First Published: AAPG June 2024 by Keyla Gonzalez, TGS, Zoltan Sylvester, Bureau of Economic Geology, The University of Texas at Austin, Alejandro Valenciano, TGS and Ben Lasscock, TGS.

Here we present a case study on semi-automated, basin-scale geomodeling.

In the current data analytics landscape of artificial intelligence, machine learning, and multi-variate modeling, a key differentiator of a determinative model is the quality and scope of the input data. Predicting well performance with a high level of accuracy requires not only well production and completion data, but also high-fidelity geologic data differentiating well landing zones and reservoir quality. TGS, with an industry leading, high-quality, comprehensive log library, is uniquely positioned to provide the geologic context for the next generation of multi-variate predictive models and subsurface interpretations.

However, correlating and interpreting well logs are necessary, labor-intensive tasks for building large scale stratigraphic models used in multi-variate analytics, geomodeling, and reservoir simulation workflows. Aside from the high resource and time constraints, manual correlation and interpretations can also vary from interpreter to interpreter and often do not make use of all well and log data available. These workflows often require interpreters to focus on fine-scale details in a limited number of logs, making it challenging and time-consuming to assess the large-scale structure of the subsurface. Furthermore, generating accurate 3D property and stratigraphic volumes from well log data, especially in horizontal sections of producing formations, faces obstacles such as data quality variability, lateral reservoir variability, and the complexity of accurately modeling these variations. There is a clear need for automation to improve efficiency and reproducibility. Various approaches have been proposed to automate geological boundary detection from well log data. Dynamic Time Warping (DTW) and artificial intelligence (AI) are promising concepts for correlating signal sequences and extending to the domain of geology for
well-to-well correlation (Zoraster et al., 2004; Lineman et al., 1987; Smith and Waterman, 1980; Le Nir et al., 1998; Baldwin et al., 1989; Luthi and Bryant, 1997; Po-Yen Wu et al., 2018; Brazell et al., 2019; Tokpanov et al., 2020).

Fig 3_Rustler_ChronostratigraphicDiagram

Study Area
The study area of interest (AOI) is the Midland Basin, spanning Glasscock, Howard, Martin, and Midland counties. We use an extensive dataset of approximately 30,000 vertical and 6,550 producing horizontal wells. The Midland Basin’s size, complex geology, stacked pay zones, and variable lithologies make extensive manual interpretation prohibitively expensive and therefore a good test case for this workflow.

The ChronoLog (Sylvester, 2023) methodology requires an initial input set of interpreted formation tops to constrain the well log correlation. We select interpreted formations tops that provide the largest span of our 3D property generation spatially and in-depth; these include the Rustler, Bone Spring/Upper Spraberry, Wolfcamp, Strawn, Devonian Carbonate, and Ellenburger.

To read the full article, click here