DERIVATION OF STRUCTURAL MODELS FROM AMBIENT VIBRATION ARRAY RECORDINGS : RESULTS FROM AN INTERNATIONAL BLIND TEST

Unfavorable site conditions may give rise to significant local amplification of ground motion during earthquakes. Thus, for an efficient mitigation of seismic risk, site-specific studies are of uttermost importance. Site effects may be characterized either by quantifying Vs30 and using empirical relationships for ground motion prediction or by forward modeling of frequency dependent amplification effects requiring a proper knowledge of the shallow and sometimes deep shear wave velocity structure. Originally proposed by Japanese authors, the use of array measurements applied to ambient vibration for estimating the subsurface S-waves velocity has spread throughout the world in recent decades. Although the processing techniques mainly f-k based and SPAC techniques are relatively well understood from the theoretical point of view, the true performance of those methods for extracting velocity models from microtremor measurements is difficult to assess. The success of shear wave velocity profiling using ambient vibration array measurements depends on the combined influence of the site structure and the characteristics of ambient vibration sources onto the observability of the microtremor wavefield. Additionally, the validity of assumptions regarding the interpretation of original phase velocity measures as mode branches is a prerequisite and the need for interpretation of results introduces a need for expertise. It is therefore important to independently check the reliability of results and their related uncertainties. Within the third international symposium on Effects of Surface Geology on seismic motion, a noise blind test was organized in order to compare the results from competing analysis approaches and to make a clear assessment regarding the potential of microtremor array studies for site effect estimation. This blind test involved both synthetic and real data sets. Synthetic data provided the opportunity to perform a benchmark test where the site structure and the wavefield situation are fully known. Real sites were used to properly assess the reliability of results for various real site conditions. Contrary to real world experiment, no prior information on site condition was provided. Nineteen groups participated to this exercise using different techniques. Regarding phase velocity, we observe a tendency for phase velocity estimates of fundamental mode Rayleigh waves to be biased to higher velocities. At high frequency, we explain this observation by insufficient resolution capabilities of the applied analysis methods with respect to the existence of higher mode contributions in the wavefield. At low frequency, overestimation of phase velocities is mainly due to insufficient resolution for multiple signals arriving from different directions, which is especially true for f-k methods while spatial autocorrelation methods seem performing better. Interestingly, Love waves phase velocity estimates are not or less biased compared to the corresponding Rayleigh wave dispersion curves. An obvious result has been the apparent difficulty in associating the estimated phase velocity samples to the correct surface wave ESG2006, Grenoble, 30/08-01/09/2006 2 mode branches when interpreting the dispersion curve results. Furthermore, we observe a rather optimistic view among participants what regards the capabilities of a specific array configuration: in most cases, phase velocities are measured in a larger frequency band than what is recommended in literature. Regarding the inverted shear-wave profiles, we observe that fine layering, basement depth and velocity were almost never retrieved. The poor bedrock resolution can be explained by the sedimentary cover high pass filtering effect that limits the analyzable lower frequency band for phase velocity measurement. Consistently with the overestimation of Rayleigh waves phase velocities, the shear-wave time-averaged velocities are systematically biased to higher velocities by about 10-15% on average. Site amplification estimation by using either empirically-based prediction or SH transfer function modelling outlined that empirical prediction that only depends on time-averaged velocity in the uppermost 30 meters seems a more robust measure than the SH transfer function (whose computation requires also a reliable estimate of bedrock depth and velocity) provided a proper design of array sizes for enabling shortest wavelengths sampling and a proper interpretation of surface wave modes. Finally, this experiment outlines that the following critical issues need to be improved in the future: 1) accurate identification and interpretation of surface wave modes; 2) introduction of prior information or combined/joint inversion with other reconnaissance data; 3) quantitative and meaningful evaluation of confidence intervals on shearwave profiles.