Vertical seismic profiling with distributed acoustic sensing images the Rotliegend geothermal reservoir in the North German Basin down to 4.2 km depth

Abstract. We performed vertical seismic profiling at the Groß Schönebeck site in order to gain more detailed information on the structural setting and geometry of the geothermal reservoir, which is comprised of volcanic rocks and sediments of Lower Permian age. During the four-day survey, we acquired data for 61 source positions with the novel method of distributed acoustic sensing (DAS), using hybrid wireline fiber-optic sensor cables deployed in two 4.3 km deep wells. We show that wireline cable tension has a significant effect on data quality. While most of the recorded data has a very good signal-to-noise ratio, individual sections of the profiles are affected by characteristic coherent noise patterns. This ringing noise is a result of how the sensor cable is mechanically coupled to the borehole wall, and it can be suppressed to a large extent using suitable filtering methods. After conversion to strain rate, the DAS data exhibits a high similarity to the vertical component data of a conventional borehole geophone. Upgoing reflections are nevertheless recorded with opposite polarity, which needs to be taken into account during further seismic processing and interpretation. We derived accurate time-depth relationships, interval velocities, and corridor stacks from the recorded data. Based on integration with other well data and geological information, we show that the top of a porous and permeable sandstone interval of the geothermal reservoir can be identified by a positive reflection event. Overall, the sequence of reflection events shows a different character for both wells, which is explained by lateral changes in lithology. The top of the volcanic rocks has a somewhat different seismic response, and no stronger reflection event is obvious at the postulated top of the Carboniferous. The thickness of the volcanic rocks can therefore not be inferred from individual reflection events in the seismic data alone. The DAS method has enabled measurements at elevated temperatures up to 150 °C over extended periods and has led to significant time and cost savings compared to deployment of a conventional geophone chain.


The Groß Schönebeck site is located 40 km N of Berlin in the state of Brandenburg, Germany. It is a research platform operated by the GFZ German Research Centre for Geosciences, which has been set up in order to test if production of geothermal energy from deep-seated reservoirs in the North German Basin is feasible. An enhanced geothermal system (EGS) has been created by hydraulic stimulation of low-permeability sedimentary and volcanic 40 rocks of lower Permian (Rotliegend) age (Huenges et al. 2006;Zimmermann et al. 2010). So far, two deep research boreholes, the former E GrSk 3/90 hydrocarbon exploration well and the Gt GrSk 4/05 geothermal well (referred to as GrSk3 and GrSk4 in the following), exist at the site. For further development of the site, the implementation of a new stimulation concept and drilling of a new well have been proposed (Blöcher et al. 2015).
In order to gain more detailed information on the structural setting and geometry of the reservoir, a 3D seismic 45 survey within an 8 km x 8 km permit area has been carried out in February and March 2017 (Krawczyk et al. 2019). In addition, vertical seismic profiling (VSP) has been performed within two research wells existing at the site. The primary aims of the VSP survey were to establish precise time-depth and velocity profiles, and to image structural elements in the vicinity of the boreholes with higher resolution in three dimensions. A special challenge is the imaging of structures within the reservoir interval of the Rotliegend at 4200 m depth, which is overlain by 50 the 1400 m thick Upper Permian Zechstein salt complex.
The VSP measurement was performed using the novel DAS method. This method is based on optical time-domain reflectometry, and enables to register strain changes along optical sensor cables with high spatial and temporal resolution (Parker et al. 2014). Within recent years, a growing number of VSP surveys has been reported, where the DAS method has successfully been applied using sensor cables permanently installed behind casing or along 55 tubing (e.g. Mestayer et al. 2011;Daley et al. 2013;Götz et al. 2018). This deployment method is very convenient as it allows for data acquisition without well intervention. In cases where such a permanent installation is not possible or has not been performed during construction of the well, a sensor cable can be lowered downhole temporarily, similar to conventional wireline logging. For this wireline deployment method nevertheless only very few experiences exist until now: First tests using an experimental optical wireline logging cable deployed in a 625 60 m well were described by Hartog et al. (2014), while a more extensive DAS walkaway VSP survey has been performed by Yu et al. (2016) in a vertical well to a depth of 4004 m. Within the current study, we report on the results of a DAS-VSP acquisition on wireline cable to a depth of 4256 m, which to the authors knowledge represents the deepest survey currently documented in literature worldwide.

Survey design and data acquisition 65
The target area was defined by the positions of the existing wells, the expected extent of the hydraulic fractures, and the trajectory of the proposed new well. It has a horizontal extent of approx. 700 m x 500 m and a vertical thickness of approx. 300 m. A spiral pattern of 61 source points with offsets between 180 m and 2000 m from the wellheads was chosen, in order to achieve a good 3D coverage of the target area with a uniform distribution of azimuths ( Figure 1). Survey planning was based on well trajectories and geometry of the major geologic units 70 (Moeck et al. 2009), taking into account DAS specific acquisition characteristics like directivity and signal-tonoise ratio. The source point positions were optimized based on ray tracing, using average acoustic properties of the major geologic units from a previous regional seismic survey (Bauer et al. 2010). The actual source point locations were then adjusted according to the conditions within the survey area, i.e. location of roads and agricultural areas, as well as required distances to sensible infrastructures like gas lines or buildings. 75 https://doi.org/10.5194/se-2020-169 Preprint. Discussion started: 12 October 2020 c Author(s) 2020. CC BY 4.0 License.
A listing of the acquisition parameters is contained in Table 1. Energy excitation was performed with four heavy vibrator trucks operating simultaneously at each source position. For acquisition of the DAS data in well E GrSk 3/90 the GFZ hybrid borehole measurement system was used, which allows for deployment of fiber-optic sensors and electric downhole tools in parallel (Henninges et al. 2011). The GrSk3 well is near-vertical (maximum inclination 7.2°), and the fiber-optic data was acquired to a measured depth (MD) of 4256 m below ground level, 80 which corresponds to a true vertical depth (TVD) of 4245.8 m below ground level (note that all depths in this study are given in MD, if not stated otherwise). Within the well Gt GrSk 4/05, which is deviated up to 49° in the reservoir interval, a second wireline cable containing optical fibers was deployed (maximum DAS acquisition depth 4196 mMD / 4126.1 mTVD). This is an experimental optical wireline cable developed by Schlumberger, referred to as optical heptacable (Hartog et al. 2014). This cable was also used to deploy a conventional three-component 85 borehole geophone with acceleration characteristics (VSI Versatile Seismic Imager tool, Schlumberger), in order to record several check shots 1 at specific depths within the GrSk4 well. DAS data was acquired on both cables using two separate Schlumberger hDVS (Heterodyne Distributed Vibration Sensing) units. Fieldwork was carried out within four days from Feb. [15][16][17][18]2017. At the beginning, we performed a start-up test (referred to as day 0 in the following), where suitable source and recording parameters were determined. As a result, we selected a sweep with 10-112 Hz (linear) and 36 s duration for acquisition. For some of the larger offsets, a sweep with reduced frequency range of 10-96 Hz was used. A gauge length of 20 m was selected for online DAS data processing during recording. This value was later adjusted to 40 m during post-processing as a result of an 95 optimization procedure (see Section 3.1). The DAS measurements were recorded with a temporal sampling of 2 ms and a spatial sampling of 5 m spacing across the entire length of the wells.
During the start-up test, we recorded several shots with variation of the wireline cable tension in the GrSk3 well, in order to test the influence on the mechanical coupling of the cable and the quality of the recorded signals (cf. Frignet and Hartog, 2014;Constantinou et al., 2016). After the bottom of the drivable depth in well GrSk3 had 100 been reached at 4259 mMD, recordings with increasing amounts of cable slack of 1 m, 5 m, 11 m, and 20 m have been performed. Based on the results, it was decided to keep the wireline cable under almost full tension for recording, as the best overall data quality was found to be achieved under these conditions (see section 4.2). 1 Here and in the following, the term "shot" is used to refer to a single vibroseis record. https://doi.org/10.5194/se-2020-169 Preprint. Discussion started: 12 October 2020 c Author(s) 2020. CC BY 4.0 License.
Within the following three days (days 1-3), acquisition was performed with a nominal number of 16 repeats for the 61 source positions distributed around the wells (see Figure 1). Nevertheless, due to a technical problem with 105 acquisition in well GrSk4 during day 1, mainly data for well GrSk3 could only be recorded during this time.
Therefore, in order to improve the reduced coverage around the GrSk4 well caused hereby, we relocated some of the original source positions from the northern to the southwestern part of the survey area.

3
Seismic data processing As one of the first processing steps, the DAS data recorded along the length of the sensor cables was correlated to 110 the measured depth along the boreholes. This depth correlation was performed using the gamma-ray logs recorded during running in hole with the sensor cables, as well as travel time data from check shots recorded at 1200 m, 2400 m, 3600 m, and 4207 m depth in the GrSk4 well. During further processing, the depths were transferred to vertical depths below the seismic reference datum, which is mean sea level (TVDSS, True Vertical Depth Sub Sea), using the geometries of the borehole trajectories. 115

Gauge length optimization
The choice of an optimized gauge length value is an essential part of the DAS data acquisition and processing.
This parameter has a significant effect on the signal-to-noise ratio of the data and on resolution in the frequency domain. Dean et al. (2017) presented an approach, which helps to maximize the signal-to-noise ratio, while keeping interfering influences on the frequency content below a desired threshold value. By selecting an optimum gauge 120 length GLopt, a favorable compromise between these two factors can be achieved, using with R the gauge length / spatial wavelength ratio, v acoustic velocity, and fp peak frequency.
The graphs presented in Figure 2 show the dependence of signal-to-noise ratio and resulting wavelength on R for the conditions of the current survey. According to this, optimum conditions within the desired limits are found for 125 R-values between 0.46 and 0.56. For an intermediate R value of 0.5, an optimum gauge length of 39 m is calculated using (1), for a velocity of 4800 m/s, which has been extracted from the interval velocities derived for the Rotliegend reservoir interval (see section 4.3), and a middle frequency of 61 Hz for the 10-112 Hz sweep. Therefore, the acquired DAS-VSP data was reprocessed accordingly, using the derived optimum gauge length value. 130

Pre-processing
An overview of the further seismic data processing steps is contained in Table 2. Seismic pre-processing included stacking and correlation with the pilot sweep. The hDVS output strain data was then transformed to strain rate by differentiation in time, resulting in a 90° phase shift. The strain-rate data is proportional to acceleration (Daley et al. 2016), and acceleration is in phase with the pilot sweep (Sallas 1984). 135 https://doi.org/10.5194/se-2020-169 Preprint. Discussion started: 12 October 2020 c Author(s) 2020. CC BY 4.0 License.

Common-source gathers and coherent noise suppression
Common-source gathers for zero offset, intermediate, and far offset source positions are displayed in Figure 3 and Several methods for elimination of this "ringing" noise like spectral balancing, deconvolution, and time-frequency domain filtering (Elboth et al., 2008) were tested. For zero offset data processing we selected to use Burg adaptive deconvolution (Griffiths et al., 1977). This method is a good compromise between computational effort, 150 robustness, and application simplicity. A more thorough description of this ringing noise and further methods of noise suppression can be found in Martuganova et al. (under review). The filtered data sets are displayed in Figure   3 and Figure 4, together with the unfiltered data sets for comparison.
Usage of self-updating linear prediction operators is the foundation of the Burg adaptive deconvolution method.
The designed filter operator is different at each trace sample. A set of filter coefficients is convolved with the data 155 in order to predict the future data values at some prediction distance. Coefficient values are re-computed for each data sample in the seismic record with the criterion of minimising the root-mean-square error. The computations are performed in forward and reverse direction in the time domain.
The application of Burg adaptive deconvolution resulted in a significant reduction of the coherent ringing noise The traces recorded at 1200 and 3600 m depth both contain direct P-wave arrivals, at 520 ms and 1090 ms, 170 respectively. The DAS trace for 1200 m depth is strongly influenced by the ringing noise described above, which is confined to a narrow frequency band between 40 and 50 Hz ( Figure 5B) at this location. Similar noise characteristics have been observed, e.g. in the study of Chen et al. (2019). This noise is however not evident in the geophone data recorded at the same depth, which suggests that the ringing noise in the DAS data is related to the different deployment methods of the acoustic receivers. While the DAS sensor cable is freely suspended inside the 175 borehole, the geophone tool is clamped to the borehole wall. Closer analysis of the ringing noise shows that the sensor cable acts like a vibrating string within the affected intervals, with resonances occurring at a fundamental frequency and higher overtones (Didraga et al., 2015;Martuganova et al., submitted).
The traces recorded at 3600 m depth ( Figure 5E) also contain strong reflected waves, which arrive at around 1190 ms and originate from the base of Zechstein reflectors, at around 3850 m depth (see Figure 4). Overall, the DAS 180 strain rate data exhibits a high similarity with the geophone measurements, except for the upgoing reflections.
Here, the DAS strain rate data displays the opposite polarity as the geophone data. This polarity reversal for reflected upgoing waves has also been observed in previous studies, e.g. by Hartog et al. (2014), Mateeva et al. (2014), or Willis et al. (2016). Frignet and Hartog (2014) note that such a polarity flip compared to geophone data is similar to the characteristics of hydrophone sensors. 185 As a test, we have also converted the DAS data to geophone-equivalent acceleration data, using the method described by Egorov et al. 2018. For this, we performed a transformation of the original DAS strain data into acceleration via filter application in the vertical wavenumber domain (k z ) and further double differentiation in the time domain. The results for the check-shot traces recorded at 2400 m and 3600 m depth are shown in Figure 6.
After conversion to acceleration, the DAS data displays the same polarity as the geophone data, also for the 190 upgoing reflections. This is in line with previous results obtained by Correa et al. (2017).

Signal quality
Common-source gathers recorded with different amounts of cable slack in well GrSk3 are displayed in Figure 7.
There is a zone with decreased amplitude of the first break signal at the bottom of the well, which increases in length with increasing amount of cable slack. While the random noise is similar, leading to an overall signal-to-195 noise ratio (SNR) drop within the affected zone, the coherent noise is changing. For the recordings with 1 m, 5 m and 11 m cable slack, a zone with ringing noise is visible at a depth of approx. 2890 mMD. This zone almost disappears in the 20 m cable slack data set, where the zone of decreased first break amplitudes is approximately approaching the same depth. So ringing noise seems to be reduced within the affected zone, likely because of improved mechanical coupling of the cable to the borehole wall. But at the same time, the signal amplitude is 200 significantly reduced within the affected zone as well.
Due to the higher first break amplitudes, the best signal quality overall was assigned to the data set recorded with 1 m cable slack, i.e. under almost full cable tension, and further recording was performed like this. Notably the best seismic record had been found to be recorded under the opposite conditions with released cable tension during the field trial reported by Frignet and Hartog (2014). Nevertheless, in their study, the optical wireline cable had For data quality evaluation, the SNR for each trace of the dataset was calculated. The energy of signal and noise was computed as the root mean square (RMS) amplitude within time windows of -10 to +30 ms around the first arrival and 150 ms at the beginning of the trace before the first arrival, respectively. The signal-to-noise ratio was 215 then calculated in dB using the formula: The calculated SNRs are displayed in Figure 8 and Figure 9. The data is sorted for the different acquisition days and with increasing source offset. Each vertical column represents a source location, and the calculated SNR for each trace is color-coded. Altogether, the data has a good SNR, with average values of approx. 40 dB to 50 dB at 220 a depth region around 1000 m for the smaller offset source locations, decreasing to approx. 4 dB to 10 dB at around 4200 m close to the final depth. There is an overall decrease of the SNR with increasing channel depth and source offset, which corresponds to the decay of signal amplitudes to be expected due to spherical divergence of the acoustic waves. The data for the first acquisition day have similar characteristics for both wells, with slightly larger SNRs for well GrSk4. 225 From the start of the second acquisition day, a sharp drop of the SNR is evident in the data recorded in GrSk3 at a depth of approx. 3400 m. In addition, there are further intervals with decreased SNR at depths of approx. 3100 m and 2600-2800 m. Curiously, the SNR for the channels below 3400 m gradually recovers again with increasing depth, until even improved SNRs in comparison to the first acquisition day are reached in the bottom interval.
The observed signal drop at 3400 m after day 1 seems to be similar to the effect of reduced signal amplitudes 230 observed during the slack test. Nevertheless, the configuration of the wireline cable remained unchanged between day 1 and day 2. Accidental introduction of additional cable slack during this time, e.g. by slipping of the wireline winch, or movement of the crane arm holding the cable sheave, can be excluded, as the position of the cables was carefully monitored by placing marks on them after running into the hole. Furthermore, no significant change of the wireline cable tension at surface has been registered between day 1 and day 2. Other causes must therefore be 235 responsible for the observed effect.
Combined with the remaining coherent noise after filtering, there is a significant heterogeneity in the data, which requires to carefully select the data to be considered during evaluation and interpretation.

Time-depth relationships and interval velocities
For every source point, the travel times of the direct downgoing waves were determined by picking of the first 240 break times (Table 2)

Corridor stacks 250
Further processing steps included separation of up-and downgoing wavefields, deconvolution, and transformation to two-way travel time (Table 2). After this, reflections are aligned horizontally and vertical reflection profiles were generated by stacking of the separated upgoing wavefield data over a defined time window after the first arrival (corridor stack). Because of the recording characteristics of the DAS data (see section 4.1), the polarity of the upgoing wavefield data has been reversed, in order to match the polarity convention of conventional geophone 255 data. The polarity convention of the data is European or EAGE normal, i.e. a negative amplitude value (trough) corresponds to an increase in acoustic impedance downwards (Simm and White 2002).
Corridor stacks for GrSk3 and GrSk4 are displayed in Figure 10. The recorded reflections are accurately correlated to depth and can therefore directly be assigned to lithology and other borehole data. The most prominent reflection events within the corridor stacks occur at the base (reflectors Z1, Z2, Z3) and top (reflectors X1, X2, X3) of the 260 Upper Permian (Zechstein), and within the Middle Triassic (Buntsandstein; reflectors S1, S2).
Larger differences between the corridor stacks are mostly related to intervals where the reflection data is disturbed by residual ringing noise. The slope of this residual noise in the common-source gathers is similar to the slope of reflected upgoing waves, leading to positive superpositions and enhancements in the corridor stack, which cannot be distinguished from real reflection events. 265 In the eastern part of the North German Basin, the deepest seismic reflections that can be readily recognized and correlated are at or close to the base of the Zechstein. The reflecting interface Z1 is at the boundary between the Stassfurt Salt and the underlying Stassfurt Anhydrite ("Basalanhydrit"). This "base Zechstein" reflector is used as a marker horizon over the entire Southern Permian Basin area .
At Groß Schönebeck, the base of the Zechstein is comprised of an 80 -90 m thick sequence of anhydrite, salt, and 270 carbonate layers, which is underlain by the sediments of the Rotliegend. This interlayered sequence of strata with high impedance contrasts gives rise to several strong and closely spaced reflection bands, which mark the base of Zechstein in the corridor stacks (see Figure 10 and Figure 11).
Reflections within the underlying Rotliegend interval are evident as well, which can now be assigned to individual sections of the reservoir. The corridor stacks for the Rotliegend reservoir interval are shown in Figure 11, together 275 with well logs and lithology data for both wells. Some of the well logs are unfortunately not available for the lower parts of the wells, especially for GrSk4. Acoustic impedance has been calculated as the product of bulk density and sonic velocity.  (Reinhardt 1993).

Summary and conclusions 315
Based on this survey, several important new experiences for DAS-VSP acquisition on wireline cable have been gathered. The presented results can be used in support of planning, execution, and evaluation of future surveys of this type.
Common-source gathers of the recorded data are dominated by arrivals of downgoing P waves, upgoing reflections, and tube waves. One characteristic of the recorded DAS-VSP data is that it is affected by a coherent noise, which is correlated among neighboring traces. This ringing noise is evident in common-source gathers as a conspicuous zigzag pattern confined to distinct depth intervals, and is occurring in narrow frequency bands. It is influenced by the cable tension and how the cable is aligned with the inner surface of the borehole, depending on changes of the borehole trajectory.
Several tests to determine the influence of the wireline cable tension on the mechanical coupling of the cable to 325 the borehole wall have been performed. The highest signal amplitudes and best overall data quality were found to be achieved under almost full cable tension, and the main part of the data was acquired under these conditions.
The results of these tests nevertheless also indicate that a reduction of coherent ringing noise can be achieved by adding cable slack. The interrelation between cable tension and configuration inside the borehole, mechanical coupling to the borehole wall, and recorded signal amplitudes needs further investigation. 330 After conversion to strain rate, the waveforms and frequency content of the DAS data display a high similarity to vertical component data of a conventional borehole geophone. However, upgoing reflections are recorded with opposite polarity, which confirms the results of earlier studies. The polarity of the reflection data was reversed during later processing, in order to match the polarity of conventional geophone data.
Most of the data has a very good signal-to-noise ratio. Nevertheless, in the GrSk3 well, a sudden reduction of SNR 335 along the deeper part of the profile after the first recording day has been observed. As a larger movement of the cable can be excluded during this time, the cause of this change of acquisition characteristics remains elusive. The ringing noise can be suppressed to a large extent by suitable filtering methods.
From the zero-offset data, accurate time-depth relationships and velocity profiles were derived. The reflectivity along the boreholes could be mapped with high resolution. The strongest reflections occur at the base and the top 340 of the Zechstein salt complex, and within the Buntsandstein. Nevertheless, in parts the interpretation of the corridor stacks is hampered by residual ringing noise, which is occurring within a short time window after the first break arrivals, and is difficult to be distinguished from true reflection events.
For the Rotliegend reservoir section, the sequence of reflection events in the corridor stacks shows a different character for both wells overall, which is explained by lateral changes in lithology. But it also displays local 345 similarities: The top of the Dethlingen sandstone interval is marked by a positive reflection event in both wells.
This information can be used to identify a related reflector and track the distribution of this reservoir layer in a 3D seismic volume. Processing and interpretation of both 3D VSP and 3D surface seismic data is currently ongoing.
The top of the volcanic rocks has a somewhat different response in both wells and no stronger event is obvious at the postulated top of the Carboniferous. The thickness of the volcanic rocks can therefore not be inferred from 350 individual reflection events in the seismic data alone.
The DAS method has enabled measurements at elevated temperatures up to 150 °C and has led to significant time and cost savings compared to deployment of a conventional geophone chain.
Data availability. The seismic survey data will be made available as data publications through GFZ Data Services 355 (https://dataservices.gfz-potsdam.de/portal/).
Author contributions. JH and CMK conceptualized the project. JH and MS planned and supervised fieldwork and data acquisition. EM, MS, CMK, and JH performed the seismic data processing and analysis. JH interpreted the data under discussion with all co-authors, and input from BN on geological well data. JH and EM prepared the 360 manuscript with contributions from all co-authors. https://doi.org/10.5194/se-2020-169 Preprint. Discussion started: 12 October 2020 c Author(s) 2020. CC BY 4.0 License.  https://doi.org/10.5194/se-2020-169 Preprint. Discussion started: 12 October 2020 c Author(s) 2020. CC BY 4.0 License.

480
First column of panels shows data after pre-processing for source positions 10 (a), 25 (b), 66 (g), and 17 (j). Second column (panels b, e, h, and k) shows the data for the same source positions after ringing-noise suppression (Burg adaptive deconvolution) and moderate coherency enhancement. For display, we applied a windowed trace equalisation. The third column (panels c, f, i, and l) shows the signal-to-noise ratio of the data after pre-processing. Colored arrows (exemplary): direct downgoing P wave (light blue), upgoing reflected P-P waves (green), tube wave (magenta), residual 485 noise after application of ringing-noise filter (dark blue). https://doi.org/10.5194/se-2020-169 Preprint. Discussion started: 12 October 2020 c Author(s) 2020. CC BY 4.0 License.

490
(panels b, e, and h) shows the data for the same source positions after ringing-noise suppression (Burg adaptive deconvolution) and moderate coherency enhancement. For display, we applied a windowed trace equalisation. The third column (panels c, f, and i) shows the signal-to-noise ratio of the data after pre-processing. Colored arrows (exemplary): direct downgoing P wave (light blue), upgoing reflected P-P waves (green), tube wave (magenta), residual noise after application of ringing-noise filter (dark blue).