Simulating H/V spectral ratios (HVSR) of ambient vibrations: a comparison among numerical models

The use of H/V spectral ratios (HVSR) of ambient vibrations to constrain the local seismo-stratigraphical configuration relies on numerical forward models able to connect observations with subsoil seismic properties. Several models were proposed to this purpose in the last decades, which are based on different assumptions about the nature of the ambient vibration wavefield. Performances of nine numerical tools implementing these models have been checked by considering 1600 realistic 1-D subsoil configurations mostly relative to A, B and C Eurocode8 soil classes. Resultant HVSR curves predicted by the models are quite similar both in their general shape and in predicting the resonant soil frequencies, possibly because all of them share the same basic representation of the subsoil as a 1-D stack of flat uniform viscoelastic layers. The common sensitivity to transmission/reflection matrices resulting from that representation explains the well-known correspondence of HVSR maxima to 1-D resonance frequency estimates, regardless of the physical assumptions (about source distribution, radiation pattern, dominating seismic phases, etc.) behind the computational model adopted for simulating HVSR curves. On the other hand, the computational models here considered provide quite different amplitudes for HVSR values corresponding to the resonance frequencies. However, since experimental HVSR amplitudes at the same site are affected by an inherent variability (e.g. due to the possible lack of ergodicity of the ambient vibration stochastic wavefield, non-ideal experimental settings, etc.) and uncertainty about the local seismo-stratigraphical profile (attenuation, 2-D/3-D effects, etc.) observations cannot be used for general scoring of the considered computational models on empirical basis. In this situation, the 'optimal' numerical tool to be considered for the forward HVSR modelling must be defined case by case.