Earthquake hazard maps that underlie catastrophe modelers’ insured loss estimates are poorly constructed and are leading to significant predictive failures, according to a study released U.S. and Japanese researchers.
Using example of the 2011 Tohoku earthquake — as well as the 2010 Haiti and 2008 Wenchuan earthquakes — the study argues that many hazard maps often underestimate or overestimate possible seismic activity, resulting in “earthquake occurrence [that] is typically more complicated than the models on which hazard maps are based.”
“[Key] aspects of hazard maps often depend on poorly constrained parameters, whose values are chosen based on the mapmakers’ preconceptions,” the study titled Why earthquake hazard maps often fail and what to do about it. “When these are incorrect, maps do poorly.”
Citing the Tokyo earthquake — which killed 16,000 and lead to losses up to $34.6 billion — the study points out that the Japanese national seismic hazard map showed that the region had significantly lower quake hazard than other parts of Japan. The report argues that this was because of a widespread view among Japanese seismologists that “large earthquakes would not occur on the Japan Trench off Tohoku.”
“Such a giant earthquake was not anticipated off Tohoku due to several incorrect assumptions that reinforced one another,” the report says, adding that the available earthquake history was coupled with an “incorrect hypothesis” of subduction dynamics that dismissed the possibility of large earthquake Japan Trench.
“The present state of hazard mapping reflects the general paradox that humans desire to predict the future so strongly that we are reluctant to ask how well or poorly predictions do,” the study adds.
In order to make better use of earthquake hazard maps, the researchers suggest that users, such as modeling firms and insurers, recognize the uncertainties of the maps and “decide how much credence to place” on them when determining the predictive use. In addition, the hazard maps should undergo strict testing to compare their predictions to other hypotheses and uniform regional seismicity or hazard data.
“If these models fundamentally diverge from the actual non-linear physics of earthquake occurrence – as may well be the case – then no amount of tweaking and tuning of models can produce hazard maps that come close to the ideal of predicting the shaking that will actually occur,” the study adds.