Eliciting the parameters of a geostatistical model directly from a geologist using a genetic algorithm

Geostatistical models are used to model the spatial relationships between geological features, and can generate stochastic realisations of this geology. Often a geologist can envisage the geology they wish the model to generate, but does not know the optimal parameter values which produce realisations of this geology. A new method of obtaining these optimal parameters, directly from a geologist, is introduced. The method requires a geostatistical model which can be used to generate a realisation of the geology for a given vector of parameters. Then, a population of parameter vectors can be iteratively improved by presenting the corresponding realisations to a geologist and asking them to rank each of them according to how well its geology matches the geology they envisage. This ranking is used as input to a genetic algorithm (GA) which improves the population at each iteration until the current population contains the optimal parameters. An example of this methodology in use by geologists is given for a geostatistical model of the pore space distribution in a rock.