BAD ASSUMPTIONS OR BAD LUCK: TOHOKU'S EMBARRASSING LESSONS FOR EARTHQUAKE HAZARD MAPPING

The Tohoku earthquake is another striking example - after the 2008 Wenchuan and 2010 Haiti earthquakes - of highly destructive earthquakes that occurred in areas predicted by earthquake hazard maps to have significantly lower hazard than nearby supposedly high-risk areas which have been essentially quiescent. Given the limited seismic record available and limited understanding of earthquake mechanics, hazard maps have to depend heavily on poorly constrained parameters and the mapmakers' preconceptions. These preconceptions are often incorrect. The Tohoku earthquake and its tsunami were much larger than expected by the mappers because of the presumed absence of such large earthquakes in the seismological record. This assumption seemed consistent with a model based on the convergence rate and age of the subducting lithosphere, which predicted at most a low M 8 earthquake. Although this model was invalidated by the 2004 Sumatra earthquake, and paleotsunami deposits showed evidence of three large past earthquakes in the Tohoku region in the past 3000 years, these facts were not incorporated in the hazard mapping. The failure to anticipate the Tohoku and other recent large earthquakes suggests that general rethinking of such hazard mapping is called for. One part of this effort should be an objective testing of hazard maps. We should compare map predictions and those of null hypotheses based on random regional seismicity to earthquakes that actually occurred after the maps were published, and use the test results to assess the maps' uncertainties and improve them. Such testing is common and useful in other fields. Weather forecasts, which are conceptually similar to earthquake hazard mapping, are routinely evaluated to assess how well their predictions matched what actually occurred. Over the years, this process has produced measurable improvements in forecasting methods and results, and yielded much better assessment of uncertainties. Earthquake hazard mapping should do the same.