Multiple estimates of soil structure at a vertical strong motion array: Understanding uncertainties from different shear wave velocity profiles

• We present VS estimates carried out at a site of deep downhole accelerometric array.
• 62 VS profiles are analyzed using several invasive and non-invasive techniques.
• VS estimates depend on the hypotheses of the methods and the model's heterogeneity.
• Intermediate VS contrasts may limit the usefulness of monotonic gradient functions.

We present a detailed study of investigation of various shear wave velocity, VS, profiles at the TST site of the Euroseistest test-site. We benefit from the availability of 62 VS models derived from earthquake records, conventional seismic prospecting, and seismic noise array measurements. Five groups of models provided from many different non-invasive and invasive methods (seismic interferometry, stress–strain analysis, annealing simulation, surface wave inversion, cross-hole and down-hole tests, and seismic noise array measurements) lead to averaged VS profiles. The estimate of VS models that we obtain differs depending on the technique used. In such cases, it becomes clear that, it is better to understand the differences and not simply compute an average. The percentage of the observed disparity with respect to the average reference model albeit is small, becomes significant at certain depths and is associated with the existence of strong vertical discontinuities, thus introducing an uncertainty on the interface definition between the main formations. On the other hand, the use of VS profiles in ground simulation studies (especially 2D or 3D) usually implies the need to represent them as gradient functions. Testing representative generalized power law functions, we found that they fail to predict reliably the VS model for the total thickness of sediments. To overcome this, a third order polynomial function is proposed. Finally, we test the sensitivity of average VSz index widely used in soil categorization and site amplification studies and find that all VS models, either measured or proposed, are equivalent in terms of VSz vanishing any discrimination between layering as well as models.