en
ter
s*,
a Stat
m 19
Earth. Orbital and airborne SAR instruments operating at had potential for penetrating loose deposits. Subsurface
Remote Sensing of Environment 8(Jordan, Huneycutt, & Werner, 1995) and AIRSAR (Lou et
al., 2001)] have returned radar images that provide more
information about a target than single frequency and
polarization systems. These data can be used to classify
surface types based on surface roughness, and they offer
insight into the potential for subsurface imaging at various
frequencies (e.g., Dabbagh, Al-Hinai, & Khan, 1997;
Schaber & Breed, 1999).
images from the 1981 NASA Shuttle Imaging Radar
mission (SIR-A) revealed buried river channels beneath
the sand of the Sahara (McCauley et al., 1982). River
channels appeared as dark, dendritic patterns in L-band
(m = 1.25 GHz; k= 24 cm) images, and sand-covered
bedrock returned a bright radar signal (McCauley et al.,
1982, 1986). Related field studies revealed that the radar
signals were able to penetrate up to 1?2 m of very drymultiple frequencies and polarizations [e.g., SIR-C/X-SAR imaging by radar gained widespread attention whenResults from the experiment were compared to field measurements of sand thickness during acquisition of airborne radar images. In AIRSAR
images, the extent of dry sand in a dune field appears best in C-band because longer wavelength L- and P-band signals penetrate thinner sand
deposits. Images of wet sand (4.9 vol.%) suggest that L-band was able to penetrate thin sand even though that sand was wet. Together, these
laboratory and field measurements contribute towards a better understanding of how a sand cover modifies the radar backscatter of a surface.
D 2003 Elsevier Inc. All rights reserved.
Keywords: Radar backscatter; SAR; Radar penetration; AIRSAR; Sand
1. Introduction
Synthetic aperture radar (SAR) has become an increas-
ingly useful remote sensing tool for studying geology on
The ability of radar to penetrate loose material such as
sand and alluvium is beneficial to studies of the geologic
and climatic history of an area. An early study (Dellwig,
1969) showed that P-band (m= 0.5 GHz; k = 60 cm) radarmoisture content of 0.28 vol.% and caused calculated average atteOver the last two decades, the use of synthetic aperture radar (SAR) to address geologic problems has expanded as new applications for
radar have been developed. One of the earliest and perhaps most surprising results from orbital SAR images of the Sahara was that, under
certain conditions, radar signals penetrated up to several meters of sand to reveal subsurface features such as ancient river channels.
Subsequent studies of radar penetration of arid sand deposits have dealt with factors that govern the ability of radar to penetrate a sand cover.
This paper presents results from a laboratory experiment in which radar backscatter from a surface of rocks was measured under controlled
conditions as a function of frequency, polarization, incidence angle, and sand cover thickness. The sand used in the experiment had a
nuations of 4.2F 1 dB/m for C-band and f 11F 2 dB/m for X-band.Laboratory and field measurem
backscat
Kevin K. William
Department of Geological Sciences, Arizon
Received 29 July 2002; received in revised for
Abstract0034-4257/$ - see front matter D 2003 Elsevier Inc. All rights reserved.
doi:10.1016/j.rse.2003.09.006
* Corresponding author. Center for Earth and Planetary Studies,
National Air and Space Museum, Smithsonian Institution, Washington,
DC 20013-7012, USA. Tel.: +1-202-633-2487; fax: +1-202-786-2566.
E-mail address: williamsk@nasm.si.edu (K.K. Williams).ts of the modification of radar
by sand
Ronald Greeley
e University, Tempe, AZ 85287-1404, USA
September 2003; accepted 19 September 2003
www.elsevier.com/locate/rse
9 (2004) 29?40sand to return a signal from buried bedrock and calcium
carbonate-cemented nodules (McCauley et al., 1986;
Schaber, McCauley, Breed, & Olhoeft, 1986). The dark
radar response from the channels was caused by reflection
of the radar signal from smoother channel fill material
and by attenuation of the signal by that material (Breed et
al., 1983; Davis, Breed, McCauley, & Schaber, 1993;
Elachi, Roth, & Schaber, 1984; McCauley et al., 1982,
1986; Schaber et al., 1986). Schaber et al. (1986) note
that the modern sand sheet in the Eastern Sahara now
covers much of the region, but the L-band radar of SIR-A
was able to penetrate sand cover to expose previously
obscured information about past fluvial activity.
After detection of subsurface penetration by radar during
SIR-A, older SEASAT radar data (also L-band) were recon-
sidered, and it was found that SEASAT data exposed
subsurface tectonic and geomorphic features in southern
California (Blom, Crippen, & Elachi, 1984). Radar penetra-
would have the potential to map the bedrock geology of that
planet (e.g., Campbell et al., 2001) by penetrating sand and
dust that obscures much of the surface at visible and thermal
wavelengths. Table 2 lists characteristics of some orbital and
airborne imaging radar systems that have been used for
geological studies over the last 25 years.
In a comparison of sand penetration by radar at various
frequencies, Schaber and Breed (1999) noted the need for
a better understanding of how sand affects a radar image at
various radar parameters. To address this issue, experimen-
tal measurements of the radar backscatter coefficient under
controlled conditions were conducted. The experimental
results were then compared to backscatter values acquired
by the NASA/JPL AIRSAR instrument during two cam-
Table 2
Characteristics of imaging radars
Sensor Band Wavelength
(cm)
Frequency
(GHz)
X-SAR X 3.1 9.6
SIR-C, RADARSAT, AIRSAR C 5.8 5.3
MAGELLAN S 12.6 2.385
SEASAT, SIR-A, -B, -C, AIRSAR L 23.5 1.25
K.K. Williams, R. Greeley / Remote Sensing of Environment 89 (2004) 29?4030tion field studies were also planned to coincide with and to
take advantage of data from the second Shuttle Imaging
Radar mission (SIR-B) in 1984. Farr, Elachi, Hartl, and
Chowdhury (1986) placed receivers at several depths in the
Nevada desert and measured power received as SIR-B
passed overhead. The data were processed to determine
attenuation of the signal as it passed through alluvium in
the area. The results of Farr et al. were significant because
they showed that radar can penetrate desert soils even in
regions where soil moisture is greater than in hyperarid
regions such as the Sahara (Table 1). SIR-B also provided
the opportunity to conduct further field measurements of
sand thickness in areas where radar could penetrate the sand.
Berlin, Tarabzouni, Al-Naser, Sheikho, and Larson (1986)
studied an area in Saudi Arabia and compared their measure-
ments of sand thickness to the L-band images from SIR-A
(Fig. 1) and SIR-B. They found that the radar signal
penetrated as much as 1.25 m of sand to return an image of
the subsurface. They also determined that the radar was
unable to penetrate f 3 m of sand accumulated in a sand
hill (Fig. 1). Measurements such as these from SIR-A and
SIR-B began to demonstrate the ability of L-band radar to
image subsurface geology and return information that is
otherwise obscured from optical sensors such as LANDSAT
and orbital photographs. The need for multifrequency imag-
ing radar systems was recognized and led to the two SIR-C/
X-SAR missions in April and October 1994.
The United States component of the mission (SIR-C)
included L-band (m = 1.25 GHz; k= 24 cm) and C-band
(m = 5.3 GHz; k = 5.6 cm) radars that were fully polarimetric,
with the ability to transmit and receive in both horizontal
and vertical polarizations (Jordan et al., 1995). The German/
Italian component of the mission (X-SAR) was an X-band
Table 1
Comparison of volumetric moisture contents in radar penetration studies
Location Study Experiment Moisture
contenta (%)
Sahara Schaber et al., 1986 SIR-A < 1.0
Saudi Arabia Berlin et al., 1986 SIR-B 0.29?0.30
Nevada desert Farr et al., 1986 SIR-B 4.2?6.9
Laboratory this study backscatter 0.28
Mojave Desert this study 1999 AIRSAR 4.9
Mojave Desert this study 2000 AIRSAR 0.26a Moisture content is given as volume percent.(m = 9.6 GHz; k = 3.1 cm) radar operating in VV polarization
(Jordan et al., 1995). SIR-C/X-SAR had the added benefit of
variable incidence angle.
Also in the early 1990s, theMagellan mission (Saunders et
al., 1992) used S-band (m= 2.385 GHz; k = 12.6 cm) to
penetrate the thick cloud cover of Venus to return a radar
map of the surface. Imaging radars are also likely to continue
playing an important role in future studies of the Earth and the
other planets. For example, an imaging radar mission to Mars
Fig. 1. SIR-A image over the Al Labbah Plateau, Saudi Arabia, where L-
band radar penetrated f1.25 m of sand to reveal dendritic patterns but did
not penetrate 3 m of sand in a sand hill (arrow) (SIR-A data take no. 28).AIRSAR P 68 0.44
paigns coordinated with field measurements of sand thick-
ness and surface properties.
2. Laboratory measurements of backscatter
2.1. Overview
In order to understand better how a cover of sand
modifies backscatter, an experiment was conducted to
measure the change in radar backscatter of a surface as
layers of sand were added. The Telecommunications
Research Center at Arizona State University houses a
large electromagnetic anechoic chamber with a radar
emitter/receiver system able to operate at frequencies that
include those of imaging radar at 0.5?12 GHz (Birtcher,
1992). The availability of the chamber presented an
excellent opportunity to measure the change in radar
backscatter for varying radar properties under controlled
conditions. Although the facility is typically used to
the targeted radar energy remained stationary as incidence
angle was varied.
The sand used in the experiment was ?#60 Silver Sand?
acquired from P.W. Gillibrand of Simi Valley, CA. A sieve
analysis showed that 96% of the sand had a grain size of
0.1?0.3 mm with 60% having a grain size from 0.2 to 0.3
mm. Thus, the sand is similar in grain size to sand classified
as the modern aeolian sand sheet, lower aeolian sand sheet,
and medium-sand alluvium in the Sahara studies of McCau-
ley et al. (1986) and Schaber et al. (1986). The sand was
composed mineralogically of f 80% quartz, f 16% feld-
K.K. Williams, R. Greeley / Remote Sensing of Environment 89 (2004) 29?40 31Fig. 2. Experiment configuration showing radar emitter and receiver abovemeasure radar cross section of aerospace objects such as
helicopter models (C. Birtcher, personal communication),
the anechoic chamber was reconfigured to measure back-
scatter over geologic materials.
The experiment consisted of a target table that held a
rough surface of rounded river gravel (f 2.5 cm diameter)
that was later covered with sand. The radar emitter and
receiver were oriented downward and were suspended over
the target on a positioning system that allowed measure-
ments at various incidence angles (Fig. 2).
Backscatter measurements were acquired at incidence
angles from 
 10j to 60j from nadir at 0.2j increments.
Measurements were made over radar frequencies centered at
5.3 (C-band) and 9.7 GHz (X-band). Each frequency range
spanned 3 GHz and consisted of 201 measurements. Inthe initial target of river gravel before sand was added.addition, orientations of the radar emitter and receiver were
adjusted to collect data at HH (horizontally transmitted,
horizontally received), HV (horizontally transmitted, verti-
cally received), and VV (vertically transmitted, vertically
received) polarizations. Table 3 lists parameters that were
varied during the experiment. Although lower radar fre-
quencies would have been able to penetrate thicker sand, the
dimensions of the experiment prevented use of the equip-
ment necessary for measurements at lower frequencies.
2.2. Experimental setup and calibration
Fig. 2 shows the configuration of the experiment with the
radar emitter and receiver suspended above the target. A
3 3-m table was constructed of dry lumber held together
with wooden dowels and served as a platform for the gravel
and sand. This table was supported above the floor by dense
foam blocks so that it was at the center of rotation of the
incidence angle positioning system. The rocks were con-
fined to a target area of 2.75 2.75 m over which sand was
emplaced. Overflow of sand due to its natural angle of
repose extended to the table edge.
Before gravel and sand were added to the table, calibra-
tion measurements were made for all radar parameter
variations. A 5-cm-thick deposit of rounded river gravel
was then added to act as the initial, sand-free target. The
gravel had an average diameter of 2.5 cm and had been
rinsed several times and dried to remove any dust that could
contain clay particles and have an effect on the radar
backscatter. The gravel was also sorted and inspected to
remove angular fragments and vesicular basalts. By design-
ing the experiment so that the top of the rock surface was at
the center of rotation of the positioning system, the center of
Table 3
Laboratory parameters for backscatter measurements
Parameter Range of measurements Increment
Frequency C-band (3.65?6.35 GHz) 0.0135 GHz
X-band (8.18?11.18 GHz) 0.015 GHz
Polarization HH, HV, VV ?
Incidence angle 
 10?60j 0.2j
Sand thickness 0?23.2 cma varies
a Specific values of sand thickness: 0, 0.3, 0.5, 1.5, 3.1, 5.8, 8.7, 11.6,
17.4, and 23.2 cm.spar, and f 4% lithic fragments and other minerals. Al-
K.K. Williams, R. Greeley / Remote Sensing of Environment 89 (2004) 29?4032though the sand is not pure quartz, the backscatter over a
small sample of the #60 Silver Sand was measured and
compared to the backscatter over a sample of pure quartz
sand. The difference between the two measurements was
negligible and verifies that the #60 Silver Sand can be used
for this experiment without errors introduced by non-quartz
components.
Because moisture content of sand affects its electrical
properties and the ability of radar to penetrate it, the
moisture content was measured before and throughout the
experiment by weighing samples before and after heating to
remove moisture. Although the water weight was measured
in grams, an average sand density of 1.47 g/cm3 was used to
convert to volume percent of moisture for comparison to
other studies of radar attenuation (e.g., Farr et al., 1986).
The amount of moisture in the sand used for this experiment
was measured at 0.28F 0.1 vol.%. This is much less
moisture than was measured in Nevada by Farr et al.
(1986), but is approximately the same moisture level mea-
sured in Saudi Arabia (Berlin et al., 1986) (Table 1). The
sand was further characterized by having its dielectric
properties measured. Complex permittivity measurements
were used to constrain the real permittivity, eV, at 2.08 for C-
band and 1.97 for X-band. The extreme dryness of the sand
leads to a low imaginary permittivity, eV, which, because of
uncertainties in measurements, is constrained at 0.01?0.02
for both C-band and X-band.
2.3. Measurements
Following the calibration measurements, backscatter of
the river gravel was measured at all combinations of radar
frequency, polarization, and incidence angle (Table 3). In
order to cover the frequency ranges at C-band and X-band,
two radar emitter/receiver sets were used. To make measure-
ments at the three polarization combinations, orientations of
the emitter/receiver sets were adjusted. Therefore, six
sweeps through the range of incidence angles were needed
to cover the full suite of measurements at each sand
thickness level (starting with no sand).
When bare rock measurements were completed, sand
was added to the first level. The sand was added in such a
way to prevent packing that could create inhomogeneities in
density. Because the area of the target table prevented access
from all sides, a walkway was constructed that spanned the
area at a height that would clear the highest sand level. The
walkway was mounted on wheels so that it could be moved
easily across the sample area, and it was removed from the
chamber prior to measurements. In order to add sand to
consistent levels over the target area, metal guide bars were
attached to the sides of the table, and a screen box was
placed on the guide bars. The screen box served as an
indicator of the level to which sand was to be added. After
sand was added across the table in this manner, a metal
straight edge was pulled across the guide bars to level the
sand, and the metal guide bars were removed beforemeasurements were made. This technique was used to add
sand to nine thickness levels up to 23.2 cm (Table 3). The
combinations of frequency, polarization, incidence angle,
and sand thickness led to 4.25 million data points.
2.4. Data analysis
The raw data were processed to remove extraneous
signals from the anechoic chamber, and the received signal
was gated over a time range corresponding to the expected
receipt time for the target. Calibration measurements were
then applied to give backscatter results in decibels per
square meter. The calibrated data display variations due to
constructive and destructive interference likely caused by
the target gravel moving in and out of phase as measure-
ments vary over frequency range and incidence angle (J. van
Zyl, personal communication). Laboratory backscatter
measurements are typically performed as the target is rotated
in the horizontal plane (Nesti, Fortuny, & Lopez-Sanchez,
2000), and the results are averaged to remove effects of
noise and variations in target shape. Because it was not
possible to rotate the target in this experiment, data are
averaged over frequency and incidence angle to determine
how backscatter varies. Several averaging techniques were
tested with respect to preserving the information in the
original data, and it was found that averaging over 0.1
GHz in frequency and 5j in incidence angle could be
applied (Fig. 3). In addition, data from 
 10j to 10j in
incidence angle were averaged to compensate for some of
the single-orientation effects.
For each combination of radar band, polarization, and
incidence angle, the averaged backscatter data were plotted
as a function of frequency for the various sand levels.
Because the sand was dry, a linear curve was fit to the
change in backscatter as a function of frequency. These data
could then be used to plot backscatter as a function of sand
thickness for any selected frequency within the C- and X-
band ranges measured in this study (e.g., Fig. 4). Because
the radar signal is refracted as it passes into the sand layer at
an angle, a dielectric constant of eV= 2 was used to calculate
the path length. The path length is used for sand thickness in
the backscatter versus thickness plots (for sand thickness
>0). These data were plotted as change in backscatter versus
sand thickness by subtracting the backscatter at each sand
level from the backscatter at the first sand level (0.3 cm).
Linear curves were then fit to these plots, and those curves
were used to calculate the attenuation by forcing the curve
to intersect the origin and extrapolating the curve. Because
the data are for two-way travel of the radar signal through
the sand, the data were extrapolated to 50 cm to give
attenuation as decibels per meter (dB/m).
2.5. Experimental results
Fig. 4 shows the decrease in backscatter with increasing
sand thickness at C-band and X-band. There is an initial
K.K. Williams, R. Greeley / Remote Sensing of Environment 89 (2004) 29?40 33decrease in backscatter with the first level of sand (f 0.3
cm). This sand level fills in the roughness of the gravel,
creating a smoother surface and decreasing the backscatter.
The decrease is f 2?6 dB for like-polarized data and is
slightly greater at f 4?8 dB for cross-polarized data. After
this first drop, the backscatter continues to decrease due to
increased attenuation of the signal as more sand is added.
Plots of further decreasing backscatter lead to calculations
of attenuation.
Table 4 gives calculated values of attenuation for the
six frequency-polarization combinations in the experi-
ment. Although measurements were made over a range
of frequencies, values given in Table 4 are for center
frequencies of 5.3 (C-band) and 9.7 GHz (X-band). The
experimental results can be compared to theoretical
attenuations by using dielectric values of the sand and
3.5
Fig. 3. (a) Example of averaged backscatter versus incidence angle for C-
band (at 5.3 GHz), hh polarization, and 11.6 cm of sand. (b) Example of
averaged backscatter versus frequency for C-band, hh polarization, 11.6 cm
of sand, and 30j incidence angle.following the technique of Schaber et al. (1986). Atten-
uation length, 1/a, is calculated by
1=a ? k0=?ptand?eV?0:5 ?1?
where a is the attenuation coefficient, wavelength, k0, is
in meters, and tand = eVV/eV is the loss tangent (Von Hippel,
1954). Signal attenuation is then calculated by
attenuation ? 8:686a: ?2?
Solving Eqs. (1) and (2) for C- and X-band wavelengths
gives predicted attenuations of 3.4?6.8 dB/m for C-band
Fig. 4. Averaged backscatter versus sand thickness for C-band (5.3 GHz)
and X-band (9.7 GHz), hh polarization, and 30j incidence angle. X-band
data shifted by 5 dB along the y-axis to demonstrate greater decrease in
backscatter at X-band compared to C-band.and 6.3?12.6 dB/m for X-band. The ranges in predicted
attenuations are due to the range in imaginary permittiv-
ity discussed above.
Comparison of these predicted values to the experi-
mental results (Fig. 5) shows that the experimental results
fall within the ranges of predicted attenuations. Fig. 5
also compares these results to attenuations calculated from
transmission experiment measurements that used the same
sand (Williams & Greeley, 2001). Attenuations calculated
for C-band agree very well with the value of 4.2 dB/m
calculated from transmission measurements. However,
results from this backscatter experiment do not agree with
the value of 5.9 dB/m for X-band calculated from
transmission measurements. Because the gravel size is
Table 4
Attenuation (dB/m) as a function of frequencya and polarization calculated
from backscatter measurements in this study
Chh Chv Cvv Xhh Xhv Xvv
5.2F 1.2 4.2F 1.5 3.4F 0.5 9.9F 1.4 12.2F 1.0 10.2F 2.6
a For center frequencies of 5.3 and 9.7 GHz for C-band and X-band,
respectively.
geologic surfaces in order to understand variables due
to system and target parameters.
3. Field research
3.1. Overview
While developing the laboratory measurements of back-
scatter, field research was also planned for comparison to
the laboratory results. The Dumont Dunes, Mojave Desert,
California, were selected because they are accessible and
contain a variety of dune forms (MacDonald, 1970).
Dumont Dunes, in the Silurian Valley 55 km north of Baker,
CA, are located outside of Death Valley National Monument
in a recreational area used by all-terrain vehicles. Dune
forms include barchan, transverse, longitudinal, and com-
plex star dunes, with the largest reaching heights of f 120
m (MacDonald, 1970) (Fig. 6). The dunes formed during the
last 18,000 years as multidirectional winds accumulated
Fig. 5. Calculated attenuation values for C-band and X-band polarization
combinations compared to range of predicted values (diagonal lines). The
K.K. Williams, R. Greeley / Remote Sensing of Environment 89 (2004) 29?4034similar to the X-band wavelength, it is likely that the
surface roughness of the gravel exceeds the ?rough?
criterion of Peake and Oliver (1971), creating a strong
specular return. This may be affecting the change in
backscatter with increasing sand thickness, leading to
greater calculated attenuation than would be expected if
the surface were not as rough. The cause of the difference
in attenuations at X-band is not known for sure, but
future experiments of this type would aid in constraining
the cause of this difference. By considering different
surface characteristics and radar properties, experiments
solid, horizontal lines are attenuation values calculated from transmission
experiments with the gray boxes denoting error bars for those calculations.such as this, conducted under controlled conditions, can
be used to develop models of radar backscatter over
Fig. 6. Dunes (f100 m high) at Dumont Dunes, CA, the fsand from southern Lake Manly in present-day Death Valley
and from the former Lake Dumont. Accumulation of sand
continues today with sand from the Armagosa River and
Salt Creek (MacDonald, 1970; Smith, 1984). Although sand
accumulates from multidirectional winds, net transport of
the sand is northward (Smith, 1984).
The substrate beneath the dunes is a lag deposit of lake
and river sediments that were deposited in this area when
Lake Dumont was present (MacDonald, 1970). Since Lake
Dumont desiccated, deflation of lake sediments has oc-
curred and continues today. During fieldwork, a hole was
dug to study change in sediment with depth near the surface
of the study area. An 8-cm-thick layer of desert pavement
(average clast size f 4 cm) covers the surface and serves as
the substrate immediately beneath the sand. Sand and finerield site used for comparison with laboratory results.
grained material forms a layer f 10 cm thick beneath the
desert pavement, and a mix of gravel and fine material is
found beneath that fine-grained layer. Dumont Dunes can
receive rainstorms, but it is typically very arid during the hot
summer months. Yearly average rainfall is f 9.33 cm at
nearby Silver Lake, but the combined average rainfall for
May to July is only 0.06 cm. The low rainfall contributes to
the paucity of vegetation.
3.2. Field measurements
Fieldwork was conducted during two flights of the
NASA/JPL AIRSAR instrument in 1999 and 2000. Field-
work consisted of sand thickness measurements and surface
characterization. During the 1999 flight, a rainstorm resulted
in a 3?4-cm layer of sand with 4.9% water content. In 2000,
the sand was extremely dry (Table 1). These two conditions
allowed comparison of radar images for two different sand
moisture conditions.
The first set of field measurements bracketed the AIR-
SAR flights on June 2, 1999. A 100 100-m grid was laid
out and located using a differential GPS system with
horizontal accuracy of f 1 m. Within this grid, sand
thickness was measured using a probe that was pushed
through the sand until underlying desert pavement was
detected by resistance when the probe hit the substrate
e fore
K.K. Williams, R. Greeley / Remote Sensing of Environment 89 (2004) 29?40 35Fig. 7. (a) Technique for measuring sand thickness. Bare gravel surface in th
to aid in registering radar images. The hypotenuse of each side triangle is f1.4ground with 5?10-cm-thick sand being measured. (b) Corner reflector used
m.
(Fig. 7a). This technique was tested at several locations by
digging through the sand to measure the depth to desert
pavement, and it was confirmed that the probe accurately
measured sand thickness. The grid area contained bare
desert pavement surface, thin (2?10 cm) sand, and sand
up to f 1 m thick. Samples of the wet sand were later
heated and dried to determine that the wet layer contained
4.9% water by volume (Table 1).
A second AIRSAR flight occurred on July 22, 2000.
With the hot summer temperatures in the Mojave Desert,
the dune sand was extremely dry, having measured
moisture content of 0.26F 0.05% to a depth of at least
50 cm. This value is almost identical to the moisture level
of sand used in the backscatter laboratory experiment
(Table 1).
During the second AIRSAR flight, another grid was set up
using the GPS unit, and sand thickness was measured. To aid
in matching the GPS points to the corresponding pixels on the
radar images, eight corner reflectors were emplaced prior to
the flight (Fig. 7b). Although these corner reflectors were not
of calibration quality, their locations showed very well in the
AIRSAR images. Other features such as bushes, prominent
dunes, and roads were also recorded with the GPS unit.
3.3. AIRSAR data
AIRSAR is a NASA/JPL radar imaging system that
(Lou et al., 2001). For this study, polarimetric radar data
(Fig. 8) were to be acquired at P-, L-, and C-bands
(k = 68, 23.5, and 5.8 cm, respectively). During the 1999
flight, P-band data were not collected because of military
operations; however, data were acquired at all three bands
during the 2000 flight. Slight differences in image char-
acteristics between the two flights resulted from system
configurations and from flight constraints imposed by
nearby Fort Irwin, CA (Table 5).
3.4. Field results
Positions of corner reflectors and bushes were used to
register the AIRSAR images to the map-projected GPS
data. Once matched to the GPS points, backscatter values
within the sand thickness measurement grids were
ause of signal attenuation by the sand. Sand ridges shown in Fig. 11 are at A. Area
is from the left. Incidence angle is 55j at the left side of the dune field and 59j at
Table 5
Radar image characteristics from AIRSAR flights over Dumont Dunes
Image
ID
Date acquired Bands Range
pixel
size (m)
Along-track
pixel size
(m)
Anglesa
(j)
cm6088 June 2, 1999 C, L 3.32 4.58 43?57
cm6093 June 2, 1999 C, L 3.32 4.61 49?57
cm6243 July 22, 2000 C, L, P 6.64 9.22 48?59
cm6244 July 22, 2000 C, L, P 6.64 9.16 55?59
a Range of incidence angles that cover dune field.
K.K. Williams, R. Greeley / Remote Sensing of Environment 89 (2004) 29?4036flies on a DC-8 aircraft and operates in several modes
Fig. 8. Portion of L-band total power AIRSAR image. Dunes appear dark bec
of sand thickness measurements shown in Figs. 7a and 9 is at B. Illumination
the right side [AIRSAR image cm6244].extracted from the images and converted into decibels
using the ENVI software package from Research Systems
Inc. GPS data of other features such as dune outlines
were used to verify the correct positions of the AIRSAR
pixels (e.g., Fig. 9).
To search for a relationship between AIRSAR data and
sand thickness, backscatter values were plotted versus sand
thickness measured in the corresponding grid section. An
example is shown by the open squares in Fig. 10. The
scatter in the points and the large error bars are likely
caused by variations in sand thickness within a grid cell or
by a bush within a cell. There does not appear to be a
strong relationship between backscatter and sand thickness;
however, the scatter in thickness caused by irregularities
may be masking a relationship.
One objective of this study was to investigate whether
laboratory backscatter results can be used to predict sand
thickness from backscatter image data. In order to test this,
attenuation was extracted from laboratory results cor-
responding to the parameters for a certain image. For
example, the backscatter values shown in Fig. 10 were
The fieldwork also allowed ground truth for compar-
ison of radar imagery of sand dunes under wet and dry
conditions. Some areas in the 1999 images were interest-
ing because they contain thin sand but showed as dark in
some of the radar images. It was speculated that the dark
radar signature was caused by increased attenuation due
to wet sand and that those areas should be brighter in
images acquired in 2000 when the sand was dry. Loca-
tions of these areas were noted in the 1999 AIRSAR
images and were visited in 2000 to characterize the
surface. It was confirmed that these areas containedf 1?6
cm of sand where moisture in 1999 caused greater
attenuation at C-band. The moisture caused increased
attenuation in both C- and L-bands, but some penetration
by L-band occurred in areas of thin sand. The moisture
also caused sides of some dunes to appear radar bright in
L-band, likely due to small local incidence angles return-
ing a bright signal from the sand. Many of these larger
dunes appear radar dark in the 2000 images because the
signal passed into the dry sand and was attenuated. In the
2000 images, extent of dry sand is shown better in C-
band because sand of intermediate thickness (few tens of
K.K. Williams, R. Greeley / Remote Sensing of Environment 89 (2004) 29?40 37taken from a C-band, hh polarization AIRSAR image where
the incidence angle at the measurement location was 58j.
The laboratory attenuation for these parameters is 6.6 dB/m.
It should be noted that this value differs from that given for
Chh in Table 4 because those values were corrected for
signal refraction and averaged to give values that could be
applied to any angle. The value of 6.6 dB/m used here was
calculated from laboratory measurements at a 58j incidence
angle. The attenuation is applied to the backscatter for bare
ground (
 19.5 dB in Fig. 10) to calculate the sand
thickness at lower backscatter values. The values of pre-
dicted sand thickness are shown as a solid line in Fig. 10
where the two-way travel of the radar signal within the sand
has been taken into account. The ?bare ground? in this area
was composed of gravel mixed with sand, much like the
Fig. 9. Portion of an AIRSAR L-band HH backscatter image overlaid with
GPS data of measurement grid and dune outline. The toe of the dune in the
left part of the measurement grid is up to 1 m thick. Lower part of the grid
contains bare gravel, and sand of intermediate thickness occupies other grid
areas. This location is to the left of B in Fig. 8. Incidence angle is 58j, and
illumination is from the bottom [AIRSAR image cm6244].first two levels of sand in the laboratory measurements.
Therefore, an initial decrease of f 3 dB (Fig. 4) was not
applied to the predicted values in Fig. 10.
Fig. 10 shows that the measurements of sand thickness
are scattered around the line of predicted thickness, but
except for two outliers, most of the measured points are
within 5 or 10 cm of the predicted values. This suggests that
laboratory measurements of attenuation may be useful for
estimating sand thickness from radar image data. However,
more comparisons of predictions to field data are necessary
to confirm a definite relationship.
Fig. 10. Comparison of predicted and measured sand thickness shown as a
function of backscatter. Open squares are sand thickness measured in the
field versus backscatter values measured from an AIRSAR image. Solid
line is predicted sand thickness from laboratory results using C-band radar
parameters specific to this example area.centimeters) attenuated the signal. The L- and P-band
signals penetrated this sand and give a brighter return, but
thicker sand still attenuates the radar signals and shows as
radar dark (Fig. 11). In the 1999 images, sand attenuated
the C-band signal, but L-band was able to penetrate the
moist sand where it was up to a few tens of centimeters
thick (Fig. 11).
Overall, C-band images are useful for determining the
extent of sand in an area, and L-band images can be used
to discern the dune morphologies. P-band signals (and
perhaps some L-band signals) likely penetrated the desert
pavement into the subsurface gravel of the lake deposits.
The penetration into the subsurface in this very dry area
by P-band signals cannot be quantified because required
field measurements were not made. Although Dumont
Dunes offers an accessible area with interesting dune
morphologies, the subsurface is largely unconsolidated,
promoting penetration by lower frequency radars. Future
field measurements coordinated with radar image acqui-
sition should be conducted in areas where the substrate is
bedrock.
4. Conclusions
Laboratory measurements of radar backscatter were
made over a surface of gravel in order to measure the
change in backscatter with increasing thickness of a sand
cover. These measurements were made at various polar-
izations and incidence angles for C-band and X-band, and
the change in backscatter was used to calculate attenuation
caused by sand (Table 4). The attenuation values calculated
here agree well with values measured in microwave trans-
ands,
uatio
Portio
in atte
K.K. Williams, R. Greeley / Remote Sensing of Environment 89 (2004) 29?4038Fig. 11. (a) Portion of AIRSAR image cm6244 from 2000 at C-, L-, and P-b
high. Note that the ridges appear wider in C-band because there is more atten
penetration. Incidence angle is 56j, and illumination is from the bottom. (b)
bands (total power) when sand was moist (4.9% H2O). Note that C-band aga
the wet sand. Incidence angle is 53j, and illumination is from the bottom.total power. These dry sand ridges are shown at A in Fig. 8 and are 2?4 m
n of the signal. The ridges are narrowest in P-band because of greater signal
n of AIRSAR image cm6093 from 1999 showing sand ridges in C- and L-
nuates more signal, but L-band appears to have still penetrated thin areas of
backscatter was caused by smoothing of initial surface
roughness by sand. As sand was added to greater thickness,
Radar system. IEEE IGARSS ?01 ( pp. 2046?2048). Piscataway, NJ:
IEEE.
MacDonald, A. A. (1970). The northern Mojave Desert?s Little Sahara.
K.K. Williams, R. Greeley / Remote Sensing of Environment 89 (2004) 29?40 39the backscatter continued to decrease due to increased
attenuation.
Field measurements of sand thickness were compared to
thickness estimates based on AIRSAR backscatter and
laboratory results. Although there is general agreement
between predicted and measured sand thickness values,
scatter in measured sand thickness caused by nonuniform
sand and sparse vegetation makes it difficult to determine
definitely whether there is a relationship. However, future
additional comparisons of laboratory results to field meas-
urements may answer this question.
Fieldwork was also used to ground truth causes of
differences between AIRSAR images of dry and wet sand.
Fig. 11 shows dependence of thin sand backscatter on
wavelength for dry and wet sand. The extent of the dry
sand coverage is shown well in the C-band images, but L-
and P-bands penetrate thin sand. L-band is better for
determining dune morphologies. P-band shows the loca-
tions of thicker sand when compared with L-band and may
have penetrated into the subsurface. When the sand was
wet, the C-band signal was attenuated, but L-band appears
to have penetrated thin sand even though it was wet. The
ability of L-band (and presumably P-band if it were
available) to penetrate thin wet sand without great atten-
uation agrees with microwave transmission experiments
(Williams, 2002; Williams & Greeley, 2001). Although
penetration would be greater in dry sand, these results
challenge previous allegations that sand must be extremely
dry for radar penetration to occur.
As future imaging radar missions are developed, it is
important to consider results of studies such as this that
combine field measurements and laboratory data to con-
struct a better understanding of the ability of radar to
penetrate sand at various radar parameters and under differ-
ent surface conditions.
Acknowledgements
The authors wish to acknowledge C. Birtcher and G.
Beardmore for their assistance with the experiment and R.
Davis, M. Kraft, and J. Moreau for help in the field. We
thank E. O?Leary and the AIRSAR team for acquiring the
radar images of the field site. We also thank J. van Zyl and
T. Farr for helpful discussions, D. Ball for photographicmission experiments at C-band but are almost twice the
attenuation at X-band measured by transmission. This
disagreement at X-band may be due to Rayleigh scattering
effects of the gravel, but it would be worthwhile to test this
in future experiments.
The laboratory measurements also showed an initial
decrease in backscatter of 2?8 dB when a thin layer of
sand filled spaces within the gravel. This steep drop insupport, and C. Santamarina for conducting the dielectricMineral Information Service, vol. 23 ( pp. 3?6 ). Sacramento: CA
Division of Mines and Geology.
McCauley, J. F., Breed, C. S., Schaber, G. G., McHugh, W. P., Issawi, B.,
Haynes, C. V., Grolier, M. J., & Kilani, A. E. (1986). Paleodrainages of
the Eastern Sahara?The radar rivers revisited (SIR-A/B implications
for a mid-Tertiary trans-African drainage system). IEEE Transactions
on Geoscience and Remote Sensing, GE-24, 624?648.
McCauley, J. F., Schaber, G. G., Breed, C. S., Grolier, M. J., Haynes, C. V.,measurements. We are grateful for reviews by T. Farr and
an anonymous reviewer that improved this manuscript.
This work was supported by a NASA GSRP fellowship
through JPL, a NASA/ASU Space Grant fellowship, and
the Dr. Robert H. Goddard Scholarship from the National
Space Club.
References
Berlin, G. L., Tarabzouni, M. A., Al-Naser, A. H., Sheikho, K. M., &
Larson, R. W. (1986). SIR-B subsurface imaging of a sand-buried land-
scape: Al Labbah Plateau, Saudi Arabia. IEEE Transactions on Geo-
science and Remote Sensing, GE-24, 595?602.
Birtcher, C. R., 1992. The electromagnetic anechoic chamber at Arizona
State University: Characterization and measurements. M.S. thesis.
Tempe: Arizona State University, 120 pp.
Blom, R. G., Crippen, R. E., & Elachi, C. (1984). Detection of subsurface
features in SEASAT radar images of Means Valley, Mojave Desert,
California. Geology, 12, 346?349.
Breed, C. S., Schaber, G. G., McCauley, J. F., Grolier, M. J., Haynes,
C. V., Elachi, C., Blom, R., Issawi, B., & McHugh, W. P. (1983).
Subsurface geology of the Western Desert in Egypt and Sudan
revealed by Shuttle Imaging Radar (SIR-A). First Spaceborne Imag-
ing Radar Symposium, JPL Pub. 83-11 ( pp. 10?12). Pasadena, CA:
Jet Propulsion Laboratory.
Campbell, B. A., Campbell, D. B., Grant, J. A., Hensley, S., Maxwell, T. A.,
Plaut, J. J., Rosen, P., Shepard, M. K., & Simpson, R. (2001). Orbital
imaging radar and the search for water on Mars. Conference on Geo-
physical Detection of Subsurface Water on Mars, Abstract #7040. Hous-
ton: Lunar and Planetary Institute.
Dabbagh, A. E., Al-Hinai, K. G., & Khan, M. A. (1997). Detection of sand-
covered geologic features using SIR-C/X-SAR data. Remote Sensing of
Environment, 59, 375?382.
Davis, P. A., Breed, C. S., McCauley, J. F., & Schaber, G. G. (1993).
Surficial geology of the Safsaf region, south-central Egypt, derived
from remote sensing and field data. Remote Sensing of Environment,
46, 183?203.
Dellwig, L. F. (1969). An evaluation of mutifrequency radar imagery of the
Pisgah Crater area, California. Modern Geology, 1, 65?73.
Elachi, C., Roth, L. E., & Schaber, G. (1984). Spaceborne radar subsurface
imaging in hyperarid regions. IEEE Transactions on Geoscience and
Remote Sensing, GE-22, 383?387.
Farr, T. G., Elachi, C., Hartl, P., & Chowdhury, K. (1986). Microwave
penetration and attenuation in desert soil: A field experiment with the
Shuttle Imaging Radar. IEEE Transactions on Geoscience and Remote
Sensing, GE-24, 590?594.
Jordan, R. L., Huneycutt, B. L., & Werner, M. (1995). The SIR C/X SAR
synthetic aperture radar system. IEEE Transactions on Geoscience and
Remote Sensing, 33, 829?839.
Lou, Y., Imel, D. A., Chu, D., Miller, T. W., Moller, D., & Skotnicki, W.
(2001). Progress report on the NASA/JPL Airborne Synthetic ApertureIssawi, B., Elachi, C., & Blom, R. (1982). Subsurface valleys and geo-
archaeology of the eastern Sahara revealed by shuttle radar. Science,
218, 1004?1020.
Nesti, G., Fortuny, J., & Lopez-Sanchez, J. M. (2000). Polarimetric micro-
wave remote sensing experiments at the EMSL. IEEE Geoscience and
Remote Sensing Society Newsletter, 113, 6?11.
Peake, W. H., & Oliver, T. L. (1971). The response of terrestrial surfaces at
microwave frequencies. Ohio State University Technical Report 2440-7,
Columbus, OH.
Saunders, R. S., et al. (1992). Magellan mission summary. Journal of Geo-
physical Research, 97, 13067?13090.
Schaber, G. G., & Breed, C. S. (1999). The importance of SAR wavelength
in penetrating blow sand in northern Arizona. Remote Sensing of En-
vironment, 69, 87?104.
Schaber, G. G., McCauley, J. F., Breed, C. S., & Olhoeft, G. R. (1986).
Shuttle imaging radar: Physical controls on signal penetration and scat-
tering in the eastern Sahara. IEEE Transactions on Geoscience and
Remote Sensing, GE-24, 603?623.
Smith, R. S. U. (1984). Eolian geomorphology of the Devil?s Playground,
Kelso Dunes, and Silurian Valley, California. Geological Society of
America, 97th, Field Trip 14 (pp. 162?174). Boulder, CO.
Von Hippel, A. R. (1954). Dielectrics and waves ( pp. 26?37). New York:
Wiley.
Williams, K. K. (2002). Radar penetration of sand and dust as determined
by laboratory and field measurements. Ph.D. Thesis. Tempe, AZ: Ari-
zona State University.
Williams, K. K., & Greeley, R. (2001). Radar attenuation by sand: Labo-
ratory measurements of radar transmission. IEEE Transactions on Geo-
science and Remote Sensing, 39, 2521?2526.
K.K. Williams, R. Greeley / Remote Sensing of Environment 89 (2004) 29?4040