A Statistical Low Noise Model of the Earth

An accurate modeling of background seismic noise is important to modern seismology in several respects, such as the identification of the best sites for permanent seismic stations, which see noise as a major nuisance; the definition of baselines for automatic detection of station malfunctioning; the definition of the instrumental requirements necessary to correctly record the lowest noises on Earth; and the burgeoning field of passive seismology, which uses noise as a source to explore the interior of the Earth. While a number of scientists have recently approached the study of defining the Earth seismic noise curve using statistics with various degrees of sophistication, the original empirical Peterson’s model (1993) is still a very popular reference. Called the new low noise model (NLNM), it is based simply on the envelope of the lowest noises recorded on the vertical component at 75 stations from eight networks. For each station, a maximum of 9.48 days of recording was considered after removing earthquakes and other transients. No statistical analysis was performed to assess the noise distribution and, as the author states and is evident in his Figure 2, due to the scarce availability of spectra from quiet sites at frequencies above 2.5 Hz, it cannot be considered reliable above this frequency. The NLNM has been followed by several noise models; the first one, proposed in 2004 by Berger et al., was based on the analysis of hourly spectra from 118 Global Seismograph Network (GSN) stations in a one-year period (mid 2001–mid 2002) in order to capture the seasonal variability and provide results for both the horizontal and vertical components. In this analysis data were not scrutinized for transients and earthquakes search and removal. As a final low noise model, the envelope of the first percentile of the empirical distribution for each frequency …