We use ambient noise correlation (ANC) to create a detailed image of the site of the Newberry EGS experiment down to 5 km. Our intent is to precisely locate and monitor the microseismicity associated with fluid injection. We collected continuous data for the month of October 2012, for the 22 stations in the Newberry network, together with 12 additional stations from the nearby CC, UO and UW networks. The data were instrument corrected, whitened and converted to single bit traces before cross correlation according to the methodology laid out in Benson (2007). There are 231 unique paths connecting the 22 stations of the Newberry network. The additional networks extended that to 402 unique paths crossing beneath the Newberry site. Because we are particularly interested in the very shallow seismic structure, we need high quality correlation waveforms at frequencies from 0.5-15 Hz. These particular data are very good and the Green\'s functions (GF) emerge quickly. We treated each GF as a seismic record and inverted for the best fitting 1D model along each path. The objective was to maximize the fit between the GF and synthetic seismograms, including the scattering energy in the coda. Short paths and high frequencies are most sensitive to the shallowest structures. Deeper structures are resolved using longer paths. We inverted simultaneously for Vp, and Vs and Qs, although Qs is poorly resolved. We broke the data into 3 groups. GFs for paths shorter than 5 km were filtered between 0.6 to 15 Hz and focused on matching details to 1 km. For paths between 5 - 10 km we filtered GFs between 0.5-8 Hz and data for the longest paths were filtered between 0.1 - 2 Hz. These longest paths, typically including at least one station outside the Newberry network, extended our coverage laterally to depths below 5km. The individual 1D models were merged into a tomogram of the region using singular value decomposition. The seismicity generally follows the most rapid changes in velocity gradient. This remarkable feature can be seen in most of the 2D slices throughout the model. To test the accuracy of our result, we calculated synthetic seismograms for local earthquakes through both the original reference 1D and final 3D models using the reflectivity method and the LLNL SW4 code, respectively. We compare the synthetics to a M1.85 earthquake that occurred on December 1, 2012. The 1D synthetics are only able to capture one or two peaks of the actual wavetrain. The 3D synthetics capture the complexity of the surface waves, such as those observed at the NB19 surface station. Finally, we compare the rate of microearthquake occurrence with the fluid injection and pressure parameters. We apply the Matched Field Processing (MFP) method to identify more and smaller microearthquakes than were identified using traditional STA/LTA earthquake detection methods. Between October – December 2012, 204 events were identified using traditional techniques. In the same 3 months the MFP technique identified 249 additional events.

Press the Back button in your browser, or search again.

Copyright 2014, Stanford Geothermal Program: Readers who download papers from this site should honor the copyright of the original authors and may not copy or distribute the work further without the permission of the original publisher.