Theoretical estimates of low-frequency radar sounding performance and its potential for mapping moist subsurface interfaces in conductive environments on Mars are controversial, with predictions of ultimate penetration depth ranging from a few meters to kilometers. To address this issue, we conducted a broadband electromagnetic field survey in which we combined ground penetrating radar (GPR) operating at multiple low frequencies with the transient electromagnetic method (TEM) to investigate the dependence of radar penetration depth on ground resistivity. Surveys were performed in the frequency range 16–100 MHz at two locations on the northwest margin of the Amargosa Desert, Nevada, where numerous Mars-analog investigations have been performed. The surveys were conducted on a 20-m-high homogenous sand dune and on the flanks of a 20-m-high scoria cone and above a buried lava flow. A wet alluvial interface was located at the bottom of each structure. GPR detected the wet alluvium contact at the base of the sand dune, but failed to penetrate to the same depth at the scoria cone under similar residual moisture content. Depths of investigation for both the scoria cone and the buried lava flow were limited to approximately 10 m owing to the presence of conductive inclusions in the first few meters, which are below the radar resolution but dramatically decreased the dynamic of the radar-backscattered echoes and hence the penetration depth. Absorption models constrained by the TEM data are in good agreement with these observations. Depths of investigation varied weakly with frequency owing to substantial, frequency-independent absorption.
 One of NASA's major science objectives for the Mars Exploration Program is to identify the three-dimensional distribution and state of subsurface water on Mars (Mars Exploration Program Analysis Group, Scientific goals, objectives, investigations, and priorities: 2003, unpublished document, 22 pp., posted April 2004 at http://mepag.jpl.nasa.gov/reports/index.html). Water has been identified as a cross-cutting theme of all goals and the common thread of Mars research, its abundance and distribution having important implications for understanding the geologic, hydrologic, and climatic evolution of the planet, the potential origin and continued survival of life, and the accessibility of critical in situ resources for sustaining future human explorers (G. J. Taylor et al., Mars Exploration Program Analysis Group's scientific goals, objectives, investigations, and priorities, 2004, available at http://mepag.jpl.nasa.gov/reports/ind).
 It has been widely suggested during the last decade that one of the technologies that may help to identify the current distribution of subsurface water is low-frequency sounding radar (defined here as the frequencies below 100 MHz), which is well adapted to the mass and power constraints of planetary missions [Barbin et al., 1993; Berthelier et al., 2003]. During summer 2005, the Mars Advanced Radar for Subsurface and Ionospheric Sounding (MARSIS), aboard the European Space Agency's Mars Express spacecraft, began acquiring the first of what may eventually be a global data set of Martian subsurface radar properties within the low-frequency band 1.3–5.5 MHz [Picardi et al., 2004]. Resulting radar data will be used to the extent practicable to assess the three-dimensional distribution of possible subsurface liquid water by its dielectric contrast with dry soil and rock, contrasts that are also potentially indicative of variations in mineralogical composition, lithology, stratigraphy, and internal structure. MARSIS will be joined by the SHAllow RADar (SHARAD) instrument onboard NASA's Mars Reconnaissance Orbiter in 2005 [Ori et al., 2002; Phillips et al., 2005]. Like MARSIS, the primary mission of SHARAD is to search for evidence of subsurface liquid water, but at shallower depth and higher resolution, a capability based on SHARAD's higher 20-MHz central operating frequency, 10-MHz bandwidth, and better dynamic range [Ori et al., 2002; Phillips et al., 2005]. Because no prototypes of the MARSIS or SHARAD radars were field tested and ground-penetrating radar (GPR) investigations of the Earth's dry and volcanic areas are not common within this frequency range, estimates of the likely performance of both sounders are based mainly on modeling and simulation using a basic full-propagation model (i.e., no scattering contribution to total estimated attenuation) and simplified geoelectrical models that assume a homogenous, low loss, parallel and stratified subsurface. On Earth, these ideal characteristics are rarely observed.
 At low frequencies (i.e., below 100 MHz) loss mechanisms in volcanic rocks may include Maxwell-Wagner-Sillars and magnetic permeability relaxation loss mechanisms as well as standard conductivity losses due to the presence of conductive inclusions such as magnetite. In an effort to better understand those losses and determine the implications of electrical resistivity on low-frequency sounding radar for Martian volcanic areas, we combined GPR and transient electromagnetic (TEM) surveys at three semiarid sites located on the northwest margin of the Amargosa Desert, Nevada. The southwestern deserts of the United States have been the subject of several Mars-analog investigations [Farr, 2004], including synthetic aperture radar and GPR [Grant et al., 2001; Campbell et al., 2002] at higher frequencies. Our particular test area was selected because its geologic and hydrologic properties have been studied in detail as part of a regional technical evaluation of Yucca Mountain, Nevada, related to its potential for serving as a geologic repository for high-level radioactive waste and spent nuclear fuel. The area is well characterized both geologically [Peterson et al., 1995; Faulds et al., 1994; Swadley and Carr, 1987; Carr and Parrish, 1985; Carr, 1982] and geophysically [Farrell and Sandberg, 2003; La Femina et al., 2002; Langenheim, 2000; Brocher et al., 1998] and hence offers a unique opportunity to understand loss mechanisms that are due to subsurface geological and geophysical complexity.
 GPR soundings in the frequency range of 16–100 MHz were conducted to assess the performance and limitations of the method and to identify complimentary geophysical investigations for unambiguous identification of moist interfaces in the near surface. It is important to keep in mind that the GPR penetration ability represent an ideal case compared to orbital radar survey where the transmitted power to the subsurface and the signal-to-noise ratio is lower owing to the difference between both sounding geometries and instrument configurations. TEM was used to determine the subsurface electrical resistivity and its impact on the penetration depth of the low-frequency sounding radar. The combined data set provides an improved understanding of both dielectric and conductive loss mechanisms resulting from the interaction between the electromagnetic radar pulses and the subsurface. In what follows we provide a brief description of the technique, test sites, results and implications for Martian subsurface exploration using low-frequency sounding radars.
 We selected three sites for investigation in the vicinity of Yucca Mountain, Nevada: (1) sand dunes overlying alluvial sediments at Big Dune, Amargosa Desert, Nevada; (2) the flank of Northeast Little Cone, a scoria cone in Crater Flat, Amargosa Desert, Nevada; and (3) an alluvium-basalt flow-alluvium sequence adjacent to Southwest Little Cone in Crater Flat, Amargosa Desert, Nevada (Figure 1).
 The majority of annual precipitation falls during the winter months in the Amargosa Desert. Our field campaign was conducted during March 2004, following the winter precipitation events. Samples representing the major geologic units were collected from each site for dielectric characterization in the laboratory over the frequency range of 1–100 MHz (Table 1) to support GPR data analysis. Samples were kept in isothermal sealed packs to maintain their residual moisture level before and during laboratory measurements. As discussed in section 5, laboratory measurements combined with electrical resistivity measurements in the field enabled us to quantify signal loss and radar penetration depth. We briefly describe the two complementary geophysical field methods used in this study next.
Table 1. Laboratory Electromagnetic Characterization of the Three Samples Representing Major Geological Units From the Study Sitesa
Big Dune Sand
Little Cone Basalt
The real and imaginary parts of the dielectric constant (ɛ′ and ɛ″) and the loss tangent (tgδ) are shown for the central frequencies, 20, 40, and 100 MHz.
2.1. Ground-Penetrating Radar
 GPR uses the propagation of electromagnetic pulses to profile the structural elements of the subsurface; as a result, interpretations and mappings of the geoelectrical properties can be made. For low frequencies below 100 MHz, the depth of investigation is determined principally by the choice of the sounding frequency, subsurface resistivity and the degree of heterogeneity of stratigraphic layers. Hence the detection of a shallow wet interface using low-frequency GPR is constrained by the total signal loss, which is a convolution of the effects of spherical spreading, antenna-ground coupling, scattering, electrical resistivity and dielectric contrast between the relatively dry and water-saturated rock at depth. The first two factors are readily calculated and observed from knowledge of the sounding geometry and the signal dynamic. Measurements of electrical resistivity and dielectric characteristics are necessary, however, in order to deconvolve the effect of attenuation from the remaining signal losses (which included the sum of attenuation and volume scattering) and hence get an assessment of the losses that are due to volume scattering.
 Previous planetary GPR field surveys have focused primarily on the potential use of radar for mapping the shallow subsurface in Mars analog environments, especially in the frequency range of 400–900 MHz [Paillou et al., 2001; Grant et al., 2004]. Results in most cases have been extrapolated to the 2- to 20-MHz band demonstrating a proof of concept for shallow subsurface sounding and suggesting a possible limitation in penetration depth in iron oxide rich soils. Previous low-frequency surveys conducted by Arcone et al.  in Alaska and in the Antarctic Dry Valleys and by Olhoeft et al.  on lava flows highlight the complexity of loss and diffraction mechanisms at lower central frequencies in the 50–100 MHz range. In most ground-penetrating radar studies, GPR has been used as a standalone investigative tool in relatively resistive environments. In our study we investigate radar performance in conductive soils, complemented by TEM to help quantify geoelectrical signal loss.
 Our GPR survey was performed using the Geophysical Survey Systems Inc. (GSSI) SIR 2000 radar unit and two antenna systems that covered the frequency range of 16–100 MHz. One system consisted of a pair of bistatic, shielded 100-MHz antennas with an effective radiated power of ∼40 W that was used to acquire data in continuous mode. The other antenna system was a Multiple Low-Frequency (MLF) system used to acquire point-to-point sounding shots with 512 to 1024 stacks at each point with a sampling rate of 2048 points per trace, allowing us to achieve good noise reduction. The MLF system is composed of two separate, parallel, unshielded dipoles whose length can be adjusted to emit signals at central frequencies of 16, 20, 40, and 80 MHz with a 3-dB bandwidth nearly equal to 35% of the value of each central frequency. The effective radiated power of this system is respectively 400, 333, 167 and 80 mW. Both antenna systems were coupled to the surface during data acquisition and configured to achieve maximum penetration and signal-to-noise ratio (maximum stacking, data sampling, and time range). To minimize the effects of surface clutter, all soundings were conducted in open areas to avoid potential surface reflections that could give rise to a backscattered signal in the same time range as the anticipated subsurface echoes.
2.2. Transient Electromagnetic Method
 In TEM soundings, a large transmitter loop is used to induce eddy currents in the ground, which induce secondary magnetic fields that are detected by a smaller receiver loop (see McNeill  and Hoekstra and Blohm  for reviews). These inducting fields penetrate (diffuse) to much greater depths than do propagative waves like GPR, albeit at lower resolution. While responding strongly to resistivity, TEM is insensitive to dielectric contrasts. In the other hand resistivity measurements help in assessing the loss tangent in the subsurface [see Hartshorn, 1927]. The full mathematical is given by Grimm et al. .
 TEM soundings were conducted with a Geonics PROTEM 47 transmitter and 47D receiver. Transmitter loop sizes of approximately 15, 20, 50, and 100 m were used with an effective receiver coil area of 31.4 m2. Current was maintained at 2 amps for all transmitter loops at all stations, and one or more of the pulse repetition rates 285, 75, and 30 Hz were used for each survey. Data were acquired with the receiver coil at the approximate center of the transmitter loop for all loop sizes; for the 15- and 50-m loops, the receiver coil was also placed 10 m outside the transmitter loop. This latter configuration was used as a precaution to eliminate induced polarization (capacitive) effects that can sometimes distort small-loop central-receiver soundings, but we observed few such effects.
 The relatively resistive quartz-rich sands of Big Dune (Figures 1 and 2a) were selected for initial calibration studies because the low moisture content, high porosity, and homogeneity of eolian sands tend to reduce both absorption and scattering losses. Big Dune, a complex assemblage of barchan, transverse, and star sand dunes with crests up to 80 m in height, is located in the Amargosa Desert, Nevada, where local surface winds vary in both direction and intensity. Big Dune overlies relatively moist valley-fill alluvium. Two-dimensional ground magnetic surveys at Big Dune [Hill et al., 2002] indicate bedrock composed of either volcanic tuff or basalt at depths of approximately 100–200 m. This depth estimate is based on the shape and amplitude of the resulting magnetic anomaly. The sand dune we studied had a moist sand/alluvium interface at its base, and whose depth varied from ∼5 to 20 m below the portions of the dune surface profiled by GPR. The dune has a steep lee face and a gentle windward slope approximately 100 m long, a geometry enabling determination of the depth of the alluvium/sand interface.
 Our first radar survey at Big Dune was performed starting at the crest of the dune (∼20 m above the underlying alluvium interface) and moved downslope toward its base (∼5 m above the alluvium) using the 100-MHz bistatic antennas (Figures 2a and 2b). A second and third survey were performed using the 20- and 40-MHz antennas and point-to-point soundings along the same transect at 5-m intervals. The 20-MHz trace provides an indication of the type of subsurface data that might be obtained from SHARAD (Figure 2c).
 The first pair of TEM soundings was conducted ∼500 m away off-horizon from the GPR measurement point (Figure 2a) to permit simultaneous acquisition of TEM and GPR without significant electromagnetic interference. The first station, On-Dune, was approximately 25 m above the sand-alluvium contact and was probed with 20- and 50-m transmitter loops. The second station, Off-Dune, was on the alluvium at the base of the dune, and only a 20-m loop was used.
3.2. Little Cones, Crater Flat, Amargosa Desert, Nevada
 The Little Cones site is located approximately 12 km north of Big Dune (Figures 1 and 3). Little Cones comprise the southernmost two of five exposed scoria cones in Crater Flat (Figure 3a). Crater Flat is a structural half graben west of Yucca Mountain, Nevada. Prior magnetometer surveys by Stamatakos et al.  (Figure 3b) revealed the presence of buried lava flows to the south and southeast of Little Cones. Relative to the other three contemporaneous scoria cones, the Little Cones have much less surface exposure, indicating a southern and western shift in deposition of alluvium during the Quaternary [Stamatakos et al., 1997] such that alluvial deposits overlay the distal portion of the lava flows. Southwest Little Cone is breached on its southern flank, and lava flows extend more than 500 m south of the cone [Stamatakos et al., 1997]. The depth of burial at the distal extent of the flow is estimated at 5–20 m, with flow thickness of 10–20 m.
 Two volcanic structures were studied using the GPR at this site. The first structure was a buried lava flow from Southwest Little Cone. The 100-MHz bistatic antenna system was used to perform soundings along a transect crossing the lateral boundary of the buried basalt flow at its distal extent into an alluvium-only deposit (Figure 3b). Two point locations were also sounded with the MLF antenna: one over the buried basalt flow and the other over the alluvium alone. MLF soundings were conducted at 20, 40 and 80 MHz (Figure 4) to compare the different backscattered echoes and signal attenuation. The second volcanic structure was Northeast Little Cone itself (Figure 3a). A 100-MHz continuous profile was performed on a road leading to the summit of the cone. The summit of Northeast Little Cone stands approximately 20 m above the surrounding terrain. The locations of the continuous radar transects and the two MLF soundings are indicated in Figure 3.
 TEM soundings were conducted both on and off the buried flow southeast of Southwest Little Cone. The On-Flow TEM station was located approximately 100 m north of Old Stagecoach Road, at a position known to be over the buried flow (Figure 3). The Off-Flow station was located approximately 400 m south of Old Stagecoach Road at a position known to be beyond the extent of the buried flow, on the basis of a prior magnetic survey (Figure 3b). At the On-Flow station, 15- and 50-m transmitter loops were used, whereas these sizes and a 100-m loop were used at the Off-Flow station.
 The principal target for our data acquisition and interpretation was the identification of the wet-alluvium interface with the dry sand above (Big Dune) or basalt below (Southwest Little Cone). An assessment of the soil dielectric properties is crucial for converting the radar range (echo delays expressed in nanoseconds) into depths. This transformation for Figures 2b, 2c, 4, and 5is based on laboratory measurements of the dielectric constant of field samples collected at each location. The complex permittivities of the four field-collected samples of the major geological units at each test site are summarized in Table 1. The laboratory measured permittivities show that we have significantly high loss tangents (tgδ) in the Little Cones basalt compared to the Big Dune sands, but both have similar residual moisture contents of approximately 6%. The moist alluvium (with saturation of approximately 35%) exhibits an important dielectric contrast with both the Big Dune sand and the Little Cones basalt, but also exhibits higher losses that significantly constrained the ability to perform deep soundings.
4.1. Big Dune
 On this site the GPR data sets illustrate different penetration depths and resolutions obtained using different sounding frequencies. The radar investigation successfully identified the sand/alluvium contact located under the dune formation as shown in Figure 2a. The topographically corrected continuous radar profile (Figure 2b) was generated using the 100-MHz shielded bistatic antenna. We used an 80-MHz low-pass filter to reduce radio communication interference present in the survey area. Direct signal and surface coupling was partially removed using a background average subtraction filter. In Figure 2b, the dune crest is at the upper left (at x = 0), and the horizontal reflector (at 4 to 17 m below the surface, and between x = 10 and 65 m laterally) is the interface between dry sand above and moist alluvium below. A buried earlier stage of the sand dune is also seen in a sloping reflector between x = 15 and 25 m. A bright reflector at a depth of approximately 2 m may be an echo from the surface, or just below the surface, that indicates higher moisture content from recent winter precipitation.
 A monostatic point-to-point 20 MHz radar profile along the same transect (Figure 2c) was obtained using the 20-MHz MLF antenna configuration. The profile is composed of one-dimensional radar soundings stacked 512 times and normalized to the first maximum reflection (corresponding to the surface coupling). The data were acquired at 5-m intervals along the profile. The red line is the surface reflection and the blue line indicates where the sand/alluvium interface is inferred to occur. The longer wavelength of the 20-MHz investigation resulted in poorer resolution in identifying the position of the reflectors, ambiguity that was further increased by the fact that the depth of the reflector came within one in-ground wavelength (15m/) of the antenna. Beneath the sand/alluvium interface, the signal was strongly attenuated owing to the high loss tangent of the moist alluvium; thus no clear reflections were identified below this depth.
 Inversions for the On- and Off-Dune TEM sounding curves were generally smooth and without obvious strong interface contrasts, including the water table. The mean resistivity of the dune was approximately 600 Ω-m, and the resistivity fell to a mean of approximately 60 Ω-m in the uppermost alluvium. The low electrical resistivity caused significant attenuation of the radar signal in the moist alluvium, dramatically decreasing the radar penetration depth within this layer to just a few meters, and making it difficult to distinguish any features beyond the contact. We treat this in more detail below in section 5.
 At approximately 35 m below the alluvium surface the resistivity is just 6–8 Ω-m, and at 65–75 m depth the resistivity is a remarkably low 1–2 Ω-m. McNeill  gives a heuristic for salinity of pore water in saturated unconsolidated materials as TDS = 25,000/ρ, where TDS is the total dissolved solids in mg/L (ppm) and ρ is the resistivity in Ω-m. The Big Dune alluvium at depths of several tens of meters, therefore, likely contains brackish pore water with TDS less than 1%.
4.2. Little Cones, Crater Flat
 Both monostatic MLF soundings and 100-MHz continuous GPR profiles from this location (Figures 4, 5a, and 5b, respectively) indicate a limited penetration depth of a few meters. The monostatic MLF soundings in Figure 4 correspond to three pairs of 20-, 40- and 80-MHz soundings, illustrating the difference in returns obtained between the alluvium alone (at left) and the alluvium over basalt (at right). These sounding comparisons suggest it would have been possible to distinguish the shallowly buried basalt at ∼8 m depth when sounding above the alluvium sediments using the 40- and 80-MHz antennas. The low resolution and the shallow location of the interface made it difficult to identify this feature in the 20-MHz sounding. Figure 4 also shows that, below 12 m, signals were severely attenuated at both locations owing to the presence of inclusions of iron oxides and residual moisture. Using the 100-MHz antennas, signal penetration was limited to approximately 4–6 m in the alluvium that is located above a buried basalt flow from Southwest Little Cone (Figure 5a), and was limited to only 3 m on the road leading from the summit of Northeast Little Cone (Figure 5b). Minor magnetite and hematite inclusions in the alluvium, and recent winter precipitation, have resulted in the depression of the electrical resistivity of the near surface, but penetration into the tephra of Northeast Little Cone was mainly limited by the presence of moderate iron content. Iron oxide as total FeO is reported to be present in the range of 10.8–11.6 weight percent in the alluvium covering the basalt Crater Flat area [Vaniman et al., 1982].
 TEM soundings provide insight into the shallow electrical resistivity, which affects signal absorption. Near-surface resistivities in the alluvium near Little Cones averaged ∼300 Ω-m. These resistivity values result in intermediate near-surface losses compared to the Big Dune sand and near-surface alluvium. At depths of 15–20 m, however, resistivities fell to tens of ohm-meters, and therefore are similar to the Big Dune alluvium at similar depths. The buried flow was not unambiguously detected in TEM, suggesting that conductive pore water infiltrates all units comparably. In both On- and Off-Flow soundings, a strong conductor was detected at a depth of 50–65 m, which agrees with other estimates of the regional depth to the water table [Farrell and Sandberg, 2003]. Resistivities <3 Ω-m are also representative of comparable depths below Big Dune and again indicate brackish to saline groundwater, but here are more likely identified with the phreatic zone given borehole information in the vicinity.
5. Absorption Analysis
 The GPR depths of investigation for which complementary TEM data exist may be summarized as follows: (1) greater than 18 m for Big Dune sand at 20 and 100 MHz; (2) less than 2 m for Big Dune alluvium beneath sand at 20 and 100 MHz; (3) 9–12 m for Little Cones alluvium at 20–80 MHz. The last figures reflect the modestly increasing depth of investigation with decreasing frequency. The relevant TEM-determined shallow resistivities are roughly 600, 60, and 300 Ω-m, respectively.
 If iron minerals or small quantities of electrically conductive pore water dominate the absorption at all frequencies, the TEM average resistivity can be directly used as a “DC” contribution. The absorption is computed as dB/m = 8.686β, where β is the imaginary part of the propagation constant [e.g., Stratton, 1941]. For the frequencies cited above, this leads to estimated absorptions of 1 dB/m for Big Dune sand, 8–10 dB/m for Big Dune alluvium, and 2 dB/m for Little Cones alluvium. The assumption of a DC resistivity implies that frequency dependence of absorption is weak in the propagative regime.
 The validity of these absorption estimates can be assessed with a radar-range analysis [e.g., Annan, 2003; Davis and Annan, 1989]. Because we cannot accurately specify all of the system and antenna parameters for an end-to-end figure-of-merit, we simply compute the effective dynamic range available for ground losses, i.e., spreading + absorption + backscatter loss, which is consistent with the observed depth of investigation and TEM-computed absorption. We assume a dielectric contrast of 2 at the target interface and that the signal is coherently reflected over the area of the first Fresnel zone. The MLF system (20–80 MHz) typically used 512 stacks or integrations per station, whereas the continuously profiled 100-MHz soundings were assumed to have an effective averaging length of 4 traces. We compute an effective ground dynamic range of 85–75 dB at 20–80 MHz, respectively, for the Little Cones alluvium. This compares well with a figure of greater than 85 dB for the Big Dune sand at 20 MHz. The 100-MHz data, however, indicate an effective ground dynamic range of greater than 110 dB, likely due to the higher signal-to-noise ratio for the shielded antennas.
 These computed effective ground dynamic radar ranges are comparable to other investigations and to other GPR systems, confirming that the TEM results may be used directly at this site to estimate GPR absorption. This approach provides only a lower bound to absorption expressed as the conductive component, however, because dielectric losses from the Debye relaxation of water can also contribute a few dB/m in this frequency range. This would require an additional 30–60 dB of dynamic range, beyond what is reasonable for these commercial systems. We therefore conclude that the water abundance is sufficient to promote strong electrolytic conduction without seeing a large contribution from dielectric relaxation. Iron minerals in the Little Cones region can also contribute to “dry” losses. With few internal structures observed in the GPR profiles, we also conclude that scattering losses are negligible at these sites.
 Note that the laboratory-measured loss tangents in Table 1 are very small, leading to absorptions of less than 1 dB/m at the frequencies investigated. While the real components were useful for computing subsurface velocities in the absence of discrete scatterers in the GPR sections, the surface-gathered samples are not fully representative of the moisture and salinity at greater depths that control GPR absorption.
 Finally, we note that the weak dependence of depth of investigation versus frequency was also adequately modeled with the simple radar-range equations [see Davis and Annan, 1989]. In the absence of attenuation, the depth of investigation for a planar reflector varies inversely with frequency (whether the entire plane or just the first Fresnel zone is integrated). The logarithmic form (dB/m) of attenuation must change proportionately to frequency to maintain the inverse-frequency dependence on depth. This translates to a constant (frequency-independent) loss tangent, as well as loss tangents that are sufficiently small so that the available dynamic range is not exhausted. The sites investigated here are closer to constant resistivity and attenuation and, therefore, produce nearly constant depth of investigation over the frequency band tested.
6. Implications for Shallow Soundings on Mars in a Conductive Subsurface Environment
 Our penetration depths were limited and nearly independent of frequency, suggesting substantial, frequency-independent absorption. TEM surveys independently measured the electrical resistivity, and absorption models constructed from these data predicted the observed depths of investigation using reasonable dynamic range constraints.
 While not representative of the global geological and environmental conditions on Mars, these sites may, however, be analogous to areas where the regolith exhibits significant conductivity due to the presence of moderate to high concentrations of iron oxides, such as hematite [Christensen et al., 2000, 2001], maghemite [Hargraves et al., 2000], and magnetite [Madsen et al., 1999; Bertelsen et al., 2004]. Such environments may pose significant challenges for GPR, and need to be explored to provide a realistic understanding of the performance of low-frequency sounders under less than optimal conditions.
 The presence of brackish groundwater at depths of a few tens of meters in this study may also be more geoelectrically analogous to the subcryosphere of Mars, or portions of the cryosphere where it is warm enough for thin films of unfrozen capillary water to exist [Grimm, 2002]. If a sharp water table below a thin capillary fringe exists [Carr, 1986, 1996], this interface may be detectable by low-frequency sounding radar operating with a higher dynamic range than what is currently suggested in radar sounding experiments. A thick capillary fringe overlying a water table will yield a gradient loss with depth that will significantly attenuate the radar signal [Simpson, 1976; Heggy et al., 2001] and would require auxiliary information before the presence of a water table could reasonably be interpreted. In either case, GPR signals will not penetrate brackish to saline aquifers on Mars, so three-dimensional groundwater mapping with low-frequency sounding radar may be difficult to achieve [Grimm, 2003].
 Our results also highlight challenges in selecting Mars analogs for deep GPR investigations. To find very low absorption (less than 1 dB/m) more representative of the Martian cryosphere, site evaluations will have to be made on the basis of geoelectrical, as well as geological, compatibility [Dinwiddie et al., 2005]. Of particular interest are those environments with fresher groundwater (or little to no water), deep water tables (∼100–1000 m) and little scattering, which we plan to investigate with sounding radars operating in the frequency band of 1–100 MHz. While GPR surveys help us improve our understanding of the losses and estimate a more realistic penetration depth, they do not totally simulate the case of an orbiter sounder at the same frequency; surface reflection and clutter increase considerably the complexity in data analysis and subsurface structure identification. Additionally, the expected geological and geophysical complexity of the Martian surface and subsurface makes it difficult to draw a simple conclusion from this study on the usefulness and performance of radar sounding techniques to globally map shallow moist interfaces on Mars. Further field surveys are needed to enable statistical studies and a more in-depth understanding of the wave and subsurface material interaction at low frequency. In the Yucca Mountain region, GPR sounding performance at 20 MHz can range from negligible penetration to as much as tens of meters at sites separated by relatively short distances. Radar sounding performance was shown to be highly dependent on the choice of frequency to minimize signal loss and maximize penetration depth. The optimal frequency varied according to the subsurface geoelectrical properties of each site. Our results show that, in conductive soils, going to a lower frequency does not necessarily result in better penetration. Hence orbital sounding radar operating on a very narrow frequency band may not be well suited for conducting both shallow and deep subsurface mapping if a significant part of the Martian surface is covered with lossy material [Heggy et al., 2001].
 Data returned from the MARSIS and SHARAD instruments should provide valuable information, however, about the extent to which high-loss materials will limit future radar sounding investigations (whether conducted from orbit or from the surface). It appears likely that shallow local variations in material properties will also greatly constrain the successful identification of water at depth. Our work in the Amargosa Desert suggests that the ambiguities associated with identification of water will be greatly reduced by use of complementary geophysical techniques, such as GPR and TEM. Hence a potential TEM experiment performed on Mars would be of a major support to both MARSIS and SHARAD data interpretations.
 Our goal was to assess the efficacy of using low- to moderate-frequency GPR radar to investigate the presence and distribution of water in the shallow subsurface of a semiarid yet electrically conductive Mars analog environment. The two sites we selected for study were located within a well-characterized region near Yucca Mountain, in the northwestern portion of the Amargosa Desert, Nevada. One principal finding of this work is the utility of combining broadband GPR and TEM to assess the extent to which conductive losses contributed to the shallow penetration achieved by the radar in a Mars analog environment. Another principal finding is the role of the first few meters in determining the ultimate penetration depth. Results from the Little Cones study show that the presence of conductive material in the first few meters, which are below the radar resolution, dramatically decreased the dynamic of the radar-backscattered echoes, and hence significantly decreased the penetration depth. Such an effect could also impact radar sounding performance on Mars, where a thin layer of conductive material, such as hematite, maghemite, clay or silt, may significantly attenuate the radar signal. We also conclude that operating at lower frequencies does not yield better penetration in electrically conductive soils. More realistic estimates of the potential effects of electrical resistivity must be considered in assessing the potential performance of future radar sounding investigations of Mars.
 The work described in this article was supported in part by NASA MARSIS Participating Scientist grant NRA-02-OSS-01-MARSIS (E. Heggy, S. Clifford) and by the SwRI Internal Research and Development Program under contract numbers R9428 (R. Grimm) and R9458 (C. Dinwiddie, J. Stamatakos, S. Gonzales). We thank Exploration Instruments, Austin, for providing the instrument support for the radar study and to D. Waiting and M. Necsoiu for their contributions to development of Figures 1–3. Additionally, we thank L. McKague for his assistance with Yucca Mountain vicinity literature, R. Green for his technical review, and G. Walter for his editorial and programmatic reviews. This paper is Lunar Planetary Institute publication 1233.