Oil exploration prospects in an area are typically dependent due to common geological factors. These dependencies can have major impact on how a drilling program should be carried out in the area in order to maximise the income. Despite this, oil companies tend to use these dependencies only to update marginal discovery probabilities after a well has been drilled. The main reason for this is that this is done ad hoc based on the geological understanding, without an explicit underlying model. Thus, exploring possibilities in advance of drilling becomes too time consuming.
We show how Bayesian networks can be used to capture and summarize the underlying geological dependencies in a consistent manner. This gives a full joint probability distribution for all prospects in the area, which easily updates when new wells are drilled. The quick and easy updating allows testing of different exploration strategies. All elements in the network have direct physical interpretation, making it simple both to build the networks and to see which geological effects that have been included. The methodology has been tested on several real world cases, and we will present one such case.