The acquisition and integrated inversion of a continuous active source seismic monitoring dataset : application to shallow hydrofracture evolution

Fast subsurface processes, which occur on the scale of seconds to minutes, present a challenging target for geophysical monitoring approaches. The initiation, propagation, and consolidation of a hydraulic fracture is one example of a process occurring on these time-scales for which a satisfactory spatio-temporal imaging approach has been difficult to develop in a field setting. Active-source seismic monitoring has the potential to dynamically characterize fracture propagation rates, spatial extent and potentially the transport properties of the newly established feature. However, even well designed timelapse seismic surveys have limited temporal resolution due to the intrinsic time required to acquire a survey. We have recently developed a technique which circumvents this limitation by using fixed semi-permanent arrays of seismic sources and receivers to allow complete active source surveys to be acquired in a matter of minutes indepedendent of human intervention; this approach, referred to as multi-level continuous active source seismic monitoring (ML-CASSM), can provide precise quantification of traveltime and attenuation changes as well as fast temporal sampling. However, the datasets generated by this acquisition approach are voluminous, consisting of 10s to 1000s of timelapse "epochs" which span the targeted event and present unique problems in the pre-processing, inversion, and integration stages if a consistent spatio-tenporal model is the desired result. We present results detailing our extended analysis of the first deployment of ML-CASSM with a dense source/receiver geometries. Our system is capable of autonomously acquiring full tomographic datasets (10 sources, 72 receivers) in 3 minutes without human intervention, thus allowing active source seismic imaging of processes with short durations. The dataset in question targets the emplacement of a single hydraulic fracture for the purpose of enhanced bioremediation of DNAPL contamination. We apply a novel inversion process to the dataset which uses a temporal regularization scheme to jointly invert 30 temporal epochs spanning the course of fracture initiation, propagation, and consolidation. The fracture, visible seismically as a localized zone of decreased P-wave velocity and increased P-wave attenuation, was effectively tracked using this approach. Joint inversion or "temporal fusion" of the large sequence of datasets improved the quality of the spatio-temporal model of the evolving fracture. The depth and position of the fracture as observed by ML-CASSM was confirmed by subsequent confirmation drilling and core acquisition.