Geophysicists still lack a comprehensive understanding of the mechanisms and stochastic dynamics behind the earthquake generation process, and so also lack an ability to make reliable predictions of the likelihood of extreme events. One valuable insight, however, is that earthquake events cluster in both space and time for a variety of physical reasons. In particular, earthquakes can trigger further earthquakes, making seismic events arrive in cascades. One model of such effects – the Epidemic Type Aftershock Sequence (ETAS) model – offers a useful way to quantify the increased earthquake hazard after a mainshock by including the triggering influence of foreshocks, mainshock and aftershocks.
A parallel stream of research has tried to estimate the largest expected magnitude of an earthquake during a future time interval, or for a given seismogenic zone. This work has relied on Bayesian methods combined with extreme value statistics. So far, however, these efforts have not attempted to fully incorporate the rich triggering structure of earthquake sequences. In a recent paper, LML External Fellow Jiancang Zhuang and colleagues have pursued this approach, aiming to constrain the magnitudes of the largest expected earthquakes in a future time interval.
The researchers employed Bayesian analysis with extreme value statistics to compute the Bayesian predictive distribution for the magnitude of the largest expected event to exceed a certain value in the near future. In the analysis, they assumed that the earthquake occurrence rate can be modelled by the ETAS process. The study also assumed that the distribution of earthquake magnitudes could be approximated by a left-truncated exponential distribution (restriction of the Gutenberg-Richter magnitude-frequency law above a certain magnitude threshold). To illustrate the power of the approach, the team analysed one recent prominent earthquake sequence – the 2016 sequence from Kumamoto, Japan – and demonstrated an ability to reliably predict the probabilities of events above specific magnitudes. The principal result of this work is an improved inference procedure to incorporate the ETAS process into the estimation of probabilities of largest expected events. Zhuang and colleagues hope the method can be implemented in current or future operational earthquake forecasting schemes.
The paper is available at https://www.nature.com/articles/s41467-019-11958-4