PERFORMANCE OF CANONICAL CORRELATION IN DEVELOPING A HIGH-RESOLUTION SITE AMPLIFICATION MAP IN BASEL

Site amplification is generally estimated using different spectral techniques based on the analysis of earthquake recordings. However, it is not always feasible to apply them in areas where only a few earthquake recordings are available due to a lack of seismic stations or seismicity, or where ground motion needs to be estimated with high spatial resolution. However, the ambient noise at single stations can be measured relatively easily, making it a good candidate for exploiting Horizontal-to-Vertical Noise Ratios (HVNRs). A multivariate statistical technique such as Canonical Correlation Analysis (CCA) can be used to evaluate the relationships between HVNRs and Empirical Amplification Functions (EAFs) from earthquakes, along with geologic and soil parameters that are hypothesized to influence site amplification, at a set of reference sites. Then they can be applied to predict amplification functions at locations where only HVNRs are available. Taking advantage of a large training dataset consisting of EAFs, shear-wave velocity (Vs) profiles, and HVNRs, available from about 200 stations of the national seismic networks of Switzerland, we apply CCA to develop a high-resolution site-amplification map, over the frequency band of engineering interest (0.5-10 Hz), for Basel, Switzerland. Adjustment factors in the form of geological and geophysical predictor proxies, such as the last glacial maximum thickness (LGM) and the average Vs down to 30 m depth (Vs30), are also considered in order to improve the predictions. Our results indicate that the developed high-resolution amplification map, at any period of interest for earthquake engineering, efficiently captures the spatial variability of the complex shallow geology in the area. However, the performance of CCA in this context can depend on several factors, including the quality and quantity of HVNRs and proxy parameters. We present a comparison of our results with amplification maps obtained by using 1) a hybrid empirical spectral method and 2) geological proxies, and provide interpretations based on the possible contribution of the surface geology of the area. We thoroughly discuss the limitations and uncertainties associated with the CCA method and show that it can be used as a reliable tool for elucidating site effects at a resolution of engineering interest.