Final time step (step 7) in the history match. We see here the variation in pressure field. Hot colors indicate high pressure, therefore less decline and more gas remaining in place.

After drilling an initial nine delineating wells in the Dutch sector of the southern North Sea, GDF-SUEZ Production knew that additional insight into the reservoir was needed to successfully place infill wells. Production levels varied from well to well and differed from expectations, suggesting thinner reservoirs and more numerous sealing faults than earlier predicted.

A more detailed reservoir model was required if the company was to have confidence in placing additional wells.

GDF-SUEZ teamed with Fugro-Jason to build a highly accurate, predictive reservoir model that incorporated all field data, including seismic, wells, and production history. The objective was to reduce uncertainty and enable better positioning of new wells. The outcome of the study was a new dynamic simulation model that honored all field data, reflected a disciplined workflow, and was sufficiently fine-scale to predict the best locations for future drilling.

A promising play

Located off the coast of The Netherlands, the play is Permian age, with the focus on Rotliegend sandstone with clay-rich sediments deposited in an aeolian and fluvial depositional environment capped by Zechstein evaporite. The field is structurally complex, with two main reservoir zones separated by a clay-rich interval. The gas-bearing sandstone is layered, with aeolian layers showing the highest permeability.

Initial wells were placed at wide intervals to a maximum depth of about 16,400 ft (5,000 m). Well logs clearly showed thin high-porosity sands in the reservoir, making it critical to precisely place each well.

The early wells that struck the aeolian layers were the best producers. Finding these highly productive layers, typically less than 16 ft (5 m) thick, would give GDF-SUEZ the biggest return. To reduce uncertainty and improve profitability potential, GDF-SUEZ was determined to construct an accurate, predictive model that was fine-scaled enough to identify thin reservoir sands so that infill well placement could have the best chance for success.

Blueprint for a fine-scale model

In order to build a reliable highly detailed model, three principles were strictly followed. First, all field data were incorporated into and honored by the model. This included wells, seismic, cores, and production data. Second, quality control was enforced throughout the model-building process. With quality control throughout the entire process, the static and dynamic models are unbiased and match the data without introducing modifiers, arbitrarily dropping inconsistent wells or “fudging” the data. Third, a disciplined and documented workflow was used throughout the modeling process to ensure repeatability and enable updates in the future as new information became available. The team adhered to a specific workflow, including constrained and unconstrained geostatistical inversion, cosimulation, uncertainty assessment, upscaling, and flow simulation.

During data preparation, a database of statistical characteristics of key rock properties was generated. This database was later used to control the output from the geostatistical inversion. Statistical analysis on the well data focused on deriving distributions and correlations for compressional (P)-impedance, porosity, and permeability.

The database also included lithotype definitions because the analysis would be done one lithotype at a time. To be successful, each lithotype has to have distinct elastic properties. In this case, four lithotypes were defined based on porosity, shale content, and production logs. These four lithotypes reside in the interval between the Top Rothliegend and Top Carboniferous.

Geostatistical inversion

Seismic inversion techniques combine well and seismic data to produce a 3-D model of the elastic properties of the reservoir. In this study, an initial static model was generated using deterministic inversion. This model provided a good overall view of the porosity over the field and served as a valuable quality-control check. However, the resolution of this model was insufficient to capture the thin, high-permeability aeolian sandstones.

The next step was to perform a geostatistical inversion, which uses a statistical approach to create multiple, equiprobable models consistent with the seismic, wells, and geology. The inversion parameters were tuned by running the inversion many times with and without well data. The blind-well mode inversions were important in checking the reliability of the constrained inversion and ensuring that the process was not biased. Forty lithology and elastic property realizations were generated for the reservoir.

From a rock physics analysis, relationships between elastic and petrophysical properties were established. Porosity and permeability were co-simulated from the elastic properties using these relationships. In this study, the elastic properties were reliable and stable, consistently predicting the properties at blind-well locations. The lithology was not as easy to predict because of the overlap in elastic properties. As a result, porosity was marginally less reliable.

The 40 realizations that were generated were ranked, and 20 were chosen for dynamic flow simulation, representing the range of possibilities and thereby accounting for uncertainty.

Building static and dynamic models

Once the inversion process was complete, a static model was built and the seismic-derived property models transferred to the geologic model using RockScale. The zonal adjustment method in RockScale transfers the properties in 3-D from the seismic grid to a corner point grid. The relative locations of properties were preserved, ensuring data points in the seismic grid arrived in the correct stratigraphic layer in the corner point grid. For this study, the corner point grid resolution was set to 10 ft (3 m) vertically and 330 by 330 ft (100 by 100 m) laterally. The lateral upscaling was performed during the zonal adjustment process to minimize upscaling errors.

Using the porosity and permeability models and a saturation height function, initial saturation models were built. The volumetric calculations performed on the 20 realizations showed a 13% variation in pore volume; some compartments indicated as much as double the gas volume estimated through material balance from well test data. To resolve this problem, additional faults interpreted from the seismic data were incorporated into the model. The updated structural model showed compartment volumes close to material balance calculations.

Flow simulation continued the integration process by bringing in the production history. This provided a further validation of the static model against history. The first set of 10 models sampled from the geostatistical inversion results were history-matched against gas production data. If the properties in the model are realistic, simulated well bottomhole pressure behavior should match historical (measured) well bottomhole pressure. One of the models is closer to measured pressure, while the others are above and below.

Vertical permeability and pore volume in communication with the well have a major impact on pressure match quality. A single higher pore volume model matching historical gas production was used from the first set of 10 realizations to condition the dynamic model. In the static model, vertical permeability was 0.0 mD for the non-reservoir lithology. No porosity, net-to-gross, or permeability modifiers were required.

All 20 models were now simulated under the same conditions. Based on the quality of the match, some models were eliminated. After the initial history match process, dynamic well parameters were adjusted for each of the remaining models to improve the match.

A final model emerged that honors all available geophysical, geological, and engineering data with a 95% match to historical data in producing wells. The fact that there was no need for porosity or permeability modifiers boosted confidence in the model. This model was used for a 10-year forward model to identify and test five potential infill well sites.

Fulfilling the promise

Taking field data all the way from seismic to simulation is a popular notion but one with many potential pitfalls. In this study, a strict workflow and leading-edge techniques for rock property estimation, geostatistical inversion, model building, and flow simulation achieved the desired result of good history match without modifying key rock properties. Preserving the integrity of the original data, known physical properties, and actual production data makes the model reliable and predictive. The model is now in the hands of the drilling team, which expects to begin drilling on the recommended location early in 2009.