Liège site

For a simulated wavefield, the theoretical model is perfectly known and the inversion reliability is easily checked by comparing the results to the known velocity structure. On real sites, the results have to be validated by external geological or geotechnical information like existing borehole descriptions, cone penetration test (CPT) or conventional geophysical prospecting data. Actually, those data are also affected by uncertainties which must be considered in the validation process. This section compares the results of the three processing techniques applied to array vibration measurements in the city of Liège, Belgium. The reliability of the techniques is evaluated using newly acquired seismic refraction data and existing borehole data. Signals generated by hammer shots were recorded on vertical sensors for measuring the first P-wave arrivals and the apparent velocity of the triggered surface waves. A special care is paid to the uncertainties of the interpretation of usual refraction data. Within an urban context, the signal to noise ratio is relatively low, and the picking of the P-wave first arrivals can be ambiguous. For each picked time, an error value is estimated. The traveltime-distance curves are then inverted using the neighbourhood algorithm (Sambridge (1999a)) to obtain one-dimensional $V_p$ profiles. This method offers the advantage over other common approaches to take into account the picking uncertainties. The artificially triggered surface waves were processed to give the high frequency part of the dispersion curve (Stokoe et al. (1989), Malagnini et al. (1995)), which might be uncertainly deduced from the processing of microtremor arrays (see below). The overlapping frequency ranges of ambient vibrations and triggered waves offer the opportunity to validate the array results. Though only the vertical components of the sensors are used for the array processing, we measured the three components of the particle motion. The horizontal to vertical spectral ratios (H/V method, Bard (1998)) were computed for all the sensors. The frequency of the peak of the H/V curve is known to be close to the resonance frequency of the site (Bonnefoy (2004)), giving an additional constraint to the $V_s$ profile.