Faire avancer la sûreté nucléaire

La Recherchev2


Variability of one-dimensional soil amplification estimates at four sites of the French permanent accelerometer network RAP)



Email :

Mot de passe :

Titre du congrès :The 14th World Conference on Earthquake Engineering Ville du congrès :Pékin Date du congrès :12/10/2008


In this study, we compute site amplification estimates and response spectral ratios for different acceleration time histories scaled to different PGA's by propagating them through 1D soil models with stochastic variation of their mechanical properties. We use three models of wave propagation: the widely known equivalent linear model (EQL) (Schnabel et al., 1972), the frequency-dependent equivalent linear model (AKEQL) (Kausel and Assimaki, 2002), and the elastoplastic nonlinear model (NL) developed by Iwan (1967). We have selected four sites from the French permanent accelerometer network (RAP) deployed in the city of Nice, which is characterized by local strong site amplification due to alluvial filling. A sensitivity study on the numerical methods shows that purely nonlinear analysis greatly depends on the input motion whatever the rock input PGA (PGAr) is; whereas equivalent linear analyses have less dependency with respect to PGAr. We present then the variability of amplification estimates among the wave propagation models. We found that frequency-dependent equivalent linear results are close to those from the traditional equivalent linear model for PGA's larger than 0.7g. Conversely, for intermediate PGAr values, the first model de-amplifies the high frequencies much lower than the second one. Furthermore, the standard deviation of amplification estimates increases as PGA and frequency increase regardless of the utilized wave propagation model. Using the previous results, the response spectral ratio as a function of PGAr has been computed and inserted in simplified site-specific probabilistic seismic hazard assessment (Cramer, 2003). Finally, a comparison with experimental data shows that 1D modeling is a first order approximation of site effects in Nice.