The estimation of shear wave velocity and attenuation in near-surface geology is of primary importance in engineering seismology. In fact, such knowledge is essential for site response studies when preparing improved seismic hazard scenarios. In this study, we propose a linear inversion of the spectra of a deconvolved wavefield collected by a borehole array in Istanbul, Turkey. The spectra are calculated using as a reference the recordings collected by a sensor at the surface. This allows us to minimize the effect of the deconvolution filter on the peaks’ amplitudes. The feasibility of the proposed inversion scheme and the parametrization of the velocity and attenuation models were tested and assessed using synthetic data. The real data inversion is carried out using observation from three weak motion events. Several starting models are used for each event, changing the values of the quality factor Qs, whereas the starting Vs profiles are fixed to what was obtained by the analysis of the deconvolved wavefield in the time domain.
Our results showed that fairly well-constrained Vs models can be obtained while the variability of the Qs profile, both between different inversions of the same data set, and between the inversions of different data sets, might be large. This indicates that Qs variations observed in models derived by strong motion data inversions should be considered with care. However, the high agreement of the estimated soil profiles in the uppermost layer, both when using different events and different starting models, suggests that any variation in the S-wave velocity that might be estimated when using strong motion data should be reliable.