Virtually all shallow earthquakes are followed by aftershocks, with the universal property that aftershock frequency decays roughly with inverse time. This means that the rate of aftershocks in the first 24 hours is about 10 times higher than in the next 10 days, and 100 times higher than in the next 100 days. But aftershock magnitudes do not decay; with time, large aftershocks just become less frequent. The largest aftershock in the first 24 hours is about the same as the largest in the next 10 days, and the next 100 days. And so, large late aftershocks are a threat that only slowly recedes. It is also widely believed that the aftershock rate melts back to the background rate (the seismicity rate in the decade preceding the mainshock), at which time the aftershock sequence can be viewed as ending. But whether it ends after months or decades is widely debated, and is certainly not universal. Similarly, there are conflicting observations about where aftershocks concentrate—on the rupture zone or surrounding it. Reconciling these contradictory observations in order to make better aftershock forecasts was our goal.
The largest aftershocks on the planet are triggered by megathrust earthquakes, M≥9 shocks in subduction zones. We find that on the rupture zone of a megathrust, aftershocks do not simply decay to the background rate as widely expected. Instead, after several years, aftershocks nearly shut down on the rupture to a rate well below background, with the shutdown lasting for at least 60 years and probably more than 300 years; we call this the aftershock ‘core.’ But in a much larger region surrounding the rupture zone, which we call the ‘corona,’ aftershocks persist for 3-6 decades, and these do decay to the background rate. Thanks to the JMA and NIED seismic networks, the 2011 M 9.0 Tohoku earthquake provides the best observations, but all four of the M≥9 shocks since 1960 broadly fit this pattern, so this behavior is not unique to Tohoku. And, in each of these cases, the rupture zone remains a seismicity hole today.
We can explain these observations by recourse to the Coulomb stress transfer hypothesis, married to the theory of rate-and-state friction. The stress on most faults near and on the rupture surface drops profoundly due to the mainshock slip, but the resulting seismicity shutdown is delayed while aftershocks strike on mis-oriented faults near the rupture zone, until these sites are consumed. Stress is also transferred to faults that surround the rupture zone, activating the corona. This model enables us to forecast the spatial and temporal distribution of seismicity with some skill. All four of the M~7 shocks that struck the Tohoku coastline in the past 18 months were brought closer to failure by the Tohoku mainshock, and all lie in the corona. So, contrary to the ‘seismic gap theory, the net seismic hazard increases after megathrusts because the corona is about 10 times larger than the core. Because the corona tends to drape over the coast and lasts for many decades, this hazard increase is critical.
Another outcome of this work is that one should be able to prospect for pre-instrumental and prehistoric megathrusts by searching for seismicity holes. Megathrusts that struck in AD 1700 (Cascadia), 1762 (Myanmar) and 1765 (Pakistan) all locate in contemporary seismicity holes, and we identify several other holes for which no historic quake is known, and so those might be sites of prehistoric megathrusts. We hope that others will join in the hunt for these holes, and for independent evidence—perhaps from tsunamis or oral histories—of these rare but massive quakes.