Stratigraphic Hierarchy, Seismic
Integration, and Up-Scaling Issues in a Low Net-to-Gross, Finely Stratified Reservoir: Beaver Lodge Devonian Unit, North Dakota
Mark D. Sonnenfeld1, Hai-Zui Meng1, K. Lyn Canter1, Brian
Rothkopf1, Michael J. Uland1,
Brad C. Watts2,
and Michele Simon2
1 iReservoir.com, Inc., 1490 W. Canal Court, Suite 2000, Littleton, Colorado 80120:
[email protected]
2 Amerada Hess Corp., 500 Dallas St., Houston, Texas, 77002: [email protected]
For the first time in its 52-year history, the Beaver Lodge Devonian Unit reservoir has a technically comprehensive, full-field 3D description built on an integrated foundation of geologic, geophysics, petrophysics, engineering, and operational data. This type of effort facilitates confident, proactive reservoir management recommendations for optimizing field development, even for mature waterfloods such as BLDU.
The workflow was not only multidisciplinary, but both parallel and iterative in nature. Preliminary full-field simulation preceded completion of core description, detailed correlation, and log conditioning thereby enhancing data quality control in addition to addressing issues such as field-scale material balance, multiple contacts, and permeability transforms. As a standalone visualization and analysis tool, and in conjunction with ongoing reservoir simulation, the model has been used to design an infill drilling and re-completion program which has yielded significant incremental oil recovery.
Facies
Approximately 2,921 feet of core were described for the BLDU project. During the core description phase thirteen depositional facies were described. The traditional emphasis for Duperow pay has been on variably dolomitized stromotoporoid bioherms. This study separates true biostromal facies which are tight lime boundstones with muddy lime matrix, from both stromatoporoid- and peloid-bearing, grain-dominated dolomitic pay facies that both succeed, and sometimes grade laterally away from true biostromes as aprons, tidal channels, and bars. We recognize three principal reservoir facies: stromatoporoid-rich, peloidal dolomitic lime packstone (f.5), micropeloid dolomite packstone (f.6), and stromatolitic dolomite (tidal flat) boundstone (f.10). These three facies occupy different positions within stratigraphic cycles and have different aerial distribution relative to paleostructure.
We recognize two types of facies associations: so-called “chemical-dominated” and “marine-dominated”. Marine-dominated facies associations are peloid and stromotoporoid-bearing and experienced more physical processes such as winnowing, as opposed to chemical depositional processes such as suspension deposition of evaporates and azoic carbonate mudstones. Marine-dominated facies by far contain the most net pay (f.5, f.6). The chemical-dominated facies model applies to rock successions that contain few fossils and are instead characterized by dolomitic and somewhat evaporitic rocks that represent deposition on a shallow protected marine shelf during times dominated by chemically restricted, poorly oxygenated waters. The chemically restricted depositional model is mud-dominated and has limited reservoir development, either at the regressive tops of cycles as porous and permeable thin beds of f.6 or f.10, or as thin beds of f.6 at the initial, transgressive portions at cycle bases.
Four lithologic facies (porous dolomite, shaley microporous dolomite, limestone, and anhydrite) were distributed throughout the gOcad model. Although the suite of depositional facies (f.1-f.15) were not distributed throughout the gOcad model, the facies description/discretization effort was nonetheless critical for our appreciation of reservoir facies and recognition of cyclic facies arrangements. Cyclic facies arrangements are particularly important not only for understanding vertical barriers to fluid-flow, but also because they provide the basis for field wide correlation of log markers and log patterns. At times of maximum transgression, both for 5th-order cycles and for larger scale, 4th-order sequences, flooding facies dominated by f.2 tend to create vertical baffles/barriers to flow at the core of thick limestone intervals.
Stratigraphic Framework
By recognizing the two types of facies associations, we have refined both the facies models and the understanding of cyclicity, with the end-result being a higher resolution, higher confidence correlation framework. Both depositional systems are highly aggradational, perhaps with limited progradation away from local paleostructural nuclei. One implication of extreme aggradation is a high degree of confidence for lateral well log correlation, both locally and regionally. This is responsible for the historically successful waterflood performance based on vertically isolated pays with very high degrees of lateral continuity.
Cycle Hierarchy
We recognize stratigraphic cycles of various scales, including four major regionally correlative 4th-order high-frequency sequences, approximately twelve additional 5th-order cycles (correlatable across BLDU), and some additional very small-scale cycles (perhaps 6th-order?) with vertical thicknesses less than 5 ft. We have found through experience over numerous projects that proportion curves help graphically distinguish very high frequency cyclicity from larger-scale vertical trends and cycles. Proportion curves are simple cumulative probablility plots (0-100% on horizontal axis), with the vertical axis representing stratigraphic position relative to one or more stratigraphic reference levels or datums. Proportion curves can highlight vertical trends and proportions for lithology, facies, or other discretized parameters. Proportion curves are most effective and precise when correlations have progressed to the point of allowing proportion curve construction using multiple, as opposed to single reference horizons. With fewer stratigraphic datums, proportion curves tend to “smear out” most small-scale cycles and would not properly represent the stratigraphic position of the Duperow’s very thin pay zones. We constrained our proportion curves with 9 datums spaced over the main pay interval.
A key point is that the 4 major cycles are each composed of only 10-20% net pay by thickness. Each cycle tends to form a natural large-scale flow unit which in some cases have separate oil-water contacts. Individual pay streaks are so thin, commonly on the order of 1’-5’, that a high degree of deterministic correlation is required to adequately constrain interwell porosity interpolation for the 3D
geomodel.
Geophysics
Geophysical study of BLDU included seismic
inversion, velocity model building and time-to-depth conversion. It also included generation of
seismic
attributes and correlation analysis of these attributes to rock properties, which allowed us to incorporate
seismic
data when building the final reservoir model for engineering simulation. In the end, inter-well rock properties were distributed though the incorporation of
seismic
data integrated with derived geostatistical relationships between
seismic
acoustic impedance (AI) and physical rock properties. An in-house technique termed Pseudo-Stochastic Inversion (PSI) was applied to increase the resolution of AI within the reservoir depth model and get higher correlation between porosity and AI values. The best correlation between porosity and AI was achieved within dolomite lithologies; therefore
seismic
AI was used as soft data
only within dolomite regions. The reservoir model was built using porosity obtained from Sequential Gaussian Simulation with Collocated Co-Kriging using
seismic
AI as soft data and only within dolomite regions.
Petrophysics
Well log and core data for 121 wells in the BLDU and Capa areas were processed to create a petrophysically consistent data set. This data set integrated with the seismic
rock property data to create a 3D geo-model of lithology and reservoir flow-unit rock properties. The petrophysical well log data was used in the geo-model as the constraining porosity, permeability, and net-to-gross parameters in the reservoir flow simulator. Due to the Duperow pay’s thin-bedded nature, iterative core-to-log shifting to a precision of +/- 1 ft was necessary. To create this quality-controlled, integrated log data set, the log curves were edited, normalized, and then processed to provide estimates of lithology, effective porosity, and water saturation that were calibrated to core analysis, core descriptions,
seismic
attributes, and cased-hole engineering PLT flow log results.
3D Reservoir Model
The final stage of the reservoir characterization is a 3D reservoir model populated with rock properties based on a prior petrophysical, geological and geophysical study of the field. Construction of the 3D reservoir model consisted of three main parts: 1) structural framework, 2) reservoir property modeling
, 3) up-scaling.
The preliminary 3D reservoir model was built using well logs only. Properties from the wells were geostatistically distributed using variograms describing the spatial relationship of the reservoir properties. Reservoir properties in the final 3D reservoir model were distributed with additional constraint from an acoustic impedance model derived from seismic
inversion, once reliable correlation between
seismic
impedance and well log porosity was established. Integration of high quality
seismic
data enabled us to take full advantage of the high aerial resolution that only
seismic
can provide, coupled with the high vertical resolution provided by well logs.
Stochastic models of reservoir porosity, permeability, and lithofacies were iteratively generated at different stages of the workflow, as progressively more data from core description, stratigraphic interpretation, logs, and engineering data were integrated. Each refined model iteration was then exported to ECLIPSE for history-match simulation.
Two scales of reservoir model were constructed for BLDU. One is a fine-scale geomodel, constructed so that all geologically significant units could be captured; the second is a coarser, up-scaled model used for reservoir simulation.
Engineering and Simulation
Though not the emphasis of this presentation, the Engineering Study first evaluated historical waterflood performance using classic reservoir engineering techniques and conceptual models. The Simulation Study was pursued with a full-field model using GeoQuest’s ECLIPSE simulator that was history matched from July 1951 through December 2001. Four prediction scenarios were run: case 1) base case, extrapolation of existing wells; case 2) workover cases, case 3) perforate unswept zones in existing wells, case 4) combination of cases 2 and 3. As a standalone visualization and analysis tool, and in conjunction with ongoing reservoir simulation, the model has been used to design an infill drilling and re-completion program which has yielded significant incremental oil recovery.