Declustering

We Drill the Best Rock

Wells are not placed at random. We drill where seismic and geology promise the best reservoir, so the samples cluster in the good zones and the simple average of the data is biased, usually too optimistic. Feeding that biased histogram into a model would carry the bias straight into every estimate of volume and flow.

Cell Declustering

Cell declustering corrects the bias by laying a regular grid over the samples and weighting each sample by the inverse of how many samples share its cell. A tight cluster of ten wells in one cell then carries about the same total weight as a single well in an empty cell, so the crowded good rock no longer dominates the statistics. In the widget the declustered mean pulls back toward the true field average as the weights take effect.

Choosing the Cell Size

The correction depends on the declustering cell size. Too small and every sample sits alone in its own cell, so the weights are all equal and nothing changes; too large and every sample falls in one cell, again equal weights. The useful size is comparable to the spacing of the clusters, where the weights genuinely separate dense areas from sparse ones. It is the declustered histogram, not the raw one, that the property model should reproduce.