Abstract
Megathrust subduction faults have caused the largest earthquakes ever recorded in human history. In addition, they are often associated with devastating tsunamis. Thus, assessing their potential maximum magnitude and respective return periods is vital for seismic-tsunami hazard assessment. However, empirical data is limited, owing to their very-long recurrence period that can be several centuries or millennia. To bridge this gap of empirical data with seismotectonic observations, a combined assessment of maximum magnitudes and return periods was undertaken applying a variety of machine learning and conventional methods. For this purpose, 76 subduction zone segments have been assessed worldwide. This includes a 3D modelling of subduction slab geometries on the basis of earthquake hypocentres and a collection of various relevant parameters e.g. those of local geology, geodesy or seismicity. These parameters have been used to assess the potential maximum magnitudes using machine learning classification and other statistical methods. The results have been combined with a tapered Gutenberg-Richter seismicity model and plate tectonic modelling to quantify earthquake return periods and their uncertainties. They show that almost all major subduction zones have the potential to produce earthquakes $M_w\ge8.5$ and that the maximum magnitude highly correlates with the subduction zone geometry. The results also highlight the potential of large megathrust earthquakes in regions where no large or significant events have been recorded during human written history. It is hoped that the results of this study support the development of tail-end hazard and risk studies for long return period events and the quantification of tsunami hazard and risk.
Highlights
Identifying the potential for very large earthquakes and their return periods has been a common research interest since the very beginning of earthquake science
A parametric approach was used on the basis of physical characteristics of subduction zones or subduction zone segments which are correlated to identify physical threshold parameters beneficial for generation of megathrust earthquakes (e.g., Heuret et al, 2011; Schellart and Rawlinson, 2013)
Within the terminology of machine learning, predicting Mmax can be described as a supervised learning approach
Summary
Identifying the potential for very large earthquakes and their return periods has been a common research interest since the very beginning of earthquake science. Great megathrust earthquakes are of primary interest. They can be defined as the largest possible earthquakes of magnitudes Mw ≥ 8.0 occurring along large thrust faults, so-called megathrusts, always found along convergent plate boundaries. The quantification of likelihood and potential for great earthquakes is a frequent but not yet well answered question (Geller et al, 2015) highlighted by the question to identify the largest possible magnitude Mmax at a fault. A parametric approach was used on the basis of physical characteristics (e.g., length, convergence rate, sediment thickness, etc.) of subduction zones or subduction zone segments which are correlated to identify physical threshold parameters beneficial for generation of megathrust earthquakes (e.g., Heuret et al, 2011; Schellart and Rawlinson, 2013)
Sign-in/Register to view highlights & summary Sign-in/Register to access full text options 
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have
Schedule a callDisclaimer: All third-party content on this website/platform is and will remain the property of their respective owners and is provided on "as is" basis without any warranties, express or implied. Use of third-party content does not indicate any affiliation, sponsorship with or endorsement by them. Any references to third-party content is to identify the corresponding services and shall be considered fair use under The CopyrightLaw.
