Selection of seismic noise recording by K-means

The automatic selection of elementary analysis windows in continuous seismic noise recordings, aiming to obtain the most precise average horizontal-to-vertical spectral ratio (HVSR, also named H/V) curve, holds great importance in conducting HVSR investigations in urban areas. To achieve this goal, an automatic procedure utilizing K-means cluster analysis is proposed. The performance of K-means and hierarchical clustering on clustering the spectral ratio curves was compared and analyzed first. Then the characteristic of spectral ratio curves generated by anthropogenic sources (EHVSR) and site information (SHVSR) was examined. Based on this , an automatic procedure for selecting SHVSR curves using K-means was presented. The procedure was applied to 24 sites. The results indicated that at 23 sites, the predominant frequency of the SHVSR curve was much closer to the predominant frequency of the site transfer function than that of the noise HVSR. Moreover, at 21 of these sites, the difference between the predominant frequency of the SHVSR curve and the predominant frequency of the transfer function was less than 2 Hz.

Automated summary

Automatic selection of analysis windows is crucial in conducting accurate HVSR investigations in urban areas. A procedure utilizing K-means cluster analysis is proposed to achieve this goal. Through comparing K-means and hierarchical clustering methods, the characteristics of spectral ratio curves generated by anthropogenic sources (EHVSR) and site information (SHVSR) were examined. An automatic procedure for selecting SHVSR curves using K-means was presented and applied to 24 sites, where the results showed close proximity between SHVSR curve and site transfer function.