A Generic Model of Global Earthquake Rupture Characteristics Revealed by Machine Learning

• Published in: Geophysical research letters Vol. 49; no. 8
• Main Author: Li, Zefeng
• Format: Journal Article
• Language: English
• Published: Washington John Wiley & Sons, Inc 28.04.2022
• Subjects: Algorithms; Earthquake magnitude; Earthquake prediction; earthquake process; Earthquakes; Learning algorithms; Learning behaviour; Machine learning; Modelling; Origins; Rupture; Rupturing; Seismic activity; Similarity; source time function; Time functions; Variability
• ISSN: 0094-8276; 1944-8007
• DOI: 10.1029/2021GL096464

Rupture processes of global large earthquakes have been observed to exhibit great variability, whereas recent studies suggest that the average rupture behavior could be unexpectedly simple. To what extent do large earthquakes share common rupture characteristics? Here, we use a machine learning algorithm to derive a generic model of global earthquake source time functions. The model indicates that simple and homogeneous ruptures are pervasive whereas complex and irregular ruptures are relatively rare. Despite the standard long‐tail and near‐symmetric moment release processes, the model reveals two special rupture types: runaway earthquakes with weak growing phases and relatively abrupt termination, and complex earthquakes with all faulting mechanisms but mostly shallow origins (<40 km). The diversity of temporal moment release patterns imposes a limit on magnitude predictability in earthquake early warning. Our results present a panoptic view on the collective similarity and diversity in the rupture processes of global large earthquakes. Plain Language Summary Over the past decades, seismologists have observed great variability in the rupture processes of many large earthquakes. However, some recent studies suggest that the average rupture behavior could be unexpectedly simple. Can the average behavior be representative of most earthquakes? To what extent do large earthquakes share common rupture characteristics? Here, we use machine learning to derive a panoptic picture, that is, a generic model of source time functions, for global earthquakes. The model shows that simple and homogeneous ruptures are pervasive whereas complex and irregular ruptures are relatively rare. Besides, it reveals two special rupture types: runaway earthquakes with weak initial phases, and complex earthquakes with all faulting mechanisms but mostly shallow origins (<40 km). Our results present a panoptic view on the collective similarity and diversity in the rupture processes of global large earthquakes, which affects how well we can predict earthquake magnitude in earthquake early warning. Key Points A generic model of characteristic source time functions is derived from global earthquake observations using machine learning The model presents a panoptic view of the similarity and the diversity in the rupture processes of large earthquakes The diversity of moment release patterns, together with absolute duration variability, limits magnitude predictability in early warning