Volume Extractions

The structural model for the SEAM LoFS model includes two reservoirs, a salt body, a sea floor, horizons and faults. The horizons and faults are supplied as triangulated meshes. The remaining surfaces for encapsulating the reservoirs and salt need to be extracted from 3D voxel data. To that end, the creation of structural model required the extraction of closed, FEM-quality triangular meshes for:

The surfaces are based on the following datasets provided by SEAM (with post-processing by MP-Geomechanics):

• Upper reservoir (SFR): Vcl_FBW30mp15cor2Salt_SCA_SFR_Upper__res_linear_salt_1_outside_1.sgy
• Lower reservoir (SCA): DW_SCA_4Petrel.sgy
• Salt: Salt_mask_2_0_cleaned.sgy
• Sea floor: extracted based on the region indicator: Regions_hires.sgy

All input data required extensive cleanup and image processing. In its original form it failed to represent the respective geological formations correctly. In the case of the two reservoirs the data in its original form did not form connected reservoir bodies in large parts of the area of interest.

Representing sloping turbidites on a structural grid is not trivial due to the individual turbidite layer being thin and being closely spaced vertically. The turbidite model is synthetically generated from a layer-cake model model (in deposition space) to which subsequently a displacement transformation is applied. There are two ways in which a good export on a structured grid can be achieved:

• The best result are obtained by using slanted grid for the shale volume. The grid is logically Cartesian and conforming, however the Z-coordinates are properly displaced to capture the surfaces. This could be obtained by displacing vertically the original layer cake model on which the turbidite is defined.
• If a sugarcube is to be produced, one must use proper anti-alias based on the actual displacement transformation combined with a grid of reasonable aspect ratio e.g. 25 x 25 x 5m or similar.

The problem of the supplied data is that it is exported on a sugarcube of poor choice of sampling along with inadequate image processing, that is, no anti-aliasing (Figure 2 [left]). The input was provided in a resolution of 50 x 50 x 2.5 m. The justification for the excessive vertical resolution and the coarse horizontal resolution is not obvious. The resulting small aspect ratio (vertical-to-horizontal dimension) sugarcube cannot properly represent sloping surfaces. Further it leads to poor representation of thin sloping surfaces, such as turbidite layers, i.e. - the connectivity of layers is broken or severely restricted at the voxel level. The problem is made worse by the lack of any anti-aliasing applied on export to seismic volumes.

An example of the actual data (lower reservoir) and the resulting topological connectivity is shown in Figure 3.

While performing anti-aliasing is easily done using the original layer-cake representation of the turbidites, once the results are deformed and exported on a sugarcube, this is no longer possible. Mitigating issues resulting from an aliased model is a complex task consisting of applying various 3D image filter to restore connectivity where possible, and removal of parts of the model that cannot be connected. The steps taken are described on an individual basis for each of the reservoirs (see Upper Reservoir and Lower Reservoir )

On the salt side, the supplied volumes Salt_model_hires_clip50m.sgy and Regions_hires.sgy , were exported on the same grid as the turbidite (50 x 50 x 2.5 m) without any image processing. Areas of thin salt sheets are affected in similar ways as the turbidites. The surface of the main salt body is very jagged.

In addition to the sampling issues described above, the salt marker and indicator cube and property volumes contain a number of numerous artefacts. Such artefacts were observed:

• At the sea floor.
• In a halo of samples around the salt body.
• In areas with salt tails (i.e. thin salt was removed).
• In first inline & crossline.
• First and second sample in each trace.

These artefacts were removed and replaced by compaction trend values or average properties from surrounding samples with the same rock type.

The processing for the reservoir layers had the following objectives:

• Restore connectivity to parts where fluid flow will be severely restricted or blocked due to stair-stepping.
• Remove parts of the turbidite which cannot be restored to a reasonable connectivity.
• Retain the layering suggested by the data.

The standard way to restore connectivity and remove stair-stepping effect is to smooth the data sufficiently. Unfortunately, due to chosen aspect ratio of the voxel cells, the required smoothing will merge vertically disconnected layers. As a result smoothing had to be kept to a minimum and biased to improve stair-steps but not fill adjacent blocks. This also lead to the need to remove partially connected or disconnected layers liberally, in order to make space vertically for adjacent, well defined reservoir layers to be improved. Finally, a number of specialized tools (including 3D-convolutional filters and analysis of connected volumes) had to be designed to remove obvious topological defects in the input cube.

The filter sequence for each of the two reservoirs are different and they are described in more detail in sections Upper Reservoir and Lower Reservoir .

The first step in extracting the Salt and Sea Floor was to fix erroneous values in the original cube. These were cleaned by replacing erroneous values with values from a general compaction trend created from sea bed. The regions where the erroneous values occur are detected by using a combination of the regions cubes and physically plausible values. The values are declared to be erroneous if they are not within 70 % of the trend values. This procedure is applied for Vp , Vs , rho- . Afterwards, a definition of salt and the sea floor which is consistent with the property volumes is produced. The preprocessed files are stored into a new volume, Salt_mask_2_0_cleaned.sgy , subsequently used for salt extraction.