Time‐domain spectral matching is the most commonly used technique in earthquake engineering to obtain accelerograms for which the response spectrum is compatible with a smooth target spectrum, be it a polyline design spectrum or a hazard spectrum. These accelerograms are used for assessing the response of structures, usually beyond their linear elastic regime. Although this practice is widespread, there is an ongoing debate on whether these matched accelerograms are legitimate substitutes for real ones and on whether they produce biased structural response estimates.
To help shed some light on this debate, we addressed two main questions:
How does the matching process modify the time–frequency properties of a real accelerogram?
How can one devise a rigorous benchmark to test the performances of spectral matched accelerograms?
We implemented a nonconventional strategy using the jagged spectra of real accelerograms for a given earthquake scenario as targets. This allows establishing a benchmark by which both seismological and engineering points of view can be addressed. Given the nonstationarity of the accelerometric signals, we used the time–frequency Stockwell transform to compare pairs of accelerograms to characterize in both time and frequency the changes induced by the matching process and to investigate the characteristics of an accelerogram that cause peaks in the response of a nonlinear single‐degree‐of‐freedom oscillator.
On the seismological side, we found that spectral matching generates spectrum‐compatible accelerograms with the same dominant frequency of the S waves of the target, and that the matched accelerograms have, on average, higher coda amplitudes with respect to the natural records with the same spectrum, over a time window of 5 s.
On the engineering side, the most important finding regards the large dispersion of peak responses to accelerograms with the same spectrum. Caution should therefore be applied in using very few spectrum‐compatible records for practical applications.