Towards Incorporating Uncertainties in a 3D Geotechnical Model of the Lower Var Valley, Nice

Accurate 3D geotechnical models of valleys are essential for seismic numerical simulations of site amplifications. Although many sources of uncertainty are present, these are rarely quantified and presented in final models. To address this, we incorporate uncertainty using 2D random field samples from the geophysical measurements of fundamental resonance frequency (f₀) applying Bayesian compressive sampling and Karhunen-Loève expansion.
The Lower Var Valley, between Nice and Saint-Laurent-du-Var, is a 2 km wide fluvial sedimentary basin where site effects have been observed in earthquake recordings. This seismically active area has a reference peak ground acceleration on rock of 1.6 m/s², per French seismic regulations. The valley consists of recent alluvial deposits overlying Pliocene delta conglomerates, sloping 15-20° southward. Quaternary sediments reach depths of 100 m at the Var estuary. The existent 3D geological model used borehole log data (335 logs, only 28 reaching substratum) and geophysical measurements of fundamental resonance frequency (f₀) from 355 ambient vibration recordings.
In this paper, we first quantify the differences between the f₀ from the existing model and those from the geophysical measurements (355 ambient vibration recordings). We derive a random field of f₀ using the Bayesian compressive sensing - Karhunen-Loève expansion (BCS-KL) method on the f₀ from the measurements. The proposed random filed model of f₀ accurately represents the measurements, with the definition of the related uncertainties quantified in terms of the coefficient of correlation. Our results are refine previous models and will support future 3D numerical simulations and basin site effects analysis.

Keywords: 3D geotechnical model, Basin amplification, HVSR, Random fields.