Supplement of Evolution of a highly dilatant fault zone in the grabens of Canyonlands National Park , Utah , USA – integrating fieldwork , ground-penetrating radar and airborne imagery analysis

Abstract. The grabens of Canyonlands National Park are a young and active system of sub-parallel, arcuate grabens, whose evolution is the result of salt movement in the subsurface and a slight regional tilt of the faulted strata. We present results of ground-penetrating radar (GPR) surveys in combination with field observations and analysis of high-resolution airborne imagery. GPR data show intense faulting of the Quaternary sediments at the flat graben floors, implying a more complex fault structure than visible at the surface. Direct measurements of heave and throw at several locations to infer fault dips at depth, combined with observations of primary joint surfaces in the upper 100 m, suggest a highly dilatant fault geometry. Sinkholes observed in the field as well as in airborne imagery give insights in local dilatancy and show where water and sediments are transported underground. Based on correlations of paleosols observed in outcrops and GPR profiles, we argue that either the grabens in Canyonlands National Park are older than previously assumed or that sedimentation rates were much higher in the Pleistocene.

of high resolution airborne imagery. GPR data show intense faulting of the Quaternary sediments at the flat graben floors, implying a more complex fault structure than visible at the surface. Direct measurements of heave and throw at several locations to infer fault dips at depth, combined with observations of primary joint surfaces in the upper 100 m suggest a model of the highly dilatant fault geometry in profile. Sinkholes 10 observed in the field as well as in airborne imagery give insights in local massive dilatancy and show where water and sediments are transported underground. Based on correlations of paleosols observed in outcrops and GPR profiles, we argue that the grabens in Canyonlands National Park are either older than previously assumed, or that sedimentation rates were much higher in the Pleistocene.

Introduction
Understanding the structure of dilatant fractures in normal fault zones is important for many applications in geoscience. Reservoirs for hydrocarbons, geothermal energy and fresh water often contain dilatant fractures (e.g. Ehrenberg and Nadeau, 2005;Wennberg et al., 2008;Jafari and Babadagli, 2011). A better understanding of the 20 internal structures of such fault zones is required to extrapolate fault geometries in typical reservoir depths below the industrial seismic resolution. This is especially complicated when failure structures of different generations produce a mechanical stratigraphy and ultimately interact with each other.
In this paper we present a case study focusing on the spectacular grabens of the Introduction Needles Fault Zone, ranging from 2 • toward NW (Walsh and Schultz-Ela, 2003;Furuya et al., 2007) to 4 • toward WNW (McGill and Stromquist, 1979;Huntoon, 1982;Trudgill and Cartwright, 1994). The stratigraphy of the study area consists of more than 300 m of Pennsylvanian evaporites at the base, mainly halite, gypsum and anhydrite (e.g. Cartwright et al., 5 1995). These units are nowadays located at depths between 300 and 500 m, and crop out only within the Colorado River incision (e.g. Mertens, 2006). These are overlain by sandstones, intercalated shales and limestones of the Pennsylvanian, and the prominently jointed Permian Cedar Mesa sandstones on top (Lewis and Campbell, 1965;Huntoon, 1982;McGill et al., 2000). This entire overlying sediment package 10 reaches up to 500 m in thickness (e.g. McGill and Stromquist, 1979;Trudgill and Cartwright, 1994;Schultz and Moore, 1996;Cartwright and Mansfield, 1998;Moore and Schultz, 1999;Fossen et al., 2010).
At the present day the Grabens are filled with unconsolidated Quaternary soft sediments. According to the geological map of Billingsley et al. (2002) most of Devil's 15 Pocket, and Cyclone Canyon as well as northern Devil's Lane are dominated by Holocene and Pleistocene alluvial deposits. The southern section of Devil's Lane is mostly filled with Holocene alluvial fan deposits. Isolated depressions permitted ponding of Holocene silt and mud in Devil's Lane and Cyclone Canyon with sediment thicknesses up to two meters. Faulting significantly influences distribution of local 20 deposition centers and the drainage system (Trudgill, 2002). All grabens show occasional rock fall debris.
Incision of the Colorado River, in combination with the local tilt is seen by most researchers as cause for the formation of the grabens, either due to salt movement or gravitational gliding. Huntoon (1982) proposed that salt movement occurs as a reaction 25 to the unloading effect of the incising Colorado expressed as the Meander Anticline, a fold structure following the Colorado River with the stream flowing in the fold axis striking roughly NNE. Furuya et al. (2007) showed by using interferometric synthetic aperture radar (InSAR) measurements that the uplift of the Meander Anticline as well Introduction as the graben formation is ongoing, which is expressed most prominently by NW oriented extensional movements of up to 3 mm a −1 in the south of the grabens. Faulting in the Canyonlands National Park occurs as aseismic creep-faulting, the only recorded earthquake activity was a series of microquakes in spring of 1987 with magnitudes M L ≤ 1.8 in depth of 6-10 km, which is below the decoupling salt layer (Wong et al.,5 1993; Moore and Schultz, 1999). To investigate sediment thicknesses and their distribution, Grosfils et al. (2003) and Abrahamson (2005) used seismic refraction surveys in northern Devils Lane and Cyclone Canyon, respectively. Grosfils et al. (2003) additionally collected gravity data. Evaluation of the data lets them estimate maximum thicknesses of Pleistocene and 10 Holocene sediments of more than 90 m in Devils Lane and 60-75 m in Cyclone Canyon. The resulting total throw (i.e. sediment thickness plus graben wall height) in both grabens exceeds earlier estimates (e.g. Cartwright et al., 1995) by a factor of up to 1.5 (e.g. ≥ 145 m instead of ≤ 105 m for the Master fault in Devil's Lane). In the following we distinguish between the terms graben-bounding fault, which is the actual fault plane 15 and fault position forming the graben, and graben wall, which is the exposed rock on either side of the graben. The latter is almost invariable an original joint surface, in many cases affected by weathering.
In the northern section, the upper hundred meters of outcropping hard rock are cut by two characteristic regular joint sets (e.g. McGill and Stromquist, 1979). One joint-20 set strikes NNE, the second one to SE. Both joint sets change their orientation slightly towards the south, the NNE set following the change of the graben orientation, the second joint set stays roughly normal to the first one with a deviation of up to 30 • in the vicinity of eastern Chesler Canyon. According to the considerations of McGill and Stromquist (1979) these latter joints are older than the grabens. Their considerations 25 are based on the variety of different angles between joint strike and graben wall orientation, which are neither consistent with shear nor extensional origin of jointing. Additionally, the observation of joints with regular spacing at exposed graben floors (e.g. northern Devil's Lane) imply that they are older than the graben faults. Several Introduction papers have interpreted that the faults developed in a way strongly influenced by the preexisting joints (McGill and Stromquist, 1979;Cartwright and Mansfield, 1998;McGill et al., 2000;Trudgill, 2002). The effect of joints on the geometry of fault tips has been described by Cartwright and Mansfield (1998), but an extensive analysis of the faultjoint relationship considering faults and joints in the entire grabens area and the related 5 structures has not been done before.

Data sources and methodology
Detailed mapping of faults and joints as well as a remote sensing analysis of topography and structures was carried out as base for geophysical data collection and field mapping. We used high resolution (25 cm pixel −1 ) airborne orthoimagery 10 (Utah Automated Geographic Reference Center, 2009) for the northern section and aerial images with 1 m resolution (National Agriculture Imagery Program, 2009) for the southern parts. These images are also used in the figures of this paper. In addition to remote sensing, classical field mapping with GPS, compass, highresolution photographs, and a laser rangefinder, we used a ground penetrating radar 15 system by GSSI (Salem, USA) to image subsurface features. The survey equipment consisted of the SIR 3000 field computer with GPS tracker, 100 and 400 MHz antennas, and a survey wheel. We aimed at reaching penetration depths up to 10 m or more with the lower-frequency antenna. Data with a surface-near resolution better than 10 cm are routinely gathered by the 400 MHz system. Data processing was performed 20 with the ReflexW software package (Sandmeier, 2011) and included static corrections, background removal, gain adjustments, and frequency filtering. Topographic correction (Neal, 2004) was applied to those profiles with significant topographic variations only.
Wave travel times were converted into depths based on literature velocity values and hyperbola analyses where possible. Values between 0.1 and 0.15 m ns −1 were expected for the relatively dry sandy soils present in the Canyonlands National Park (Smith and Jol, 1995;Heteren et al., 1998) single blocks in several profiles revealed velocities of 0.125 ± 0.005 m ns −1 . This value was used for the time-depth-conversion in the profiles. Vertical resolution of the data depends on the wave velocity as well. Maximum achievable vertical resolution is a quarter of the wavelength (wavelength equals wave velocity divided by frequency: λ = v/f ; Neal, 2004), and thus approximately 8 cm for 5 the 400 MHz antenna in our study.
A total of 66 GPR profiles were collected with a cumulative length of more than 7 km, distributed over Devils Pocket, Devils Lane, and Cyclone Canyon. The different antenna types were used to achieve either higher penetration depth or better resolution, respectively. Profiles normal to fault strike were investigated with both antennas as we

Field observations
We focused on Devil's Pocket, Devil's Lane and Cyclone Canyon because in these grabens the walls are less weathered and jointing is most distinct, but included an excursion to Cross Canyon in the south and Lens Canyon in the west. Besides the GPR data that are described later we also collected basic geometric information of 5 these grabens using a laser range-finder and visual observations. Three major findings are of importance: 1. Along the 4wd track leading from Devils Lane southwards along the Bobby Jo Camp to Bobby's Hole we observed a number of localized depressions at graben boundaries and faults, interpreted as sinkholes in agreement with work by (Biggar 10 and Adams, 1987). The sinkholes can reach from few meters to tens of meters in length and several meters in depth. They were mapped from field observations and orthoimages (for map distribution see Fig. 3). Numerous sinkholes observed in airborne imagery are not located at graben boundaries, but coincide with faults within or across grabens. Figure 5a shows an example of an airborne image 15 of such a sinkhole located at the eastern graben wall in a graben east of Cow Canyon (see arrow "1" in Fig. 3). A channel like structure protruding towards the graben wall indicates sediment transport towards the depression at the graben wall. The corresponding field photograph ( Fig. 5b) shows the geometry and extent of this sinkhole. Apparently, water as well as transported sediment disappears 20 underground at these locations. Due to their depth of several meters they often act as a trap for the abundant tumble weed. It is unclear though, whether there is an interconnected system of open fractures that allows a transport of sediments over larger distances, or the opening rate of these fissures is larger than sediment input, so that they do not fill up. Figure 5c shows a sinkhole that is not located Introduction underground. According to National Park staff, this sinkhole opened up in 2011 during heavy rainfall, which washed away the sediment cover.
2. At three locations we were able to exactly measure heave and throw of offset blocks that remained visible within the grabens. While throw can be determined precisely by horizontal marker beds, heave is affected by an unknown amount 5 of erosion, and thus measured values are considered overestimates. One measurement was done in northern Devil's Pocket (Fig. 6a). The throw at the western graben wall could be measured quite precisely (38.5 m) using the laser range finder. The heave is estimated about 25 m but rock fall from the graben wall has likely affected the heave here. In Devil's Pocket south of Devil's Kitchen a huge 10 offset block is located within the graben (Fig. 6b) and marker horizons allowed to measure a throw of 31 m and a heave of 8 m. Again the heave is affected by erosion and is therefore slightly overestimated. A second block further west is relatively higher than the previous, implying a second fault in between. A third outcrop is located right at the previously described sinkhole east of Cow Canyon

15
(cross in Fig. 3) and depicted in Fig. 6c. The throw is 2.4 m and the heave 7.8 m.
Note that the heave is more than two times larger than the throw, indicating a very shallow fault.
3. To test hypotheses on dilatant faulting we looked for slickensides or tool marks at not weathered fault/joint surfaces. Moore and Schultz (1999) described 20 slickenlines at calcite-coated joint surfaces, but we did not find any examples ourselves. The majority of our observations point to extensional fracturing, without frictional sliding (Fig. 7). We confirm the observation of calcite coatings covering the faulted joint surfaces at many places. Occasionally these are formed by rainwater flow over the surfaces in streams, thus producing individual trails

GPR observations
A number of profiles selected for their descriptive value are shown and interpreted in the following paragraphs. Especially the profiles crossing from a graben to a horst, thus being crosscuts of a graben-bounding fault, show the most interesting results.
The given examples illustrate sediments dipping towards the graben wall, sediments 5 dipping towards the graben center, changing deposition rates within a graben and graben internal faulting. Dip angles of the faults cannot be determined exactly since depth migration of the GPR data was not possible due to amplification of artefacts during migration. The location of all described profiles is illustrated in Fig. 3a with numbers according to the subheadings.

Devil's Lane WE-profile
The profiles were acquired with both the 400 ( profile, respectively. Both antennas image sediments dipping towards the graben bounding fault over a length of 30-40 m, but differing in the depth at which they can be observed. Whereas layers may be interpreted to become horizontal at a depth of around 2.5 m in the 400 MHz profile, the 100 MHz antenna shows dipping structures down to 5 m depth.

20
The discrepancy is assumed to be the result of different resolutions. In the airborne imagery (a) this area coincides with darker colors, indicating more humidity and a slight surface depression, which is also observable in the field.
More localized areas of dipping strata are observed in three other positions. Dips from the west towards the 60 m mark can be observed in the 400 MHz profile as which the observations can be made.
In addition to dipping sediments we interpret a number of faults in these profiles, which cannot be seen at the surface ( Fig. 8c and e). These faults mostly correlate between both profiles, although some smaller faults seen in the 400 MHz profile could not be resolved in the 100 MHz profile. The average spacing between the faults is about 10-15 m. Many faults reach up close to the surface, but there are also some that show no more offset in several meters depths.

Northern Cyclone Canyon WE-profile
This 100 MHz profile in northern Cyclone Canyon starts at the western graben wall and follows the trail towards SE (see Fig. 9a). It shows a sediment package of about 15 10 m thickness dipping from the surface down to 10-20 m depth over a length of 100 m along the profile. This dip is only an apparent dip, as the actual structure is unknown (e.g. alluvial fan). The overlying horizontal layers gain thickness along the profile and finally reach down to 10 m.
In this profile there are also faults to be found as illustrated by dashed lines and 20 named in Fig. 9c. We interpret three major faults and four associated minor faults. Fault 1 is clearly visible due to terminating reflectors and a change in reflector intensity at the 16 m mark. This fault is interpreted to be the graben bounding fault of the northern graben segment. Faults 2 and 3 are located more distal from the graben-wall and both show small associated faults to their west. Again, terminating reflectors and changes in 25 reflector intensity indicate the existence of these faults. The strike direction of faults 2 and 3 is uncertain, but we assume them to be parallel to fault 1, following the general Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | regional trend of fault strikes. The three major faults are indicated as interpreted with white dashed lines in the airborne image (Fig. 9a).

Central Cyclone Canyon WE-profile
Evidence for changes in either deposition or deformation rate were found in a profile in central Cyclone Canyon, starting at the western graben wall and then changing 5 to a NNE direction as illustrated in Fig. 10a. This 85 m long profile was shot using the 400 MHz antenna to achieve a high resolution in the upper 4 m of sediments. Figure 10b shows the profile without interpretation and Fig. 10c indicates the prominent sedimentary features as our interpretation. Three distinct horizontal beds can be seen here. The first one being within the upper 20-40 cm, the second one in a depth of about

Devil's Kitchen WE-profile
This 400 MHz profile was taken along a WNW-ESE path crossing Devil's Kitchen as shown in Fig. 11a. It starts right behind the western graben-bounding fault of 20 the northern section of Devil's Pocket graben and then leads to the eastern graben wall, again not crossing the bounding fault. Both bounding faults are indicated by white dashed lines in Fig. 11a. Due to the high resolution, the profile provides useful information only down to 5-6 m. Figure 11b depicts the raw profile after processing while in Fig. 11c we present our interpretation. Two major faults are visible by

Interpretation and discussion
Combined with the field and remote sensing data (observations, mapping, airborne imagery and digital elevation models), the information derived from the GPR profiles 20 support and enhances the knowledge of how the grabens form and which processes are involved:

Evidence for dilatant faulting
Evidence for dilatant faulting is the occurrence of sinkholes that we observed along graben bounding faults in the more eastern grabens (cf. Fig. 3 allow water and sediments to be transported underground (Fig. 5). Analysis of airborne imagery showed several sinkholes located within grabens at locations where a fault is crossing. These sinkholes are comparable to pit craters described by Ferrill et al. (2004Ferrill et al. ( , 2011. Also, during our field work we never observed slickensides at unweathered faulted 5 joint surfaces. Slickenlines should be abundant at discrete slip surfaces in an area with so many faults as the Needles Fault Zone. The absence of such striations indicates dilatant faulting by reactivation of the preexisting joints. Layers distinctly dipping towards the graben wall require a depression into which they can dip. The GPR profile from Devil's Lane (Fig. 8) shows an example of this situation. 10 At both ends of the profile layers dip towards the respective graben wall. The profile from Devil's Pocket (Fig. 11) partially shows this situation. Layers to the east of fault 1 dip towards the western graben wall and at fault 2 dip occurs from east and west.

Fault shape
The combination of opening of preexisting joints during faulting and the considerable 15 vertical displacement in the grabens requires a change in dip of the faults in depth. The measurements of heave and throw in Devil's Pocket, Devil's Lane and the graben east of Cow Canyon provide direct evidence of fault dip at depth according to geometric considerations illustrated in Fig. 12a. The fault dip α is defined as α = arctan (throw/heave). Joint surfaces that were once connected are now separated by several 20 meters horizontally while also showing distinct vertical offset. Estimates of this are highly variable and show required fault dips of 60-80 • at depth in Devil's Pocket to produce the observed offsets, whereas in other cases very shallow dipping extensional faults are assumed. Disregarding the erosional effect on the heave, in the graben east of Cow Canyon a fault dip of less than 20 • is calculated. This can be explained by 25 a secondary low-angle slip on a lubricating layer due to removing the confinement in the initial tensile deformation (see Fig. 12b). Other authors found field evidence for such a vertical change of the fault dips in the grabens of the Canyonlands by investigating crosscut grabens along Y Canyon, Cross Canyon and Lower Red Lake Canyon. Based on observations in these crosscuts McGill and Stromquist (1979) proposed that faults are vertical over about 100 m followed by dips of about 75

SED
• down to the evaporite interface. This observation was confirmed by This "pull-apart" model in mechanically layered stratigraphy causing opening in the more competent layers is well known from outcrops (e.g. Peacock and Sanderson, 1992;Peacock, 2002;Ferrill and Morris, 2003;Crider and Peacock, 2004;Ferrill et al., 15 2014) and shown in DEM models as well (e.g. Schöpfer et al., 2007a-c;Abe et al., 2011). We suggest that in the Canyonlands this effect is additionally controlled by the abundant vertical joint sets that are defining the present day graben walls.
Analogue modeling of the interaction of preexisting joints and normal faults in brittle rocks was performed by Kettermann (2012). The basic deformation box, setup and 20 scaling relations were the same as in the work of Holland et al. (2011). Additionally, joints were introduced to the system, by hanging sheets of paper into the box during filling, and carefully removing them before the deformation. As the hemihydrate is cohesive, open joints are formed and can then affect the faulting. The maximum depth of the joints as well as the joint spacing was chosen to represent the situation in the 25 Canyonlands. In a series of experiments the angle between joint strike and fault strike was varied in a range observed in the field. Figure 13 shows a top view of an experiment with 8 • angle between joint strike and basement-fault strike. Observed structures are

Fault zone complexity
Fault zones are usually complex in geometry and strain partitioning (McClay, 1990;Childs et al., 1996a, b;Walsh et al., 1999Walsh et al., , 2003Mansfield and Cartwright, 2001;15 Peacock, 200115 Peacock, , 2002Wilkins and Gross, 2002;Soliva et al., 2008). For the Canyonlands grabens Baker (1933) first described a graben internal deformation by a number of graben-parallel faults in the downthrown blocks where they were exposed, although he did not define the location of his observations precisely. Walsh and Schultz-Ela (2003) showed field evidence from a crosscut in Y Canyon, indicating at least 20 one graben internal fault and successfully compared this observation with numerical models. They also note incremental multiple faulting from aerial photographs. This graben internal faulting is consistent with our observations from crosscuts in Cross Canyon as illustrated in Fig. 14. The black lines depict the location of the main graben-bounding faults of this graben (for location see cross in Fig. 3), and the white 25 lines show two conjugate faults affecting the graben floor. A number of GPR profiles in all three investigated grabens, i.e. Devil's Pocket (Fig. 11), Devil's Lane (Fig. 8) and Cyclone Canyon (Fig. 10 The faults in Devil's Pocket and northern Cyclone Canyon additionally show more than one fault strand. Our observations show that even flat and apparently undisturbed graben floors can contain complex fault systems. As these are not visible at the surface the faults are presumably not active any more. Combining all the information presented we propose a model for the formation of 5 a typical Canyonlands graben as shown in Fig. 15. Faults dip with 60-80 • and intersect the vertical joints. Hence, the outcropping graben walls are joint surfaces rather than fault planes. Inclined faults reactivate the vertical joints at surface, and this change in fault dip forms large open voids at the graben walls, which are then filled with rock fall debris and Quaternary sediments. Additional graben internal faulting causes a complex 10 hard rock topography and requires reactive diapirism of the salt as also suggested by Schultz-Ela and Walsh (2002) and Walsh and Schultz-Ela (2003). As sinkholes are preferably found in the presumably younger grabens in the east and southeast and ponded deposits are observed in the older, western grabens (Cyclone Canyon), this might also be a criterion of graben maturity.

Graben age vs. deposition rates
The GPR profile of central Cyclone Canyon revealed detailed sedimentary structures to a depth of about four meters (Fig. 10). We noticed three distinct horizons of horizontal reflectors and in between delta-like sequences of foreset beds and onlaps. The interpretation of lacustrine delta sediments is supported by typical layering of the 20 dipping sediments, visible by changing reflector intensities, that implies sedimentation in water due to sudden events such as flash-floods. The geological map of Billingsley et al. (2002) shows that ponded deposits are common in Cyclone Canyon and the authors state that these ponds were capable of sustaining water for several months to years. The thickness of the ponded deposits of up to 2 m fits well to the observed 25 thickness in the radar profiles. Especially the three horizontal layers that we interpret as paleosols fit very well to a graben fill east of Virginia Park, described and dated by Reheis et al. (2005) location is marked with a black star in Fig. 3b. Thanks to the incision of a stream they were able to visually study a soil-profile down to 6 m depth and apply optically stimulated luminescence (OSL) dating. They described three paleosols, one close to the surface, one in two meters depth, and a third one in about three meters depth. Dating right above the second paleosol results in an age of about 15-16 ka and 5 sediments above the third paleosol were dated to 27-28 ka. Figure 16 shows the GPR profile of the central Cyclone Canyon with the stratigraphic column and OSL-ages of Reheis et al. (2005) as an overlay. The depths of all paleosols match with deviation of only a few cm. Based on the proposed age of the grabens (Biggar and Adams, 1987;Schultz and 10 Moore, 1996;Trudgill, 2002), the sediments thickness in Cyclone Canyon (Grosfils et al., 2003) and the known age of these paleosols (Reheis et al., 2005) we propose three possible scenarios of graben formation and sedimentation: 1. The horizons in GPR and outcrop are not the same. Still we would interpret the observed horizons as paleosols. Assuming a faster deposition rate in Cyclone

15
Canyon due to its preferred role as a sediment trap would then suggest longrange climate changes within the recorded four meters of the profile and distinct differences in deposition over short distances.
2. The horizons in GPR and outcrop are the same. An explanation for the small discrepancy in the depth of the paleosols would be a faster sedimentation 20 between the first and second paleosol. This is likely, since Cyclone Canyon is a quite deep graben compared to the one east of Virginia Park and thus a better sediment trap. The delta-like foreset beds observed between the upper two paleosols indicate a high deposition rate in the presence of water, i.e. extensive flash floods as they occur from time to time depositing in ponds. Additionally, slight 25 differences in depth can result from imperfect time-depth conversion. Applying the OSL ages from Reheis et al. (2005) to the depths in Cyclone Canyon allows an estimation of sedimentation rates. Assuming the first OSL age of roughly 16 ka in a depth of 2.2 m leads to a sedimentation rate of 0.14 mm a −1 for this section.
The distance between the second and third paleosol is approximately the same as in the Virgina Park Graben -one meter -and the age of about 28 ka leads to a sedimentation rate of 0.08 mm a −1 . Seismic refraction data (Abrahamson, 2005) suggest a sediment thickness of about 50 m at this location. Following the 5 suggested ages of graben initiation of 65-85 ka (Campbell, 1987;Schultz and Moore, 1996), the deposition rates below the third paleosol must have been one order of magnitude higher. These would still be reasonable deposition rates, but would suggest a drastic change in the deposition system at one point.
3. In the third case the paleosols are the same, a sediment package of 50 m is 10 given and we assume that the rate of deposition has not changed dramatically in the past. That would point towards an earlier initiation of graben formation than previously estimated by termoluminescence dating of sediments found in sinkholes (Campbell, 1987), although their results may not apply for all grabens. Using a deposition rate of 0.15 mm a −1 as an average for the remaining 47 m 15 beneath our profile results in an age of about 300 ka. This is supported by comparing recent deformation rates with estimates of total strain in the Needles Fault Zone. Westwards movements of 3-9 mm a −1 were proposed from InSAR data by Furuya et al. (2007). These fit to the GPS data of Marsic (2003), who reported 5 mm a −1 . According to Moore and Schultz (1999) a total extension 20 of 1.29 km took place on the northern section of the grabens. Assuming an average extension of 5 mm a −1 would require an age of graben initiation of about 260 ka, which would fit well to the observed sediment thickness and deposition rates. Schultz-Ela and Walsh (2002) used geomechanical numerical modeling to reproduce the grabens and proposed an age of 135 ka for their models, which is To resolve the remaining problems, more data are needed from different grabens and locations. Especially case (1) can only be validated or rejected by drilling in Cyclone Canyon and actually comparing and dating the sediments directly. However, it seems legitimate to assume an intermediate situation of cases (2) and (3). Some authors (Trudgill and Cartwright, 1994;Schultz-Ela and Walsh, 2002) propose graben 5 initiation at 500 ka or more, following the timing of the Colorado River incision. Higher deposition rates are also possible. A mixture of both would be a good explanation for the intermediate age of 260-300 ka proposed in our model.

Conclusions
By applying ground penetrating radar surveys combined with field observations in the Needles Fault Zone of Canyonlands National Park, Utah/USA, we draw the following conclusions: 1. Our GPR profiles revealed faults in the subsurface that were overprinted and not visible at the surface. We can therefore conclude that the fault zone of the Canyonlands grabens is more complex than obvious. Since this area is an 15 analogue for reservoirs, knowledge of the fault zone complexity can be used to enhance seismic interpretations and fault zone permeability estimates.
2. Dilatant faulting is evident in many places, even if the surface expression is overprinted by sedimentation. Measurements of heave and throw at some faults additionally provide evidence for vertical fractures at the surface in combination 20 with joint reactivation and inclined faults at depth.