A geological model of a Jurassic reservoir formation was developed by Pangea experts in November-December of 1994. The major objective was the determination of the formation parameters of the reservoir  "P+T" at the Western part of an oil field, which had not been covered by exploratory drilling.

Multi-D Interpretation technique was applied for processing the dataset of well logs and the following maps:
       Top structural map of the "P2" Horizon;
       Base Structural Map of the "T" Horizon;
       Topographical map;
       Structural map of the "A" Horizon;
       Gravity field anomalies map;
       Isochrones map of the "A" Horizon;
       Magnetic field anomalies map;
       Structural map of the "B" Horizon;
       Isochrones map of the "B" Horizon;
       Structural map of the "M" Horizon;
       Structural interpretation results scheme;
       Bottom structural map of the "P" Horizon.

The number of maps was increased by means of special mathematical transformations of the initial fields. The geological characteristics of the area under investigation adopt the following formation structural fields to become highly informative:
       Local components with the window width of 5 and 7 nodes of the regular grid;
       Second derivative on XY direction;
       Standard deviation.

Normalized gradients, as well as gradients of the first and second vertical derivatives, were calculated based on the gravity and magnetic fields. The thicknesses between the structural surfaces P2, T, A, B, M, were also calculated. This resulted in the increase of the number of maps up to 51. The values of predicted parameters in 17 wells were used as the standard ones, of which 7 were located within the area of investigation, and 10 just outside of it.

The analysis of correlation relationships between various field combinations, their transforms and the predicted parameters allowed to choose the most informative map combinations  for predicting different formation parameters.

Figure 1 (left) shows the results of contouring and extrapolation of effective thickness values in the well area. The statistical characteristics of the accuracy are: MC=0.96 and SD=3.1 m. The proximity of multiple correlation (MC) coefficient to 1, indicates a high reliability of the prognoses (the predicted thicknesses values coincide with the available measurements). The magnitude of the standard deviation SD=3.1 is equal to 10% of the maximum predicted value of the reservoir effective thickness.

The porosity, oil saturation, and linear reserve maps, along with the local uncertainty maps for each prediction, were created. Figure 2 (left) shows oil saturation map.

After Pangea had completed the modeling project, two more wells were drilled, as shown in Figure 1 and 2 (right). Well 7 proved to be dry, exactly as predicted. The following table compares the predicted values of formation parameters with the results of sampling in wells 8 and 9, drilled with Pangea's recommendations taken into account (Prediction/Sampling):

Sampling data of formation parameters in wells 8 and 9 were recorded into the database of the area.

Figure 1 and 2 (right) show the resulting model recalculated using the additional data. Due to a good agreement of the drilling data with the initial model, all variations have a characteristic of further detailing and more accurate definition of the contours. Slight variations in statistical characteristics also confirm the reliability of the initial model.

The mathematical routines of the MDI software allow to calculate the local uncertainties of any predicted parameter. Such analysis for the improved model showed that one could not confidently consider the "stripe" of increased (more than 10 meters) pay zone thickness to the south of the main structure (Fig.1, right) to be a perspective zone. Additional drilling in the Southern part of this area is necessary in order to evaluate its perspectives.

Construction of parameter maps for the initial model took two months (including the time for material digitizing), while the improvement of the model accuracy took only 2 hours.