Reservoir modeling workflows have remained essentially unchanged for the past decade, facilitated by commercially available software packages such as Petrel and IRAP RMS. These workflows begin with the construction of a geo-cellular reservoir model, in which a largely deterministic structural and stratigraphic framework is used to define the overall reservoir volume, and also zones within the reservoir. A grid is then constructed within each zone, and geostatistical methods are used to populate each grid-cell with a geological indicator (such as facies or rock type) and associated petrophysical properties. The resulting models typically contain several millions to tens of millions of cells, and are typically upscaled onto a coarser grid prior to flow simulation. Despite its ubiquity, there are a number of shortcomings with this workflow, including (but not limited to):

1. Conventional modeling workflows are slow, often requiring many months from the development of initial model concepts to flow simulation or other outputs.
2. Conceptual geological models become fixed early in the modelling process, with uncertainty explored using geostatistical methods within the framework of a single conceptual model, rather than across a range of possible geological concepts.
3. It is difficult or impossible to rapidly explore a range of conceptual models, and test how these might impact on reservoir behavior.
4. Geostatistical modeling methods are often non-intuitive, and require inputs that are not closely linked to the underlying geological concept.
5. Integration across different disciplines is made more difficult by the use of different software tools, and also by different model grid types and resolution (e.g. fine- versus coarse-grids for geological and flow simulation models; unstructured meshes for geomechanical modeling).