Dept. Earth and Planetary Sciences
Harvard University
Email:
ekstrom@seismology.harvard.edu
poster/oral: oral
| The term ``seismic noise'' is commonly used to describe signals in a seismogram that have an unknown or uninteresting source. At long periods, the minimum achievable level of noise at quiet stations is well described by the (empirical) low-noise model of Peterson (1993). Though the basic character of the low-noise curve is well explained by consideration of atmospheric and oceanic loading, there are still some unexplained details. For example, a local noise maximum between 50 and 400~seconds, with a peak around 120~seconds, is not well understood. Recent studies have found that some of the ``noise'' in this period band has a distinct structure and corresponds to a low-level excitation of Earth's normal modes. The detection of these oscillations has been accomplished through analysis of low-frequency amplitude spectra, which provides good resolution of the excited frequencies but poor resolution of temporal variations and signal amplitude. The normal-mode oscillations can be represented equivalently as traveling surface waves, and I have used a time-domain detector of long-period (200--400~seconds) surface-wave energy to investigate the character of this background seismic radiation. The time-domain method, based on the autocorrelation function and fundamental properties of multiple-orbit Rayleigh waves, allows for better observations of temporal variations of the signal. Analysis of vertical-component seismograms recorded at quiet seismic stations leads to nearly continuous detections of coherent Rayleigh wave energy using correlation time windows that are only several hours in duration. Stacking of the autocorrelation functions from several stations enhances the detections. I analyzed five years of vertical, long-period data from the Global Seismographic Network. All detections with an amplitude equivalent to that of an earthquake of MW>/=6.0 can be correlated with known earthquakes; no evidence is found for the existence of anomalously slow or ``silent'' earthquakes in the period band 200--400~seconds. The background excitation level is remarkably constant, corresponding approximately to the maximum amplitude detection generated by a MW=5.75 earthquake. Coherent Rayleigh wave energy is detected 94% of the time analyzed. A seasonal component with a period of 6 months is well resolved, with maxima in mid-January and mid-July, consistent with the hypothesis that atmospheric processes are the main cause of the phenomenon. Variations in the background signal level on timescales of several days to weeks are also documented. The background radiation makes the detection of slow earthquakes smaller than MW=6.0 at long periods very difficult with a normal-mode or autocorrelation method. I have developed a new cross-correlation and stacking technique to search for earthquakes at specified locations. The algorithm continuously analyzes the long-period wavefield recorded on the global array and determines whether it is consistent with a localized excitation. Application of this technique to the episode of magmatic unrest near Miyake Island, Japan, in the summer of 2000 leads to the detection of several MW=5.7 slow earthquakes associated with caldera collapse on Miyake Island. |