creator
-
- Bountzis, Polyzois
- Μπουντζής, Πολυζώης
Subjects
dc:subject × 24- Στοχαστική μοντελοποίηση
- Κρυφά μακροβιανά μοντέλα
- Συσταδοποίηση
- Πρόγνωση σεισμών
- Ανίχνευση σημείων αλλαγής
- Στατιστική σεισμολογία
- Stochastic modeling
- Hidden Markov models
- Cluster analysis
- Earthquake forecasting
- Change point analysis
- Statistical seismology
- Μαθηματικά
- Φυσικές Επιστήμες
- Μοντελοποίηση και Προσομοίωση
- Στατιστική και Πιθανότητες
- Γεωεπιστήμες και Επιστήμες Περιβάλλοντος
- Γεωφυσική
- Mathematics
- Natural Sciences
- Modeling and Simulation
- Statistics and Probability
- Earth and Related Environmental Sciences
- Geophysics
Rights
- Language dc:language
- eng
Identifiers
dc:identifier.*- Identifier
- 10.12681/eadd/52043