Content uploaded by Joshua King
Author content
All content in this area was uploaded by Joshua King on Mar 26, 2015
Content may be subject to copyright.
Spatio-temporal influence of tundra snow properties on Ku-band
(17.2 GHz) backscatter
Joshua KING,1Richard KELLY,1Andrew KASURAK,1Claude DUGUAY,1
Grant GUNN,1Nick RUTTER,2Tom WATTS,2Chris DERKSEN3
1Department of Geography, University of Waterloo, Waterloo, Ontario, Canada
2Department of Geography, Northumbria University, Newcastle upon Tyne, UK
3Climate Research Division, Environment Canada, Toronto, Ontario, Canada
Correspondence: Joshua King <jmking@uwaterloo.ca>
ABSTRACT. During the 2010/11 boreal winter, a distributed set of backscatter measurements was
collected using a ground-based Ku-band (17.2 GHz) scatterometer system at 26 open tundra sites. A
standard snow-sampling procedure was completed after each scan to evaluate local variability in snow
layering, depth, density and water equivalent (SWE) within the scatterometer field of view. The shallow
depths and large basal depth hoar encountered presented an opportunity to evaluate backscatter under
a set of previously untested conditions. Strong Ku-band response was found with increasing snow depth
and snow water equivalent (SWE). In particular, co-polarized vertical backscatter increased by 0.82 dB
for every 1 cm increase in SWE (R2= 0.62). While the result indicated strong potential for Ku-band
retrieval of shallow snow properties, it did not characterize the influence of sub-scan variability. An
enhanced snow-sampling procedure was introduced to generate detailed characterizations of
stratigraphy within the scatterometer field of view using near-infrared photography along the length
of a 5 m trench. Changes in snow properties along the trench were used to discuss variations in the
collocated backscatter response. A pair of contrasting observation sites was used to highlight
uncertainties in backscatter response related to short length scale spatial variability in the observed
tundra environment.
KEYWORDS: remote sensing, snow
INTRODUCTION
In the microwave spectrum, changes to the dielectric and
physical properties of snow can influence directional
scattering components, providing tangible quantities from
which to derive information (Ulaby and others, 1984). Given
the dynamic nature of terrestrial snow accumulation and
metamorphosis, the measured microwave response is not
only a function of aggregate snow properties such as snow
water equivalent (SWE), but often of stratigraphy and lateral
snowpack heterogeneity (Colbeck, 1991). In practical
application, the complexities of these interactions are
difficult to account for when scales of spatial variability are
smaller than the spacing and/or support of the in situ
observations used for interpretation. The influence of snow
property variability local to an observing instrument can be
significant where placement and thickness of dielectric
discontinuities can dominate observed backscatter or bright-
ness temperature (e.g. Marshall and others, 2007; Montpetit
and others, 2013). Improving understanding of these local-
scale interactions specific to distinct snow-cover types is an
important step towards the development of robust methods
for microwave-based snow property retrieval.
Complex spatio-temporal patterns are common to snow
around the world (Deems and others, 2008; Jonas and others,
2009; Scipión and others, 2013), but tundra environments in
particular present a challenging target of analysis, with
numerous environmental agents acting on shallow deposi-
tions of snow on an often rough underlying soil and shrub
surface (Derksen and others, 2009; Sturm and Wagner,
2010; Domine and others, 2012). In general, tundra
environments north of the treeline are characterized by low
air temperature, exposure to high wind, and little precipi-
tation input (Sturm and others, 1995). Under these condi-
tions, tundra snowpack in open areas forms with contrasting
depth-hoar and wind-slab components. Dry snow subjected
to strong vertical temperature gradients will undergo
metamorphosis, redistributing mass by processes of vapour
transport and subsequent grain growth (Sturm and Benson,
1997). Basal layers in tundra environments are often
dominated by this process, composed of recrystallized snow
microstructure, with grain diameters in excess of 2 mm, and
preferential orientation. When sustained winds are present,
deposition is quickly redistributed, forming fine-grain
(<1 mm) crust and slab features. Over short distances,
changes in wind and topography can produce lateral
heterogeneity of <1 m up to >100 m (Sturm and Benson,
2004). Electromagnetically, these local variations in bulk and
stratigraphic properties have the potential to generate diverse
microwave responses from outwardly simplistic targets of
similar bulk characteristics (e.g. depth, density, SWE).
A number of purpose-built systems for ground-based radar
observation of snow and ice have been introduced in recent
years to address questions regarding local-scale interaction
(e.g. Marshall and Koh, 2008; Willatt and others, 2010; King
and others, 2013; Morrison and Bennett, 2014). These
systems have been successfully deployed in support of large
field campaigns, including the Cold Land Processes Experi-
ment (CLPX; Marshall and others, 2004), and under-flight
testing of proposed satellite missions, such as CoReH
2
O (Rott
and others, 2010; Chang and others, 2014). Despite recent
Journal of Glaciology, Vol. 61, No. 226, 2015 doi: 10.3189/2015JoG14J020 267
advancements in snow-radar observation, evaluation of the
response from snow and substrate in tundra-specific condi-
tions remains incomplete. During the 2010/11 boreal winter,
the Canadian Snow and Ice Experiment (CASIX) was initiated
to collect coincident backscatter and snow property meas-
urements in a previously unevaluated sub-arctic environ-
ment near Churchill, Manitoba, Canada. The field-observed
dataset provides a unique opportunity to examine radar
sensitivity to snow properties under challenging conditions
including shallow depth and prevalent depth hoar.
In this study, we present two destructive sampling
procedures to evaluate backscatter against in situ snow
properties for the purpose of quantifying and evaluating Ku-
band (17.2 GHz) sensitivity in a unique terrestrial tundra
environment. As a first case, traditional pit and bulk snow
measurements are made within the field of view of an
observing radar instrument at a spatially distributed set of
snow-covered open tundra sites. Summary and statistical
analysis are used to identify and discuss potential drivers of
backscatter variability and to quantify Ku-band sensitivity to
selected snow properties, including SWE. A second case
study presents an enhanced observation protocol using
trench excavation and near-infrared (NIR) photography to
characterize snow stratigraphy within the radar field of
view. Observed heterogeneity of snow properties and
stratigraphy are discussed in relation to coincident radar
returns measured across the length of the trench. Two sites
of contrasting snowpack composition are used to identify
potential drivers of backscatter variability and discuss future
direction for study. Finally, the observed spatio-temporal
backscatter is discussed in the context of previous field
studies and electromagnetic models to relate physical
contributions to observed backscatter.
STUDY AREA
Churchill (58.7692° N, 94.1692° W) is located on the
southwest shore of Hudson Bay at the mouth of the Churchill
River. Proximity to the arctic treeline divides the local
environment into a number of distinct tundra/forest transition
zones, each with characteristics typical of a larger domain
within the circumpolar north (Kershaw and McCulloch,
2007). Access to a number of distinct environments over
short distances made Churchill an ideal location to satisfy the
diverse observational requirements of CASIX. Measurements
collected as part of this study were made in an area to the
east of Churchill, primarily composed of open areas (62%)
with smaller portions occupied by forest (27%) and lake
(11%) features (Derksen and others, 2012). Large expanses of
graminoid and shrub tundra were found throughout the study
area, with limited vegetation height (<30 cm), little topo-
graphic relief (0–30 m a.s.l.), and underlying organic soils
(5–30 cm in depth) common among them. The prevailing
climate conditions in Churchill are best described as sub-
arctic, where strong wind (>5 m s1), low air temperature
(<–20°C) and limited snowfall (201 cm) are defining char-
acteristics of the winter accumulation period.
Backscatter measurements were made between 15 No-
vember 2010 and 1 March 2011 at a distributed set of snow-
covered open tundra sites along the Hudson Bay coast
(Fig. 1). In total, 26 independent radar measurements were
completed, along with a suite of coincident snow property
measurements to characterize physical processes and vari-
ability within the bounds of the observing instrument
footprint. By standardizing the observation protocol, a
framework for evaluation of inter- and between-site back-
scatter was established. The mixed land-cover environment
local to Churchill provided access to a variety of tundra-
class snow conditions, limited in depth by wind exposure
and lack of standing vegetation. To minimize environmental
and observational complexities, sites were kept free of
standing vegetation and anthropogenic modifiers. As such,
observations were primarily collected in graminoid-domin-
ated environments where such complexities could be
minimized. The following provides a brief overview of the
theory and methodology used to characterize the local
Fig. 1. Churchill study area, with measurement locations indicated in red.
King and others: Tundra snow 17.2 GHz backscatter268
dynamics of Ku-band radar interaction for the explicit
purpose of evaluating sensitivity to tundra snow properties.
DATA AND METHODS
Background
Radar-based snow property retrieval exploits subtle changes
in backscatter resulting from variations in accumulation and
metamorphosis to derive snowpack information without
physical contact. In terrestrial environments, a simple
conceptualization of total backscatter (tot) can be made
with snow- and soil-scattering contributions in a particular
transmit-and-receive polarization combination (pq) (Ulaby
and others, 1984; Rott and others, 2010):
tot
pq ¼as
pq þvol
pq þgv
pq þgnd
pq ð1Þ
where the air/snow interface (as), snow/ground interface
(gnd), snow volume (vol ) and higher-order interactions
among the ground and snow volume (gv) contribute to total
backscatter (Fig. 2). The magnitude of each contribution in
Eqn (1) is determined by the dielectric and physical state of
the snow-covered environment and the parameters of the
observing instrument. With parameters of the observing
instrument held constant, spatio-temporal changes in SWE
can be expected to exert influence on backscatter because
of its inherent relationship with both physical and dielectric
properties of the snowpack.
Dielectric properties of target media are commonly
described using the complex form of relative permittivity "r:
"r¼"0i"00 ð2Þ
where the real component "0describes the ability of a
material to polarize and store energy and the imaginary
component "00 is a loss factor describing dissipation of
energy, both relative to that of air. For dry snow, values of
the imaginary component of permittivity are generally low
at microwave frequencies ("0<2 and "00 <103), resulting
in negligible absorption of incident energy, dominant
scattering losses, and potential for penetration through
meters of snow. If the observed snow becomes wet, the
presence of liquid water causes increased reflectivity at the
snow surface ("020–80), greatly increased attenuation
within the snow volume ("00 5–40) and, therefore,
diminished potential to retrieve snow properties at micro-
wave frequencies.
Incident energy from a ground-based radar system
directed towards a dry snowpack interacts first at the air/
snow interface, where it is partially reflected in different
directions and partially transmitted into the snowpack. The
magnitude of each component depends on the dielectric
mismatch between the surface media and air, the roughness
at the interface, and the geometry of the radar. In the case of
dry snow, the vast majority of incident energy propagates
into the volume at incident angles greater than nadir ( > 0)
because of the small difference in permittivity between the
air ("01) and the snow ("01.05–2). Assuming a smooth
surface, incident energy reflected at the air/snow interface
can be estimated using the square of the Fresnel reflection
coefficients, where if "0
snow ¼2 and ¼40, only 5.9% of
the horizontally polarized and <1.0% of the vertically
polarized incident energy is reflected. As a result, as is
generally considered to be a minor contributor to dry snow
backscatter and is often neglected.
Within the snow volume, variability in vol results from
changes in physical properties including depth, density and
microstructure (Fung, 1994; Tsang and others, 2007; Du and
others, 2010). As a mixed medium, dry snow consists of ice
crystals in an air background, with ice structures contributing
volume scatter at microwave wavelengths. As depth in-
creases, the path length of the propagating wave is extended,
increasing the potential for scattering within the volume. A
theoretical 17.2 GHz response to increasing depth is
demonstrated in Figure 2, where when the quasi-crystalline
approximation/dense-media radiative transfer model (QCA/
DMRT) of Tsang and others (2007) is parameterized with
conditions common to tundra environments, co- and cross-
polarized vol is found to increase by >10 dB with depths up
to 1 m. Open tundra observations in Churchill span a
relatively limited range of the demonstrated sensitivity,
which coincides with strong increases in vol as depth
departs from 5 cm. Snow density () contributes to vol
through variation in the number of scatterers, and
Fig. 2. Left: set-up of UW-Scat, illustrating first-order backscatter for snow-covered terrain (adapted from Rott and others, 2010). Right:
composition of theoretical single-layer Ku-band scattering contributions for a snow-covered tundra environment (modelled with the QCA/
DMRT implementation of Tsang and others, 2007). Model input parameters: f¼17:2 GHz, "r, gnd ¼5þ0:5i,= 250 kg m3,d= 1 mm,
¼0:1 and = 40°. Observed range of depth shown in grey.
King and others: Tundra snow 17.2 GHz backscatter 269
consequently the permittivity and the extinction coefficient
of the snow volume. A functional relationship exists where
"0¼1þ0:0019, independent of frequency in the micro-
wave spectrum (Hallikainen and others, 1986). According to
this relationship, seasonal snow occupies a narrow range of
"0between approximately 1.19 and 1.70 (100 and 400 kg
m3). Despite the limited range, variation in snow permittiv-
ity resulting from seasonal and spatial changes in density can
contribute internal reflections and therefore influence vol.
Higher-order interactions contributing to vol are often
difficult to quantify, as complexity of the signal is exacer-
bated by seasonal accumulation and metamorphosis which
drive intricate layering, thereby evolving vertical and hori-
zontal heterogeneity of scattering within the snowpack.
The influence of snow metamorphosis on radar response
is dictated by a complex set of interactions controlled by
grain size, grain shape and aggregate structure. In the case of
seasonal snow, processes of vapour transport and subse-
quent grain growth can exert significant influence on
volume scatter as the ratio between grain diameter and
wavelength (d=) is increased. This relationship is classic-
ally defined by Rayleigh scattering theory where scattering is
proportional to the fourth power of frequency and the third
power of size. In practical application, dependence on grain
diameter alone does not account for real-world variations in
shape and aggregate structure that may contribute signifi-
cant portions to observed scatter in addition to multiple
scattering between densely packed grains. For example, Du
and others (2010) suggest that as the ratio between the long
and short axis of a theoretical ellipsoid grain is reduced,
cross-polarized response may increase in low to moderate
snow volume-scattering environments (i.e. increased depo-
larization with prolate-shaped grains in shallow snow
environments). Similar observations have been made in
field studies such as Yueh and others (2009) where a strong
seasonal depolarization response was attributed to the
development of non-spherical depth-hoar grains. In addition
to grain shape, recent studies have suggested that bonding
and evolving microstructure can play an important role in
observed scatter where aggregate grain structures contribute
stronger effective scattering than their individual constitu-
ents (Tsang and others, 2007; Chang and others, 2014).
Given the potential for dynamic grain growth (0.5 mm
>d<6 mm) and development of aggregate depth-hoar
structures, it can be expected that metamorphosis of the
observed open tundra snowpack will be a notable con-
tributor to the volume-scattering response in addition to
bulk properties such as depth or SWE.
If incident microwave energy successfully traverses the
snow volume, it is subjected to reflection at the ground
surface. Unlike as, spatio-temporal variability in gnd is
common in terrestrial dry snow environments, where local
surface properties can exert influence on total backscatter
due to the larger dielectric contrast of the surface media
("0>3) and typically large standard deviation of surface
height relative to the wavelength (h> =32 cos ; Ulaby,
1981). Soil moisture and state play an important role in
determining reflection where a larger portion of incident
energy may be returned into the snow volume when soil is
moist and "0is elevated. Seasonally, this is an important
consideration when accumulation occurs before the soil has
frozen and free moisture is present. Spatially, variation in
surface scattering generally results from differences in
surface roughness, soil and vegetation type, and hetero-
genous soil moisture patterns. In general, gnd can be
expected to contribute less to total backscatter than vol at
Ku-band in snow-covered tundra environments (Fig. 2).
Energy reflected back into the snowpack may again scatter
within the volume and/or scatter within volume followed by
reflection from the ground surface, generating higher-order
scatter contributions (gv).
Backscatter measurement
To characterize backscatter response, a frequency-modu-
lated continuous wave radar system known as the University
of Waterloo scatterometer (UW-Scat) was deployed (King
and others, 2013). Pulled by snow machine, the Ku-band
unit of UW-Scat was transported in a sled-based configur-
ation to terrestrial locations and configured to collect
backscatter measurements over a 15 m 20 m area in
<20 min. In the microwave remote sensing of snow, choice
of frequency is often a compromise between penetration
depth and sensitivity to desired snow properties. At Ku-band
(12–18 GHz), the corresponding wavelengths are able to
penetrate dry snow to depths >1 m (Mätzler, 1987; Marshall
and others, 2004). For the purpose of observing tundra-class
snow, Ku-band was therefore sufficient to penetrate the
expected range of depths (0.10–0.75 m; Sturm and others,
1995). Moreover, the proximity of the observing instrument
wavelength to the scale of individual snow grains was close
enough to solicit a large volume response without saturating
within the upper bounds of the snowpack. Backscatter
measurements made with UW-Scat were collected at a
centre frequency of 17.2 GHz across a narrow bandwidth of
0.5 GHz. From the sled-mounted position, the antenna and
radio-frequency (RF) hardware operated at a height of
2.0 m with a narrow beamwidth of 4.3°. The resulting
ground-projected radar footprint was <30 cm in azimuth
and range directions. Additional operational parameters of
use during the experiment are listed in Table 1.
Once an observation site was selected, the scatterometer
was levelled to the local terrain and RF hardware was
allowed to stabilize at 35°C to minimize temperature effects
on the hardware. A two-axis positioning system was then
used to direct the radar antenna hardware through a series of
azimuth sweeps to collect measurements at elevation angles
between 30° and 45° in 3° increments. Measurements at
each elevation angle were integrated over a 60° azimuth
sweep to improve the number of independent samples and
ensure an adequate signal-to-noise ratio. In post-processing,
the backscattering coefficient, 0, was estimated to express
the ratio of power reflected to power transmitted as
Table 1. UW-Scat operational parameters
System parameter Operational value(s)
Height (m) 2.0
Transmit power (mW) 10
Centre frequency (GHz) 17.2
Bandwidth (GHz) 0.5
Range resolution (m) 0.3
Beamwidth (°) 4.3
Incident angle (°) 30–45
Footprint at 30° (m) 0.15 0.17
Footprint at 45° (m) 0.20 0.28
Polarization VV, VH, HV, HH
King and others: Tundra snow 17.2 GHz backscatter270
normalized to the ground-projected radar footprint. Co-
polarized vertical (0
vv), horizontal (0
hh) and cross-polarized
(0
vh) backscatter coefficients were derived based on the
averaged range profiles and system geometry. Detailed
discussion of signal-processing procedures can be found in
Geldsetzer and others (2007) and King and others (2013).
To calibrate the scatterometer system, an accompaniment
of internal and external procedures was used. Prior to each
site scan, a trihedral corner reflector was erected in the field
of view and observed to collect a short time series of meas-
urements. Differences between the observed polarimetric
measurements and reference target backscatter were cor-
rected offline using a procedure described by Geldsetzer and
others (2007). To characterize system and platform noise,
open sky measurements were collected after each corner
reflector measurement. Sky measurements were coherently
subtracted from target range profiles to reduce the influence
of system noise and isolate the desired target signal. Estimates
of worst-case measurement error from the post-processing
procedure were, on average, 2.0 dB including a 0.5 dB
addition for random error. In general, the observed errors
were a function of the limited number of independent
samples collected under the beam-limited conditions of the
sled-based configuration as well as bias introduced by
human and environmental inaccuracies with misestimation
of the deployed sensor height and wind buffing of the
reference target. A detailed discussion of the error estimation
process can be found in King and others (2013).
Snow measurements
Destructive sampling within the scatterometer field of view
was completed immediately after each scan to gather
information about local snow properties and to evaluate
snow variability at the sub-scan level. A standard procedure
was used where horizontal heterogeneity in depth, density
and SWE was quantified in addition to point measurements
of vertical snow properties. On completion of the scatter-
ometer measurement, a 15 m 20 m box was measured
within the field of view of the scatterometer to bound a
snow-sampling area (Fig. 3). Within the sampling area, a
grid of snow depth measurements was collected using a
GPS-enabled Snow-Hydro MagnaProbe. Each grid con-
tained a minimum of 200 measurements spaced across
several transects parallel to the scatterometer range dir-
ection. Pairs of snow-core measurements were also col-
lected at six locations within the sampling area to estimate
bulk density and SWE. The snow-coring instrument had a
cross section of 30 cm2and is often referred to in the
literature as the ESC-30 (Pomeroy and others, 1997; Derksen
and others, 2012; Rees and others, 2014). Using a handheld
spring scale, the weight of each snow core was recorded,
along with the depth at the measurement site.
In addition to bulk snow measurements, two snow pits
were excavated at 3 and 5 m outwards from the scatter-
ometer using standardized procedures (Fierz and others,
2009). Multiple snow pits were completed to avoid
mischaracterization related to pit placement and large
variations in topography (e.g. hollows between hummocks).
Once excavated, structural and textural discontinuities were
used to define and record layer locations, thickness and
general composition. A snow sample from each layer was
extracted and placed on a grid comparator card for visual
analysis. Using a stereo-microscope, grain origin types were
identified and estimates of axial diameter were recorded.
Density within each snow pit was measured along two
continuous vertical profiles extracted from the pit face using
Fig. 3. Destructive snow-sampling protocol plan view. At each site a distributed set of bulk and stratigraphic measurements was made to
quantify variability within the scatterometer field of view.
King and others: Tundra snow 17.2 GHz backscatter 271
a 100 cm3box-style cutter and digital scale. Temperature
profiles were constructed using a string of thermistors
inserted into the snowpack with a vertical spacing interval
of 4 cm.
Trench measurements
To evaluate the influence of lateral heterogeneity on
backscatter, detailed measurements of snowpack stratig-
raphy were excavated within the scatterometer field of view.
Following a protocol detailed by Tape and others (2010), a
5 m trench was dug 3 m outwards from the scatterometer
across the azimuth look direction to correspond with
scatterometer measurements collected at elevations be-
tween 30° and 45°. The excavated face was prepared
manually with hand tools to create a flat surface per-
pendicular to the ground surface. Once prepared, 850 nm
centre frequency NIR photography was completed along the
length of the trench using a linear rail system to maintain a
constant distance to the trench face. Sequential photographs
taken along the length of the trench were georeferenced
using an in-scene cm-scale measuring staff positioned
horizontally above the trench face. A single continuous
image was stitched together from the collected set of
photographs, and stratigraphy was manually extracted at a
1 cm horizontal resolution (Fig. 4). In addition to NIR
photography, inter-layer snow properties were character-
ized with a set of five snow pits completed at 1 m intervals
along the trench. Vertical transects of density, temperature
and grain size were measured using the previously de-
scribed standard snow-pit protocol.
RESULTS
Seasonal and spatial snow properties
Accumulation within the Churchill study area evolved with
spatio-temporal complexity related to variations in vegeta-
tion, topography and climate. Given the selected land-cover
composition of flat open terrain and dominant graminoid
vegetation, observation sites were exposed to a common set
of sub-arctic environmental agents including strong wind
and sustained low air temperature. The majority of accumu-
lation within the study area was deposited between Decem-
ber and January, with a limited number of precipitation
events occurring thereafter. In the observed open areas,
snowpack development was heavily influenced by sustained
winds, where average speeds commonly exceeded 5 m s1,
rapidly redistributing accumulated snow. Mean snow depth
at the observed tundra sites was shallow, ranging from
approximately 4 to 25 cm, generally increasing through the
season (Table 2). Examples at the extreme of this range were
associated with changing land-cover characteristics found in
close proximity to the observation site (e.g. forest transition
zones situated near sites 14 and 15). Inter-site standard
deviation of snow depth ranged between 20% and 64% of
the mean, in most cases declining with increasing depth
(R= –0.5, n= 26; correlations and coefficients of determin-
ation were significant at the 0.05 level). At the open tundra
sites, microtopographic elements, including hollows be-
tween hummocks, trapped early-season snow and created
deviations by up to double the inter-site mean. The influence
of these features on total depth was reduced as accumulation
increased relative to variations in surface height, effectively
masking the smaller ground features. Despite the sustained
influence of wind, bulk snow density remained low (<250 kg
m3) and did not vary substantially through the season.
Compared to depth, inter-site variability of density was
smaller at a majority of sites, with standard deviations
accounting for 9–46% of the mean. As a product of depth
and density, the observed range of SWE was seasonally and
spatially limited to 70.0 mm or less. At the tundra observation
sites, variation in SWE showed a stronger linear relationship
Fig. 4. A series of 850 nm NIR photographs taken along the length of an excavated 5 m snow trench. Photographs were referenced and
stitched together for analysis. Black lines show manually identified snow stratigraphy.
Table 2. Inter-site snow properties measured at each tundra
observation site. Dates of observation are presented as day of year
(DOY) spanning the 2010/11 observation period. Bulk SWE and
density measurements were not available on DOY 60
Depth Density SWE
Site DOY Mean SD Mean SD Mean
cm cm kg m3kg m3mm
1 310 5.7 2.9 104 10 5.9
2 319 5.0 2.9 139 40 6.9
3 320 5.3 1.8 129 51 6.9
4 320 3.9 1.0 154 54 6.0
5 322 3.5 1.3 139 40 4.8
6 324 4.0 1.5 155 71 6.3
7 328 6.2 2.7 258 70 15.9
8 328 6.9 2.8 262 41 18.1
9 329 3.9 1.4 112 26 4.3
10 329 4.1 1.4 130 31 5.3
11 334 8.1 1.9 206 41 16.7
12 337 11.6 3.7 288 91 33.3
13 337 10.9 4.6 211 32 22.9
14 346 23.2 6.7 302 40 70.0
15 346 24.1 6.6 207 29 49.9
16 352 14.0 3.6 186 22 26.0
17 4 16.2 4.3 253 30 41.0
18 16 20.5 13.2 163 41 33.3
19 17 14.0 3.6 219 33 30.8
20 24 16.9 4.6 264 43 44.6
21 33 19.0 6.0 248 42 47.0
22 44 16.8 6.0 280 34 46.9
23 49 19.2 7.2 233 36 44.6
24 50 21.9 6.0 216 34 47.2
25 56 15.2 4.3 209 29 31.8
26 60 25.0 5.0 – – –
King and others: Tundra snow 17.2 GHz backscatter272
with depth (R2¼0:98) than density (R2¼0:46) as a result of
the smaller range of bulk density encountered.
Snow stratigraphy excavated within the scatterometer
field of view increased in complexity from an early point in
the experiment. Sequential periods of accumulation and
wind transport created two to seven distinct horizontal
layers of varied thickness and texture. A basal depth-hoar
layer was the most consistent feature excavated among the
open tundra sites. In the 50 pits completed, depth hoar
comprised, on average, 52% of total depth. The rapid
development of the basal hoar layer was driven by a sharp
early-season decline in air temperature sustained through
the end of the experiment with limited increases in depth.
Vertical temperature gradients measured through the shal-
low snowpack exceeded 20°C m1in 40 of 50 open tundra
pits, a strong indicator of sustained kinetic growth. Well-
defined cups were a common feature of the basal depth-
hoar layer, with large aggregations present from DOY 334
onward. By the end of the observation period, grains within
the depth-hoar layer had exceeded 3 mm in major axis
diameter, with poly-aggregates reaching 6 mm or larger (Fig.
5). Thickness of the basal depth-hoar layer was heavily
influenced by inter-site changes in ground surface height. As
a result, the thickness of the basal hoar layer between inter-
site snow pits was found to vary by up to 10 cm where
periodic hummock features were present.
Contrasting snow surface features were heavily influ-
enced by wind transport, featuring multiple slab layers
separated by hard thin crusts. Grain diameter within the
surface slab layers was generally small (<1 mm) due to
persistent wind action and subsequent rounding of grains
with transport. A vertical gradient of grain diameter became
apparent in the latter part of the experiment, where wind-
rounded grains transitioned into solid facets with sustained
temperature gradient metamorphosis. In several of the late-
season snow pits, large faceted grains were found inter-
mixed all the way to the surface of the snowpack. Few ice
features were identified, with those found being very thin
snow-ice layers situated at the base of the pack from early-
season melt processes. Rapid redistribution of snow accu-
mulation limited the number of fresh-snow observations
collected during the experiment; in most cases, surface
layers identified with fresh snow were <2 cm thick and
intermixed with wind-rounded grains.
Overall, the rapid and often subtle changes within the
shallow tundra snowpack provided a challenging target of
analysis and a unique opportunity to evaluate distinctive
features previously unaddressed at Ku-band. The predom-
inant local processes of wind redistribution and temperature
gradient metamorphism were characteristic of conditions
previously described along the Hudson Bay coast and of
general conditions present with tundra-type snow (Sturm
and others, 1995; Kershaw and McCulloch, 2007; King and
others, 2013). Electromagnetically, the early and rapid
development of large depth hoar had the potential to drive
strong increases in backscatter as seasonal and spatial
processes of accumulation and metamorphosis progressed.
Comparison of backscatter measurements and snow
properties
To evaluate spatio-temporal backscatter sensitivity, snow
property measurements collected at each site were
compared against coincident estimates of 0. In an
effort to reduce the influence of sub-scan variations in
microtopography, estimates of 0were averaged to integrate
a number of azimuth sweeps across a small range of
incident angles between 30° and 45°. In the averaged
scatterometer measurements, the dynamic range of 0was
found to be >8 dB. Co-polarized 0
vv and 0
hh ranged in
magnitude from approximately –13 to –5 dB while cross-
polarized 0
vh and 0
hv responses were measured across a
lower and smaller range between –26 and –15 dB. For the
entirety of the experiment period, cross-polarized 0
vh and
0
hv were found within 0.1 dB and therefore were considered
reciprocal for the purpose of analysis.
The observed relationships between depth, SWE and 0,
as shown in Figure 6, indicate Ku-band sensitivity to
evolving snow properties at the 26 open tundra sites. Across
the small range of encountered depth, a clear linear
relationship with 0
vv emerged, where vertically polarized
backscatter increased by 0.22 dB for every 1 cm increase
in depth (R2¼0:67). A slightly less sensitive 0
hh response,
not shown in Figure 6, increased by 0.17 dB for every 1 cm
in depth (R2¼0:44). The observed sensitivity agrees with
the theoretical relationship demonstrated in Figure 2, where
increasing path length generates a strong volume response
in shallow snow with large grains (>1 mm). Preferential
vertical scattering was apparent throughout the experiment,
where 18 of 26 observations demonstrated a co-polarization
ratio (0
hh/0
vv) less than 0. The Ku-band co-polarization
ratios were moderately correlated with the number of layers
observed within the scatterometer field of view (R¼ 0:64)
suggesting that rough surface interactions within the layered
accumulation played a role in the observed horizontal
response. Alternatively, it is possible that the observed
vertical arrangement of depth-hoar aggregates along the
path of vapour transport contributed to the stronger vertical
response. The cross-polarized 0
vh response was comparable
in slope to the co-polarized, increasing by 0.17 dB for
every 1 cm increase in depth (R2¼0:39). The coevolution
Fig. 5. Basal depth hoar at site 25 on a 2 mm grid comparator card.
The large aggregations found at site 25 were common to most
observation sites from DOY 334 forward.
King and others: Tundra snow 17.2 GHz backscatter 273
of the co- and cross-polarized response was reflected in a
small standard deviation of the depolarization ratio (0
vh/
0
vv), where 0
vh was generally 9–14 dB lower than 0
vv. The
strong cross-polarized backscatter is indicative of multiple
scattering between density-packed grains and/or the pres-
ence of non-spherical snow grains (Tsang and others, 2007;
Yueh and others, 2009; Du and others, 2010).
Similar to snow depth, spatio-temporal variation in SWE
was limited by persistent wind action and negligible
precipitation input. Despite this, the observed increase in
0was large relative to the small range of observed SWE.
Sensitivity to SWE was strongest with 0
vv, increasing by
0.82 dB for every 1 cm in SWE (R2¼0:62). Similar to the
observed relationship with depth, 0
hh showed a lower but
significant sensitivity to SWE at 0.62 dB for every 1 cm
increase (R2¼0:42). Finally, the cross-polarized 0
vh sensi-
tivity was observed at 0.80 dB for every 1 cm increase in
SWE (R2¼0:36). In the relationship between SWE and 0,
backscatter diversity often separated sites of similar aggre-
gate composition, in some instances by several decibels.
This finding highlights the complexity of the observed
interaction space, where density, grain size and vegetation/
soil properties are important contributors to backscatter in
addition to variation in depth. Seasonality and metamorphic
state of the snowpack appeared to play a role in the
observed backscatter diversity where increases in the major
axis diameter of grains measured with a grid comparator
card were moderately correlated with increased vertical
backscatter (R¼0:39). The identified relationship with grain
diameter was not found to be significant with horizontal or
cross-polarized backscatter. While this finding is of interest,
the seasonality of the measurements and, as a result,
covariance of depth and grain growth made it difficult to
decompose causation between the evolving snow proper-
ties. Moreover, the previously demonstrated spatial vari-
ability in stratigraphy made it extremely difficult to interpret
the inter-site snow pits in the context of the larger scatter-
ometer field of view. Overall, the distributed measurement
protocol was able to identify Ku-band sensitivity to bulk
snow properties when both in situ measurements and radar
observations were averaged up to the site scale, but did not
provide the means to quantify the role of stratigraphy, grain
size and surface properties in observed backscatter diversity.
Evaluation of trench backscatter response
Direct comparison of in situ sampled snow properties and
0revealed Ku-band sensitivity to increasing snow volume.
Despite the positive result, snow pits completed within the
scatterometer field of view yielded insufficient information
to evaluate the influence of sub-scan lateral heterogeneity
on observed backscatter. To address this open question, two
5 m trench experiments were completed on 7 and 8 January
to characterize variation in stratigraphy within the scatter-
ometer field of view. Collocated radar returns measured
across the azimuth range of the scatterometer were
compared with excavated snow stratigraphy to evaluate
potential sub-scan influences on backscatter. The first trench
was excavated in an open area adjacent to a transition zone,
hereafter referred to as the forest edge site. The forest edge
site possessed snow features common to the previous open
tundra sites including limited depth, predominant wind-
slab/depth-hoar composition and a rough underlying ground
surface. The second trench was completed within a sparsely
populated tree stand where the sheltered snowpack was
composed of a larger number of lower-density layers.
Comparison of the two targets provided an opportunity to
evaluate a range of snow conditions and their influence on
backscatter at scales local to the observing instrument.
A complex arrangement of internal layers with several
discontinuous features was observed at the forest edge site
(Fig. 7a). Within the shallow snowpack, three to five distinct
layers were found at any given horizontal position, and, in
total, six unique layers were identified. Total snow depth
along the trench varied between 16.4 and 37.2 cm with a
standard deviation of 5.6 cm (Fig. 7b). The observed
variation in depth was in agreement with inter-site measure-
ments completed at the previous open tundra sites, where
large changes in total depth were encountered over short
distances. The most prominent change in snowpack com-
position was associated with a protruding hummock feature
centred at the midpoint along the trench. Prior to excav-
ation, the smooth surface of the mid-season snowpack
Fig. 6. Comparison of 0averaged for incident angles between 30°
and 45° against snow depth (a) and SWE (b) at each tundra
observation site. Solid circles show 0
vv response, and hollow
circles show 0
vh response. Measured 0
hh and 0
hv responses not
shown, to improve readability.
King and others: Tundra snow 17.2 GHz backscatter274
provided no indication of the heavy influence of this feature
on both total depth and stratigraphy.
Internal stratigraphy of the forest edge snowpack con-
sisted of several wind-influenced features underlain with a
continuous layer of depth hoar (Table 3). A number of thin
wind crusts (<0.5 cm), harder in texture than the surround-
ing snow, separated the larger internal layers of the
snowpack. The upper and most recent feature, noted as
layer 6 in Figure 7, was a smooth continuous surface layer
composed of small wind-rounded grains intermixed with
decomposing needle-like precipitation (0.1–0.4 mm). Be-
neath layer 6, a set of thin soft slabs, noted as layers 4 and 5,
surrounded the apex of the prominent hummock feature
where wind scouring had created a notable horizontal
discontinuity. Each slab was composed of slightly larger
wind-rounded grains (0.4–1.0 mm) and was similar in
density to the surface slab layer (220 kg m3). The largest
continuous feature identified within the trench was a mixed-
type layer composed primarily of faceted grains identified as
layer 3. Within this large continuous layer, a wide range of
grain size was found (0.2–1.5 mm). On the right-hand side
of the trench, a hard relict wind slab was identified as layer 2
(3–5 m) in which densities were the highest within the
snowpack, increasing to 280 kg m3. The basal surface of
layer 5 was composed of high-density indurated hoar with
well-developed cup structures >1 mm in diameter. Finally,
the basal depth-hoar layer, noted as layer 1, contained
striated cup-shaped grains and large poly-aggregates ranging
in diameter from 1 to 4 mm. As a fraction of total depth, the
depth-hoar layer composed on average 33% or 9.0 cm of
the snowpack (Fig. 8b). The thickness of the basal layer was
controlled by the ground surface height and location of the
buried hard slab.
The excavated trench revealed an interesting target for
radar observation where large changes in snowpack proper-
ties were found over short distances. Radar measurements
collected as part of the forest edge scan were processed as
individual returns relative to response of the collocated
corner reflector and given approximate horizontal positions
in relation to the trench face (Fig. 7c). As a general
limitation of the observation protocol, uncertainty in the
estimated horizontal radar position was greatest near the
edges of the scan due to the arc of ground projected
scatterometer sweep. The measured co-polarization re-
sponse across the trench showed variability in relation to
changing snow and soil conditions. In particular, variability
Fig. 7. Stratigraphy (a), depth (b) and relative backscatter (c) observed along the length of the forest edge trench. Layer notation in (a)
corresponds to description in Table 3.
Table 3. Description of stratigraphy excavated at the forest edge
site. Layer numbers correspond to Figure 7
Layer ICSSG* code Description
6 RGwp (DFbk) Recent mixed layer with wind-rounded grains
and decomposing needles
5 RGwp (FCsf) Mixed soft slab layer composed of mostly
wind-rounded grains
4 RGwp (FCsf) Mixed soft slab layer composed of mostly
wind-rounded grains
3 RGwp (FCsf) Mixed soft slab layer with wind-rounded and
faceted grains
2 FCso (RGxf) Buried hard slab. Mixed composition of
indurated hoar
1 DHch Depth hoar. Large faceted cups and
poly-aggregate structures
*International Classification for Seasonal Snow on the Ground.
King and others: Tundra snow 17.2 GHz backscatter 275
was noted in relation to the previously identified hummock
feature where co-polarized vertical backscatter decreased
by 16% relative to the deeper accumulation situated at
horizontal positions to the left of the feature. The horizontal
backscatter response to the large change in snow depth was
muted in comparison, emulating previously observed
behaviour of the low co-polarization ratio observed in the
distributed set of tundra snow targets. Across the width of
the trench, co-polarized horizontal response was found to
be 12% lower then the vertical response. To the right of
the hummock, backscatter increased, reaching a trench
maximum near the horizontal position of 3 m. The rapid
increase in co-polarized backscatter coincided with
increasing snow depth and complexity in layering. In
particular, layer 5 was a unique dielectric and geometric
discontinuity where the indurated depth hoar was both of
higher density and larger grain size compared to the
surrounding layers. Coupled with the increased depth to
the right of the hummock, there was potential for multiple
scattering, which appeared to be supported by the increased
depolarization. Overall, the forest edge trench demon-
strated the presence of dynamic changes in snowpack
structure over short distances, which appear to have had a
large influence on backscattered energy.
The second trench was completed within a sparsely
populated tree stand, sheltered from strong local winds.
Total depth along the trench ranged from 30.0 to 51.6 cm,
with a smaller standard deviation of 3.1 cm. The contrasting
influence of the local environmental agents was evident in
the excavated stratigraphy (Fig. 8). Here layers were
comparatively homogeneous, greater in number and, in
most cases, found continuously through the 5 m trench. In
total, nine unique layers were identified, with five present
across the entire trench (Table 4). As in the previous
excavation, a number of thin crusts (<1 mm) were found
separating the major internal layers. Fresh precipitation,
identified as layer 9, was the most recent layer and was
composed of a combination of large stellar plates (1 mm)
and much smaller graupel (0.1 mm). Below the fresh snow
surface, a decomposing precipitation layer was found with
smaller round grains (layer 8). Lacking influence from local
wind, the two most recent layers were very low in density,
ranging from approximately 100 to 150 kg m3. The
remaining layers of the snowpack were heavily influenced
by temperature gradient metamorphosis, with grain size
increasing towards the base of the snowpack. Layer 7, the
more recent of two mixed-type layers, was composed of
Fig. 8. Stratigraphy (a), depth (b) and relative backscatter (c) observed along the length of the forest trench. Layer notation in (a) corresponds
to description in Table 4.
Table 4. Description of stratigraphy observed along the length of
the forest trench. Layer numbers correspond to Figure 8a
Layer Code Description
9 PPsd (PPgp) Fresh snow. Stellar plates and graupel present
8 DFdc (RGlr) Recent snow. Decomposing precipitation and
clusters of rounded grains
7 FCso Mixed layer. Primarily composed of small
rounded grains with some faceted grains present
6 FCso Mixed layer. Primarily composed of larger loose
faceted grains with some round grains
5 DHcp Unconsolidated depth hoar. Mix of cups and less
developed forms
4 DHcp Unconsolidated depth hoar. Large cup-shaped
grains
3 DHch Consolidated depth hoar. Hard texture
2 IFbi Icy layer. High-density hard-snow feature
1 IFbi Icy layer. High-density extremely hard-snow
feature
King and others: Tundra snow 17.2 GHz backscatter276
smaller rounded grains with some larger faceted grains
(0.3–1.0 mm). A similar mix of grain types was found
in layer 6, with a bias towards larger faceted grains
(0.6–1.8 mm). Density increased within the more mixed
layers, reaching a maximum of 270 kg m3. Again, basal
depth-hoar features dominated the remainder of the snow-
pack, with layers 4 and 5 composed of large cup-shaped
grains up to 4 mm in diameter. Density within the
unconsolidated depth-hoar layers was lower, at 170 kg
m3. A number of discontinuous icy hoar features, identified
as layers 1, 2 and 3, were found at the base of the pack.
These thin features were very hard in texture and were
composed of large grains up to 4 mm in diameter.
Compared to the forest edge trench, the backscatter
response from the forest trench was stable across the scan.
The reduced variability in the co- and cross-polarized
backscatter corresponded well to reduced variation in snow
depth and increased homogeneity of the internal stratigraphy
relative to the previous trench. In clear contrast to the forest
edge site, there was no observed preference in orientation of
the co-polarized backscatter. This observation separates the
forest response from the open tundra sites where vertical
preferential scattering was consistently observed. Across the
full length of the trench, a decline was identified in both the
co- and cross-polarization response. This observed variation
in backscatter was small compared to the previous trench,
but did indicate a changing interaction with the physical and/
or dielectric properties of the sub-scan target. The most
evident change in snowpack composition along the trench
that could have produced an attributable change in back-
scatter was the increasing thickness of layer 7 and corres-
ponding reduction of layer 6. These two mixed-layer types
were composed of different grain sizes, those within the
more recent layer 7 being significantly smaller in diameter.
From left to right, layer 7 gradually became the dominant
snowpack layer, effectively reducing the mean grain size of
the snowpack. Unfortunately, the coarse range resolution of
UW-Scat did not allow for characterization of individual
layer contributions, and so did not offer definitive evidence
of why the change had occurred.
In general, the contrasting backscatter responses of the
two trench sites provided practical examples of the
influence of sub-scan snow properties on Ku-band back-
scatter. The findings suggest that variability in tundra snow
and soil properties over short distances are significant
modifiers of Ku-band backscatter and, in the future, should
be considered in the development of tundra snow property
retrievals. When in situ and radar measurements were
averaged up to the trench (5 m) scale, there was clear
sensitivity of backscatter to SWE. Comparisons at smaller
scales likely require high-resolution radar systems to separ-
ate contributions to backscatter as a function of depth and
horizontal location.
DISCUSSION AND CONCLUSIONS
During CASIX, a distributed set of 26 backscatter measure-
ments were collected across a range of tundra snow
conditions influenced by spatio-temporal processes of
accumulation and metamorphosis. A standardized sampling
protocol was introduced to measure snow properties within
the scatterometer field of view immediately after each scan
to evaluate local influences on backscatter. Predominant
agents of wind transport and temperature gradient
metamorphosis governed the spatio-temporal development
of the observed snowpack at the selected open tundra sites.
Sustained exposure to strong winds quickly redistributed
accumulation, and combined with the dry polar environ-
ment, limited total depth to typically <30 cm. Inter-site
variations in snow depth were often large as a result of
changes in underlying ground surface and vegetation height
that varied over short distances. Given the anticipated
sensitivity of Ku-band to changing snow conditions, the
inter-site variations in depth had the potential to drive
variability in sub-scan backscatter measurements. In add-
ition to wind action, steep vertical temperature gradients
were common to the open tundra sites, which enabled
periods of substantial grain growth. A thick basal depth-hoar
layer was found throughout the experiment, with large cup-
shaped grains with diameters in excess of 4 mm. By the end
of the observation period, faceted grains were found
throughout the full volume of the snowpack, with large
aggregations present in the basal layers.
Despite limited accumulation, a large dynamic range of
co- and cross-polarized 0was observed at the open tundra
sites. Analysis of the spatio-temporal response indicated Ku-
band backscatter sensitivity to increasing snow depth and
SWE. This finding is in agreement with previous airborne
measurements made by Yueh and others (2009), where
strong co- and cross-polarized Ku-band (13.4 GHz) re-
sponses were observed at a number of CLPX-II terrestrial
sites in Colorado (0.5 dB for every 1 cm SWE). In general, the
Churchill response to SWE appears to have been stronger
but shows similarity to the shallow snow observations at the
CLPX-II North Park site (0.15–0.5 dB for every 1 cm SWE).
Here steep temperature gradients through the shallow
snowpack (<37 cm) at both sites provided the conditions
necessary for the development of depth hoar, and, as a
result, strong snow volume responses over a relatively
limited range of SWE. The development of large and/or
clustered depth-hoar grains such as those shown in Figure 5
had the potential to enhance the radar response with strong
forward scattering and increased multiple scattering within
the snow volume (Marshall and others, 2004; Du and
others, 2010). Recent advancement in DMRT theory
supports this finding, where clustering of grains, described
as stickiness (), has been shown to generate strong
scattering effects comparable to grains of a much larger
effective diameter (e.g. Tsang and others, 2007; Chang and
others, 2014). Prominent aggregations clustered along paths
of vertical vapour transport within the Churchill snowpack
are a likely contributor to the enhanced backscatter
response and will be a subject of future study.
While the distributed open tundra scatterometer obser-
vations showed sensitivity to evolving snow properties, in
agreement with previous field and model observations, inter-
site measurements were insufficient to characterize the
influence of sub-scan variations in snow properties. With
demonstrated variability in depth and stratigraphy over short
distances, a refined methodology for detailed characteriza-
tion of snow stratigraphy within the field of view was
implemented. Two detailed trench experiments, completed
in contrasting environments, exemplified the highly variable
nature of tundra snow despite its shallow accumulation. The
influence of variability in snow properties was apparent in
the backscatter signal at the forest edge site where deviations
of depth and stratigraphy created strong deviations in signal.
Conversely, the relative homogeneity of the forest trench site
King and others: Tundra snow 17.2 GHz backscatter 277
responded with reduced variability in backscatter. In the
future, snow property information extracted from trench
experiments will be of use to initialize electromagnetic
models and for diagnostic analysis where sensitivity of high-
resolution snow properties can be evaluated over the entire
trench. Such studies will be particularly useful for clarifying
the role of grain size where subtle changes in layer structure
have appeared to influence backscatter.
This study has presented a novel set of ground-based
experiments to evaluate the influence of spatio-temporal
changes in snow properties on Ku-band backscatter. In
doing so, a previously untested sub-arctic tundra environ-
ment has been characterized, providing a starting point for
the development of robust retrieval methods. The findings of
this study support the use of Ku-band backscatter measure-
ments for observation of snow properties in tundra environ-
ments. Future study will be needed to assess the
implications of the demonstrated backscatter variability
with snow accumulation and its influence on scaling to
airborne and satellite-based observation approaches.
ACKNOWLEDGEMENTS
CASIX field activities were supported by the European Space
Agency, Environment Canada and the Natural Sciences and
Engineering Research Council of Canada (NSERC). We
thank Arvids Silis, Cristina Surdu, Dave Halpin, Homa
Kheyrollahpour, Jason Oldham, Mel Sandells, Nic Svacina,
Niina Luus, Peter Toose, Ryan Ahola and Steve Howell for
their contributions to the field campaign. Logistical support
for CASIX was provided by the Churchill Northern Studies
Centre. The University of Waterloo scatterometer system
was developed with support from the Canadian Foundation
for Innovation and the Ontario Ministry of Research and
Innovation. The work of Joshua King, Richard Kelly, Grant
Gunn and Claude Duguay was supported by the NSERC.
We thank two anonymous reviewers who greatly improved
the quality of the paper.
REFERENCES
Chang W, Tan S, Lemmetyinen J, Tsang L, Xu X and Yueh SH
(2014) Dense media radiative transfer applied to SnowScat and
SnowSAR. IEEE J. Select. Topics Appl. Earth Obs. Remote Sens.,
7(9) (doi: 10.1109/JSTARS.2014.2343519)
Colbeck SC (1991) The layered character of snow covers. Rev.
Geophys.,29(1), 81–96 (doi: 10.1029/90RG02351)
Deems JS, Fassnacht SR and Elder KJ (2008) Interannual consistency
in fractal snow depth patterns at two Colorado mountain sites.
J. Hydromet.,9(5), 977–988 (doi: 10.1175/2008JHM901.1)
Derksen C and 6 others (2009) Northwest territories and Nunavut
snow characteristics from a subarctic traverse: implications for
passive microwave remote sensing. J. Hydromet.,10(2), 448–
463 (doi: 10.1175/2008JHM1074.1)
Derksen C and 6 others (2012) Evaluation of passive microwave
brightness temperature simulations and snow water equivalent
retrievals through a winter season. Remote Sens. Environ.,117,
236–248 (doi: 10.1016/j.rse.2011.09.021)
Domine F, Gallet J-C, Bock J and Morin S (2012) Structure, specific
surface area and thermal conductivity of the snowpack around
Barrow, Alaska. J. Geophys. Res.,117(D14), D00R14 (doi:
10.1029/2011JD016647)
Du J, Shi J and Rott H (2010) Comparison between a multi-
scattering and multi-layer snow scattering model and its
parameterized snow backscattering model. Remote Sens.
Environ.,114(5), 1089–1098 (doi: 10.1016/j.rse.2009.12.020)
Fierz C and 8 others.(2009) The international classification for
seasonal snow on the ground. (IHP Technical Documents in
Hydrology 83) UNESCO–International Hydrological Pro-
gramme, Paris
Fung AK (1994) Microwave scattering and emission models and
their applications. Artech House, Norwood, MA
Geldsetzer T, Mead JB, Yackel JJ, Scharien RK and Howell SEL
(2007) Surface-based polarimetric C-band scatterometer
for field measurements of sea ice. IEEE Trans. Geosci.
Remote Sens.,45(11), 3405–3416 (doi: 10.1109/TGRS.2007.
907043)
Hallikainen MT, Ulaby FT and Abdelrazik M (1986) Dielectric
properties of snow in the 3 to 37 GHz range. IEEE Trans.
Antennas Propag.,34(11), 1329–1340 (doi: 10.1109/TAP.1986.
1143757)
Jonas T, Marty C and Magnusson J (2009) Estimating the snow
water equivalent from snow depth measurements in the Swiss
Alps. J. Hydrol.,378(1–2), 161–167 (doi: 10.1016/j.jhydrol.
2009.09.021)
Kershaw GP and McCulloch J (2007) Midwinter snowpack
variation across the Arctic treeline, Churchill, Manitoba,
Canada. Arct. Antarct. Alp. Res.,39(1), 9–15 (doi: 10.1657/
1523-0430(2007)39[9:MSVATA]2.0.CO;2)
King JML, Kelly R, Kasurak A, Duguay C, Gunn G and Mead JB
(2013) UW-Scat: a ground-based dual-frequency scatterometer
for observation of snow properties. IEEE Geosci. Remote Sens.
Lett.,10(3), 528–532 (doi: 10.1109/LGRS.2012.2212177)
Marshall H-P and Koh G (2008) FMCW radars for snow research.
Cold Reg. Sci. Technol.,52(2), 118–131 (doi: 10.1016/
j.coldregions.2007.04.008)
Marshall HP, Koh G and Forster RR (2004) Ground-based
frequency-modulated continuous wave radar measurements in
wet and dry snowpacks, Colorado, USA: an analysis and
summary of the 2002–03 NASA CLPX data. Hydrol. Process.,
18(18), 3609–3622 (doi: 10.1002/hyp.5804)
Marshall H-P, Schneebeli M and Koh G (2007) Snow stratigraphy
measurements with high-frequency FMCW radar: comparison
with snow micro-penetrometer. Cold Reg. Sci. Technol.,
47(1–2), 108–117 (doi: 10.1016/j.coldregions.2006.08.008)
Mätzler C (1987) Applications of the interaction of microwaves
with the natural snow cover. Remote Sens. Rev.,2(2), 259–387
(doi: 10.1080/02757258709532086)
Montpetit B, Royer A, Roy A, Langlois A and Derksen C (2013)
Snow microwave emission modeling of ice lenses within a
snowpack using the microwave emission model for layered
snowpacks. IEEE Trans. Geosci. Remote Sens.,51(9),
4705–4717 (doi: 10.1109/TGRS.2013.2250509)
Morrison K and Bennett J (2014) Tomographic profiling: a
technique for multi-incidence-angle retrieval of the vertical
SAR backscattering profiles of biogeophysical targets. IEEE
Trans. Geosci. Remote Sens.,52(2), 1350–1355 (doi: 10.1109/
TGRS.2013.2250508)
Pomeroy JW, Marsh P and Gray DM (1997) Application of a
distributed blowing snow model to the Arctic. Hydrol. Process.,
11(11), 1451–1464 (doi: 10.1002/(SICI)1099-1085(199709)
11:11<1451::AID-HYP449>3.0.CO;2-Q)
Rees A, English M, Derksen C, Toose P and Silis A (2014)
Observations of late winter Canadian tundra snow cover
properties. Hydrol. Process.,28(12), 3962–3977 (doi:
10.1002/hyp.9931)
Rott H and 13 others (2010) Cold Regions Hydrology High-
resolution Observatory for snow and cold land processes. Proc.
IEEE,98(5), 752–765 (doi: 10.1109/JPROC.2009.2038947)
Scipión DE, Mott R, Lehning M, Schneebeli M and Berne A (2013)
Seasonal small-scale spatial variability in alpine snowfall and
snow accumulation. Water Resour. Res.,49(3), 1446–1457 (doi:
10.1002/wrcr.20135)
Sturm M and Benson CS (1997) Vapor transport, grain growth and
depth-hoar development in the subarctic snow. J. Glaciol.,
43(143), 42–59
King and others: Tundra snow 17.2 GHz backscatter278
Sturm M and Benson C (2004) Scales of spatial heterogeneity for
perennial and seasonal snow layers. Ann. Glaciol., 38, 253–260
(doi: 10.3189/172756404781815112)
Sturm M and Wagner AM (2010) Using repeated patterns in snow
distribution modeling: an Arctic example. Water Resour. Res.,
46(W12), W12549 (doi: 10.1029/2010WR009434)
Sturm M, Holmgren J and Liston GE (1995) A seasonal snow cover
classification scheme for local to global applications. J. Climate,
8(5), 1261–1283
Tape KD, Rutter N, Marshall, HP, Essery R and Sturm M (2010)
Recording microscale variations in snowpack layering using
near-infrared photography. J. Glaciol.,56(195), 75–80 (doi:
10.3189/002214310791190938)
Tsang L, Pan J, Liang D, Li Z, Cline DW and Tan T (2007) Modeling
active microwave remote sensing of snow using dense media
radiative transfer (DMRT) theory with multiple-scattering effects.
IEEE Trans. Geosci. Remote Sens.,45(4), 990–1004 (doi:
10.1109/TGRS.2006.888854)
Ulaby FT, Moore RK and Fung AK (1982) Microwave remote
sensing, active and passive. Vol. II. Radar remote sensing and
surface scattering and emission theory. Artech House,
Norwood, MA
Willatt RC, Giles KA, Laxon SW, Stone-Drake L and Worby AP
(2010) Field investigations of Ku-Band radar penetration into
snow cover on Antarctic sea ice. IEEE Trans. Geosci. Remote
Sens.,48(1), 365–372 (doi: 10.1109/TGRS.2009.2028237)
Yueh S, Dinardo SJ, Akgiray A, West R, Cline DW and Elder K
(2009) Airborne Ku-band polarimetric radar remote sensing of
terrestrial snow cover. IEEE Trans. Geosci. Remote Sens.,47(10),
3347–3364 (doi: 10.1109/TGRS.2009.2022945)
MS received 23 January 2014 and accepted in revised form 12 October 2014
King and others: Tundra snow 17.2 GHz backscatter 279
