A volcanic reservoir: Facies distribution model accounting for pressure communication

Abstract

Abstract A TCF-class gas field has been producing over decades in Japan. The reservoir body comprises stacked Rhyolite lava domes erupted under submarine environment. Porous network developed in each dome and rapid chilling by seawater caused Hyaloclastite to deposit over it. Although Hyaloclastite is also porous in this field, its permeability has been dramatically reduced by clay minerals. These reservoir facies are interbedded with Basaltic sheets erupted alternately and Mud sedimented while the volcano was dormant. Based on stratigraphic correlation, multiple reservoirs were originally interpreted. Gas had been produced according to the priority assigned to each. However, it was noticed after 10-20 years of production that pressures of all unexploited units had been declining with variety of rates. We confirmed that by subsequent survey and decided to remodel the whole pressure system. As seismic data had not been informative, we made maximum use of the pressure data to resolve the nature of the communications. We employed multi-point geostatistics to capture the complicated facies distribution patterns. Realizations are then calibrated against pressure history through a probability perturbation technique. A common difficulty of building proper stationary training image is further pronounced in modeling a volcanic reservoir. We solved this by iteratively adjusting a training image inferred from literatures until acceptable history match was reached with reasonable number of perturbations. Another issue was undetermined field extent. We settled this by stochastically populating a rectangular-parallelepiped modeling space of regular cells with pay and non-pay facies. Resulted realizations closely simulate pressure history and look realistic in both facies distribution and field extent. They ascribe the uneven pressure decline to narrow channels of Rhyolite and confinements by Hyaloclastite. Finalized training image indicates more intensity in facies spatial variation than had been expected. Based on 20 realizations, roughly 15% of scatter was estimated in OGIP around the mean.