The application of earthquake recordings to the estimation of an event’s magnitude and the construction of rapid‐response ground‐motion maps requires an adequate classification of the recording stations in terms of their site response. For permanent stations, this information can be obtained from a sufficiently large database of past recordings.
In this work, we analyze more than 7300 three‐component recordings collected between 1996 and 2017 by 67 permanent stations in northeastern Italy to assess their site amplification. The signals come from 368 earthquakes with a magnitude range of M 3.2–5.8 and a distance range of 10–300 km. We evaluate the frequency‐dependent amplification function with respect to a reference station with a flat seismic noise horizontal‐to‐vertical spectral ratio. The evaluation relies on the decomposition of the S‐wave amplitude spectra in terms of source, propagation, and site response. We solve the decomposition with a nonparametric, single‐step generalized inversion in the frequency band 0.5–20 Hz. In addition, we compute the amplification factors for peak ground acceleration and velocity with respect to a well‐established ground‐motion prediction equation. The results highlight that only 11 stations show a relatively flat unitary response with respect to the reference site, whereas the frequency‐averaged amplification function at 23 out of 67 stations exhibits a value larger than 2. We classified the sites according to their surface geology and geomorphological scenario and found that amplification affects not only stations installed on the alluvial soil but also several stations installed on what are assumed to be rock sites. Sites in caves and mines exhibit deamplification, whereas the stations with sensors in boreholes exhibit the typical interference pattern. A good correlation between the amplification factors and the frequency‐averaged amplification functions suggests the possibility of predicting time‐domain peak ground‐motion values from amplification functions estimated by generalized inversion.