Multiple-Point Simulation Applied to High Sinuosity Fluvial Systems Modeling: Mina El Carmen Formation, San Jorge Gulf Basin, Argentina

CLAUDIO LARRIESTRA and HUGO GOMEZ

This research was presented at 2009 AAPG International Conference and Exhibition held in Río de Janeiro, Brasil, and it was distinguished with Top Ten Poster Presentation Award

Introduction: Diadema field is located in the north flank of San Jorge Gulf Basin, Patagonia region, southern Argentina. The field is producing since 1920 from the upper Cretaceous El Trebol, Comodoro Rivadavia and Mina El Carmen Formations. The uppermost formations (El trebol and Comodoro Rivadavia) are currently producing by secondary and tertiary recovery. The main reserves of the field are in the meandering fluvial system beds of Mina El Carmen formation. From 1500 wells drilled in the field only 230 wells reached Mina El Carmen formation (Fig. 1). The purpose of the study is to evaluate the probability of intersecting channels in new appraisal wells to be drilled.

Figure 1: Location Map showing wells used in the models.

Data features: Well data present several characteristics, which make modeling jobs particularly difficult. Data points control distribution are irregularly spaced (clustered), an important drawback for estimation and simulation algorithms applications. Moreover, logs available have very different technologies and only shale and sand discrimination was possible.The reservoir sands are under seismic resolution (thickness average 5 mts) and it is impossible to separate sand from shale areas using seismic attributes in time or frequency domains. Additionally, channel widths are less than the average of inter-well distance as it was shown in well correlation. Finally, to emphasize the modeling effort, channels belong to a high sinuosity meandering system.

Methodology: The study was performed using a sand/shale indicator curve, derived from a normalized SP curve. The methodology used consisted of a comparison of 30 realizations results using Sequential Gaussian Simulation (SGS), Sequential Indicator Simulation (SIS), Object-Based Simulation (OBS) and Multiple-point Sequential Simulations (MPS) performed over the same facies data. Cross-validation was the criteria to evaluate the results of each modeling process. Multiple-point stochastic sequential simulation is a statistical procedure that allows to simulate categorical variables, particularly curvilinear facies structures, from relative frequencies observed in training images. The SNESIM algorithm (Strebelle 2002), one of the MPS methods, was used in Multiple-point stochastic simulation process.The Training image was built by taking a current high sinuousity meandering river, as an example of the probable sedimentary environment for Cretaceous Mina El Carmen formation.

 The training model: The model used to guide the MPS simulation was built using the deposits shapes of Sanborombón meandering river, located on the east of Buenos Aires province, Argentina (fig. 2).

Figure 2: Satellite view of San Borombón River used as model and digitized interpretation of meandering system. Drawing details show point bars as result of adding elemental chords. 

The model was made using simple sinuous and non-regular shapes drawn by hand. After that, layer by layer were digitized and converted into a tridimensional volume in ASCII format to be used in the SGems public software (Remy, et al., 2009). The model is statistically stationary because the probability to find different shapes is approximately the same in the whole volume (fig. 3). 

Figure 3: Part of Training Volume made with simple shapes. Grouping simple shapes will result in more complex meandering system shapes

Simulation details: Sequential Gaussian Simulation was made using SP curve normalized geometrically ranging from cero units (clean sand) to 80 units (shale). Variogram used was exponential type with N-S anisotropy ratio of 1.25. Sequential Indicator Simulation was used with indicator sand curve built with normalized SP curve with 70 unit cut-off. Variogram used was exponential type with N-S anisotropy ratio of 1.25. Object Based Simulation was made using a commercial modeling package and the geometrical parameters used (channel width, orientation, sinuosity, repulsion, thickness, etc.) was derived from geological experience in Mina El Carmen formation. Maximum data points honored was 90% and convergence was difficult to reach.

Multiple Point Simulation: Several attempts were made with SNESIM algorithm of SGems software. The most satisfying results were obtained without zones, affinities changes or training image rotation. We think the training image itself keeps the most important characteristics of the meandering system.

 Stochastic Simulation Results: Comparison between methods

Figure 4 shows the same layer of the sand probability volume after 30 realizations, estimated with above mentioned methods. Pixel based methods (SGS and SIS) overestimated the size of sand bodies and they made a unrealistic sandy forms.Object based simulation was unable to honor the whole data and probability sand volume is non representative of Mina El Carmen reservoirs. 

Figure 4: Stochastic simulation results after 30 realizations. Figures show one layer of the resulting volume.

On the other hand, Multiple Point simulation represents very well the sedimentary concept assigned to Mina El Carmen formation, i.e. high sinuousity meandering system with channel width less than inter-well distance in average. When we compare the probability of sand volume with the current model used, we can see correspondence between shapes, widths and sinuosities. Although we use simple shapes, simulation control of hard data and the size of search templates allowed the addition of simple curves generating more complex shapes such as point bars (Fig. 5). Moreover, Training Image stationarity allows to see probable changes in the channel mean directions, from N-S in the north of the field towards NW-SE in the south of the field. Appraisal wells drilled after modeling showed a very close estimation between sand probability curve (derived from MPS) and recorded SP logs (Fig. 6)

 Figure 5: Multiple point simulation results. Three layers of sand probability volume compared with several views of the current model used

Conclusions: It is possible to model high sinuosity meandering system using simple training image with the condition of stationarity satisfied in the whole volume. For data configuration and characteristics similar to those of Diadema field, SNESIM algorithm is the most simple and efficient solution.

Figure 6: Appraisal well drilled after modeling. Uncertainty increases with depth due to less quantity of control data.

References

Remy, N., Boucher A., and Wu J., 2009, Applied Geostatistics with SGeMS, Cambridge Univ. Press, NY

Strebelle, S., 2002, Conditional simulation of complex geological structures using multiple-point statistics, Math. Geol., v. 34, nro. 1, p.1-21