Use Vs,30 to characterise sites. Distribution w.r.t. NEHRP classes: A (6 records), B (399), C (2405), D
(1022) and E (42).
Study the influence of uncertainties in estimates of Vs,30 on model (coefficients and σ). Propose regression
technique, based on generalized least-squares and maximum-likelihood approaches, to account for variable
accuracy of Vs,30 estimates.
Approach allows estimation of site-specific σ for use when Vs,30 is known to different accuracies (e.g.
specified only by class or specified by measured Vs profile). Find that this approach leads to lower σ for
well-characterised site than one that is poorly known.
State that objective is not to develop models for use in hazard assessments but to show impact of
measurement uncertainty on model.
Undertake various regressions assuming different uncertainties in Vs,30, including uniform for all sites and
individual. When uniform uncertainties are assumed, coefficients unchanged but σs (inter-site) decreases
but the other components are unaltered. Find coefficients largely insensitive to assumption of non-uniform
uncertainties but again σs decreases.
Discuss the use of the variance components in the context of seismic hazard assessment. Even if σs has
decreased have to consider the measurement error when computing total σ, which is unchanged in the
case of uniform uncertainties in Vs,30 but decreases if know Vs,30 more accurately. Allows discrimination
between aleatory variability and epistemic uncertainty.