Using ambient vibration data for predicting site amplification for the city of Lucerne (Switzerland): comparison between canonical correlation and hybrid standard spectral ratio methods

Assessment of seismic risk should include estimation of site response variability, especially in densely populated cities situated in soft sedimentary basins. Earthquake recordings can be used to derive empirical amplification functions (EAF), e.g. based on empirical spectral modelling or the standard spectral ratio (SSR) method. However, due to the high background noise in urban areas, a seismic monitoring network can take several years to record a statistically sufficient number of events, particularly those that are situated in low-to-moderate seismicity zones. Meanwhile, ambient vibrations can be recorded everywhere quickly and at a low cost. However, evaluating the site amplification using approaches that merely take into account background noise, such as noise-based spectral ratios (SSRn) or horizontal-to-vertical spectral ratios (HVSR), is still very challenging. Therefore, we tested and compared two different approaches based mainly on ambient vibration recordings. However, in both techniques, a number of sites with earthquake-derived amplification functions are needed for the calibration of the amplification models.

In the frame of the Horizon 2020 ITN-funded URBASIS-EU project, which focuses on urban seismology, the Swiss city of Lucerne, located on soft soil deposits, is used as a test site to obtain local amplification models. Although the seismicity in the region is low to moderate, there have been a few significant earthquakes in the area (including one with a magnitude of 5.9 in 1601). Therefore, the long-term seismic hazard cannot be ignored. For our purpose, the hybrid standard spectral ratio approach (SSRh) is used, in which spectral ratios based on ambient vibrations at 100 sites are adjusted with spectral ratios based on weak-ground motion records from small earthquakes at 10 temporary seismic stations. Another approach uses the statistical method of canonical correlation (CC) to correlate the amplitudes of HVSR and EAF, allowing for the reconstruction of the amplification from HVSR. CC derives the correlation from a large number of sites, where both EAF and HVSR are available. Using this method, the amplification functions are estimated for 320 sites from HVSR in the Lucerne area.

Both approaches are compared separately to EAF, mainly the SSR method -  where available - showing a good agreement, hence, they can be used to predict EAF from ambient vibration data. Both models indicate high amplification factors in some parts of the city reaching 10 at about 1 Hz. However, the results from SSRh and CC are not identical. We thoroughly analyzed the spatial and frequency distribution of the differences between models to assess the reliability and limitations of both methods. Generally, the SSRh approach provides higher amplification factors in the deep basin while CC gives higher estimates at the basin borders. We test also how several factors affecting the models, for example, the length and time of the recordings, influence the model differences.