Site‐specific ground‐motion hazard, as assessed by a probabilistic seismic‐hazard analysis one‐step approach that handles a single‐site sigma and its uncertainties and uses a simple logic tree, is compared with a two‐step approach that includes bedrock motion evaluation and wave propagation through a local soil profile with consideration of the main epistemic uncertainties. The one‐step analysis relies on accelerometer data from the Po Plain, a sedimentary basin in northern Italy where an earthquake sequence with two Mw∼6.0 events was extensively recorded in 2012. Uniform hazard spectra (UHS) on soil and exposed bedrock are evaluated at three deep‐soil accelerometer sites (MRN, NVL, and T0821), using residual measures available from other studies, by which uncertainties in site terms (δS2S) and single‐site sigma (σss,s) are estimated. Despite similarity in geologic conditions, at least one out of three sites in the one‐step analysis displays substantial differences in mean level (T0821) or in the (84–16) percentile spread (NVL), depending on differences in site terms and single‐site sigma, possibly caused by source‐to‐site propagation effects.
The two‐step approach was applied to the remaining site (MRN) using carefully selected and broadband‐matched acceleration signals and linear, equivalent‐linear, and nonlinear approaches as excitation in propagation analyses. We found that assumptions on soil degradation curves dominate the variability of results. The linear approach provided the best results, based on (1) the similarity of the one‐step nonergodic UHS with the two‐step result based on the linear approach; (2) the comparison with observed records at MRN during the 2012 sequence mainshocks, showing peak ground acceleration and short‐period spectral levels well beyond those predicted by different nonlinear assumptions; and (3) similar evidence from a set of 21 stations at deep soil sites of the Japanese KiK‐net.