Statistical Clustering of Microseismic Event Spectra to Identify Subsurface Structure

摘要

Understanding subsurface structure by studying microseismicity influences a wide range of activities, including energy extraction, aquifer storage, carbon sequestration, and seismic hazard assessment. Identifying individual fractures in a larger fault system is key to characterizing, understanding, and potentially mitigating risks of natural or induced seismicity.A year-long study associated with a carbon dioxide (CO2) sequestration project was conducted at the Aneth oil field in southeast Utah to record microseismicity at a single downhole geophone array. A previous analysis located events by first identifying event multiplets consisting of highly correlated time-domain waveforms on receivers shallower than the depth of the microseismic events. Then, a relative location algorithm was used within each multiplet. Hypocenters turned out to be in a layer not directly impacted by either water or CO2 injection or oil extraction. Nevertheless, the locations outlining faults are consistent with the geology of the basin.In this thesis, hierarchical agglomerative clustering is used to identify subtle differences for one multiplet at the deepest receiver in the array, whose waveforms might include guided waves. Each event starts out as its own cluster, after which events are iteratively combined based on a dissimilarity metric until a single, final cluster results. Two distance measures are defined, spectral and temporal distance, and used to calculate dissimilarity in the clustering algorithm.While time-domain clustering was inconclusive, clustering in the frequency domain reveals first spectral differences between two groups of events in multiplet 18, which may originate in different lithologies. A more detailed look identifies subclusters in one of these groups that organize spatially. Subtle spectral differences are detected that are not the result of attenuation and may identify individual en echelon fractures within the same lithological unit.More investigation into the application of hierarchical agglomerative clustering to event spectra and waveforms is needed to identify geophysical conditions where the method could be further utilized. Additional station components, stations, and multiplet analysis could further characterize the methodu27s strengths and constraints, as well as refinement of the geophysical interpretation of results.