Tormann et al.1 evaluate changes in the b-value, a measure of the relative abundance of small and large earthquakes, as a function of space and time in northeast Japan. Near the high-slip zone of the 2011 magnitude 9 Tohoku-oki rupture, they document a sudden increase in the b-value at the time of the earthquake and its return to about 77% of its average 1998–2003 value within about two years. Assuming that the b-value is negatively correlated with differential stress2, they infer that the stress relieved by the earthquake has very rapidly recovered to levels found before the earthquake. Consequently, the authors propose that megathrust earthquake occurrences follow a Poisson process and are effectively random in time. However, we argue that the hazard of a great tsunami-producing earthquake in this section of the Japan megathrust zone is unlikely to already be almost as high as it was just before the 2011 event. Instead, many centuries of stress accumulation are required to rebuild stress and produce another great event.

Prior to the earthquake, the seismicity in northeast Japan was predominantly contractional and involved reverse faults (Fig. 1a, c). Since the Tohoku-oki earthquake, there have been few reverse events near the high-slip zone of the rupture (Fig. 1b, d) and aftershocks primarily occurred as extensional (normal faulting) events in both the overriding and down-going plates of the subduction zone. The dominance of normal faulting suggests that the stress drop during the mainshock was large enough, relative to the background stress, to reverse the style of faulting near the high-slip zone3, 4. Because events with different faulting types in northeast Japan have different b-values (Supplementary Information), we suggest that the observed coseismic rise and postseismic decay of b-values (Fig. 1f) partly reflect the fact that Tormann and colleagues sampled event populations with different mechanisms and from different source volumes in the subduction zone (Fig. 1e).