Best-practice stochastic facies modelling from a channel fill turbidite sandstone analog (the 'Quarry outcrop', Eocene Ainsa Basin, NE Spain)

Abstract

Using data from an outcrop characterization of a sandstone-rich turbidite channel fill (the so-called ‘‘Quarry outcrop’’ in the Ainsa basin), several stochastic facies models were constructed at bedscale resolution (cells 2.5 m [8 ft] wide and 0.05 m [2 in.] thick).</p><p>Several industry-standard reservoir-modeling algorithms were employed: truncated Gaussian simulation, sequential indicator simulation, multiple-point geostatistics, and object-based methods with varying degrees of complexity. </p><p>The degree of similarity (i.e., realism) between realizations and the outcrop characterization was quantified through the use of several responses: (1) static connectivity, (2) effective permeability, and (3) recovery efficiency from waterflood simulations.</p><p>Differences in the responses measured from the outcrop and facies models were observed: these are mostly algorithm related, instead of caused by soft data or different stochastic realizations. Differences increase greatly when the permeability of the heterolithic</p><p>packages and mudstone beds (Ht-M) decreases and reflect the methods’ ability to model the inclined and undulating Ht-M packages and beds that occur in the outcrop. These packages and beds can drape scours and sandstone beds with depositional topography</p><p>and pinch-outs, producing sandstone thinning and dead ends.</p><p>Object-based methods capable of introducing highly undulating Ht-M beds provided the most realistic models. Variogram-based and simple object-based methods failed to capture and reproduce the whole length of undulating beds. </p><p>Multiple-point geostatistics provided realizations with responses intermediate between variogram-based and simple object-based methods and the more successful advanced object-based methods. The conditioning-to-harddata capabilities of multiple-point geostatistics are higher than those of the object-based methods, which give them an added advantage.