Remote Sensing of Environment 184 (2016) 288–301

Contents lists available at ScienceDirect

Remote Sensing of Environment

journal homepage: www.elsevier.com/locate/rse

Estimating the age of desert alluvial surfaces with spaceborne radar data
Guy Hetz a,b, Amit Mushkin a,c,⁎, Dan G. Blumberg b, Gidon Baer a, Hanan Ginat d
a
Geological Survey of Israel, Jerusalem, Israel
b
Ben-Gurion University of the Negev, Be'er Sheva, Israel
c
University of Washington, Seattle, WA, USA
d
Dead Sea and Arava Science Center, Israel

a r t i c l e i n f o a b s t r a c t

Article history: We present a new dating approach for desert alluvial surfaces that utilizes radar image data to obtain regional-
Received 14 January 2015 scale correlations between dated surfaces and surfaces of unknown ages. The study was carried out along the
Received in revised form 2 June 2016 Arava segment of the Dead Sea Transform north of the Gulf of Elat (Aqaba) using Advanced Land Observation Sat-
Accepted 6 July 2016
ellite (ALOS) Phased Array type L-band Synthetic Aperture Radar (PALSAR) data with 6–12 m/pixel resolutions.
Available online 21 July 2016
Nineteen independently dated Quaternary morphostratigraphic alluvial units (5 to 1910 ka) from eight different
sites with variable lithologies (carbonate/magmatic/chert) were examined in order to develop, test and validate
the proposed approach. Expanding on previous studies that established the characteristic smoothing of desert al-
luvial surface through time and the capability of measuring such roughness variations with radar, we employed a
pixel-aggregation approach that yielded a robust (R2 = 0.95) power-law relation between the average radar
backscatter value from the surface (‘AR’) and the abandonment age (‘T’) of the unit as measured with indepen-
dent methods such as luminescence and/or cosmogenic radionuclide dating. We found that: 1) This AR-T corre-
lation can be inverted to obtain calibrated radar-based T estimates as old as ~1.5 Ma with 35% uncertainty; 2)
ALOS PALSAR 6 m/pixel images acquired at incidence angles of 34–38° yielded optimal results; 3) a sample of
at least 70 contiguous radar pixels was required to capture the natural roughness variability of the surfaces ex-
amined, and 4) surface lithology did not exert a primary effect on T inversions. Validation experiments yielded
radar-based ages of 3 ± 1, 91 ± 32, 484 ± 169 and 1604 ± 561 ka that compare to Holocene, 56 ± 14,
540 ± 60 and 1590 ± 250 ka ages previously determined for the same surfaces, respectively. We propose that
with region-specific calibrations, spaceborne radar data can be used to quantitatively estimate the age of aban-
doned alluvial units across regional scales.
© 2016 Elsevier Inc. All rights reserved.

1. Introduction suitable dating material in desert environments can also limit the appli-
cation of in-situ dating approaches in arid landscapes. As a result, efforts
1.1. Motivation to determine the age of alluvial surfaces in desert environments are
often restricted by the number of surfaces that can be effectively
Abandoned alluvial landforms provide effective recorders of past dated rather than by the number of surfaces of scientific interest identi-
tectonic activity, climatic variations and landscape evolution processes fied. In this context, we present a new surface dating approach that uti-
(e.g., Bull, 1977, 1991). Consequently, determining the surface age of lizes spaceborne radar data to quantitatively estimate the age of
morphostratigraphic alluvial units often comprises a pivotal component abandoned alluvial surfaces that could not be otherwise dated using
in quantitative studies of Earth's surface and its dynamics over recent in-situ approaches.
geologic time-scales. Whereas in-situ radiometric dating approaches
such as 14C, luminescence and Cosmogenic Radionuclides (CRN) (here- 1.2. Mapping desert alluvial landforms with remotely sensed data
after “in-situ dating”) are routinely applied in this context (e.g., Noller et
al., 2000) the substantial resources required for application of these ap- Remotely sensed data provide an effective resource for mapping
proaches, such as sample collection and laboratory analyses, typically morphostratigraphic alluvial units in bare desert terrains: Farr and
restrict the number of surfaces that can be dated in a single study. Inher- Chadwick (1996) demonstrated the use of panchromatic satellite im-
ent analytical complexities as well as accessibility issues and lack of ages together with spaceborne radar data to distinguish between late
Quaternary alluvial units in central Asia and Death Valley, USA. Others
⁎ Corresponding author at: University of Washington, Department of Earth and Space
showed the benefits of combining visible - shortwave infrared (VIS-
Science, Box 351310, 98195 Seattle, WA, USA. SWIR: 0.4–2.5 μm) data with radar images to map time-progressive
E-mail address: [email protected] (A. Mushkin). compositional and roughness variations amongst abandoned alluvial

http://dx.doi.org/10.1016/j.rse.2016.07.006
0034-4257/© 2016 Elsevier Inc. All rights reserved.
 G. Hetz et al. / Remote Sensing of Environment 184 (2016) 288–301 289

units (Weeks et al., 1996, 1997; Kierein-Young, 1997). Satellite stereo 1.4.2. Radar measurements of surface roughness
imaging was also established as a complementary and effective measure Radar backscatter, which primarily responds to sensor-surface
for mapping surface roughness and relative age variations amongst phase angle, surface dielectric constant and sub-pixel surface rough-
morphostratigraphic alluvial units in Death Valley, USA and the Negev ness/structure (e.g., Ulaby et al., 1978), is a widely utilized tool for map-
desert in southern Israel (Mushkin and Gillespie, 2005). ping spatial variability in soil roughness and moisture content (Ulaby et
Airborne sensors provide additional effective tools for such mapping al., 1982). In the absence of significant soil moisture and vegetation, sur-
purposes. Multi-spectral thermal infrared images from Death Valley face roughness integrated over all sub-pixel scales (down to ~the wave-
(Gillespie et al., 1984) and hyperspectral VIS-SWIR images from the length of the incident radar radiation) becomes the primary parameter
Negev desert of Israel (Crouvi et al., 2006) were used to successfully modulating pixel-to-pixel variations in radar backscatter from low-
map variations in composition and age-progressive varnish develop- lying desert surfaces (e.g., Evans et al., 1992; Farr, 1992; Weeks et al.,
ment on late Quaternary alluvial terraces. Airborne LiDAR data at 1 m/ 1996, 1997; Farr and Chadwick, 1996; Kierein-Young, 1997). Of the
pixel resolution were also successfully applied to capture age-depen- commonly used radar wavelengths, i.e., X (~ 3 cm), C (~ 5.6 cm), L
dent roughness variations amongst late Quaternary alluvial terraces (~ 24 cm) and P (~ 68 cm) bands, L-band appears to best capture the
and to delineate and map morphostratigraphic alluvial units in Death roughness variations between abandoned alluvial surfaces (e.g., Ulaby
Valley (Frankel and Dolan, 2007) and the Sonoran desert in SW USA et al., 1978; Farr and Chadwick, 1996).
(Regmi et al., 2013).
1.4.3. Radar backscatter as a calibrated proxy for surface age
Expanding on the characteristic smoothing of desert alluvial surface
1.3. Post-abandonment evolution of desert alluvial surfaces
as a function of surface age (Section 1.4.1) and the capability of radar
images to effectively quantify such roughness variability (Section
In-situ weathering of young and rough boulder-strewn deposits into
1.4.2) we examine the use of radar backscatter as a readily calibrated
smooth and persistent desert pavements presents a common post-
quantitative proxy for the age of morphostratigraphic alluvial units in
abandonment evolution trajectory for alluvial surfaces in desert envi-
arid environments (Fig. 1).
ronments worldwide (e.g., Bull, 1977). Desert pavement development
Four objectives were outlined to achieve this goal:
is typically driven by a suite of processes that include reg soil formation
in the upper decimeters – meters of the deposits (e.g., Amit and Gerson, A) Determine the optimal radar backscatter metric that displays a
1986), subaerial clast weathering (Wells et al., 1995), varnish develop- systematic and predictive relation with surface age.
ment on surface clasts and accumulation of eolian dust as a clast-free Av B) Evaluate the uncertainty of this radar vs. surface-age regression
horizon beneath the pavement rocks (e.g., McFadden et al., 1987). The and its sensitivity to key variables such as measurement noise,
initial roughness of desert alluvial surfaces primarily results from the natural roughness variability and surface composition.
wide range of boulder to pebble sized clasts found in such deposits C) Validate the performance of the proposed radar-based surface
and from larger-scale depositional elements such as boulder clusters, le- dating approach.
vees and bar and swale features. Gravity-driven diffusion of material D) Demonstrate the use of radar backscatter to infer the age of
from bars into swales and in-situ subaerial weathering of clasts typically morphostratigraphic alluvial units that could not be successfully
reduce this initial roughness asymptotically over ~ 104 yr time scales dated with in-situ approaches.
until mature and smooth desert pavements are established (Amit et
al., 1993; Matmon et al., 2006; Frankel and Dolan, 2007; Regmi et al.,
2013; Mushkin et al., 2014). This predictable smoothing trajectory of
desert alluvial surfaces has been widely recognized as an effective mea- 2. Study sites
sure for relative age determination (e.g., McFadden et al., 1989;
McDonald et al., 2003). Our study area is located along the Arava valley segment of the Dead
Typically, mature desert pavements, which consist of an Av silt hori- Sea Transform (DST) (Fig. 2), which separates the Arabian and African
zon and an overlaying interlacing pavement of angular rock fragments, plates and accounts for ~ 105 km of left-lateral slip since the Miocene
develop over time into an impermeable surface-armoring layer that (Garfunkel et al., 1981). Asymmetric vertical tectonics along this seg-
promotes runoff accumulation during precipitation events. Such well- ment of the DST (e.g., Garfunkel and Ben Avraham, 1996) have led to
established (late-stage) desert pavements commonly display a dendrit- deeper exhumation of crustal lithologies along the eastern side of the
ic network of erosional rills that drain the surface internally and incise Arava valley and together with the ~ 105 km of lateral offset result in
into it over time (Christenson and Purcell, 1985; Wells et al., 1987;
Farr, 1992) to the point where only remnants of the original pavement
surface are preserved. In hyper-arid low-relief and tectonically stable
settings, where such late-stage incision is minimal, extensive and
smooth desert pavement surfaces can persist for 106 yr time-scales
(e.g., Matmon et al., 2009; Guralnik et al., 2010).

1.4. Approach

1.4.1. Roughness as a remote measure for the age of desert alluvial surfaces
Recent studies demonstrated that airborne LiDAR can be used to
measure the age-dependent smoothing of alluvial surfaces due to
diminishing bar-swale relief at larger than meter (‘supra-meter’) length
scales (Frankel and Dolan, 2007; Regmi et al., 2013). However, these
studies also showed that late stage (greater than ~105 ka in the cases ex-
amined) incision rills can ‘reverse’ the age-progressive smoothing tra-
Fig. 1. The relation between radar, surface roughness and age of abandoned alluvial
jectory of undisturbed surfaces at supra-meter length scales and can surfaces in desert environments. Previous studies demonstrated robust radar-roughness
therefore introduce ambiguities into straight-forward surface-age in- and roughness-age correlations for such landforms. In the present study we test the
versions from such roughness data. relation between radar and surface age.
290 G. Hetz et al. / Remote Sensing of Environment 184 (2016) 288–301

Fig. 2. Location map. Solid white boxes – main study sites, dashed boxes – validation sites. White arrows mark the DST lineament that crosses the Risha and Mazla-Fuhud sites. Base map
derived from 90 m/pixel SRTM data (http://www2.jpl.nasa.gov/srtm).

juxtaposition of different crustal lithologies within the drainage basins Raham, Risha, Mazla-Fuhud and Paran). The three additional sites
of this region. Hyperarid Quaternary conditions (Noy-Meir, 1973; Muhtadi, Shitta and Ashosh were used for validation.
Amit et al., 2006), sparse vegetation cover and a fairly large number of
readily accessible and previously dated alluvial surfaces in this area 2.1. Shehoret
(e.g., Matmon et al., 2009; Porat et al., 2010; Guralnik et al., 2010; Le
Béon et al., 2010, 2012) provide an ideal setting for developing and test- This was our primary site and is located ~ 7 km north of the Gulf of
ing the radar-based surface dating approach. Elat/Aqaba along the western side of the Arava Valley (Fig. 2). This
Altogether we examined nineteen previously dated chronosequence has been the focus of several field-based geomor-
morphostratigraphic alluvial units at eight different sites along the phic studies (e.g., Amit et al., 1993, 1996; Mushkin et al., 2014) and
Arava Valley in Israel and Jordan (Fig. 2). Surface ages ranged from the ages of its main morphostratigraphic units have been deter-
Holocene to 1.9 Ma (Table 1). Field work for calibration and sensitiv- mined with luminescence techniques (Porat et al., 2010). Of these
ity analyses purposes was carried out at five of these sites (Shehoret, units, we examined four mid-fan, predominately carbonate-clast
 G. Hetz et al. / Remote Sensing of Environment 184 (2016) 288–301 291

Table 1
Alluvial surfaces examined in this study.

a
Porat et al. (2010); bCrouvi et al. (2006); cMushkin et al. (2014); dLe Béon et al. (2012); eGuralnik et al. (2010); fLe Béon et al. (2010); ⁎Relate to the sample identifier used in the respective
study.
292 G. Hetz et al. / Remote Sensing of Environment 184 (2016) 288–301

surfaces with b15% magmatics and sandstones, dated to 5 ± 1, 14 ± interpreting CRN exposure ages and that luminescence (deposition)
2, 62 ± 5 and 87 ± 2 ka (Table 1). Ground-based LiDAR measure- ages provide maximum age constraints for surface abandonment (e.g.,
ments and rock counts at this site revealed systematic surface Guralnik et al., 2011), we adopt the geologic considerations presented
smoothing and clast-size reduction at sub-meter length scales as a in the respective dating studies above and regard these two different
function of surface age (Mushkin et al., 2014). types of age estimates (i.e., luminescence and CRN) as 1st-order approx-
imations for the time elapsed since these surfaces became fluvially
2.2. Raham inactive.

This site is located ~10 km north of the Shehoret site along the west- 3.2. Radar data
ern side of the Arava Valley (Fig. 2). Of the five late Quaternary
morphostratigraphic units previously mapped within this We examined ALOS PALSAR (http://www.eorc.jaxa.jp/ALOS/en/
chronosequence (Crouvi et al., 2006), we examined a Holocene unit about/palsar.htm) L-band data with different polarizations (HH, HV),
and two older terraces dated with luminescence techniques to 24 ± 4 incidence angles (24°, 38°), and pixel resolutions (6.25 and 12.5 m/
and 56 ± 10 ka (Mushkin et al., 2014). Clast composition on the pixel) that were available for the study area (Table 2). All images were
Raham surfaces is similar to that of the Shehoret units examined with radiometrically calibrated to backscatter values (R) expressed in dB
85–90% carbonate and 15–10% magmatic and sandstone clasts (Table units using:
1).
R ¼ 20 log10 ðDNÞ−CF ð1Þ
2.3. Risha and Mazla-Fuhud
where DN is the digital number in the ‘raw’ image and CF is a calibration
These sites are located ~70–110 km north of the Gulf of Elat/Aqaba factor obtained from http://earth.eo.esa.int/pcs/alos/palsar/articles/
along the eastern side of the Arava Valley (Fig. 2). These fan sequences Calibration_palsar_products_v13.pdf.
are sinistrally offset southwards relative to their original catchments
(Ginat et al., 1998) along a prominent lineament of the DST. Le Béon 4. Methods
et al. (2012) mapped eight Quaternary morphostratigraphic alluvial
units common to both sites (F0 – active wash to F7 the oldest) and 4.1. Field work and image analysis
used CRN to date units F2, F3, F4 and F5 to 48 ± 5, 95 ± 25, 160 ± 20
and 320 ± 24 ka, respectively. Units F6 and F7, which were determined Field work was carried out at the five study sites in order to establish
to be older than F5 by their geomorphic characteristics, could not be the optimal predictive relation between radar backscatter and surface
successfully dated with CRN. The Risha and Mazla-Fuhud age. The primary field site was the Shehoret chronosequence where a
chronosequences display a predominately magmatic clast-assemblage geomorphic map of the main morphostratigraphic mid-fan units was
with ~20–30% carbonates, sandstones and cherts (Table 1). produced (Fig. 3). At the other sites, field work was focused on detailed
geomorphic characterization of the morphostratigraphic units that
2.4. Paran were examined with radar (Table 1). Image analysis included character-
ization of the limitations imposed by noise levels in the radar data used
This site is located ~60 km NW of the Gulf of Elat/Aqaba and ~30 km and manual selection of representative samples of radar pixels for each
west of the DST axis (Fig. 2). The extensive low-relief alluvial plains at of the morphostratigraphic units.
this site have been previously studied as examples for some of the oldest Measurement noise is a widely-acknowledged problem in radar im-
preserved desert pavements on Earth and were dated with CRN to ages. Here, it is addressed through a pixel aggregation approach in-
1160 ± 160 (PS8) 1610 ± 260 (PS4) and 1940 ± 340 ka (PS8) formed by ancillary morphostratigraphic maps (Figs. 3, 4) and high-
(Matmon et al., 2009; Guralnik et al., 2010). These ancient alluvial plains resolution optical images (e.g., ‘Google Earth’) used to delineate repre-
are characterized by very mature and well-developed desert pavements sentative surfaces (referred to herein as ‘polygons’) for the
consisting primarily of chert and carbonate gravels (Table 1). morphostratigraphic units. Two key assumptions are implied: 1) All
pixels comprising a selected polygon belong to the same
3. Materials morphostratigraphic unit, i.e. are of the same (possibly unknown) age;
and 2) the measured variability in radar backscatter is primarily driven
3.1. In-situ surface ages by variability in surface roughness, which for undisturbed desert alluvial
surfaces changes in a predictable trajectory as a function of time (Fig. 1).
The surface ages for the morphostratigraphic units we examined for Manual extraction of polygons using the high-resolution optical im-
calibration, testing and validation are all based on previous studies ages was required in order to detect and exclude various surface distur-
(Table 1). At the Shehoret and Raham sites luminescence ages were re- bances including man-made features (e.g., roads/paths, fences and
ported as surface abandonment ages (Porat et al., 2010; Mushkin et al., structures) as well as natural disturbances such as vegetation or late-
2014) and at the Paran, Risha, Mazla-Fuhud, Muhtadi, Ashosh and Shitta stage incision rills. To distinguish between depositional bar-swale line-
sites CRN surface ages were all reported as exposure ages neglecting aments and similar-scale late-stage incision rills, which commonly dis-
inherited signal and/or erosion (Matmon et al., 2009; Guralnik et al., sect mature desert pavements, we examined the plan-view pattern of
2010; Le Béon et al., 2010, 2012). Recognizing the complexities of these features in the high-resolution optical images (Fig. 3). Whereas

Table 2
ALOS-PALSAR data used.

Date Polarization Incidence angle (°) [center] Ground range resolution (m) Coverage: sites

06/06/07 HH ~24 6.25 South Arava:

19/11/07 HH ~38 6.25 Shehoret,
24/08/09 HH ~38 12.5 Raham,
24/08/09 HV ~38 12.5 Muhtadi
19/11/07 HH ~34 6.25 Central Arava: Paran, Hiyyun, Risha, Mazla, Ashosh, Shitta
02/11/07 HH ~34 6.25 North Arava: Fuhud
G. Hetz et al. / Remote Sensing of Environment 184 (2016) 288–301 293

active fluvial channels and bar-swale features typically display a distrib-

utary pattern, late-stage incision rills commonly display a dendritic pat-
tern reflecting down-slope concentration of run-off into increasingly
higher order channels (e.g., Christenson and Purcell, 1985). Polygons
were delineated to exclude pixels affected by late-stage incision rills.
To determine the optimal radar acquisition configuration for surface
age inversions, we projected the polygons selected at the Shehoret site
onto four different ALOS-PALSAR data configurations (Table 2). Statisti-
cal characteristics for the distribution of R values within the polygons,
which include the histogram, mean, median, mode, standard deviation
(1σ), kurtosis, range, maximum and minimum dB, were calculated for
each image configuration and the co-variance of each statistical param-
eter with surface age was examined (Table S1).

4.2. Sensitivity analyses

We examined the impact of measurement noise in the ALOS PALSAR

images, polygon size, natural variability of surface roughness and litho-
logical differences on surface-age inversions. ‘Leave-one-out’ cross-val-
idation (Shao, 1993) was employed to evaluate the confidence bounds
for the empirical relation (‘calibration curve’) established between
radar backscatter and surface age within the Arava region (Section
5.2.4 below).

4.3. Validation

Four previously dated morphostratigraphic units at three additional

sites within the Arava valley, which were not included in the calibration
curve, were examined for validation purposes. At these validation sites
ALOS PALSAR radar polygons were extracted following the methodolo-
gies described above and the radar-inferred surface ages were then
compared to the published in-situ surface ages of the respective unit.
Considerations for radar polygon selection (Section 4.1) differ from
those that typically drive sample collection for in-situ dating. These dif-
ferences preclude straight forward geographic linkage between the in-
situ dating location and radar backscatter from the immediate area sur-
rounding it. Instead, we compared between the radar age estimate for
the morphostratigraphic unit and the age assigned to the unit from in-
situ dating.

4.4. Application of radar-based surface dating

The radar-based surface dating approach described herein was used

to estimate the age of tectonically offset units F6 and F7 at the Risha-
Mazla site. As previous in-situ dating of these units was unsuccessful
(Le Béon et al., 2012) this study provides the first quantitative age esti-
mate for tectonically deformed alluvial deposits older than ~ 320 ka
along the DST.

5. Results

5.1. Radar backscatter vs. surface age

5.1.1. Shehoret
Field-based mapping at the Shehoret site yielded the
morphostratigraphic map presented in Fig. 3, which outlines the extent
of the Q4, Q3, Q1b and Q1a units previously dated by Porat et al. (2010)

Fig. 3. A) The Shehoret site. B) Geomorphic map overlain on a subset of A (white box in A)
and representative polygons (in yellow) for the four morphostratigraphic units examined.
Active fluvial features and bar-swale lineaments are marked in white. Late-stage incision
rills are marked in blue. C) ALOS-PALSAR image (HH, 38°, 6.25 m/pixel) of the same area
as B with projected polygons from B. D) Histograms for backscatter values within the
radar polygons in C. The median (‘Me’; red) and average (blue) consistently decrease as
the surfaces become older. Envelopes mark best-fit normal distributions.
294 G. Hetz et al. / Remote Sensing of Environment 184 (2016) 288–301

Fig. 4. Radar backscatter and surface age along the Arava valley. Left – R histograms for each of the examined units. All units display normally distributed R values (marked by a thin
envelope) and AR values (solid bar) that systematically decrease with increasing surface ages (determined with independent in-situ dating methods; Table 1). Clast-composition at the
respective sites is marked by black circles (Carbonate), grey triangles (Magmatics) and black squares (Chert-Carbonate). Top Right - Google Earth image, morphostratigraphic map
(modified from Crouvi et al., 2006) and radar image for the Raham site. Center right - Google Earth image, morphostratigraphic map (modified from Le Béon et al., 2012) and radar
image for the Risha-Mazla site. White boxes outline the Mazla (north) and Risha (south) fan complexes within the Risha-Mazla site. Bottom right - Google Earth and radar images for
the Paran site.

to 5, 14, 62 and 87 ka, respectively. Regardless of surface age we found Of the statistical parameters examined AR and Me displayed the most
that R values within the representative polygons typically display a nor- consistently monotonic relation to surface age (Fig. S1). Optimal separa-
mal distribution meeting the Kolmogorov-Smirnov test (Fig. 3). All tion between units was obtained using the HH polarization, 38° inci-
polygons display a slight negative skewness manifested by a small dence angle and 6.25 m/pixel image configuration, in which AR values
shift of the median (Me) towards higher values in respect to the average of −12.3 ± 3.3 dB (1σ) for the 5 ka surface decrease to −13.5 ± 3.8,
(AR). ΔAR-Me, defined here as 100∙(AR − Me) ∕ AR and used to quantify −16.5 ± 3.2 and −17.6 ± 3.6 dB for the 14, 62 and 87 ka units, respec-
the minor negative skewness of R values, remains ~constant regardless tively (Fig. 3). Considering that all pixels within a given polygon are as-
of surface age and ranges between 1.5 and 2.7%. sumed to be of the same age (Section 4.1), uncertainties in AR are
 G. Hetz et al. / Remote Sensing of Environment 184 (2016) 288–301 295

regarded as the standard error of the mean rather than the standard de- and T measured with CRN/luminescence (Fig. 5), can be approximated
viation of R values within the polygon. as power law function with a fairly robust fit of R2 = 0.95:

5.1.2. Raham AR ¼ −3:81  log10 ðT Þ−9:69 ð2Þ

Polygons were extracted for three morphostratigraphic units
mapped at this site by Crouvi et al. (2006) and previously dated as Ho- where AR is in units of dB and T is in ka. Uncertainty in T is regarded as
locene, 24 ka and 56 ka (Fig. 4). AR values derived after projection of the the error associated with the in-situ age estimates and uncertainties in
polygons onto an HH polarization, 38° incidence angle and 6.25 m/pixel AR are estimated as the standard error of the mean (Section 5.1.1).
ALOS PALSAR image, display an age-progressive decrease from
− 12.8 ± 3.1 dB for the Holocene unit to − 14.23 ± 3.4 dB for the 5.2. Sensitivity analyses
24 ka unit and − 15.4 ± 3.1 dB for the 56 ka unit. The histograms of
the Raham polygons displayed similar patterns to those of Shehoret 5.2.1. Measurement noise
and also met the Kolmogorov-Smirnov test for normal distribution A large polygon (N = 730) from a homogeneous dry segment of the
with a slight negative skewness of ΔAR-Me = 2.6, 1.8 and 2.6% for the Yotvata playa was examined in order to estimate the noise level of the
Holocene, 24 ka and 56 ka units, respectively. ALOS PALSAR data used (Fig. S2). R values for the homogeneous playa
surface were normally distributed with a standard deviation of
5.1.3. Risha and Mazla-Fuhud ± 3.4 dB about the average and a minor negative skewness of ΔAR-
Polygons for morphostratigraphic units F0–1 (Holocene), F2 (48 ± Me = 1.5%. It therefore appears that the characteristic normal distribu-
5 ka), F3 (95 ± 25 ka), F4 (160 ± 20 ka) and F5 (320 ± 24 ka) mapped tion of R values about AR, the age-invariant standard deviations and
and dated by Le Béon et al. (2012) were delineated and projected onto the minor negative skewness we found for the fifteen Quaternary allu-
ALOS PALSAR image acquired at HH polarization, 38° incidence angle vial surfaces examined (Figs. 3, 4) primarily reflects measurement
and 6.25 m/pixel (Fig. 4). AR values systematically decrease from noise rather than physical roughness variability on the surface.
− 13.3 ± 3.8 dB for F0, 1 to − 14.4 ± 3.7 dB for F2, − 18.1 ± 3.7 dB
for F3, −18.8 ± 3.5 dB for F4 and −20.3 ± 3.5 dB for F5. The histograms 5.2.2. Polygon selection and roughness variability
of all the Risha and Mazla-Fuhud polygons meet the Kolmogorov- The dependence of AR values on polygon size (N) is plotted in Fig. 6a.
Smirnov test for normal distribution with a slight negative skewness We find that for both the rough Holocene unit at Risha and the smooth
ΔAR-Me = 1.2–3.2%. 1.6 Ma surface at Paran (not shown), AR values typically converge to a
constant value at N ≥ 70. We therefore regard N = 70 as the threshold
5.1.4. Paran polygon size required to capture the natural roughness heterogeneity
Polygons for the previously dated (Matmon et al., 2009: Guralnik et of desert alluvial units with the ALOS PALSAR data used (6.25 m/
al., 2010) surfaces at Paran were projected onto an HH polarization, 34° pixel). We also found that regardless of surface age (i.e., roughness)
incidence angle, 6.25 m/pixel ALOS PALSAR image available for this site intra-unit variability of AR values were typically within ±~0.5 dB (Fig.
(Fig. 4). AR values of −21.0 ± 3.6 dB for the 1160 ka surface decreased 6) and comparable to the standard error of the mean.
to −22.1 ± 3.4 dB for both the 1610 ka and 1940 ka surfaces. The histo- All the polygons examined regardless of surface age, display normal-
grams of the Paran polygons also display a normal distribution with a ly distributed dB values, standard deviations that range between 3.1 and
slight negative skewness of ΔAR-Me = 1.7–1.9%. 3.8 dB, and a minor negative skewness within a limited range of ΔAR-
Me = 1.5–3.0%. As these appear to be the statistical characteristics of
5.1.5. Radar backscatter vs. surface age along the Arava valley ALOS PALSAR returns from natural, undisturbed pavement surfaces in
Average radar backscatter values for the fifteen morphostratigraphic the study area we require R distribution within polygon used for testing,
units examined along the Arava region monotonically decrease as a validation and/or surface dating described below to meet comparable
function of surface age (‘T’) (Fig. 4). The relation between these two in- statistical characteristics. Polygons that did not meet these characteris-
dependently measured surface attributes, i.e., AR measured with radar tics were discarded.

5.2.3. Lithology
The primary physical processes that govern the time-progressive
smoothing of young boulder-strewn alluvial deposits into mature and
smooth desert pavements in the Arava region are: 1) In-situ, subaerial
weathering of surface clasts (Wells et al., 1995), 2) gravity-driven diffu-
sion of material from bars into swales (e.g., Matmon et al., 2006) and 3)
accumulation of eolian dust in the Av horizon directly below the surface
clasts (McFadden et al., 1987). Whereas the latter two processes are not
expected to be significantly affected by clast lithology, weathering rates
of surface clasts likely vary with lithology. To evaluate the effect of this
compositional dependence on the decrease of AR over time we examined
chronosequences consisting of three distinct lithological assemblages:
predominately carbonate at Shehoret and Raham, mainly magmatic at
Risha and Mazla-Fuhud and chert-carbonate at Paran (Table 1).
We found that despite the variable clast lithology of the fifteen
morphostratigraphic units examined all plot along a common AR-T
curve with a robust R2 of 0.95. This 2nd order dependence of the AR-T
regression on surface composition is further illustrated in Fig. S3,
where separate AR-T regressions constrained for the carbonate-chert,
Fig. 5. Average Radar backscatter vs. surface age along the Arava valley. Surface ages were carbonate-magmatic and magmatic-chert units compare to the all-in-
independently determined with in-situ methods (Table 1). Squares mark the youngest
morphostratigraphic units examined at Raham and Risha-Mazla, which were classified
clusive regression in Fig. 5. It therefore appears that for the Arava valley
as Holocene (i.e., younger than 10 ka) by Crouvi et al. (2006) and Le Béon et al. (2012), and the specific lithologies examined, surface composition exerts only a
respectively. secondary effect on the observed decrease in AR values as a function of T.
296 G. Hetz et al. / Remote Sensing of Environment 184 (2016) 288–301

Fig. 6. Sensitivity analyses. A) Left – ALOS PALSAR image with different size polygons (white boxes) for the Risha Holocene (F0–1) unit. Right - AR values as a function of polygon size from
the radar image on the left. AR values converge to a constant value for polygons larger than ~70 pixels. Similar analysis for the smooth 1610 ka surface at Paran (not shown here) displayed
a similar pattern. B–D) Multiple polygons (N N 100) extracted from the Shehoret Qa3 (14 ka), the Risha F3 (95 ka) and the Paran PS4 (1610 ka) surfaces demonstrate the limited variability
of AR values (solid bar) within each unit.

5.2.4. Regression uncertainty curve translating AR measurements into quantitative T estimates for al-
‘Leave-one-out’ cross-validation (preformed using the fifteen data luvial morphostratigraphic units along the Arava valley:
points in Fig. 5) indicates an average uncertainty of 34% for predictions
based on this regression. Accordingly, we assume a 35% uncertainty in
the radar-inferred surface-age estimates we obtain here for the Arava T ¼ 10
ðARþ9:69Þ
=−3:81 ð3Þ
region.

where T is in ka and AR is in units of dB. Radar-based age estimates ob-

5.3. Validation tained using this equation for the four morphostratigraphic units at the
three validation sites described below, were found to be all within mea-
Based on the robust regression established between AR and T (Fig. 5) surement uncertainty of previously published CRN ages for these units
we propose that Eq. (2) can be inverted to yield a regional calibration (Fig. 7).
 G. Hetz et al. / Remote Sensing of Environment 184 (2016) 288–301 297

5.3.1. Muhtadi Béon et al. (2012) that sinistral slip rates of ~ 5–10 mm/yr they mea-
A sequence of three morphostratigraphic units (F1–3) was previous- sured for DST since ~320 ka extend out to ~1 Ma time-scales.
ly mapped at this site (Le Béon et al., 2010) located along the eastern
side of the Arava valley across from the Shehoret site (Fig. 2). Le Béon 6. Discussion
et al. (2010) reported a CRN exposure age of 56 ± 14 for unit F3 and a
Holocene age for unit F1 (Fig. 7B). Nine radar polygons selected for 6.1. Radar backscatter as a proxy for surface age
unit F3 yielded AR values tightly clustered between − 16.7 and
− 17.6 dB with an average of − 17.2 dB. Using Eq. (3) we obtain a Measurement noise in the ALOS PALSAR images used (i.e., ~ ±
radar estimated age of 91 ± 32 ka for unit F3. Five radar polygons select- 3.4 dB; Fig. S1) spans up to ~two thirds of the ~9 dB difference found be-
ed for unit F1 yielded AR values tightly clustered between −11.2 and tween the average backscatter value from rough Holocene units to
− 12.2 dB with an average of − 11.6 dB. Using Eq. (3) we obtain a smooth 1600–1900 ka desert pavements (Fig. 5). This low signal-to-
radar estimated age of 3 ± 1 ka for unit F1. noise ratio implies that the R value of a random radar pixel from an al-
luvial surface can be statistically associated with a large range of
5.3.2. Ashosh morphostratigraphic units. Consequently, the value of an individual
This site is located in the central Negev desert ~ 15 km west of the radar pixel in itself has little significance in terms of inferring the age
Arava valley (Fig. 2). Guralnik et al. (2010) constrained a CRN exposure of the morphostratigraphic unit it is on. In contrast, we find that the av-
age of 540 ± 60 ka for this abandoned alluvial surface (Fig. 7C). Five erage R from a morphostratigraphic unit (i.e., AR) systematically de-
radar polygons selected for this unit yielded AR values tightly clustered creases as a power-law function of time (Fig. 5) and can therefore be
between −19.6 and −20.2 dB with an average of −19.9 dB. Using Eq. effectively utilized as a proxy for surface age.
(3) we obtain a radar estimated age of 484 ± 169 ka. Power-law relations are commonly found in a wide range of time-
dependent geomorphic processes. In the case examined here it appears
5.3.3. Shitta that the measured power-law decrease of AR as a function of surface age
This site is located in the southern Negev desert ~20 km west of the likely relates to a complex combination of two surface processes: The
Arava valley (Fig. 2). Guralnik et al. (2010) obtained a CRN exposure age non-linear decrease in the rate of gravity-driven diffusion of material
of 1590 ± 250 ka for this abandoned alluvial surface, which is character- from bars into swales, which reduces bar-swale relief over time
ized by a well-developed desert pavement dominated by chert clasts (Matmon et al., 2006), and to the power-law decrease in subaerial
(Fig. 7D). Six radar polygons selected for this surface yielded AR values rock weathering rates over time, which was specifically measured at
clustered between −21.4 and −22.4 dB with an average of −21.9 dB. the Shehoret and Raham sites using ground-based LiDAR (Mushkin et
Using Eq. (3) we obtain a radar estimated age of 1604 ± 561 ka. al., 2014).
Recognizing that surface smoothing rates of desert alluvial surfaces
5.4. Radar age estimates for undated morphostratigraphic units in the can vary as a function of additional physical drivers, such as salt
Arava valley weathering (e.g., Farr and Chadwick, 1996), concentration of resistant
lithologies (Gillespie et al., 1984) and climate, we point towards the re-
In their study of lateral tectonic slip rates along the DST Le Béon et al. gion-specific nature of the AR-T calibration curve tested here for the
(2012) obtained CRN model exposure ages for units F2–5 at the Risha- Arava valley (Fig. 5). Application of the proposed surface dating ap-
Mazla sites (Fig. 2; Table 1) in order to constrain tectonic lateral offset proach in other locations and/or with other radar data sets likely re-
rates for the DST. As units F6 and F7 could not be successfully dated quires region/sensor-specific calibrations and sensitivity analyses as
with CRN, DST Pleistocene lateral offset rates beyond ~ 320 ka (unit described above.
F5) remain loosely constrained. Below we use radar-based dating to de- Application of the surface dating approach presented here could
termine the age of units F6 and F7. conceivably be also extended to other remote sensing data that re-
Ten polygons selected for unit F6 at Mazla (Fig. 8) displayed AR spond to surface roughness, e.g., airborne LiDAR. However, consider-
values that tightly clustered between −20.8 and −20.1 dB with an av- ing that the roughness of un-disturbed mature desert pavements is
erage of −20.5 dB. All polygons display normally distributed R values, primarily governed by the size of their surface clasts, which are typ-
standard deviations that typically range between ~3–4 dB and a minor ically within the pebble size-range (i.e., smaller than 6.4 cm), sug-
negative skewness. AR values for unit F6 translate to a surface-age esti- gests that radar data are likely to be most suitable for capturing
mate of 694 ± 243 ka (Eq. (3)). The nine polygons selected within the such age-dependent roughness variations. Accordingly, we point to-
smoothest and best-preserved surfaces of the highly dissected F7 unit wards C-band (~ 6 cm) and X-band (~ 3 cm) radar data as potentially
south of Fuhud (Fig. 8) also display normally distributed R values, stan- favorable options for additional application of remote dating of ma-
dard deviations that typically range between 3 and 4 dB and a minor ture alluvial surfaces.
negative skewness. AR values for unit F7 range between − 20.4 and
−21.8 dB with an average of −21.1 dB and translate to a surface-age es- 6.2. Relation to in-situ surface dating methods and remote sensing mapping
timate of 997 ± 349 ka (Eq. (3)). approaches
The radar-based surface age estimates of 694 ± 243 and 997 ±
349 ka obtained here for units F6 and F7, respectively, are consistent Remote sensing mapping approaches/algorithms are typically
with the relative age assignment implied by the geomorphic classifica- designed to identify commonalities amongst image pixels in order
tion previously established for these units by Le Béon et al. (2012), i.e., to map and delineate boundaries between surface units. Such
older than 320 ka (F5). Whereas the radar-based age estimates for F6 ‘pixel-based’ approaches have been successfully applied for identify-
and F7 overlap with each other in absolute terms their AR values ing and mapping morphostratigraphic desert alluvial units with
imply that F6 is younger than F7, as also indicated by the field-based spaceborne thermal data (Gillespie et al., 1984), combination of
geomorphic classification of Le Béon et al. (2012). radar and optical imagery (Farr and Chadwick, 1996; Weeks et al.,
Based on the amount of slip estimated for units F6 and F7, Le Béon et 1996, 1997; Kierein-Young, 1997), stereo imaging (Mushkin and
al. (2012) suggested that ages of 510–880 and 780–1190 ka for F6 and Gillespie, 2005, 2010) airborne hyperspectral images (Crouvi et al.,
F7, respectively, would be consistent with the tectonic slip rates they 2006) and high-resolution (1 m/pixel) airborne LiDAR data
determined for the younger units at the Risha and Mazla-Fuhud sites. (Frankel and Dolan, 2007; Regmi et al., 2013). Radar-based surface
Thus, our radar-based age estimates, which are independent from dating presented herein differs from and complements these ap-
slip-rate considerations, provide new support for the hypothesis of Le proaches in that it is not designed for delineating and mapping
298 G. Hetz et al. / Remote Sensing of Environment 184 (2016) 288–301
 G. Hetz et al. / Remote Sensing of Environment 184 (2016) 288–301 299

Fig. 8. Radar dating of middle to early Pleistocene units at the Mazla and Fuhud sites. A) Google Earth image. B) Subset of the geomorphic map published for this site by Le Béon et al.
(2012). Units F6 and F7 could not be dated with CRN and were estimated by Le Béon et al. (2012) to be older than 320 ka based on their geomorphic characteristics. C) Top - Google
Earth image for the Mazla F6 surface with ten delineated and numbered polygons. Center - Polygons 1 through 10 projected onto an ALOS PALSAR image (HH, 34°, 6.25 m/pixel).
Bottom - Histograms for R values within the respective radar polygons yield an estimated surface age of 694 ± 243 ka for F6. D) Top - Google Earth image for the Fuhud F7 surface
with nine delineated and numbered polygons. Center - Polygons 1 through 9 projected onto an ALOS PALSAR image (HH, 34°, 6.25 m/pixel). Bottom - Histograms for R values within
the respective radar polygons yield an estimated surface age of 997 ± 349 ka for F7. Solid bars mark the average (AR) in all histograms.

boundaries between units, but rather as a tool for quantitatively es- surface ages. However, inversion of AR values into quantitative age esti-
timating the age of morphostratigraphic alluvial units that were in- mates requires rigorous calibration using independently obtained age
dependently identified and mapped as such. determinations, such as those obtained with in-situ dating methods.
The systematic decrease of AR values from abandoned desert alluvial Thus, radar-based surface dating does not replace in-situ dating
surfaces over Quaternary time scales (Fig. 4) offers an effective tool for methods, but rather expands on them by facilitating quantitative corre-
establishing relative chronologies amongst morphostratigraphic alluvi- lations and interpolations between dated and undated alluvial units of
al units across regional scales with lower AR values implying older interest across regional scales.

Fig. 7. Validation. A) Surface ages for four Quaternary surfaces along the Arava valley inferred from AR values (using Eq. (3)) plotted against CRN age estimates for the same surfaces. B) The
Muhtadi site. Top - The sequence of three morphostratigraphic units (F1–3) previously mapped at this site (modified from Le Béon et al., 2010) and their CRN age estimates (from Le Béon
et al., 2010). Left – Google Earth image with nine and five polygons delineated for units F3 and F1, respectively. Bottom – ALOS PALSAR image (HH, 38°, 6.25 m/pixel) with projected
polygons 1–9 for unit F3 and 10–14 for F1. Right - Histograms for R values within the respective radar polygons. B) The Ashosh site. Left – Google Earth image with five polygons
delineated for the morphostratigraphic unit dated by Guralnik et al. (2010) to 540 ka. Center - ALOS PALSAR image (HH, 34°, 6.25 m/pixel) with projected polygons 1–5. Right -
Histograms for R values within the respective radar polygons. D) The Shitta site. Left – Google Earth image with six polygons delineated for the morphostratigraphic unit dated by
Guralnik et al. (2010) to 1590 ka. Center - ALOS PALSAR image (HH, 34°, 6.25 m/pixel) with projected polygons 1–6. Right - Histograms for R values within the respective radar
polygons. Solid bar marks AR for all histograms.
300 G. Hetz et al. / Remote Sensing of Environment 184 (2016) 288–301

The uncertainty in the radar-based age estimates (~35%) is signifi- Amit, R., Harrison, J.B.J., Enzel, Y., Porat, N., 1996. Soils as a tool for estimating ages of Qua-
ternary fault scarps in a hyperarid environment- The Southern Arava valley, the Dead
cantly higher than the typical range of ~ 10% uncertainty reported for Sea Rift, Israel. Catena 28, 21–45.
in-situ dating methods. However, the ability to quantitatively deter- Amit, R., Enzel, Y., Sharon, D., 2006. Permanent Quaternary hyperaridity in the Negev, Is-
mine surface ages from space and over large areas although with a rael, resulting from regional tectonics blocking Mediterranean frontal systems. Geol-
ogy 34 (6), 509–512.
somewhat higher uncertainty complements the ‘point’ nature of Bull, B.W., 1977. The alluvial fan environment. Prog. Phys. Geogr. 1, 222–270.
ground-based age estimates, which may be more accurate but are in- Bull, B.W., 1991. Geomorphic responses to climatic change. Oxford University press.
herently associated with the local scale of the sample. Radar-based sur- Christenson, G.E., Purcell, C., 1985. Correlation and age of Quaternary alluvial fan se-
quences, Basin and Range province, southwestern United States. Soils and quaternary
face dating may therefore also be used to identify surfaces of particular geology of the southwestern United StatesGSA Special Paper 203. The Geological So-
interest for further examination with more accurate in-situ dating. ciety of America, Boulder, Colo., pp. 115–122.
Crouvi, O., Ben-Dor, E., Beyth, M., Avigad, D., Amit, R., 2006. Quantitative mapping of arid
alluvial fan surfaces using field spectrometer and hyperspectral remote sensing. Re-
7. Summary and conclusion mote Sens. Environ. 104, 103–117.
Evans, D.L., Farr, T.G., van Zyl, J.J., 1992. Estimates of surface roughness derived from syn-
thetic aperture radar (SAR) data. IEEE Trans. Geosci. Remote Sens. 30 (2), 382–389.
This study explored the use of ALOS PALSAR backscatter data to ob- Farr, T.G., 1992. Microtopographic evolution of lava flows at Cima volcanic field, Mojave
tain quantitative estimates for the age of abandoned desert alluvial sur- Desert, California. J. Geophys. Res. 97, 15171–15179.
faces along the Arava region of the Dead Sea Transform. Our results Farr, T.G., Chadwick, O.A., 1996. Geomorphic processes and remote sensing signatures of
alluvial fans in the Kun Lun Mountains, China. J. Geophys. Res. 101, 23091–23100.
demonstrate that whereas characteristic noise levels in radar images Frankel, K.L., Dolan, J.F., 2007. Characterizing arid region alluvial fan surface roughness with
preclude effective surface-age inversions from the backscatter value of airborne laser swath mapping digital topographic data. J. Geophys. Res. 112, 1–14.
individual pixels, the average backscatter value (‘AR’) from abandoned Garfunkel, Z., Ben Avraham, Z., 1996. The structure of the Dead Sea basin. Tectonophysics
226, 155–176.
alluvial surfaces along the Arava Valley monotonically decreases as a
Garfunkel, Z., Zak, I., Freund, R., 1981. Active faulting in the Dead Sea rift. Tectonophysics
power-law function of surface age from ~− 13 dB for Holocene units 80, 1–26.
to ~−22 dB for 1.6–1.9 Ma surfaces. Gillespie, A.R., Kahle, A.B., Palluconi, F.D., 1984. Mapping alluvial fans in Death Valley, Cal-
ifornia, using multispectral thermal infrared images. Geophys. Res. Lett. 11 (11),
To determine the AR of a given surface we used a pixel aggregation
1153–1156.
approach that requires independent delineation/mapping of the Ginat, H., Enzel, Y., Avni, Y., 1998. Translocated Plio-Pleistocene drainage systems along
morphostratigraphic unit and manual extraction of representative poly- the Arava fault of the Dead Sea transform. Tectonophysics 284, 151–160.
gons consisting of at least ~70 radar pixels from undisturbed areas on Guralnik, B., Matmon, A., Avni, Y., Fink, D., 2010. Be exposure ages of ancient desert pave-
ments reveal Quaternary evolution of the Dead Sea drainage basin and rift margin
the surface. Rigorous evaluation criteria for the distribution of backscat- tilting. Earth Planet. Sci. Lett. 290, 132–141.
ter values about the AR of the polygon were employed to test the integ- Guralnik, B., Matmon, A., Avni, Y., Porat, N., Fink, D., 2011. Constraining the evolution of
rity of each selected polygon. A regional-scale calibration curve river terraces with integrated OSL and cosmogenic nuclide data. Quat. Geochronol.
6, 22–32.
regressing the AR of fifteen morphostratigraphic units against their sur- Kierein-Young, K.S., 1997. The integration of optical and radar data to characterize miner-
face age as determined with independent in-situ dating methods was alogy and morphology of surfaces in Death Valley, California, USA. Int. J. Remote Sens.
constructed to allow radar-based dating of abandoned alluvial surfaces 18, 1517–1541.
Le Béon, M., Klinger, Y., Al-Qaryouti, M., Mériaux, A.S., Finkel, R.C., Elias, A., Mayyas, O.,
in this study area. Sensitivity analyses and validation experiments dem- Ryerson, F.J., Tapponnier, P., 2010. Early Holocene and late Pleistocene slip rate of
onstrated that AR values could be inverted to obtain surface ages rang- the southern Dead Sea fault determined from 10Be cosmogenic dating of offset alluvi-
ing from Holocene to 1.6 Ma with uncertainty levels of ~35%. al deposits. J. Geophys. Res. 115, B11414.
Le Béon, M., Klinger, M.Y., Mériaux, A., Al-Qaryouti, M.R.C.F., Mayyas, O., Tapponnier, P.,
Global-scale coverage of radar image data and high-resolution opti- 2012. Quaternary morphotectonic mapping of the Wadi Araba and implications for
cal imagery (e.g., Google Earth) facilitate application of radar-based sur- the tectonic activity of the southern Dead Sea fault. Tectonics 31, TC5003.
face dating in additional desert environments worldwide. This new Matmon, A., Nichols, K., Finkel, R., 2006. Isotopic insights into smoothening of abandoned
fan surfaces, Southern California. Quat. Res. 66, 109–118.
ability to understand the age variability of abandoned alluvial surfaces
Matmon, A., Simhai, O., Amit, R., Haviv, I., Enzel, Y., Porat, N., McDonald, E.V., Benedetti, L.,
across large areas now offers a unique regional perspective to better de- Finkel, R.C., 2009. Desert pavement-coated surfaces in extreme deserts present the
cipher preserved alluvial records in desert environments in the context longest-lived landforms on Earth. Geol. Soc. Am. Bull. 121, 688–697.
of past climatic changes as well as regional-scale tectonic and landscape McDonald, E.V., McFadden, L.D., Wells, S.G., 2003. Regional response of alluvial fans to the
Pleistocene-Holocene climatic transition, Mojave Desert, California. GSA Special Pa-
evolution processes. pers 368, 189–206.
McFadden, L.D., Wells, S.G., Jercinovic, M.J., 1987. Influences of eolian and pedogenic
processes on the evolution and origin of desert pavements. Geology 15, 504–508.
Acknowledgements McFadden, L.D., Ritter, J.B., Wells, S.G., 1989. Use of multiparameter relative-age methods
for age estimation and correlation of alluvial fan surfaces on a desert piedmont, east-
ern Mojave Desert, California. Quat. Res. 32, 276–290.
The authors thank E. Trabelci for his assistance with field work and Mushkin, A., Gillespie, A.R., 2005. Estimating sub-pixel surface roughness using remotely
the Dead Sea and Arava Science Center for supporting field efforts in sensed stereoscopic data. Remote Sens. Environ. 99, 75–83.
Jordan. We are also greatly thankful to the comments by B. Guralnik, Mushkin, A., Gillespie, A.R., 2010. Using ASTER stereo images to quantify surface rough-
ness. In: Ramachandran, B., Justice, C., Abrams, M. (Eds.), Invited chapter in – Land re-
T. Farr, an anonymous reviewer and the editorial staff at RSE, which
mote sensing and global environmental change: NASA's earth observing system and
significantly improved the quality and clarity of the manuscript. This re- the science of ASTER and MODIS. Springer-Verlag Publishing.
search was funded by Israel Science Foundation (ISF) grant #12206/09 Mushkin, A., Sagy, A., Trabelci, E., Amit, R., Porat, N., 2014. Measuring the time and scale-
to Mushkin and Baer. ALOS PALSAR images were provided by the dependency of subaerial rock-weathering rates over geologic timescales with
ground-based LiDAR. Geology 42 (11), 1063–1066.
European Space Agency (ESA) under project C1P.5544. Noller, J.S., Sowers, J.M., Lettis, W.R., 2000. Quaternary geochronology- Methods and ap-
plications. Wiley.
Noy-Meir, I., 1973. Desert ecosystems: Environment and producers. Annu. Rev. Ecol. Syst.
Appendix A. Supplementary data 4, 25–51.
Porat, N., Amit, R., Enzel, Y., Zilberman, E., Avni, Y., Ginat, H., Gluck, D., 2010. Abandon-
ment ages of alluvial landforms in the hyperarid Negev determined by luminescence
Supplementary data to this article can be found online at http://dx. dating. J. Arid Environ. 74, 861–869.
doi.org/10.1016/j.rse.2016.07.006. Regmi, N.R., McDonald, E.V., Bacon, S.N., 2013. Mapping quaternary alluvial fans in the
southwestern United States based on multi-parameter surface roughness of LiDAR
topographic data. J. Geophys. Res. Earth Surf. 119, 12–27. http://dx.doi.org/10.1002/
References 2012JF002711.
Shao, J., 1993. Linear model selection by cross-validation. J. Am. Stat. Assoc. 88 (422),
Amit, R., Gerson, R., 1986. The evolution of Holocene Reg (gravelly) soils in deserts– An 486–494.
example from the Dead Sea region. Catena 13, 59–79. Ulaby, F.T., Batlivala, P.P., Dobson, M.C., 1978. Microwave backscatter dependence on sur-
Amit, R., Gerson, R., Yaalon, D.H., 1993. Stages and rate of the gravel shattering process by face roughness, soil moisture, and soil texture: Part 1- Bare soil. IEEE Trans. Geosci.
salts in desert Reg soils. Geoderma 57, 295–324. Remote Sens. 16 (4), 286–295.
 G. Hetz et al. / Remote Sensing of Environment 184 (2016) 288–301 301

Ulaby, F.T., Moore, R.K., Fung, A.K., 1982. Microwave remote sensing: Active and passive, Wells, S.G., McFadden, L.D., Dohrenwend, J.C., 1987. Influence of late quaternary climatic
2, surface scattering and emission theory. Addison-Wesley, Reading, MA. change on geomorphic and pedogenic processes on a desert piedmont, eastern
Weeks, R.J., Smith, M., Pak, K., Li, W.H., Gillespie, A.R., 1996. Surface roughness, radar back- Mojave Desert, California. Quat. Res. 27, 130–146.
scatter, and visible and near-infrared reflectance in Death Valley, California. Wells, S.G., McFadden, L.D., Poths, J., Olinger, C.T., 1995. Cosmogenic 3He surface-expo-
J. Geophys. Res. 101, 22,077–23,090. sure dating of stone pavements: Implications for landscape evolution in deserts. Ge-
Weeks, R.J., Smith, M., Pak, K., Gillespie, A.R., 1997. Inversions of SIR-C and AIRSAR data for ology 23 (7), 613–616.
the roughness of geological surfaces. Remote Sens. Environ. 59 (2), 383–396.