Anatomy of sigma of a global predictive model for ground motions and response spectra

We present a comprehensive scrutiny of the components of the total uncertainty (sigma, σ) of our recent global predictive model (Cauzzi et al. in Bull Earthq Eng 13:1587–1612, 2015. https://doi.org/10.1007/s10518-014-9685-y), with emphasis on their possible dependence on basic model predictors and source region. Our dataset does not allow us to clearly support or reject the adoption of a magnitude-dependent ϕ or τ description, although there is evidence to suggest that τ of large-magnitude events is lower than that of moderate- and small-magnitude events for vibration periods T < ~ 3 s. The distance dependence of ϕ and ϕss in our data is unclear, but near-source residuals exhibit comparatively larger variability, especially at intermediate and long periods, most likely due to the absence of near-source terms (e.g., hanging-wall, directivity) in our predictive model. The variability of the δWes residuals segregated by ground type is magnified at the dominant amplification periods of the site response, and the residuals on EC8 ground-type A are associated with the lowermost spread. The regional dependence of the δWes residuals in our dataset is small up to intermediate periods, and the offset of regional sub-populations with respect to the overall mean of the residuals is practically null. ϕS2S and ϕSS computed based on stations with at least four records are in good agreement with previously published global and regional models, confirming the limited dependence of ϕSS on region and ground type. Compared to other studies, our τ model is enlarged by Pan-European event terms associated to reverse faults, especially those of the 2012 Emilia (Northern Italy) sequence. We propose an alternative τ model that neglects the large spread associated to these event terms. The contribution to τ of poorly recorded events (with less than three records) is effectively minimised by the weighting scheme of Joyner and Boore (Bull Seismol Soc Am 83:469–487, 1993; Bull Seismol Soc Am 84:955–956, 1994) that we used to develop our model.