The Impact of Dipping Velocity Models on Microseismic Event Locations

Microseismic monitoring of hydraulic fracturing treatments is a powerful stimulation diagnostic technique when properly implemented. This paper describes one of the many parameters having a significant impact on the location of each microseismic event, the subsurface velocity model, which ultimately shapes the microseismic analysis. Despite its relevance, the importance of using accurate velocity models is sometimes underestimated. Several authors have already emphasized the significance of choosing an appropriate subsurface velocity model, and its impact on microseismic event location and subsequent interpretation. For instance, Warpinski (2009) points out the velocity model as "the most important element in the whole process of microseismic mapping", and describes how differences between vertical velocities (measured by logs) and horizontal velocities (used for ray path modeling), could originate event misplacements in the order of hundreds of feet. Wilson et al. (2008) also presented a case study on the variability originated by modified velocity structures in the presence of high velocity layers, thus underscoring the importance of the velocity model on any microseismic interpretation. Maxwell (2009) and Maxwell et al. (2010), on the other hand, recognized that the velocity model, often built following calibration procedures such as the one described by Warpinski et al. (2005), can result in oversimplified versions of the subsurface structure, which may carry significant uncertainty into the final event mapping. None of these previous studies, however, have addressed the effect of dipping velocity models in the event location and subsequence interpretation. This work highlights how geometric differences of velocity models, associated to known geological characteristics of the subsurface, may significantly affect the location of microseismic events and the overall stimulation diagnostic.

The main objective of the microseismic monitoring experiment described on this publication was to diagnose the performance of a multistage hydraulic fracturing treatment designed to stimulate a hydrocarbon bearing, low permeability reservoir. Surface reflection seismic and open hole logs across the area of interest suggest an image of the subsurface structurally and stratigraphycally simple. No major faults are believed to cut through the zone of interest. Moreover, the target zone consists of a predictable succession of thinly laminated beds dipping at a relatively constant angle of 4.5 degrees.

The treatment and monitor wells were both drilled parallel to one another with an 85.5 degree inclination, 1000 ft apart, as shown in Figure 1. Microseismic events were registered in the monitor well via a receiver array consisting of eight multi-component seismic sensors, each separated 100 ft. This array of seismic sensors was sequentially positioned along three different locations within the observation well during the job execution, in order to keep a reasonable separation distance between the monitoring tool and the stimulated zones. For the first nine fracturing stages the sensor array was positioned along two different sections of the horizontal portion of the monitor well, while the last four stages were monitored with the sensor array positioned in the vertical section.

Figure 1 illustrates schematically one of the positions of the sensor array, represented by green triangles, as well as the perforation clusters planned, each one color-coded relative to its associated fracturing stage. Different treatment designs and perforation clusters were also evaluated during this multistage hydraulic fracturing job, but their details are out of the scope of this paper.

Time signals (i.e. seismograms) registered in the monitor well are associated to a specific microseismic event in the subsurface via the use of an appropriate velocity model, which must be properly scaled in vertical and horizontal directions. The detection range of the sensor array determines the maximum length at which a signal can be registered (typically 2000-5000 ft). Horizontal scaling of the velocity model is then achieved by means of an anisotropic analysis. There are a number of ways in which an anisotropy analysis can be performed. In this study we chose to manipulate Thomsen (1986) parameters to account for known differences between horizontal and vertical velocities. The vertical component of the velocity model, on the other hand, is impacted by the subsurface dip angle and therefore, it must be considered. Although in this case the dip angle of the formation is fairly constant, showing relatively small values within the target zone, it is taken into account in the velocity model. Thus, in this study, an anisotropy calibrated velocity model consisting of a constant dip layered velocity sequence, based on a 1D sonic log, is used. This may be called a 1.5D anisotropic model. A fully 3D orthorhombic anisotropic approach for model building would be preferred. However, the methodology presented within this publication, is considered a good approximation (based on the relatively simple subsurface geometry) and serves for underscoring how minor geometric differences on the velocity model have a significant impact on the proper location of events and subsequent microseismic analysis, which is one of the main objectives of the paper.

The first step in the velocity model creation is smoothing and upscaling P-wave and S-wave sonic logs acquired in a vertical well (pilot hole). These sonic logs provide a good representation of the subsurface surrounding the zone of interest.

The pilot hole was drilled to support the design of directional plans and for geological control. The bottom section of this borehole was later plugged and abandoned. The treatment well was drilled from the same surface location of the pilot hole as illustrated in Figure 1.

Since 50 ft is a typical sonic log wavelength, a smoothing of 50-100 ft is commonly used to work with a simplified data set. Original sonic logs (black), and a series of color-coded upscaled versions, generated by averaging across different length intervals, are shown in Figure 2. The selected 1D model (green), resulting from the use of an 80 ft smoothing window, is also shown in this figure.

The horizontally layered model generated by using the selected upscaled sonic logs is shown in Figure 3. True well paths are overlaid in Figure 3a. For this particular study, 5000 ft laterals are drilled parallel to bedding along the zone of interest and both, the zone of interest and the treatment lateral well, dip 4.5 degree. This implies a velocity model depth error of at least 390 ft from landing point to bottom hole location (almost twice the thickness of the zone of interest) if the model shown in Figure 3a is used. A tilted velocity model with a 4.5 degree dip is required to keep the entire borehole in the correct layer as shown in Figure 3b.

Anisotropic calibration of the velocity model was accomplished by recording perforation shots from the treatment well. Modeled times were obtained by ray-tracing with a title anisotropy vertical transverse isotropy ray tracer engine in one dimension. Mismatch between the observed arrival onsets and model times (residual times) is minimized by interactive adjustment of anisotropy parameters defined by Thomsen (1986). The P-wave moveout observed from the receiver arrays determines Thomsen-epsilon, while the SH-wave moveout and S-P time determines Thomsen-gamma. When SV-wave information is present, it constrains Thomsen-delta, otherwise it is left as zero. This is a good approximation in most hydrocarbon bearing shales. Figure 4a shows the mismatch of isotropic model times with the perforation shot waveforms. After adjustment for Thomsen epsilon and gamma, the arrivals are well matched by the selected anisotropic model, as observed in Figure 4a. Anisotropic travel times for P-, SH-and SV-waves are calculated for all the grid points of a subsurface space to all receivers and stored in a lookup-table (LUT) utilized to automatically locate the microseismic events in the subsurface.

The total data set was scanned with an event detector algorithm which first detects signals above a given signal-to-noise threshold on each shuttle. For each detected signal, the known velocity model can be used to calculate a range of expected signal arrival times for the other shuttles. Signals arriving in these expected time ranges are said to be associated, meaning they come from the same microseismic event. If signals are detected on a given minimum number of shuttles, an event is declared and stored in an event file for satellite transmission to a remote processing center.

The traditional Geiger method requires many iterations of ray-tracing for each event to be located in real time. This means a simplified velocity model with only two or three layers must be used to reduce the calculation time. In contrast, we determined event locations following the Coalescent Microseismic Mapping technique described by Woerpel et al. (2010). Implementing a look-up-table becomes very practical in that all ray-tracing is done before the treatment starts, allowing for an anisotropic velocity model with more than 100 layers, without reduction in processing speed as the job is being executed.

For each gridpoint in the LUT, that is for each possible hypocenter, the time dependent signal to noise ratio (SNR) of the P-wave and SH-wave traces are time shifted and stacked. The time shifts come from the stored travel times in the LUT. The interpolated gridpoint with the best stack is the most likely hypocenter. Only events exceeding an SNR threshold after stacking are retained.

Travel times are calculated by a 1D anisotropic ray-tracer, which assumes horizontal layers. Therefore, a mathematically equivalent method of rotating the coordinate system by 4.5 degree is used, following the wells azimuth direction. All wellhead locations and deviation logs are first rotated to a flattened coordinate system (Figure 3b) before loading them into the ray-tracer. At the end of the location process, located hypocenters, the positions from where the microseism is believed to be originated, are rotated back to the true coordinate system.

As each event is auto-located, its hypocenter is displayed in a 3D visualization software. From here, it is easy to retrieve waveforms associated to a given hypocenter with modeled times plotted for quality control as shown in Figure 4. Outlier events are then reviewed following this step. If modeled times do not fit the arrival onsets, the arrival times are then re-picked by hand, and the event may be relocated by a conventional Geiger least squares inversion method of arrival times. The model times associated to outlier events are then revised, and if the model times fit valid arrivals, the hypocenter location is retained. This process increases the level of confidence associate to each event location, resulting in higher quality processing of the data and all subsequent analyses.

The results from the microseismic monitoring job are presented in Figures 5 and6. Figure 5, shows the hypocenters plotted relative to a cross-section view of the wells path. A horizontally layered velocity model with zero dip was used to generate this data set in real time, as the fracturing treatment was being executed. Figure 6, on the other hand, shows the results obtained a few weeks after the treatment was completed and the data reprocessed using an anisotropic velocity model with a 4.5 dip as previously explained in this work.

Direct comparison of Figure 5 and6 shows that the location of microseismic events is significantly affected by the velocity model used, which then impacts the interpretation of the estimated stimulated volume (ESV). This is particularly evident in the spread of hypocenters along the vertical direction, with the majority of the events lying across the target horizon when a 4.5 dip is taken into account in the velocity model (Figure 6). In contrast, a much larger spread in the vertical location of microseismic events is observed in Figure 5, which can potentially lead to misleading conclusions regarding fracture containment.

In general, layered subsurface velocity models cause the microseismic events to preferentially align along the interface between velocity layers with significant velocity contrast. Thisis observed in Figure 7 (detailed view of Figure 5), where a sequence of horizontally aligneddots (hypocenters) line up along the layer boundaries due to this processing artifact. The presence of these well-aligned dots could potentially be misinterpreted as a consequence of the hydraulic treatment, instead of a layer within the velocity model.

It is important to note that the artifacts originated by the implementation of an incorrect velocity model were observed throughout the job execution, independently of the placement of the sensor array within the wellbore. Nonetheless, as expected, the impact of the artifact was less noticeable when the sensor array is placed in the vertical section of the monitor well, due to its location relative to the source of hypocenters.

The workflow followed to generate the results shown in this paper required additional time to re-process the data gathered during the fracturing treatment. Nowadays, however, tilted velocity models can be implemented real time, thus allowing for more accurate placement of microseismic events and better decisions regarding fracturing diagnostics and overall field development strategies.

This work showed how implementing a tilted velocity model, based on a simple coordinate rotation during processing steps, made significant changes to previously estimated microseismic event locations. When targeted hydrocarbon bearing beds have even a relatively small dipping angle, this observed subsurface characteristic should be accounted for during the event location process, particularly in the vicinity of significant velocity contrasts.

Given these results, it is our recommendation that the microseismic interpreter fully understands the assumptions made to locate microseismic events so that subsequent treatment analyses are supported by robust data.

The authors would like to thank the management of Hilcorp Energy Company and Schlumberger for giving permission to publish this paper. We would also like to recognize valuable contributions provided by Mike Craven during the acquisition and processing of the data presented in this publication.