Understanding hydraulic fracturing is crucial to improving the stimulation of unconventional reservoirs and increasing fluid production. This study develops a novel seismic monitoring technology using distributed acoustic sensing (DAS) and surface orbital vibrators (SOV) to capture fracture seismic response and mechanical properties at high temporal intervals. We analyze continuous time-lapse vertical seismic profiling (VSP) data acquired every hour during the first nine days of treatment of an unconventional reservoir in the Austin Chalk/Eagle Field Laboratory. The VSP data contain clear seismic signals scattered from the activated fractures. The spatiotemporal changes of the fracture reflectivity revealed by the SOV/DAS data correlate well with the observations of fracture locations inferred from low-frequency DAS data. These results capture the fracture opening and closure processes, as well as highlight potential prestage activations of the fractures due to hydraulic connectivity with preexisting fracture systems. Therefore, analysis of the presented data set provides a unique opportunity to understand fracture initiation and subsequent evolution, not only in the context of unconventional resources but also in enhanced geothermal systems.

The monitoring of hydraulic fracturing processes in unconventional reservoirs plays a crucial role in understanding subsurface hydromechanical response and optimizing fluid production, particularly in systems with minimal matrix permeability (e.g., Zoback & Kohli, 2019). The growth of a hydraulic fracture is usually predicted using modeling algorithms based on rock mechanics and subsequently characterized using direct field measurements, such as offset cores, and indirect measurements, such as borehole imaging and microseismic monitoring (Maxwell, 2014; Birkholzer et al., 2021; Maity and Ciezobka, 2021). Microseismic monitoring is a particularly useful technique in the context of hydraulic fracture characterization. By analyzing the distribution of hundreds of discrete induced events and identifying clusters, such measurements can constrain fracture planes at a distance from stimulation wells. Although microseismic monitoring is undeniably critical in the context of hydraulic fracturing, microseismic clouds typically provide only a vaguely defined stimulated reservoir volume (SRV) (Mayerhofer et al., 2010). However, the linkage between the fracture growth and seismicity is typically very complex and depends on many factors apart from the fracture geometry, such as mechanical properties of the reservoir and its stress state, seismogenic capacity of preexisting faults and fractures, and many others (Eisner et al., 2010; Maxwell, 2014; Kuang et al., 2017; Glubokovskikh et al., 2023). Furthermore, not all microseismic events indicate fractures hydraulically connected to either production wells or other fractures.

Time-lapse vertical seismic profiling (VSP) can provide insights into fracture geometry and fracture compliance (Turpening, 1984; Wills et al., 1992; Meadows and Winterstein, 1994; Willis et al., 2008). In the past several decades, advances in fiber-optic sensing have changed the paradigm of VSP acquisition through the availability of distributed acoustic sensing (DAS) data (Miller et al., 2012; Parker et al., 2014), a dramatic departure from prior sparse wireline geophone arrays because of its wide acoustic aperture. DAS uses conventional single-mode fibers to create an array of distributed strain-rate sensors that record seismic data. The method sends a series of laser pulses along a fiber-optic cable and monitors changes in the phase of the backscattered light caused by elongation/deformation of the cable (Hartog, 2017). The use of DAS VSP demonstrates significant advantages over the use of conventional technology, as DAS provides on-demand seismic data along a single cable, generating thousands of receivers spanning the entire length of the well. DAS can be used with permanently installed cables, which makes it an ideal alternative for permanent reservoir monitoring (Mateeva et al., 2017). The popularity of DAS extends to other areas of seismology, such as ambient noise analysis, microseismic detection, or more specific analysis such as the use of deep-sea fiber-optic cables for characterizing oceanic noise and marine processes (Dou et al., 2017; Lindsey et al., 2017, 2020; Martin et al., 2017; Lellouch et al., 2020).

There has been a large increase in the use of DAS for VSP acquisition in the past decade, with DAS VSP being applied widely in the oil and gas industry for characterization of hydrocarbon reservoirs and, more recently, in carbon capture and storage projects for CO2 plume tracking (Pevzner et al., 2021; Isaenkov et al., 2022). In the context of characterization of hydraulic fracturing, the use of VSP data recorded with DAS has been explored in a small number of prior studies. Bakku et al. (2014) show that good-quality VSP data can be acquired with DAS in a hydraulic fracture treatment well. Byerley et al. (2018) provide an estimate of fracture geometry, such as fracture height and compliance, from DAS VSP data and demonstrate that DAS VSP is sensitive to P-wave time delays. More recently, Titov et al. (2022) demonstrate the use of interstage DAS VSP to analyze the amplitudes of observed PS waves scattered from the SRV.

Continuous seismic monitoring can allow for real-time observations of the hydraulic fracturing, such as the fracture initiation process and fracture closure. In this paper, we explore the use of a novel approach for continuous seismic monitoring using permanent surface seismic sources, called surface orbital vibrators (SOV), in combination with DAS to build a continuous and autonomous approach for acquiring high temporal-resolution time-lapse seismic data. As part of the Austin Chalk/Eagle Ford Field Laboratory (ACEFFL), we demonstrate the use of SOV/DAS to acquire nine days of continuous VSP data, recorded during the hydraulic fracturing treatment of an unconventional reservoir. We aim to (1) demonstrate the use of SOV/DAS for continuous time-lapse seismic monitoring of hydraulic fracturing processes at high temporal resolution and (2) provide qualitative insights into hydraulic fracture behavior, such as initiation, closure, and compliance, through the analysis of scattered signals in the SOV/DAS data and low-frequency strain-rate anomalies. Our work builds on the results of Byerley et al. (2018) and Titov et al. (2022) by capturing fracture reflectivity changes at high spatial-temporal resolution. By acquiring SOV/DAS data without interruption throughout the fracturing process, the data revealed unique observations on fracture stimulation. Overall, our results highlight the added value of continuous active seismic monitoring for understanding the sophisticated hydromechanical processes that occur during fracturing.

Continuous seismic monitoring with SOV/DAS

SOVs are alternating current induction motors that generate seismic energy by rotating eccentric weights, inducing circularly polarized waves that travel through the earth, which can then be decomposed into vertical and horizontal components by either summing or subtracting each direction of rotation (Daley and Cox, 2001). They are designed to sweep through a range of frequencies, accelerating until the peak frequency and then decelerating. Peak frequency is limited by the size of the motor as rated by the manufacturer and can range from approximately 50 Hz to 200 Hz for a 15 to 225.5 kN (approximately 1.5 to 23 tonne-force). The sweep of the source is recorded by a near-field 3C geophone, where the pilot can be used for further processing. Because SOVs are rotary sources, the force of the signal increases as frequency squared, meaning that the higher the frequency of the sweep, the higher the energy generated. A deconvolution process is preferred to crosscorrelation to collapse the pilot sweep during processing because the frequency spectrum is unbalanced, thus minimizing the occurrence of correlation sidelobes.

Similar rotary sources have been demonstrated successfully in previous studies. Kasahara et al. (2017) demonstrate the application of the accurately routinely operated signal system (ACROSS). One other notable rotary source is the controlled accurate seismic source (CASS) (Wang et al., 2020). The main difference between the SOVs used in our study and other rotary sources, similar to the ACROSS and CASS, is that the SOVs are not phase controlled, which reduces the complexity of the system. The SOV sources are based on off-the-shelf components, which improves the affordability and accessibility of the system.

An SOV source is ideal for time-lapse seismic monitoring, especially at the long-term time scale. They can be installed permanently, eliminating one of the biggest uncertainty components of time-lapse seismic data, which is location inaccuracy (Lumley, 2001). The sources can also be programmed to be autonomous and continuous, generating highly repeatable seismic energy at high temporal sampling with minimal human intervention, as demonstrated by Cheng et al. (2021) and Correa et al. (2021). Cheng et al. (2021) demonstrate the application of the SOV/DAS combination during an instrumentation test at another Eagle Ford unconventional production site and observe a phase repeatability over sequential temporal epochs of approximately 200 μs while recovering reflection energy from intervals below 2.5 km.

Given the unique characteristics of DAS and SOV for long-term seismic surveillance, the combination of the two technologies, with SOVs as the seismic sources and DAS as the seismic receivers, can allow for continuous and autonomous seismic monitoring. With this, SOV/DAS technology can revolutionize the way time-lapse data is acquired. In the context of hydraulic fracturing, the use of SOV/DAS represents a unique opportunity to “listen” to reservoir failure/fracture not only between stages but also during the fracturing process. In this paper, we demonstrate the first application of SOV/DAS technology for monitoring fracturing initiation and fracture evolution during and after hydraulic fracturing treatment.

The data presented in this paper were acquired as part of the ACEFFL project, sponsored through a joint partnership between SM Energy and the United States Department of Energy (DOE) with the objective of demonstrating a multidisciplinary approach to map hydraulic fractures and the SRV. The project seeks to use advanced diagnostic methods and new monitoring alternatives to characterize the SRV and to improve the effectiveness of shale oil production (Hill et al., 2020).

The main objectives of the project are to improve understanding and gain insights into the flow, transport, and mechanical processes occurring during and after stimulation. To achieve this, we aim to integrate multiphysics data sets, such as microseismic monitoring, VSP, fiber-optic strain monitoring, and bottomhole pressure measurements, to capture a holistic view of the fracturing process. We demonstrate the application of real-time monitoring for tracking fracture propagation through the application of SOV/DAS as an innovative active seismic monitoring methodology. In addition, the project aims to evaluate the viability of using fiber-optic sensing technologies in capturing continuous and simultaneous data streams from distributed temperature sensing (DTS) and distributed strain sensing (DSS), in addition to DAS.

A zipper-fracking style stimulation was performed within the Austin Chalk Formation along three boreholes, two of which, well 5 and well 3 (Figure 1), are instrumented with fiber-optic cables. The cables are permanently installed behind the casing, consisting of conventional single-mode fiber, multimode fiber, and enhanced sensitivity single-mode fibers (Constellation, Silixa LLC). We use a combination of fiber-optic sensing technologies and recorded continuous temperature (DTS) on the multimode fiber, static strain (DSS) on the single-mode fiber, and low-frequency strain rate (DAS) on the constellation fiber. The constellation fiber was chosen for the acquisition of DAS instead of standard single mode, as the fiber is engineered to enhance the light backscatter and thus improve the signal-to-noise ratio of the seismic signal up to 20 dB (Naldrett et al., 2020).

We deployed an array of SOV sources at five locations (SOV1, SOV2, SOV3, SOV4, and SOV5) along the horizontal (lateral) section of the well (Figure 1). As discussed previously, DAS senses changes in the axial elongation of the fiber; thus, it is a 1C measurement, and it is mostly sensitive to particle motion polarized along the fiber axis (Correa et al., 2017). The challenge in our case was to provide a source geometry that would generate sufficient seismic energy to be detected at approximately 2 km depth on a horizontal fiber along the lateral section of the well, wherever treatment occurred. Therefore, the location of the SOV sources was decided based on the expected sensitivity of the direct P wave along different lateral sections.

To increase source strength and bandwidth, we deployed a combination of motors consisting of 100 kN (approximately 10 tonne-force) and 150 kN (approximately 15 tonne-force) motors at each source location. Both motors were set to run simultaneously to augment source strength. We selected a different sweep design for each motor size; the 100 kN motor was used to improve high-frequency response and was set to accelerate up to 80 Hz, whereas the 150 kN motor focused on the low-frequency band; the larger motor was set up to generate a sweep with quadratic acceleration up to 55 Hz. Figure 2 shows an example of one single sweep recorded by the vertical component of the near-field geophone as both motors spin simultaneously. The spectrogram shows the fundamental frequency generated by each motor, as well as the combinational frequencies and harmonics of both motors. These harmonics are likely generated through nonlinear mixing at the plate/motor or plate/foundation interface. One sweep is 2.5 min in total.

Data acquisition

The main objective of the SOV/DAS survey was to acquire high temporal-resolution VSP data to allow frequent snapshots of the hydraulic fracturing process, from fracture opening to closure. In this section, we detail the acquisition design and parameters for the SOV/DAS survey.

We recorded continuous and simultaneous DAS data along well 5 and well 3 interrogator units connected to engineered single-mode fibers deployed in each well. The units were set to acquire data at 1 kHz time sampling and 1 m spatial sampling, using a fixed 10 m gauge length. The DAS interrogator units were global positioning system (GPS)-synchronized to allow time coordination with the SOVs. Using the same cable, we also recorded simultaneous and continuous DTS data on both wells and Brillouin DSS data on well 3. The data were saved continuously to high-performance redundant array of independent disks storage units on site. Approximately 2 Tb of combined data sets were recorded per day during treatment. Treatment was initiated in well 5, and after seven days of treatment, zipper-fracking-style treatment commenced in the other two wells. Therefore, the first seven days essentially consisted of a single well operation in well 5, whereas well 3 functioned as a monitoring well.

Due to the location of each SOV relative to the fiber, the fiber’s sensitivity to the direct P wave varies along the lateral section of the well. As the incident P waves arrive parallel to the fiber axis, the sensitivity of the DAS measurement for the first arrival is optimized. In this context, we decided to only operate the SOV locations where the recording of the incident direct P wave would be optimized for the stages around the toe of the well. Therefore, during the first days of treatment, we only used the SOV3 and SOV5 source locations.

As mentioned in the previous section, one sweep is 2.5 min duration, and we operate each of the two SOV sources for 10 consecutive sweeps, pausing for 30 s between sweeps, for a total of 30 min. The SOV sources were programmed to run continuously and automatically, and after the end of a 30 min cycle, the next SOV source would start its 30 min cycle. As a result, SOV3 and SOV5 operated for 30 min at every hour.

During the first nine days of treatment, we acquire continuous VSP surveys using DAS data collected at well 3 and SOV3 and SOV5. After the first nine days of continuous 24/7 operation, concerns regarding interference of the SOV source signal with microseismic monitoring were raised; as a result, an operational decision was made to stop the continuous operation of the SOV during treatment. Therefore, in this paper, we will focus on the high temporal-interval time-lapse VSP data acquired using SOV3 and SOV5 during the first nine days of treatment.

On-site data processing

We designed an edge computing solution to automate the processing of continuous data acquisition. The objective of the edge computing was to generate data products that could be easily transferred on the cloud, as well as generate daily diagnostic products that could be used to ensure quality of the data acquisition during treatment. Due to limitations in telemetry bandwidth, transfer of the raw data off site was not possible, necessitating either edge analysis or manual transport of raw data. To circumvent these constraints, we built a quasi-real-time processing flow that was automatically triggered after a day of acquisition.

Figure 3 shows the different streams of data, from acquisition to processing. After acquisition of the DAS, DTS, and DSS data, the data were saved to local storage servers and automatically processed using the edge computing system. The DAS data undergo separate processing flows to generate two data products. One DAS product is the low-frequency DAS (LFDAS), which was generated after resampling of the raw DAS data and converting the interrogator unit to strain rate unit (Zhu et al., 2023). The LFDAS product refers to the low frequency range (<0.5 Hz) of the DAS data, and it shows slow mechanical and thermal disturbances in the dynamic strain field. The use of LFDAS for hydraulic fracture diagnosis has become relatively popular in recent years for the detection of fracture hits in monitor wells (Jin and Roy, 2017).

The raw DAS data are also processed as a conventional VSP survey to generate a series of seismic gathers for each SOV sweep acquired. For this processing flow, given that the DAS data is acquired continuously, we first match the DAS GPS time to each sweep start time. After matching the correct start time of the sweep to the DAS data, we deconvolve the DAS signal with the sweep signal. Deconvolution is preferred in this case in comparison with crosscorrelation because it minimizes the effect of correlation sidelobes caused by the unbalanced frequency spectrum of the pilot sweep, as the amplitude of the signal is proportional to frequency squared.

After the data products are generated, they are automatically uploaded to the cloud. We should note that the processing steps reduce data volume by almost two orders of magnitude, thus allowing transfer even at sites with bandwidth-limited telemetry links. The subsequent phases of the seismic data processing are done in house. After obtaining the deconvolved sweeps, each 10-sweep cycle is stacked, resulting in one VSP gather per hour for each source position. The stacked VSP gathers go through a noise attenuation flow consisting of applying a band-pass filter of 5 to 80 Hz, followed by a 2D median spatial filter to remove the common-mode noise caused by ambient vibrations in the vicinity of the DAS interrogator. Given that we have hourly VSP gathers during nine days of operation, the large amount of data allows for filtering in the receiver domain. Therefore, an additional 2D median spatial filter is applied after sorting the data into the receiver domain to minimize the effect of noisy traces caused by disturbance in the well due to stimulation and surface activities.

The scattering generated by a newly created fracture is expected to have reflections of relatively small amplitudes in comparison with the direct wavefield. To improve the visibility of possible reflections from fractures, we remove the direct P-wave and multiples by flattening the direct arrival and applying a 2D median spatial filter. The result is a VSP shot gather containing only the down- and upgoing reflections. As part of the time-lapse analysis, we take the difference of each VSP gather with the baseline. The units are proportional to phase change and were not converted to strain rate.

Baseline data

Figure 4 shows the baseline data for SOV3 and SOV5, acquired in well 3. Well 5 is not displayed because the waveforms are practically identical to well 3 due to its close proximity. The baseline data have a total of 10 repeated sweeps stacked. The vertical portion of the fiber extends to approximately 2 km depth, after which it transitions to mostly horizontal, and extends to approximately 5 km measured depth. The seismograms for SOV3 and SOV5 exhibit a rich wavefield along the entire fiber-optic cable. We see a combination of direct P (blue arrow) and S waves (green arrow), as well as reflected wavefields (orange arrow) that can be observed. Note that the fiber is almost insensitive to the direct P wave along the vertical portion of the well for SOV5 due to lack of broadside sensitivity inherent to the DAS measurement (Correa et al., 2017). The baseline data is used subsequently in the processing to subtract from the monitor data sets to identify time-lapse variations. The initial stage of treatment occurs at the toe of the well, therefore, we expect changes within approximately 4 to 5 km measured depth.

Seismic scattering signatures from SOV/DAS shot gathers

In this paper, we focus on the analysis of well 3 and SOV3. The data acquired in well 3 for source SOV5 exhibit weak scattered energy in the monitor records, possibly due to the geometry of the well in relation to the source and fractures.

The offset VSP data acquired with the SOV/DAS system was processed daily on site by applying an automated processing flow using an edge computing structure and processing framework, as described in the previous section. The processed results, or data products, were subsequently uploaded automatically to the cloud for further in-house processing. Figure 5 shows an example of the processed seismograms for well 3, using SOV3 source location, which reveals a series of scattered signals from the direct P wave.

The data set shown in Figure 5a is the output, as acquired from the DAS interrogator unit, after deconvolution of the source signal and stack of 10 repeated sweeps. The data set contains clearly visible direct waves; however, it presents strong high-frequency random noise and common-mode noise, a type of noise commonly seen on DAS data due to movement of the interrogator unit. The data set improves significantly after band-pass filtering to remove the high-frequency noise and 2D median spatial filtering to remove the common-mode noise, though random noise and bad traces are still present in the data (Figure 5b). At this stage in the processing, the noise still masks potential scattered waves from fractures. Further in-house processing, with the application of additional 2D median spatial filtering in the receiver domain (Figure 5c), uncovers a series of scattered events from the direct P wave. The noise attenuation process in the receiver domain is only possible due to the high volume of data in the slow time domain. A median filter is then applied along the direct arrival to isolate the direct P wave and remove it from the entire wavefield. This process reveals the existence of multiple coherent scattered events (Figure 5d) along the toe of the well.

The example shown in Figure 5 shows the data acquired after the completion of seven fracturing stages. Note that the reflections seen on Figure 5d include down- and upgoing wavefields. At approximately 4.5 km measured depth, a strong upgoing reflection is observed, which could be interpreted as scattered P and PS waves. Before 4 km measured depth, some downgoing reflections also emerge despite no preceding stimulation in the area, suggesting the potential existence of other reflectors, such as natural faults and fractures. This interpretation will be confirmed in the next section after the subtraction of the seismograms of the baseline data.

Analysis of the time-lapse seismic data

Figure 6 shows 10 postprocessing snapshots from the sequence of hourly SOV/DAS seismograms acquired during the first nine days of stimulation. During this period, 12 fracturing stages were completed, starting from the toe of well 5. To highlight the scattered signals from fractures, the baseline survey is subtracted from the seismograms, and they are magnified at the toe of well 3. The hourly VSP data revealed a series of coherent scattered arrivals from either reactivated preexisting fractures or newly created hydraulic fractures from the injection well, well 5, to the monitor well, well 3. The scattered events appear post stage and typically decrease in apparent amplitude within days.

Note the absence of scattering energy on day one, before the start of treatment, which indicates the good repeatability of the signal. By the end of the second day, several strong reflections emerge between the toe and a measured depth of 4800 m, after the completion of four stages (examples of reflections indicated by orange arrows in Figure 6). Although all four stages are located close to the toe of well 5, we observe a few traveltime curves between 4400 and 4600 m (green arrow), which implies that there might be a fracture system that was activated by relatively remote stimulations. On day three, the reflections observed on the previous day increase in amplitude. By the end of the fourth day, a strong reflection appears at approximately 4550 m measured depth, while the previous reflections decrease in intensity. The reflection at approximately 4550 m measured depth gains in intensity by the next day, lessening by the end of day six. On day seven, a reflection appears at approximately 4400 m measured depth, while the previous reflections are almost imperceptible. The scattered events appear as the well 5 stimulation stages are completed, sequentially toward the heel. By the ninth day, most of the scattered events are receding and appearing dimly.

To understand the dynamics of fracture behavior through seismic measurements, we analyze the variations in fracture reflectivity over time. To compute this parameter, we use fully processed shot gathers after baseline subtraction and flatten the reflections along the time coordinate, so that each reflection is aligned horizontally along time. After flattening, a top mute is applied above the first break, and a bottom mute is applied 100 ms after the first break. Each 1 hr flattened snapshot is then stacked into one single trace. The stacked trace then contains the averaged amplitude of each reflection along relative time. The time-depth relationship obtained from the VSP is used to convert the relative time to depth.

Figure 7a shows the evolution over time of the SOV/DAS fracture scattering amplitudes in well 3, as well 5 is undergoing treatment, where each wiggle represents the amplitude of the scattered events at each hour. Figure 7b shows the simultaneously acquired LFDAS data, where blue indicates compressive strain and red indicates extension. Hydraulic fractures initiated from well 5 and intersecting well 3 show up as frac-hit patterns in LFDAS, where we can observe extension strain at the tip of a fracture and compression of its sides (Zhu et al., 2023). Figure 7c shows the pressure in the treatment well, well 5, acquired with a wellhead pressure gauge; each increase in pressure marks the start of a fracturing stage. The next section discusses the observed patterns in the data in detail.

Hydromechanical causes for geophysical anomalies

A rigorous interpretation of the seismic and strain anomalies shown in Figures 6 and 7 requires application of sophisticated rock-physics models. In addition, seismic and strain anomalies have limited resolution, thus they may represent an interference of various colocated events that occur simultaneously (Sherman et al., 2019; Zhang et al., 2020). Here, however, we may discuss some hydromechanical processes that explain the observed signals.

Figure 8 emphasizes the consistency between the anomalies detected by the LFDAS and the seismic scattering in the SOV/DAS data. Figure 8a combines the seismic reflectivity evolution (Figure 7a) with the strain-rate anomalies (Figure 7b). The next three subplots (Figure 8b8d) magnify the three peculiar intervals indicated by the yellow arrow. The curves, called strain-rate and normalized reflectivity, correspond to the signed root-mean-squared values computed in a 20 m interval in the middle of the magnified section. Although the geophysical signals induced by reservoir stimulation operations vary for different depth intervals, the data reveal a repeated pattern:

  • Phase 1: Seismic scattering appears along the entire segment of the fiber-optic cable, around the time of a borehole operation that manifests itself by a strong strain anomaly on the first day.

  • Phase 2: The strength of the seismic signals decreases in the parts of the reservoir that are located further away from the toe and are not reactivated by the fluid injections.

  • Phase 3: The apparent changes of the seismic reflectivity are consistently preceded by bursts of strain rate, which indicate an onset of reservoir stimulation.

  • Phase 4: Finally, the seismic anomaly gradually fades away once the stimulation stages progress toward the heel.

  • Phase 5: The seismic amplitudes flatten out after another strain anomaly extended through the entire borehole.

To explain the origin of phases 1–4, we rely on the linear slip model of fracture deformation (Schoenberg, 1980; Schoenberg and Sayers, 1995). Due to high compliance of fractures, propagating seismic waves become discontinuous across the fractures; the waves induce a significant displacement of the opposite surfaces of the fracture. The size of the displacement discontinuity depends on the moduli of the fracture-filling material, the fracture aperture (Fehler, 1982; Oelke et al., 2013), and the contacts between the fracture surfaces (Sevostianov and Kachanov, 2008; Glubokovskikh et al., 2016). Thus, we expect to see a strong response from a pressurized or hydropropped fracture, which is wide open, has only a few small contacts, and contains proppant grains suspended in fracturing fluid without direct load-bearing contact. Otherwise, contacting asperities at fracture surfaces and dense proppant packs transmit the normal and tangential displacements through the narrow fracture in a stress-dependent manner. At higher effective stress states, the propped fracture may have a minimal seismic contrast with the surrounding formation.

We hypothesize that phase 2 in Figure 8c and 8d corresponds to pressure dissipation inside fractures opened on the first day of the treatment of well 5 and a corresponding increase in normal stress across the fracture plane. In the interval 4900–5020 m (Figure 8b), phase 2 is absent because this interval remains pressurized, as it is close to the first stimulation stage (the first stage happens before the time frame displayed at a section of the toe without fiber). Once the fracturing operations approach a particular interval, the fracture-associated reflectivity slowly increases due to hydraulic connectivity with adjacent fracturing operations, until the interval is stimulated directly and existing fractures are pressurized again (phase 3). After that, the pressure dissipates and the proppant is slowly loaded within the fracture, effectively reducing the impedance contrast visible to seismic waves (phases 4 and 5).

The preceding interpretation ignores the fact that fractures may grow during stimulation. Each fracture patch will contribute to the observed scattered amplitudes depending on its position and orientation relative to the receiving fiber-optic cable, and thus fracture configuration plays an important role in the observed seismic amplitudes. In general, a larger fracture produces a stronger response than a smaller fracture with the same compliance. Also, fractures that intersect the monitoring borehole laterally and have sufficient height above the borehole will produce a strong response. If we only consider fracture geometry, phase 1 may indicate creation/activation of relatively small fractures that do not intersect well 3, as we do not see a clear burst of strain rate in this interval. Then, the fracture gradually closes from its edges as pressure dissipates during phase 2. During a direct stimulation stage, fractures rapidly grow laterally and vertically and intersect the borehole (phase 3). Such behavior is most obvious in the interval 4580–4700 m (Figure 8a and 8c): the seismic scattering is apparent (phase 3.1), even though the LFDAS measurements indicate compression, not extension. Then, a strain-rate anomaly manifests as a boost of reflectivity (phase 3.2), which is followed by another similar event (phase 3.3) induced by the next stimulation stage.

The last aspect that we discuss here involves the limited seismic resolution of the SOV/DAS data. The dominant seismic wavelength is approximately 100 m, which implies that seismic scattering from adjacent fractures, separated by less than 25 m, is indistinguishable. Thus, some of the events in Figure 8b8d may correspond to strain-rate anomalies produced by new hydraulic fractures that propagated from well 5 to well 3, parallel to the ones that were activated by earlier fluid injections. For example, interval 4400–4520 m (Figure 8d) only has a detectable reflectivity increase after a fracture hits the monitoring borehole. The amplitude increases in phases 3.2 and 3.3 in Figure 8b and 8c may have the same explanation.

Although the presented data set is extremely rich in terms of the subsurface characterization, we still may not unambiguously constrain the geometry of the fractures and their permeability. This would require a comprehensive, almost forensic, rock-physics analysis of the seismic anomalies in combination with the strain rates and microseismic clusters. We mentioned the likely candidates to explain the observed data: fracture opening, deposition of stiff clusters of interconnected proppant grains, contacting asperities on the fracture surfaces, vertical and lateral growth of the existing fractures, and formation of the new fractures in between the existing ones. Typically, VSP interpretation only considers the last aspect and pore pressure effects. In reality, all of these processes affect the seismic response simultaneously and may not be decoupled. Therefore, we believe that for our field experiment, the standard approaches to the analysis of fracture scattering (Binder et al., 2020; Titov et al., 2022) may be an oversimplification. One must rely on sophisticated coupled geomechanical simulations to estimate the hydromechanical fracture properties based on the pressurization/relaxation times and number of fracture hits. Such estimates would aid a robust prediction of the performance of stimulated reservoirs.

The use of SOV/DAS technology for autonomous and continuous seismic data acquisition is a unique approach to monitoring, offering insights into the hydraulic fracturing process with rapid snapshots of the subsurface. This is particularly useful for monitoring aseismic fracture processes such as pore-pressure diffusion and closure.

One point of consideration when using permanent systems for fracture monitoring is the reduction in time-lapse noise traditionally associated with source position and coupling variations. This not only improves the quality of data but also enhances the ability to observe small changes, paving the way for detecting subtle shifts in subsurface properties. Having said this, the VSP data acquired in this project using the SOV/DAS system underwent a rather simple data processing flow, and yet it was able to detect considerable changes in the subsurface due to fracturing.

We have detected coherent scattered arrivals from newly created fractures, initiated after multiple fracturing stages. The high temporal sampling of the VSP data allowed observation of fracture opening and activation, as well as fracture relaxation and pressure diffusion over time. We observed a strong event associated with the stage where the operator reported the largest volume injected.

Furthermore, we analyzed the amplitude increase and decay of the scattered events as a function of time. As fractures hit the monitoring well, the strain response from the LFDAS and the scattered wave amplitudes were perturbed. This association not only aids in the deeper understanding of the hydraulic fracturing process in terms of frac hits but also validates the consistency and reliability of SOV/DAS as a monitoring tool. An interesting observation was the rapid increase in amplitude from scattered waves in synchrony with rising pressure, immediately after the well treatment started. This pattern, particularly evident during the sixth stage, persisted for approximately six days, showcasing the longevity and after-effects of the hydraulic treatment.

The correlation between the low-frequency strain-rate anomalies and changes in seismic reflectivity in our data set is apparent, indicating a direct relationship between fracture activation and changes in their reflectivity. We discussed a few potential hydromechanical mechanisms that may underly the observed geophysical response. These include fracture opening/closure, proppants deposition, lateral/vertical growth of preexisting fractures, and creation of new fractures. An interplay between these processes could easily match the observed evolution of the fiber-optic signals but any such interpretation would have high uncertainty without careful reservoir simulations, which could provide characteristic temporal and spatial scales for various hydromechanical processes.

We demonstrated the application of a novel technology for the acquisition of high temporal-resolution time-lapse VSP data using fiber-optic DAS and permanent seismic sources, SOVs. The SOV/DAS system allowed for high-quality VSP seismic data acquired at every hour during the first nine days of stimulation of an unconventional reservoir as part of the ACEFFL. The deployment of edge computing infrastructure, coupled with further data processing, has provided a comprehensive view of hydraulic fracturing dynamics in close to real time. As a result, we detected coherent scattered arrivals from newly created fractures and observed their dynamics, with fracture opening and relaxation over time, as well as potential prestage activations due to hydraulic connectivity with preexisting fracture systems.

In conclusion, the use of SOV and fiber-optics DAS is a novel strategy for hydraulic fracturing monitoring, potentially transforming our understanding of fracture dynamics with broader applications across tight formations. The findings of the paper can also be applied to other synergistic topics, such as enhanced geothermal systems. We hope these findings can improve future operations, illustrating the importance of continuous monitoring and the integration of multiple data streams to gain a holistic understanding of subsurface dynamics during hydraulic fracturing operations.

This project was funded by the U.S. DOE, Assistant Secretary for Fossil Energy and Carbon Management, Office of Fossil Energy and Carbon Management, Resource Sustainability Program through NETL, under contract no. DE-AC02-05CH11231. The authors are grateful to SM Energy for providing the data and permission to publish this work, and in particular Robert Bohn and Erich Kerr for their significant assistance. The authors thank D. Hill and D. Zhu (Texas A&M University) for their leadership in the ACEFFL project.

Data associated with this research are confidential and cannot be released.

Biographies and photographs of the authors are not available.