Szerbiak1583

AAPG Bulletin, v. 85, no. 9 (September 2001), pp. 15831608 1583

Detailed internal architectureof a uvial channel sandstonedetermined from outcrop, cores,and 3-D ground-penetratingradar: Example from themiddle Cretaceous FerronSandstone, east-central UtahRucsandra M. Corbeanu, Kristian Soegaard,Robert B. Szerbiak, John B. Thurmond,George A. McMechan, Deming Wang, Steven Snelgrove,Craig B. Forster, and Ari Menitove

ABSTRACT

Ideally, characterization of hydrocarbon reservoirs requires infor-mation about heterogeneity at a submeter scale in three dimensions.Detailed geologic information and permeability data from surfaceand cliff face outcrops and boreholes in the alluvial part of the Fer-ron Sandstone are integrated here with three-dimensional (3-D)ground-penetrating radar (GPR) data for analysis of a near-surfacesandstone reservoir analog in uvial channel deposits. TheGPR sur-vey covers a volume with a surface area of 40 16.5 m and a depthof 12 m. Five architectural elements are identied and described inoutcrop and well cores, using a sixfold hierarchy of bounding sur-faces. Internally, the lower four units consist of ne-grained,parallel-laminated sandstone, and the upper unit consists ofmedium-grained, trough cross-bedded sandstone. The same sedi-mentary architectural elements and associated bounding surfacesare distinguished in the GPR data by making use of principles de-veloped in seismic stratigraphic analysis.

To facilitate comparison of geologic features in the depth do-main and radar reectors in the time domain, the radar data aredepth migrated. The GPR interpretation is carried out mainly onmigrated 100 MHz data with a vertical resolution of about 0.5 m.Measures of the spatial continuity and variation of the rst- andsecond-order bounding surfaces are obtained by computing 3-D ex-perimental variograms for each architectural element (each radar

Copyright 2001. The American Association of Petroleum Geologists. All rights reserved.

Manuscript received October 27, 1999; revised manuscript received September 14, 2000; nal acceptanceNovember 9, 2000.

AUTHORS

Rucsandra M. Corbeanu University ofTexas at Dallas, 2601 N. Floyd Road, Richardson, Texas,75080; [email protected]

Rucsandra M. Corbeanu received her B.Sc. degree ingeoscience from the University of Bucharest, Faculty ofGeology and Geophysics, Romania, in 1991 and iscurrently working toward her Ph.D. in geology at theUniversity of Texas at Dallas. Rucsandras interests includeall aspects of reservoir characterization, geostatistics, andground-penetrating radar applications.

Kristian Soegaard E&P Research Centre,Norsk Hydro ASA, N-5020 Bergen, Norway;[email protected]

Kristian Soegaard received his high school degree inDenmark in 1974, his B.Sc. honors degree in geologyfrom the University of the Witwatersrand inJohannesburg, South Africa, in 1980, and his Ph.D. fromVirginia Polytechnic Institute in Blacksburg, Virginia, in1984. Kriss interests are in description and interpretationof sedimentary systems at all scales and of all ages.

Robert B. Szerbiak University of Texas atDallas, 2601 N. Floyd Road, Richardson, Texas, 75080

Robert Szerbiak received his B.S. degree (1971) ingeoscience at Michigan State University and an M.S.degree (1980) in geophysics from Texas A&M Universityand is currently working toward his Ph.D. in geophysics atthe University of Texas at Dallas. His interests includereservoir characterization and shallow geophysics, uid-ow modeling, geostatistics, ground-penetrating radar,and effective medium theory.

John B. Thurmond Massachusetts Institute ofTechnology, 77 Massachusetts Avenue, 54-913,Cambridge, Massachusetts, 02139

John Thurmond is currently working on his Ph.D, incarbonate sedimentology at the Massachusetts Institute ofTechnology. He received his B.S. degree in geology withhighest honors from the University of Texas at Dallas in1997. His work currently involves 3-D mapping ofcarbonate stratigraphy to understand evolvingmorphologies and the processes that control them.

George A. McMechan University of Texas atDallas, 2601 N. Floyd Road, Richardson, Texas, 75080;[email protected]

George McMechan received a B.A.Sc. degree ingeophysical engineering from the University of BritishColumbia in 1970 and an M.Sc. degree in geophysicsfrom the University of Toronto in 1971. His main researchinterests are waveeld imaging, 3-D seismology, reservoircharacterization, and ground-penetrating radar.

Deming Wang University of Texas at Dallas,2601 N. Floyd Road, Richardson, Texas, 75080

Deming Wang received a B.S. degree with honors (1986)in exploration geophysics from Hefei PolytechnicUniversity, China, an M.S. degree (1993) in geophysicsfrom Peking University, China, and an M.S. degree (2000)in geosciences from the University of Texas at Dallas. Hehas done research on prestack imaging and crossholeimaging.

1584 Ferron Sandstone Internal Architecture

facies). The maximum correlation length of the dominant internalfeatures ranges between 4 and 6 m, and the anisotropy factor rangesbetween 0.6 and 0.95.

INTRODUCTION

Over the past 15 years an acute realization of the limitations of one-dimensional (1-D) facies models (i.e., frommeasured sections, coredescriptions, and well logs) in reconstruction of depositional sys-tems architecture has led to studies of continuous two-dimensional(2-D) outcrop facies maps (Miall and Tyler, 1991). Facies mappingof outcrop analogs yields reliable sedimentologic and stratigraphicdetail that, in conjunction with outcrop permeability and porosityinformation, may be used for characterizing subsurface reservoirsin three dimensions (e.g., Flint and Bryant, 1993). As is the case for1-D stratigraphic sections, however, 2-D outcrop facies maps alsofall short of providing continuous empirical information regardingsedimentary deposits in the third dimension.

A new technology for characterizing sedimentary rocks in threedimensions is now emerging through the use of ground-penetratingradar (GPR) (Baker and Monash, 1991; Gawthorpe et al., 1993).A high-resolution geophysical technique, GPR can provide indirectinformation on lithologic and petrophysical properties of shallowsubsurface rock units. The vertical resolution of GPR is on the orderof a few decimeters, and the depth of penetration is in the range ofmeters to tens of meters (Davis and Annan, 1989). The GPR an-tennas send electromagnetic pulses into the ground to image thesubsurface through the energy reected and diffracted by spatialchanges in electromagnetic properties. Depending on the desiredresolution and depth of penetration, frequencies from 25 MHz to1 GHz can be used. The maximum penetration depth depends onthe attenuation of the GPR signal, which is inversely proportionalto the effective electrical resistivity. The propagation velocity andamount of reected energy depend mainly on the complex dielec-tric permittivities of the materials encountered (Davis and Annan,1989).

The data from GPR have the same potential for describingstratigraphic geometries in specic depositional environments asseismic data have had in providing understanding of larger-scalestratigraphic sequences (Vail, 1977; Posamentier and Vail, 1988;VanWagoner et al., 1990; Weimer and Posamentier, 1993). Unlikeconventional seismic data used in oil exploration, which generallyhave vertical and horizontal resolutions no better than upward of5 and 25 m, respectively, GPR is capable of resolving sedimentaryfeatures at the decimeter scale necessary for describing and inter-preting depositional paleoenvironments. To date, most GPR sur-veys have been performed on unconsolidated, recent sedimentsrather than on consolidated sedimentary sequences in which hy-drocarbon accumulations occur (e.g., Bridge et al., 1995). In near-surface settings, a shallow water table is signicant because theGPR

ACKNOWLEDGEMENTS

The research leading to this article was funded primarilyby the U.S. Department of Energy under Contract DE-FG03-96ER14596 to McMechan and Soegaard with auxil-iary support from the University of Texas at DallasGround-Penetrating Radar Consortium. The migrated GPRdata were interpreted using the PC-based seismic inter-pretation software WinPICS of Kernel Technologies Ltd.The geostatistical analysis was done using the Geostatisti-cal Software Library (GSLIB) programs. The outcropgamma-ray scintillometer was provided by ARCO, andgamma-ray measurements on split cores and Hassler cellpermeability/porosity testing were performed by TerraTek Labs in Salt Lake City.

Gerard Neil Gaynor initiated the use of GPR on out-crop of the Ferron Sandstone for reservoir analog studies.We thank John S. Bridge for his insight into the uvialbarform in the upper 5 m of the channel complex andCoco van den Bergh and Jim Garrison from The FerronGroup Consultants for discussions in the eld and forproviding insight into the position of the Coyote basin sitein the greater depositional framework of the Ferron Sand-stone. We acknowledge Marie D. Schneider for help in in-tegrating geologic outcrop and geophysical data. We alsothank Janok Bhattacharya for his review of an earlier ver-sion of the manuscript. AAPG reviewers Bruce S. Hart,Peter J. McCabe, and Keith W. Shanley provided manycomments that improved the nal version of the article.This article is contribution No. 927 from the GeosciencesDepartment of the University of Texas at Dallas.

Steven Snelgrove University of Utah, 423Wakara Way, Salt Lake City, Utah, 84108

Stephen H. Snelgrove received a B.S. degree in geophysicsand an M.S. degree in geological engineering from theUniversity of Utah. He is currently completing his Ph.D. incivil engineering at the University of Utah. His researchinterests include characterization of aquifers andpetroleum reservoirs using geophysics and geostatistics,and numerical modeling of subsurface ow.

Craig B. Forster University of Utah, 423Wakara Way, Salt Lake City, Utah, 84108;[email protected]

Craig Forster holds degrees in geology and hydrogeologyfrom the University of Waterloo, Canada (M.S. degree),and the University of British Columbia (B.S. degree andPh.D.). His current research program employsinterdisciplinary outcrop-to-simulation studies to assesshow geologically derived permeability heterogeneityshould be incorporated in numerical models ofsubsurface uid ow, mass transport, and heat transfer.

Ari Menitove University of Utah, 423 WakaraWay, Salt Lake City, Utah, 84108

Ari Menitove is currently working as a geological engineerfor Kleinfelder, Inc., in Salt Lake City, Utah. He receivedhis B.S. degree in geophysics from Bates College inLewiston, Maine, in 1993 and his M.E. degree ingeological engineering from the Colorado School of Minesin Golden, Colorado, in 2000.

Corbeanu et al. 1585

signal will be strongly attenuated in the saturated zoneand under the water table. In arid environments, suchas the site studied for this article, the water table is notencountered at depths of GPR penetration.Diagenesis,subsequent fracturing, and preferential weathering ofexposed outcrops are factors that inuence the elec-trical properties of consolidated rock. These factorsoverprint the response of primary sedimentary featuresand complicate the GPR signature. Data preprocess-ing, velocity analysis, and depth migration can focusthe GPR image and reduce artifacts that are unrelatedto primary lithology (Szerbiak et al., in press).

The GPR data commonly consist of 2-D proles(Alexander et al., 1994) or pseudo-3-D grids of widelyspaced intersecting 2-D lines (Bristow, 1995). Notableexceptions are 3-DGPR surveys of recent-delta gravelsin Switzerland (Beres et al., 1995), and Cretaceousshoreface sandstone bodies in Utah (McMechan et al.,1997). If the signal attenuation and dispersion are suf-ciently low, the kinematic properties of GPR data aresimilar (except for scale) to seismic reection data(Fisher et al., 1992a, b). This implies that many of theprocessing techniques developed in contemporary seis-mic studies, and the techniques and facilities devel-oped for interpreting 3-D seismic stratigraphic data(Brown, 1996) may also be used in 3-D GPR investi-gations. Acquisition of GPR data on 3-D grids, and 3-DGPR migration enhance the horizontal and vertical ac-curacy of the GPR image (Szerbiak et al., in press).

To facilitate the integration of geologic and geo-physical data, geostatistical prediction and simulationalgorithms are applied to interpolate the sparse geo-logic control data. Geostatistical analysis of radar re-ections has previously been used to quantify the cor-relation structures found in 2-D GPR proles and toprovide a means for interpretation based on the as-sumption that correlation structures in GPR data aredirectly related to lithologic variation and the internalstructure of different depositional environments (Reaand Knight, 1998). In this article, quantitative com-parative evaluation of the spatial variability of hetero-geneities encountered in uvial deposits is based oncomputation of 3-D experimental variograms of theGPR amplitude data corresponding to each architec-tural element.

The objective of this article is to evaluate the po-tential of 3-D GPR surveys for investigating ancientsedimentary systems and constructing accurate 3-Dreservoir analog models suitable for subsequent hydro-carbon ow simulation. This evaluation is madethrough a case study of a uvial channel sandstone at

Coyote basin, in the Cretaceous upper Ferron Sand-stone Member of the Mancos Shale in east-centralUtah. The procedure for, and utility of, applying 3-DGPR data to ancient siliciclastic rocks is demonstrated.

GEOLOGIC SETT ING

The Coyote basin eld site is located in east-centralUtah, in the upper part of the Cretaceous Ferron Sand-stone known as the Last Chance Delta (Garrison etal., 1997) (Figure 1). The Ferron Sandstone crops outalong the southwestern ank of the San Rafael swelland is the product of a series of uvial-deltaic com-plexes that prograded toward the northeast. Excellentexposures are present along vertical cliffs parallel withthe progradational direction. Exposures perpendicularto the progradation direction are afforded by east-westoriented canyons. The outcrop at Coyote basinincludes a cliff face oriented northwest-southeast andextends approximately 45 m laterally and approxi-mately 12 m vertically. The surface above the cliff faceis a relatively at and barrenmesa top favorable toGPRsurveys. Seismic surveys near cliff faces typically con-tain strong reections from features at the cliff face;however, GPR acquisition design, with dipole antennasoriented perpendicular to the acquisition lines that areparallel with the cliff face, produces and records energythat is polarized near the plane below the survey lineand discriminates against energy coming from the sidesof the line. Thus cliff face reections are less of a prob-lem in GPR data than in seismic data.

Stratigraphic Setting

The Cretaceous Ferron Sandstone Member is one ofseveral northeastward-thinning clastic wedges thatprograded into the Mancos Sea along the western mar-gin of the Cretaceous Interior Seaway during the mid-dle to late Turonian (Ryer, 1981; Gardner, 1992). Theupper part of the Ferron Sandstone is a thick uvial-deltaic complex deposited during a third-order sealevel rise combined with a progressively decreasing rateof sedimentation (Gardner, 1992, 1995).

The Ferron Sandstone is subdivided into seven dis-crete delta lobes (genetic sequences GS1 to GS7)(Ryer, 1981) or major stratigraphic cycles (SC1 toSC7) (Gardner, 1992, 1995). The lower three se-quences (SC1 to SC3) are interpreted as prograda-tional with sea level constant or slowly falling, andexceeded by sediment input. The following two


Figure 1. Location of theCoyote basin site in the FerronSandstone outcrop (the shadedareas) along the southwesternank of the San Rafael Swell ineast-central Utah.

sequences (SC4, SC5) are considered aggradational,with sea level slowly rising and balanced by sedimentinput. The nal two sequences (SC6, SC7) are retro-gradational with relative sea level rising at an increasingrate (Gardner, 1995).

Each sequence or stratigraphic cycle is capped bya major coal bed or coal zone. Recent work of Garrisonet al. (1997) identied at least 12 parasequence setsthat appear to form four high-frequency, fourth-orderdepositional sequences (FS1 to FS4) within the upperpart of the Ferron Sandstone clastic wedge (Figure 2).The uvial channel complex at Coyote basin is locatedat the top of stratigraphic cycle SC3 of Gardner (1995)or parasequence set 3, in the FS2 sequence of Garrisonet al. (1997) (Figure 2). SC3 is capped by coal zone Cand is represented at Coyote basin by nonmarine faciesassociations composed of large distributary channelbelts (Garrison et al., 1997). The paleoshoreline during

deposition of parasequence set 3 is more north-southoriented (approximately 345 azimuth) than that ofthe underlying and overlying parasequence sets (Gar-rison et al., 1997). The channels at Coyote basin aregenerally straight or of low sinuosity (Garrison et al.,1997).

F IELD DATA

The Coyote basin site contains a surface area of 4016.5 m on the mesa top (Figure 3) and 45 12 mvertical exposure at the adjacent cliff face. The dataconsist of detailed sedimentologic, stratigraphic, andpetrophysical data and 3-D GPR data. A leveling sur-vey provided accurate topographic corrections and areference datum for all data sets. For reference, thevolume extent of the survey is roughly equal to the size

Corbeanuet

al.1587

Figure 2. Generalized cross section of upper part of the Ferron Sandstone clastic wedge (modied from Garrison et al., 1997). Stratigraphic location of survey site at Coyotebasin is illustrated. See Figure 1 for location of cross section. Letters A to M identify marker coal horizons; SB1 to SB5 are sequence boundaries; FS1 to FS4 are fourth-ordersequences; 1a to 8b are parasequence sets.


Figure 3. Surface geology of the GPR survey site at Coyote basin. Heavy black lines represent conjugate fracture set orientednorthwest-southeast and northeast-southwest, and the cliff face. CB1 through CB5 are locations of measured stratigraphic sections atthe cliff face. A through D are locations of boreholes from which cores were extracted. The map shows the location of the cliff face(Figure 4), trough cross-bed outcrop (Figure 12), the 3-D grid, and a 200 MHz GPR crossline at x 31.5 m (Figure 13). The origin(x,y) (0,0) is at the southeast corner of the GPR grid; the total grid size is (x,y) (40.0,16.5) m.


of a single voxel in contemporary reservoir owsimulators.

Geologic Data

A wide spectrum of geologic and petrophysical datawere collected at Coyote basin. The surface geologywas mapped on the top of the outcrop where the GPRsurvey was conducted, and the map includes cross-bedding, fractures, and soil cover (Figure 3). A faciesmap with architectural elements and bounding sur-faces of sedimentary deposits was made along the east-facing cliff face (Figure 4). Paleocurrent orientationsfrom 81 trough cross-beds inside the GPR grid andfrom an additional 130 trough cross-beds adjacent tothe survey yielded information about depositionaltrends (Figure 5). Five stratigraphic sections, evenlyspaced along the 45 m long outcrop, provided detailedsedimentologic information (Figure 4). Four 15 mlong, 2.5 in. (6.3 cm) diameter cores were obtainedfrom wells drilled behind the outcrop (Figure 3). Per-meability measurements were performed on 485 coreplugs extracted from the outcrop along the strati-graphic sections at a sample spacing of 10 cm and onthe well cores at a sample spacing of 5 cm; permeabilitymeasurements on the outcrop core plugs were ob-tained using a probe permeameter to test one end ofeach core plug, and along the well cores using acomputer-controlled, stage-mounted, electronic probepermeameter (Snelgrove et al., 1998). Total gamma-ray measurements at a sample spacing of 25 cm alongthe stratigraphic sections were obtained using a hand-held scintillometer. Full spectral gamma-ray measure-ments were made along the well cores at a sample spac-ing of about 3 cm. Measurements of electricalproperties (dielectric permittivity and electrical con-ductivity) were performed on a set of 33 cylindricalplugs, 1 in. (2.5 cm) in diameter, drilled orthogonallyto the well-core axes, providing GPR velocity and at-tenuation information as a function of water satura-tion. Hassler cell permeability/porosity tests were per-formed on the same 33-sample set. Petrographicanalysis of thin sections (Snelgrove et al., 1998) pro-vided quantitative mineralogy information for dielec-tric constant modeling, and parameters such as claycontent and porosity for GPR modeling.

Ground-Penetrating Radar Data

Three 3-D common-offset digital GPR data sets wererecorded using antenna frequencies of 50, 100, and

200 MHz. Interpretation was performed mainly on the100 MHz data to obtain a good compromise of verticalresolution (0.5 m) and depth of penetration (15m). The 200 MHz GPR data set, which has higher res-olution (0.3 m) but shallower penetration (10 m),was used only for detailed interpretation of the upper5 m of the uvial sandstone. The 50 MHz data (1 mvertical resolution and 20 m depth of penetration)were too coarse to be of use at the scale of interest.

The 3-D GPR survey at Coyote basin was per-formed on a rectangular grid of 34 approximatelynorth-southoriented lines (azimuth 350) at a spacingof 0.5 m between adjacent lines (Figure 3). Each GPRline contains 81 traces, equally spaced at 0.5 m. TheGPR equipment used in the survey was a PulseEKKOIV system with a transmitter voltage of 1000 V.Dipoleantennas were oriented parallel with each other andperpendicular to the in-line direction. A commonmid-point (CMP) gather, covering an offset range of 26 m,was recorded for each data set. The CMPs providedinitial velocity control and helped optimize the source-receiver offset for the 3-D data acquisition. The offsetsused were 3 m at 50 and 100 MHz and 2 m at 200MHz. Vertical and crosshole GPR surveys were alsorecorded at 100 MHz using boreholes A, C, and D(Figure 3); the results of analysis of these data, thepetrophysical data, and the ow modeling will be re-ported elsewhere.

GEOSTATIST ICAL METHODOLOGY

Geostatistics is used to estimate the spatial variabilityof different geologic and GPR parameters, based on theassumption that properties in the earth are not ran-dom, but have spatial continuity and are correlatedover some distance. Variogram modeling has been suc-cessfully used by Rea and Knight (1998) to quantifythe correlation length of radar reections to character-ize heterogeneities of the subsurface in two dimen-sions. The main assumption is that there exists a linkbetween the lithology of layers and their electricalproperties, and thus a relationship between the corre-lation structure of radar reections and lithology. Thisspatial relationship is expressed through standardvariograms (Rea and Knight, 1998).

An essential assumption in the calculation of thevariograms is that the data are stationary in space,which means that any subset of the data has the samestatistics as any other subset. For GPR data, the sta-tionarity requirement is not satised because of the

1590Ferron

SandstoneInternalArchitecture

Ripple Cross-Laminated Siltstone

MudstoneMudstone-Intraclast Conglomerate

Massive & Parallel-Laminated Fine-Grained Sandstone

Trough Cross-Bedded Medium-Grained Sandstone

C

10 meters

SOUTH

B

NORTH

3

0

0

1

5

0

0

PERMEABILITY(md)

GAMMA RAY(Total Count)

400

800600

CB4

3

0

0

2

0

0

1

0

0

D

F

3

0

0

2

0

0

1

0

0

3

0

0

2

0

0

1

0

0

CB2

UNIT 3

UNIT 2

CB5

3

0

0

2

0

0

1

0

0

UNIT 5

3

0

0

2

0

0

1

0

0

400600800

UNIT 4

400600800

400600800

400600800

400600800

E

1 meter

UNIT 1A

CB3 CB1

Figure 4. Sedimentary facies map of the cliff face at Coyote basin. Higher-order bounding surfaces (A through E, in red) outline major architectural elements (units 1 through5). Surface F is the topographic surface. Less-signicant, lower-order bounding surfaces are in black. Exposed surfaces are shown as solid lines; dashed lines are inferred whereoutcrop is covered. Also shown are ve measured stratigraphic sections (CB1 through CB5) in which primary sedimentary structures, textural information, permeability, andgamma-ray data were recorded. The position of the outcrop relative to the 3-D GPR grid is shown in Figure 3.


Figure 5. Paleocurrent mea-surements for the upper surfaceof the uvial sandstone. Thepaleocurrent rose diagrams areexclusively for the uppermostunit, unit 5, of the channelcomplex and represent ow di-rection inferred from medium-scaled trough cross-beds. Thegeneral progradation directionof the parasequence set 3, inthe upper part of FS2 sequenceof the upper Ferron Sandstonedelta complex (Garrison et al.,1997) is illustrated using theheavy arrow. The orientation ofthe 211 cross-beds in relationto parasequence set 3 paleo-geography is explained as a lo-cal phenomenon of the radialsediment dispersal pattern indelta systems. The trough cross-bedded sandstone in unit 5 isinterpreted as a channel bar.The GPR survey site is locatedin an upstream position on thebar. A possible areal extent ofthe barform (the stippled re-gion) and the correspondingcross sectional geometry andinternal reectors (cross-bedcosets), illustrated below, areschematically extended fromoutcrop facies maps.

strong decay of the amplitude down a radar trace dueto radar signal attenuation, and also by changes in radarfacies both vertically and laterally (Rea and Knight,1998). To compensate for radar signal attenuation anautomatic gain control (AGC) with a window lengthof 2.5 m was applied to each GPR trace after migra-tion. Between proles, the GPR amplitudes were nor-malized relative to the maximum amplitude value inthe survey. To account for changes in radar facies, themigrated GPR volume was subdivided into four units(referred to as units 2 to 5) dened by specic radarfacies, prior to the variogram computations.

The experimental variograms were computed forthe 3-D GPR relative amplitude data within eachGPRfacies using the equation (Deutsch and Journel, 1998)

2(x y )i ic(h) (1)2N(h)

where h is the separation distance between two datapoints (the lag), N(h) is the number of pairs of datapoints separated by h, xi is the data value at one of thepoints of the ith pair, and yi is the corresponding datavalue at the second point. Equation 1 can be appliedto 1-D, 2-D, or 3-D data sets. For 3-D data, the sep-aration vector h is specied together with its directiondened by three angles, azimuth, dip, and plunge(Deutsch and Journel, 1998).

In most geologic data sets, the data values alongcertain directions are more coherent than along others.The direction with best continuity represents the max-imum correlation direction of the data set. Theminimum correlation direction is perpendicular to themaximum correlation direction. The ratio betweenminimum and maximum correlation lengths is theanisotropy factor (Isaaks and Srivastava, 1989). Com-monly, variograms are presented as 1-D curves alonga particular direction. A more global view of the


variogram values in all directions is achieved by com-puting variogram volumes. A variogram volume is a3-D plot of the sample variogram c(h) computed in alldirections for all available separation vectors h (hx,hy,hz). The lowest values of c(h) generally form anellipsoid centered at the value c(o) 0, which is alsothe symmetry center (Deutsch and Journel, 1998).Variogram volumes are used to determine the orien-tation and dip of vector h for which data sets show bestspatial continuity. Directions and amount of aniso-tropy are given by the orientation of the major andminor axes of the ellipsoid. The major axis is coinci-dent with the maximum correlation direction.

DATA PROCESSING AND ANALYSIS

Context: Hierarchy of Bounding Surfaces

Miall (1985, 1988) emphasized the importance ofidentifying and correlating bounding surfaces at variousscales rather than simply documenting vertical faciestransitions to clearly understand the complexities ofuvial depositional systems. He developed a sixfoldhierarchy of bounding surfaces. First-order surfacesseparate similar sedimentary features such as cross-bedset bounding surfaces (Allen, 1983; Miall, 1985).Second-order bounding surfaces outline cosets of ge-netically related facies without signicant evidence oferosion, but with dissimilar lithofacies above and be-low the surface (Miall, 1985). Third- and fourth-orderbounding surfaces envelop larger-scale architecturalelements constituting facies associations (Miall, 1985;Soegaard, 1991). A third-order bounding surface en-velops any architectural element with uniform com-position of facies or facies sequences such as a bar orchannel element (Soegaard, 1991). Fourth-order sur-faces envelop a complex of stacked architectural ele-ments composed internally of similar facies sequencessuch as composite bars (Soegaard, 1991). Fifth-orderbounding surfaces outline larger depositional systemscomposed of diverse but related architectural ele-ments. They are marked by erosion and local cut-and-ll relief and basal gravel lags (Miall, 1985). Sixth-order surfaces separate depositional sequences whosedistribution is generally dictated by allogenic effects.Fifth- and sixth-order surfaces can be mapped usinghigh-resolution 3-D seismic data (Miall, 1988; Thomasand Anderson, 1994). Lower-order surfaces, generallyobserved only at the outcrop, can be imaged usingGPRtechnology.

GPR Facies

Interfaces that generate GPR reections can includebedding planes, fracture planes, or any other boundaryseparating rock types with different electrical proper-ties. Electrical properties of a rock correlate mainlywith lithologic composition (sand/clay ratio, grain size,sorting, etc.) and water saturation (Knight and Nur,1987; Annan et al., 1991). Generally, saturation is ameasure of permeability and porosity of rocks, whichin turn, are generally consistent with lithology (Rea andKnight, 1998).

Identication of bounding surfaces using GPR re-ections is based not only on the contrast in electricalproperties above and below surfaces that produce sig-nicant reection amplitudes, but also on the hierarchyof reection terminations, reection continuity, andgeometrical congurations above and below the sur-face (see the radar facies of Gawthorpe et al. [1993]).First-order surfaces separate similar lithofacies belowand above the surface and present a contrast in elec-trical properties only if there is a change in petrophys-ical properties (e.g., permeability, porosity) across thesurface (e.g., due to a change in grain size). In this casea rst-order surface correlates with a single continuousGPR reection that truncates against higher-orderbounding surfaces. If no contrast is present, the posi-tion of the bounding surface does not correspond to areector and must be inferred from the attributes pre-viously listed. Second-order surfaces separate differentlithofacies above and below and are thus more likelyto have disparate electrical properties. Therefore, asecond-order bounding surface is almost everywhererepresented by a continuous GPR reection (see Gaw-thorpe et al., 1993). First- and second-order surfacesare generally several decimeters to several meters inlength (Miall, 1985, 1988). Third- and fourth-ordersurfaces should give rise to continuous GPR reectionswhere different electrical properties are encounteredabove and below the surface but commonly are denedby characteristic reection terminations (truncation,onlap or downlap) against a surface, and also by theexistence of different radar facies (specic patterns ofreection continuity, conguration, amplitude, andfrequency) above and below the surface (see Gaw-thorpe et al., 1993; Alexander et al., 1994; Bridge etal., 1998). Third- and fourth-order surfaces are gen-erally several tens of meters in length (Miall, 1985,1988). Fifth-order surfaces are represented by contin-uous, through-going reections where they are char-acterized by sharp contacts but may be more complex


or completely obscured where gradational contacts oc-cur. Fifth-order surfaces clearly separate different ra-dar sequences (see Gawthorpe et al., 1993). No sixth-order surfaces are present in the study volume.

GPR Data Preprocessing

Several processing steps were applied to the 3-D GPRdata before depth migration. Preparation and pre-processing of the GPR data consisted of trace editing,time-zero corrections, air-wave removal (to reducenear-surface interference), bandpass lter analysis (todiscriminate high-frequency events associated withsmall sedimentary structures from the high-amplitudeenergy near the median signal frequency), gain analy-sis, and predictive deconvolution. Detailed informa-tion on this processing was given by Szerbiak et al. (inpress).

The most important step in processing GPR datais 3-D depth migration, which allows direct and ac-curate comparison (in 3-D space) between geologicdata and radar data, especially where velocity variessignicantly in three dimensions (Szerbiak et al., inpress). An initial migration, using a single average in-terval velocity function, produced a poorly migratedGPR image and also poor ties with the borehole depthcontrol points. These poor results are explained by sig-nicant lateral variation in velocity that is produced,not only by the spatial variation of lithologic facies, butalso by the fracture systems at the site. The fracturesystems are oriented northwest-southeast and north-east-southwest (Figure 3) and inuence the amountand pattern of weathering in each block bounded bythe fractures. Depending on the amount and type ofweathering, different parts of the same lithologic unitcan have different electrical properties, which in turn,may substantially change the GPR propagation veloc-ity. Also, the permeability values at the outcrop aredetermined to be signicantly higher than permeabilityvalues in well cores because of increased weathering atthe cliff face (Snelgrove et al., 1998). Because surfacewaters have been moving along the fractures in recenttimes, the rock adjacent to the fractures has likely beenexposed to weathering in a way similar to the rock atthe present cliff face.

Velocity Model Building and MigrationThe 3-D velocity model was obtained in two steps:(1) obtaining vertical velocity proles at control points,and (2) spatially interpolating between these velocitiesby kriging (Deutsch and Journel, 1998).

In the rst step, synthetic GPR traces are simu-lated to estimate vertical velocity functions at the fourwells and the ve stratigraphic sections. Bounding sur-face depths and two-way reection traveltimes wereavailable at the wells, but only depths were availableat the cliff face stratigraphic sections. Reection traveltimes were estimated at the cliff face by extrapolationof the two-way traveltime surface observed in the 3-Dgrid to the cliff face. Each 1-D model was parameter-ized by electrical properties based on correlation oflithology and permeability, and lab measurements ofdielectric permittivity (which is the main determinantof the velocity of the GPR wave) and electrical resis-tivity (which is the main determinant of GPR signalattenuation). The nite-difference modeling algo-rithm used was described in detail by Xu and Mc-Mechan (1997). Figure 6 shows the simulated GPRresponse at well A; the synthetic radargrams havematched the reection amplitude, polarity, and fre-quency content of the main events in the 3-D data.This modeling procedure yields a robust velocity es-timate in depth at the control sections in the surveyand also provides a direct correlation of major bound-ing surfaces identied at the cliff face and in bore-holes, with reections in GPR proles in the time do-main (Figure 6).

The second step in building the 3-D velocitymodel consisted of spatially interpolating and extrap-olating the vertical velocity proles obtained by mod-eling. The interpolation procedure was based onbuilding 3-D experimental variograms from two mainaverage velocity facies (one facies above surface E andthe other one between surfaces E and A/B in Figure6), and then simulating 2-D velocity surfaces at reg-ular depth intervals from the vertical velocity controlproles. The complete procedure of building a smooth3-D velocity model based on geologic control and geo-statistical techniques was discussed by Szerbiak et al.(in press).

A 3-D Kirchhoff algorithm (Epili and McMechan,1996) was used to migrate the GPR data into a depthimage. Depth migration provided high-resolution im-ages of the sedimentary features and also relative am-plitude data for the geostatistical correlation analysis.Because 3-D GPR data volumes have a format similarto that of 3-D seismic data, 3-D seismic interpretationsoftware provides a exible and efcient means fordisplay, attribute computation, and analysis of the3-D GPR data. Figure 7 shows representative slicesthrough the 3-D GPR data volume, without and withinterpretive labels.


Figure 6. Input and output of the synthetic radargram modeling at well A. Panel (A) shows details 1 m from the core emphasizingthe correlation between lithology and permeability. Panel (B) shows the lithofacies model and the permeability prole on which thesynthetic radargram was built, together with the interval velocity prole resulting from the radargram modeling. Panel (C) containsthe synthetic radargram for well A (in the middle) and ve traces from the 3-D GPR volume adjacent to well A (on either side). E andA/B are two major bounding surfaces interpreted in outcrop and boreholes and identied using GPR reections in time proles. Thesetwo surfaces provide ties that control the average velocity facies from which velocity correlation functions were obtained.

Geostatistical Analysis

Figure 8A shows three orthogonal slices from the vario-gram volume of the GPR relative amplitudes from theuppermost interpreted unit (unit 5 in Figure 7) of themigrated GPR volume. The azimuth, dip, and plungeof the maximum correlation direction (the longest axesof the ellipsoid) can be computed from the projectionsonto the three orthogonal planes of the variogram vol-ume (the red arrows in Figure 8A). The dip angles ex-tracted from the variogram volumes for each unit werealso compared with the dip of GPR reections fromthe migrated GPR volume. The resulting parametersfor each unit are presented in Table 1.

The experimental variograms were computedalong the maximum and minimum correlation direc-tions for each radar facies identied in theGPR volume(units 2 to 5). Satisfactory tting of the experimentalvariograms commonly requires use of nested struc-tures containing a linear combination of two basicmodels, rather than a single model (Isaaks and Srivas-tava, 1989). Each model in the nested structure pro-vides different contributions to the nal compositemodel. The tting was done by iterative manual trialsuntil the best nested structure tting was obtained. Fig-ure 8B shows an example of an experimental vario-gram and the nested model tted to it, from the upper-most unit of the GPR volume. The results frommodeling the experimental variograms in each unit arealso given in Table 1 and are interpreted in the follow-ing section.

INTEGRATED INTERPRETATION OFSEDIMENTOLOGIC AND GPR DATA

Five architectural elements are identied in the out-crop at Coyote basin and referred to as units 1 to 5 inascending stratigraphic order. Five bounding surfacesseparate these units and are referred to as surfaces Ato E, also in ascending order (Figure 4). The same unitsand bounding surfaces are mapped in the GPRmigrated-data volume, except for units 1 and 2, which

are grouped together and referred to as unit 2. SurfacesA and B at the outcrop are also mapped together inthe GPR data volume and referred to as surface A/B(Figure 7). The contour maps with the depths of thefour bounding surfaces that resulted from the inter-pretation of the GPR volume are presented in Figure 9.

Fifth-Order Bounding Surface: A/B

The sharp erosional contact between the underlyingextensive, thick mudstone and the overlying 12 mthick sandstone is interpreted as a fth-order boundingsurface that separates the uvial ood-plain mudstonefrom the channel sandstone and is referred to as A/Bin outcrop and in the GPR interpretation (Figures 4,7). The sandstone above this surface and the mudstonebelow it have very different electrical properties so thatthe GPR reection at the boundary should be strongand continuous. Surface A/B is dened locally bymud-stone intraclast conglomerate and small-scale scour-and-ll relief. The mudstone intraclast conglomerateand associated siltstone deposits have average electricalproperties between the sandstone and mudstone endmembers, causing the A/B surface to be less sharplydened by the radar signal. Depending on the thicknessand complexity of the transition zone, surface A/Bproduces locally dispersed reections with reducedamplitudes analogous to the transition that occurs at awater table (Annan et al., 1991).

Tracking the continuous, strong GPR reection(the dashed red line on the interpreted prole inthe lower panel of Figure 7, which correlates with theA/B surface in wells A, C, and D) toward the northernpart of the survey, there is an apparent mistie aroundwell B. The strong GPR event correlates with themud-stone intraclast conglomerate layer (B) at about 14 min well B, rather than with the A surface (top of ood-plain mudstone). Around well B and measured sectionCB1, surfaces A and B delineate a local scour-and-llelement, which in outcrop was originally interpretedas unit 1 (Figure 4) and in the GPR interpretation laybetween the two red lines (A and B in Figures 7,10A). The presence of the conglomerate obscures the

1596Ferron


10 meters

13.1

0

6.66

-6.66

-13.1x 1000

0

2

4

6

8

10

12

14

16

0

2

4

6

8

10

12

14

16

A C D B

0

2

4

6

8

10

12

14

16

0

2

4

6

8

10

12

14

16

D

e

p

t

h

(

m

)

Depth (m)

Depth (m)

D

e

p

t

h

(

m

)

Relativeamplitude

S N

Unit 5

Unit 4

Unit 3

Unit 2

A

B

E

D

C

A/B

Well Well Well Well

A C D BWell Well Well Well

Figure 7. Uninterpreted (upper panel) and interpreted (lower panel) GPR proles from the migrated 3-D 100 MHz volume, connecting wells A, C, D, and B. Lithologic columnsand permeability proles from each well are shown for correlation with the GPR reectors. Colored lines in the lower panel show the interpreted bounding surfaces (A/B to E);red arrows below interpreted surfaces C to E show the truncation of the GPR reections against the third- and fourth-order erosional surfaces. The dashed rectangle in the lowerright corner shows the location of the area analyzed using instantaneous frequency in Figure 9. The dashed red line marks the continuous, strong GPR event tracked from wellsA, C, and D, which correlates with the top of unit 1 (surface B) and obscures below the reection corresponding to the upper bounding surface of the ood-plain mudstone(surface A). The black arrow in the upper panel marks the reduced amplitude reection correlated with surface A.

0.80

0.40

0.00

1.20

1.60

(h)(A)

N

0.01.0

-1.0

2.0

-2.0

3.0

-3.0

4.0

-4.0

0.0-1.0

1.0

2.0

-2.0

3.0

-3.0

0.00.2

-0.2-0.4

X

Y

Z(XY)

(XZ)

(YZ)

0.00

0.20

0.40

0.60

0.80

1.00

1.20

(h)

h (m)0.0 5.0 10.0 15.0 20.0 0.0 4.0 8.0 12.0 16.0

h (m)

(B)

direction of maximum correlation direction of minimum correlation

N

Nested structure = spherical + exponential Nested structure = gaussian + exponential

hmax = 5.75 m hmin = 3.4 m

Figure 8. Example of variogram analysis in unit 5. (A) three orthogonal slices through the center of symmetry of the variogramvolume displaying the variogram values computed along all directions and for all available separation lags. The direction of themaximum correlation of the GPR amplitudes projects on the three slices along the longest axis of the central dark blue ellipses (redarrows) dened by the lowest values of c as a function of the separation vector (blue colors on the color bar). These projectionsindicate the azimuth, dip, and plunge of the direction of maximum correlation. Azimuth is measured in the horizontal symmetry planeclockwise from the y axis; dip and plunge are measured in the vertical symmetry planes clockwise toward the z axis (Deutsch andJournel, 1998). The parameters inferred from the variogram volume analysis are given in Table 1. (B) Experimental variograms alongdirections of maximum and minimum correlation of the GPR amplitudes in the uppermost unit interpreted in the GPR volume. Thered squares are data points of the experimental variograms, whereas the green continuous lines are the nested model tted to eachvariogram; the results of the variogram analysis are presented in Table 1. For denitions of symbols used in the gure see equation1 in the text.



Table 1. Semivariogram Analysis: Parameters and Results

Unit Facies Correlation Direction* Model Range Nugget Sill Anisotropy Factor

Unit 5 Trough cross-bed Maximum Azimuth 90Dip 7

Spherical exponential 5.7515.00

0.00.0

0.700.35

0.59

Minimum Azimuth 0Dip 0

Gaussian exponential 3.408.00

0.00.0

0.800.30

0.53

Unit 4 Scour and ll Maximum Azimuth 90Dip 0


0.00.0

0.700.30

0.6



0.00.0

0.650.45

0.8



0.00.0

0.700.35

0.75



0.00.0

0.650.32

0.53



0.00.0

0.650.35

0.95



0.00.0

0.800.40

0.67

*By convention the azimuth is measured clockwise from the y axis, whereas dip is measured clockwise toward the z axis (Deutsch and Journel, 1998).

GPR reection from surface A, and where unit 1pinches out against surface A and becomes thinnerthan one-quarter of the wavelength, differentiating be-tween the two surfaces A and B becomes more difcultbecause of tuning effects. Displaying radar data withother attributes, such as instantaneous frequency, clar-ies the position of the A/B surface at well B andthroughout the northern part of the survey. The in-stantaneous frequency attribute is the time derivativeof the instantaneous phase and represents a measure ofthe frequency of the waveform at every sample. Lateralheterogeneity, including pinch-outs or abrupt changesin lithofacies, tends to change the instantaneous fre-quency more rapidly. If this is the case, then the GPRreection of surface A/B around well B is not a mistiebut the product of a composite reection due to abruptlateral changes in facies not resolved by the 100 MHzGPR. Figure 10 shows a comparison of the instanta-neous frequency attribute for the GPR data in two pro-les, one running through well B (see also Figure 7)and the other located farther eastward, nearer the cliffface. The dashed line in Figure 10A delineates thestrong continuous GPR event B (interpreted on therelative amplitude display in Figure 7 as correspondingto the mudstone intraclast conglomerate at 14m depthin well B), and the attenuated reection from the topbounding surface of ood-plain mudstone is shown bysolid line A. In Figure 10B, the two GPR events cor-responding to surfaces A and B become coincident, as

unit 1 pinches out or is thinner than the vertical reso-lution and is no longer resolved by the 100 MHz GPRdata.

The contour map with the depths of surface A/B(Figure 9) shows a general dip of the surface towardthe northwest and an erosional depression in the north-ern part of the survey, more accentuated around wellB where the scour-and-ll element 1 has its maximumthickness.

Fine-Grained, Parallel-Laminated Sandstone FaciesAssociation: Units 1 to 4

Sedimentologic DescriptionUnits 1 to 4 cover approximately the lower 7 m ofthe channel complex and consist of ne-grained len-ticular sandstone bodies that pinch out over distancesof several tens of meters parallel to the cliff face (Fig-ure 4). Internally, these architectural elements consistof low-angle, parallel-laminated, ne-grained sand-stone that scour into underlying, similar parallel-laminated sandstone. The base of each of unit (1 to4) is erosional and commonly has mudstone intraclastconglomerate along the basal scour. Locally, the ero-sional scours can have a steep cut relief of almost 1m lled with mudstone intraclast conglomerate (seeFigure 4 near section CB1 at depths of 9 and 12 m)resulting in abrupt lateral changes in thickness ofconglomerate layers. The upper part of units 1 to 4

Corbeanuet

al.1599

Figure 9. Depth contour maps of the four surfaces (A/B to E) that bound the major architectural elements in the uvial sandstone at Coyote basin; depths are in meters, andthe depth contour increment is 0.25 m. These contours maps are generated from the 100 MHz migrated GPR data and are relative to the GPR horizontal datum. A to D arelocations of the wells inside the GPR grid. Notice the abrupt erosional depression around well B on surface A/B, the relatively at character of surfaces C and D, and the erosionalscour oriented parallel with the paleoow on surface E.


111.00

72.00

33.10

5.88

150.00

Instantaneous F

requency (M

Hz)

X (m)

X (m)20.0 30.0 40.035.025.0

20.0 30.0 40.035.025.0

Depth (m)De

pth (m

)

7.5

10.0

12.5

15.0

17.5

7.5

10.0

12.5

15.0

17.5

Depth (m

)

7.5

10.0

12.5

15.0

17.5

7.5

10.0

12.5

15.0

17.5

Depth (m)

B

AB

A/B

S N

S N(B)

(A)

Figure 10. Instantaneous frequency displays of the northern half and lower 10 m of two GPR proles. (A) The GPR prole throughwell B at y 12.0 m; (B) the prole at y 7.0 m. The continuous line in (A) is the interpretation of the A/B surface revealed asa composite reection due to the gradational character of the contact; the dashed line marks the continuous GPR reector, which isinterpreted as the top of unit 1 (see also Figure 6). In panel (B) the two continuous lines show complete coincidence as apparentlyunit 1 is pinched out or thins beyond the vertical resolution of the GPR data.

are capped by mudstone layers, generally 510 cmthick, which are also laterally discontinuous becauseof truncation by the overlying unit (Figure 4). Per-meabilities measured in units 1 to 4 are on the orderof tens of millidarcys with very low values (a fewmillidarcys) in the mudstone and mudstone intraclastconglomerate intervals. The lowest average perme-ability is measured in unit 4, but the highest disper-

sion about the mean value is observed in units 2 and3 (see Figures 4, 7; Table 2).

Units 1 to 4 are interpreted as scour-and-ll ele-ments deposited during ood events within a uvialchannel. Because units 1 to 4 are covered beyond theextent of the 45 m survey area, the larger-scale sedi-mentologic architecture is not revealed in outcrop forthe lower part of the channel complex.


Table 2. Characterization of Major Units of Fluvial Channel by Means of Lithofacies, Permeability Values, and Range, and theCorresponding Radar Facies

~ 1 to 270 md

~ 1 to 80 md

~ 1 to 90 md

~ 1 to 80 md

Mean ~50 mdStd ~40 md




Sedimentologic Description

Unit 5Medium- to large-scale, trough cross-

bedded, medium-grained sandstone

Units

Unit 4

Unit 3

Unit 2

Low-angle, parallel-laminated,

fine-grained lenticular sandstone

capped by mudstone layers and

with mudstone intraclast

conglomerates on the basal scours

Permeability Statistics Radar Facies

GPR InterpretationGPR reections in approximately the lower 7 m ofthe data volume correlate well with second-orderbounding surfaces between sandstone layers andmudstone or mudstone intraclast conglomerate lay-ers because these correspond to a signicant changein electrical properties between the three lithologies.The irregularity in thickness and shape of thesemudstone and conglomerate layers is evident in theGPR images as discontinuous, irregular reections(Figures 7, 11). Many layers that are signicantlythinner than 0.5 m are not resolved using the 100MHz GPR data (see Figure 7 at 10.5 m deptharound wells C and D).

Third-order bounding surfaces C and D, inter-preted in the GPR proles, are continuous surfacesdened by downlap or truncation of second-orderreections above and below the third-order surfaces,respectively (Figures 7, 11). Both surfaces C and Ddip gently toward the north, following the regional

structural dip, and have limited erosional relief (Fig-ure 9).

A cube view of the 3-D GPR data shows, on ahorizontal amplitude slice cut at constant depththrough unit 3 (Figure 11), high-amplitude zonescorrelated with mudstone and mudstone intraclastconglomerate layers striking approximately north-south and dipping slightly toward the east.

The mudstone and conglomerate layers insideunits 2 to 4 could affect uid ow if these layersare continuous. Commonly, the mudstone and con-glomerate layers are laterally discontinuous in out-crop, and GPR reectors display the same pattern.The GPR radar facies identied in units 2 to 4 con-tain subparallel, discontinuous GPR reections (Fig-ures 7, 11; Table 1). Units 2 to 4 are characterizedby generally similar radar facies in terms of conti-nuity and conguration of reections, with more dis-continuous GPR reections in units 2 and 3 relatedto higher variability in permeability values (Table 2).

1602Ferron


0.0

10.0

20.0

30.0

40.0

15.0

10.0

5.0

15.0

10.0

5.0

X (m)

Y (m)

D

e

p

t

h

(

m

)

Unit 5

Unit 4

Unit 3

Unit 2

E

D

C

A/B 0.0

0.5

-0.5

-1.0

1.0

Relativeamplitude

S

N

Figure 11. Cube display of the 3-D GPR data, made of two lines at y 1.5 m and y 10.5 m, two crosslines at x 18 m and x 40 m, and two horizontal slices at z 4 m and z 9.5 m. The x and y axes coincide with the long and short axes of the GPR grid (Figure 3). Red, blue, orange, and green labels on the left side of the cube markthe interpreted A/B, C, D, and E bounding surfaces, respectively. Inside the vertical GPR proles, the purple arrows mark downlap, onlap, and truncation of the GPR reectionsagainst the major bounding surfaces. The relation between high-GPR-amplitude zones on the horizontal slice and the inclined reections on the vertical proles in unit 5 isillustrated using thin black lines portraying the climbing cross-beds in the vertical plane and their shape on the surface. In unit 5, the black arrows show paleoow direction, andin unit 3 they show the dip direction of the mudstone and mudstone intraclast conglomerate layers.


Geostatistical InterpretationTo quantify the lateral extent of mudstone and con-glomerate layers, from the continuity of the corre-sponding GPR reections, experimental variograms arecomputed for each unit from the GPR relative ampli-tude data, along both maximum and minimum corre-lation directions. The maximum correlation directionsof the GPR amplitude data coincide with the long sideof the GPR grid in all units (Table 1). The data in theexperimental variogram are tted with a nested struc-ture composed of two basic models: Gaussian andexponential.

The correlation lengths (or ranges) of the Gaussiancontribution range from 4 to 5 m in the maximumcorrelation direction and from 3 to 4 m in the mini-mum correlation direction (Table 1). These correlationlengths are interpreted as characterizing the lateralcontinuity of themudstone ormudstone intraclast con-glomerate layers with thicknesses comparable to ver-tical resolution of the GPR (0.5 m), and envelopedby second-order bounding surfaces, inside each unit.The anisotropy factors of these short-wavelength struc-tures are 0.95, 0.75, and 0.6, respectively, for units 2,3, and 4. These anisotropies imply that mudstone andmudstone intraclast conglomerate inside the channellls have more elongated shapes toward the upper partof the channel (unit 4) and more isometric shapes atthe base of the channel (unit 2), but all have a maxi-mum lateral extent of 5m. These results compare fairlywell with the facies map at the cliff face, especially themudstone intraclast conglomerates in units 2 and 3.Where making direct comparisons of the mudstoneand mudstone intraclast conglomerate layers with theGPR reections, one should consider the limitation ofthe 100 MHz GPR data on resolving features signi-cantly thinner than about 0.5 m. Sometimes mudstonelayers are interpreted in the outcrop to be laterally con-tinuous over more than 10 m (e.g., at the base of unit4 and the top of unit 2 in the southern part of theoutcrop in Figure 4) but are relatively thin and irregularin thickness and may not be well resolved by the GPRreections. These layers are described by longer cor-relation lengths (the exponential model in the nestedstructure), but they have a smaller contribution to thecombined model (Table 1).

Medium-Grained Trough Cross-Bedded Sandstone FaciesAssociation: Unit 5

Sedimentologic DescriptionUnit 5 (the uppermost 4.55.5 m of the sandstone

complex) consists exclusively of medium- to large-scale, trough cross-bedded, medium-grained sandstone(Figure 4). Permeabilities in unit 5 are in the range offew hundreds of millidarcys with high dispersion aboutthe mean (Table 2).

This trough cross-bedded unit is lenticular in ge-ometry with a relatively at (erosional) base and a con-vex upper surface (Figure 5). The base of unit 5 is de-ned by surface E both in outcrop (Figure 4) and inthe interpreted GPR volume (Figures 7, 11). Unit 5has been mapped outside the GPR survey area and ex-tends about 640 m to the south in a downcurrent di-rection before pinching out (Figures 3, 5). The up-current (northward) extent of unit 5 is not determinedbecause of the lack of a cliff face exposure, but the unitis present at least 30 m outside the 3-D GPR grid,based on information from a 2-D GPR prole extend-ing toward the north beyond the 3-D grid. Troughcross-beds in the upcurrent position are clearly climb-ing, with rst-order coset bounding surfaces truncatingagainst the fourth-order E surface in an upcurrent di-rection (Figures 4, 5). In the downcurrent part of unit5, south of the survey area, trough cross-bed coset sur-faces truncate against the lower E surface in the down-current direction (Figure 5). The thickness of thetrough cross-beds in the lower half of unit 5 is 1030cm, with a signicant proportion of the cross-beds be-ing preserved. In the upper half of unit 5, trough cross-bed sets tend to be less than 10 cm thick (Figure 4).The smaller cross-bed sets are either a result of smalleroriginal bedforms on the upper bar surface or due toscouring by overlying cross-beds (Figure 4).

On the upper surface of the survey site, troughcross-beds are up to 1.5 m wide and extend in a down-current direction for a distance of up to 7 m (Figure3). Along the cliff face, similar lateral extents of severalmeters are seen for individual cross-bed sets (Figures4, 12). The paleocurrent measurements from the up-per surface of unit 5 (Figure 5) show a more east-southeastern (115 to 150 azimuth) paleoow for thedistributary channels at Coyote basin than the generaleast-northeastern (075 azimuth) progradational direc-tion of delta lobes forming parasequence set 3 (Garri-son et al., 1997). This change in ow may be due toactive bifurcation as the main distributary channels ap-proach the coastline.

Unit 5 is interpreted as a channel barform. Cosetboundaries outline the geometry of the upper surfaceof the barform. The survey site at Coyote basin is inthe upcurrent part of the barform on the northwesternside of the channel bar based on dip orientations of


Figure 12. Detailed outcrop map of trough cross-beds on two orthogonal outcrop faces in unit 5 immediately southeast of the GPRsurvey area (Figure 3). The north-south panel shows the geometry of trough cross-beds parallel with ow direction, whereas the east-west panel is perpendicular to ow. Heavy lines mark the coset bounding surfaces that are most likely resolved by GPR data; theseare compared in the text with the maximum correlation lengths from the geostatistical analysis.

cross-bed cosets seen both at the outcrop and in theGPR data (Figure 5). The upward climb of cross-bedcosets in the upcurrent part of the barform implies thatsedimentation rates were high and that bar accretionoccurred both in an upcurrent and a downcurrent di-rection. More commonly, barforms tend to experienceerosion in an upcurrent direction and bar growth in adowncurrent direction (Bridge, 1986; Miall andTurner-Peterson, 1989). In these last instances, rst-order cosets truncate against the upper surface of bar-forms (i.e., fourth-order bounding surfaces) in an up-current direction.

GPR InterpretationThe base of unit 5 is a fourth-order bounding surface(E) separating a medium-grained trough cross-beddedsandstone with high permeabilities (hundreds of mil-lidarcys) from underlying ne-grained, parallel- toslightly obliquely laminated sandstone with low per-meabilities. Locally, discontinuousmudstone intraclastconglomerate lies immediately above surface E (Figure4). On GPR proles, surface E is dened by a changeof geometry from baselapping reections above totruncated reections below the surface, rather than asingle continuous reection (Figures 7, 11). This pat-tern in the GPR data is consistent with the truncationrelationships between bounding surfaces seen at theoutcrop. Based on interpretation of the GPR data, thegeometry of surface E has an erosional scour orientedapproximately north-south, with a northward dip (Fig-

ure 9). The orientation of this scour is also parallel withthe paleoow indicators at the site (Figure 5).

The internal conguration of radar facies insideunit 5 along proles is generally parallel with thepaleoow (see the GPR section between wells D andB in Figure 7 and the north-south faces of the datacube in Figure 11) and show continuous, slightlyoblique reections (Table 2). These reections are in-terpreted as rst-order bounding surfaces inside unit5. A horizontal amplitude slice cut at a constant depthof 4 m through unit 5 (the uppermost face of the GPRcube in Figure 11) shows high-amplitude zones cor-relating with rst-order cross-bed cosets strikingnortheast-southwest, perpendicular to the ow direc-tion as measured from the trough cross-beds at thesurface. These high-amplitude zones are a result of theintersection between the horizontal slice and the up-ward climb of trough cross-bed cosets to the southeast(Figure 11).

Migrated 200 MHz GPR data are useful for inter-preting detailed sedimentologic structures of about 0.3m thickness inside unit 5. The GPR prole transverseto the paleoow direction at the position x 31.5 m(Figure 3) from the 200 MHz migrated GPR datashows cross-bed cosets of medium scale interpreted inthe lower part of unit 5 (Figure 13). Upwardly concavediscontinuous reectors truncate against adjacent oroverlying reectors, thus mimicking the geometriesseen in the nested trough cross-beds in the facies map.The GPR reections in areas with thin trough cross-


beds outline cosets of several such cross-beds ratherthan individual trough bed sets. These reections arethe GPR expression of the rst-order boundingsurfaces.

Geostatistical InterpretationTo quantify the lateral extent of cosets of trough cross-beds bounded by rst-order surfaces, experimentalvariograms were computed along both the maximumand minimum correlation direction on GPR relativeamplitude data. The maximum correlation direction ofthe GPR amplitudes corresponds to the long side ofthe GPR grid. The data in the experimental variogramswere tted with a nested structure composed of twobasic models (spherical/Gaussian and exponential)(Table 1).

In the maximum correlation direction, the shorterrange (corresponding to the spherical model from thenested structure tted to the experimental variogram)is 5.75 m and represents the main contribution to thecombined model (Table 1). This correlation length isin good agreement with the length of trough cross-bedsets measured at the outcrop of up to 7 m. In the min-imum correlation direction, the range of 3.4 m (Table1) is almost twice the maximum width of the troughcross-bed sets measured in outcrop (up to 1.5 m). The100 MHz GPR data have a horizontal spacing betweentraces of about 0.5 m, so laterally and vertically stackedcross-bed cosets with dimensions less than a meter orso are not resolved, and a direct comparison with in-dividual sets is no longer possible (Figure 12).

The longer correlation lengths resulting from thenested structure tted to the experimental variogramsare 15 and 8 m, respectively, for the spherical andGaussian models (Table 1); these correlation lengthshave a smaller contribution to the combined nestedmodel and are interpreted as probably the net result ofthe lateral and vertical stacking of some of the cross-bed sets.

DISCUSSION AND CONCLUSION

The 3-DGPR data are used together with detailed sed-imentologic and stratigraphic information to analyzethe detailed 3-D architecture of a uvial channel res-ervoir analog in the Ferron Sandstone beneath a surfacearea of 40 16.5 m, at Coyote basin in east-centralUtah. The uvial channel at Coyote basin belongs tothe seaward-stepping parasequence sets and is straightor slightly sinuous.

The 100 MHz data are a good compromise be-tween vertical resolution (0.5 m) and depth of pene-tration (15 m) for the scale and detail studied at theoutcrop. The 200MHzGPR data have a better verticalresolution (0.3 m) but are not useful at depthsgreater than 910 m where the signal is strongly atten-uated. The bulk of our interpretation was carried outon migrated 100 MHz GPR data, and only our inter-pretation of the upper 5 m of the stratigraphic succes-sion used information from the migrated 200 MHzGPR 3-D images.

To effectively integrate geologic and GPR data,3-D migration of the GPR data from the time do-main into the depth domain was essential. A gooddepth migration was obtained only after constructinga detailed velocity model containing both verticaland lateral changes in electrical properties of therock volume surveyed. Synthetic radargrams weregenerated to estimate vertical velocity proles andto correlate key GPR reections in the time domainto geologic boundaries in the depth domain. Krigingwas used to interpolate the lateral distribution of thevelocity.

To identify and separate architectural elementsand bounding surfaces in outcrop and well cores, thesixfold hierarchy of bounding surfaces developed byMiall (1985) was used together with techniques forinterpreting stratigraphic sequences from seismic data.Five architectural elements, referred to as units 1through 5 in ascending stratigraphic order, and theirbounding surfaces, referred to as surfaces A throughE, were correlated in outcrop and well cores. Units 1through 4 are scour-and-ll elements deposited duringood events within a uvial channel, and unit 5 is achannel barform accreting in both upcurrent anddowncurrent directions. The same architectural ele-ments and bounding surfaces were interpreted in themigrated GPR data.

Radar facies characteristic to each element wereinterpreted based on the internal conguration andcontinuity of reections as well as reection termi-nation patterns against higher-order bounding sur-faces. First- and second-order surfaces generally cor-relate directly with the GPR reections. Where thecontact between two elements is gradational ratherthan sharp, the GPR expression is a composite reec-tion that can be resolved using information from ad-ditional attributes such as instantaneous frequency.Abrupt lateral changes in lithofacies (e.g., unit 1around well B in Figure 10) are effectively addressedthrough instantaneous frequency attribute analysis.

1606Ferron


E

Y (m) (m)0.016.5 0.016.5

1

m

e

t

e

r

(A) (B) (C)

EW E

EW

0

1.6

3.2

4.8

6.4

8.0

D

e

p

t

h

(

m

)

1 meter

W Y

Figure 13. Upper 7 m of the uninterpreted (A) and interpreted (B) versions of the migrated 200 MHz GPR prole, at x 31.5 m (Figure 3). Cross-bed sets and cosets can beinterpreted as upward-concave reections in the GPR data and are marked with continuous orange lines in (B). For comparison, (C) shows the cliff face map of trough cross-beds from unit 5, perpendicular to ow as illustrated in Figure 12, for comparison. The sketch of trough cross-beds appears distorted because of a two-time vertical exaggerationfor direct comparison with the GPR proles. The dashed lines at the top of (A) and (B) represent the topographic surface.


Because of their high variability in thickness andlateral extent, ow barriers inside uvial reservoirscannot be condently mapped in 100 MHz data setsover large areas away from geologic control points.A quantitative description of the distribution of owbarriers inside each unit is achieved by modeling3-D experimental variograms of GPR amplitude.The assumption is that GPR amplitude is an indirectfunction of changes in permeability and, ultimately,of existence of ow barriers. Correlation lengths ofthe nested model tted to variograms in unit 5 aresimilar to dimensions of trough cross-bed sets andcosets measured in outcrop. The nested models inunits 2 to 4 suggest that the channel at Coyote basincontains discontinuous and randomly distributedmudstone barriers and bafes, as is expected forstraight distributary channels in a progradationalparasequence set such as SC3 in the Ferron Sand-stone (Barton, 1994).

Detailed mapping of rst- and second-orderbounding surfaces needs additional information fromhigher-frequency (e.g., 200 MHz) GPR surveys. Thetradeoff is that higher frequencies provide higher res-olution (though higher bandwidth) at the expense ofa reduction in depth of penetration. First-order bound-ing surfaces from unit 5 in the upper 5m of the channelcomplex are successfully imaged in the migrated 200MHz data and compare well with rst-order boundingsurfaces mapped in outcrop.

The area studied at Coyote basin represents a smallfraction (40 16.5 15 m) of a uvial reservoir an-alog in which we show that depth-migrated GPR datafrom closely spaced 3-D grids can be successfully usedto image individual architectural elements and hetero-geneity at the scale of a single voxel cell in a reservoirow simulator. Szerbiak et al. (in press) have used per-meability measurements from cores and outcrop totransform the GPR data into the permeability domainthroughout the survey volume discussed herein. Al-though correlation lengths for the permeability struc-ture within the survey volume are far too short to bebuilt into a full-eld petrophysical model of uvial res-ervoirs, ow simulations performed by Snelgrove et al.(1998) do provide effective permeability in X, Y, andZ. These permeability tensors may be used directly inlarger ow simulation models with a coarser grid.Moreover, the ow simulations performed on the datavolume at Coyote basin indicate that 5% less oil is re-covered from the uvial reservoir if the detailed het-erogeneity described in this article is considered com-pared with a homogeneous model (Snelgrove et al.,

1998). This type of information will assist in bettervolumetric calculations and history matching for thelarger, more-crude ow simulation models needed forentire elds. In light of the efciency of GPR surveys(modest costs and acquisition time), larger grids can beemployed in the future to extend the detailed inter-pretation presented herein to larger volumes approach-ing the scale of the interwell spacing in actual hydro-carbon reservoirs (Corbeanu et al., 2000).

REFERENCES CITED

Alexander, J., J. S. Bridge, M. R. Leeder, R. E. L. Collier, and R. L.Gawthorpe, 1994, Holocene meander-belt evolution in an ac-tive extensional basin, southwestern Montana: Journal of Sed-imentary Research, v. B64, p. 542559.

Allen, J. R. L., 1983, Studies in uviatile sedimentation: bars, bar-complexes and sandstone sheets (low-sinuosity braidedstreams) in the Brownstones (L. Devonian) Welsh Borders:Sedimentary Geology, v. 33, p. 237293.

Annan, A. P., S. W. Cosway, and J. D. Redman, 1991, Water tabledetection with ground-penetrating radar (abs.): Society of Ex-ploration Geophysicists Annual International Meeting and Ex-position Program with Abstracts, p. 494497.

Baker, P. L., and U. Monash, 1991, Fluid, lithology, geometry, andpermeability information from ground-penetrating radar forsome petroleum industry applications: Society of PetroleumEngineers, paper SPE 22976, p. 277286.

Barton, M. D., 1994, Outcrop characterization of architecture andpermeability structure in uvial-deltaic sandstones, CretaceousFerron Sandstone, Utah: Ph.D. thesis, University of Texas atAustin, Austin, Texas, 260 p.

Beres, M., A. Green, P. Huggenberger, and H. Horstmeyer, 1995,Mapping the architecture of glaciouvial sediments with threedimensional georadar: Geology, v. 23, p. 10871090.

Bridge, J. S., 1986, Paleochannel pattern inferred from alluvial de-posits; a critical review: Journal of Sedimentary Petrology,v. 55, p. 579589.

Bridge, J. S., J. Alexander, R. E. L. Collier, R. L. Gawthorpe, and J.Jarvis, 1995, Ground-penetrating radar and coring used to studythe large-scale structure of point-bar deposits in three dimen-sions: Sedimentology, v. 42, p. 839852.

Bridge, J. S., R. E. L. Collier, and J. Alexander, 1998, Large-scalestructures of Calamus River deposits (Nebraska, U.S.A.) re-vealed using ground-penetrating radar: Sedimentology, v. 45,p. 977986.

Bristow, C. S., 1995, Internal geometry of ancient tidal bedformsrevealed using ground penetrating radar, inB.W. FlemmingandA. Bartholoma, eds., Tidal signatures in modern and ancientsediments: International Association of Sedimentologists Spe-cial Publication 24, p. 313328.

Brown, A. R., 1996, Interpretation of 3-dimensional seismic data:AAPG Memoir 42, 4th ed., 444 p.

Corbeanu, M. R., M. C. Wizevich, X. Zeng, G. A. McMechan, J. P.Bhattacharya, S. H. Snelgrove, and C. B. Forster, 2000, 3-Dinternal geometry of delta plain channel deposits, CretaceousFerron Sandstone, Utah (abs.): AAPG Annual ConventionAb-stracts with Programs, v. 9, p. A31.

Davis, J. L., and A. P. Annan, 1989, Ground penetrating radar forhigh resolution mapping of soil and rock stratigraphy: Geo-physical Prospecting, v. 37, p. 531551.


Deutsch, C. V., and A. G. Journel, 1998, GSLIB Geostatistical Soft-ware Library and Users Guide: Oxford, Oxford UniversityPress, 340 p.

Epili, D., and G. A. McMechan, 1996, Implementation of 3-D pre-stack Kirchhoff migration, with application to data from theOuachita frontal thrust zone: Geophysics, v. 61, p. 14001411.

Fisher, E., G. A. McMechan, and P. A. Annan, 1992a, Acquisitionand processing of wide-aperture ground-penetrating radar data:Geophysics, v. 57, p. 495504.

Fisher, E., G. A. McMechan, and P. A. Annan, 1992b, Examples ofreverse-timemigration of single-channel ground-penetratingra-dar: Geophysics, v. 57, p. 577586.

Flint, S. S. and D. Bryant, eds., 1993, The Geological modeling ofhydrocarbon reservoirs and outcrop analogs: International As-sociation of Sedimentologists Special Publication 15, 269 p.

Gardner, M. H., 1992, Sequence stratigraphy of the Ferron Sand-stone, east-central Utah, in N. Tyler, M. D. Barton, and R. S.Fisher, eds., Architecture and permeability structure of uvial-deltaic sandstones: a eld guide to selected outcrops of the Fer-ron Sandstone, east-central Utah: University of Texas at Austin,Bureau of Economic Geology Guidebook, p. 112.

Gardner, M. H., 1995, Tectonic and eustatic controls on the stratalarchitecture of mid-Cretaceous stratigraphic sequences, centralWestern Interior foreland basin of North America, in S. L. Do-robek and G. M. Ross, eds., Stratigraphic evolution of forelandbasins: SEPM Special Publication 52, p. 243281.

Garrison Jr., J. R., T. C. V. van den Bergh, C. E. F. Barker, and D. E.Tabet, 1997, Depositional sequence stratigraphy and architec-ture of the Cretaceous Ferron Sandstone: implications for coaland coalbed methane resourcesa eld excursion, in P. K. Linkand B. J. Kowallis, eds., Mesozoic to recent: Geological Societyof America Field Trip Guidebook for 1997 Annual Meeting,Brigham Young University Geological Studies, v. 42, part 2,p. 155202.

Gawthorpe, R. L., R. E. L Collier, J. Alexander, J. S. Bridge, andM. R. Leeder, 1993, Ground penetrating radar: application tosandbody geometry and heterogeneity studies, in C. P. Northand D. J. Prosser, eds., Characterization of uvial and aeolianreservoirs: Geological Society Special Publication 73, p. 421432.

Isaaks, E. H., and R. M. Srivastava, 1989, An introduction to appliedgeostatistics: Oxford, Oxford University Press, 561 p.

Knight, R. J., and A. Nur, 1987, The dielectric constant of sand-stones, 60kHz to 4MHz: Geophysics, v. 52, p. 644654.

McMechan, G. A., G. C. Gaynor, and R. B. Szerbiak, 1997, Use ofground-penetrating radar for 3-D sedimentological character-ization of clastic reservoir analogs: Geophysics, v. 62, p. 786796.

Miall, A. D., 1985, Architectural-element analysis: a newmethod offacies analysis applied to uvial deposits: Earth-Science Re-views, v. 22, p. 261308.

Miall, A. D., 1988, Reservoir heterogeneities in uvial sandstones:lessons from the outcrop studies: AAPG Bulletin, v. 72, p. 682697.

Miall, A. D., and C. E. Turner-Peterson, 1989, Variations in uvialstyle in the Westwater Canyon Member, Morrison Formation(Jurassic), San Juan basin, Colorado Plateau: Sedimentary Ge-ology, v. 63, p. 2160.

Miall, A. D., and N. Tyler, eds., 1991, The three-dimensional faciesarchitecture of terrigenous clastic sediments and its implicationfor hydrocarbon discovery and recovery: SEPM Concepts inSedimentology and Paleontology, v. 3, 309 p.

Posamentier, H.W., and P. R. Vail, 1988, Eustatic controls on clasticdeposition IIsequence and systems tract modeling, in C. K.Wilgus, B. S. Hastings, C. G. S. C. Kendall, H.W. Posamentier,C. A. Ross, and J. C. Van Wagoner, eds., Sea-level changes: anintegrated approach: SEPM Special Publication 42, p. 109154.

Rea, J., and R. Knight, 1998, Geostatistical analysis of ground-penetrating radar data: a means of describing spatial variationin the subsurface: Water Resources Research, v. 34, p. 329339.

Ryer, T. A., 1981, Deltaic coals of Ferron Sandstone Member ofMancos Shale: predictive model for Cretaceous coal-bearingstrata ofWestern Interior: AAPGBulletin, v. 65, p. 23232340.

Snelgrove, S. H., G. McMechan, R. B. Szerbiak, D. Wang, R. Cor-beanu, K. Soegaard, J. Thurmond, C. Forster, L. Crossey, andK. Roche, 1998, Integrated 3-D characterization and ow simu-lation studies in a uvial channel, Ferron Sandstone,east-centralUtah (abs.): AAPG Annual Convention Abstracts with Pro-grams, v. 2, p. A617.

Soegaard, K., 1991, Architectural elements of fan-delta complex inPennsylvanian Sandia formation, Taos trough, northern NewMexico, in A. D. Miall and N. Tyler, eds., The three-dimensional facies architecture of terrigenous clastic sedimentsand its implication for hydrocarbon discovery and recovery:SEPM Concepts in Sedimentology and Paleontology, v. 3,p. 217223.

Szerbiak, R. B., G. A. McMechan, R. M. Corbeanu, C. Forster, andS. H. Snelgrove, in press, 3-D characterization of a clastic res-ervoir analog: from 3-DGPR to a 3-D uid permeabilitymodel:Geophysics.

Thomas, M. A., and J. B. Anderson, 1994, Sea-level controls on thefacies architecture of the Trinity/Sabine incised-valley system,Texas continental shelf, inR.W. Dulrymple, R. Boyd, and B. A.Zaitlin, eds., Incised-valley systems: origin and sedimentary se-quences: SEPM Special Publication 51, p. 6382.

Vail, P. R., 1987, Seismic stratigraphic interpretation procedure, inA. W. Bally, ed., Atlas of seismic stratigraphy: AAPG Studiesin Geology 27, p. 110.

Van Wagoner, J. C., R. M. Mitchum, K. M. Campion, and V. D.Rahmanian, 1990, Siliciclastic sequence stratigraphy in welllogs, cores and outcrops: concepts for high-resolution correla-tion of time and facies: AAPG Methods in Exploration 7, 55 p.

Weimer, P., and H. W. Posamentier, eds., 1993, Siliciclastic se-quence stratigraphy: AAPG Memoir 58, 545 p.

Xu, T., and G. A. McMechan, 1997, GPR attenuation and its nu-merical simulation in 2.5D: Geophysics, v. 62, p. 403414.

Szerbiak1583

Documents

radar data aredepth

permeability data

robert szerbiak

degree ingeoscience

corbeanu university

radar reectors

university of bucharest

honors degree