This study presents the first demonstration of the transferability of a convolutional neural network (CNN) trained to detect microseismic events in one fiber-optic distributed acoustic sensing (DAS) data set to other data sets. DAS increasingly is being used for microseismic monitoring in industrial settings, and the dense spatial and temporal sampling provided by these systems produces large data volumes (approximately 650 GB/day for a 2 km long cable sampling at 2000 Hz with a spatial sampling of 1 m), requiring new processing techniques for near-real-time microseismic analysis. We have trained the CNN known as YOLOv3, an object detection algorithm, to detect microseismic events using synthetically generated waveforms with real noise superimposed. The performance of the CNN network is compared to the number of events detected using filtering and amplitude threshold (short-term average/long-term average) detection techniques. In the data set from which the real noise is taken, the network is able to detect >80% of the events identified by manual inspection and 14% more than detected by standard frequency-wavenumber filtering techniques. The false detection rate is approximately 2% or one event every 20 s. In other data sets, with monitoring geometries and conditions previously unseen by the network, >50% of events identified by manual inspection are detected by the CNN.
This paper presents a novel approach to predict the propagation of hydraulic fractures in tight shale reservoirs. Many hydraulic fracture modelling schemes assume that the fracture direction is pre-seeded in the problem domain discretisation. This is a severe limitation as the reservoir often contains large numbers of pre-existing fractures that strongly influence the direction of the propagating fracture. To circumvent these shortcomings, a new fracture modelling treatment is proposed where the introduction of discrete fracture surfaces is based on new and dynamically updated geometrical entities rather than the topology of the underlying spatial discretisation. Hydraulic fracturing is an inherently coupled engineering problem with interactions between fluid flow and fracturing when the stress state of the reservoir rock attains a failure criterion. This work follows a staggered hydro-mechanical coupled finite/discrete element approach to capture the key interplay between fluid pressure and fracture growth. In field practice, the fracture growth is hidden from the design engineer and microseismicity is often used to infer hydraulic fracture lengths and directions. Microseismic output can also be computed from changes of the effective stress in the geomechanical model and compared against field microseismicity. A number of hydraulic fracture numerical examples are presented to illustrate the new technology.
2 February 2022 in Journal of Geophysical Research: Solid Earth
Distributed Acoustic Sensing (DAS) involves the transmission of laser pulses along a fiber‐optic cable. These pulses are backscattered at fiber inhomogeneities and again detected by the same interrogator unit that emits the pulses. Elastic deformation along the fiber causes phase shifts in the backscattered laser pulses which are converted to spatially averaged strain measurements, typically at regular fiber intervals. DAS systems provide the potential to employ array processing algorithms. However, there are certain differences between DAS and conventional sensors. While seismic sensors typically record the directional particle displacement, velocity, or acceleration, the DAS axial strain is inherently proportional to the spatial gradient of the axial cable displacement. DAS is therefore insensitive to broadside displacement, for example, broadside P‐waves. In classical delay‐and‐sum beamforming, the array response function is the far‐field response on a horizontal slowness (or wavenumber) grid. However, for geometrically non‐linear DAS layouts, the angle between wavefront and cable varies, requiring the analysis of a steered response that varies with the direction of arrival. This contrasts with the traditional array response function which is given in terms of slowness difference between arrival and steering. This paper provides a framework for DAS steered response estimation accounting also for cable directivity and gauge‐length averaging – hereby demonstrating the applicability of DAS in array seismology and to assess DAS design aspects. It bridges a gap between DAS and array theory frameworks and communities, facilitating increased employment of DAS as a seismic array, while providing building blocks for the development of DAS array design tools.
25 January 2019 in Journal of Geophysical Research: Earth Surface
The crystal orientation fabric of ice reflects its flow history, information which is required to better constrain projections of future ice sheet behavior. Here we present a novel combination of polarimetric phase‐sensitive radar and seismic anisotropy measurements to provide independent and consistent constraints on ice fabric at Korff Ice Rise, within the Weddell Sea sector of West Antarctica. The nature and depth distribution of fabric in the ice column is constrained using the azimuthal variation in (1) the received power anomaly and phase difference of polarimetric vertical radar soundings and (2) seismic velocities and shear wave splitting measurements. Radar and seismic observations are modeled separately to determine the nature and strength of fabric within the ice column. Both methods indicate ice fabric above 200‐m depth which is consistent with present‐day ice‐divide flow. However, both measurements also indicate an oblique girdle fabric below 230‐m depth within the ice column, inconsistent with steady state divide flow. Our interpretation is that this deeper fabric is a remnant fabric from a previous episode of flow, which is currently being overwritten by ongoing fabric development associated with the present‐day flow regime. The preexisting fabric is consistent with ice flow from the south prior to ice‐divide formation, in agreement with models of Holocene ice sheet evolution. These findings apply new constraints to the flow history at Korff Ice Rise prior to divide formation and demonstrate the capacity of radar and seismic measurements to map fabric and thus constrain past ice flow.
7 August 2013 in Geophysical Journal International
Hydraulic overpressure can induce fractures and increase permeability in a range of geological settings, including volcanological, glacial and petroleum reservoirs. Here we consider an example of induced hydraulic fracture stimulation in a tight-gas sandstone. Successful exploitation of tight-gas reservoirs requires fracture networks, either naturally occurring, or generated through hydraulic stimulation. The study of seismic anisotropy provides a means to infer properties of fracture networks, such as the dominant orientation of fracture sets and fracture compliances. Shear wave splitting from microseismic data acquired during hydraulic fracture stimulation allows us to not only estimate anisotropy and fracture properties, but also to monitor their evolution through time. Here, we analyse shear wave splitting using microseismic events recorded during a multistage hydraulic fracture stimulation in a tight-gas sandstone reservoir. A substantial rotation in the dominant fast polarization direction (ψ) is observed between the events of stage 1 and those from later stages. Although large changes in ψ have often been linked to stress-induced changes in crack orientation, here we argue that it can better be explained by a smaller fracture rotation coupled with an increase in the ratio of normal to tangential compliance (ZN/ZT) from 0.3 to 0.6. ZN/ZT is sensitive to elements of the internal architecture of the fracture, as well as fracture connectivity and permeability. Thus, monitoring ZN/ZT with shear wave splitting can potentially allow us to remotely detect changes in permeability caused by hydraulic stimulation in a range of geological settings.