3D numerical simulation and ground motion prediction: Verification, validation and beyond – Lessons from the E2VP project

Highlights

• Limitations to validation mainly come from source and model uncertainties.

• Validation easier for distant and/or deeper events (R>10–20 km, Z>8–10 km).

• High sensitivity of ground motion to depth and distance of local events, and high sensitivity of 3D site response to back-azimuth.

• Very significant impact of location and magnitude uncertainties on the between-event variability.

• Impact of epistemic variations of site response (related to back-azimuth, distance and depth) on the within-event variability.

Abstract

The Euroseistest Verification and Validation Project (E2VP) is part of a series of complementary benchmarking exercises launched to better assess the ability of numerical simulation to accurately predict seismic ground motion. E2VP targeted more specifically the current, most-advanced numerical methods applied to realistic 3D, linear models of sedimentary basins through a quantitative comparison of the recorded and numerically-simulated ground motions. The target site, located within the Mygdonian basin near Thessaloniki, Greece, has been thoroughly investigated for two decades and a detailed, realistic 3D model has been derived from geological, geophysical and geotechnical investigations, while a dedicated instrumentation provided a significant number of surface and borehole recordings. Verification and validation tests up to a frequency of 4 Hz, much beyond the 0.4 Hz fundamental frequency of the deepest part of the graben, have been performed for a set of 19 local, small to moderate magnitude events. For careful and accurate enough computations, the model-to-model differences are smaller than the model-to-observations differences, the latter being controlled by uncertainties primarily in the crustal propagation model and source properties, and secondarily in the shallow structure. It is therefore recommended to prefer distant and/or deep events (R>10–20 km, Z>8–10 km) for validation exercises. Additional sensitivity tests illustrate the ability of carefully verified numerical simulation tools to provide an instructive insight at the structure of the so-called “aleatory” variability of ground motion, for both its within- and between-event components. The between-event variability is shown to be very sensitive to hypocenter location errors (even as low as ±2 km), and to uncertainty in magnitude estimates. It explains the increase of aleatory variability for small magnitude events and emphasizes the usefulness of dense seismological networks. The within event, single-site variability is shown to be associated to an “epistemic” dependence of the 3D site response on the event back-azimuth, distance and depth, and calls for caution when interpreting single-station variabilities derived from a too small number of events.

Introduction

The rapid development of the simulation codes and computational facilities allowed considering the use of numerical-simulation tools as a valid option for predicting seismic ground motion, especially for poorly instrumented or moderate-seismicity countries lacking representative earthquake recordings. However, such an approach requires a careful evaluation of the actual performance of numerical simulation codes. This issue has been the topic of a few international studies, including blind prediction tests or comparative exercises, focused on various sites. It started with the Turkey Flat, California (Cramer [1]), and Ashigara Valley, Japan (e.g., Bard [2]), blind tests focusing on effects of surface sediments, the results of which were presented during the first ESG conference in Odawara (Japan) in 1992. It was followed by more comprehensive comparison exercises on the Osaka/Kobe basin area in Japan (Kawase and Iwata, 1998 [3]), and on the Southern California area within the SCEC framework (Day et al. [4], [5], [6]]; Bielak et al. [7]), which also included the effects of extended sources and regional propagation in the low frequency range (f<1 Hz). Each of these cases had its own specificities (for instance, very low frequencies for the Osaka and SCEC exercises). A request issued in late 2003 by the French Nuclear Authority (ASN) to perform a 3D, NL simulation of site response for specific sites, was the initial impetus for a dedicated R&D program funded by CEA Cadarache and ILL (Laue-Langevin Institute, an international research center on neutron science based in Grenoble, and operating the most intense neutron source on Earth). It started with an international benchmarking exercise on the Grenoble basin (Chaljub et al. [8]; Tsuno et al. [9]; Chaljub et al. [10]), and was further deepened through the Euroseistest Verification and Validation Project (E2VP). Considering the lessons of the ESG2006 Grenoble benchmark, the E2VP project was launched in 2007 with two main objectives: (a) a quantitative analysis of the accuracy of current, most-advanced numerical methods applied to realistic 3D models of sedimentary basins, in the linear, small strain domain (3DL verification); (b) a quantitative comparison of the recorded and numerically-simulated ground motions (3DL validation). The selected target site was an extensional graben located in the Mygdonian basin near Thessaloniki, Greece, located in a seismically active zone, belonging to both Serbomacedonian massif and Circum Rodope zone (Fig. 1). A detailed, realistic 3D model of the basin and surrounding area had already been derived from a comprehensive set of geological, geophysical and geotechnical investigations, and the site instrumentation installed for about two decades provided a significant number of surface and borehole recordings.

This paper is intended to present a concise overview of the work accomplished since the launching of the E2VP project. This project has been organized in two phases, E2VP1 (2007–2010) and E2VP2 (2012–2014). As the main results of the first phase are reported in two recent papers (Chaljub et al. [11]; Maufroy et al. [12]), the present article puts more emphasis on the latest results, while reminding the overall process. The first section shortly reminds the main learnings of E2VP1, and its shortcomings as well. A few key issues were identified, which shaped the second phase E2VP2: its main components are presented in the following section, including an improvement of the source parameters for a larger set of validation events, an enlargement and refinement of the 3D model on the basis of newly compiled information and sometimes new measurements, and a comprehensive set of numerical simulations for close to 2000 point source locations and 15 receivers. These simulations aim first at the validation up to a frequency of 4 Hz, much beyond the 0.4 Hz fundamental frequency of the deepest part of the graben, for a set of 19 local, small to moderate magnitude events. The corresponding results are described in the following section, distinguishing the rock and sediment stations, and for the latter the absolute ground motion and the 3D site response. The next section is dedicated to the presentation of additional sensitivity tests, which illustrate the ability of carefully verified numerical simulation tools to provide an instructive insight at the structure of the so-called “aleatory” variability of ground motion: the between-event component is shown to be highly impacted by uncertainties in hypocentral location and magnitude, while the within event component is affected by the epistemic dependence of site response on source back-azimuth. The conclusion summarizes the main outcomes from the whole E2VP project, including recommendations regarding the organization of further validation exercises, the use of numerical simulation for ground motion prediction in engineering projects, and the analysis, interpretation and reduction of the aleatory variability in GMPEs.