Innovations in seismic tomography, their applications and induced seismic events in carbon sequestration

This dissertation presents two innovations in seismic tomography and a new discovery of induced seismic events associated with CO2 injection at an Enhanced Oil Recovery (EOR) site. The following are brief introductions of these three works. The first innovated work is adaptive ambient seismic noise tomography (AANT). Traditional ambient noise tomography methods using regular grid nodes are often ill posed because the inversion grids do not always represent the distribution of ray paths. Large grid spacing is usually used to reduce the number of inversion parameters, which may not be able to solve for small-scale velocity structure. We present a new adaptive tomography method with irregular grids that provides a few advantages over the traditional methods. First, irregular grids with different sizes and shapes can fit the ray distribution better and the traditionally ill-posed problem can become more stable owing to the different parameterizations. Second, the data in the area with dense ray sampling will be sufficiently utilized so that the model resolution can be greatly improved. Both synthetic and real data are used to test the newly developed tomography algorithm. In synthetic data tests, we compare the resolution and stability of the traditional and adaptive methods. The results show that adaptive tomography is more stable and performs better in improving the resolution in the area with dense ray sampling. For real data, we extract the ambient noise signals of the seismic data near the Garlock Fault region, obtained from the Southern California Earthquake Data Center. The resulting group velocity of Rayleigh waves is well correlated with the geological structures. High velocity anomalies are shown in the cold southern Sierra Nevada, the Tehachapi Mountains and the Western San Gabriel Mountains. The second innovated work is local earthquake tomography with full topography (LETFT). In this work, we develop a new three-dimensional local earthquake tomography algorithm with the inclusion of full topography that is integrated from the Digital Elevation Model data. We present both synthetic and real data tests based on the compressional (P) wave arrival time data for Kilauea volcano in Hawai'i. A total of 33,768 events with 515,711 P-picks recorded by 35 stations at the Hawaiian Volcano Observatory are used in these tests. The comparison between the new and traditional methods based on the synthetic test shows that our new algorithm significantly improves the accuracy of the velocity model, especially at shallow depths. In the real data test, the P-wave velocity model of Kilauea shows some intriguing features. Velocity decrease from the surface to 2 km depth beneath Kilauea caldera indicates a state change of the basalt. Low velocity zones beneath Pu'u'O'o, Heiheiahulu and the Hilina fault system between 5 and 12 km are possible partial melting zones. High velocity anomalies are resolved below 6 km depth beneath the summit caldera, which may suggest the presence of consolidated gabbro-ultramafic cumulates. In the third work, we installed three broadband seismic stations (Test1, Test2 and Test3) in an Enhanced Oil Recovery field to monitor the potential seismic events associated with CO 2 injection. In the two years of continuous seismic data between October 2011 and October 2013, we observed a type of long duration (LD) events instead of typical micro earthquakes, with an average daily rate of 12. The LD events have the following characteristics: (1) their duration varies from ∼30 to ∼300 sec; (2) the amplitude changes smoothly from the beginning to the end of the LD event window; (3) they are local seismic events and were not recorded by regional seismic stations (e.g., ∼200 km away); (4) the waveforms are very different from those of typical earthquakes, but similar to volcanic tremors; (5) the frequency content is mainly concentrated between 0.5 and 6 Hz, which is similar to the frequency band of volcanic tremors; and (6) the source of the LD event is not a single source and could migrate to complex fractures. We picked the LD events in the two-year time period, calculated their daily rate, and compared the results with the reservoir pressure data measured in the north block. The LD event daily rates of Test1 and Test2 have a similar variation pattern as the reservoir pressure. The peak of the LD event daily rate at Test2 is about two months delayed from the peak of the pressure, whereas the LD event daily rate at Test3 does not show similar pattern. We interpret that this is because Test3 is located in the south block and a sealing fault blocks the migration of the injected CO2 from the north to the south block.