Influence of processing technique on the agreement of site fundamental frequency (f0) from earthquake and microtremor horizontal-to-vertical spectral ratio

Abstract

Leveraging a database of earthquake recordings and microtremor measurements collected at seismic stations in California, this research explores the influence of processing decisions on the site fundamental frequency (f(0)) obtained from the horizontal-to-vertical spectral ratio (HVSR) of earthquakes (eHVSR). The study systematically evaluates different approaches for processing eHVSR, including considering signal-to-noise ratio (SNR) to determine usable frequency range, the use of the Fourier amplitude spectrum (FAS) or pseudo-spectral acceleration (PSA), and the impact of using the full earthquake record or selecting the S-wave portion. The SESAME clearness and reliability checks on eHVSR reveal that an SNR-based frequency range outperforms a total frequency range, and FAS outperforms PSA. Across different eHVSR, f(0) was determined to be consistent with microtremor HVSR (mHVSR) (Pearson correlation coefficient, r > 0.95) while revealing strong differences in amplitude (r approximate to 0.01-0.7), with 30-40% of the mHVSR-eHVSR pairs disagreeing regarding the occurrence of peaks and resultant median curves being flat. However, when peaks are identified, f(0) from various eHVSR matches with mHVSR (r > 0.90), but their amplitudes do not (r r = 0.89). Lastly, while selecting the S-wave window manually versus automatically using existing machine-learning algorithms, they occasionally did not identify identical portions of the earthquake recordings; however, both methods produced very similar eHVSR. Therefore, while additional study is necessary to understand the source of these differences, existing machine algorithms for S-wave selection show promise for use as part of eHVSR processing. Hence, the FAS method employing the manually picked S-wave window and/or full earthquake, along with the calculation of SNR-based frequency range, may be favored for determining f(0) from eHVSR curves. The source of inconsistency between mHVSR and eHVSR should be further investigated.