Echo source discrimination in single-pass airborne radar sounding data from the Dry Valleys, Antarctica: Implications for orbital sounding of Mars

Authors


Abstract

[1] The interpretation of radar sounding data from Mars where significant topographic relief occurs will require echo source discrimination to avoid the misinterpretation of surface echoes as arising from the subsurface. This can be accomplished through the identification of all radar returns from the surface in order to positively identify subsurface echoes. We have developed general techniques for this using airborne radar data from the Dry Valleys of Antarctica. These data were collected in a single pass, including Taylor Glacier, ice-covered Lake Bonney, and an ice-free area of Taylor Valley. The pulsed radar (52.5–67.5 MHz) was coherently recorded. Our echo discrimination techniques included a radar simulator using a digital elevation model (DEM) to predict the location and shape of surface echoes in the radar data. Real and simulated echo strengths were used to calculate a signal-to-clutter ratio. This was complemented by the cross-track migration of radar echoes onto the surface. These migrated echoes were superimposed on the DEM and imagery in order to correlate with surface features. Using these techniques enabled us to identify a number of echoes in the radar data as arising from the surface and to identify subsurface echoes, including a continuous reflector under the main trunk of Taylor Glacier and multiple reflectors beneath the terminus of Taylor Glacier. Surface-based radar confirms the thickness of the glacier at three crossing points. The results illustrate the importance of using complementary techniques, the usefulness of a DEM, and the limitations of single-pass radar sounding data.

1. Introduction

[2] The primary goals of ongoing and upcoming orbital missions to Mars are to detect and characterize subsurface features, with liquid and frozen water being the main targets. This is being attempted through the use of long-wavelength (i.e., meters to hundreds of meters) radar sounding. Layered polar deposits and subice topography are also of great interest, so the characterization of any subsurface reflectors in the radar data will be of considerable interest. To date, the only orbital radar sounding data acquired anywhere have been of Earth's moon during the Apollo program [Phillips et al., 1973]. On Earth, airborne radar sounding is routinely used to penetrate several kilometers of cold, meteoric ice in the polar regions [e.g., Blankenship et al., 2001]; however, airborne radar sounding data over land have been nonexistent or scarce. Thus there have been little means by which to evaluate models of radar propagation through Mars-like materials under conditions analogous to those expected on Mars or at the approximate scale at which those data will be collected. This need was the impetus for the collection and analysis of airborne radar sounding data in the McMurdo Dry Valleys, Antarctica. Climatic, hydrological, and geological conditions in the Dry Valleys are analogous in many ways to those on Mars, and a number of ice-related features in the Dry Valleys may have direct morphologic and compositional counterparts on Mars.

[3] Given the long wavelengths involved, radar sounders employ antennas with broad beamwidths that are typically nadir pointing. Energy can be returned from the surface directly below the radar, from the off-nadir surface, and from the subsurface. Off-nadir surface returns, often referred to as “surface clutter,” can mask the subsurface returns arriving at similar time delays. Therefore a crucial step in the data interpretation process is the discrimination of echo sources in the radar data. The goal is to identify all returns from the surface in order to positively identify subsurface echoes. Benham and Dowdeswell [2003] demonstrated that mapping radar echoes to a flat surface combined with simple echo prediction using the known location of discrete surface features is useful for echo source discrimination in airborne radar sounding data. We have improved upon these techniques to include the effects of topography on the migration of echoes to the surface and the generation of simulated radar data for the entire illuminated surface, allowing for the quantitative comparison of real and simulated surface echoes. Using these techniques we show that most surface clutter can be identified in radar sounding data from Taylor Valley, Antarctica. The use of these or similar techniques will be critical for the correct interpretation of radar data in areas of Mars exhibiting significant topographic relief, so that subsurface echoes are identified unambiguously.

2. Radar Data Acquisition

2.1. Study Area

[4] The Dry Valleys of Antarctica encompass a 4000 km2 region consisting primarily of shallowly dipping Devonian-Triassic metasedimentss of the Beacon Supergroup pervasively intruded by the Jurassic Ferrar dolerite [Allibone et al., 1991]. The regional climate is that of a polar desert, with temperatures rarely exceeding 0°C, and averaging about −23°C in the lower reaches [Thompson et al., 1971]. Largely free of ice, the Dry Valleys have long been considered one of the best terrestrial analogs to conditions on Mars [e.g., Gibson and Ransom, 1981; Malin, 1985; Marchant et al., 2002; Sletten et al., 2003]. The cold, dry conditions produce weathering products similar to those observed on Mars (patterned ground, ventifacts, desert pavements, etc.) and waters contained within closed basins are mostly saline. Ice exists in a variety of deposits both above and below the surface. Long, deep troughs (up to 6 km long, tens of meters deep and 100 meters wide) on the surface of Taylor and Columbia Glaciers are unique to terrestrial glaciers, yet they resemble features observed on the polar caps of Mars [MacClune et al., 2003].

[5] For this study, a 24 km segment of a flight conducted on 25 November 2001 was analyzed (Figure 1). This segment spans a variety of features including the ice-free floor of Taylor Valley, ice-covered (and saline) Lake Bonney, and the lowermost ∼10 km of Taylor Glacier. Taylor Glacier, with its grooved surface, steep valley walls and subglacial ice-rock interface which intersects the surface at the terminus provides a good test case for surface clutter removal. Radar studies conducted on the surface of Taylor Glacier provide multiple constraints on ice thickness in this interval. Flight elevation ranged from 400 m to 700 m above the surface (Figure 1c). At the time of the flight, temperature at Lake Bonney was −2.9°C. The maximum and minimum temperatures that day were −0.4°C and −6.5°C, respectively.

Figure 1.

Taylor Valley test segment, with flight path (red solid line, flight direction indicated by arrows) and cross-track-migrated echoes (multiple colors, short segments) shown on (a) surface topography with contour interval of 50 m from the digital elevation model (DEM) [Csatho et al., 2005] and (b) optical satellite imagery (declassified image from 5 December 1975). (c) Aircraft height above surface. Direction of flight is from right to left.

2.2. Radar System

[6] A Twin Otter airborne platform was used for our 60 MHz (5 m wavelength) radar sounder (Figure 2). The radar system consisted of a programmable signal source with a coherent dual-channel receiver and data acquisition [Moussessian et al., 2000] linked to a high-power transmitter. The transmitted chirp pulse was centered at 60 MHz with 15 MHz bandwidth and 1 microsecond duration. The pulse repetition frequency was 6400 Hz and peak output power was 8 kW. The dual-channel receiver was configured with high and low gain channels for simultaneously recording weak subglacial bed echoes and strong surface echoes without receiver saturation. The low-gain data are used for all results presented in this manuscript since the surface return saturates the high-gain channel, causing other echoes to be obscured. The receiver downconverted the radar signals by mixing with a 70 MHz local oscillator yielding output signals with frequency range 2.5–17.5 MHz. These downconverted signals were coherently digitized with 3200 12 bit samples recorded at 20 ns intervals after each pulse, creating a record length of 64 microseconds. Along-track coherent integration of 16 radar signals resulted in 400 radar observations per second, or about every 17 cm along track. Pulse compression and unfocused SAR implementation (additional along-track integration) were part of data postprocessing (described below).

Figure 2.

Twin Otter survey aircraft. Flat-plate dipole radar antennas (mounted within airfoils) are visible under each wing.

[7] The antenna system comprises two flat-plate dipole antennas, one mounted beneath each aircraft wing (cross-track polarization), connected to a transmit/receive switch and a power splitter/combiner which enables the use of both antennas for both transmit and receive. The antenna patterns were predicted from a moment method solution of the dipoles on the Twin Otter aircraft [Peters and Newman, 1994] and the peak nadir antenna gain is 9 dB. The antenna 3 dB beamwidths are 152° along track and 12° cross track. The first cross-track sidelobes have peaks at 22° and are down 2.4 dB; the second cross-track sidelobes have peaks at 48° and are down 8 dB (Figure 3). As discussed below, the along-track beamwidth is modified by SAR processing.

Figure 3.

Modeled cross-track antenna pattern for the radar system as deployed on the Twin Otter platform.

[8] A laser altimeter (fixed relative to the aircraft frame) was used to augment radar-derived surface elevations. Precise positioning was accomplished through the use of two carrier-phase GPS receivers on the aircraft and two at McMurdo Station. Postprocessing of the positioning data yields accuracies of ∼10 cm with a sampling interval of ∼15 m.

3. Radar Data

3.1. Processing

[9] The effective along-track radar beamwidth was narrowed by unfocused SAR processing (i.e., coherent along-track integration), suppressing surface clutter in the along-track direction. Typical aircraft heights of 400–700 m produce unfocused SAR aperture distances of 45–60 m using sqrt(height × wavelength) [Cutrona, 1990]. Thus for a typical Twin Otter speed of 66 m/s, we integrate for 0.75 s (50 m) including 4800 individual pulses (300 × 16) for each unfocused SAR waveform. This result is generated 6.25 times per second, or about every 11 m along track. The predicted 3 dB along-track SAR beamwidth is 3°, yielding 20–35 m along-track resolution. We have also numerically simulated the corresponding SAR pattern from our modeled antenna pattern; this result is used in the simulation of anticipated surface echoes as discussed below.

[10] Pulse compression processing was then performed on the unfocused SAR data using an echo recorded from a smooth air to sea ice interface near McMurdo as the reference chirp. Cosine-squared frequency weighting with pedestal height 0.08 was applied across the 15 MHz bandwidth after downconversion. The compressed pulse width is 100 ns at the −3 dB point and 280 ns at the first nulls, and the first pulse compression sidelobes are down approximately 20 dB from the echo maxima. The pulse compression improved the radar system response by about 10 dB.

[11] The identification of specific radar echoes is accomplished through an interpretation of radar-grams similar to that described by Blankenship et al. [2001]. For reference in this paper, this process is referred to as “radar picking.” An echo is defined as a picked set of multiple radar returns that are contiguous (i.e., have approximately the same travel time, without along-track gaps), each return being a stack of radar records as described previously. During the radar picking process, a user sets upper and lower bounds on echoes that appear distinct and continuous in the radar record. An auto tracker then identifies the time of maximum signal amplitude within these bounds for each return, using a five-point parabolic fit to achieve subsample resolution. Determining the start and stop of a contiguous echo (along track) is sometimes trivial, but it can also be a subjective process. We attempted to delineate as many distinct echoes as possible, knowing that adjacent echoes could then later be combined. Known surface topography along the flight path was not used to guide the picking process.

3.2. Results

[12] Radar data for the flight segment analyzed in this study are shown in Figure 4a. A number of echoes that range from sharp to diffuse appear beyond the first echo (generally the nadir surface return) throughout the interval. An aircraft surface double bounce is evident in most of the record and easily identified as a sharp echo that follows the shape of the surface topography, but with steeper slopes. It is generally apparent that there is a sloping subglacial interface (presumably ice-rock) below Taylor Glacier, although its exact location is not clear, especially approaching the terminus.

Figure 4.

The acquired and processed radar data corresponding to the flight track shown in Figure 1. (a) Pulse-compressed data. (b) Identified (“picked”) echoes superimposed on the data. Each echo is a contiguous set of strong returns and is given a color to distinguish it from the others.

[13] The picked echoes are shown in Figure 4b. The aircraft double bounce echo is not picked, nor is a rather diffuse area of returned energy in the 19–23 km interval at approximately 2–3 ms (discussed below).

4. Echo Discrimination

4.1. Overview

[14] To determine which echoes are surface clutter and which could be from subsurface interfaces, two techniques are used together. In the first technique, surface returns are predicted by simulating radar data. We use aircraft position data, the modeled radar antenna pattern, and surface topography from a digital elevation model (DEM). The simulated data are compared both qualitatively and quantitatively with the actual data to reveal echoes that are clearly from off-nadir topography. The left-right ambiguity is addressed through a comparison of radar echo strength with the echo expected from each side of the aircraft, taking into account the antenna pattern, aircraft position, and known topography. The second technique identifies all echoes in the radar data and then maps them into possible correlative surface features to the sides of the aircraft through range estimation and cross-track migration. This assumes the echoes are from the surface and uses the measured travel times to map the echoes onto the DEM and optical imagery.

4.2. Radar Data Simulation (Clutter Modeling)

[15] The simulated radar data were produced using a simplified model of radar behavior based on geometrical optics. The primary inputs were aircraft position and bearing, antenna pattern, and surface topography. Phase interference, multiple reflections, and frequency-dependent effects are ignored. The modeled radar antenna pattern is used in the simulator, including the effects of unfocused SAR processing. Although the antenna pattern is somewhat range-dependent and aircraft motion results in range changes during a SAR window, we use a fixed range of 450 meters to calculate the antenna pattern. Spreading losses are accounted for separately. For surface topography, we use a high-resolution (2 m postings) digital elevation model (DEM) based on scanning lidar data acquired approximately two months after our radar overflight [Csatho et al., 2005].

[16] A simulated radar return was computed for each aircraft position. The roughness of Taylor Valley was assumed to be a significant fraction of the radar wavelength (5 m) over the radar footprint (>100 m), hence diffuse scattering was assumed to dominate the returns. Triangular facets, for which range, orientation, and projected area were computed, were generated between neighboring DEM data points within a specified region along track. Since the projected area affects both the facet's received and reradiated energy, this quantity was squared when computing the diffusely scattered component. In addition, a small specular reflection component was added. The specular component was computed in the same manner as the diffuse reflection, but was multiplied by the square of the cosine of the off-normal angle. The specular component was weighted at one tenth of the diffuse component; therefore, diffuse scattering dominates the simulated radar images.

[17] Owing to antenna pattern symmetry in the cross-track plane, a left-right ambiguity exists for off-nadir echoes. To address this, we treat left- and right-sided data separately and generate simulated data for each side of the aircraft (Figures 5a and 5b). These can be used to qualitatively identify matches in the real data or to quantify the expected difference in signal strength between a potential surface echo from the left and right sides.

Figure 5.

Simulated radar data for (a) left of flight path and (b) right of flight path. All echoes seen in the simulated data are from the surface, so any real echoes (Figure 4) matching the simulated data are likely to be from a surface source. For the production of simulated radar data and relative echo strengths (Table 1) the fore-aft antenna pattern and three-dimensional topography are used.

4.3. Comparison of Simulated Data With Real Data

[18] A qualitative comparison of the resulting simulated data (Figures 5a and 5b) with the radar data (Figure 4a) shows several features in common. Since the simulator only produces echoes from the surface, any features in the radar data that match those in the simulated data are most likely derived from the surface. In order to facilitate a comparison, the picked returns in the radar data are superimposed upon the simulated data in Figures 6a and 6b. These comparisons reveal clear correlations of some radar echoes with surface features. For example, the upwardly concave echo at approximately 13 km (labeled “b1” in Figure 4b) is readily apparent in the right-sided simulated data (Figure 5b, match shown in Figure 6b). In addition to well-defined echoes, there are other more subtle characteristics evident in both the real and simulated data. For example, the diffuse region of energy returned in the real data in the interval 19–23 km (Figure 4a) also appears in the left-sided simulated data (Figure 5a). Examination of the aircraft's position along the flight path (upper right corner of Figure 1) shows that the radar illuminated a region of hummocky moraines on the valley floor (shown in Figure 7). This demonstrates that the simulator works well for these types of surface features.

Figure 6.

Picked radar echoes (from Figure 4b) superimposed on (a) left- and (b) right-sided simulated data.

Figure 7.

Photo from left side of airplane over Taylor Valley (see Figures 1 and 9 for location). Hummocky moraines on valley floor are likely source of surface scattering apparent at approximately 19–23 km in both real data (Figure 4) and left-sided simulated data (Figure 5a).

[19] There are other echoes in the radar data which do not match anything obvious in the simulated data. These could be surface echoes that for some reason are not evident in the simulated data, or they could be true subsurface echoes. The relatively deep echoes in the 0–7 km interval (e.g., “c17a”) have no correlation in the simulated data and appear to result from the interface at the base of Taylor Glacier.

[20] A quantitative comparison between the real and simulated data is desirable for multiple reasons including the eventual automation of surface clutter detection and removal. For this we use echo strengths in both the real and simulated data. All echo strengths are normalized to the average strength of the first return (nadir surface echo) for the 24 km interval. A “signal-to-clutter ratio (SCR)” can then be calculated to compare the real echo (the signal) with its counterpart in the simulated data (the clutter).

[21] Echo strength calculations for the real data are more straightforward than for the simulated data. Real echo strengths are determined by summing the individual returns within each echo as defined in the picking process. Since echoes in the simulated data are not always aligned with real echoes to the precision inherent in the data, and since the real signals are more distributed in time, erroneous results could occur if the exact positions of the signal peaks in the real data were used when calculating echo strengths in the simulated data. Therefore a scheme is needed to quantify the strength of simulated data that corresponds to the real echoes. We chose to implement an automated version of our radar picker using the locations determined for the real echoes as starting points to create new upper and lower bounds from the real echoes (plus or minus six samples, or a 240 ns window). The algorithm then used those bounds in the simulated data to search for a local maximum for each return in the echo. Echo strengths were then calculated in the same manner as for the real echoes and were used to determine left, right and combined SCR values for each echo. A low SCR implies that surface clutter is strong relative to the real radar echo, and hence the echo is more likely to arise from a surface feature. Low energy in the simulated data (i.e., no surface echo source) would result in a high SCR value. Table 1 includes surface-normalized radar echo strengths and SCR values for all of the identified echoes.

Table 1. Echo Analysis Resultsa
PickEcho Strength Calculations, dBDiffCross-Track Migration CorrelationsComb. Result
RawSCR LeftSCR RightSCR BothLeft-Sided Visual CorrelationCR (Left)bRight-Sided Visual CorrelationCR (Right)bMax CR
  • a

    Name of pick (as shown in Figure 4b) in order of decreasing total SCR. Echo Strength Calculations: Surface-normalized echo strength in real radar data (RAW); Signal to Clutter Ratio (SCR) using left-sided (LEFT), right-sided (RIGHT), and combined (BOTH) simulated echoes (SCRs are determined by the Raw values divided by echo strengths for each simulated echo, also surface-return normalized); Left-Right SCR Difference (DIFF, positive value indicates surface source on the left side of the aircraft). Cross-track migration correlations: Visual correlations with possible surface sources using DEM and optical imagery, as discussed in text; correlation rating (CR, 0–5, see table footnote for explanation) for each side of aircraft, and maximum value (Max CR). Combined Result for each echo: subsurface echoes defined by SCR value of 20 dB or higher (asterisked entries in “BOTH” column) and maximum correlation rating of 0–2 (asterisked entries in “max rating” column).

  • b

    Visual correlation rating (CR) determined as follows: 0, no correlation with surface features; 1, shape of echo corresponds to general patterns only; 2, correlation for part of echo (33% or less); 3, correlation for part of echo (33–67%) or combination of 1 and 2; 4, correlation for more than 67% but not complete; 5, complete correlation.

c17b−63.95 58.07* on glacier, no feature correlation0on glacier, crosses flow lines00*subsurf.
c4b−39.9671.3771.3271.04*−0.05follows glacier/wall contact, and wall4no correlation04surf.
c5−44.5365.3265.4665.08*0.15partial match with glacier/wall contact2no correlation02*subsurf.
c4a−47.3563.5164.2063.43*0.69partial match with glacier/wall contact2no correlation02*subsurf.
c8−40.1263.5759.0059.00*−4.57matches glacier/wall edge, lower 3/44on glacier, not coincident with flow lines04surf.
c15a−54.2357.5558.1957.46*0.65valley wall, crosses onto glacier1generally follows glacier flow lines11*subsurf.
c9a−54.0558.5256.5756.56*−1.95on glacier, parallel with flow lines1on glacier, no feature correlation01*subsurf.
c9b−51.9453.0252.0852.03*−0.94somewhat follows wall, onto glacier2generally follows glacier flow lines12*subsurf.
c14−53.5351.6550.5750.53*−1.08partially follows glacier/wall contact1partially follows gulleys, small drainages44surf.
c12−44.7551.9050.1850.17*−1.72crosses features on valley wall1matches gulleys on glacier55surf.
c10a−52.7950.6649.9549.88*−0.71no correlation0matches general flow, wiggles in DEM44surf.
c15b−59.5744.6944.8144.44*0.11valley walls, no obvious correlation1parallels gully, follows wall/ice contact33surf.
c3−28.8259.9240.7640.76*−19.15mid 1/3 matches glacier/wall contact2some matches with gulleys on glacier12*subsurf.
c6−29.9031.1238.8431.12*7.72follows some small gulleys on glacier2no correlation02*subsurf.
c2−28.3825.5639.9725.56*14.41crosses wall contours, onto glacier1mid 1/3 matches gulleys on glacier22*subsurf.
c7−33.7823.8335.9023.83*12.07possible corr. w/gulley on glacier2valley wall, crosses topography02*subsurf.
c17a−59.4165.9022.4122.41*−43.50on glacier, parallel with flow lines1on glacier, crosses flow lines01*subsurf.
c10b−47.1628.0418.8218.82−9.21no correlation0valley wall and topography on glacier44surf.
b1−48.4636.1616.6916.69−19.47crosses contours, valley wall0follows lakeshore55surf.
a7−52.4816.0940.5416.0924.45may follow channel on valley floor2crosses multiple gulleys on valley wall02*surf.
c16−44.1424.2415.5915.59−8.65no correlation, surface of glacier0follows gully on glacier.44surf.
a13−39.9214.8524.2814.859.43lower valley wall, little correlation1valley wall, crosses contours01*surf.
c1−8.1037.1314.5414.54−22.59valley wall, good corr. w/glacier edge4almost follows drainage gully on glacier14surf.
a6−19.9612.1651.9512.1639.79first 1/3 colocated w/rough topog.3crosses contours on glacier03surf.
l1−11.5212.0411.6911.53−0.35crosses lakeshore, no match0follows lakeshore features, ridges in ice55surf.
c18−59.656.116.375.920.25match with linear feature on glacier5on glacier, crosses flow lines05surf.
a3−20.531.803.891.802.08channel on valley floor4valley wall, crosses contours04surf.
a2−29.64−5.7244.99−5.7250.71lineated ridge/hummock, valley floor5valley wall, crosses glacier, contours05surf.
a4−26.29−6.4528.94−6.4535.39valley floor and lake0follows contour of wall, parallel to ledge44surf.

4.4. Cross-Track Migration

[22] Cross-track migration was employed as a complementary technique to determine potential surface sources of echoes in the real radar data. This can further confirm an echo source determination based purely on comparisons of the real data with the simulated data. Alternatively, it can provide further insight into real echoes that do not correspond to anything significant in the simulated data. In this technique, radar echoes that have been identified are migrated in the cross-track direction to the surface locations from which the radar energy might have been reflected.

[23] Unfocused SAR processing results in an antenna pattern that is essentially in the plane perpendicular to the direction of travel, so we only considered topography in that plane. We assumed that all picked echoes are surface echoes. Since the range is known, the reflection point must lie on a circular arc with radius equal to that range, centered on the airplane and perpendicular to the direction of travel. At each aircraft position, we computed the points where the arc and the surface topography intersect. Owing to the left-right ambiguity, there are intersection points on each side of the aircraft. This is demonstrated graphically in Figure 8 for one aircraft position. This computation was performed for each radar record comprising all picked echoes along the flight track. The cross-track-migrated echoes were then superimposed on the DEM (Figure 1a) and on optical imagery (Figure 1b).

Figure 8.

Cross-track geometry using actual elevation data and aircraft position with radar antenna pattern depicted below aircraft (profile location shown in Figure 10). For each echo at that position, points of intersection with the surface at the appropriate range in air are determined.

4.5. Application of the Techniques

[24] Proceeding with each echo identified, we evaluated the superposition of each cross-track-migrated echo onto both the DEM and optical imagery. Candidate surface features for each echo were identified and described, if they existed. A rating of 0–5 was given for the perceived degree of correlation with surface features on each side of the aircraft. The results from this process and the criteria for the correlation rating are listed in Table 1.

[25] Combining these results with the forward simulation reveals that SCR values less than ∼20 dB correspond well to echoes with easily identifiable surface sources. The difference between the left and right SCR values are also good predictors for which side of the aircraft the surface echo originated from. For example, echo a2 (at 20 km in Figure 4b) has a left-right SCR difference of 50.71 dB (Table 1), strongly indicating that the echo is from the left side. This is confirmed by examining the cross-track-migrated echoes superimposed on the DEM (Figure 9). This shows that the left-sided echo matches topographic lineations on the valley floor, while the right-sided echo does not correlate with the topography, crossing contour lines on a glacier whose surface features do not match the echo superposition.

Figure 9.

Expanded view of DEM with cross-track-migrated echoes superimposed. Echo a2 shows a clear correlation with topography on the valley floor (left side of the aircraft) with a corresponding low signal-to-clutter ratio (SCR) value for that side (−5.72 dB; Table 1), while the right side, which has little correlation with surface features, has a high SCR value (44.99 dB; Table 1). Echo a6 is more difficult to correlate visually but also has a relatively low SCR value for the left side (12.16 dB versus 51.95 dB for the right side; Table 1), indicating that the echo arises from the surface on the left side of the aircraft. For context, see Figure 1.

[26] The most difficult part of the flight path in which to identify echo sources is the interval 7–12 km, which is close to the terminus of Taylor Glacier. The cross-track surface migrations are shown for this interval in Figure 10. A number of echoes exhibit similar patterns, and it appears that some echoes that were picked as contiguous are actually produced from multiple sources that happened to align. This may have resulted from a combination of the U-shaped morphology of the glacial valley, along-flow lineations (e.g., drainage gulleys) on the surface of the glacier, and the flight path causing a variety of surfaces and features to lie at similar ranges in this interval of the flight.

Figure 10.

Expanded view of terminus region of Taylor Glacier with flight path (single red line, flight direction indicated) and cross-track-migrated echoes (multiple colors) depicted on (a) surface topography and (b) optical imagery. The complexity of possible surface echo sources is evident, especially over the glacier, where drainage gulleys are at similar ranges to valley walls on the opposite side. For context, see Figure 1. Profile A–A′ is depicted in Figure 8.

4.6. Results

[27] The most conservative approach for classifying echoes as arising from the subsurface would require both a high SCR value (i.e., no strong echo in the simulated data for that position and time delay) and little or no visual correspondence of cross-track-migrated echoes with surface features. Using these criteria with the empirically derived threshold of 20 dB for SCR and maximum cross-track migration correlation ratings of 0–2 (Table 1), 11 of the 28 identified echoes could result from subsurface reflectors. These are identified in Table 1 and plotted on the radar data in Figure 11. In order to verify the subsurface reflector known to exist at the base of Taylor Glacier (an ice-rock or ice-water interface), we compared our results to the those of surface-based ice penetrating radar studies undertaken on Taylor Glacier. In our 24 km test segment, three surface transects roughly perpendicular to the flight path were collected. The surface-based radar data showed a clear subice reflection below the glacier, and the locations of these echoes correspond to some of our candidate subsurface echoes (Figure 11).

Figure 11.

Final results showing the most likely subsurface echoes in this test segment. White diamonds indicate the position of the bed interface below Taylor Glacier, as determined from ice-penetrating radar studies conducted on the ice surface. The steeply concave-down echoes at approximately 7 km (c4 and c5) are likely to be off-nadir subsurface echoes (see text for discussion).

[28] The steeply curved, concave down echoes underneath Taylor Glacier (c4a and c5, Figure 11) are likely due to off-nadir topography beneath the glacier, perhaps a continuation of a small ridge emanating from the valley wall on the north side of the glacier (Figures 1a and 1b). This interpretation is supported by the surface-based radar control point at 8 km. Some echoes (i.e., c10a, c12, c14, and c15b) have good visual correlations between their cross-track-migrated echoes and surface features (specifically, drainage gulleys on the glacier surface and the intersection of the valley wall with the glacier surface), yet have SCR values in the 40–50 dB range indicating a subsurface origin. Because of these strong visual correlations it is unlikely they have a subsurface origin; however, the simulator does not produce significant echoes corresponding to these surface features.

[29] In the remaining part of the test segment, there are no convincing subsurface reflectors. The only possibilities, based on lack of correlation with surface features in the cross-track migrations, are echoes a13 and a7 (18 km and 22 km, respectively (Figure 4b)). The flight path in these locations is over dry ground: the valley floor just east of Lake Bonney for a13, and a ridge on the north side of the valley for a7 (Figure 1). Because both of these echoes have very low left-sided SCR values, it is likely that subtle features on the valley floor are responsible for the reflections.

[30] The net result of our echo discrimination process (Figure 11) is a set of echoes that collectively defines a sloping interface below Taylor Glacier, as expected. Although this process has reduced the number of potential subsurface echoes considerably (comparing Figure 11 with Figure 4b), some questions remain. For example, the surface control point at 8 km aligns with echo c9b, indicating that echo c15a, directly below c9b in the radar record, results from an unidentified surface feature. The remaining echoes near the terminus (c2, c3, c6, and c7) could also be surface reflections or perhaps reflections from ice cored layers of lacustrine sediments such as those observed at the base of the Taylor Glacier terminus and within thrust moraines along the front margins of the glacier [Higgins et al., 2000]. This complexity demonstrates the difficulty of fully resolving the source of every echo in this type of environment using single-pass radar data.

5. Implications for Mars Orbital Sounding

[31] The current emphasis on locating reservoirs of both solid and liquid water in the subsurface of Mars has led planners of current and upcoming missions to make use of radar sounding to begin exploration of the subsurface from orbit. The Mars Advanced Radar for Subsurface and Ionosphere Sounding (MARSIS) instrument is currently orbiting Mars on the Mars Express spacecraft. Operating with center frequencies of 1.8–5 MHz (60–167 m wavelength), this long-wavelength sounder is expected to probe on the order of a few kilometers into the subsurface [Picardi et al., 2004]. MARSIS employs a secondary, nadir-pointing monopole antenna to help identify off-nadir surface echoes viewed by the primary antenna [Picardi et al., 2004]. Launched in August of 2005 with an estimated mission start in November 2006, the Mars Reconnaissance Orbiter carries the Shallow Radar (SHARAD) instrument, a radar sounder with a 20 MHz center frequency (15 m wavelength) and 10 MHz bandwidth. SHARAD is expected to be capable of detecting dielectric interfaces to a few hundred meters depth [Seu et al., 2004]. It is hoped that data from these two sounders will provide the context needed to plan subsequent missions for exploring and eventually sampling the subsurface using landed spacecraft. Brines or other conductive materials would of course inhibit any radar's ability to sound the subsurface, as may be the case for the ice-free portion of our Taylor Valley test segment where no positive subsurface reflectors were identified (although this could also be due to a lack of any significant dielectric discontinuities in the subsurface). However, recent findings from Mars Global Surveyor and Mars Express indicate the importance of near-surface massive ice (as opposed to dispersed ice as in permafrost) in the form of rock glaciers [Head and Marchant, 2003], ice-rich floes and debris aprons [Head et al., 2005]. Possible ice deposits at high elevations on Olympus Mons [Neukum et al., 2004] provide evidence for glaciations at low latitudes within the past few million years due to large changes in Mars's obliquity [Laskar and Robutel, 1993; Head et al., 2003]. On the basis of surface radar studies of analogous features on Earth such as rock glaciers [e.g., Berthling et al., 2000; Degenhardt and Giardino, 2003] it is quite possible that radar sounding could confirm the existence of ice in such features on Mars; however, given their likely thicknesses of a few tens to a few hundred meters, surface clutter removal will be a crucial step in interpreting orbital or airborne radar sounding data acquired over such features.

[32] To prepare for upcoming radar sounding on Mars, models have been developed to predict the effects of surface roughness on MARSIS data [Plaut et al., 2001; Picardi et al., 2004] and it appears that MOLA-derived topography with its 500 meter postings will be sufficient for estimating surface clutter over most of the planet. Nouvel et al. [2004] have developed a radar signal simulator for this purpose using a facet model based on the assumption that topography is relatively smooth at the wavelength of the radar. Their facets are therefore defined to be larger than the radar wavelength. This assumption may be sufficient for MARSIS; however, surface topography is likely to be rough at the scale of the radar wavelength of SHARAD. The Nouvel et al. [2004] approach also coherently integrates the responses from neighboring facets, whereas our model sums all returned energy incoherently. Since the surface of Taylor Valley is assumed to be rough at the wavelength of our radar, and our facet size is less than the radar wavelength, the integration over many facets will statistically produce a diffusely scattered result. Thus incoherent integration should be adequate for simulating radar data for Taylor Valley and for similar conditions on Mars.

[33] It should be noted that the resolution requirement of a DEM is not as demanding for cross-track migration as for the forward simulation of radar data, since long-wavelength topography generally determines the range (and hence, surface location) of a given echo. Benham and Dowdeswell [2003] demonstrated that the assumption of a flat surface can yield useful results for this technique under certain conditions; however, we found that assumption to be inadequate for our studies in the Dry Valleys. MOLA-derived DEMs are insufficient for detailed surface clutter prediction for SHARAD, but they should be sufficient for performing cross-track migration of echoes in the radar data for superposition onto other imagery. As we have demonstrated, this can be a very useful technique that is complementary to, and largely independent of, surface clutter simulation.

6. Conclusions

[34] It is evident that for surface clutter removal in data acquired by a radar sounder moving above the surface, whether airborne or orbital, the use of multiple techniques should be employed to reduce the uncertainties inherent to using a single method. This study used the forward simulation of surface clutter and the visual correlation of cross-track-migrated echoes with multiple representations of the surface. Each method has its own merits but neither is sufficient alone, and the combination is not a complete solution for all cases examined here. For Mars orbital data, it is clear that the accurate identification and characterization of subsurface features will depend on the availability of a topographic model of the surface at the appropriate wavelength along with other high-resolution imagery of the surface. In order to confirm questionable subsurface reflectors, or to identify them without the use of a topographic model, techniques using multiple passes over the same target (first attempted on Lunar Sounder data [Peeples et al., 1978]), will be required and need further development. Such techniques also have the potential to positively identify off-nadir subsurface echo sources, a problem that cannot be addressed in single-pass techniques such as those described here.

Acknowledgments

[35] This work was supported by NASA grant NAG5-12693 and the John A. and Katherine G. Jackson School of Geosciences, University of Texas at Austin. Data acquisition was supported by NSF grants OPP-9814816, OPP-9319379, and OPP-0126202. Thanks to Beata Csatho for early access to the Dry Valleys DEM and to Anatoliy Mironov and Jody Sturdy for assistance with field operations. Comments and suggestions from an anonymous reviewer were much appreciated. This is University of Texas Institute for Geophysics contribution 1761.

Ancillary