Improving the representation of apparent anelastic attenuation variability in regionalised Ground Motion Models in Europe with a focus in mainland France

The complex physics of earthquake rupture, wave propagation and site effects are simplified and modelled into generic Ground-Motion Models (GMMs) for use in seismic hazard and risk assessment. However, the complexity of geology and seismicity in Europe leads to a high variability in the ground motion prediction compared to observed data. Recent GMMs partially resolved this variability by regionalising their model and adapting to the specificity of each region. We focus this study on the apparent anelastic attenuation variability, to understand the underlying physics of wave propagation in the crust and better model the variability in GMMs. The regionalisation model used in the recent European Seismic Hazard Maps 2020 (ESHM20) divides Europe in several polygons, each with a specific apparent anelastic coefficient adjustment. This model has certain limitations due to ambiguity in the criteria to define these polygons, which have led to non-reproducible maps and geologically diverse regions being grouped into a single large polygon, especially in France. Two hypotheses have been made to improve the current regionalisation. Firstly, France was divided following the contrast of Rayleigh wave velocity in the ambient noise tomography of France. Secondly, a null hypothesis was defined where no prior geological information is used, and Europe is simply regionalised into a regular grid. Linear mixed-effects regressions were performed on the pan-European ground-motion dataset, complemented with a French dataset, to quantify the apparent anelastic attenuation variability. The regionalisation based on Rayleigh wave velocity captures attenuation variability better than the current model for France. However, the grid-based regionalisation is more accurate in the representation of the attenuation variability which leads to keep this choice as the best regionalisation. Analyses of variance statistics confirmed this result. The size of the grid was also discussed based on these statistical tests and the number of records. The apparent anelastic attenuation variability captured on a regular grid can now be examined for a physical meaning producing this variability and improve the parametrisation of GMM.