GEOPHYSICS, VOL. 70, NO. 6 (NOVEMBER-DECEMBER 2005); P. K53–K61, 15 FIGS., 1 TABLE. 10.1190/1.2122414
Acquisition and processing strategies for 3D georadar surveying a region characterized by rugged topography
Bjoern Heincke1 , Alan G. Green1 , Jan van der Kruk1 , and Heinrich Horstmeyer1 characterize the shallow subsurface to depths of 10 to 50 m. Under favorable conditions, these methods can provide highresolution structural information in a nondestructive and costeffective manner. Until about 10 years ago, surface GPR studies were based on sparse 2D profiles. More recently, they have been extended to three dimensions, thus allowing highly heterogeneous structures to be imaged accurately for stratigraphic studies (Beres et al., 1995), for the detection of fractures and faults in crystalline rock (Grasmueck and Green, 1996), for archaeological surveys (Pipan et al., 1999), and for paleoseismic investigations (Gross et al., 2002, 2004). Most 3D GPR data sets have been recorded across topographically flat or gently dipping smooth surfaces. To our knowledge, none have been collected across topographically rugged terrain because of technical reasons; conventional procedures that involve recording GPR data and coordinate information separately are too time consuming for this purpose, and there is no commercially available software for correctly migrating shallow-wavefield data recorded on undulating surfaces. To address these issues, we present data acquisition and processing strategies used for 3D GPR surveys in an unstable rugged mountainside that is certain to produce a major rockslide. The survey was part of a large, multidisciplinary research project designed to investigate sliding processes within crystalline rock and thus to derive a better understanding of rock failure mechanisms (Willenberg et al., 2002). The purpose of the GPR survey was to supply information on the distribution and orientation of fracture zones to approximately 40 m depth. We begin by describing the survey location and our dataacquisition strategy, which involves using a GPR system that records GPR traces and corresponding coordinates simultaneously. The data processing is complicated by inaccurate coordinates in the vicinity of abrupt topographic discontinuities, changes in signal character created by antenna-ground coupling variations, and unavoidable data gaps associated with large boulders and undulating topography. To suppress these
ABSTRACT Efficiently performing 3D ground-penetrating radar (GPR or georadar) surveys across rugged terrain and then processing the resultant data are challenging tasks. Conventional approaches using unconnected GPR and topographic surveying equipment are excessively time consuming for such environments, and special migration schemes may be required to produce meaningful images. We have collected GPR data across an unstable craggy mountain slope in the Swiss Alps using a novel acquisition system that records GPR and coincident coordinate information simultaneously. Undulating topography (dips of 8◦ to 16◦ ) and boulders with diameters up to about 2 m complicated the field campaign. After standard processing, the data were found to be plagued by time shifts associated with minor coordinate inaccuracies, uneven antenna-ground coupling, and numerous small gaps in data coverage. These problems were resolved by passing the data sequentially through an adaptive f -xy deconvolution routine and f -kx and f -ky filters. This filtering also reduced incoherent noise. Finally, the data were migrated using a 3D algorithm that accounted for the undulating topography. The nonmigrated and migrated images contained gently and moderately dipping reflections from lithological boundaries and actively opening fracture zones. A suite of prominent diffraction patterns was generated at a steeply dipping fracture zone that projected to the surface. Through this case history we introduce a general strategy for 3D GPR studies of topographically rugged land.
INTRODUCTION Over the past two decades, surface ground-penetrating radar (GPR or georadar) methods have been widely used to
Manuscript received by the Editor May 7, 2004; revised manuscript received September 17, 2004; published online October 21, 2005. 1 ¨ ¨ Swiss Federal Institute of Technology, Institute of Geophysics, ETH-Honggerberg, CH-8093, Zurich, Switzerland. E-mail:
[email protected];
[email protected],
[email protected];
[email protected]. c 2005 Society of Exploration Geophysicists. All rights reserved. K53
Downloaded 24 Apr 2009 to 132.198.136.75. Redistribution subject to SEG license or copyright; see Terms of Use at http://segdl.org/
K54
Heincke et al.
negative effects and improve the general quality of the images, a combination of f -xy deconvolution and low-pass spatial filtering proves to be effective. A 3D version of Lehmann and Green’s (2000) topographic migration algorithm is then used to generate representative images of the subsurface. Finally, the shallow fracture distribution is illustrated on vertical sections and horizontal slices extracted from the nonmigrated and migrated 3D data volumes.
SURVEY SITE The Randa investigation site is situated on a Swiss Alpine mountainside at an altitude of 2350 m, immediately above a 1-km-high scarp that resulted from a 30 million-m3 rockslide that blocked the only land route through the Matter Valley to Zermatt in 1991 (Figure 1). Shortly after this event, a 100 × 150 m area of the rock mass above the scarp became unstable. This mass is moving centimeters per year in the direction of the 1991 rockslide scarp (Willenberg et al., 2002), such that open fractures are now visible at the surface (see Figure 2c). Our GPR data were recorded across two overlapping survey areas, A1 and A2 (40 × 35 m and 22 × 22 m, respectively), located on a natural terrace about 50 m from the 1991 rockslide scarp (Figure 2a). The site is covered with low-lying vegetation and a thin soil layer with a maximum thickness of a few meters, typical of Alpine pastures. Isolated boulders (e.g., I and II in Figures 2a and 2b) with diameters up to 2 m are scattered across survey area A1. Although the general slope of the mountainside is easterly directed, the rugged and craggy ter-
Figure 1. Photograph of the 1991 Randa rockslide in the Swiss Alps (courtesy Heike Willenberg). The investigation site is located immediately above a rockslide scarp at an altitude of 2350 m. The inset locates the site in southern Switzerland.
rain of the GPR survey site dips 8◦ to 16◦ in a southwesterly direction. The foliation of the crystalline rock mass, which comprises a complex mixture of gneisses, schists, amphibolites, and granitic intrusions (Willenberg et al., 2002), dips roughly 25◦ west-southwest. Along the eastern edges of the survey areas, the outcropping crystalline basement steps approximately 8 m down to the next terrace (Figure 2a). One open fracture zone, z1 (Figure 2c), and two surface lineaments that likely define fracture zones, z2 and z3, occur within or just outside the survey areas (Figure 2a). Debris and overburden made it difficult to determine the full extent and dips of the fracture zones from surface observations.
DATA ACQUISITION We used Lehmann and Green’s (1999) acquisition system to collect the GPR and topographic data. It included a standard pulse EKKO 100A GPR unit, a Leica TCA 1800 self-tracking theodolite, and two field laptop computers (Figure 3).
Recording parameters Unshielded 100-MHz transmitter and receiver antennas separated by 1.0-m and a 2.15-m mast holding the theodolite target prism were mounted on a sled. The long mast ensured
Figure 2. Photographs of the investigation site. (a) Viewed from the west, the dashed black lines outline the 3D GPR survey areas A1 and A2. The solid red line delineates surface fracture z1. Surface lineaments that may delineate fractures are shown by dashed red lines. Numerous large boulders (diameters of 1 to 2 m) are scattered across area A1, the largest of which are marked I and II. Crystalline basement rocks outcrop along the eastern edges of the survey areas. (b) Viewed from the south, highly uneven topographic relief and numerous boulders complicate the 3D GPR data acquisition. I and II are as for (a). (c) Surface fracture z1.
Downloaded 24 Apr 2009 to 132.198.136.75. Redistribution subject to SEG license or copyright; see Terms of Use at http://segdl.org/
3D Georadar Surveying Rugged Terrain
continuous line of sight between the theodolite and prism in regions with large boulders and other topographic undulations. Fiber-optic cables connected all components of the acquisition system. While towing the sled, the laser theodolite automatically followed the prism, such that the GPR traces and coordinates of the prism were recorded simultaneously. The same acquisition parameters were used for the two surveys, which were conducted two months apart. Since the rock mass was generally dry (no standing water was observed in a nearby 120-m-deep borehole), the recording conditions were very similar for the two surveys. To take advantage of the approximately 40-m depth penetration offered by the dry-rock environment, a long 1050.4-ns recording window was adopted (Table 1). Each recorded trace was the result of vertically stacking 32 individual traces. To ensure the nonaliased recording of steeply dipping events, the spacing d between traces in common-offset data should be
d ≤
vmin , 4fmax sin amax
(1)
where fmax is the maximum dominant signal frequency, vmin is the lowest expected GPR velocity, and amax is the steepest expected reflector dip. Dominant signal frequencies at Randa Table 1. Acquisition parameters for the 3D GPR surveys A1 and A2. Parameter
Value
Nominal antenna frequency Recording length Sampling interval Vertical stack Nominal spacing in x-direction (crossline) Nominal spacing in y-direction (inline) Estimated coordinate accuracy (except for regions covered by large boulders)
100 MHz 1050.4 ns 0.8 ns 32 ∼0.2 m ∼0.25 m x ∼ 0.03 m y ∼ 0.04 m z ∼ 0.01 m
K55
are 1 m). The vertical-tohorizontal exaggeration is 2:1. (b) Plot showing elevations determined from the midpoint coordinates.
The f-xy deconvolution The f -x and f -xy deconvolution routines were originally introduced as means to reduce random noise in seismic data (Canales, 1984; Chase, 1992). Their potential for interpolating data was recognized somewhat later (Spitz, 1991; Wang, 2002). They operate on the assumption that deterministic signals in any trace are predictable on the basis of information contained in adjacent traces, whereas random noise is normally unpredictable from trace to trace. All traces are first divided into a number of overlapping time windows that are transformed into the frequency domain. For each time window and for each frequency of the complex amplitude spectra, a rectangular spatial prediction filter is applied. By using a least-mean-squares adaptive algorithm (Widrow et al., 1967, 1975), the prediction filter automatically adapts to spatial variations of reflection strike and dip. During f -xy filtering, each time window of an input trace is effectively replaced by a phaseshifted suite of values determined from adjacent input traces (Stearns and David, 1993). An individual input trace contributes to neighboring prediction-filtered traces but not to its own prediction-filtered trace. Since the prediction-filter procedure is directional, the input data are filtered in the positive and negative x- and y-directions, and the results of the four filtering operations are stacked. On completing the prediction filtering, the data are transformed back into the time domain and summed to yield the filtered 3D data set.
Figure 8. An x-directed cross section (y = 18 m) extracted from the 3D GPR data set at different processing stages. Vertical pale blue lines delineate data gaps (blank traces), and red arrows indicate positions of boulders. (a) Cross section after standard processing. Imprecise coordinates result in artificial discontinuities. (b) After f -xy deconvolution, most artificial discontinuities are markedly reduced and traces are interpolated to fill the narrow gaps. (c) High-wavenumber noise is suppressed after low-pass filtering in the f -kx and f -ky domains [an example is shown in the red circled region in (b); see also Figures 9 and 10].
Figure 9. The f -kx spectra of cross sections shown in Figure 8. (a) Immediately after application of conventional static corrections, artificial discontinuities are mainly responsible for energy (partially aliased) in the high-wavenumber region of the f -kx spectrum. (b) The f -xy deconvolution eliminates most high-wavenumber noise. (c) Remaining high-wavenumber noise is reduced after low-pass filtering in the x- and y-directions. Low-frequency (
