3-D characterization of a clastic reservoir analog: From 3-D GPR data to a3-D fluid permeability model

Citation
Rb. Szerbiak et al., 3-D characterization of a clastic reservoir analog: From 3-D GPR data to a3-D fluid permeability model, GEOPHYSICS, 66(4), 2001, pp. 1026-1037
Citations number
40
Categorie Soggetti
Earth Sciences
Journal title
GEOPHYSICS
ISSN journal
00168033 → ACNP
Volume
66
Issue
4
Year of publication
2001
Pages
1026 - 1037
Database
ISI
SICI code
0016-8033(200107/08)66:4<1026:3COACR>2.0.ZU;2-Y
Abstract
A three-dimensional (3-D) 100 MHz ground-penetrating radar (GPR) data volum e is the basis of insitu characterization of a fluvial reservoir analog in the Ferron Sandstone of east-central Utah. We use the GPR reflection times to image the bounding surfaces via 3-D velocity estimation and depth migrat ion, and we use the 3-D amplitude distribution to generate a geostatistical model of the dimensions, orientations, and geometries of the internal stru ctures from the surface down to similar to 12 m depth. Each sedimentologica l element is assigned a realistic fluid permeability distribution by krigin g with the 3-D correlation structures derived from the GPR data and which a re constrained by the permeabilities measured in cores and in plugs extract ed from the adjacent cliff face. The 3-D GPR image shows that GPR facies changes can be interpreted to locat e sedimentological bounding surfaces, even when the surfaces do not corresp ond to strong GPR reflections. The site contains two main sedimentary regim es. The upper similar to5 rn contain trough cross-bedded sandstone with ave rage permeability of similar to 40 Ind and maximum correlation lengths simi lar to (5.5-12.5) x (3.5-8.0) x (0.2-1.5) m. The lower similar to7 m contai n scour and fill fluvial deposits with average permeability varying from si milar to 30 Ind to similar to 15 md as clay content increases, and maximum correlation lengths similar to (4.0-12.5) x (3.0-10.0) x (0.5-1.0) m. These representations are suitable for input to fluid flow modeling.