<img height="1" width="1" style="display:none;" alt="" src="https://px.ads.linkedin.com/collect/?pid=2603577&amp;fmt=gif">
Geoteric | November 14, 2014

IFC+: Class definition in the distribution model

IFC+: Class definition in the distribution model

Geoteric on 14 Nov 2014
Find me on:

IFC+ offers the interpreter an advanced method of Seismic Facies classification in a rich multi-attribute environment. IFC+ utilises a sophisticated attribute distribution model, which provides the ability to detect subtle trends, to enable very precise differentiation of different facies classes.

Within the IFC+ tool, classes can be defined either through zones of interest between well markers, or by manually selected sample areas. The example shows 1D Gaussians for ease of drawing/understanding, but the concept extends to the multi-dimensional Gaussians used in IFC+.  The below example shows how three classes interact and are defined.

It shows 3 Gaussians each with a different standard deviation/variance (width).  Unit-area normalisation means that where the power is higher, the narrower the standard deviation. This means that where two groups overlap, the one with higher power will “win” the classification.  In the case below even if the green group was centered in the same place as the red group rather than on its down-slope it would still dominate it where they overlap.

Increasing “acceptance level” distorts the unit-normalisation because it widens all groups (or all sub-groups of the parent group since acceptance level is per parent group rather than global) but does not lower the power.  So the key thing is the decision boundary between two overlapping groups (obvious below as where the green group over-rides the red where it is higher power).  

Now even though all groups widen by the same variance modifier parameter, the widening of the red group will have little effect except for previously unclassified voxels on the far right, the blue group will widen and begin to chew up the left side of the red group initially, but the wider the acceptance the green group will get wider and wider and progressively chew up more and more of the red group before starting on the blue group as it spreads left.  So in this case as you slide acceptance level up, initially blue would steal some red voxels while green would steal others, but after some point green would then also start stealing the blue voxels it previously stole from the red group (then some of blue’s original voxels).

Hints and Tips, Blog