Uncertainties and bias in ground motion estimates from ground response analyses.

摘要

Ground motions intensity measures (IMs) are typically estimated using probabilistic seismic hazard analyses (PSHA), which combine the effects of source, path, and site on the IM. Hazard analyses use attenuation relationships to define the probability density function (PDF) for IM conditioned on earthquake magnitude and site-source distance. These PDFs are log-normal, being defined by a median and standard deviation. When ground response analyses are performed to evaluate site effects in lieu of more approximate methods, it is with the expectation that the standard deviation would be reduced and any bias in the median would be removed. This study investigates the degree to which these benefits of ground response analyses are realized as a function of site condition, and outlines how ground response analyses can be implemented within PSHA.; Suites of input motions for ground response analyses were selected and scaled in a manner that accounts for magnitude, distance, and rupture directivity effects, while retaining natural aleatory uncertainty. Input motions were developed in this manner for ground response analyses for 50 sites having 93 recordings. IMs from recordings were compared to predictions from ground response and alternative models such as attenuation relations. Prediction residuals were evaluated using data from sites within categories to evaluate the models' bias and dispersion.; Spectral ordinates from ground response analyses are unbiased at low period (T ≤ ∼1 s), but underestimate long-period ( T ≈ 3 s) spectral ordinates from deep basin sites. At soft clay sites, ground response analyses reduce the dispersion in spectral accelerations at T 1 s relative to alternative models. This dispersion reduction is not observed for stiff soil sites or at longer periods.; The results of the ground response analyses are interpreted to identify as a function of site category the combined uncertainties associated with the inaccurate physics of the ground response model and unknown features of the input motions. This uncertainty can be combined with the standard error of the median ground response prediction to estimate the full dispersion. This dispersion can then be coupled with the median to define the PDF of the spectral ordinate for use in PSHA.