A geostatistical modelling technique called cokriging is used to describe the lateral variations of porosity, f, in a synthetic and a real reservoir. Using this method, an error-qualified porosity model is estimated for each of the two reservoirs from sparse well porosity measurements and seismically derived velocities. The method capitalizes on the high spatial density of the seismic measurements and on their correlation with f. Compared with conventional reservoir models derived solely from sparse well control, the seismically consistent models are better spatially constrained and, hence, provide more detailed and accurate reconstructions of the porosity variations.

In areas of rapid lithologic variations, the areal extent of sand and shale units usually cannot be inferred from sparse well data alone. Seismically derived interval velocity data can be used to help predict lithologic variations away from wells. However, in general, the overlap of the velocity ranges for sands and shales is such that seismic discrimination of lithology is ambiguous. We present a Monte Carlo technique for numerically simulating the spatial arrangement of sand/shale units. This technique accounts for the ambiguous nature of the seismic velocity information. Rather than calculating a unique sand/shale model, the Monte Carlo method provides a family of alternative lithologic images, all of which are consistent with the data. The range of models reflects the uncertainty of the lithologic classification and is used to assess risk in reservoir development. The sand/shale simulation technique is illustrated using a data set from an oilproducing channel-sand reservoir. Sand/shale cross-sectional simulations are generated along a seismic traverse that intersects three wells. The simulated models reproduce the logderived lithologic sequences at the wells; they are conditioned by interval velocities inverted from the seismic amplitude data and are consistent with the spatial autocorrelation and crosscorrelation structures of the seismic and well data. The seismically derived lithologic models of the reservoir are better spatially constrained than models solely conditioned by well data. However, in keeping with the inherent ambiguity of the seismic information, the exact location of the lateral truncation of the channel sand is not precisely defined in the Monte Carlo lithologic simulations.

Predicting reservoir parameters such as lithology, porosity, permeability, and fluid saturation is essential for estimating reserves, selecting placement of development wells, and forecasting production. To this end, new methods are presented that integrate three-dimensional (3-D) seismic data with well-log, core, and geological information and significantly improve the spatial characterization of reservoirs. Seismic data are used not only to derived detailed 3-D depth images of the reservoir geometry and faulting, but also to infer spatial and temporal variations of rock parameters relevant to production.