TCThe CryosphereTCThe Cryosphere1994-0424Copernicus PublicationsGöttingen, Germany10.5194/tc-10-401-2016Late-summer sea ice segmentation with multi-polarisation SAR features in C
and X bandForsAne S.ane.s.fors@uit.noBrekkeCamillaDoulgerisAnthony P.EltoftTorbjørnRennerAngelika H. H.https://orcid.org/0000-0002-9997-6366GerlandSebastianDepartment of Physics and Technology, University of Tromsø – the Arctic University of Norway, 9037 Tromsø, NorwayInstitute of Marine Research, 9294 Tromsø, NorwayNorwegian Polar Institute, FRAM Centre, 9296 Tromsø, NorwayAne S. Fors (ane.s.fors@uit.no)17February201610140141530June20151September201518December201518January2016This work is licensed under a Creative Commons Attribution 3.0 Unported License. To view a copy of this license, visit http://creativecommons.org/licenses/by/3.0/This article is available from https://tc.copernicus.org/articles/10/401/2016/tc-10-401-2016.htmlThe full text article is available as a PDF file from https://tc.copernicus.org/articles/10/401/2016/tc-10-401-2016.pdf
In this study, we investigate the potential of sea ice segmentation by C- and
X-band multi-polarisation synthetic aperture radar (SAR) features during late
summer. Five high-resolution satellite SAR scenes were recorded in the Fram
Strait covering iceberg-fast first-year and old sea ice during a week with
air temperatures varying around 0 ∘C. Sea ice thickness, surface
roughness and aerial photographs were collected during a helicopter flight at
the site. Six polarimetric SAR features were extracted for each of the
scenes. The ability of the individual SAR features to discriminate between
sea ice types and their temporal consistency were examined. All SAR features
were found to add value to sea ice type discrimination. Relative kurtosis,
geometric brightness, cross-polarisation ratio and co-polarisation
correlation angle were found to be temporally consistent in the investigated
period, while co-polarisation ratio and co-polarisation correlation magnitude
were found to be temporally inconsistent. An automatic feature-based
segmentation algorithm was tested both for a full SAR feature set and for a
reduced SAR feature set limited to temporally consistent features. In C band,
the algorithm produced a good late-summer sea ice segmentation, separating
the scenes into segments that could be associated with different sea ice
types in the next step. The X-band performance was slightly poorer. Excluding
temporally inconsistent SAR features improved the segmentation in one of the
X-band scenes.
Introduction
A decline in the Arctic sea ice extent has been observed during
the last decades, together with a large reduction in sea ice thickness and
sea ice volume . At the
same time, the melt season has lengthened at a rate of about 5 days per
decade since 1979 . To understand the processes governing
these changes, and to meet the needs of shipping, oil and gas industries in
an increasingly accessible Arctic, more detailed mapping and monitoring of
the summer sea ice cover are required .
Synthetic aperture radar (SAR) is widely used in operational sea ice
monitoring. The Canadian Ice Service alone processes 10 to 12 000
SAR images every year . Operating in the microwave frequency,
SAR has the advantage of providing all-weather and day-and-night imagery. At
present, operational sea ice services use single and dual polarimetric SAR
images (HH+HV or VH+VV) in sea ice
monitoring due to their wide swath widths and good temporal coverage.
However, on a local scale, more information and improved sea ice segmentation
can be retrieved from full polarimetric SAR imagery (HH+HV+VH+VV). Today, such data are in limited use mainly due to its
reduced coverage. The recent development of compact polarimetry could open
the way for more polarimetric radar information to be retrieved at larger
swath widths .
C band (5.4 GHz) is considered the preferred frequency in operational
sea ice satellite monitoring, offering good all-season capability
. With the launch of TerraSAR-X (2007) and COSMO-SkyMed
(2007), new opportunities to investigate the potential use of X band
(9,6 GHz) in sea ice satellite monitoring appeared. Several studies have investigated the
application of X-band radar for sea ice mapping through ground-based,
airborne and satellite-borne platforms. X band is found to have good
separation capabilities between first-year ice and old ice
, between water and sea ice and in
detection of thin ice . Results from the Baltic Sea
suggest that the information content in C and X band are largely equivalent
, while X band was found to add information
when used in combination with C band in the Arctic Ocean .
Several techniques for automatic segmentation of sea ice in SAR scenes exist.
Methods consist of gamma distribution mixture models, thresholding of
polarimetric features, k-means clustering, neural networks, Markov random
field models, Gaussian mixture models, Wishart classifiers and iterative
region growing using semantics seeand references therein.
Several of these methods are feature-based methods, making use of a feature
set in the segmentation. They have the advantage of being flexible as the
input features used can be varied with, e.g. location and seasonal
conditions, and the features offer possible post-segmentation information as
an interpretation and labelling source. showed promising
results in segmenting a full polarimetric sea ice scene taken under winter
conditions (low temperatures and snow cover) with a simple feature-based
multi-channel SAR segmentation method described in and
, utilising six polarimetric features derived from the
covariance matrix.
Research has been conducted on SAR and microwave scatterometer measurements
of sea ice since the early 1990s . Most of the conducted
studies have been in winter and late fall, and the number of studies in the
melt period is limited. In winter, differences in salinity content and degree
of deformation of sea ice make it possible to separate multiyear ice (MYI)
and different stages of first-year ice (FYI) from each other. During summer,
smaller differences in salinity between MYI and FYI and the presence of moist
snow on the sea ice surface make monitoring with SAR challenging. SAR is
sensitive to the large changes in relative permittivity connected to air
temperatures close to 0 ∘C and to
variation in moisture content in the sea ice volume caused by freeze and melt
cycles . Early studies on the use of SAR and microwave
scatterometer data for summer sea ice applications can be found e.g. in
, , ,
and . Newer studies include
examination of backscatter signatures of multiyear sea ice with ship-based
scatterometer and investigation of the use of
a supplementary frequency of either X or Ku band in addition to C band in
late-summer sea ice classification with an airborne scatterometer
. Satellite-based studies include separation of MYI and FYI
by dual polarisation intensity data from RADARSAT-2 ,
classification potential of polarimetric features from RADARSAT-2
and investigations of melt pond fraction retrieval from
co-polarisation ratio data acquired by RADARSAT-2
. Separating different sea ice types during
summer melt is still a challenge.
The objective of this study is to investigate the potential of sea ice
segmentation using C- and X-band multi-polarisation SAR features during late
summer. A data set consisting of five high-resolution C- and X-band scenes
recorded on iceberg-fast first-year and old ice in the Fram Strait in August
and September 2011 is employed in our study. The satellite data are combined
with airborne measurements from a helicopter flight at the site. We explore
how the features and feature-based automatic segmentation successfully
employed on FYI during winter conditions in perform on
late-summer sea ice with air temperatures close to 0 ∘C. Our study
consists of two parts. Firstly, the suitability of the individual features
for use in late-summer sea ice segmentation is evaluated. This is done by
investigating the ability of the individual features to discriminate between
sea ice types and their temporal consistency during changing temperature
conditions. A reduced set of the four most temporally consistent features is
suggested for use in segmentation. Secondly, a feature-based automatic
segmentation algorithm is tested on the data set. We investigate whether it
groups the scenes into reasonable segments, which are possible to associate
with distinct sea ice types. The algorithm is tested both with a full feature
set and with the reduced feature set suggested in the first part of the
study. The segmented images are evaluated both visually and by pixel-wise
evaluation of regions with known geophysical properties.
Methods
In this study, we examine the potential of six polarimetric SAR features for
use in late-summer sea ice segmentation. To simplify the study, five regions
of interest (ROIs) with different sea ice types were defined based on
information from the satellite scenes and the helicopter flight at the site.
The first part of this section describes the data set utilised in our study.
In the second part we explain the design of the study, including the choice
of ROIs, the generation of polarimetric SAR features and the methodology of
the analysis.
Map of the western Fram Strait showing the location of the satellite
scenes included in the study and the track of the helicopter flight
collecting airborne measurements for the study. The red box in the inset map
of the Northern Hemisphere displays the geographical position of the area
displayed. At the time of the flight, R/V Lance was slightly north
of this map section.
Overview of the data set.
DateTimeSceneSatellite, mode and polarisationIncidencePixel spacing(UTC)IDangle(azimuth × slant range)29 Aug 201117:41R1RADARSAT-2, Fine Quad, HH, HV, VH, VV38.2∘5.0m×5.0m30 Aug 201118:23T1TerraSAR-X, StripMap, HH, VV29.4∘2.4m×1.9m31 Aug 201118:23R2RADARSAT-2, Fine Quad, HH, HV, VH, VV48.2∘4.7m×5.1m3 Sep 201114:09–Helicopter flight––4 Sep 201118:07R3RADARSAT-2, Fine Quad, HH, HV, VH, VV44.4∘5.1m×6.8m5 Sep 201117:00T2TerraSAR-X, StripMap, VH, VV25.9∘2.3m×2.1mStudy site
Fram Strait is a dynamic region characterised by the outflow of sea ice from
the central Arctic Ocean e.g.. The sea ice
cover is therefore highly variable with both FYI and MYI and contains a
large fraction of deformed ice. In late summer, the snow cover has usually
melted completely, leading to melt ponds on top of the ice
e.g.. While in most parts of Fram Strait southward
drift leads to fast movement of the sea ice, a region with iceberg-fast ice
forms in some years in western Fram Strait . In this
region, the ice cover varies between rough ice due to deformation and very
level ice where the ice is formed during winter and protected from impact
(unpublished data; Beckers et al., 2015). The study site was situated in this
area (Fig. ). Both FYI and old sea ice in different stages of
development were represented at the site.
Data set
The data used in this study were collected from ship, helicopter and
satellite platforms during a coordinated campaign in Fram Strait in late
summer 2011. The data set consists of several high-resolution
multi-polarimetric SAR scenes, together with airborne observations collected
from a helicopter (Table ). In addition, meteorological
observations from the scientific vessel R/V Lance provided
information about the changing weather conditions during the campaign. The
area covered by the satellite scenes could not be reached by the ship, and
the helicopter did not have the opportunity to land within the area;
therefore no in situ measurements from the sea ice surface were retrieved.
Satellite measurements
For this study, three quad polarimetric C-band scenes from the Canadian
RADARSAT-2 (RS-2) satellite (denoted R1, R2 and R3) and two dual polarimetric
X-band scenes from the German TerraSAR-X (TS-X) satellite (denoted T1 and T2)
were used. More details about the scenes can be found in
Table , and the positions of the scenes are displayed in
Fig. . All scenes were acquired during ascending orbits. The
RS-2 scenes have a coverage of 25km (range)×25km
(azimuth), while the TS-X scenes have a coverage of 15km (range)×50km (azimuth).
Airborne measurements
Airborne measurements were conducted during a helicopter flight out from
R/V Lance within the period of the satellite campaign (see
Table ). They include sea ice thickness, relative
surface roughness and aerial images. The track of the flight is displayed
together with the location of the satellite scenes in Fig. .
Measurements of total snow plus sea ice thickness (from now on referred to as
sea ice thickness) were performed with an electromagnetic induction sounder
(EM-bird), which was towed underneath the helicopter and flown at a height of
about 15 m above the surface. More details about the EM-bird can be
found in and . From this
device, the difference in conductivity between sea ice and water is used to
find the height of the EM-bird above the ice/water interface, and a laser
altimeter integrated in the EM-bird detects the distance between the EM-bird
and the snow/ice surface. The difference between the two measures gives the
sea ice thickness. The footprint of the EM-bird has a diameter of about
50 m (depending on the height of the instrument). At the time of the
acquisition there was very little or no snow on top of the sea ice, confirmed
by the aerial photos and observations from scientists onboard the helicopter.
The data from the laser altimeter mounted on the EM-bird can be used to
extract surface roughness . Calibration is
needed to remove helicopter altitude variations. This was done by the
three-step high- and low-pass filtering method described in
. The resulting surface elevation profiles are relative to
the level ice. Surface roughness is in this study presented as the standard
deviation of the profile surface elevation about the mean (root mean square
height), Rq:
Rq=1N∑i=1N(yi-y‾)2,
where N represents the number of measurements, y‾ the mean
height above level ice and yi the height above level ice of sample i.
Each ROI profile is 400 m long, and N varies between 960 and 1067,
depending on the speed of the helicopter.
The helicopter was equipped with a digital camera (GoPro YHDC5170, focal
length 5 mm, view angle 127∘), taking downward looking
photographs of the sea ice surface. The area covered by each image was about
85m (length)×110m (width) and the sampling rate
was 0.5Hz. The images were processed with a semi-automatic
classification algorithm, separating classes of open water, submerged ice,
melt ponds, very thin ice and thicker ice, as described in .
In an accuracy assessment of the method performed in ,
76 % of the melt pond pixels were correctly classified. The melt pond
fraction, i.e. the percentage coverage of melt ponds retrieved from each
image, is used in our description of the sea ice types. No additional ground
information could be retrieved about the state of the melt ponds at the site
of the satellite scenes during the campaign; hence, we do not know whether
the melt ponds were open or refrozen at the time of the acquisitions.
According to the cruise report, open melt ponds were observed during the
first days of the cruise, but from 26 August a major part of the melt ponds
had started to freeze over. Melt pond measurements in open melt ponds at the
ice edge were, however, performed until 31 August.
Meteorological information
SAR scattering properties of sea ice are highly affected by temperature and
humidity, and meteorological information can therefore aid the interpretation
of SAR satellite scenes. Meteorological measurements were performed on
R/V Lance during the campaign (Fig. ). An automatic
weather station at R/V Lance consisting of an air temperature sensor
(3455), an air pressure sensor (2810) and a relative humidity sensor (3445),
all from Aanderaa (numbers refer to model), were recording meteorological
information during the campaign (Fig. ). The height of the
station was 22 m a.s.l. R/V Lance was sailing during this
period and its route was located in the Fram Strait within 100 km
west and north of the position of satellite scenes. During the week of data
collection, the weather conditions were varying and the temperature was
fluctuating around 0 ∘C. We have no recorded information about the
amount of precipitation during the campaign, but the cruise report describes
long periods with fog until 2 September. To investigate how the distance
between R/V Lance and the position of the satellite scenes
influenced the meteorological information, 2 m air temperature and
surface pressure were extracted from the European Centre for Medium-Range
Weather Forecasts (ECMWF) re-analysis ERA-interim;. The
parameters were extracted in 6 h increments for both the position of
R/V Lance and the satellite scenes (79.25∘ N,
14.25∘ W). There was good agreement between ERA-interim air
temperature and surface pressure at the two locations (Fig. ).
The re-analysis seems to overestimate the air temperature during the start of
the campaign.
Air temperature (a), air pressure (b) and relative
humidity (c) during the campaign. The grey vertical lines represent
the time of the acquisition of the satellite scenes.
Study design
In the following subsections, the design of our study is presented.
Position of regions of interest and helicopter thickness
measurements displayed on the RADARSAT-2 scene from 31 August 2011 (R2). The
polarimetric image is a Pauli composite, the intensity channel combinations
|HH-VV|, 2|HV| and |HH+VV|
are assigned to the red, green and blue (RGB) channels
respectively.
Example photos from the five regions of interest: (a) ROI1,
(b) ROI2, (c) ROI3, (d) ROI4 and
(e) ROI5. The photos are captured during the helicopter flight on
3 September 2011, and the EM-bird can be seen in the lower centre part of
each photo.
Histograms of sea ice thickness (m) measured during the helicopter
flight 3 September 2011 for each of the five regions of interest
(ROIs).
Regions of interest
The area covered by the satellite scenes consists of sea ice with different
geophysical properties. Some regions were homogeneous and some contained
mixtures of different sea ice types. To simplify our study we focus on five
different sea ice regions, representing the most typical sea ice types in the
scenes (Fig. ). The regions of interest were chosen to be as
homogeneous as possible and to represent five distinctly different sea ice
types. All ROIs are situated along the helicopter flight track and are
400m (along track)×200m (across track) in size.
The selection of the ROIs was performed manually, based on colour-coded
polarimetric images (Pauli and composite representations) of the satellite
scenes together with photos, sea ice thickness, surface roughness and melt
pond fraction retrieved from the helicopter overflight. Example photos from
each ROI are presented in Fig. and sea ice thickness
histograms for each ROI can be found in Fig. .
Table presents helicopter measurements for each ROI, including
mean and modal sea ice thickness, mean melt pond fraction, surface roughness
and sea ice class labels according to WMO sea ice nomenclature
. ROI1 represents an area with
level medium thick FYI, found in the upper left part of the scene in
Fig. . The sea ice in ROI1 was relatively smooth and had
a moderate melt pond fraction. ROI2 represents the area of level thin FYI
located in the middle of the scene. The sea ice in ROI2 was smooth with a
high melt pond fraction. ROI3 and ROI4 represent areas of weathered deformed
old ice, situated in the lower middle part of the scene. ROI3 represents
thinner ice with a higher melt pond fraction than ROI4. ROI5 represents
heavily deformed old ice, located in the lower part of the scene. Note that
other areas of deformed ice can be seen as light-coloured regions in the
right part of the scene possibly forming a shear ridge.
Detailed information about the regions of interest (ROIs) from
helicopter-borne measurements and the corresponding sea ice class labels (WMO
nomenclature).
Polarimetric featureDefinitionExtracted for sceneRelative kurtosisRK=1L1d(d+1)∑i=1Lsi∗TC-1si2All scenesGeometric brightnessB=det(C)dAll scenesCross-polarisation ratioRVH/VV=〈SVHSVH∗〉〈SVVSVV∗〉R1, R2, R3, T2Co-polarisation ratioRVV/HH=〈SVVSVV∗〉〈SHHSHH∗〉R1, R2, R3, T1Co-polarisation correlation magnitude|ρ|=〈SHHSVV∗〉〈SHHSHH∗〉〈SVVSVV∗〉R1, R2, R3, T1Co-polarisation correlation angle∠ρ=∠〈SHHSVV∗〉R1, R2, R3, T1Polarimetric SAR features
Polarimetric SAR features combine information from the channels of
a multi-polarisation SAR system, and they represent information about the
scattering properties of the surface. The features studied were previously
successfully used in segmentation of a wintertime sea ice scene
. An overview of the features and their definitions is
presented in Table . The features consist of relative
kurtosis (RK), geometric brightness (B), cross-polarisation ratio
(RVH/VV), co-polarisation ratio
(RVV/HH), co-polarisation correlation magnitude
(|ρ|) and co-polarisation correlation angle (∠ρ).
RVH/VV is used instead of
RHV/HH as T2 has the polarisation combination
VH/VV. By inspection, these two features show similar
values in our data set. ∠ρ is equivalent to the more frequently
used term co-polarisation phase difference (ϕHH-VV).
A full-polarimetric SAR system is transmitting and receiving both horizontal
(H) and vertical (V) polarised electromagnetic waves, resulting in d=4
possible polarimetric channels (SHH, SHV, SVH
and SVV). Assuming reciprocity (SHV=SVH), the
Lexicographic feature vector, s, is given by equation
s=SHH2SVHSVVT,
where T denotes transpose . The covariance matrix,
C, is defined as the mean outer product of the Lexicographic
feature vector
C=1L∑i=1Lsisi∗T,
where si is the single look complex vector corresponding to pixel
i, L is the number of scattering vectors in a local neighbourhood and
∗T denotes the Hermitian transpose . Hence,
C can be written as
C=〈SHHSHH∗〉〈SHHSVH∗〉〈SHHSVV∗〉〈SVHSHH∗〉〈SVHSVH∗〉〈SVHSVV∗〉〈SVVSHH∗〉〈SVVSVH∗〉〈SVVSVV∗〉,
where the 〈⋅〉 is the sample mean over L scattering
vectors and ∗ denotes the complex conjugate.
The TS-X scenes included in our study are dual-polarimetric. The covariance
matrix then reduces to a 2×2 matrix. This implies that the full
feature set of six features could not be achieved for these scenes since the
achievable feature set depends on the scenes' polarimetric channel
combination (see Table ). Note that RK and B in the TS-X
scenes are calculated from reduced covariance matrices, and should not be
directly compared to the similar RS-2 features.
RVH/VV, RVV/HH, |ρ| and
∠ρ are well-known polarimetric features in sea ice applications
, while RK and B have seen less attention in the
literature. RK is a measure of non-Gaussianity and is defined as Mardia's
multivariate kurtosis of a sample divided by the expected multivariate
kurtosis of a complex normal distribution (d(d+1))
. RK<1 points towards
a distribution with broader shoulders and lighter tails than for Gaussian
data, while RK>1 implies a sharp peak close to the mean and heavy
tails relative to Gaussian distribution . Large values of
RK are expected for deformed sea ice due to scattering from a few strong
reflections and for inhomogeneous areas due to differences in intensity
mixtures . B represents the intensity of the multichannel
radar backscatter. It is closely related to the more familiar feature SPAN,
i.e. trace (C), as they both represent the eigenvalues of the
covariance matrix. B is, however, more sensitive to the smaller
eigenvalues. RVH/VV is known as a measure of
depolarisation . In microwave scattering of sea ice,
depolarisation is expected related to multiple scattering within the sea ice
volume or to surface roughness .
RVV/HH is only dependent on the relative permittivity
for very smooth surfaces within the Bragg regime . For
rougher surfaces, the feature is expected to increase with incidence angle
and relative permittivity and decrease with increasing surface roughness
. With volume scattering,
RVV/HH (dB) tends toward 0 .
|ρ| is a measure of the proportion of polarised backscatter, reaching
unity when the co-polarisation channels are perfectly correlated
. The feature is expected to decrease with incidence
angle, at an increasing rate for high salinity ice . ∠ρ is the relative difference in phase between the
co-polarisation channels, describing the sea ice scattering history
. The feature depends on both the sea ice relative
permittivity and surface roughness.
Data analysis
In this study, the sea ice type discrimination ability is evaluated through a maximum
a posteriori (MAP) supervised classifier, using Bayes' decision rule
. The classifier assigns pixel x to class
ωj if
Pωj|x>Pωi|x∀j≠i,
where P(ωj|x) is the probability of class ωj given the
feature value x. The probability density functions (PDFs) are estimated
with a Parzen kernel density estimator, using a Gaussian kernel function
. The bandwidth used is a function of the number of
points in the sample and their distribution, as described in
. The pixels in the five ROIs are used as training areas,
and each of the satellite scenes is classified individually. As the ROIs
investigated are small, resulting in small sample sizes, leave-one-out cross
validation is used in training and testing the classifier. A 7×7
pixels neighbourhood, L=49, is used in the classification and a stepping
window with steps of 5×5 pixels is employed to reduce neighbourhood
overlap. The resulting classification accuracies obtained for each individual
feature are used to evaluate the discrimination abilities of the features in
each of the five scenes.
The temporal consistency of the individual features is studied qualitatively
for the three RS-2 scenes, by inspecting the mean ROI values of each feature.
We consider a feature temporally consistent if the ranking of the mean ROI
values of the feature is similar in all three scenes. For example, the ROI with the
highest mean value for a specific feature has the highest mean value of that
feature in all the three investigated scenes. Based on the result of temporal
consistency, a reduced feature set of four features is suggested.
A feature-based automatic segmentation algorithm is tested on the five scenes
in the data set. It is tested both with the original full feature set, and
with a reduced feature set excluding the most temporally inconsistent
features. The segmentation uses multivariate Gaussian mixture models to model
the features' PDF, and employs an expectation-maximisation algorithm. Markov
random fields are used for contextual smoothing. Further description of the
segmentation approach is given in and .
A 21×21 pixel neighbourhood, L=441, was used performing the
segmentation. The size of the neighbourhood does not take into account the
difference in resolution between the scenes but does assure an equal sample size
in the extraction of the features. The algorithm was set to segment the
scenes into six different segments. The number is chosen to allow for the
five sea ice types described by the ROIs, in addition to one extra segment to
allow for detection of other sea ice types and to assure some flexibility for
the algorithm. For easier comparison, the area used in the segmentation is
confined to the intersection of the individual scenes' geographical location
(see the pink patch in Fig. ). For each scene, the
segmentation's performance is evaluated visually on its ability to separate
the four main sea ice types represented in the ROIs (medium thick FYI, thin
FYI, old ice and old deformed ice) and based on its ability to discriminate
the pixels of the five ROIs into different segments.
Results
This section consists of three parts. The first two parts examine the
individual sea ice type discrimination ability and the temporal consistency
of six polarimetric SAR features. In the third part, an automatic
segmentation algorithm based on the investigated features is tested on the
data set. Results for C and X band are presented separately, as differences
in incidence angle, resolution and polarimetric channel combinations make
a direct comparison inappropriate (see Table ). The
features in C band are based on the full covariance matrix, while those in
X band are based on reduced covariance matrices as the TS-X scenes are dual
polarisation scenes (see Table ). Note that ROI5 is only
present in the RS-2 scenes.
Individual features discrimination ability
The polarimetric features' individual capacity of classifying the
investigated ROIs into separate classes is presented in
Tables and , for RS-2 and TS-X
respectively. The presented values represent the diagonal values of the
confusion matrices, i.e. the percentage of true classification. The best
result for each ROI is highlighted in bold. All pixels from the five ROIs
were included in the classification, and the experiment was performed
separately for each of the scenes included in the study. From the two tables
we note that none of the features individually were able to classify all the
five ROIs in a single scene with high accuracy. All features do however give
satisfying classification results for some of the sea ice types represented
by the ROIs, in some of the scenes. Hence, by combining the features, all
features could add value to a feature-based sea ice type segmentation
algorithm. The best feature for discriminating a given ROI varies from scene
to scene. In all scenes except T1, ROI4 seems to be the most challenging to
separate from the others. ROI4 consisted of old ice, as did ROI3. An overlap
between the PDFs of these two ROIs could be a reason for the poor
discrimination result.
Classification accuracy of individual polarimetric features in the
three RADARSAT-2 scenes derived from MAP classification. The best result for
each ROI in each scene are highlighted in bold.
SceneFeatureSea ice type classification IDaccuracy (%) ROI1ROI2ROI3ROI4ROI5R1RK52242064B169711078RVH/VV23038051RVV/HH74915040|ρ|04131546∠ρ03070041R2RK4128078B3163753223RVH/VV198701844RVV/HH07040026|ρ|5700034∠ρ511192744R3RK020604055B3845242654RVH/VV34024074RVV/HH61350410|ρ|31506258∠ρ14002351
Classification accuracy of individual polarimetric features in the
two TerraSAR-X scenes derived from MAP classification. The best result for
each ROI in each scene is highlighted in bold.
SceneFeatureSea ice type classification IDaccuracy (%) ROI1ROI2ROI3ROI4RK3532417B5402160T1RVV/HH54171619|ρ|5144019∠ρ59122218RK440326T2B41235910RVH/VV16611923
The result of the MAP classification for C and X band does not show large
differences. The best classification accuracies in the C-band scenes are
slightly higher than those in the X-band scenes, indicating a larger
discrimination potential in C band. This difference is not necessarily
a result of different frequencies. RK and B are calculated from a reduced
covariance matrix in the X-band scenes and therefore contain less
information. The lower incidence angles and higher resolution of the TS-X
scenes could also contribute to the observed differences.
Mean values of the features in the regions of interest in the three
RADARSAT-2 scenes (R1, R2 and R3). The error bars are 2 standard deviations
long.
Mean values of the features in the regions of interest in the two
TerraSAR-X scenes (T1 and T2). The error bars are 2 standard deviations
long.
Segmentations of the three RADARSAT-2 scenes (R1, R2 and R3) into
six segments. To the left: segmentation with full feature set. To the right:
segmentation with reduced feature set consisting of relative kurtosis,
geometric brightness, cross-polarisation ratio and co-polarisation
angle.
The segments assigned to the pixels in the five regions of
interest by the segmentation of the three RADARSAT-2 scenes (R1, R2
and R3). To the left: segmentation with full feature set. To the
right: segmentation with reduced feature set.
Segmentations of the two TerraSAR-X scenes (T1 and T2) into six
segments. To the left: segmentation with full achievable feature set. For T1
the feature set consists of relative kurtosis, geometric brightness,
co-polarisation ratio, co-polarisation correlation magnitude and
co-polarisation correlation angle. For T2 the feature set consists of
relative kurtosis, geometric brightness and cross-polarisation ratio. To the
right: segmentation of T1 with the reduced feature set consisting of relative
kurtosis, geometric brightness and co-polarisation correlation
angle.
The segments assigned to the pixels in the four regions of interest
by the segmentation of the two TerraSAR-X scenes (T1 and T2). To the left:
segmentation of T1 and T2 with full achievable feature set. To the right:
segmentation of T1 with reduced feature set.
Temporal consistency of features
The temporal evolution of the feature means from each ROI are displayed in
Figs. and for RS-2 and TS-X
respectively. The variances of the features within each ROI are displayed as
error bars equivalent to 2 standard deviations. Due to different
polarisation channel combinations (see Table ),
different features are displayed for T1 and T2 in
Fig. . This also limits a temporal investigation in
X band, and we will in the following focus on the results in C band.
As weather conditions and incidence angles are different for the RS-2 scenes
in the data set (see Table ), the mean ROI values of the
features are expected to vary between the scenes even when sea ice conditions
are the same or very similar. Hence, when searching for temporally
consistent features, we look at the evolution of the ranking of the mean ROI
values of each feature. For instance, studying RK in
Fig. , the mean value within each ROI varies between the
scenes. However, the relative relationship between the different mean values
is almost constant. The RK of ROI5 does for instance take values between 1.05
and 1.15, but the RK value is always highest in this ROI. The same
between-ROI consistency during the investigated period can also be found for
B, RVH/VV and ∠ρ
(Fig. ). The relative relationship of the mean ROI value
of RVV/HH and |ρ| changes from scene to scene,
and
hence no temporal consistency can be observed.
T2 shows similar relationships between the mean ROI values of the features as
the RS-2 scenes for all three features extracted
(Figs. , ). The same between-ROI
relationship cannot be found for T1. The error bars in the TS-X ROIs are in
general larger than in the RS-2 ROIs, which may indicate poorer
discrimination ability of the TS-X scenes.
A feature-based sea ice segmentation algorithm depends on features with good
discrimination ability and temporal consistency to give consistent results
during changing geophysical conditions. This is especially important in the
Arctic, as in situ information is often not available. Excluding temporally
inconsistent features could help achieve a temporally stable segmentation
during changing conditions. We therefore suggest a reduced feature set,
consisting of RK, B, RVH/VV and ∠ρ for late-summer sea
ice segmentation. A reduction of features in the feature set could of course
also imply loss of important information and hence degradation in the
segmentation performance. The following subsection will further explore the
use of a reduced feature set.
Segmentation
From Fig. , the segmentations of R1 and R2 look
reasonable compared to the information from the helicopter flight, both for
the full (right) and reduced (left) feature set. The different segments seem
to be associated with distinct sea ice types. One can recognize the thin FYI
ice area in the middle of the scenes (violet), the heavily deformed old ice
areas in the diagonal bottom-left part of the scenes (blue and turquoise)
and two different sea ice types north (medium thick FYI, orange) and south
(old ice, yellow) of the middle region. The segmentation of R3
(Fig. e) has a more granular appearance, and the areas
with medium thick FYI are confused with the areas consisting of old ice
(yellow, orange, grey). The differences between the segmentations with full
and reduced feature sets for the three RS-2 scenes are small, but the
segmentation of R3 becomes slightly noisier with the reduced feature set.
Figure displays which segments the pixels of each of the ROIs
were assigned to in all three RS-2 scenes, both for the full (left) and the
reduced (right) feature sets. In general, the segmentations with the full
feature set give good distinction between the different ROIs included in this
study. In particular, the thin FYI in ROI2 and the deformed old ice in ROI5
were separated with an accuracy above 71 % from the other ROIs in all
three scenes. In R1 and R2, the segmentation was not able to separate ROI3
and ROI4 clearly (Fig. a, c). These ROIs both contain old ice,
with different thicknesses and melt pond fractions; hence, the ice types in
the ROIs are quite similar. In R3, the medium thick FYI in ROI1 was segmented
to three different segments. Reducing the feature set by excluding the
temporally inconsistent features does not affect the results for R1 and R2
(Fig. b, d). In R3, it improves the separation of medium thick
FYI in ROI1 and reduces the discrimination between the thin FYI in ROI2 and
the old ice in ROI3 (Fig. f).
The segmentations of the two TS-X scenes, based on the achievable features
limited by their polarisation channels (see Table ), are
presented to the left in Fig. . In addition, T1 was
segmented with a reduced feature set presented to the right in the same
figure. The segmentation of T1 with a full achievable feature set gives a
poor and granular impression. The area of thin FYI in the middle of the scene
was not discriminated from the rest of the scene, and the deformed sea ice
areas in the bottom-left diagonal were not fully segmented (green). Also the
segmentation of T2 gives a slightly granular impression, but the areas of
thin FYI in the middle of the scene (violet), and the areas of deformed ice
in the bottom-left diagonal (blue and turquoise) were well segmented.
Reducing the feature set in the segmentation of T1 seems to improve the
segmentation of the area with thin first-year ice in the middle of the scene
(violet), but granular noise is still present.
Figure displays which segments the pixels in each of the
ROIs were assigned to in the segmentations of the two TS-X scenes. For T1,
both for the full achievable (left) and the reduced (right) feature set.
Figure a confirms the poor impression of the segmentation
of T1 with full achievable feature set, giving minimal discrimination between
the four ROIs. In the segmented image of T2, the thin FYI in ROI2 is
separated from the other ROIs, while the rest of the ROIs are confused.
Reducing the feature set in the segmentation of T1 (Fig. b)
does not improve the segmentation performance, even though the visual inspection
of Fig. b suggested a slight improvement for the
whole scene.
Discussion
Among the six investigated features, RVV/HH and
|ρ| were found to be temporally inconsistent during the study. The
temporal inconsistency could have several reasons. These features might have
a stronger sensitivity to sea ice relative permittivity than the others. As
stated in the introduction, relative permittivity will vary largely with
temperature during warm conditions , and small
temperature differences between the scenes could cause large differences in
relative permittivity. In Bragg scattering theory,
RVV/HH is only dependent on the relative permittivity
of the surface for smooth surfaces . Another possible reason
for the inconsistency of these two features is a stronger sensitivity to
changes in incidence angles than for the rest of the features. The incidence
angle of the three RS-2 scenes varies between 38 and 48∘ (see
Table ). |ρ| varies linearly with incidence angle,
according to Fig. , but the same dependency cannot be seen
for RVV/HH. did a study on feature
temporal consistency in C band between a winter and a spring scene on FYI
north of Canada. They found, similar to this study, that ∠ρ showed
high consistency during changing temperature conditions. In contradiction to
our findings, they also found RVV/HH to have high
temporal consistency. RK and B were not included in their study. Different
incidence angles, sea ice types, snow conditions and season may explain the
differences in results.
Choice of features and their temporal consistency is not the only factor
affecting the results of the segmentation algorithm. Differences in incidence
angle and resolution between the scenes, changing meteorological conditions
and choice of segmentation parameters do all affect the outcome of our study.
The incidence angles in our study vary between 26∘ (T2) and
48∘ (R2). As the backscatter signature from a sea ice surface depends
on incidence angle, this is expected to affect the results. Between the RS-2
scenes, the incidence angle variation is small with a 10∘ difference.
From Fig. , the influence of the changing incidence angle
is limited, except for |ρ|. The pronounced difference in incidence angle
between the RS-2 and TS-X scenes could contribute to the poorer performance
of the segmentation algorithm in X band, but a larger number of scenes with
overlapping incidence angle are needed to confirm this. To obtain equal
sample sizes in our study, the same neighbourhood size was used in filtering
all scenes even when the scene resolution differed. The scenes with highest
resolution would therefore have smaller spatial filter sizes. This difference
in scale possibly influences the signature of physical properties of the
surface, like surface roughness variation. Filter sizes adjusted to the
resolution were tested during our investigations, but this made little
difference to the results.
During the week of data collection, the air temperature was varying around
0 ∘C, introducing difficult conditions for sea ice information
retrieval from SAR. The distance between the meteorological measurements
retrieved from R/V Lance and the study site makes detailed analysis
of SAR weather dependence difficult. Some general events observed in the
meteorological data could, however, help explain our results. Both T1 and R2
were acquired during a period with air temperatures close to or above
0 ∘C, conditions that are on the limit of suitability for sea ice
type discrimination by SAR. As reported by , moisture in
the upper sea ice layer could mask out volume scattering and hence lower the
backscatter contrast between different sea ice types. The difficult
conditions could explain the poor segmentation performance of T1. However, R2
was acquired during similar meteorological conditions with good segmentation
results. Lower frequency, higher incidence angle and extra information
contained in the cross-polarisation channel (lacking for T1) could all have
contributed to a better segmentation of R2. The segmentation of R3 was poorer
than those of the two other RS-2 scenes. Prior to the acquisition of R3, a
drop in temperature and relative humidity could have caused rime on the sea
ice surface or draining and refreezing of freshwater
in the upper layers of the sea ice . Both processes could
cause a lower contrast between different sea ice types and hence hamper the
segmentation results. A refreeze of the sea ice could, however, also result in
the opposite; enhanced volume scattering could lead to increased sea ice type
discrimination.
Choice of sliding window size and number of segments are important for the
segmentation results. The use of window size of 21×21 pixels or
larger showed the best results in our data set. The size of the window was in
our case a trade-off between resolution details (small window) and
segmentation with little speckle and larger continuous regions (large
window). The choice of window size will also determine which kind of
information one can retrieve about the sea ice surface. If information about
small-scale structure like ridges, melt ponds and small leads is important,
this requires a small window. Larger window sizes could be more appropriate
to gain information about, for instance, sea ice age or type. Choice of
sensor restricts how high a resolution is possible to achieve, and high
resolution is at present coupled to small swath width. The number of segments
was set in advance, based on visual inspection of the scenes and information
retrieved from the helicopter-borne measurements. Choosing too few segments
could force different sea ice types into a common segment, while increasing
the number of segments could split an ice type into several segments.
Conclusions
We examined the potential of sea ice segmentation by C- and
X-band multi-polarisation SAR features during late summer in the Fram Strait.
Firstly, the individual features sea ice type discrimination ability and
their temporal consistency were investigated. Secondly, an automatic
feature-based segmentation was tested.
The ability of the individual features to discriminate five sea ice types
during changing temperature conditions was evaluated by a MAP supervised
classifier and by a qualitative study of the temporal consistency of the
features. The classification results revealed a potential in all individual
features for discriminating some of the sea ice types from each other, but
none of the individual features could separate the total set of sea ice types
in any of the scenes. Hence, a combination of the features has the potential
of segmenting the different sea ice types included in our study. Temporal
consistency was evaluated by studying the ability of the features to rank the
mean value of the five sea ice types in the same order throughout the three
RS-2 scenes. Relative kurtosis, geometric brightness, cross-polarisation
ratio and co-polarisation correlation angle were found to give good temporal
consistency during changing temperature conditions. These features were
suggested as a reduced feature set. Co-polarisation ratio and co-polarisation
correlation magnitude were found to be inconsistent through the period
investigated. Possible reasons for the two features' inconsistency could be a
higher sensitivity to changes in relative permittivity or incidence angles.
Our study demonstrates some of the difficulties of sea ice discrimination at
temperatures close to 0 ∘C and highlights that it is important to
cautiously select features for consistent sea ice monitoring during late
summer. However, our study shows as well that it is possible to retrieve
valuable information from multi-polarisation SAR imagery, even under these
difficult conditions.
An automatic feature-based segmentation algorithm was tested on the data set
and evaluated for its ability to discriminate the five investigated sea ice
types. The segmentation was tested for a full feature set of six features and
for a reduced feature set of the four features showing the best temporal
consistency. The segmentation in general performed well on the three RS-2
scenes. It showed good temporal consistency between the scenes, both for the
full and for the reduced feature set. However, reducing the feature set
slightly degraded the segmentation performance for one scene. The
segmentation succeeded in segmenting some of the sea ice types in one of the
two TS-X scenes. In the other scene the segmentation performed poorly. The
poor performance might be a result of air temperatures above 0 ∘C
combined with low incidence angle and polarimetric channel combination HH-VV.
Reducing the feature set introduced a slight improvement in this poorest
segmented scene. In total, the automatic feature-based segmentation algorithm
demonstrates a potential of sea ice type discrimination during late summer,
and our results indicate that an exclusion of temporally inconsistent
features could improve the segmentation results in some cases. To confirm
this, more scenes need to be investigated.
Both C- and X-band scenes were included in the study, but differences in
incidence angle, resolution and number of polarisation channels made a direct
comparison with respect to frequency inappropriate. One of the X-band scenes
showed promising results regarding sea ice type discrimination, close to
those achieved for the quad polarimetric RS-2 scenes, even though it was
a dual polarimetric scene. However, investigations of more scenes with
different incidence angle and polarisation combinations are necessary to
assess the potential of X band in sea ice discrimination.
Future studies should also focus on a better physical understanding of the
relation between SAR polarimetric features and geophysical properties. This
could improve the interpretation of the segmented sea ice scenes, and
possibly lead to an automatic labelling of the segments, a classification. The
suitability of other features in late-summer sea ice segmentation should also
be explored. Multi-polarisation SAR images offer good possibilities for sea
ice segmentation, but due to their limited swath width they are not suitable
for operational ice charting. The development of compact polarimetry modes on
new satellite missions, e.g. RISAT-1, PALSAR-2 and RADARSAT
Constellation Mission and the new wide quad polarimetric mode in RS-2, could
increase the amount of polarimetric information on larger swath widths, and
the possibilities of late-summer sea ice investigations in these modes should
be investigated.
Acknowledgements
The authors thank the captain, crew and scientists from the Norwegian Polar
Institute onboard R/V Lance in the Fram Strait 2011 for data
collection. Thanks also to Justin Beckers at the University of Alberta, Canada,
for preprocessing the laser altimeter measurements. RADARSAT-2 data are
provided by NSC/KSAT under the Norwegian–Canadian RADARSAT agreement 2011
and TerraSAR-X data are provided by InfoTerra. This project was supported
financially by the project “Sea Ice in the Arctic Ocean, Technology and
Systems of Agreements” (“Polhavet”, subproject “CASPER”) of the Fram
Centre and by the Centre for Ice, Climate and Ecosystems and the long-term
ocean and sea ice monitoring programme in the Fram Strait of the Norwegian
Polar Institute. This project was also supported financially by Regional
Differensiert Arbeidsgiveravgift (RDA) Troms County.
Edited by: C. Haas
References
Barber, D. G., Yackel, J. J., Wolf, R. L., and Lumsden, W.: Estimating the
thermodynamic state of snow covered sea ice using time series Synthetic
Aperture Radar (SAR) data, in: International Society of Offshore and Polar
Engineers Vol. III, The Eighth International Offshore and Polar Engineering
Conference, 24–29 May 1998, Montreal, Canada, 50–54, 1998.Beckers, J. F., Renner, A. H. H., Spreen, G., Gerland, S., and Haas, C.:
Sea-ice surface roughness estimates from airborne laser scanner and laser
altimeter observations in Fram Strait and north of Svalbard, Ann. Glaciol.,
56, 235–244, 10.3189/2015AoG69A717, 2015.
Bowman, A. W. and Azzalini, A.: Applied Smoothing Techniques for Data
Analysis, Oxford University Press, New York, 1997.
Brath, M., Kern, S., and Stammer, D.: Sea ice classification during freeze-up
conditions with multifrequency scatterometer data, IEEE T. Geosci. Remote,
51, 3336–3353, 2013.Carlstrom, A. and Ulander, L.: C-band backscatter signatures of old sea ice
in the central Arctic during freeze-up, IEEE T. Geosci. Remote, 31, 819–829,
10.1109/36.239904, 1993.Dabboor, M. and Geldsetzer, T.: Towards sea ice classification using
simulated RADARSAT Constellation Mission compact polarimetric SAR imagery,
Remote Sens. Environ., 140, 189–195, 10.1016/j.rse.2013.08.035, 2014.DeCarlo, L. T.: On the meaning and use of kurtosis, Psychol. Methods, 2,
292–307, 10.1037/1082-989X.2.3.292, 1997.Dee, D. P., Uppala, S. M., Simmons, A. J., et al.: The ERA-Interim
reanalysis: Configuration and performance of the data assimilation system,
Q. J. Roy. Meteor. Soc., 137, 553–597, 10.1002/qj.828, 2011.
Doulgeris, A. P.: A simple and extendable segmentation method for
multi-polarsation SAR images, in: Proc. of PolInSAR 2013, 6th International
Workshop on Science and Applications of SAR Polarimetry and Polarimetric
Interferometry, 28 January–1 February 2013, Frascati, Italy, ESA SP-713,
European Space Agency, Paris, 2013.Doulgeris, A. P. and Eltoft, T.: Scale mixture of Gaussian modelling of
polarimetric SAR data, EURASIP J. Adv. Sig. Pr., 2010, 1–13,
10.1155/2010/874592, 2010.
Drinkwater, M.: Airborne and satellite SAR investigations of sea–ice surface
characteristics, in: Oceanographic Applications of Remote Sensing, CRC Press,
Boca Raton, Florida, 339–357, 1995.Drinkwater, M., Kwok, R., Winebrenner, D. P., and Rignot, E.: Multifrequency
polarimetric synthetic aperture radar observations of sea ice, J. Geophys.
Res., 96, 20679–20698, 10.1029/91JC01915, 1991.Drinkwater, M., Kwok, R., Rignot, E., Israelsson, H., Onstott, R. G., and
Winebrenner, D. P.: Potential applications of polarimetry to the
classification of sea ice, in: Microwave Remote Sensing of Sea Ice, edited
by: Carsey, F. D., American Geophysical Union, Washington, DC, Geoph. Monog.
Series, 68, 419–430, 10.1029/GM068, 1992.
Eriksson, L. E. B., Borena, K., Dierking, W., Berg, A., Santoro, M.,
Pemberton, P., Lindh, H., and Karlson, B.: Evaluation of new spaceborne SAR
sensors for sea-ice monitoring in the Baltic Sea, Can. J. Remote Sens., 36,
S56–S73, 2010.
Fung, A. K.: Microwave Scattering and Emission Models and their Applications,
Artech House Inc., Norwood, MA, 1994.Gill, J. P. and Yackel, J. J.: Evaluation of C-band SAR polarimetric
parameters for discrimination of first-year sea ice types, Can. J. Remote
Sens., 38, 306–323, 10.5589/m12-025, 2012.Gill, J. P., Yackel, J. J., and Geldsetzer, T.: Analysis of consistency in
first-year sea ice classification potential of C-band SAR polarimetric
parameters, Can. J. Remote Sens., 39, 101–117, 10.5589/m13-016, 2013.
Gogineni, S. P., Moore, R. K., Grenfell, T. C., Barber, D. G., Digby, S., and
Drinkwater, M.: The effects of freeze-up and melt processes on microwave
signatures, in: Microwave Remote Sensing of Sea Ice, edited by: Carsey,
F. D., v American Geophysical Union, Washington, DC, 68, 329–341, 1992.Haas, C., Lobach, J., Hendricks, S., Rabenstein, L., and Pfaffling, A.:
Helicopter-borne measurements of sea ice thickness, using a small and
lightweight, digital EM system, J. Appl. Geophys., 67, 234–241,
10.1016/j.jappgeo.2008.05.005, 2009.Hajnsek, I., Pottier, E., and Cloude, S. R.: Inversion of surface parameters
from polarimetric SAR, IEEE T. Geosci. Remote, 41, 727–744,
10.1109/TGRS.2003.810702, 2003.Hibler, W. D.: Removal of aircraft altitude variation from laser profiles of
the arctic ice pack, J. Geophys. Res., 77, 7190–7195,
10.1029/JC077i036p07190, 1972.Hughes, N. E., Wilkinson, J. P., and Wadhams, P.: Multi-satellite sensor
analysis of fast-ice development in the Norske Øer Ice Barrier, northeast
Greenland, Ann. Glaciol., 52, 151–160, 2011,
10.3189/172756411795931633, 2011.
Isleifson, D., Langlois, A., Barber, D. G., and Shafai, L.: C-band
scatterometer measurements of multiyear sea ice before fall freeze-up in the
Canadian Arctic, IEEE T. Geosci. Remote, 47, 1651–1661, 2009.Jeffries, M. O., Schwartz, K., and Li, S.: Arctic summer sea-ice SAR
signatures, melt-season characteristics, and melt-pond fractions, Polar Rec.,
33, 101–112, 10.1017/S003224740001442X, 1997.Kwok, R.: Outflow of Arctic Ocean Sea Ice into the Greenland and Barents
Seas: 1979–2007, J. Climate, 22, 2438–2457, 10.1175/2008JCLI2819.1,
2009.Kwok, R., Cunningham, G. F., Wensnahan, M., Rigor, I., Zwally, H. J., and Yi,
D.: Thinning and volume loss of the Arctic Ocean sea ice cover: 2003–2008,
J. Geophys. Res., 114, C07005, 10.1029/2009JC005312, 2009.Laxon, S. W., Giles, K. A., Ridout, A. L., Wingham, D. J., Willatt, R.,
Cullen, R., Kwok, R., Schweiger, A., Zhang, J., Haas, C., Hendricks, S.,
Krishfield, R., Kurtz, N., Farrell, S., and Davidson, M.: CryoSat-2 estimates
of Arctic sea ice thickness and volume, Geophys. Res. Lett., 40, 732–737,
10.1002/grl.50193, 2013.
Lee, J. S. and Pottier, E.: Polarimetric radar imaging: from basics to
applications, CRC Press, Boca Raton, Florida, 2009.Mäkynen, M. and Hallikainen, M.: Investigation of C- and X-band
backscattering signatures of Baltic Sea ice, Int. J. Remote Sens., 25,
37–41, 10.1080/01431160310001647697, 2004.
Mardia, K. V.: Measure of multivariate skewness and kurtosis with
applications, Biometrica, 57, 519–530, 1970.
Matsuoka, T., Uratsuka, S., Satake, M., Kobayashi, T., Nadai, A., Umehara,
T., Maeno, H., Wakabayashi, H., Nakamura, K., and Nishio, F.: CRL/NASDA
airborne SAR (Pi-SAR) observations of sea ice in the Sea of Okhotsk, Ann.
Glaciol., 33, 115–119, 2001.Meier, W. N., Hovelsrud, G. K., van Oort, B. E., Key, J. R., Kovacs, K. M.,
Michel, C., Haas, C., Granskog, M. A., Gerland, S., Perovich, D. K.,
Makshtas, A., and Reist, J. D.: Arctic sea ice in transformation: a review of
recent observed changes and impacts on biology and human activity, Rev.
Geophys., 52, 185–217, 10.1002/2013RG000431, 2014.Moen, M.-A. N., Doulgeris, A. P., Anfinsen, S. N., Renner, A. H. H., Hughes,
N., Gerland, S., and Eltoft, T.: Comparison of feature based segmentation of
full polarimetric SAR satellite sea ice images with manually drawn ice
charts, The Cryosphere, 7, 1693–1705, 10.5194/tc-7-1693-2013, 2013Onstott, R. G.: SAR and scatterometer signatures of sea ice, in: Microwave
Remote Sensing of Sea Ice, edited by: Carsey, F., American Geophysical Union,
Washington, DC, 68, 73–104, 1992.
Parkinson, C. L. and Comiso, J. C.: On the 2012 record low Arctic sea ice
cover: combined impact of preconditioning and an August storm, Geophys. Res.
Lett., 40, 1356–1361, 10.1002/grl.50349, 2013.
Raney, R. K.: Hybrid-polarity SAR architecture, IEEE T. Geosci. Remote, 45,
3397–3404, 2007.Renner, A. H. H., Dumont, M., Beckers, J., Gerland, S., and Haas, C.:
Improved characterisation of sea ice using simultaneous aerial photography
and sea ice thickness measurements, Cold Reg. Sci. Technol., 92, 37–47,
10.1016/j.coldregions.2013.03.009, 2013.Renner, A. H. H., Gerland, S., Haas, C., Spreen, G., Beckers, J. F., Hansen,
E., Nicolaus, M., and Goodwin, H.: Evidence of Arctic sea ice thinning from
direct observations, Geophys. Res. Lett., 2012, 1–8,
10.1002/2014GL060369, 2014.Scharien, R. K., Geldsetzer, T., Barber, D. G., Yackel, J. J., and Langlois,
A.: Physical, dielectric, and C band microwave scattering properties of
first-year sea ice during advanced melt, J. Geophys. Res., 115, C12026,
10.1029/2010JC006257, 2010.Scharien, R. K., Yackel, J. J., Barber, D. G., Asplin, M., Gupta, M., and
Isleifson, D.: Geophysical controls on C band polarimetric backscatter from
melt pond covered Arctic first-year sea ice: Assessment using high-resolution
scatterometry, J. Geophys. Res., 117, C00G18, 10.1029/2011JC007353,
2012.Scharien, R. K., Hochheim, K., Landy, J., and Barber, D. G.: First-year sea
ice melt pond fraction estimation from dual-polarisation C-band SAR –
Part 2: Scaling in situ to Radarsat-2, The Cryosphere, 8, 2163–2176,
10.5194/tc-8-2163-2014, 2014.Stephenson, S. R., Smith, L. C., Brigham, L. W., and Agnew, J. A.: Projected
21st-century changes to Arctic marine access, Climatic Change, 118, 885–899,
10.1007/s10584-012-0685-0, 2013.Stroeve, J. C., Markus, T., Boisvert, L., Miller, J., and Barrett, A.:
Changes in Arctic melt season and implications for sea ice loss, Geophys.
Res. Lett., 41, 1216–1225, 10.1002/2013GL058951, 2014.
Theodoridis, S. and Koutroumbas, K.: Pattern Recognition, Academic Press,
London, 2009.Vant, M. R.: Dielectric properties of fresh and sea ice at 10 and 35 GHz,
J. Appl. Phys., 45, 4712, 10.1063/1.1663123, 1974.
von Saldern, C., Haas, C., and Dierking, W.: Parameterization of Arctic
sea-ice surface roughness for application in ice type classification, Ann.
Glaciol., 44, 224–230, 2006.Warner, K., Iacozza, J., Scharien, R., and Barber, D.: On the classification
of melt season first-year and multi-year sea ice in the Beaufort Sea using
Radarsat-2 data, Int. J. Remote Sens., 34, 3760–3774,
10.1080/01431161.2012.760855, 2013.
World Meteorological Organisation: Sea-ice nomenclature and interational
system of sea ice symbols, Tech. rep., World Meteorological Organisation,
Geneva, Switzerland, 1989.Yackel, J. J. and Barber, D. G.: Melt ponds on sea ice in the Canadian
Archipelago: 2. On the use of RADARSAT-1 synthetic aperture radar for
geophysical inversion, J. Geophys. Res., 105, 22061,
10.1029/2000JC900076, 2000.