The use of computer-aided seismic facies classification techniques continues to grow and play a vital role in interpretation workflows within the exploration and production (E&P) industry. In recent years there has been an explosion in the number of seismic attributes available for use in E&P. These attributes are used to help analyze the subsurface and can reveal important features, from regional geology to detailed reservoir properties. To effectively understand the multitude of seismic attributes, classification techniques have been developed to support the quantitative assessment of exploration targets and to improve reservoir characterization within field development projects. The objective of the facies classification process is to describe subtle characteristics within the seismic data and relate this to lithology and ultimately rock properties to help identify potential hydrocarbon accumulations.

In some cases, however, the facies classification results are affected by noise from the input seismic data and/or horizon surfaces used to define the interval of analysis. These and other issues with input data (e.g. phase or time shifts in different seismic attributes) can generate isolated pixels or voxels of anomalous facies classes, which change rapidly and arbitrarily, making the results less interpretable. These "noisy" class codes must be changed in order to obtain more continuous facies distributions for improved stratigraphic and other types of geological interpretation. Classic smoothing operations based on averaging of neighboring points cannot be used for this purpose since facies classes are non-ordered discrete values (each facies class is unique in terms of data definition).

Intelligent facies map and facies volume smoothing techniques have been introduced to remove noise and spurious anomalies resulting from the facies process. This unique methodology is implemented and performed in two basic steps. First, the smoothing operation detects isolated classes or holes using a mathematical morphology tool.

Second, new classes are assigned using different estimating techniques based on the spatial distribution of the map (or volume). In reality, the holes are estimated only if their similarity is inferior to the similarity threshold chosen by the user.

Facies smoothing technology

Although there are many techniques for estimating facies values, three methods are presented in this paper: Bayes, Dominant and Maximum Similarity Criteria. The detection of "isolated" pixels/voxels is accomplished using a derivation of a classic mathematical morphology tool: the White Top Hat Operator. A re-estimation of the facies class is then performed for each isolated pixel/voxel.

In the first stage of the process we attempt to isolate pixels/voxels that are anomalous. We start by generating binary maps for each of the facies. The original facies map is transposed (translated) into actual class values, and binary maps are generated based on each facies location. These binary maps are basically pixels or voxels containing either ONES (facies exists) or ZEROS (facies does not exist). Once we have the binary maps for each facies, we perform the Opening Operation.

The Opening Operation looks for the areas with isolated ZEROS by analyzing binary maps for each facies. If there is at least one pixel with no facies within a given neighborhood, it is considered an isolated ZERO. The neighborhoods are usually the four and eight connected pixels or voxels. The user can choose the shape of the neighborhoods (type of structural element). Three common types of the structural element are shown on Figure 1.

The difference between the Original binary map and the Opened map produces the Holes map (Top Hat Operator). This map is constituted by the "facies noise," i.e. pixels where facies need to be re-estimated.
Three methods of estimating facies values for the identified noisy areas are presented in this paper. The Dominant Facies Criteria method is a fairly straightforward method where the most dominant facies in the neighborhood, again, based on the selected structural element, prevails. For the Maximum Similarity Criteria method, the facies chosen is the facies which has the maximum similarity among the facies in the structural element neighborhood. For this method of estimation, facies similarity maps or volumes are required. For the Bayes Criteria method, the facies estimated is based on knowing the facies transition probabilities, i.e. the probability to pass from a given facies to another facies following a given direction, and individual facies similarity distributions.

These innovative smoothing techniques have proven successful in both stratigraphic and structural plays. Conventional smoothing techniques fail when applied to seismic facies data.

Case study

A typical workflow normally begins with well data analysis and interval definition for the seismic facies classification process. Calibration of the well information to the seismic attribute data is a vital component of the workflow. Fluid substitution and modeling techniques utilizing crossplot analysis are used iteratively to help analyze the resultant facies volumes and maps. The facies smoothing techniques are applied throughout this process in order to achieve improved continuity of the facies distribution for more accurate well calibrations and enhanced interpretation of the subsurface. The results presented in this paper are from a case study utilizing data from the LaPalma field, which was published in The Leading Edge magazine (V. Linary et al, 2002). This field lies within the Colón Unit oil province (southwest corner of the Maracaibo Basin) in Venezuela. The objective of this work was to delineate the facies distribution and predict the fluid content of the Mirador I Massive sand (Eocene) based on the integration of the information from multiple seismic attributes (amplitude variations with offset, acoustic impedance, coherency) and well data.

A multi-attribute seismic hierarchical classification method was used for this study to delineate the facies distribution. The classification process consists of basically two steps. First, meaningful subsets of the input data are defined based on the multi-dimensional crossplot. A representative (cluster node) is assigned to each subset during this step. The subsets are ordered in accordance with their location in the multi-dimensional crossplot and assigned a class number and a color. Second, individual samples are assigned to the appropriate subsets based on the Euclidean distance.

Figure 2 shows a traditional amplitude map juxtaposed to a comparable horizon slice from the resultant facies volume. The amplitude map in Figure 2a fails to show any differentiation between the wet and oil wells. The wet and oil wells were assigned different seismic facies classes, as illustrated by the horizon slice from the facies volume in Figure 2b. Note the NE-SW trend assigned yellow facies, which is penetrated by the three producing wells. It was interpreted as an oil-bearing sandbar (see geological model in Figure 3). Figure 2c shows the same horizon slice from the facies volume as in Figure 2b, with intelligent smoothing applied. The interpretation depicted here is more easily attainable on this slice versus the original horizon facies slice. Figure 4 shows a comparison of a facies volume profile
before and after intelligent smoothing. The use of the classification methods greatly enhanced the understanding of the reservoir, while the intelligent smoothing improved the interpretability.

Conclusion

Facies classification results are sometimes affected by noise from the input seismic data (e.g. phase or time shifts in different seismic attributes) and/or horizon surfaces used to define the interval of analysis. These "noisy" class codes must be changed in order to obtain more continuous facies distributions for improved stratigraphic and other types of geological interpretation. Awareness of the benefits, and pitfalls, of classification technology, as applied to 3-D seismic data, is critical for successful stratigraphic and structural interpretation. Through sophisticated multi-attribute facies classification methods, rock physics and innovative facies smoothing technology, we were able to validate and fine-tune the original interpretation.

Comments