In this paper we introduce a parameter for the retrieval of the thickness of undeformed first-year sea ice that is specifically adapted to compact polarimetric (CP) synthetic aperture radar (SAR) images. The parameter is denoted as the “CP ratio”. In model simulations we investigated the sensitivity of the CP ratio to the dielectric constant, ice thickness, ice surface roughness, and radar incidence angle. From the results of the simulations we deduced optimal sea ice conditions and radar incidence angles for the ice thickness retrieval. C-band SAR data acquired over the Labrador Sea in circular transmit and linear receive (CTLR) mode were generated from RADARSAT-2 quad-polarization images. In comparison with results from helicopter-borne measurements, we tested different empirical equations for the retrieval of ice thickness. An exponential fit between the CP ratio and ice thickness provides the most reliable results. Based on a validation using other compact polarimetric SAR images from the same region, we found a root mean square (rms) error of 8 cm and a maximum correlation coefficient of 0.94 for the retrieval procedure when applying it to level ice between 0.1 and 0.8 m thick.
Sea ice covers about one-tenth of the world ocean surface and significantly affects the exchanges of momentum, heat, and mass between the sea and the atmosphere. Not only is sea ice extent a significant indicator and effective modulator of regional and global climate change, but sea ice thickness is also an important parameter from a thermodynamic and kinematic perspective (Soulis et al., 1989; Kwok, 2010). The decline of sea ice extent recently observed in the Arctic, e.g., is linked with a decrease of ice thickness and increasing fractions of seasonal ice areas (e.g., Kwok et al., 2009). Measurements of sea ice thickness are compared with model results to control and validate the model capabilities of reproducing recent and predicting future trends of sea ice conditions in the Arctic (e.g., Laxon et al., 2013). Although sea ice thickness is only several meters at most, it forms an effective thermal insulation layer due to its high albedo and low thermal conductivity, leading to a significant reduction in the heat flux from the ocean to the atmosphere, especially in winter (Vancoppenolle et al., 2005). Besides investigations focusing on the entire Arctic or Antarctic region, other studies analyze ice thickness variations on local scales to improve regional ice thickness retrievals (e.g., Haapala et al., 2013). Operational services charged with providing sea ice maps and forecasting ice conditions for marine transportation and offshore operations need near-real-time regular information about local and regional ice thickness distributions. The use of sensors providing high spatial resolutions on the order of 100 m or better for ice thickness retrieval, such as synthetic aperture radar (SAR), is an important topic of recent research (Dierking, 2013).
Unfortunately, the sea ice thickness distribution is also one of the most
difficult parameters to measure. The most direct and accurate measurement
technique is in situ drilling with an ice auger. Although it provides data
with sufficient accuracy (in the range of centimeters), it is time-consuming
and spatially limited. Therefore, this method is mainly used for the calibration
of other sensors or methods. To obtain ice thickness distributions at larger
spatial scales, remote sensing methods are requisite tools. There are
generally different strategies.
Measurements of ice draft using upward-looking sonar on ocean moorings or
submarines (Wadhams, 1980; Behrendt et al., 2013) from which thickness is
estimated based on assumptions about buoyancy, ice density, and snow load
(e.g., Rothrock et al., 1999) are used. Such data provide information about detailed
temporal thickness variations (daily or even hourly) at a fixed location. An
example for using in situ measurements of ice thickness from the New Arctic
Program initiated by the Canadian Ice Service (CIS) starting in 2002, and
sea ice draft measurements from moored upward-looking sonar (ULS) instruments in the Beaufort Gyre
Observing System for testing a method of ice thickness retrieval from
optical methods is provided by Wang et al. (2010). Measurements of sea ice freeboard (i.e., the part of the ice above the
water level) plus snow layer thickness with laser altimetry (e.g., Wadhams
et al., 1992; Dierking, 1995) are used. From such data, the average ice thickness can
be estimated, or the probability density function (PDF) of ice freeboard can
be converted to a PDF of ice thickness. However, the estimation of ice
thickness from freeboard data is less reliable than from ice draft because
of a relatively stronger impact of errors in the freeboard measurements
(Goebell, 2011). The distance between snow surface and ice bottom is measured with
electromagnetic induction sounders (EMSs) mounted on sledges, ships, or
helicopters/airplanes (Goebell, 2011; Haas et al., 1997; Prinsenberg et al.,
2012a, b). With such systems, spatial ice thickness variations measured
at horizontal distances of a few 10 m were obtained in various regions
(Kovacs et al., 1987; Rossiter and Holladay, 1994; Haas et al., 2006;
Hendricks et al., 2011).
Although ULSs and EMSs have all contributed greatly to our knowledge about ice
thickness distributions on local and regional scales, such data can only be
obtained at specific locations over a limited time period. Satellite
remote sensing, on the other hand, is useful to monitor ice thickness
variations regularly over much larger areas.
On a still experimental basis, data of L-band passive microwave sensors, such as for example the Soil Moisture and Ocean Salinity mission (SMOS) radiometer, have been employed to retrieve thickness of sea ice thinner than about 0.5 m. The limitation of this approach is that it is only possible for very high (almost 100 %) sea ice concentration and in cold freezing conditions (Tian-Kunze et al., 2014; Huntemann et al., 2014). A space-borne altimeter has been used primarily to map ice thickness, and to monitor and study trends in thickness changes. The capabilities of laser and radar altimeter systems (such as CryoSat-2 and ICESAT) of measuring ice freeboard have been extensively investigated during the last decade (e.g., Kwok and Cunningham, 2008; Kwok et al., 2009; Laxon et al., 2013). Compared with radiometers, which only collect data at a coarse spatial resolution of a few to tens of kilometers (e.g., 25 km for the Special Sensor Microwave Imager (SSM/I) 37 GHz data), the spatial resolutions of radar altimeter systems are about 250 m along-track and 1.5 km across-track for CryoSat-2, and the footprint for ICESAT is about 70 m diameter. The sea ice products derived from altimeters usually focus on large-scale spatial and temporal variations. While the large-scale ice thickness product is important for climate research, the support of marine navigation and offshore operations in polar areas are crucially dependent on precise and reliable sea ice thickness maps with spatial resolutions better than 1 km.
Space-borne synthetic aperture radar (SAR), which operates in the microwave
frequency band, provides all-weather and day–night high-resolution imagery
(within a range of 1–100 m) with 1–3 days' temporal coverage.
Hence, SAR is in general very useful for operational mapping tasks on
regional and local spatial scales (Dierking, 2013). The disadvantage of SAR
systems is that higher spatial resolutions are linked with a limited
coverage between 10 and 500 km, compared, for example, to more than 1000 km
for passive microwave radiometers. SAR measures the intensity of the radar
signal backscattered from the ice surface and volume at different
polarizations. The backscattered intensity depends on the dielectric
constant of the ice and small-scale (mm–dm range) ice properties such as
ice surface roughness and air bubble fractions and sizes. If at least two
polarizations are measured simultaneously, the SAR, which is a coherent
device, can also provide the phase difference between the differently
polarized channels. The most recent SAR sensors have polarimetric
capabilities. A fully polarimetric radar transmits and receives both linear
horizontal (H) and vertical (V) polarized electromagnetic waves. Amplitude
and phase information of the backscattered signal are recorded for four
transmit/receive polarizations (HH, HV, VH, and VV). This mode is commonly
referred to as “quad-pol”. Quad-pol scenes are usually acquired at very
high spatial resolution. A RADARSAT-2 quad-pol scene has a spatial
resolution of 4.7 m (slant range)
Recently a number of investigators noted correlations between ice thickness and the co-polarization ratio, which is the ratio of measured intensities at VV and HH polarization (here we use VV / HH). The sensitivity between co-polarization ratio and thin ice thickness was first demonstrated by Onstott (1992), based on C-band radar data from the eastern Arctic region. Kwok et al. (1995) estimated the thin ice thickness (0 to 0.1 m) from L- and C-band fully polarimetric airborne SAR data acquired over the Beaufort Sea. Their approach included the training of a neural network. L-band polarimetric characteristics of ice in the Sea of Okhotsk were investigated by Wakabayashi et al. (2004), and the L-band co-polarized ratio was used to estimate ice thicknesses between 0 and 2 m (their Fig. 13). The investigation was further extended to other sensors, e.g., to the airborne Pi-SAR (X- and L-band data from the Sea of Okhotsk; Nakamura et al., 2009a; Toyota et al., 2009) and to Envisat ASAR, using radar intensity and ice thickness data from 0.2 to 2.5 m, the latter acquired from a research vessel in Lützow-Holm Bay, Antarctica (Nakamura et al., 2009b). The good correlations were attributed to the fact that the co-polarized ratio values are sensitive to the dielectric constants of the ice surface layer which change due to the process of desalination during ice growth. The relationship between relatively thick multi-year ice (thickness between 2 and 5 m), on the one hand, and co-polarized correlation and cross-polarized ratio HV/HH or VH/VV, on the other hand, was also investigated in the Arctic Ocean employing RADARSAT-2 and TerraSAR-X data (Kim et al., 2012). They found that the degree of depolarization is linked to the thickness of the multi-year ice as ice surface roughness increases and salinity decreases.
Although the above-mentioned parameters derived from polarimetric SAR imagery have shown the potential for estimating sea ice thickness under certain conditions, polarimetric SAR data can presently only be acquired at limited swath-widths. The quad-pol mode on RADARSAT-2 has a swath width of only 25–50 km, as mentioned above. The swath width of the VV/HH dual-polarization StripMap mode on TerraSAR-X is 15 km. Therefore, they are insufficient for operational use which requires a large-scale coverage (Scheuchl et al., 2004). The limited swath width also restricts scientific investigations to local domains. An alternative is to use compact polarimetry.
The methods of generating compact polarimetric (CP) information (explained below) are based on receiving data at two different polarizations (Souyris et al., 2005; Raney, 2007). Compared with the “traditional” dual-polarization modes described above, CP data include a greater amount of polarization information (but less than quad-polarization data). They can cover much greater swath widths compared to quad-polarization modes due to reduced power consumption and data storage requirements.
The term “CP system” refers to a unique polarization in transmission and
coherent dual-orthogonal polarizations in reception. There are three
different CP configurations (Nord et al., 2009). The first one is
the
However, one apparent disadvantage of the CP mode as compared to dual- or quad-polarization mode is the fact that the HH, VV, and HV signal combinations are not directly measured. This means that the co-polarized ratio (Wakabayashi et al., 2004; Nakamura et al., 2009a; Toyota et al., 2009) and the cross-polarized ratio (Kim et al., 2012) which are often used as an ice thickness proxy cannot be directly calculated from CP-mode SAR data. Although CP SAR images have been used to distinguish sea ice types (Dabboor and Geldsetzer, 2014; Charbonneau et al., 2010; Geldsetzer et al., 2015), to our knowledge there have been no published studies on its use for ice thickness detection in the open literature until now. Therefore, in this study, we considered the CTLR mode and developed an approach to directly retrieve the thickness from CP SAR data (hereafter we assume that the CP SAR is operated in CTLR mode). The paper is organized as follows: in Sect. 2 we introduce a new parameter to estimate ice thickness and demonstrate its sensitivity to different ice parameters by numerical modeling in Sect. 3. In Sect. 4, an empirical relationship based on a comparison of CP-SAR signatures with ice thickness data obtained from electromagnetic induction sounding is presented, and the retrieval performance of this algorithm is described. Further discussions and conclusions are presented in Sect. 5.
The full polarimetric radar scattering return can be represented by the
scattering matrix
We consider the CTLR mode for which the scattering vectors are given by (e.g., Nord et al., 2009)
According to the results obtained by the Cold Region Research and
Engineering Laboratory (CRREL'88), the typical ranges of root mean square (rms) height and
correlation lengths for smooth level sea ice are 0.02–0.143 and
0.669–1.77 cm respectively (Fung, 1994). For C-band
SAR, the small perturbation method (SPM) can be applied for explaining the
surface scattering characteristics from smooth level sea ice. By doing so,
the underlying assumption is that the received radar signatures are typical
for Bragg scattering. However, the SPM fails to describe cross-polarization
and de-polarization effects that are observed in real SAR data. In order to
overcome these limitations and to widen the SPM range of validity, an
extended Bragg model (termed X-Bragg model) was presented by Hajnsek et al. (2003).
In the X-Bragg model, the scattering surface is composed of rough
randomly tilted facets that are large with respect to the wavelength but
small with respect to the spatial resolution of the sensor (for RADARSAT-2
fine-quad mode, the wavelength is 5.6 cm and the resolution is 8 or 25 m).
Scattering from each rough facet is evaluated by employing the SPM, whereby
for the facets, a random tilt is assumed which causes both a random
variation
In order to improve the range of validity of the X-Bragg model, different
approaches (termed X-SPM model) were proposed by del Monaco et al. (2009)
and Iodice et al. (2011). In those studies, more realistic distributions of
Variations of the CP ratio as a function of the standard deviation of surface
slope
For ice thickness retrievals we propose to exploit the ratio between
CP ratio as a function of dielectric constants for different
We calculated the CP ratio as a function of the standard deviation of surface
slope
For our analysis we use the fact that a dry snow layer is transparent at C and L frequencies (meaning that our method is only applicable under freezing conditions), and we do not consider metamorphosis of the basal snow layer due to brine wicking effects or due to melt–freeze cycles. We focus on undeformed Arctic young and first-year ice for which volume scattering is low because of the relatively high ice salinity, which means that the ice surface is the dominant scattering source. Then the backscattering coefficients depend on the small-scale surface roughness and the dielectric constant of the ice surface. Desalination of the ice occurs parallel to its growth due to brine drainage (Kovacs, 1996). The desalination process causes a decrease of the dielectric constant. Hence the basic idea of our method for retrieving ice thickness is to relate changes of the dielectric constant to ice thickness growth. Because the CP ratio is sensitive to the variation of the dielectric constant, it is well-suited for detecting changes of the ice thickness of smooth first-year level ice.
In this section, we describe the combined use of an ice growth model and an electromagnetic scattering model for level sea ice to study sensitivities of the CP ratio to different ice and radar properties. We applied the scattering model proposed by Nghiem et al. (1995) to simulate the sea ice volume scattering and absorption by brine inclusions. The surface contribution was calculated with the polarimetric two-scale model (Iodice et al., 2011, 2013) and incoherently added to the volume term.
The sea ice scattering model configuration is presented in Fig. 3. Note that we do not explicitly include a snow layer (see also Sect. 2.3). The effects of a dry snow layer are that (1) the dielectric contrast between ice and snow is lower than between ice and air, hence the reflectivity of the ice surface is lower; (2) the radar wavelength in the snow is shorter than in air, hence the ice surface appears rougher to the radar; (3) the incidence angle gets steeper (depending on the dielectric constant of the snow), which (relatively) causes a stronger backscattering. Since we carry out simulations with different dielectric constants (by varying temperature and brine volume fraction), surface roughness parameters, and radar incidence angles, the results obtained without snow can be transferred to cases with dry snow layers.
In our model, the uppermost layer is air with permittivity
For ice growth simulations we use a 1-D thermodynamic model developed by
Maykut (1978, 1982) based on the energy balance equations at the
atmosphere–ocean boundary. The balance of the heat fluxes at the upper
surface of the ice can be expressed as
Structure and geometric model of the configuration of sea ice.
Equations and parameters used for the sea ice thermodynamic model.
Equations and parameters used for the sea ice properties.
Substituting the equations and parameters listed in Table 1 into Eq. (13)
and using the Newton–Raphson iteration method, the sea ice surface
temperature
To assess theoretical possibilities and limitations of ice-thickness
measurements by CP ratio, we simulated the evolution of ice growth for given
temperature and wind conditions based on the growth model described in
Sect. 3.1. The air temperature and wind speed were set to
The simulated sea ice growth process. Blue: sea ice thickness; red: sea ice surface temperature; green: the volume fraction of brine inclusions.
At this point we note that a systematic relationship between small-scale surface roughness and ice thickness has never been reported. Weathering effects, melt events, and snow metamorphism influence the millimeter-to-centimeter ice surface roughness to a highly variable extent, independent of ice thickness. As we will show below, the influence of the small-scale roughness on the CP ratio is moderate to low; hence the issue of varying small-scale surface roughness is not very critical.
Figure 4 illustrates the simulated sea ice thickness as a function of time and ice temperature, and the volume fraction of brine inclusions as functions of ice thickness. Figure 4 clearly shows that the volume fraction of brine inclusions reduces due to desalination processes as the ice thickness increases.
The relationship between the CP ratio and ice thickness at different incidence
angles for C-band radar (
To investigate the dependence of the CP ratio on the radar incidence angle and ice
thickness, the complex scattering coefficients (
Sensitivity of the CP ratio to the standard deviation of the surface
slope
Figures 6 and 7 indicate the roughness dependencies of the CP ratio. In Fig. 6 the
standard deviation of surface slope
Sensitivity of the CP ratio to the small-scale roughness (
Location of the study site in the Labrador Sea, with Pauli RGB
(HH
On 19–20 March 2011, a field program was conducted by the Department of
Fisheries and Oceans Canada (DFO) along the mid-Labrador coast (Fig. 8)
(Prinsenberg et al., 2012a). As part of the field survey, snow thickness and
ice thickness were measured with a helicopter-borne sensor package which
consists of a laser altimeter, an electromagnetic induction sounder (EMS),
and a ground-penetrating radar (GPR). The laser altimeter provides the
distance to the snow or ice surface, whereas the induction sounder measures
the distance from the sensor to the ice–water interface. Hence the
snow plus ice thickness can be obtained (Prinsenberg et al., 2012a, b).
Comparisons with drill hole data showed that the ice thickness values
derived from such soundings agree well within
All data are available on the website of DFO including pictures, notes, and
reports of the survey (
During the field survey, four C-band RADARSAT-2 quad-polarization images
were acquired nearly coincident with the DFO airborne survey flight lines
(Fig. 8). The RADARSAT-2 data were provided by the MacDonald, Dettwiler and
Associates Ltd (MDA). Important SAR parameters are listed in Table 3. For
our processing we used the RADARSAT-2 single-look slant range complex format
as starting point. A speckle reduction filter (13
Specifications of the qual-pol RADARSAT-2 SAR data.
Figure 8 presents the ice condition at the study site, flight paths and four
nearly coincident RADARSAT-2 fine quad-polarization images. Eight EMS
profiles were measured within the coverage of the four SAR images, and the
time differences between the SAR acquisitions and EMS flights are summarized
in Table 4. The images in Fig. 8 show the RADARSAT-2 data overlain with the
EMS flight tracks over the fast ice and drifting pack ice. According to the
ice charts, the total ice concentrations in fast ice and pack ice regions
are 10/10 and 9/10, respectively. The main ice type in land fast is
first-year ice of 70–120 cm in thickness, and the drift ice region contains
gray ice (10–15 cm thick), gray-white ice (15–30 cm), thin first-year ice
(30–70 cm), and again first-year ice, 70–120 cm thick. In the drifting ice
region several openings can be seen in the SAR images. The extent of
land-fast ice evolves in the offshore direction and can be visually
separated from the pack ice. Most of the rougher land-fast ice is brighter
in the SAR images than the thinner undeformed land-fast ice. According to
the meteorological data archive from Makkovik station
(
Specifications of helicopter-borne EMS ice thickness data sets.
Histogram of ice and snow thickness in the Labrador Sea.
A direct comparison between SAR imagery and flight profiles' data may cause
errors due to the time differences of the data acquisitions (the time
difference between SAR and flight data is shown in Table 4). In addition,
spatial differences may be caused by the different sampling and spatial
resolutions of the measurement instruments. The sampling rate for the EMS and
the laser is 10 Hz, which, given a typical helicopter survey speed of 80 mph,
corresponds to a spatial sampling interval of about 3–4 m. While the
footprint size of the laser is very small (several centimeters), the
footprint of the EMS is around 20 m at a typical operation height of 5–6 m.
For this experiment, the GPR was configured to a scan rate of approximately
30 scans per second. When flying at 60–80 knots, the ground sample spacing
is approximately one sample per 1.0–1.5 m. Moreover, according to the DFO
survey report, the floating ice drifted 1.4–1.8 knots towards the southeast, as
measured by ice beacons (Prinsenberg et al., 2012a). In order to mitigate
the errors caused by time and spatial resolution differences, we used
the following processing chain for linking SAR and airborne data.
The correction of the time difference was only implemented for the
drifting ice region. The boundary between fast ice and drifting pack ice was
taken from ice charts of the Canadian Ice Service (Fig. 8). Of the eight
EMS profiles, P1, P2, P5, and P7 are in or near the land-fast ice region,
whereas P3, P4, P6, and P8 are from the drift ice zone. With an ice drift
speed of 1.5 knots, and drift direction southeast taken from the DFO survey
report and considering the respective time differences, the profiles P3, P4,
P6, and P8 are shifted to their approximate positions at the acquisition
time of the SAR images. The shifted profiles are presented in Fig. 8 (dotted
line). It should be noted that 28 h passed between the acquisition times
of the P8 and SAR data, and the corrected location of P8 is beyond the
coverage of the SAR image. Hence P8 was discarded from further analysis. The EMS (ice plus snow) thickness values below 0.1 cm were removed to
consider the measurement accuracy of the EMS. Regions for which only EMS
data but no GPR data are available were also removed. Regions with GPR snow thickness values higher than 0.20 m were removed,
because snow layers thinner than 0.20 m are nearly transparent to C-band
radar waves, and the backscatter from the snow surface and volume can be
neglected (Hall et al., 2006). By combining the field survey data (ice charts and aerial photos), a
visual interpretation of RADARSAT-2 SAR was made, and regions of open water,
land, and deformed ice were masked in the SAR images. Land was identified
using the coastal line; open water areas were interpreted via backscattering
and texture. Deformed ice was brighter than level ice in single-polarization
SAR images, and revealed a higher entropy, which was extracted using
For ice zones of 50 m in length, averages of different parameters were
evaluated. Firstly, we used the The sea ice thickness was extracted from the averaged GPR snow depth and
EMS snow plus ice thickness values. Finally we calculated the CP ratio from Eq. (11) using the averaged complex
backscattering coefficients.
This processing chain ensures that only level ice is considered for which
the EMS system delivers reliable thickness data with an acceptable accuracy.
The total length of the profile segments that we used in this study amounts
to about 16 km (320 samples). Compared with the original data, almost 60 %
of the data were discarded in this processing chain (step 1: 17 %, step 2:
10 %, step 3: 23 %, step 4: 10 %).
To investigate the possibility of using the proposed polarimetric parameter
CP ratio to estimate sea ice thickness from SAR images, we plotted ice thickness
values obtained during the field campaign against the corresponding
values of the CP ratio derived from the RADARSAT-2 images in Fig. 10 (using all
320 samples). It can be seen that at C-band, the CP ratio shows a negative trend relative
to the ice thickness as the simulated results given in Sect. 3.2 predicted.
Figure 10 reveals that the highest sensitivity occurs between 0 and 0.5 m and
saturates with thickness values exceeding 1.5 m. As shown in Figs. 5 to 7,
the sensitivity should be smaller for ice thickness exceeding 0.4 m.
However, the slope change of the curves at 0.4 m is not as abrupt as in the
theoretical curves predicted in Sect. 3.2. This can be presumably explained
by the fact that we average over segments with different values of ice
roughness parameters
Regressions relating ice thickness to CP ratio at different incidence
angles. The solid lines represent the fits, dashed lines the 90 % confidence
intervals. The black, green and red colors are used for the incidence angles
of 29, 42 and 49
Since our data comprise different incidence angles (29,
42, and 49
We found that the level of the CP ratio increases as the incidence angle increases
at a given value of the sea ice thickness. This observation compares well
with the forward simulation studies as shown in Fig. 5. These high
correlations enable us to derive reliable thickness information for smooth
level ice from radar images, assuming winter conditions (dry snow, no brine
wicking). The ice thickness can be estimated using an exponential function,
which can be described as follows:
At the next stage, we focused on the RADARSAT-2 images no. 2 and no. 3
(which have the same incidence angle of 42
This paper provides a first analysis of sea ice thickness retrieval using
compact polarimetric SAR. We developed a new parameter that we call the
CP ratio to estimate the thickness of undeformed first-year level ice from C-band
radar images, under dry snow conditions (snow depth
Since the thickness of deformed ice can be underestimated by the EMS measurements by as much as 50 or 60 % in the worst cases, we could only study the case of level ice. The capability of CP SAR to retrieve the thickness of deformed ice, which reveals a larger variation of large-scale roughness with respect to the sensor resolution, needs to be further discussed and studied.
Although our tests are performed on a limited sample of images, our findings demonstrate that the C-band compact polarimetric SAR has a potential for sea ice thickness retrievals over level first-year ice covered by a thin dry snowpack. The issue of environmental factors affecting the retrieval accuracy, e.g., brine wicking in the snow, or snow layers with different dielectric properties, has to be investigated further in more detail. The several planned Earth-observing satellite missions supporting compact polarimetry (e.g., the RCM operated at C-band) will provide the wide swath coverage necessary for operational sea ice monitoring. Hence our approach potentially provides a new operational tool for sea ice thickness measurements with a large areal coverage. In this case, the resulting thickness products are also of interest for the development, improvement, and validation of forecast models for the prediction of ice conditions, or of interest for seasonal and climate simulations that consider Arctic and Antarctic ice conditions.
This study was supported by the National Nature Science Foundation of China under grant 41306193, and the R & D Special Foundation for Public Welfare Industry (201305025). This work was carried out as part of the Dragon-3 Programme (10501) by the Ministry of Science and Technology of the P. R. China and the European Space Agency. The authors would like to thank the Canadian Space Agency (CSA) and MDA for providing the RADARSAT-2 data, and we are very thankful to the Department of Fisheries and Oceans Canada for their support in providing valuable snow and sea ice field data. We gratefully acknowledge the detailed comments of Stefan Kern and two anonymous reviewers which helped to considerably improve the readability of the article. Edited by: C. Duguay