I use the 10 February, 2006 .Green-Canyon earthquake. as a serendipitous source to study how well the Atlantis survey''s standardbandwidth ocean-bottom nodes recorded the very low frequencies generated by the earthquake. As recorded by the nodes, the amplitude of the loudest part of the earthquake signal (at about 3 Hz) was about 54 dB weaker than the peak amplitude of the air-gun direct arrival (at about 35 Hz). Despite this disparity in amplitudes the earthquake was well recorded, and below 2 Hz is almost completely separable from the airgun signal. The nodes even recorded the background ocean wave noise, which as recorded by the nodes was another 30 dB down from the earthquake peak (i.e., 84 dB weaker than the airgun direct arrival). The ocean wave noise was still about 18 dB above the amplitude resolution limit of the Atlantis nodes. By degrading the dataset I demonstrate that even if the nodes'' amplitude resolution had been far less, we would likely still have recorded the earthquake and ocean wave noise anyway. This apparently works for the same reason that the .sign-bit. recordings made in the early 1980''s worked. The much higher-frequency energy from the source is uncorrelated with the lowfrequency earthquake signal, and thus functions as the .noise. required by the sign-bit method. I _nd that just as for the sign-bit case, a lowamplitude signal can be recovered even if its peak amplitudes are smaller than the instrument''s amplitude resolution.
Sign-bit recording was seen as a way of dramatically increasing the number of recording channels per shot. Only the sign bit at each time sample for each receiver was recorded. In its typical implementation, a vibroseis sweep would provide the necessary redundancy by spreading the source energy over a large number of time samples, which would then be averaged together by the process of cross correlation with the source sweep. Counterintuitively, the presence of signi_cant random noise was essential to the method. Far from being a hindrance, it was the random noise that allowed the desired signals to be recovered with true amplitude (O''Brien et al., 1982).
Currently there is considerable interest in generating and recording low frequencies (Dragoset and Gabitzsch, 2006). One dif_culty is that our traditional receivers (optimized for recording traditional seismic frequencies) are relatively insensitive at low frequencies, resulting in low recorded amplitudes. I show that the principles of sign-bit recording should be applicable to the problem. The high-frequency signals from a traditional airgun survey function as the .random noise. Required by the sign-bit method. The massive temporal oversampling of the data at low frequencies provides the necessary data redundancy, and low-frequency _ltering averages the redundant samples together. I use the pressure component of the Atlantis ocean-bottom-node dataset (Ross and Beaudoin, 2006) for this demonstration. The Green-Canyon earthquake of 10 February, 2006 (MS magnitude 5:3) serendipitously occurred during the survey, and the Atlantis observations were used to help locate the event (Dellinger and Ehlers, 2007; Dewey and Dellinger, 2008).