Area-based tests of long-term seismic hazard predictions

Abstract

Abstract This article develops several area-based tests of long-term seismic hazard predictions. The tests stem from the hypothesis that the observed fractional area of hazard exceedence should follow in proportion to the region's predicted likelihood of exceedence. For example, a prediction is successful if roughly 30% of the area mapped as having a 30% likelihood of exceeding some hazard threshold in a certain time interval, actually does suffer this level of shaking. Although tests of earthquake predictions are always equivocal, the success or failure of a forecast hazard map can be argued strongly from a statistical assessment of the hundreds of individual point predictions comprising the map. The specific forecasts to be examined are the 30-yr probabilities of peak ground acceleration exceedence that were presented by Ward (1994) for southern California. In lieu of a lengthy historical set of recorded peak accelerations, observational data for the tests were derived from the 150-yr earthquake catalog and standard attenuation relationships. In all cases, the observed hazard ratios were highly correlated with the predictions. Within 95% confidence bounds, the observed levels of hazard coincided with the predicted hazard from models that included a low-magnitude cutoff of M = 5.5 to 6, approximately the same limit as enlisted in the earthquake catalog. I propose a pass/fail criterion for acceleration hazard maps and I suggest that all seismic hazard predictions submit to area-based tests.