While site effects are accounted for in most modern U.S. seismic design codes for building structures, there exist no standardized procedures for the computationally efficient integration of nonlinear ground response analyses in broadband ground motion simulations. In turn, the lack of a unified methodology affects the prediction accuracy of site-specific ground motion intensity measures, the evaluation of site amplification factors when broadband simulations are used for the development of hybrid attenuation relations and the estimation of inelastic structural performance when strong motion records are used as input in aseismic structural design procedures. In this study, a set of criteria is established, which quantifies how strong nonlinear effects are anticipated to manifest at a site by investigating the empirical relation between nonlinear soil response, soil properties, and ground motion characteristics. More specifically, the modeling variability and parametric uncertainty of nonlinear soil response predictions are studied, along with the uncertainty propagation of site response analyses to the estimation of inelastic structural performance. Due to the scarcity of design level ground motion recording, the geotechnical information at 24 downhole arrays is used and the profiles are subjected to broadband ground motion synthetics. For the modeling variability study, the site response models are validated against available downhole array observations. The site and ground motion parameters that govern the intensity of nonlinear effects are next identified, and an empirical relationship is established, which may be used to estimate to a first approximation the error introduced in ground motion predictions if nonlinear effects are not accounted for. The soil parameter uncertainty in site response predictions is next evaluated as a function of the same measures of soil properties and ground motion characteristics. It is shown that the effects of nonlinear soil property uncertainties on the ground-motion variability strongly depend on the seismic motion intensity, and this dependency is more pronounced for soft soil profiles. By contrast, the effects of velocity profile uncertainties are less intensity dependent and more sensitive to the velocity impedance in the near surface that governs the maximum site amplification. Finally, a series of bilinear single degree of freedom oscillators are subjected to the synthetic ground motions computed using the alternative soil models, and evaluate the consequent variability in structural response. Results show high bias and uncertainty of the inelastic structural displacement ratio predicted using the linear site response model for periods close to the fundamental period of the soil profile. The amount of bias and the period range where the structural performance uncertainty manifests are shown to be a function of both input motion and site parameters.

La prédiction des mouvements du sol générés par un séisme est un enjeu majeur pour la prise en compte du risque sismique. C’est l’un des objectifs du projet SIGMA dans le cadre duquel j’ai réalisé ma thèse. Celle-ci se compose de deux parties. La première se concentre sur la dépendance à la magnitude de la décroissance des paramètres des mouvements du sol avec la distance. Celle-ci est un sujet de préoccupation aussi bien pour l’utilisation des relations d’atténuation (GMPEs), que pour les méthodes basées sur l’utilisation de petits évènements en tant que fonctions de Green empiriques. Nous avons démontré qu’aux distances les plus faibles (inférieures à la longueur de la faille), l'effet de saturation dû aux dimensions de la faille est prépondérant. Aux distances plus importantes, l'effet de l’atténuation anélastique devient prépondérant. Nous avons donc montré qu’il pouvait être délicat de mélanger des données de différentes régions dans les GMPEs, et validé l’utilisation des fonctions de Green empiriques à toutes les distances. Dans la deuxième partie sont testées 3 différentes méthodes de simulations dans un contexte complexe : un code combinant une source étendue en k2 et des EGFs, un code point-source EGFs et un code stochastique. Nous avons choisi de travailler sur le séisme de magnitude Mw 5.9 (29 mai 2012) situé dans un bassin sédimentaire profond (la plaine du Po), et qui a engendré des sismogrammes souvent dominés par les ondes de surface. On y démontre que sans connaissance à priori du milieu de propagation, les méthodes basées sur des EGF permettent de reproduire les ondes de surface, les valeurs de PGA, de PGV, ainsi que les durées des signaux générés. / The prediction of ground motion generated by an earthquake is a major issue for the consideration of seismic risk. This is one of the objectives of SIGMA project in which I realized my thesis. It consists of two parts. The first focuses on the magnitude dependence of the ground motion parameters decay with distance. This is a concern both for the use of relation of attenuation (GMPEs) than methods based on the use of small events as empirical Green functions. We have shown that as the shorter distances (less than the length of the fault), the saturation effect due to the fault size is preponderant. For larger distances, it’s the eanelastic attenuation effect which becomes predominant. So we have shown that it can be tricky to mix data from different regions in GMPEs and we validated the use of empirical Green functions at every distance. In the second part are tested three different simulation methods in a complex context: a code combining finite fault source in k2 and EGFs, a point-source code with EGFs and a stochastic code. We chose to work on the Mw 5.9 earthquake (May 29, 2012) which occurs in a deep sedimentary basin (the Po plain), and which has generated seismograms often dominated by surface waves. We show that without a priori knowledge of the propagation medium, methods based on EGFs can reproduce surface waves, the values of PGA, PGV, and the durations of the signals generated.