Reducing the standard deviation of ground motion models (GMMs) is a challenge that has engaged the seismic hazard scientific community for many years with results that unfortunately do not match the dedicated efforts. The work of the last few years, allowing the processing of an ever-increasing amount of data with machine learning approaches, has allowed great progress towards the reduction of these uncertainties, nevertheless, the high frequency uncertainty remains large, especially for GMMs developed in the Fourier domain.While these machine learning approaches are particularly promising, it is nevertheless necessary for the seismologist to provide them with the relevant parameters to test. In this work, we inventory a set of phenomena that disturb high frequency signals and that are not or rarely taken into account. We present a synthesis of recent results on different phenomena: small-scale soil-structure interaction generated by slabs or pillars often used to couple seismometers and accelerometers, depth effect, small-scale topography effect, seasonal variations. We discuss the interest of a better documentation of the installation conditions within the metadata associated to seismic motion databases.
The 28th IUGG General Assembly (IUGG2023) (Berlin 2023)