Reservoir Characterization and Stochastic Uncertainty, Its Influence from Development Strategies. Case of Cañadon Amarillo Field, Neuquén Basin, Argentina

Diego E. Velo, Fredy Garzon, and Rafael Vela
YPF S.A., Neuquen, Argentina

The Cañadon Amarillo Field is a combination trap and is located in the southern part of the Mendoza province within the Neuquen Basin. It is on appraisal phase with 59 wells drilled over an area of 11500 hectares. The structural component of the trap is a smooth monocline with closure to the south and west. Faults mainly exhibit low throw and are sub-vertical. Stratigrafically the field is composed of several reservoirs where fluid distribution appears conditioned by sedimentary facies variations. In order to assess resources and uncertainty in an area currently under appraisal, a predictive model was made integrating high resolution stratigraphy, seismic elastic inversion, well logs, rock cores and advanced petrophysics. This model was complemented with engineering and production data for dynamic characterization. The combination of cutoff maps, probability and standard deviation resulted in the identification of regions of stochastic uncertainty and allowed for the placing of development wells in low risk areas and appraisal wells in regions with high oil in place and greater uncertainty.

The Neuquen Basin occupies a wide region of westerm Argentina and is located in the Andean foothills between 35° and 40° south latitude. During the post-rift stage more than 4 km of sediments where accumulated through an arrangement of transgressive and regressive cycles due to variations in sea level. This process was highly influenced by the activity of a magmatic arc and local episodes of tectonic inversion (Figure 1).

Cañadon Amarillo was discovered in 1969 and first began to produce in 1977. It has a complex production reporting history because of multi productive horizons and operator changes. Covering an area of 11500 hectares and with only 59 wells drilled, mainly in the southeastern portion, it is still on appraisal stage.
The Mulichinco formation (Weaver, 1931), of Valanginian age, is the main reservoir in the Cañadon Amarillo field and was deposited during periods of relative low sea level. This unit, which is bounded by large packages of black shale, shows facies heterogeneity vertically as well as along its depositional axis (Schwarz and Howell, 2005; Schwarz et al, 2006). In the area of the Cañadon Amarillo it is composed entirely of marine deposits with shallow and deep facies, including sandstones, carbonate and mixed facies.
The sandstones were deposited in an open sea where normal and storm waves were the main sediment transporting agent. Facies associations grade from low energy in the lower shoreface near the base of fair weather waves to proximal shoreface above the wave base. Carbonates and mixed facies associations are interpreted to be deposited in a marine environment under low to high energy conditions. High energy carbonates have been generated above the abrasion zone and are identified by the presence of intra-basin elements, especially ooids.

Sandstones are predominant in the eastern part of the Cañadon Amarillo area and were developed when clastic sediment deposition inhibited carbonate production (Figure 2). The thickness of the proximal shoreface facies, the main reservoir in the field, appears reduced to the northwest and is greater in the Upper Mulichinco. Carbonates facies are important in the western part of the field where these units are present in three discrete carbonate levels. The distribution of productive facies is very irregular in these rocks.
Regionally the Mulichinco formation has been divided into Lower, Middle and Upper sections, comprised of different lithological characteristics and coincident with the low, transgressive and high system tracks, inside the Mulichinco Lowstand Sequence Set (Schwarz et al., 2006; Gulisano and Gutierrez Pleiming, 1994).

The scheme used todiscriminate Mulichinco formation in the geocellular model is based on a high resolution sequence stratigraphy study which includes surface outcrops and well data (Schwarz, 2008). Under this scheme the silicoclastic facies associations are present in upward grading sequences. These sequences are interpreted as swallowing from offshore to shoreface, indicating a progradation of the facies belts. Those sections without internal discontinuities can be defined as parasequences (Van Wagoner et al., 1990).
Carbonate intervals are in net contact above the silicoclastic rocks. Evidence in the base of the carbonate sequences suggests an initial period of low sedimentation rate with substrate erosion. Also the retrograding arrangement of the parasequences indicates that these intervals were formed mainly during periods of relative rising sea level.

There is a strong correlation between the textural-depositional character of the sedimentary facies and the petrophysical properties of the rocks. Facies that have a better probability of being good reservoir rocks are mainly proximal shoreface and high energy carbonates.

The petrophysical model, based upon capillary pressure curves (Purncell method), indicates that the reservoir rocks of the shoreface facies are characterized by macropores and subordinated mesopores (Figure 3). This was used to calibrate a Leverett J function (Leverett, 1941) for estimation of water saturation above free water level (FWL).

Since Cañadon Amarillo is presently in apprasisal stage, it has large undrilled areas. This means that the use of seismic is necessary in order to guide the modeling process away from the wells. All the decisions made upon this model were taken considering its limitations, in order to make the best use of the seismic information.
In order to make the structural model, 43 wells were used after first being depth tied using a converted Amplitude cube and an Acoustic Impedance volume corresponding to Mulichinco formation. Wells were correlated and all markers were defined using sequence stratigraphic criteria as previously described, with horizons tied to the depth converted seismic. In order to get all possible faults to be considered for the structural model, an “Ant tracking” process was used to extract patches that were compared with the hand picked faults and the acoustic impedance using geomechanical criteria. All the seismic attributes were resampled into 3D cells and then upscaled to the final structural model.
The use of the seismic for modeling requires a simple facies model. Therefore, a mineralogical electrofacies model such as Carbonate, Sandstone and Clay was applied directly from well log data. The method used clay and carbonate volume cutoffs in a first instance and then manual draw of the electrofacies.
After generation all the electrofacies were upscaled at grid resolution. A data analysis process then was executed based upon the upscaled facies data and the previously upscaled Acoustic Impedance.
Vertical proportion curves, 3D variograms and correlations between probabilities of occurrence vs. acoustic impedance were made separately for each zone and for each facies. The 3D model was then populated with facies data using Sequential Indicator Simulation using attribute probability curves. 30 realizations were made in order to be able to select P50 based on volumetric after pore volume generation.

For modeling, each of the petrophysical properties was treated differently according to its character and regional considerations. All quality controlled PHIE logs were upscaled to the grid, facies biased, using Random Pick algorithm, which picks a log point at random from anywhere within the cell. After upscaling, a process of data analysis and data transformation for each facies and for each separate zone in order to set boundaries and condition data for further processing was performed. All the 3D variograms were made using spherical models separated by facies and by zones. The method considered for 3D Porosity modeling is Gaussian Conditional Simulation because it honors the reservoir heterogeneity. The acoustic impedance property was used as a rough tendency to guide the model by the use of a low correlation factor for Collocated Co-kriging. Since this value is low it does not mean that there is no relation between AI and the porosity, in fact the correlation factor does not consider that the relationship between both variables could be stronger in points that are spatially closer. Regionally the correlation factor between AI and Porosity can be affected by changes in the mineralogy and fluid content.
The arithmetic mean of the 30 porosity realizations was generated in order to use as a reference for further processing and volumetrics.
As the nature of the data distribution and the problem to solve required the use of AI as a general tendency for Porosity modeling a blind test was made to asses the reliability of the method in areas away of the wells. In a region yet developed with several wells it can be found that the porosity features are still defined using AI, even though partially, even after the removal of 8 of 19 wells in a region were a stratigraphic trap is present.
All quality controlled PERM logs were upscaled to the grid, facies biased, using Geometric Average. For Vertical Permeability estimation PERM logs were upscaled using Harmonic Average. After upscaling become a process of data analysis and data transformation for each facies, each property and each separate zone in order to set boundaries and conditioning of data for further processing. This resulted in all the variograms necessary to guide the Kriging process in a Gaussian Conditional Simulation. Multiple realizations have been made using the arithmetic mean porosity as a collocated co-kriging tendency for Permeability and NTG, and then mean values were estimated for further processing. The correlation between modeled Permeability and modeled Porosity intends to reproduce its original dispersion.

There is some uncertainty in the placement of the free water level (HFWL) due to low net thickness and high heterogeneity in the individual zones,. Therefore several SW models have been made using the previously generated Leveret J function.
The standard original oil in place (STOIP) was estimated based on the P50 model using conservative criteria for the water-oil contact according to Bo and Rs parameters. Volume height maps were made based upon the STOIP 3D property.

Conditional Gaussian Simulation techniques bring multiple equi-probable realizations so it is possible to analyze the variations between each other, thus identifying uncertainty zones. This is useful since there are wide regions without hard data. It is expected to have more uncertainty in these areas.

Standard deviation maps were made using properties estimated from 30 STOIP, based on porosity realizations (Figure 4 b). These, expressed in the same units as the STOIP, are a measure of the variability of the results between all the realizations. Low standard deviation indicates that the data points tend to be very close to the mean, while high standard deviation indicates that the data are spread out over a large range of values, it represents the amount of risky STOIP.
Since the current economical limit for the development of the Mulichinco formation is already known, it is possible to make a map showing the probability of getting oil above the cutoff (Figure 4 c), by generating logical forms for each STOIP map, then estimating the arithmetic mean.
A stochastic uncertainty can be expressed also by a map showing the probability of occurrence of the mean STOIP value. This is P = (STOIP - Std_deviation) / STOIP. In order to indentify the areas for placing development drilling and to estimate “safe” volumes was used as a coefficient map to get areas and volumes beyond the standard deviation. This is the oil present in case of the minimum real occurrence.

For dynamic characterization a sector model was extracted from the region that is already developed and has enough well control and production history. The coherence in the upscale was verified in every well; also a sensitivity analysis to the upscale algorithm was performed for the permeability. The STOIP of the upscaled model was verified against the static model.

In order to initialize, the model was refined and regularized into a coarser grid, keeping the original vertical scale. WOC, GOC and original reservoir pressure were adjusted and matching of cumulate and initial rates were assessed. The model achieved a history matching in at least 80 % of production from well to well basis. It allowed for the forecast of best locations for producers spaced away some 225 m. The generated files were then used as an input for the economical evaluations.

All the information, including well data and outcrop analogs, indicates that the best reservoir quality is given by the uppermost sequence; however there are other sequences suitable for production in some regions of the model. The potential of the lower sequences is more localized although it could increase in the presence of natural fractures.
The model was fully predictive and helped the placing of appraisal wells based on geology and production forecasts suing an integral project scenario.
We conclude that the making of a stratigraphical and sedimentological concept, using all available data, was the key to achieving a representative model to be used as a planning tool for the development strategies in the Cañadon Amarillo field. In this case, the extensive studies for the sedimentological and petrophysical characterization of the Mulichinco formation, its concepts and the integration with seismic inversion allowed identifying a stratigraphic trap and its sequential framework. These proximal shoreface sandstones, corresponding to mesopores and macropores rock types, are distributed in northwest-southeast oriented belts, and dominant in the southern part of the analyzed region.
As a consequence of this model all the development strategies for the Mulichinco formation were reformulated in order to identify the best locations for the placing development wells in low risk areas, and appraisal in regions with high oil in place and high uncertainty.
The resulting model was fully capable of reproducing well to well production history in Cañadon Amarillo.

Figure 1: Paleogeography ans stratigraphic column of the Neuquen basin.