1、Geological Background

The Wattenberg Field, located within the greater DJ Basin, is situated just to the east of the Rocky Mountain Front Range of Colorado and covers an area of approximately 2,000 square miles between Denver and Greeley (Figure 1). The field experienced its first large-scale development in the 1970s with the discovery of the J Sand, followed by discoveries in the overlying Terry, Hygiene, Niobrara, and Codell formations (Sonnenberg and Weimer, 2005). Driven by advancements in drilling and completions technology, the Wattenberg Field has been under continuous development for nearly 50 years and is ranked one of the top five largest US onshore oil fields (U.S. Energy Information Administration, 2015). The location of the field is influenced by the presence of an abnormally high present-day geothermal gradient.

Figure1-A) Diagrammatic cross-section of the DJ Basin (Sonnenberg and Weimer, 2005) highlighting the asymmetrical geometry and basin-center accumulation of hydrocarbons. B) Map view of the areal extent of Wattenberg Field (red) within the greater DJ Basin (dark blue); 1,000 ft contours show depth to basement within the basin. The stratigraphic column (right) shows historical hydrocarbon production zones and potential source rocks (Sonnenberg and Weimer, 2005).

This thermal anomaly, presumed to be associated with the NE-trending Colorado Mineral Belt (Meyer and McGee, 1985; Higley et al., 1992), has been present since the start of Early Tertiary hydrocarbon generation (Ladd, 2001). Recent development of the Wattenberg Field over the past decades has focused on the Upper Cretaceous interval, specifically the Niobrara chalk benches and the Codell Sandstone,where as many as 20 horizontal wells can be drilled off a single pad to target each of the multiple stacked pay intervals.

2、Regional Tectonics

The DJ Basin owes its existence to the creation of the greater Western Interior Basin (Figure 2), which arose as a consequence of the convergence between the North American and Farallon plates in the late Mesozoic (Pang and Dag, 1995). During much of the Mesozoic, the Western Interior Seaway expanded and contracted, at times covering nearly half of North America and extending from the Artic to the Gulf of Mexico (Blakey, 2014). During the Campanian age of the Late Cretaceous, the Western Interior Seaway began to experience a transition from Sevier to Laramide deformation, which corresponded to a shift from flexural to dynamic subsidence and the onset of basin partitioning by Laramide uplifts, leading to the eventual withdrawal of the Western Interior Seaway (Leary et al., 2015). The DJ Basin, one of many intermountain basins created by the Laramide Orogeny, is bound to the west by the Rocky Mountains, to the northwest by the Hartville Uplift, to the northeast by the Chadron Arch and to the south by the Las Animas Arch and Apishapa Uplift.

Figure 2-Paleogeographic and tectonic reconstruction of the Middle Cretaceous showing the subduction of the Farallon Plate during the Sevier orogeny. The approximate location of Wattenberg Field is shown by the black arrow. Image courtesy of Halliburton-Landmark.

3、Stratigraphy

Figure 3 summarizes the stratigraphy of Wattenberg Field. The Cretaceous Carlile and Niobrara formations were deposited within the Western Interior Seaway and are separated by an unconformity that defines the top of the Codell Sandstone of the Carlile Formation and the base of the Fort Hays Limestone Member of the Niobrara Formation.

The Carlile Formation is comprised of four major members, however, only the Fairport and Codell Sandstone members are present in the Wattenberg area. The Codell Sandstone is a gray, highly bioturbated, very fine-grained sandstone that pinches out because of non-deposition to the south and east of the field but within Wattenberg is generally 15- to 20-ft thick (Ladd, 2001).

The Niobrara Formation is broken into two members, the lowermost Fort Hays Limestone and the upper Smoky Hill Member. The Smoky Hill Member is commonly subdivided into 10 alternating chalk and marl intervals (Longman et al., 1998), with the chalk-rich intervals referred to as benches. The three primary horizontal target benches within the Niobrara are generally referred to in descending order as the A, B, and C chalk benches. The A chalk is absent across the middle of the field along the Wattenberg High where the Sharon Springs Member of the Pierre Shale sits unconformably atop the Niobrara. The Sharon Springs Member of the Pierre Formation is a black, organic-rich, marine shale, and represents a well-known drilling hazard because of its reactivity with water and oil-based drilling mud systems.

Figure 3-Type log of horizontal target zones A, B, and C within the Niobrara Formation and the Codell Sandstone of the Upper Carlile Formation (Sonnenberg and Weimer, 2005).

4、Faulting

One of the major challenges to horizontal drilling and development in Wattenberg Field are the abundant small-scale normal faults that are pervasive through the Niobrara and Codell interval of the stratigraphic section. First identified in 1925 (Twenhofel, 1925), these small-scale, detached normal faults were not fully understood until they were studied using 3D seismic data (Sonnenberg and Weimer, 2005), which revealed the complex and layer-bound nature of faulting when viewed along time slices. Layer-bound normal faults, also commonly known as polygonal faults, occur in very-fine-grained sediments ranging from smectitic claystones to carbonate chalk and marl of high porosity and extremely low permeability (Cartwright and Dewhurst, 1998). These layer-bound normal faults are understood to have formed at relatively shallow burial depths of <1000 m by the process of volumetric reduction through bed-parallel and vertical compaction (Cartwright and Lonergan, 1996; Ireland et al., 2011; Goulty and Swarbrick, 2005; Ghalayini et al., 2017). Characterized in more than 20 basins globally (Cartwright and Dewhurst, 1998), layer-bound normal faults have throws generally in the range of 10 to 100 m with a spacing of 100 to 1000 m. The faults are planar or gently listric with dips ranging from 30 to 70° and a mean fault dip of ~45°. Layer-bound normal faults are interpreted to form in passive, non-tectonic settings in near-isotropic horizontal stress states where principal horizontal stresses are nearly indistinguishable. These stress conditions lead to the characteristic polygonal orientation of the fault planes in map view.

Figure 4-East/West seismic cross section (see line A-A’ on Figure 5 and Figure 6) from near-surface to basement. The Parkman, Niobrara, and Codell surfaces are respectively shown by the magenta, light blue and yellow lines, and demonstrate the layer-bound nature of the faulting. Tier 1 faults (highlighted in yellow) are observed to offset the Niobrara and Codell surfaces. Tier 2 faults (highlighted in black) are observed to offset the Parkman and other shallow horizons within the Pierre Formation. Basement faults are highlighted in green.

Figure 5-Depth structure map (10-ft contour interval) of the Parkman surface. Major fault planes are shown as bold black lines. Seismic section A–A’ from Figure 4 is indicated by the yellow line running from west to east across the study area. Township and Range grid system is displayed for scale, with each grid square representing ~1 square mile.

Figure 6-Depth structure map (10-ft contour interval) of the Niobrara surface. Major fault planes are shown as bold black lines. Seismic section A-A’shown in Figure 4 is indicated by the yellow line running from west to east across the study area. Township and Range grid system is displayed for scale, with each grid square representing ~1 square mile.

In the Wattenberg Field and the surrounding DJ Basin, layer-bound polygonal faulting is observed primarily in the Upper Cretaceous interval and is best represented in the Niobrara carbonate mudstones and silty shale of the lower Pierre Formation. In the Wattenberg Field, there are two prominent tiers of detached (i.e., layer-bound) normal faults (Figure 4).

The lower tier (Tier 1) is centered over the Niobrara interval of the section with the largest faults reaching up into the Lower Pierre shale and extending down into the lower Greenhorn. The total thickness of the Tier 1 faulted interval is approximately 2,000 ft. The upper tier (Tier 2) is centered over the Parkman Formation and covers an interval thickness of approximately 4,000 ft. The two tiers of layer-bound faults do not appear to be genetically related. In cross-section, it often appears that some fault planes connect up and cut across the boundary between the two tiers; but, on further inspection in map view, it is observed that the two apparently related fault planes are actually striking in drastically different orientations.

Tier 1 faults that offset the Niobrara surface strike roughly north–northeast (Figure 6), with a secondary fault set striking roughly east–west. Tier 2 faults that offset the Parkman surface are noticeably less organized (Figure 5), showing a wide range of orientations with the dominant set striking roughly north–northwest and two secondary sets striking northeast–southwest and east–west, accordingly. This faulting pattern effectively refutes the hypothesis that a single fault initiated in one of the layers and then progressively propagated through the section. Rather, it is more likely that the layer-bound normal faults initiated independently of each another within their discrete sequences and then propagated towards each other and, in some cases, coincidentally linked up across the separating boundary. Because both of the primary horizontal drilling targets are contained in the lower tier of faults, the focus for the remainder of this paper is these Tier 1 faults.

The population of Tier 1 faults that offset the Niobrara and Codell members of the Upper Cretaceous are well developed in both map and cross section view. The throw of these faults in this study area ranges between 25 and 120 ft with the maximum throw observed near the top of the Niobrara. Graben complexes are commonly observed to form en echelon sequences, where the fault planes overlap and link up with one another in a consistent left- or right-stepping sense in map view.

5、Geosteering and Fault Interpretation

Geosteering is the term used to describe the process of actively targeting a specific stratigraphic interval, known as the “target zone,” through the use of real-time electronic log data correlation to assist drillers in precisely locating the drill-bit. In the Wattenberg Field, for example, horizontal wells are geosteered to land in the Codell or Niobrara chalk benches by correlating the real-time MWD gamma ray logs to a nearby offset vertical well log (Figure 7). By matching the gamma log character between the vertical and lateral wells whilst accounting for the differing inclination of the wellbores, it is possible to identify where the wellbore being actively drilled is located within the stratigraphic section. Using this information, one can then attempt to model and predict the future location of the wellbore within the target interval, and make target adjustments to stay in zone.

The process of horizontal well correlation can also be used to identify the location and offset of faults that are encountered along the wellbore. Fault picks are made using the same principles as are applied to identify faults in vertical well logs, where missing or repeated section is identified at the location of the fault through log correlation. In the study area, horizontal image logs were also incorporated into the final geosteering interpretation which, when compared to using gamma ray logs alone, enabled fault and fracture zone locations encountered during drilling to be identified with an increased level of resolution and confidence (Figure 7C).

For example, the Wishbone 29N-E24HZ well was initially interpreted using gamma ray data alone, and two large faults were easily identified using this method (Figure 7A). When image log data was incorporated into the interpretation, several more small faults with throw < 5 ft were identified (Figure 7B). By integrating the horizontal gamma correlations and image log interpretations, it is possible to accurately identify the precise location and offset of all the faults encountered by the horizontal well. This collation of high resolution fault data is ultimately what allows us to ground-truth the fault likelihood attribute to assess its ability to predict small-scale faults and value for horizontal well planning.

Figure 7-A) Geosteering interpretation of the Wishbone 29N-E24HZ horizontal well showing location and estimated offset of faults. B) Horizontal image log interpretation of the Wishbone 29N-E24HZ horizontal well showing location and orientation of interpreted faults and fractures. C) Horizontal gamma ray from Wishbone 29N-E24HZ correlated against a vertical offset well. Shaded colors correspond to intervals shown in the geosteering interpretation shown in (A).

6、Fault Likelihood Attribute: Theory and Computation

The mapping of geological faults in 3D seismic data is a key component of subsurface interpretation workflows. Manual fault interpretation is a laborious and subjective process, and so methods for assisting or automating fault interpretation are of great potential value to the seismic interpreter. Multiple discontinuity and coherence-based seismic attributes have been proposed to help detect faults in 3D seismic data (e.g., Bahorich and Farmer, 1995; Marfurt et al., 1998). Changes in reflector orientation as measured by curvature attributes have also been proposed for the same purpose (e.g., Chopra and Marfurt, 2007). While the first generation of fault attributes were extremely successful and widely adopted, as Figure 8 shows, their direct applicability to automated fault interpretation is limited because of sensitivity to localized noise and stratigraphic discontinuities, such as channels.

Hale (2013) provided a breakthrough improvement in the quality of fault attributes derived from 3D seismic data, using fault-oriented semblance scans to detect faulting. Hale’s fault likelihood algorithm computes structure tensors (e.g., Weickert, 1999) at every image sample location in a 3D seismic volume, from which local structural slopes are extracted. These slopes are firstly used to smooth the input seismic volume along structure, and secondly to align adjacent traces in 3D after smoothing to compute a local (i.e., 9-point) structure-oriented semblance. Smoothing is necessary to condition the semblance calculation, which can become unstable if the semblance denominator approaches zero, and to reduce sensitivity of the fault likelihood attribute to discontinuities introduced by spatially and temporally localized noise. The extent of this smoothing is controlled by a parameter σ_structure, which defines the half-width in sample space of the smoothing operator.

Having computed structure-oriented semblance, the fault likelihood algorithm then performs a fault-oriented semblance scan, searching for the fault strike and dip that minimizes semblance and thus maximizes fault likelihood, which is defined as 1 − S⁸ where S is the fault-oriented semblance. This scan requires smoothing of the structure-oriented semblance along all potential fault plane orientations such that coherent features offset by an actual fault are in some sense realigned by the smoothing. The extent of smoothing along fault planes is controlled by a parameterσ_faults which defines the half-width of the fault-oriented smoothing operator in sample space.

The final step of the fault likelihood attribute calculation is to mask the fault likelihood attribute volume to preserve only the ridges corresponding to the most likely location of faults, a process known as“thinning.” Throughout this paper, the term fault likelihood is used to refer to the thinned fault likelihood attribute. An example map and section view through a fault likelihood volume computed in the Wattenberg Field is shown in Figure 8, and clearly shows the superiority of fault likelihood over traditional fault attributes for clear imaging of faults.

Figure 8-Map (top row) and section-view (bottom row) comparisons of discontinuity (left column), curvature (middle column) and fault likelihood (right column) attributes.

An immediate benefit of the fault likelihood algorithm is that it provides information not only about the location of faults but also about fault dip and strike, which is of great value if an automated extraction of discrete fault objects is to be performed. A less immediately obvious benefit of fault likelihood is that, because it uses semblance, it allows detection of very subtle fault signatures. For example, faults with small fault throws that are essentially invisible to the naked eye are detected by the algorithm due to the change in reflector coherence (as seen by structure-oriented semblance) across the fault plane.

Given these benefits, one might wonder why the fault likelihood attribute is not being applied more widely or even routinely in industry today. The main challenge to user adoption of fault likelihood is simply the fact that the fault-oriented semblance scan is computationally expensive and hence time intensive, which fundamentally limits the ability of interpreters to generate this attribute in a timely manner for 3D seismic datasets of any reasonable size on their workstations.

To address this issue, a cloud-native and fully parallelized implementation of the fault likelihood attribute has been developed. The computational resources available to an end-user to perform the attribute calculation automatically scale up appropriately with the size of the input dataset, allowing performant execution of the fault likelihood attribute calculation, regardless of the input volume size (Figure 9). The speed of the cloud implementation is important for the present study, as there is a need to perform several fault likelihood calculations with different parameterizations to test the ability of fault likelihood to detect faults with small throws.

Figure 9 – Comparison of fault likelihood attribute computation time for 3D seismic volumes of various sizes for a typical interpretation workstation versus on the cloud.

Figure 10-Typical Niobrara amplitude spectrum and synthetic used to calculate seismic tuning thickness.

7、Detecting Subseismic Faults in Wattenberg Field with the Fault Likelihood Attribute

The Anatoli 3D seismic survey was acquired in August 2013 and is located in Weld County Colorado. The N-S receiver lines had a spacing of 660 ft with geophone groups located at intervals of 110 ft. The E-W source lines had a spacing of 820 ft with shot points located at intervals of 110’ ft. The average fold of 180 resulted in very good imaging of the Niobrara interval.

Frequency and velocity information extracted from well and seismic data over the Niobrara interval allow us to estimate that the seismic tuning thickness is approximately 60 ft at the Niobrara. The analysis presented in Figure 10 shows that the average velocity through the Niobrara interval is 15,500 ft/s with the amplitude spectrum dominated by energy in the 18 to 65 Hz range. The high quality of the Anatoli survey means the upper bound of the dominant frequency range can be used to calculate seismic tuning, yielding a tuning thickness of 60 ft.

Sensitivity of the fault likelihood attribute to faults with small throws is controlled by two key parameters σ_structure and σ_faults. The structure-oriented smoothing operator half-widthσ_structure should be chosen such that it is large enough that reliable structural dips can be estimated locally given the amount of noise in the input dataset.

The fault-oriented semblance scan smoothing operator half-width σ_faults should be chosen such that it is at least as large as the maximum fault throw of interest to the interpreter. This parameter is sometimes referred to as the fault stiffness, as larger operator sizes imply more smoothing along the scanned fault-plane orientations, which results in smoother and “stiffer” features in the output fault likelihood volume. In contrast, a smaller value of σ_faults results in fault likelihood volumes that can reveal complex small-scale faulting patterns.