Aftershocks are a response to stress changes generated by large earthquakes and represent the most common observations of earthquake triggering. But explaining and forecasting the spatial distribution of aftershocks has proved difficult. Phoebe DeVries and co-authors use a machine-learning approach to identify a stress-based law that forecasts aftershock locations without prior assumptions about failure criteria or fault orientation. They show that a neural network trained on more than 131,000 mainshock–aftershock pairs can explain independent aftershock locations more accurately than the classic Coulomb failure stress change criterion can. The authors conclude that their approach provides improved forecasts of aftershock locations and enables the identification of physical quantities that may control earthquake triggering.
- Machine learning improves forecasts of aftershock locations (News & Views p556, doi: 10.1038/d41586-018-06030-y)
- Deep learning of aftershock patterns following large earthquakes (Letter p632, doi: 10.1038/s41586-018-0438-y)
Recent Hot Topics
Sign up for Nature Research e-alerts to get the lastest research in your inbox every week.