Abstract

Reservoir continuity is often problematic in reservoirs of fluvial or deltaic origin, where individual sand bodies are of limited lateral extent and cannot be correlated between wells. Evaluating such reservoirs is difficult, especially during field appraisal, when the distance between wells is large. Accurate prediction of the actual positions and optimal spacing of future producer wells, is precluded by lack of information. Uncertainty is an inherent characteristic of traditional deterministic reservoir models.

Probabilistic modelling techniques have therefore been used to reduce and quantify such uncertainties in various discontinuous reservoirs. Stochastic models were generated which honour the available well data but also incorporate general knowledge on the shape and architecture of similar reservoirs elsewhere. Instead of one model, a set of possible models is generated via a Monte Carlo approach, so that the full range of variability in the subsurface can be investigated. The technique was validated by modelling the well-documented sub-recent delta plain deposits of the Rio Grande (Texas) using a limited well data set. The effective properties of the models appear to compare well with the actual configuration in the subsurface as indicated by a much denser well coverage.

Statistical evaluation of a set of such models makes it possible to make quantified statements about connectivity, optimal well spacing and other effective properties of the reservoir. In addition, the most representative models can be selected for numerical simulation to quantify the effect of reservoir architecture on recovery efficiency.

During all stages of the development of a siliciclastic reservoir the knowledge of sand body architecture and of the properties between wells is usually insufficient to enable complete prediction of the fields potential behaviour. Geological uncertainties are therefore an inherent element of reserves estimation, development planning and field management. These uncertainties may influence important investment decisions, such as well spacing or the size of offshore facilities.

Such unknowns are especially important in reservoirs where sand bodies are narrower than development well spacings, as in reservoirs of fluvial origin, where variations in both reservoir quality and architecture cannot be predicted by routine, deterministic, correlation. To solve this dilemma other, probabilistic, techniques have been developed which aim at quantifying and reducing the uncertainties in our knowledge of the interwell space. These techniques are based on combining 'hard' data obtained from existing wells in a reservoir via stochastic methods with 'soft' data derived from a knowledge of comparable reservoirs and geological studies of analagous outcrops. The aim of the modelling is to quantify the effects of alternative well spacing scenarios and to generate input for use in numerical simulation (fluid flow) studies. Both 2 and 3-D representations of a reservoir can be produced; the examples discussed below refer to 2-D cases which were used to develop the modelling technique. The approach is similar to that described by Haldorsen and Lake (1982) for shale modelling.