Multiple-Point Statistics

Beyond Two Points

A variogram measures only how pairs of points relate, so it cannot tell a connected channel from a string of blobs that happen to share the same two-point statistics. Curvilinear, connected geology (channels, lobes, fracture corridors) needs information about many points at once. That is what multiple-point statistics (MPS) provides.

Learn From a Training Image

MPS replaces the variogram with a training image: a conceptual, gridded picture of what the geology should look like, under no obligation to honor any well. MPS scans it to learn multi-point patterns, then builds realizations that reproduce those patterns while honoring the actual data. The widget steps through realizations that all share the training image channel style yet differ in layout.

Power and Price

MPS captures shapes and connectivity that pixel-based methods lose, and connectivity is what controls flow. The price is the training image: it has to be stationary and genuinely representative, and a poor or biased one yields a confident but wrong model. Choosing the training image is the central judgment in MPS.