What do the best horizontal wells in the Midland Basin Spraberry and Wolfcamp tight oil plays have in common? What differentiates them from less productive wells? We used a geomodel containing 10,064 tight oil horizontal wells to answer these questions, focusing on the top 1 percent to see what differentiated them.
We found Wolfcamp A wells were more likely to be in the top 1 percent, especially those drilled in carbonate-poor rock. Average hydrocarbon-filled pore volume and oil formation factors still led to high-performing wells. High pressure was an important consideration for Wolfcamp B top producers. All of the best oil producers had low producing gas oil ratios (GORs). The top 1 percent of wells were usually the first well in the pad to be completed, and they came from above-average pads.
Operators can apply these results to further development of the Midland Basin. Specifically, they can prioritize development areas in the Wolfcamp A with low carbonate abundance while considering targets outside of the beaten path.
Much of the recent production and optimization literature utilizes machine learning. However, modern machine learning training minimizes the root-mean-squared residual or mean absolute residual. Machine learning practitioners use training-testing splits and regularization to increase model robustness. These two factors can lead to models that accurately predict the productivity of wells near the mean but fail for high-performing wells.
In contrast, we highlight the high-performing wells in the data exploration stage, plotting them on distribution and cross plots against the rest of wells to inform our statistical analyses. Only after generating and interpreting these plots do we draw on statistics to support conclusions.
High performing wells achieved a top-1-percent status (the best 101 wells) by leading in the absolute nonnormalized first 12 months of oil production (“raw” wells) or by leading in the first year of production divided by the lateral length. This led to a total of 145 unique wells because 57 wells were in the top 1 percent in both categories.
Methods: Geomodeling
Per well data was stored in, interpreted within and extracted from a giga-scale geomodel. We constructed a three-dimensional, faulted Midland Basin geomodel, containing more than 1 billion cells, including stratigraphic, petrophysical, core description and production data for the Spraberry and Wolfcamp intervals (figure 1). The model is based on more than 1,500 correlated wells, 700 wells with petrophysical and facies interpretations and approximately 10,700 horizontal production wells with more than 9,300 decline curve and completion data analyses.
As the initial step in building the 3-D model, we generated surface models of the stratigraphic horizons for the 1,500 wells with stratigraphic tops we interpreted. We mapped eleven stratigraphic horizons, from the top of the Spraberry, through the Wolfcamp, and down to the Strawn, across the Midland Basin. We constrained the model using conformable mapping techniques, incorporated faults interpreted by Ewing and Ruppel, and corrected these using more recent horizontal well trajectory data and stratigraphic interpretations. We combined all the interpretations into a sealed faulted framework, which forms the basis for the 3-D geocellular model. Cells had an average thickness of 2.5 feet.
Our petrophysical analysis consisted of a stepwise solution (RHOMMA-UMMA) using triple combo well logs including the photoelectric factor. Results were calibrated with core measurements. The resulting petrophysical curves include total organic content, kerogen density, clay volume, non-clay lithology, total porosity, total water saturation and lithofacies.
Per-well fluid properties included the first-year gas:oil ratio, API gravity and initial reservoir pressure. We extracted facies, porosity, water saturation, hydrocarbon-filled pore volume, total organic carbon and clay and limestone fractions averages along the lateral length of each well using the petrelpy library. The completion properties included fluid and proppant injected (figure 2).
Results
Well properties that impact production fall into three categories: fracability, producibility and completions. We examined each in turn, then how they interplay; after which, we highlighted some outliers. High performing producer (HPP) wells are concentrated in Midland, Martin and Howard counties (79 percent of the wells) (figure 3).
The key properties for assessing fracability include lithofacies and volume fractions of clay, quartz and calcium carbonate. A significant number of HPPs in the Wolfcamp A and B have far below-average carbonate content. The highest proportion of HPPs produce from silaceous mudrock facies.
Oil producibility can be related to original oil in place, reservoir quality (rock properties) and reservoir drive (fluid properties). We captured rock and fluid properties by examining water saturation, porosity, hydrocarbon-filled porosity and oil formation volume factor. In the Wolfcamp A, HPPs tend to have average-to-high hydrocarbon-filled porosity and about average formation volume factor.
While gas presence plays a vital role in reservoir drive, large producing GOR values are detrimental to well performance, as can be seen clearly in a Wolfcamp A production versus GOR cross plot (figure 4). The high-performing producers all had first-year GORs less than 2 million cubic feet per stock tank barrel, with most of them around 1 mcf/stb. Going from 1 to 4 mcf/stb degraded average performance by 30 percent.
We compared the mean values of each producibility metric for top producers versus wells that were not top producers for the three most developed landing zones (Middle Leonard, Wolfcamp A and Wolfcamp D). The uncertainties for these mean values were calculated through bootstrapping. Mineral content, porosity, water saturation, hydrocarbon pore volume and oil density were not significant. The only consistently significant factor is the GOR. Top producers consistently have a lower GOR than the average for each landing zone.
The hydrofracture intensity in fluid and proppant per lateral length were investigated. We found that some exceptionally productive wells had correspondingly large completions (figure 5). In general, high-performing wells had above-average completions and only a few wells, especially in the Wolfcamp B, found success with less intense completions. Some Wolfcamp B wells found success with unbalanced fluid-to-proppant ratios.
Most high performing wells were drilled from a multi-well pad. All these high-performing wells were drilled within six months of the first well in that pad. The vast majority of these pads had above-average median production. This is in spite of highly varying per-well performance even at the pad level (figure 3).
Unique Wells
Some wells bear further mentioning. Several high-performing wells were drilled well outside of the geomodel we used, on the Eastern Shelf in Scurry County. These wells are in the Garden City South field. Some operators classify the target as the Cline Shale, which under our nomenclature is the Wolfcamp D.
In addition, one HPP was drilled below the Wolfcamp in Martin County. The operator listed the Atoka as the target for this well. It is 1,100-feet in lateral length, has a GOR of 4,300 scf/bbl, and has a modest completion of 500 thousand gallons of fluid and 400 thousand pounds of proppant. Remarkably, this well produced 187 bbl/1000 feet per day for the first year, making it the best well by that metric.
Conclusions
In this study, we identified the best wells in the Midland basin and then looked at their properties.
The Wolfcamp A had the largest number of HPPs, followed by the Middle Leonard. Siliceous mudstones were the most common reservoir lithology. Geologic controls vary by zone. For example, in the Middle Leonard, average reservoir properties, coupled with intense completions, led to HPPs. In the Wolfcamp A, HPPs often had high hydrocarbon-filled porosity coupled with average-to-intense completions.
Lastly, we found that a few wells defy expectations to stand out. Some wells with below-average hydrocarbon-filled porosity and formation volume factor were among the top 1 percent. Several Wolfcamp D wells on the eastern shelf were high performing. The best well in production per length produces from below the Wolfcamp.
Acknowledgments
We thank Interpretation for choosing to publish our article “Properties of high-performing horizontal wells in the Midland Basin,” from which this article is adapted. The authors wish to thank the sponsors of the Tight Oil Resource Assessment Consortium at the Bureau of Economic Geology, Jackson School of Geoscience, University of Texas at Austin and the State of Texas Advanced Resources Recovery project for their support. F. Male was also supported by the ExxonMobil grant “Heterogeneity and Unconventional Production” (principal investigator: M. Marder) and by internal funds from Pennsylvania State University. H. Rogers assisted with data collection. Valuable discussion came from M. Marder, L. Lake, G. Hunter, L.R. Maraggi, E. Turkoz, Z. Jabeen and D. Ertas. Schlumberger donated Petrel licenses to UT.