Dear Colleagues:

It's been alleged by many that a systematic error runs through the practice of reservoir modeling.

The alleged error involves use of the variogram. Briefly stated, it's alleged that geo-modelers simply assume spatial dependency of rock properties over distances of 100s to 1000s of meters (within a given facies) based on the well spacing that happens to be available to them when modeling petroleum reservoirs, whereas property correlation lengths revealed via evaluation of outcrops, quarry sites and geo-hydrological projects are measured and much shorter, ranging over distances of meters to 10s of meters. It seems correct to state that correlation length within the petroleum industry is simply assumed, not measured, and that this important modeling step is arbitrary, leading often to delivery of overly “smooth” (connected) models that underestimate permeability variance and overestimate recovery efficiency.

Colleagues who understand this issue avoid this practice by using fine grids or pattern element models that result in the use of reasonable relative correlation lengths (correlation length / grid block size). It's alleged that both the software and its application are flawed, the software in the sense that it was not practical to introduce the variogram to the oil industry. How is it that we thought it possible to measure property correlation lengths at depth in petroleum reservoirs with probes, core holes or wells spaced at distances of meters to 10s of meters? It seems this is the reason for the outcry from many in the industry upon the introduction of geostatistics during the 1990s-"use of the variogram rests on the gratuitous assumption of spatial dependency."

Might we hear how geo-modelers come up with estimates of property correlation lengths if it’s other than simply using the well spacing that happens to be available to them?

Jim Dietrich (Member SPE/AAPG)
Pasadena, CA, USA