Available via license: CC BY 4.0
Content may be subject to copyright.
1
Potential of X-band polarimetric SAR co-polar phase difference for
Arctic snow depth estimation
Joëlle Voglimacci-Stephanopoli1, 2, Anna Wendleder3, Hugues Lantuit4, 5, Alexandre Langlois1, 2, Samuel
Stettner3, 6, Jean-Pierre Dedieu7, 2, Achim Roth3 and Alain Royer1, 2
1 Centre d’Applications et de Recherches en Télédétection, Université de Sherbrooke, Sherbrooke, J1K 2R1, Canada
5
2 Centre d’Études Nordiques, Université Laval, Québec, Québec, G1V 0A6, Canada
3 German Remote Sensing Data Center, German Aerospace Center, Oberpfaffenhofen, Germany
4 Institute of Geosciences, University of Potsdam, Potsdam, Germany
5 Alfred Wegener Institute Helmholtz Centre for Polar and Marine Research, 14473 Potsdam, Germany
6 German Space Agency, German Aerospace Center, Bonn, Germany
10
7 Institute of Environmental Geosciences, Université Grenoble-Alpes/CNRS/IRD, 38058 Grenoble, France
Corresponding author: Joëlle Voglimacci-Stephanopoli (joelle.voglimacci-stephanopoli@usherbrooke.ca)
Abstract. Changes in snowpack associated with climatic warming has drastic impacts on surface energy balance in the
cryosphere. Yet, traditional monitoring techniques, such as punctual measurements in the field, do not cover the full snowpack
spatial and temporal variability, which hampers efforts to upscale measurements to the global scale. This variability is one of
15
the primary constraints in model development. In terms of spatial resolution, active microwaves (synthetic aperture radar—
SAR) can address the issue and outperform methods based on passive microwaves. Thus, high spatial resolution monitoring
of snow depth (SD) would allow for better parameterization of local processes that drive the spatial variability of snow. The
overall objective of this study is to evaluate the potential of the TerraSAR-X (TSX) SAR sensor and the wave co-polar phase
difference (CPD) method for characterizing snow cover at high spatial resolution. Consequently, we first (1) quantified the
20
spatio-temporal variability of the geophysical properties of the snowpack in an Arctic catchment, we then (2) studied the links
between snow properties and CPD, considering ground vegetation. Snow depth (SD) could be extracted using the CPD when
certain conditions are met. A high incidence angle (> 30 °) with a high Topographic Wetness Index (TWI) (> 7.0) showed
correlation between SD and CPD (R-squared up to 0.72). Further, future work should address a threshold of sensitivity to TWI
and incidence angle to map snow depth in such environments and assess the potential of using interpolation tools to fill in gaps
25
in SD information on drier vegetation types.
1. Introduction
Snow cover is a key component of the cryosphere which plays an essential role for ecological processes and hydrological
dynamics. In arctic ecosystems, those processes include species survival (Dolant et al., 2018; Poirier et al., 2019), thermal
ground regime (Goodrich, 1982; Gouttevin et al., 2012; Stieglitz et al., 2003) or vegetation colonization and growth (Berteaux
30
et al., 2017; Kankaanpää et al., 2018; Myers-Smith et al., 2011a). In the past 40 years, we observed a pan-Arctic reduction in
https://doi.org/10.5194/tc-2021-314
Preprint. Discussion started: 13 October 2021
c
Author(s) 2021. CC BY 4.0 License.
2
the snow cover duration of 2–4 days per decade (AMAP, 2017) and maximum Arctic snow depth trend show a consistent
decrease since 1980 (AMAP, 2017; IPCC, 2019). These trends will undeniably change the arctic landscape. For instance,
duration of snow patches impacts vegetation phenology (Kankaanpää et al., 2018) and controls shrubs’ growth (Myers-Smith
et al., 2011b; Pomeroy et al., 2006). Hence, patterns of vegetation densification (also called greening) arise, and dense
35
vegetation such as shrubs impact the snowpack physical properties. Twigs induce a decrease in snow density and an increase
of depth hoar formation (Domine et al., 2016; Gouttevin et al., 2018; Sturm et al., 2001). By protruding above the snowpack
surface, shrubs reduce surface albedo and advance the snow melt timing (Sturm et al., 2001). Coupled to a decreasing trend
on maximum snow depth and snow cover duration observed (AMAP, 2017; IPCC, 2019), the greening of the Arctic is likely
to lead to drastic modification of the snowpack. A recent update on the classification of Sturm et al. (1995) suggested by Royer
40
et al. (2021) demonstrates a positive feedback on climate warming owed to snow-vegetation interaction. High resolution land
cover classification is therefore needed to address changes in the snowpack in a warming climate.
Current snow modules used in Earth System Models are based on coarse spatial resolution of tens of kilometres (Bokhorst et
al., 2016). Coarse special resolution hampers our efforts to understand the dynamics driving snowpacks at the landscape scale.
Indeed, snow is characterized by a high spatial and temporal heterogeneity ( e.g.: Rutter et al., 2014; Thompson et al., 2016;
45
Wilcox et al., 2019). Traditional approaches using in situ measurement can provide very detailed spatial information on snow
properties, but cannot be deployed over large areas. There is therefore a strong need to bridge these two scales and provide
means to monitor the temporal and spatial variability of the snowpack over larger areas.
Earth observation satellites can provide frequent measurements over larger areas. Space borne platforms are widely used to
monitor snow on local, regional, and global scales. Yet, they also suffer from strong limitations. Optical sensors allow
50
measurements on surface characteristics of snow, but do not provide direct measurements of the properties of the snowpack
and are often limited by cloud cover. Passive microwave monitoring methods are operational and provide continuous data, but
suffer from the coarse spatial resolution of satellite observations (e.g.: Frei et al., 2012)). Active microwave observations with
synthetic aperture radar (SAR) can overcome these issues in providing high resolution frequent snow measurements over large
areas.
55
SAR can “see” through clouds while being independent from solar illumination. SAR sensors are interesting to collect data
from the snowpack because they can, on the one hand, transmit and receive microwaves in horizontal (H) and vertical (V)
polarization, and on the other hand, their microwaves can interact with and penetrate into the observed material. The main
challenge related to the use of SAR is the lack of a reliable method to relate satellite data to physical measurements in snow-
impacted environments.
60
The objective of this paper is therefore to evaluate the potential of polarimetric method co-polar phase difference (CPD)
produced with the X-band satellite TerraSAR-X to retrieve SD from an arctic snowpack where vegetation is highly variable.
This general objective requires a complete characterization of the snowpack from field data to fully understand the sensitivity
of CPD to various snow characteristics. This requirement motivates the following two specific objectives: (1) investigate SD
variability between different vegetation classes in the Ice Creek catchment (Qikiqtaruk-Herschel Island, Yukon, Canada) using
65
https://doi.org/10.5194/tc-2021-314
Preprint. Discussion started: 13 October 2021
c
Author(s) 2021. CC BY 4.0 License.
3
in situ measurements collected over the course of a field campaign in 2019 and (2) evaluate linkages between SD and CPD
distributions considering meteorological data over the 2015–2019 period.
2. Background: Co-polar phase difference—snow structure
2.1. Arctic snow properties
Snow cover in the Arctic is mostly characterized by two main layers (Domine et al., 2016; Royer et al., 2021; Sturm et al.,
70
2008). The upper layer, the wind slab, is very compact as it is subject to sustained winds and cold temperatures that promote
cohesion of snow grains (Domine et al., 2018b; Sturm et al., 2008). The basal layer generally consists of depth hoar (DH)
grains that develop under a kinetic metamorphic regime in dry snow conditions with a sustained strong temperature gradient
(Domine et al., 2016).
The snowpack is driven by two types of metamorphic regimes, namely wet and dry snow metamorphism (Bernier et al., 2016).
75
These regimes develop according to the temperature gradient in the snowpack and to its liquid water content (Colbeck, 1973).
Wet snow metamorphism, with liquid water available in the snowpack, will lead to different metamorphic processes for
saturated and unsaturated conditions (Colbeck, 1982). As a result, there will be a major impact on microwave radiative transfer
given that wet snow acts as a blackbody in such frequencies (e.g. Rott and Matzler, 1987). In the case of an arctic snowpack,
a regime of dry snow metamorphism is generally found when sustained cold temperatures last during most of the winter
80
(Domine et al., 2018a). Schneebeli et Sokratov (2004) found that snow crystals are highly anisotropic (dependency to a
direction of an object), which is correlated with snow metamorphism (Calonne and al., 2014; Gouttevin and al., 2018). As
such, over the time, snow crystals become elongated to a vertical direction after their setting up in the snowpack.
The geometrical structure of the snow will characterize the electromagnetic wave propagation through the snowpack by
scattering and absorption processes within each layer (Mätzler, 1987). Given the dry nature of the arctic snowpack, the main
85
source of backscattering should occur at the snow-ground interface for frequencies in X-band ( = 3.1 cm) such as used in this
study or below as dry snow can be considered as a homogeneous, “non-scattering” and non-absorbing volume (Leinss et al.,
2014). This said, inhomogeneous layers such as ice layers, melt/freeze crust and any strong vertical change in dielectric
properties (i.e. density, wetness) can also affect the signal.
2.3. Co-polar coherence
90
The co-polar coherence (CCOH) indicates the correlation coefficient of HH and VV phase centers. The magnitude of the
CCOH ranges between 0 and 1 where a weak correlation (< 0.5 as defined by Leinss et al. [2014]) indicates a low scattering
with a more chaotic and randomly phase shifts between HH and VV waves and are hence omitted. Such weak correlations will
occur when HH and VV waves have different phase centers and different scattering targets. A decrease in correlation can also
be induced by a strong surface scattering caused by rough or wet surfaces, volume scattering during winter or during snow-
95
https://doi.org/10.5194/tc-2021-314
Preprint. Discussion started: 13 October 2021
c
Author(s) 2021. CC BY 4.0 License.
4
free conditions where vegetation is exposed (Fig.1). The equation of CPD is only valid where no volume scattering occurs
(Leinss et al., 2014) since an increase in volume scattering will lead to a decrease of the CCOH.
Figure 1. Phase shift can also be caused by scattering effects within the snowpack or by surface roughness (including vegetation) (Credit:
A.Wendleder).
100
2.4. Co-polar phase difference
CPD is a polarimetric method using difference in the phase between HH and VV polarization channels. The phase difference
refers to the difference in the propagation speed of a wavelength in a material as a function of polarization, which then causes
a phase difference in the electromagnetic wave between polarizations. The phase of a single polarization is assumed to have a
uniform distribution over [-π, π] (Leinss et al., 2014; Patil et al., 2020).
105
A relationship was found between CPD and snowfall by Chang et al. (1996) and Leinss et al. (2014) which induces a
propagation delay among horizontal and vertical phases due to horizontal alignments of fresh snow crystals. Recent studies
focused on the boreal region (Leinss et al., 2014, 2016) or were applied in arctic region with no or sparse vegetation (Dedieu
et al., 2018) so the application of the CPD method in the Arctic remains poorly documented. It could be hypothesized that the
CPD can describe the entire snowpack in such cold and dry environments. Strong vertical changes in density and grain size
110
could also lead to a decrease in coherence so that the use of CPD information might not be suitable (i.e. when CCOH < 0.5,
see Fig. 1).
3. Data and Methods
3.1. Study site
Qikiqtaruk-Herschel Island (69° 35’ N, 139° 06’ W) is located about 2 km off the Yukon Coast in the northwestern Canadian
115
Arctic (Fig. 2). With an approximate area of 108 km2, this island has a rolling topography (max. altitude: 183 m a.s.l.),
dissected by numerous geomorphological forms such as gullies, valleys and polygonal soils (Short et al., 2011; Stettner et al.,
2018). The permafrost on Qikiqtaruk-Herschel Island is continuous with a high ice content. Ground ice can be observed on the
island in the form of ice wedges, ice lenses, or buried snowbanks, as observed by Pollard (1990). Results from Wolter et al.
(2016) suggested that geomorphological processes, such as permafrost degradation, are strongly related to vegetation
120
https://doi.org/10.5194/tc-2021-314
Preprint. Discussion started: 13 October 2021
c
Author(s) 2021. CC BY 4.0 License.
5
composition on Qikiqtaruk-Herschel. The active layer thickness varies from 45 cm to 90 cm in marine deposits (silty
diamicton) and can reach over a 110 cm in porous deposits (Lantuit and Pollard, 2008; Smith et al., 1989). A thickening of the
active layer by 15 cm to 25 cm was documented on the island during the period from 1985 to 2005, as well as an increase in
the mean annual air temperature by 2.7 °C between 1970 and 2005 (Burn and Zhang, 2009).
Spatial distribution of snow on the island is primarily based on topography, due to the low tundra-type vegetation (Burn and
125
Zhang, 2009). The snow is blown away from the uplands and accumulates in topographic depressions such as valleys and
hummocky terrain (Burn and Zhang, 2009). The dominant wind direction is northwest with frequent storms in late August and
September (Solomon, 2005). A study by Myers-Smith et al. (2011) indicates an increase in the canopy and vegetation height
over the last century that can be expected to have an impact on the snow cover structure.
In this paper, we performed our measurements in the Ice Creek catchment (area of 1.54 km2) located at the eastern end of the
130
island (Fig. 2). The digital elevation models from ArcticDEM (2 m res.) indicates average slopes of 2.9° with maximums of
13.2° at altitudes ranging from 5 m to 94 m (Porter et al., 2018).
Figure 2. (a) Location of the study site in the Arctic (b) Visual extent of TerraSAR-X passages on Qikiqtaruk Herschel Island (c) Ice Creek
study site including the location of measurements. The black crosses were revisited during each TerraSAR-X acquisitions (Imagery provided
135
by Worldview 01.01.1001, True Color). The meteorological station belongs to Environment and Climate Change Canada (ECCC).
https://doi.org/10.5194/tc-2021-314
Preprint. Discussion started: 13 October 2021
c
Author(s) 2021. CC BY 4.0 License.
6
3.2. Snow distribution over Ice creek
3.2.1. Snow measurements
Two sampling strategies were used for the snowpit characterization (Table 1). First, detailed snowpit measurements were
conducted along predefined locations at an average distance of 200 m between each site (Fig. 2c). The snowpit locations in
140
the centre of the Ice Creek catchment as well as location at the outlet of the catchment were revisited during each TerraSAR-
X (TSX, see 3.3.) acquisition so that soil characteristics remain unchanged between snow sampling and satellite measurements.
Snow depths were measured using a GPS snow depth probe around the snowpits, ensuring the representativeness of the snowpit
location. This was conducted by measuring depths in a growing circle moving away from the snowpit location until an
approximate diameter of 30 m was reached, which is typically the area required to ensure representativeness in tundra
145
environments (Clark et al., 2011). Snowpits and SD measurements were then distributed spatially elsewhere in the catchment
to refine the characterization of snow within the catchment. Additionally, two SD transects were conducted across the
catchment to analyze the SD distribution in the study site. Both transects were established from the east side to the west side
of the Ice Creek catchment.
Detailed snow profiles were acquired in spring 2019 (mid-April to early May). In each site, we dug snowpits in a way to avoid
150
direct solar illumination of the snow wall. High resolution vertical profiles of density, temperature, grain size and type were
conducted according to Fierz et al. (2009, see Table 1). Specifically, layered density profiles were obtained by extracting snow
samples from each identified layer using a 100 cm3 density cutter and weighed using a Pesola light series scale. Temperature
profiles were measured at 3 cm intervals using a Cooper digital thermometer, and profile measurements included shadowed
surface temperature as well as soil-snow interface.
155
From the above observations, each layer was classified according to their density and snow grain type across 5 classes
following Fierz et al. (2009): 1) Depth Hoar, 2) windslab, 3) surface hoar, 4) fresh snow, 5) melt-freeze crust and ice layer.
The snow depth, mean density of each layer classified, was compiled for later linear regression analysis with TSX data.
https://doi.org/10.5194/tc-2021-314
Preprint. Discussion started: 13 October 2021
c
Author(s) 2021. CC BY 4.0 License.
7
160
Table 1: In situ measurements during the 2019 field campaign
Snowpits
See Fig 2c
Stratigraphy
Snow height
Size and grain type (visual estimation)
Temperature profile (measurement at 3 cm, ± 0.1 °C)
Snow density by layers (measurements at 5 cm when possible, ± 0.5 kg m-
3)
Environment and Climate Canada
(ECCC) meteorological station
69.5682° N, 138.9134° W
Wind speeds at 10 m and 2 m (ms-1, hourly)
Precipitation gauge for total precipitation (mm) and rate (mm h-1),
Temperature (°C) and relative humidity (%).
Datalogger — Campbell Scientific CR3000E
3.2.2. Vegetation units
The classification of the different vegetation units was obtained from Eischeid (2015) following the initial definition developed
by Smith et. al. (1989). The classification was determined by the soil type, vegetation observed and geomorphological features.
The dataset used in this study was derived from 2015 GeoEye satellite data (res.: 1.65 m) (Eischeid, 2015). For the specific
165
needs of this paper, we focused on the following specific classes: Arctic Willow and Dryas Vetch (hereinafter referred as
Dryas), Arctic Willow and Lupine (Lupine), Shrub Zone (Shrub) and Willow Saxifrage Coltsfoot (Coltsfoot). These classes
were selected given that they are physically and spatially different (see Fig. 3d), which is of primary importance from a snow
microstructure and radar backscattering perspective.
The Lupine class is associated with an irregular and hummocky terrain (Eischeid, 2015, see Fig. 3d). High variability in
170
microtopography results in equally heterogeneous SD at a similar scale (Sturm and Holmgren, 1994). Erosion rates and
moisture content will vary greatly following terrain instability (Eischeid, 2015). The Coltsfoot class is common in wetlands,
where the ground is generally saturated and composed of shrubs (Eischeid, 2015). This vegetation class is located at the bottom
of valleys, which is suitable for snow accumulation (Burn and Zhang, 2009). The Shrub class was added by Eischeid (2015).
to the original classification by Smith et al. (1989) to reflect the growing importance of shrubs on the island. It is characterized
175
by non-hydrophilic vegetation with lower soil moisture. Finally, the Dryas class is common on the gently undulating upland
slopes (Smith et al., 1989). The associated soil type is a moderately well-drained Turbic Cryosol, which shows evidence of
cryoturbation, as well as bare soil. Each snowpit characteristics and SD measurement were grouped by vegetation units to
extract means and standard deviation by vegetation classes. The snowpits made along the two transects were grouped when
https://doi.org/10.5194/tc-2021-314
Preprint. Discussion started: 13 October 2021
c
Author(s) 2021. CC BY 4.0 License.
8
they were at a distance less than 30 m and statistics of distribution (average and standard deviation) were extracted to complete
180
the data analysis.
Figure 3. Vegetation classes occurring in the Ice Creek catchment: (a) irregular and hummocky terrain observed in Lupine class (Credit: M.
Fritz) (b) Low vegetation on well drained area (such as Dryas) (Photo by the author) and (c) Shrub and wetland (such as Coltsfoot class)
(Credit: M. Fritz). (d) Vegetations unit’s distribution in the study area (from Obu et al. (2017) as defined initially by Smith et al. (1989). The
185
classes in grey are not include in the analysis.
3.3. Snow- SAR correlation
3.3.1. SAR acquisition and preprocessing
A total of five TSX acquisitions in HH and VV polarizations over three different orbits were obtained during spring 2019,
encompassing areas where snow measurements and vegetation information was available (Table 2). Snowpits and SD
190
measurements taken before and after (± 2 days) each TSX acquisition were included in the analysis as no precipitation occurred
and air temperature was stable during the field campaign. Additionally, a time series of TSX acquisitions for the 2014–2019
period (orbit 24, θ= 31°) was analyzed to evaluate the inter-annual variability of snow conditions on the island. The full TSX
https://doi.org/10.5194/tc-2021-314
Preprint. Discussion started: 13 October 2021
c
Author(s) 2021. CC BY 4.0 License.
9
dataset was first processed at the DLR (German Aerospace Center). The preprocessing is described in Schmitt et al. (2015),
and includes the determination of the Kennaugh elements, their radiometric calibration and orthorectification. The images
195
were georeferenced in UTM with a ground sampling distance of 5 m. To reduce speckle noise, we used the multi-scale multi-
looking algorithm developed by Schmitt (2016; Schmitt et al., 2015). CPD and CCOH can directly be derived from the
radiometric and geometrically calibrated Kennaugh elements. The Kennaugh matrix describes the polarimetric information
and allows to differentiate the physical scattering mechanisms (e.g. double bounce, volume and surface scattering) affecting
the signal, which in turn can be linked to snow characteristics. The following Kennaugh elements were used in the CPD
200
equation:
(1)
where K7 is the phase shift between HH, and VV phase centre described as
(2)
and where K3 is the scattering difference between surface to double bounce:
205
(3)
To reduce the loss of coherence, a threshold was applied to CPD pixels where CCOH was less than 0.5, following Leinss et
al. (2014). Again, the Kennaugh elements were used in the CCOH equation:
(4)
210
A total of 32 pixels had a CCOH less than 0.5, hence showing a random phase shift between waves which is not optimal for
CPD applications. These pixels were therefore removed from the analysis. To discard the potential effect of slope on crystal
grains orientation, 5 pixels with a slope greater than 10° were subsequently extracted (3 in the Dryas class, 2 in the Lupine).
To assess the temporal variability of CPD signal, pixels were divided by vegetation class for the period 2015–2019.
215
Table 2: TSX acquisition on Qikiqtaruk-Herschel Island. All orbits were used for linear regression with in situ snow measurements.
Orbit 24 has a sufficient time series and was used to extract temporal evolution of CPD.
Relative
orbit
Flight
direction
Polarization mode
Incidence
angle
Observation period
Number of
scenes
In situ data
24
Descending
HH, VV
31°
2014.12.26—2018.03.06
2019.04.17—2019.05.20
104
2019.04.17
2019.04.28
152
Ascending
HH, VV
24°
2019.04.15—2019.05.18
24
2019.04.26
115
Descending
HH, VV
38°
2019.04.23—2019.05.15
24
2019.04.23
2019.05.04
https://doi.org/10.5194/tc-2021-314
Preprint. Discussion started: 13 October 2021
c
Author(s) 2021. CC BY 4.0 License.
10
3.3.2. Linking snow depth to CPD
Implication of Snow Geometry
We focused on the SD variability between vegetation classes. We also evaluated depth hoar fraction (DHF) given that King et
220
al. (2018) found that X-band backscattering is highly sensitive to depth hoar grains. This allowed us to assess if any
discrepancies in SD retrieval can be linked to large grain size. In addition, horizontal structures such as ice layers and
melt/freeze crusts were identified for the same purpose of testing the SD retrieval capabilities in different stratigraphic contexts.
Dedieu et al. (2018) showed that the attenuation of the SAR signal was caused by ice layers of 3 to 5 cm thick, but lingering
uncertainties remain with regards to the contribution of thinner ice lenses such as the ones found on Qikiqtaruk-Herschel
225
Island.
Topographic Wetness Index as a proxy
SD retrieval is challenging because it is impacted by snow surface properties. SD retrieval with CPD may be impacted by the
dielectric properties of the snow surface, since the main backscatter signal is expected at the snow ground interface. High
moisture content at the soil surface would potentially improve the performance of SD retrieval, because the presence of ice
230
leads to better reflection conditions for the microwave. The Topographic Wetness Index (TWI) was chosen as a proxy to
analyze the variance between vegetation groups. Given the high sensitivity of microwaves to wetness, the high variability of
TWI between each vegetation class will lead to different responses in backscattering through changes in the dielectric constant
of the soil. The TWI was first developed by Beven and Kirkby (1979) within the runoff model TOPMODEL using the
following equation:
235
(5)
where is the local slope and a is the upslope area per unit which is obtained with the upslope area (the cells contributing
to the runoff to the cells of interest, A) and the contour length (L) following a=A/L. The upslope area calculated is based on
D8 flow direction algorithm (O’Callaghan and Mark, 1984) and the TWI values were computed based on the ArticDEM
constructed from the DigitalGlobe Constellation (Porter et al., 2018, res: 2 m). Each TWI values derived from the catchment
240
was combined to vegetation classes and CPD cells as described above.
Analysis
The Shapiro-Wilk test was used to test the normality of distributions for SD and TWI. Since TWI and SD distributions did not
respect a normal distribution, the variance in TWI and SD between each group was tested with the non-parametric test Welch
ANOVA in conjunction with a post-hoc Games-Howell test. We use the Games-Howell test as it does not assume equal
245
variances and sample sizes (Games and Howell, 1976).
We evaluated the correlation between snow characteristics and CPD using a linear regression analysis. The median value was
extracted when more than one snow measurement was found in the same TSX pixel (5 m). Thus, a total of 371 pixels was used
in the analysis (average number of snow measurements per pixel: 1.7). The median SD by pixels were grouped by vegetation
https://doi.org/10.5194/tc-2021-314
Preprint. Discussion started: 13 October 2021
c
Author(s) 2021. CC BY 4.0 License.
11
classes and orbit. Durbin-Watson’s test and Breusch-Pagan’s were used to assess autocorrelations and homoscedasticity of
250
distribution data. Significance for all tests were calculated with α = 0.05.
4. Results
4.1. Snow distribution
4.1.1. Snow depth and depth hoar fraction along transects
Measurements of SD from transect #1 (Fig. 4) varied between 20 and 250 cm where the peak was measured at the valley
255
bottom. Further west, SD values decreased significantly on the slope with values between 20 and 50 cm. Highest DHF along
the transect were found on the west side of the transect and on the slopes with an average of 0.76 while an average of 0.39 was
observed on the east side of the catchment. Along transect #2 (Fig. 4), and snow cover was also deeper at the bottom of the
valley (from 120 cm up to 200 cm) and decreased significantly on slopes and higher elevation areas (30 cm to 50 cm).
260
Figure 4. Snow depth (SD) transects surveyed in the Ice Creek catchment. Transect #1 and transect #2 shows snow depth (solid line) and
altitude (m.a.s.l., dashed line) along transect. Mean Depth hoar (DH) ratio are indicated along transects by proportional size. 1a, 1b and 2c
contain one observation. See Fig. 2c for their location in the catchment.
https://doi.org/10.5194/tc-2021-314
Preprint. Discussion started: 13 October 2021
c
Author(s) 2021. CC BY 4.0 License.
12
4.1.2. Snow characteristics by vegetation classes
Snow Depth
265
The average SD within Ice Creek catchment was 47.4 cm ± 39.6 cm. The range of variability was substantial, with minimum
value at 8.0 cm and a maximum at 212.0 cm. The standard deviation of the snow depth was variable yet strong among all
classes (Table 3). The largest standard deviation measured was over Lupine (22.3 cm or 57 % of the mean SD) followed by
Coltsfoot (67.6 cm or 54 % of the mean SD). Coltsfoot was by far characterized a greater SD than any other class.
Despite the great deviation around the mean for each class, the Welch ANOVA (Table 4) shows that SD is significantly
270
different between all vegetation groups (p-value = 5–13). Games-Howell post-hoc revealed that Coltsfoot SD measurement is
significantly different from the other classes as well as Dryas. The difference between Lupine and Shrub is not significant.
Topographic Wetness Index
The average TWI was 6.1 ± 1.6. The minimum and the maximum ranged between 2.5 and 14.7 while the average by vegetation
classes range between 7.4 and 5.5 with a weak deviation around the means (Table 3). Coltsfoot showed the highest TWI which
275
is consistent with its location in the valley (Fig. 3) and its vegetation group type, characterized by hydrophilic vegetation. The
Welch ANOVA shows that wetness index extracted with TWI is significantly different between all vegetation groups with a
p-value <0.001 (p-value = 6–14), which again is expected to lead to different responses from TSX. The Games-Howell post-
hoc test revealed that difference in TWI is not significant between Coltsfoot and Shrub units and between Dryas and Lupine
(Table 5).
280
Depth Hoar Fraction
The DHF of the snowpack was larger than 0.5 for all vegetation classes. However, standard deviations of classes with shallow
snow such as Lupine and Dryas were greater whereas the standard deviation was lower than 0.1 for Coltsfoot where the SD is
the highest. At least one horizontal structure (ice layer or melt-freeze crust layer) was found in each snowpit. The average
thickness of each ice or melt-freeze crust layer was 1.6 cm ± 0.7 cm and the cumulative thickness average by snowpit was
285
4.5 cm ± 2.8 cm. The maximum ice thickness (4 cm) was found at the station downstream, which would suggest that the
sensitivity of the CPD to ice layers should be generally lower than what was found by Dedieu et al. (2018). The high
stratification otherwise may attenuate the signal.
https://doi.org/10.5194/tc-2021-314
Preprint. Discussion started: 13 October 2021
c
Author(s) 2021. CC BY 4.0 License.
13
Table 3: Averaged SD and DHF for each of the vegetation class
290
Table 4: Post-hoc analysis with Games-Howell for snow depth in vegetation classes (non-parametric test). Each row present variance
between snow depth means from two different groups. All vegetation groups were tested on each other.
Snow depth
Class 1
Class 2
Mean Difference
(cm)
Standard error
(cm)
p-value
Coltsfoot
Dryas
+94.2
12.60
0.001
Coltsfoot
Lupine
+87.1
12.71
0.001
Coltsfoot
Shrub
+81.9
13.14
0.001
Dryas
Lupine
-7.14
2.36
0.014
Dryas
Shrub
-12.34
4.07
0.015
Lupine
Shrub
-5.2
4.41
0.622
Table 5: Post-hoc analysis with Games-Howell for the TWI (non-parametric test). Each row present variance between TWI means
from two different groups. All vegetation groups were tested on each other.
TWI
Group 1
Group 2
Mean Difference
(Wetness index)
Standard error
(cm)
p-value
Coltsfoot
Dryas
+1.56
0.20
0.001
Coltsfoot
Lupine
+1.97
0.22
0.001
Coltsfoot
Shrub
+0.63
0.30
0.147
Dryas
Lupine
+0.41
0.17
0.076
Dryas
Shrub
-0.93
0.26
0.003
Lupine
Shrub
-1.34
0.28
0.001
Vegetation class
SD and
TWI nb of
samples
Averaged SD
±σ (cm)
Averaged
TWI ±σ
DHF nb of
samples
Averaged DHF ±σ
(%)
Coltsfoot
29
126.0 ± 67.6
7.4 ± 0.9
8
0.55 ± 0.06
Dryas
146
31.8 ± 14.1
5.9 ± 1.2
16
0.62 ± 0.31
Lupine
118
38.9 ± 22.3
5.5 ± 1.5
21
0.60 ± 0.21
Shrub
28
44.1 ± 20.6
6.8 ± 1.3
6
0.51 ± 0.18
Other units
50
69.3 ± 47.2
6.8 ± 2.4
7
0.58 ± 0.20
Average in catchment
371
47.4 ± 39.6
6.1 ± 1.6
58
0.59 ± 0.22
https://doi.org/10.5194/tc-2021-314
Preprint. Discussion started: 13 October 2021
c
Author(s) 2021. CC BY 4.0 License.
14
4.2. TerraSAR-X results
295
4.2.1. Spatial and temporal evolution of CPD
Figures 5a and 5c show the averaged temporal evolution of CPD and CCOH with descending orbit 24 (incidence angle = 31°)
for each vegetation class as well as the confidence interval (95%). The period with presence of snow was set between mid-
September and mid-May based on prior observations (Burn and Zhang, 2009; Stettner et al., 2018). Figure 5c shows the
monthly average temperature and cumulative monthly precipitation on Qikiqtaruk-Herschel Island.
300
A periodicity was observed with the CPD signal, with on one side, the period of snow-free condition where the signal oscillates
around zero, and on the other side, the period with snow where the signal decreased over the season suggesting an influence
from the snowpack. For the 2014–2019 period, the mean CPD value during the snow season was -8.59° with annual mean
ranging between 13.41° (2014–2015) and -6.42° (2017–2018). During the snow-free con<