Data-driven and machine learning identification of seismic reference stations in Europe

Abstract

SUMMARYThe growing seismic networks and the increasing number of permanent seismic stations can help in improving the physical basis of seismic hazard assessment. For this purpose, the definition of reference site conditions is of great significance. If a reliable estimate of the reference ground motion is known, its modification at any given site can be modelled with respect to that reference site. Since the choice of a well-characterized reference site is not straightforward, mainly due to the high variability in the shallow layers, such choices prove to be affected by large uncertainties. While proxy parameters like the average S-wave velocity over the uppermost 30 m (vS30) might help in characterizing reference site conditions, such parameters are neither available at all sites nor do they allow concluding that the site is not affected by amplification and attenuation effects. In this study, we identify prospective reference sites across Europe in a harmonized and fully data-driven way. All analysis is based on freely available geological and geophysical data and no on-site measurements or site-specific proxies are required. The study accounts for both the influence of amplification and attenuation in a large frequency range. To address the key conceptual issues, we verify our classification based on machine learning techniques in which the influence of the individual site characterization parameters is investigated. Our study indicates that around 250 sites in Europe over more than 2000 investigated are not affected by local site effects and can de facto be considered as reference sites based on the criteria applied.