Why earthquake hazard maps often fail and what to do about it

The 2011 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 be relatively safe. Here, we examine what went wrong for Tohoku, and how this failure illustrates limitations of earthquake hazard mapping. We use examples from several seismic regions to show that earthquake occurrence is typically more complicated than the models on which hazard maps are based, and that the available history of seismicity is almost always too short to reliably establish the spatiotemporal pattern of large earthquake occurrence. As a result, key aspects of hazard maps often depend on poorly constrained parameters, whose values are chosen based on the mapmakers' preconceptions. When these are incorrect, maps do poorly. This situation will improve at best slowly, owing to our limited understanding of earthquake processes. However, because hazard mapping has become widely accepted and used to make major decisions, we suggest two changes to improve current practices. First, the uncertainties in hazard map predictions should be assessed and clearly communicated to potential users. Recognizing the uncertainties would enable users to decide how much credence to place in the maps and make them more useful in formulating cost-effective hazard mitigation policies. Second, hazard maps should undergo rigorous and objective testing to compare their predictions to those of null hypotheses, including ones based on uniform regional seismicity or hazard. Such testing, which is common and useful in similar fields, will show how well maps actually work and hopefully help produce measurable improvements. There are likely, however, limits on how well hazard maps can ever be made because of the intrinsic variability of earthquake processes.