Identification Protocols for Horizontal-to-Vertical Spectral Ratio Peaks

ABSTRACT Peaks in horizontal-to-vertical spectral ratios (HVSRs) of Fourier amplitudes from three-component recordings of ground vibrations without undue local anthropogenic influence are used to identify site resonances, which are an important component of site response. We address two topics: (1) how should HVSR peaks be identified and (2) are there appreciable differences in HVSR derived using different instruments recording microtremors and seismic strong ground motions? We propose identifying peaks by considering peak amplitudes relative to neighboring ordinates and peak width. The procedure incorporates a regression tree algorithm that can be tuned to conform with user preferences toward relatively “conservative” or “liberal” peak identification (producing few or many sites with peaks, respectively). We then investigate the consistency of microtremor-based HVSRs (mHVSRs) derived from seismometers and accelerometers, which show a high rate of false negatives (missed peaks) from accelerometers with a full scale of ± 2g or greater. In contrast, mHVSRs derived from collocated temporary and permanent seismometers (optimized to record teleseismic signals) have about 60%–80% consistency (with no apparent bias in peak attributes). This indicates that mHVSRs from accelerometers having a broad full scale are unreliable but that mHVSRs can be reliably obtained from temporary or permanent seismometers. Finally, we compare seismometer-based HVSR from microtremor and earthquake (eHVSRs) sources. Results are consistent for 60%–70% of sites (i.e., both either do or do not have significant peaks, and when peaks are present, they occur at similar frequencies, 20% change). For sites with an mHVSR peak, the rate of corresponding eHVSR peaks is nearly 50%, whereas for sites without an mHVSR peak the eHVSR peak rate is low (about 20%). The mismatch rate for mHVSR peak sites is sufficiently high that the use of eHVSR to derive site response models is likely too optimistic (overestimates model effectiveness); mHVSR is preferred for consistency with information available in forward applications.