Are Earthquake Sizes Correlated? Insight From Neural Temporal Point Process Models
Description:
Seismology is witnessing an explosive growth in the diversity and scale of earthquake catalogs owing to improved seismic networks and increasingly automated data augmentation techniques. Hopefully, this community effort to produce more detailed observations should translate into improved earthquake forecasts. Current operational earthquake forecasts build on seminal work designed for sparse earthquake records that combine the canonical statistical laws of seismology. Here, we explore the potential of a neural-network based earthquake forecasting model that leverages the new data in an adaptable forecasting framework: the Recurrent Earthquake foreCAST (RECAST). Our previous research shows that RECAST achieves robust improved forecasting skill against the standard benchmark model (ETAS). Here, we test the capabilities of RECAST beyond the standard earthquake-rate forecasting task. Whether or not the ultimate size of an earthquake is influenced by patterns of the event history is a fundamental question in seismology. A standard assumption is that the size of an earthquake is independent of the event history yielding the well-known Gutenberg-Richter model. Some attempts to move beyond this assumption include reference to previous event magnitudes with moving windows. How to translate the characteristics of timing in an event sequence into a magnitude forecasting model is less clear. RECAST enables an exploration of this relationship without ascribing or exhaustively searching for the appropriate parametric model. Preliminary results indicate that even without the inclusion of event timing a non-parametric and time-varying magnitude forecast is an improvement over a single regional Gutenberg-Righter model. Importantly, improvements also arise when including the timing of events indicating that the magnitude and timing are not independent.
Session: New Methods and Models for More Informative Earthquake Forecasting
Type: Oral
Date: 4/19/2023
Presentation Time: 10:45 AM (local time)
Presenting Author: Kelian D. Cousineau
Student Presenter: No
Invited Presentation: Yes
Authors
Kelian Cousineau Presenting Author Corresponding Author kdascher@berkeley.edu University of California, Berkeley |
Emily Brodsky brodsky@ucsc.edu University of California, Santa Cruz |
Oleksandr Shchur shchur@in.tum.de Technical University of Munich |
Stephan Günnemann guennemann@in.tum.de Technical University of Munich |
|
|
|
|
|
Are Earthquake Sizes Correlated? Insight From Neural Temporal Point Process Models
Category
New Methods and Models for More Informative Earthquake Forecasting