On the effect of noise on κ estimation

Thumbnail Image
Pikoulis, Erion-Vasilis
Ktenidou, Olga-Joan
Psarakis, Emmanouil
Abrahamson, Norman
Journal Title
Journal ISSN
Volume Title
The amplitude of the Fourier acceleration spectrum decays rapidly at high frequencies. This has been modeled by the parameter κr (Anderson and Hough, 1984). The site-specific component of κr, named κ0, describes the contribution of the first few km of the geological profile beneath a certain site. κ0 is an important parameter in characterizing high-frequency ground motion, crucial for certain structures such as nuclear facilities and small concrete dams. Large scatter has been observed in the values of κ0 published in literature, which may be due to different analysts, different approaches and frequency bands used for its estimation, and the different regions data may come from, the trade-offs with the source (stress drop) or the site (amplification). As it is the case with any parameter estimation problem, the presence of noise in the data constitutes an important factor of uncertainty in the obtained estimates. Nevertheless, despite the extensive research on the topic of kappa, a detailed study treating the impact of noise on the problem of kappa estimation is missing from the literature. This is exactly the goal of the present work. To this end, we conducted a series of experiments using various synthetic noise models, both correlated and uncorrelated, as well as noise from real records. The kappa estimations were based on ‘noisy’ versions of a high-quality real dataset. We used recordings from an accelerometric downhole array in the Corinth Gulf, Greece, whose high sampling rate and high SNR render it ‘best-case’ scenario in practice. Kappa estimates obtained from this ‘pure’ dataset constituted our control sample. The results confirm our belief that the present study constitutes a significant step towards the ultimate goal of quantifying the effect of noise on the estimation of kappa and correcting for the biases it introduces on the obtained estimates.
High-frequecy attenuation, Noise contamination, Stochastic modeling, ML estimation, Seismogram fitting