Forecasting Strong Aftershocks in Greek Earthquake Clusters Using the NESTORE Machine Learning Algorithm

Postgraduate Thesis uoadl:3376985 46 Read counter

Unit:
Κατεύθυνση Εφαρμοσμένη Γεωλογία - Γεωφυσική
Library of the School of Science
Deposit date:
2023-12-28
Year:
2023
Author:
Anyfanti Eleni-Apostolia
Supervisors info:
Φίλιππος Βαλλιανάτος, Καθηγητής, Τομέας Γεωλογίας και Γεωπεριβάλλοντος, ΕΚΠΑ
Stefania Gentili, Researcher, Seismological Research Center, OGS
Γεώργιος Καβύρης, Αναπληρωτής Καθηγητής, Τομέας Γεωλογίας και Γεωπεριβάλλοντος, ΕΚΠΑ
Original Title:
Forecasting Strong Aftershocks in Greek Earthquake Clusters Using the NESTORE Machine Learning Algorithm
Languages:
English
Translated title:
Forecasting Strong Aftershocks in Greek Earthquake Clusters Using the NESTORE Machine Learning Algorithm
Summary:
Any predictive algorithm, or set of algorithms, that learns from data and produces predictions, is known as machine learning (ML). Science is using ML techniques more and more, with applications in seismology and geophysics ranging from pattern recognition to feature extraction that may deepen our understanding of the physical world. In this work, we use the Greek seismicity and the NESTORE machine learning algorithm to predict when a severe earthquake would strike following a mainshock since aftershocks can exacerbate the harm already done to city infrastructure. The approach shows the growth of knowledge over time by extracting machine learning characteristics from the mainshock and evaluating them at increasing time intervals. The characteristics of seismicity during a cluster are described by the features. Depending on the size of the strongest aftershock, NESTORE classifies clusters into two types, type A or type B. Using Uhrhammer's (1986) law, a window-based approach was used to define a cluster. In order to evaluate a significant number of clusters, we used the AUTH earthquake database from 1995 to 2022 over a considerable part of Greece. NESTORE's successful overall implementation in Greece demonstrated the algorithm's capacity to automatically adapt to the research field. The approach is particularly appealing for use in the field of early warning since it allows assessing the chance of a subsequent hazardous earthquake happening after a strong initial event. The greatest performance was obtained for a time interval of 6 hours following the major earthquake.
Main subject category:
Science
Keywords:
Αλγόριθμος NESTORE, Μηχανική Μάθηση, Επιβλεπόμενη, Μετασεισμοί, Χαρακτηριστικά, Ελληνική Σεισμικότητα, Συστάδες, Πρόγνωση, Ενότητα εκπαίδευσης, Ενότητα δοκιμής, Αναγνώριση συστάδων, Ενότητα ταξινόμησης σε σχεδόν πραγματικό χρόνο
Index:
Yes
Number of index pages:
8
Contains images:
Yes
Number of references:
143
Number of pages:
177
File:
File access is restricted until 2024-06-28.

Master_Thesis_Anyfanti_Eleni-Apostolia.pdf
21 MB
File access is restricted until 2024-06-28.