Numerical simulation of seismic Ambiant noise in the 10-50s period band using the Spectral Element Methods

Nowadays seismic ambient noise correlations are commonly used to develop tomographic models of the crust. In most studies that use this method, the travel time of the surface waves visible on the correlations are inverted 1) using ray theory and 2) assuming that the noise sources are evenly distributed in the medium so that the correlations contain the surface wave part of the real Green function of the medium. However this is not true, since the seismic ambient noise originates mostly from a few discrete locations at the free surface in the 5-50s period band. For this reason, depending of the path considered, ambient noise correlations can either be close or significantly different from the real Green function. To investigate the importance of these two limitations, it is a key point to be able to simulate seismic ambient noise in any 3D medium using a powerful numerical tool such as the Spectral Element Method. More precisely this would be useful to 1) compute synthetic correlations in a 3D starting model and 2) to take into account the uneven distribution of noise sources when inverting the correlations. To achieve these goals, the first step is to see how much data of noise (day, month, year) is required to get a correlation that is close enough to the Green function. Thus we use RegSEM to simulate seismic noise sources and to compute synthetic correlations at the regional scale in the 10-50s period band. We study the convergence time of the correlation towards the Green function and the kind of information (travel time of body and surface waves, amplitude, ...) that is accurately retrieved depending on the distribution of noise sources and the way the noise correlations are computed ("raw" correlation, one-bit normalization, ...). Our results are compared to the theoretical prediction of Weaver et al 2005.