ALBERT

Description: We examine spatio-temporal patterns of microseismicity recorded during one month in an underground mine by addressing three key questions: (1) where does the seismicity occur? (2) Why does it occur in these locations? and (3) what triggers it? To obtain accurate locations, we perform a multiplet analysis and use a modified version of the double-difference (DD) relocation method. This approach leads to highly accurate relative event locations and requires groups of multiplets only. Most of the 281 relocated events are close to the main shaft and tunnels; thus we postulate seismicity is facilitated by stresses associated with the potential for subsidence in addition to the hoop stresses acting on the two vertical shafts. Most events occurred during certain hours of the day and there is a 68 per cent correlation with reported rock removal; therefore, it is likely they were triggered by static and dynamic stress perturbations caused by the transportation of debris along tunnels instead of our initial guess that blasting was the principal causative mechanism. Given that seismicity is present around the main shaft but absent close to the second one, we conclude that for seismicity to occur both a favourable stress state and additional external perturbing forces must exist, thus leading to dynamic event triggering in an initially stable stress situation. This analysis provides more insight into anthropogenic processes that might trigger seismicity, thereby facilitating identification of hazardous and potential damage areas in mine settings.

Description: A new Matched Filtering Algorithm (MFA) is proposed for detecting and analysing microseismic events recorded by downhole monitoring of hydraulic fracturing. This method requires a set of well-located template (‘parent’) events, which are obtained using conventional microseismic processing and selected on the basis of high signal-to-noise (S/N) ratio and representative spatial distribution of the recorded microseismicity. Detection and extraction of ‘child’ events are based on stacked, multichannel cross-correlation of the continuous waveform data, using the parent events as reference signals. The location of a child event relative to its parent is determined using an automated process, by rotation of the multicomponent waveforms into the ray-centred co-ordinates of the parent and maximizing the energy of the stacked amplitude envelope within a search volume around the parent's hypocentre. After correction for geometrical spreading and attenuation, the relative magnitude of the child event is obtained automatically using the ratio of stacked envelope peak with respect to its parent. Since only a small number of parent events require interactive analysis such as picking P - and S -wave arrivals, the MFA approach offers the potential for significant reduction in effort for downhole microseismic processing. Our algorithm also facilitates the analysis of single-phase child events, that is, microseismic events for which only one of the S - or P -wave arrivals is evident due to unfavourable S/N conditions. A real-data example using microseismic monitoring data from four stages of an open-hole slickwater hydraulic fracture treatment in western Canada demonstrates that a sparse set of parents (in this case, 4.6 per cent of the originally located events) yields a significant (more than fourfold increase) in the number of located events compared with the original catalogue. Moreover, analysis of the new MFA catalogue suggests that this approach leads to more robust interpretation of the induced microseismicity and novel insights into dynamic rupture processes based on the average temporal (foreshock–aftershock) relationship of child events to parents.

Description: 〈span〉〈div〉SUMMARY〈/div〉We investigate the possibility of passive monitoring of a salt-water disposal well in British Columbia, Canada, using continuously recorded ambient seismic noise. We find seismic velocity variations induced by a reduction of injection pressure in an effort to mitigate an elevated level of seismicity, most likely associated with the disposal of salt water. The relative velocity variations are derived from time-shifts measured between consecutive cross-correlation functions for each station pair in a surface array composed of five broad-band seismometers. The probable driving mechanisms responsible for the velocity changes are reduced pore pressures and/or lowered poroelastic stresses beyond the injection wellbore, respectively. Hydrologic data (e.g. snow and rainfall), noise energy trends and fluctuations in the incident direction of dominant noise sources do not correlate with the estimated relative velocity variations. Velocity variations are detected ahead of the zone of induced seismicity, thus indicating that seismic interferometry may aid in mitigation efforts to reduce the risk of induced seismicity by (1) providing verifiable and repeatable measurements of physical changes within the surrounding area and (2) providing hard constraints for modelling efforts to constrain how and where pore-pressure fronts change.〈/span〉

Description: 〈span〉〈div〉ABSTRACT〈/div〉Waveform enhancement methods generally explore lateral coherency in arrivals, often assuming a linear moveout across an array, as exhibited by plane waves. We illustrate how unsupervised dictionary learning combined with orthogonal matching pursuit for feature extraction can be used for signal‐to‐noise ratio (SNR) enhancement. In this strategy, waveform characteristics are directly learned from provided data samples; the created dictionary is then used for signal extraction. This combination prevents the need to set a predefined dictionary, and it becomes computationally efficient because learning is only done on smaller data portions. Because the dictionary is learned from data, there is no assumption regarding wavefront shape or form. Tests on synthetic and field data demonstrate the better denoising performance in terms of SNR enhancement compared to other methods.〈/span〉

Description: Fluids play a critical role in natural and human-induced rock failure. It is unclear, however, if propagation of a tensile fracture is inherently an episodic or continuous process. For example, typical average propagation speeds of hydraulic fracture tips on the order of 1–10 m/min suggest continuous crack growth, possibly at subcritical stress intensities. In contrast, using field observations and numerical and mathematical analyses, we show that fracture growth due to anthropogenic hydraulic fracturing is most likely to occur in an episodic fashion, characterized by stick-split behavior that is analogous to stick-slip motion of earthquakes. The stick-split mechanism is regulated by cyclic variations in fluid pressure near the crack tip, in which each successive failure produces a local pressure drop that temporarily halts or slows fracture propagation. A pressure drop results in partial fracture closure, producing noncontinuous fracture propagation through a process that is reminiscent of hand clapping. Rupture speeds for individual failure events are on the order of the shear-wave velocity of the medium; thus, continuous crack growth is not a likely mechanism for anthropogenic hydraulic fracturing treatments despite slow average tip propagation speeds.

Description: The most common approach to seismic triggering is to compare short-term averages (STA) with long-term averages (LTA) of transformed amplitudes. In recording environments where this technique is of limited use, hidden Markov models (HMMs) are increasingly used for statistical event detection and classification, but these require training data and are often susceptible to false positive detection errors. In this work, we introduce an adaptive STA–LTA triggering algorithm that uses STA and LTA of state probabilities defined by restricting an HMM to a two population model of outliers in background noise. Monte Carlo simulations of noise and synthetic events are used to investigate detector sensitivity using statistical properties of latent states. We compare our method with traditional STA–LTA triggering on real data recorded by a 12-station vertical borehole array near Hoadley gas field, Alberta, Canada. These tests suggest that our method is more accurate when dealing with closely spaced events and is less susceptible to false positive detection errors. When existing picking algorithms are adapted for HMM STA–LTA, the result is improvement in total picks, accuracy, and consistency. A narrow range of detection thresholds is optimal for a wide range of signal-to-noise ratios; this suggests HMM STA–LTA may be less sensitive to analyst parameter choices than even traditional STA–LTA.

Description: Long-period long-duration (LPLD) events are tremorlike signals that have been observed during monitoring of hydraulic-fracture treatment programs. LPLD events have been interpreted to reflect slow deformation processes on fractures or faults that are misoriented for reactivation with respect to the present-day stress field. Regional earthquakes could easily be mistaken for LPLD events because both are characterized by similar frequency content ( $$ 〈 100\hbox{ \hspace{0.17em} }\hbox{ \hspace{0.17em} }\mathrm{Hz}$$ ) and duration ( $$\sim 1\hbox{ \hspace{0.17em} }\hbox{ \hspace{0.17em} }\mathrm{min}$$ ). Using data from a 10.5-month continuous downhole deployment of a 15-Hz geophone array in a tight-sand gas field in western Canada, we compared recordings of small earthquakes with previously published LPLD events. We determined that regional earthquakes can show similar waveform characteristics to LPLD events, underscoring the importance of distinguishing regional earthquake signals from LPLD events to ensure robust interpretation of reservoir deformation processes.

Description: In this paper, we document the early stage of fault-zone development based on detailed observations of mesocale faults in layered rocks. The vertical propagation of the studied faults is stopped by layer-parallel faults contained in a weak layer. This restriction involves a flat-topped throw profile along the fault plane and modifications of the fault structures near the restricted tips, with geometries ranging from planar structures to fault zones characterized by abundant parallel fault segments. The ‘far-field’ displacement (i.e. the sum of the displacement accumulated by all the fault segments and the folding) measured along the restricted faults exhibiting this segmentation may have flat-topped shapes or triangular shapes when fault-related folding is observed above the layer-parallel faults. We develop a model from the observations. In this model, during the course of restriction, a fault forms as a simple isolated planar structure, then parallel fault segments successively initiate to accommodate the increasing displacement. We assume that, eventually, the fault propagates beyond the layer-parallel fault. This model implies first that fault widening is controlled by the fault capacity to propagate vertically in the layered section. Likewise, owing to restriction, fault growth occurs with non-linear increases in maximum displacement, length and thickness.

Description: Robust event detection and picking is a prerequisite for reliable (micro-) seismic interpretations. Detection of weak events is a common challenge among various available event detection algorithms. In this paper we compare the performance of two event detection methods, the short-term average/long-term average (STA/LTA) method, which is the most commonly used technique in industry, and a newly introduced method that is based on the power spectral density (PSD) measurements. We have applied both techniques to a 1-hr long segment of the vertical component of some raw continuous data recorded at a borehole geophone in a hydraulic fracturing experiment. The PSD technique outperforms the STA/LTA technique by detecting a higher number of weak events while keeping the number of false alarms at a reasonable level. The time–frequency representations obtained through the PSD method can also help define a more suitable bandpass filter which is usually required for the STA/LTA method. The method offers thus much promise for automated event detection in industrial, local, regional and global seismological data sets.

Description: 〈span〉〈div〉ABSTRACT〈/div〉Waveform enhancement methods generally explore lateral coherency in arrivals, often assuming a linear moveout across an array, as exhibited by plane waves. We illustrate how unsupervised dictionary learning combined with orthogonal matching pursuit for feature extraction can be used for signal‐to‐noise ratio (SNR) enhancement. In this strategy, waveform characteristics are directly learned from provided data samples; the created dictionary is then used for signal extraction. This combination prevents the need to set a predefined dictionary, and it becomes computationally efficient because learning is only done on smaller data portions. Because the dictionary is learned from data, there is no assumption regarding wavefront shape or form. Tests on synthetic and field data demonstrate the better denoising performance in terms of SNR enhancement compared to other methods.〈/span〉