Assessing the epistemic uncertainty of ground-motion predictions

Due to the limited observational datasets that are available for the derivation of ground-motion prediction equations (GMPEs) for the estimation of ground shaking from future earthquakes there is always epistemic uncertainty in the estimated median ground motion. Because of the increasing quality and quantity of strong-motion datasets it would be expected that the epistemic uncertainty in ground-motion prediction (related to lack of knowledge and data) is decreasing. In this study the predicted median ground motions from over 200 GMPEs for two scenarios (M _w 6 at r _jb=20km and M _w 7.5 at r _jb=10km) are plotted against date of publication to examine whether the scatter in the predictions (a measure of epistemic uncertainty) is decreasing with time for two regions. In addition, these predictions are compared to the median ground motion (and its confidence limits) for these scenarios based on averaging observations from a large strong-motion dataset at different dates. The epistemic uncertainty in predicting the median ground motions for a given scenario using an empirical GMPE are related to the confidence limits of the regression results, which is in turn dependent on the quantity and distribution of the underlying dataset. Confidence limits for three generations of two series of GMPEs are derived and examined for evidence that these confidence limits are narrowing. It is found that there are still considerable differences in predicted ground motions from the various GMPEs and that estimates do not seem to be converging over time, although for western North America (WNA) predictions for moderate earthquakes have show a high level of consistency since the 1980s. A good match is observed between the predictions from GMPEs and the median ground motions and their confidence limits based on observations from similar scenarios, although the confidence limits from the observations are tighter than the variation between predictions from GMPEs would suggest they should be. The confidence limits derived for the series of GMPEs from WNA are, as expected, getting narrower with time but even for moderate earthquakes there is still roughly a 20% uncertainty in the median predicted ground motions, which is consistent with the scatter in predictions from various recent GMPEs for WNA.