Earthquakes can be triggered by stress changes induced by large earthquakes. Physics-based earthquake forecasting models compute the elastic stress transfer induced by a mainshock and estimate the probability of triggered earthquakes in space and time. However, simpler statistical models often outperform more complex physical models.
Hardebeck  suggests several mechanisms to explain the poor performance of physics-based forecasts using tests on synthetic aftershock sequences. Aftershocks are strongly clustered in space, due to cascades of aftershock triggering and due to the heterogeneity of the background physical conditions. This clustering is difficult to explain using physical models but can be easily predicted using statistical models. A better understanding of the interaction between mainshock stress changes and variable background physical conditions is thus needed to improve the performance of physics-based models.
Citation: Hardebeck, J. L. . Spatial clustering of aftershocks impacts the performance of physics‐based earthquake forecasting models. Journal of Geophysical Research: Solid Earth, 126, e2020JB020824. https://doi.org/10.1029/2020JB020824
—Agnes Helmstetter, Associate Editor, JGR: Solid Earth