Content uploaded by A. Malin Johansson
Author content
All content in this area was uploaded by A. Malin Johansson on Sep 21, 2021
Content may be subject to copyright.
Available via license: CC BY 4.0
Content may be subject to copyright.
4996 IEEE JOURNAL OF SELECTED TOPICS IN APPLIED EARTH OBSERVATIONS AND REMOTE SENSING, VOL. 13, 2020
Can Mineral Oil Slicks Be Distinguished From Newly
Formed Sea Ice Using Synthetic Aperture Radar?
A. Malin Johansson , Member, IEEE, Martine M. Espeseth , Member, IEEE, Camilla Brekke , Member, IEEE,
and Benjamin Holt , Member, IEEE
Abstract—In this feasibility study discriminating oil slicks and
newly formed sea ice using synthetic aperture radar (SAR) imagery
is investigated, using imagery from the L-band high-resolution
uninhabited aerial vehicle synthetic aperture radar (UAVSAR)
airborne and the satellite C-band RADARSAT-2 (RS-2) systems. To
determine the separability of these two varying but similar appear-
ing low backscatter ocean surfaces, multipolarization features are
utilized from both SAR datasets. The discrimination is evaluated
using the Kolmogorov–Smirnov separability test. All imagery was
obtained during several sea ice campaigns in the Arctic Ocean and
separate oil spill campaigns in Norway and the Gulf of Mexico,
with each campaign collecting in situ observations. We observe that
the polarization difference (VV-HH) reliably separates the mineral
oil slicks and newly formed sea ice areas using UAVSAR images,
due to the low noise floor and subsequent high signal-to-noise
ratio (SNR) radiometric performance of the airborne system. The
comparably higher noise floor and related lower SNR hampers
the separability in the RS-2 images. Simulated noise floors were
generated by adding white Gaussian noise to the UAVSAR data,
which show that discrimination between the two low backscatter
phenomena using multipolarization features is possible, provided
that both datasets are still well above the noise floor. The pixel
resolution has a limited effect on the separability. The results of
this study provide an approach to distinguish oil slicks from newly
formed sea ice, which might be of special interest should an oil spill
occur within the marginal ice zone.
Index Terms—Oil slick, newly formed sea ice, synthetic aperture
radar (SAR), polarimetry, uninhabited aerial vehicle synthetic
aperture radar (UAVSAR), RADARSAT-2, signal-to-noise ratio
(SNR), L-band, C-band.
I. INTRODUCTION
THE thinning of sea ice and the reduced sea ice cover in
the Arctic [1] over the last decade has led to an increase in
maritime traffic and petroleum-related activities, activities that
Manuscript received March 31, 2020; revised June 12, 2020 and August 12,
2020; accepted August 13, 2020. Date of publication August 18, 2020; date of
current version September 21, 2020. This work was supported in part by the
Research Council of Norway (RCN) through “Oil spill and newly formed sea
ice detection, characterization, and mapping in the Barents Sea using remote
sensing by SAR” (OIBSAR) (RCN project 280616) and in part by the “Centre
for Integrated Remote Sensing and Forecasting for Arctic Operations” (CIRFA)
(RCN project 237906), and in part by the Jet Propulsion Laboratory, California
Institute of Technology, under contract with the National Aeronautics and Space
Administration. (Corresponding author: A. Malin Johansson.)
A. Malin Johansson, Martine M. Espeseth, and Camilla Brekke are
with the Department of Physics and Technology, UiT The Arctic Univer-
sity of Norway, 9037 Tromsø, Norway (e-mail: malin.johansson@uit.no;
martine.espeseth@uit.no; camilla.brekke@uit.no).
Benjamin Holt is with the Jet Propulsion Laboratory, Califor-
nia Institute of Technology, Pasadena, CA 91125 USA (e-mail:
benjamin.m.holt@jpl.nasa.gov).
Digital Object Identifier 10.1109/JSTARS.2020.3017278
have associated potential risks of oil spills in this challenging en-
vironment [2]–[5]. There is a wide spread international concern
of a potential spill in the Arctic environment, where detection of
oil in ice-infested waters using synthetic aperture radar (SAR)
is difficult during conditions of both new ice formation and
ice concentrations greater than 40%, for smaller spills [6]. An
oil spill within or near newly formed sea ice has not yet been
known to occur nor captured on radar imagery. Therefore, in this
feasibility study we examine the situation in radar imagery when
sea ice appears most similar to an oil spill in an otherwise clean
ocean, using both fully polarimetric (FP) and multifrequency
data, in order to investigate the possibilities of discriminating
between new ice formation and oil spills using uniform datasets.
FP (HH/HV/VH/VV) satellite images have a small areal cover-
age and are not suitable for regular monitoring of the vast Arctic
region. However, these images can be used to investigatewhether
various sea ice types and ocean conditions, including oil slicks
and the similar-appearing sea ice can be discriminated using all
three operationally available dual-polarimetric modes provided
by presently operational satellite sensors such as Sentinel-1
(C-band), the RADARSAT Constellation Mission (C-band) and
ALOS-2 (L-band).
We examine FP SAR images from spaceborne C-band
RADARSAT-2 (RS-2) SAR images as well as airborne L-band
SAR images acquired onboard the National Aeronautics and
Space Administration (NASA) uninhabited aerial vehicle
synthetic aperture radar (UAVSAR) sensor, to identify
multipolarization features that can separate oil slicks from
newly formed sea ice types over a range of incidence angles
(IA). UAVSAR images were chosen due to the low instrument
noise floor as well as the high resolution, and the RS-2 data
based on the operationally used C-band frequency. The SAR
data were acquired during sea ice field campaigns in the Arctic
Ocean and oil spill field campaigns in the open ocean, i.e., the
discrimination is tested between scenes containing either oil
slicks or newly formed sea ice. In situ observations were also
obtained during the respective campaigns which can be used for
validation of both the sea ice and oil slicks properties. In Fig. 1,
two photographs depicting the oil slicks from the oil-on-water
(OOW) campaign in 2015 in the North Sea and in the Gulf
of Mexico Mississippi Canyon 20 block (MC-20) are shown
together with a photograph of newly formed sea ice observed
during the Beaufort-Chukchi Sea 2015 campaign.
The article starts with Section II covering a review of related
studies, followed by a description of the SAR data processing
This work is licensed under a Creative Commons Attribution 4.0 License. For more information, see https://creativecommons.org/licenses/by/4.0/
JOHANSSON et al.: CAN MINERAL OIL SLICKS BE DISTINGUISHED FROM NEWLY FORMED SEA ICE USING SAR? 4997
Fig. 1. (a) Closeup of an oil emulsion from the OOW2015 campaign on 10 June 2015, photo by Øyvind Breivik. (b) Oil slick observed in MC-20, photo by
Benjamin Holt. (c) Newly formed sea ice from the Beaufort–Chukchi Sea campaign on 5 October 2015, photo by Benjamin Holt.
and the multipolarization features. In Section III the SAR data
covering the OOW exercises and the sea ice campaigns are
introduced. The methods used here are described in Section IV.
Results and discussions are presented in Section V and finally
Section VI concludes this article.
II. BACKGROUND AND POLARIMETRY
Low backscatter ocean signatures related to marine slicks and
in young sea ice have been extensively studied for decades [7]–
[10]. However, studies on discriminating oil slicks from newly
formed sea ice using remote sensing instruments are sparse.
Most studies, e.g., [11]–[14] focus on oil injection underneath
the sea ice, investigations of the migration pattern of the oil
up through newly formed sea ice, how oil is altered in cold
temperatures and within sea ice and its subsequent impacts
on the dielectric constant, and when the oil can be detected
using various remote sensing instruments. One exception is [15]
where they investigated utilizing FP SAR data, in this case the
copolarization ratio and the phase difference for the two different
mediums, based on a simple theoretical model of the dielectric
constant.
With low backscatter we mean areas that relative to either a
wind-roughened ocean surface or older/thicker sea ice, have a
lower radar return, as well as being on the lower end of ocean-ice
radar signatures. With respect to low backscatter for sea ice, the
newly formed sea ice has been observed to be around 4–5 dB
lower than, e.g., first-year ice in both C- and L-band SAR [16].
Similar results have been found in many other studies, primarily
based on C-band SAR, e.g., [17], [18]. In [19] newly formed sea
ice was reported to have backscatter values of −25 dB.
The damping of the surface waves from mineral oil slicks
varies with type and thickness of oil (crude oil, water content in
emulsified oils), age of the slick (weathering and emulsification),
and of drift and mixing speed (wind and currents) see, e.g., [10],
[20], [21]. The low backscatter may also be a result of a reduction
of the complex dielectric constant () [22]. The values for the
mediums in L-band, are for clean water 58.26 −i41.48 at −1.8 °C
and a salinity of 34.42 [23], for oil slick 2.3-i0.01 [24], [25] and
for newly formed sea ice 4.23-i0.12 [26]. A study by [27] found
that both the real and the imaginary part of the dielectric constant
were affected by oil within sea ice (a reduction in permittivity
and reduced volume scattering), and were different from the
uncontaminated sea ice (higher permittivity values and higher
volume scattering). This finding indicates that this parameter
may be useful to identify oil within sea ice.
Significant effort has gone toward establishing oil slick de-
tection algorithms as well as developing discrimination tools
relying on SAR data, e.g., [28]. In these studies, the shape of
slicks is commonly used to separate the oil slicks from other
low backscatter phenomena, but should an oil slick occur, e.g.,
at the ice margin, the sea ice edges may influence the spatial
extent of the slick and, feathering as a descriptive feature may
not occur. Out of the confirmed oil slicks in the HELCOM report
[29] 76% were elongated, and thereby have a shape comparable
to leads [30], an area where new ice regularly forms. Sea ice
floes may also affect the shape of a slick and the oil may migrate
underneath the sea ice. Therefore, the study here focuses on the
backscatter and polarimetric parameter values and will not take
into account the shape of the oil slicks, nor will it consider oil
slicks under sea ice.
To establish multipolarization features that can be used for
the separation between newly formed sea ice and open water,we
use FP data that enables extraction of multipolarization features
using the full scattering matrix Sas below
S=SHH SVH
SHV SVV =|SHH|eiφHH |SVH|eiφVH
|SHV |eiφHV |SVV|eiφVV .(1)
The first letter in the subscript refers to the transmitted polar-
ization and the second to the received polarization. |·|and φ
denote the amplitudes and the phases of the measured complex
scattering coefficients. The images are first calibrated to sigma
nought, and thereafter multilooked to reduce speckle.
A set of six different multipolarization features was selected
for the separability analysis, listed in Table I. The selected
multipolarization features emphasize different aspects of the
backscatter information, such as the scattering processes and
textural variation, and they have a proven suitability in discrim-
inating oil slicks from surrounding open water (e.g., [22], [28],
[31]) or in discriminating newly formed sea ice from surrounding
thicker sea ice and open water (e.g., [16], [18], [32]). When
4998 IEEE JOURNAL OF SELECTED TOPICS IN APPLIED EARTH OBSERVATIONS AND REMOTE SENSING, VOL. 13, 2020
TAB L E I
MULTIPOLARIZATION FEATURES USED IN THIS STUDY
The · means that the data is averaged over a neighborhood of N pixels. We
have assumed reciprocity HV =VH. PD is defined on a linear scale and
may suppress nonpolarized scattering [38] and the other multipolarization
features are defined on a log-scale.
analyzing SETHI airborne L-band SAR and UAVSAR data
[28] found that out of 12 multipolarization features, σHV and
σVV were the most suitable parameters for oil slick detection
in open water, and the normalized radar cross section (NRCS)
data are therefore included in the analysis. As available satellite
data, such as from RS-2 and Sentinel-1 (S-1), provide the dual-
polarization mode of HH/HV or VV/VH the cross-polarization
ratio (γcr) is also investigated. For sea ice classification the
copolarization ratio (γco) is a useful parameter for separating
the thicker sea ice from thinner sea ice and open water areas,
e.g., [16], [19], [32]. However, γco increases with IA for both
the newly formed sea ice and mineral oil slicks [15], [18]. This IA
dependence leads us to also include the polarization difference
(PD) as it was shown to be less sensitive to IA variations
compared to γco, to be less sensitive to additive noise [33]–[35]
and less sensitive to the wind direction [34]. Initial studies using
features derived from the covariance and coherency matrices
were carried out, though these results were not as promising
compared to the parameters included here and were therefore
excluded from the study.
According to two-scale Bragg scattering models like tilted-
Bragg model [36] or X-Bragg model [37], the multipolarization
features above the line in Table I are dependent on both large
and small scale surface roughness and below the line on large-
scale roughness. All parameters are dependent on the IA and the
dielectric constant.
III. EXPERIMENTAL SETUP AND DATA COLLECTION
In this study we use 11 multilooked complex (MLC)
UAVSAR images, see Table II. The UAVSAR images were
collected during the OOW2015 exercise in the North Sea [39],
over a crude oil seep in the Gulf of Mexico in 2016, located
at the MC-20 site [40], and during a sea ice campaign in
Beaufort–Chukchi Seas in 2015 [41]. The images were chosen
to cover a range of different IAs, spanning the entire range
from 30 ° to 55 °, though not all images contained dark features
spanning the entire range. The upper restriction of 55owas set
due to the increased influence from cross-talk at the higher
IAs in spaceborne SAR systems, even though the UAVSAR
has a reliable calibration up to 58 degrees [42]. The MC-20
average slick size (approximately 14.9 km2[43]) and the newly
formed sea ice areas were larger than those observed during the
OOW2015 campaign, and therefore a larger number of images
were required from the latter campaign to fully cover the same
IA range. The UAVSAR images were not corrected for cross-talk
[34].
In addition, we use 12 single-look complex (SLC) RS-2
images, separated into six different pairs. The oil slick RS-2
images were collected during four OOW exercises (2011–2013)
and one was acquired near simultaneously with a UAVSAR
image over MC-20. RS-2 images with overlapping IAs covering
newly formed sea ice were selected from three different sea
ice campaigns around Svalbard, in 2015 and 2019 respectively,
for more details see [44] and the Research Council of Norway
project Nansen Legacy website1. The specifics for the UAVSAR
and RS-2 data are presented in Table III.
A. Oil Slick Data
The OOW exercises are yearly oil spill response exercises
conducted by the Norwegian Clean Seas Association for Op-
erating Companies (NOFO). The exercises took place in the
abandoned Frigg field in the North Sea (59 o59N, 2 o27E),
where controlled releases of mineral oil emulsions and plant
oil into the open sea were conducted to test different recovery
systems. The amount of oil released in each exercise was 20 m3
(2011), 17 m3and 10 m3(2012), 50 m3(2013) and 43 m3and
35 m3(2015). Thickness estimates of 0.1–1.5 mm are given for
the OOW 2011 [45]. The slicks were thin mineral oil emulsion
films with an average water content of 58% – 69%. For further
information about the specific mixtures and the exercises in
2011–2013, see, e.g., [45]–[47], and for the 2015 exercise see
[39].
The other oil slick dataset was collected in 2016 over a
persistent crude oil slick in MC-20. The crude oil slick undergoes
emulsification upon surfacing and rapidly spreads as patchy
areas of emulsified oil as well as most extensive sheen. Here
four images from low to medium wind speed conditions are used
[40]. In situ data indicate that the crude oil in MC-20 ranged in
thickness from sheen to a thickness of 0.02–2 mm [40], [48].
B. Newly Formed Sea Ice Data
The UAVSAR sea ice data were collected during the Sea
State and Boundary Layer field campaign which took place in
the Beaufort–Chukchi Seas in October–November 2015 [41].
The thin ice thickness was calculated from skin temperatures
measured by two KT-15 infrared thermometers, mounted on
the ship and directed toward the sea surface at a distance of
25 m from the side of the ship. Each thermometer viewed an
approximately 3-m wide area, with the two areas separated by 24
m. These ice thickness measurements have been validated with
other related measurements and are accurate up to a thickness
limit of 0.5 m [49]. The thickness measurements from October
6 are plotted on-top of a RS-2 ScanSAR Wide image in Fig. 2.
1Online. [Available]: https://arvenetternansen.com
JOHANSSON et al.: CAN MINERAL OIL SLICKS BE DISTINGUISHED FROM NEWLY FORMED SEA ICE USING SAR? 4999
TAB L E II
UAVSAR AND RS-2 IMAGE DATASETS
−indicates not applicable or not available information. For 17 November spatially and near temporally overlapping UAVSAR and RS-2
images were used over MC-20. RS-2 images are divided up into pairs (#), where each pair has similar IA, one containing oil slick and
one newly formed sea ice. As MC-20 is a continuous seep no age of the spill is given. Temporally overlapping weather observations were
measured onboard ships that participated in the OOW and sea ice exercises, though for 15 June 17:48 UTC 2012 wind speed observations
from the closest platform is used. For MC-20 data from buoy 42020 from National Oceanic and Atmospheric Administration (NOAA)
National Data Buoy Center are used. Differences in accuracy exist between the different campaigns, when possible one decimal is used.
TABLE III
SPECIFICS OF THE UAVSAR AND RS-2 DATA U S ED I N THIS STUDY.RANGE
(RG)AND AZIMUTH (AZ)VALUES ARE GIVEN AS NOMINAL VALUES
The mean thin ice thickness during the day of the UAVSAR
flights was 0.16 m and overall during the campaign the thinner
sea ice thickness averaged 0.20 m. The air temperature ranged
from −5oCto−10 oC from October 4 through October 6, the
date of the UAVSAR flight, with October 4 being the date when
the ship entered the ice field. The water temperature remained
below freezing −1.8 oC throughout the campaign.
The RS-2 FP images were acquired during three different sea
ice campaigns, Norwegian Young sea ICE cruise (N-ICE2015)
[44], Fram Strait (FS) observatory cruise 2015, and the fourth
Nansen Legacy campaign in 2019. The N-ICE2015 and Nansen
Legacy campaigns were located north of Svalbard and FS15
close to the east coast of Greenland. During N-ICE2015 the
newly formed sea ice observed in leads was in the form of
nilas, young grey ice and young white ice [50]. During the three
weeks of the FS15 campaign, sea ice had just started to form and
newly formed sea ice with a thickness up to 4 cm was observed,
primarily grease ice and nilas without frost flowers. During the
Fig. 2. Measurements of sea ice thickness were obtained from the Beaufort–
Chukchi Seas 2015 campaign, utilizing downward-looking radiometers mounted
on the ship as it traveled through the sea ice [49]. Sea ice thickness measurements
(dots, m) are shown for October 6, which are overlaid on a RS-2 ScanSAR Wide
HH-channel image taken on October 6, 2015 at 17:15 UTC. The greyscale on
the RS-2 image ranges from −10 to −25 dB. The pairs of colored lines indicate
the swath extent of the four separate UAVSAR acquisitions, and the arrows on
the lines the flight direction.
Nansen Legacy campaign north of Svalbard in December 2019
new ice regularly formed during the freezing conditions.
IV. METHOD
UAVSAR and RS-2 images, presented in Table II, containing
either mineral oil slicks or newly formed sea ice are used
5000 IEEE JOURNAL OF SELECTED TOPICS IN APPLIED EARTH OBSERVATIONS AND REMOTE SENSING, VOL. 13, 2020
to characterize the two types of low backscatter phenomena,
with the aim to establish multipolarization parameters that can
separate the two across a range of IAs. Six multipolarization
features and their response to oil slicks and newly formed sea ice
are investigated. A noise analysis is carried out to assess the SAR
data quality and to identify potential limitations with the data
for discrimination. The discrimination analysis is performed
using the Kolmogorov–Smirnov (K–S) test [51] and finally,
the importance of the signal-to-noise ratio (SNR) and the pixel
resolution for the separability are assessed.
A. Feature Extraction
The multipolarization features are calculated using the equa-
tions outlined in Table I. The UAVSAR images were provided as
multilook complex (MLC) data and were multilooked by 3×12
pixels, with an additional smoothing of 7×7pixels performed
to reduce radar speckle. The RS-2 SLC data were multilooked
by 3×8pixels.
Regions of interest (ROI) were manually identified and se-
lected within each of the images to be representative of the
mineral oil slicks and the newly formed sea ice areas, with
example shown in Fig. 3(a) and (c). The ROIs for the MC-20
study area correspond to the areas with the highest damping
ratios, shown in [52]. Also shown are the PD values for both
areas [Fig. 3(b) and (d)]. In the ROI selection the edges of
the respective low backscatter areas were avoided to reduce
contamination from the surrounding open water and thicker sea
ice. Structures such as ships and platforms were also avoided to
mitigate effects from side-lobes that might otherwise smear the
signal around the target. The UAVSAR images were collected
during three separate campaigns and the low backscatter phe-
nomenon covered a significant range of IAs. For this analysis,
all pixels within the UAVSAR ROIs were separated into oil
slick or newly formed sea ice pixels, and thereafter binned into
five different IA ranges, where each range corresponds to 5 o.
The limited IA range covered by each RS-2 image prevented
matching IA groupings. Instead, the six RS-2 image pairs, where
each pair contained one oil slick and one newly formed sea ice
image, were identified with overlapping IAs and all ROI pixels
within each image were used for the analysis. The ROIs in the
RS-2 images were equally sized for all pairs, though the shape
was constrained by the slick or newly formed sea ice geometry
and the 3×8pixel multilooking. The same number of pixels was
used for each pair, with >800 pixels for sets #1–4. A smaller
number of pixels (>500) was used for sets #5–6 due to the
smaller size of the oil slicks.
To quantify the heterogeneity within the different multipolar-
ization features the coefficient of variation (
CV ) was calculated
for all pixels within each IA range (UAVSAR) or image dataset
(RS-2).
CV is defined as
CV =σX/μX(2)
where σis the standard deviation, μis the mean value of X,
the polarimetric feature. A low
CV value indicates a largely
homogeneous sample and a high value indicates large variations
within the sample.
B. SAR Data Quality
The low radar return signal of the investigated mediums
implies that the SNR needs to be considered when assessing
the robustness of any multipolarization features. The NESZ is
affected by many factors, including the SAR frequency, the inci-
dence angle, and the azimuth resolution [53]. Here we observe a
higher NESZ for the RS-2 images than for the UAVSAR images
(Table III), controlled primarily by the two different altitudes
of the sensors, as well as the increased NESZ with increasing
frequency (see (5) in [53]). The higher NESZ for the RS-2
images and corresponding lower SNR has resulted in a lower
separability between the ice/oil signatures and the clean ocean
using the RS-2 data compared to the UAVSAR data. The effect
of SNR on oil slick detection is extensively discussed in, e.g.,
[28], [33], and [54]. The level above the noise floor needed for
the NRCS to safely contain a valid signal and be useful for po-
larimetric scattering analysis varies in the literature, e.g., at least
6 dB [22], 7–9 dB [34] up to 10 dB [19], [33]. When analyzing
the NRCS values for the respective channels increments of 2dB
above the NESZ are therefore used, as this enables us to capture
all three above mentioned ranges. The RS-2 NESZ values are
calculated using the beam specified metadata information as well
as the local IA [55]. The UAVSAR NESZ profile and further
information is found in [42]. As the backscatter signals decrease
with increasing IA, thereby reducing the SNR, the IA must also
be considered in the analysis.
When comparing discrimination between the oil slicks and the
newly formed sea ice and the potential influence of the SNR we
define, for either of the two media, the difference (Δ) between
the mean value, of the respective backscatter intensities and the
NESZ as
Δ=μX−NESZ. (3)
C. Separability Analysis
To investigate the separability between the mineral oil slicks
and the newly formed sea ice we use the two sample K–S tests
[51]. These tests are based on determining if two probability
distributions are from the same distribution or not, by estimating
the maximum difference between their cumulative distributions
[56]. The test enables comparison between parameters that may
not be normally distributed. The K–S test has values between 0
and 1 and here, similarly to [57] and [58], we define a K–S value
above 0.7 to mean some separability, a value above 0.8 indicates
reasonable separability and values above 0.9 represents good
separability.
D. Effect of Additive Noise and Pixel Resolution
This step is only performed on the UAVSAR data, due to
the sensors inherently good SNR and high pixel resolution. To
test the influence of additive noise on the separability measure,
we add simulated noise to the UAVSAR measurements, and
compare the simulated noise to the true noise floors reported for
RS-2 ScanSAR, S-1 Extra Wide (EW), and ALOS-2/PALSAR-2
Stripmap. The noise simulation is performed in the same manner
as described in [33], where complex system noise is added to
JOHANSSON et al.: CAN MINERAL OIL SLICKS BE DISTINGUISHED FROM NEWLY FORMED SEA ICE USING SAR? 5001
Fig. 3. Intensity (VV) data in dB for UAVSAR images depicting (a) oil slicks in open water within the MC-20 area in the Gulf of Mexico and (c) newly formed
sea ice among thicker sea ice in Beaufort Sea. The red areas indicate the ROIs used. The PDvalues for the respective UAVSAR images are shown in (b) for the
oil slicks and in (d) for the newly formed sea ice. The same ROIs are here indicated with black polygons. Note that the colorbar is non-linear to enable thesame
color scale for both images.
5002 IEEE JOURNAL OF SELECTED TOPICS IN APPLIED EARTH OBSERVATIONS AND REMOTE SENSING, VOL. 13, 2020
the normalized radar backscatter coefficients. Expressed as
M=S+N(4)
where Mis the measured radar cross section, Sis the scattering
matrix without noise (1), and N=[nHH,n
VH,n
HV ,n
VV]T
represents the complex additive noise. The additive noise is
assumed to be complex Gaussian white noise with a mean around
zero. From this procedure several new simulated noise floors or
NESZ are generated. The separability estimates are thereafter
carried out again using the range of new radar cross section
values, starting from the polarimetric feature extraction stage.
The importance of the pixel spacing is investigated by varying
the speckle filtering window size for the UAVSAR data. The
original MLC UAVSAR images have been multilooked by 3×
12 pixels, and here a running averaging filter was applied to the
images, ranging from a 1×1to a 20 ×20 pixel window size.
The averaging will reduce the standard deviation of the values
but not significantly affect the mean values. The new effective
pixel spacings are also compared with the RS-2 ScanSAR, S-
1 EW, and ALOS-2/PALSAR-2 Stripmap configurations. The
original MLC UAVSAR images have a pixel spacing equivalent
to the SLC RS-2 and ALOS-2 Fine Stripmap multipolarization
data.
V. R ESULTS AND DISCUSSION
In Fig. 3(a) and (b) we observe that there is a good oil–sea
contrast in both σVV and PD for the UAVSAR data. This high
oil–sea contrast for PD was also observed for three mineral
oil emulsions (40%,60%, and 80%volumetric oil fraction) in
[34], and moreover, PD has a reduced dependence on IA for
the oil slick compared to the open water. In Fig. 3(c) and (d), the
newly formed sea ice can be detected in the σVV data though
has a lower contrast to the surroundings in PD. Note that the
PD values for the sea ice areas are lower than for the oil slick
and open water areas, though for easier interpretation the same
scale-bar is used in both sets of images.
A. Overall Observations
In Fig. 4(a) and (b), the mean σHH and σVV backscatter
values for the respective ROIs and the dark slicks are plotted
against mean PD, using the UAVSAR data. A good discrim-
ination between the newly formed sea ice (blue dots) and the
oil slicks (red triangles and magenta squares) is provided by
PD. Moreover, the newly formed sea ice has overall lower
backscatter, and lower PD values compared to the mineral oil
slicks. Fig. 4(c) shows σVV plotted against γco, where a larger
variability in the newly formed sea ice γco values compared to
the PD values [Fig. 4(b)] can be observed. The corresponding
results for the σHH and γco combination are not shown but
provide a very similar outlook. The RS-2 data are shown in
Fig. 4(d), where mean σVV is shown against mean PD values.
A lower degree of separability can be observed compared to the
UAVSAR data, though note that all ROI’s from the different pairs
are included and IA is not accounted for. For the RS-2 ice–oil
pairs with the highest IAs the values approach the noise floor,
and hence a reduced discrimination between the σHH and σVV
may be expected. Skrunes et al. 2018 [34] found that the PD
values had a large decrease from 24° to 30°, though above this
upper range the IA dependency was limited. From this it appears
that PDversus σVV gives the overall cleanest separation for the
UAVSAR data. Similar trends can be observed for the RS-2 data
provided that the data are separated into distinct IA ranges.
There are many different environmental scenarios with re-
spect to varying winds and currents, temperature, ice/oil, and
water mixtures that need to be examined for the interdisciplinary
topic of separating oil slicks from newly formed sea ice. Some
of the scenarios are covered within this study, e.g., SAR images
were acquired under low to medium wind conditions. For the
RS-2 pairs care was taken to ensure similar wind conditions
for the different pairs and also similar incidence angle, though
small differences exist between the respective pairs. The air
and water temperature variations are larger and may affect the
dielectric constant values for the crude oil and oil–ice emulsions
[27]. Excellent work has been carried out by, e.g., [11]–[14] to
investigate the effect on oil spills underneath sea ice. Here we
choose to focus on separability between oil slicks and newly
formed sea ice, and assume to be located in the vicinity of each
other within the marginal ice zone but not yet mixed in. This
study is a feasibility study, since multipolarization SAR images
covering both oil slicks and sea ice to the authors knowledge
do not presently exist. Moreover, the work presented here re-
quires that the newly formed sea ice and oil slicks have been
detected and discriminated from their respective surroundings,
and thereafter the separability between the two is assessed. For
this we choose to include multipolarization parameters that [28]
found to be suitable for oil spill detection and discrimination,
and for sea ice discrimination [32]. Further investigations could
be carried out using spaceborne L-band SAR data from, e.g.,
the ALOS-2/PALSAR-2 sensor, as it has both a low NESZ as
well as high spatial resolution, though limited data are available
overlapping oil spills.
B. Coefficient of Variation
The textural heterogeneity is quantified using
CV . The cor-
relation between sea ice thickness, IA, and the relative contri-
butions of surface and volume scattering for different IAs, e.g.,
[18], [57], [59], means that we separate the
CV values for the
UAVSAR data into five IA ranges, presented in Table IV. The
NRCS values for the dark features have low variability with IA,
and γco,γcr , and PD have a higher dependency of IA. For all
parameters the oil slick values are more homogeneous than the
newly formed sea ice values for all IAs except the lowest one.
The mineral oil slick in the MC-20 area had less variation with
IA than the OOW2015 slicks. A potential reason for this is the
higher wind speeds observed during the OOW2015 campaign,
or the thinner slick may be more susceptible to breaking up.
Skrunes et. al. [34] observed the lowest
CV values for PD for
open water areas under OOW2015, and combined with the low
IA dependency above 30 ° indicated that PD was one of the best
operational multipolarization feature due to the stability under
varying imaging geometry.
JOHANSSON et al.: CAN MINERAL OIL SLICKS BE DISTINGUISHED FROM NEWLY FORMED SEA ICE USING SAR? 5003
Fig. 4. UAVSAR mean PD is plotted against (a) mean σHH and (b) mean σVV , and in (c) mean σVV vs. mean γco. In (d) RS-2 mean PD against mean
σHH. One marker indicates one individual ROIs, with no consideration for the IA bracket into which it falls.
TAB L E IV
CV V
ALUES FOR THE MULTIPOLARIZATION FEATURES ESTIMATED FROM THE
UAVSAR DATA
The oil slick values are highlighted in light cyan and the newly formed sea ice in white. The
data are split into IA ranges of 5 ° increments.
TAB L E V
CV V
ALUES FOR THE MULTIPOLARIZATION FEATURES ESTIMATED FROM THE
RS-2 DATA
The oil slick values are highlighted in light cyan and the sea ice in white.
The
CV values for the RS-2 data are shown in Table V.
The low values for all NRCS values in pairs #5–6 are likely
5004 IEEE JOURNAL OF SELECTED TOPICS IN APPLIED EARTH OBSERVATIONS AND REMOTE SENSING, VOL. 13, 2020
a consequence of the low SNR. Notably higher values than for
the other parameters were observed for PD. For pairs #3, 5
and 6 the
CV (PD) values for the newly formed sea ice are
higher than for the oil slicks. Common for these three sea ice
images is the recent drop in air and sea water temperature [60]
and the measured sea ice thickness of 2–4 cm. Whereas for #2
the thickness was estimated to between 10–20 cm and #1 and 4
lacks thickness measurements. Studies by, e.g., [19], [61], [62],
identified that when the sea ice thickness exceeds 2–4 cm, liquid
brine forms at the sea surface, thereby changing the dielectric
constant, and also changing the NRCS. The oil slicks studied in
pairs #2 and 4 are the youngest slicks (<9 h), and in pair #1
is the large continuous release studied in MC-20. The relatively
large variability may therefore be an effect of a more contained
oil slick and a larger volume. The
CV values here may have
been affected by the low SNR and variability within the studied
oil slicks and newly formed sea ice areas.
For the UAVSAR data we observe that
CV of multipolar-
ization features, e.g., PD, may be suitable for oil slick and
newly formed sea ice discrimination, provided that the IA is
>35o. Most operational SAR sensors provide data with IAs up
to 47 °–50 °, and hence such restrictions on the use of IAs are
certainly possible. The two different UAVSAR oil slick datasets
show similar results, this may be a result of the larger volumes
of oil released within the MC-20 block, possibly resulting in a
more coherent oil slick, or the different wind conditions under
which the two slick areas were observed. The RS-2 data show
higher
CV values for PDwhen the measured sea ice thickness
indicate values below 5 cm, though no consistent variability is
observed.
C. Noise Analysis
Poor SNR may reduce the characterization ability and cause
misinterpretations of the data [22], [28], [33]. To assess the data
quality, the mean backscatter intensity values are compared to
the NESZ at the same IA or range of IAs. In [19] and [33],
a limit of 10 dB above the NESZ was determined to be at a
level when polarimetric scattering properties of the target can
be reliably assessed, [34] suggests using 7–9 dB, while [22] sets
a value of 6 dB as being needed. In Fig. 5(a)–(c), the mean
and one standard deviation backscatter values for the combined
UAVSAR ROIs separated into 5° IA ranges are shown together
with the NESZ values and 2dB increments above the NESZ,
enabling comparison with the three suggested limits.
We observe that for the UAVSAR data the copolarization
channels always meets the >NESZ +10 dB criteria for both
the mineral oil slicks and the newly formed sea ice areas. The
HV-channel meets this criteria for all the OOW2015 images and
for the IA bracket of 30 °–35 ° for the Beaufort images and for the
40 °–45 ° IA for the MC-20 images. The OOW2015 campaign
has the highest wind speeds of the included data (see Table II),
and this is likely a reason for the reduced effect of the sensor
noise on this dataset. It may be possible to use the HV-channel
for IAs up to 50 ° as it fulfills the criteria of >NESZ +6dBfor
all used datasets, though care must be taken in the interpretation
of the results.
Fig. 5. UAVSAR backscatter σXX mean±std values plotted together with the
NESZ for (a) OOW2015, (b) MC-20, and (c) Beaufort Sea. The NESZ values are
illustrated using grey shading, where the darkest color represents values below
the NESZ and white >10 dB above the NESZ. The shading is given in 2dB
increments.
The noise analysis result for the RS-2 images are presented
in Fig. 6. Note that for pair #1 the oil slick data was acquired
in wide fine quad-polarimetric beam mode and the NESZ is
therefore significantly higher than for the equivalent fine quad-
polarimetric beam mode. For image pairs #1–2 both copolarized
channels fulfil the >NESZ +10 dB. When the mean plus one
standard deviation fulfils >NESZ +10 dB we define the SNR
as good, when the values are below NESZ +10 dB but above
NESZ we define the SNR as moderate and when the values are
below NESZ, as poor. The SNR values were good for the newly
formed sea ice in pairs #1–3 and moderate in #4–6, and for the
oil slick good in #1–2, moderate in pairs #3 and 6 and poor for
pairs #4 and 5. For the cross-polarized channel the values were
poor to moderate for all image pairs. The backscatter intensity
values for the oil slicks in pair #6 are relatively high despite
the high IA, this is possibly an effect of the age (25–29 h) of
the observed oil slicks that may therefore have undergone more
emulsification than the other slicks. In [35], they argue that the IA
should, if possible, be kept below 35◦to ensure a sufficient SNR
JOHANSSON et al.: CAN MINERAL OIL SLICKS BE DISTINGUISHED FROM NEWLY FORMED SEA ICE USING SAR? 5005
Fig. 6. RS-2 backscatter σXX mean±std values plotted together with the
NESZ for (a) oil slicks and (b) newly formed sea ice. The NESZ values are
illustrated using grey shading, where the darkest color represents values below
the NESZ and white >10 dB above the NESZ. The shading is given in 2dB
increments.
for newly formed sea ice areas, and the results here corroborate
those findings.
From this section, we conclude that the UAVSAR data have
sufficiently good SNR for the copolarization channels (due to
the lower NESZ), though care must be taken when interpreting
the result for the RS-2 images when the IA is >30o. For both the
C- and the L-band data used here part of the cross-polarization
data lie below the recommended limit of 10dB set by [19], and
[33]. The low SNR for the cross-polarized data in RS-2 limits
the usability of the σHV and γcr features. A detailed scattering
analysis is not the objective of this study, and a lower SNR may
be acceptable. However, the proximity to the noise floor and
possible effects on the results should always be kept in mind
when analyzing this type of data.
D. Kolmogorov–Smirnov Test: Discrimination of Oil and
New Ice
To investigate separability between the mineral oil slicks and
the newly formed sea ice, the K–S test for each IA range in
the UAVSAR data is calculated (Table VI). For the MC-20 and
Beaufort datasets the best separability was found using the PD
and γco values. For both combinations of mineral oil slicks
and newly formed sea ice, UAVSAR data PD provides good
separability with only one value (0.77) below 0.9. It should be
noted that the Beaufort and MC-20 data had similar wind speeds
and that during the OOW2015 campaign the wind speeds were
higher. The two highest IA bands (45 °–50 ° and 50 °–55 °) for the
TAB L E VI
K–S TEST BETWEEN THE MINERAL OIL SLICKS AND THE CORRESPONDING
NEWLY FORMED SEA ICE FOR THE UAVSAR DATA
We adopt the separability measures from [57], [58] where a K–S valueabove
0.7 means some separability (light green), a value above 0.8 indicates reason-
able separability (mid green), and values above 0.9 have good separability
(dark green).
TAB L E VII
K–S TEST BETWEEN THE MINERAL OIL SLICKS AND THE CORRESPONDING
NEWLY FORMED SEA ICE FOR THE RS-2 DATA
IA is the mean IA. We adopt the separability measures from [57], [58] where a K–S
value above 0.7 mean some separability (light green), a value above 0.8 indicate
reasonable separability (mid-green) and values above 0.9 have good separability
(dark green).
Beaufort and MC-20 data are dominated by data with slightly
higher wind speeds, approximately 6 m/s compared to 4 m/s.
In these two cases we also observe lower K–S values for the
single channel-based parameters. σVV and σHV overall provide
some (>0.7) to good (>0.8) separability for the Beaufort and
OOW2015 data. For both datasets σHH and γcr was found to be
less suitable for discrimination between mineral oil slicks and
newly formed sea ice, which is similar to findings in [28]. σHH
had better overall separability in the OOW2015 and Beaufort
dataset compared to the MC-20 and Beaufort dataset. The lower
separability in γcr is important as most operational SAR sensors
acquire wide-swath coverage images in the HH/HV mode in
the Arctic Ocean context, and based on the findings here these
images may not be suitable for discrimination between oil slicks
and newly formed sea ice considering only the backscatter
intensity values. For a discrimination using these type of images,
such as S-1, inclusion of textural and spatial information is
necessary.
The RS-2 data K–S distance are presented in Table VII. It
can be observed that for some of the multipolarization features
the K–S values are >0.7 for certain IAs, though no individual
parameter was found to work for all datasets. The difference
in separability compared with the UAVSAR data might be
explained by the difference in SNR. In Fig. 7(a) and (b) the
K–S values for the different image pairs are plotted against Δ
(3) for the two different mediums, and in Fig. 7(c) the differences
5006 IEEE JOURNAL OF SELECTED TOPICS IN APPLIED EARTH OBSERVATIONS AND REMOTE SENSING, VOL. 13, 2020
Fig. 7. RS-2 data K–S test values for σHH (red squares), σHV (cyan diamonds) and σVV (blue triangles) plotted versus Δin dB for oil slicks (a), newly
formed sea ice (b) and the difference in Δfor the newly formed sea ice and oil slicks (c). The vertical lines represent 0, 6, and 10 dB above the NESZ, respectively.
in Δbetween the two different mediums are plotted against the
K–S values. In the cases with the highest separability the Δice
values fulfils the criteria of at least NESZ +6 dB. This may
indicate a discrimination between a signal in the newly formed
sea ice data and the higher noise component in the oil slick data.
For the HV-data the Δoil values indicate that the oil slicks are
close to or below the NESZ. Even though the Δice values do not
always have values of at least NESZ +6 dB, they are higher than
the Δoil values, again likely indicating a separation between a
higher signal component due to a higher noise component.
PD can be used to reliably separate the investigated mineral
oil slicks and newly formed sea ice using UAVSAR images,
though for the RS-2 images the discrimination is hampered by
the low SNR. PD has a high correlation with γco though the
advantages with the reduced imaging geometry dependency of
PD, the lower false alarm detection rate for oil slicks [28], the
reduced dependency on additive noise [33], and the reduced
influence of the wind to sensor direction [34] means that PD
is preferable for discrimination between mineral oil slicks and
newly formed sea ice for both datasets used here. However, the
lower contrast between newly formed sea ice and thicker sea ice
may limit the possibility of using PD for their discrimination,
but γco has been proven to be useful for separating thinner sea
ice from thicker sea ice [16], [18], [19], [32], [63]–[65]. Using
a linear mixture model, the authors in [15] demonstrated that
for L-band SAR the γco values for young ice are closer to the
values for oil–ice emulsions than to open water values, and that
newly formed sea ice have γco values higher than crude oil
and oil–ice emulsions at −20 °C. This was partially attributed
to the higher dielectric constant for the newly formed sea ice.
Using five complex permittivity mixture models [27] found that
the temperature was very important for the dielectric constant
values for crude oil and oil–ice emulsions and at −1.8 °C the real
permittivity values were similar to those for the newly formed
sea ice. The dielectric loss factor was on the order one magnitude
smaller than for the newly formed sea ice in [27] though one
order of magnitude larger than modeled in [15].
Following the findings in this study, we suggest that a fully
automatic detection and discrimination algorithm should first
identify low σHH and σVV values, and thereafter do an assess-
ment of the PD and γco values, similar to the discrimination
presented in [32] where two thresholds were set for γco to
separate the thin ice from open water (<2.8 dB) and from thicker
sea ice types (>1.3 dB).
E. Additive Noise
The overall good separability using PDin the UAVSAR data
meant that we select this parameter for further studies regarding
the effect of additive noise to the NRCS values. In Fig. 8 the
K–S test versus the simulated noise floors are shown for the
Beaufort and MC-20 data. The results are comparable for the
discrimination between the Beaufort and the OOW2015 data
(not shown). The NESZ values for RS-2 multipolarization data,
RS-2 dual-polarimetric data, and for S-1 EW are indicated with
vertical lines, even though these sensors are C-band sensors,
as they are commonly used for operational ocean surveillance.
The NESZ values for the L-band ALOS-2/PALSAR-2 Stripmap
high resolution mode are indicated for the HH-channel. When
the simulated additive noise is approximately at the same level
as the NESZ for S-1 EW the K–S discrimination is starting to be
negatively affected by the additive noise. From this we conclude
that significant noise levels can be added to the UAVSAR data
before the discrimination between the newly formed sea ice and
the oil slick is compromised. To ensure good separability and
low false alarm rates in the detection of dark features the NESZ
should if possible be below −24 dB. The backscatter values for
the oil slicks and the newly formed sea ice will be affected by
the frequency, and hence the sensitivity to the additive noise
should be investigated for the different frequencies, particularly
the commonly used C-band. Ideally, such an investigation should
therefore be carried out using a multifrequency platform.
In addition to RS-2 and Sentinel-1, other current SAR satel-
lites that are regularly used for ocean and sea ice surveillance
JOHANSSON et al.: CAN MINERAL OIL SLICKS BE DISTINGUISHED FROM NEWLY FORMED SEA ICE USING SAR? 5007
Fig. 8. K–S test for the Beaufort (ice) and MC-20 (oil) backscatter values with additive noise. The different colors represent different IA ranges. The x-axis shows
the simulated noise floor levels that are derived from the UAVSAR measurements after Gaussian white noise is added to the measured signal. The original NESZ
levels for the multipolarization and dual-polarimetric RS-2 modes, S-1 EW, and ALOS-2/PALSAR-2 Stripmap are indicated with vertical lines.
Fig. 9.
CV values for PD for newly formed sea ice Beaufort (ice, solid lines) and MC-20 (oil, dashed lines) data with different sized averaging window. The
different colored lines represent different smoothing window size. The x-axis show the different IA ranges.
include TerraSAR-X (NESZ =−17 dB [66]), ALOS-2 (NESZ
for HH and VV =−36 dB, NESZ for HV =−46.0 dB
[67]), and RADARSAT Constellation Mission (NESZ =−19 to
−25 dB). The PALSAR-2 sensor onboard the ALOS-2 satellite
has already been shown to have good potential for improved sea
ice characteristics [68], [69]. Upcoming missions also include
the NASA-ISRO SAR (NISAR) mission that will have sea ice
coverage but also the capability to cover oil slicks, and offers
both L- and S-band frequencies with NESZ of −25 dB [70].
Comparing the simulated additive noise level with the mean
copolarization backscatter values, shown in Fig. 5, we observe
that the mean backscatter values for the OOW2015 and MC-20
datasets are between −30 and −10 dB, whereas the Beaufort
dataset has backscatter values between −18 and −33 dB. This
seems to indicate that a good discrimination can be achieved as
long as one of the two investigated media has backscatter values
above the noise floor, whereas when both media have values
approaching the noise floor discrimination may be impaired.
This is similar to the results for the RS-2 data shown in Fig. 7.
The required backscatter values to separate low backscatter
signatures including oil spills and newly formed sea ice (6–10 dB
[19], [22], [33]) will vary in each of these systems, depend-
ing on the NESZ, available polarizations, and incidence angle
examined.
F. Averaging Filter Size
The K–S test show a slight improvement for PD when the
smoothing window size increases, though separability between
0.7–0.9 was found for all window sizes. The
CV (PD) values
5008 IEEE JOURNAL OF SELECTED TOPICS IN APPLIED EARTH OBSERVATIONS AND REMOTE SENSING, VOL. 13, 2020
in Fig. 9 indicate that the newly formed sea ice values remain
similar once the window size is 5×5or larger, whereas the
oil slick values are stable regardless of window size. The pixel
spacing corresponding to those of the RS-2 dual-polarimetric
mode (6×6), S-1 EW mode (4×4), and ALOS-2 ScanSAR
mode (5×5) and should, therefore, enable good separability
provided sufficient SNR.
From this we conclude that for discrimination between oil
slicks and newly formed sea ice, SNR is the most important
parameter, and that for our dataset a pixel size similar to the one
provided by S-1 EW did not hamper discrimination.
VI. CONCLUSION
We observe that PD can be used to reliably separate the
investigated mineral oil slicks and newly formed sea ice using
UAVSAR images, though low SNR hampers the separability in
the RS-2 images used. The recently launched RADARSAT Con-
stellation Mission (RCM) provides low resolution, high areal
coverage acquisitions using both the copolarization channels as
well as a dual polarization option, i.e., HH/HV, VV/VH, and
HH/VV. Moreover, it has a compact polarimetry SAR mode
where the 2nd component of the Stokes vector is similar to
PDassuming reflection symmetry and reciprocity. The HH/VV
ScanSAR option might be suitable for large-scale discrimina-
tion, and the compact polarimetric SAR mode is very promising
for the detection and separation case of newly formed sea ice
and oil slicks.
The high oil–sea contrast for PD in both the UAVSAR and
RS-2 data indicates that it can be used both to detect oil slicks
and separate them from the surroundings. However the lower
newly formed sea ice and thicker sea ice contrast may limit the
possibility of using PDboth for detection and discrimination of
oil slicks and newly formed sea ice. PD has a high correlation
with γco though the advantages with the reduced imaging geom-
etry dependency and the reduced dependency on additive noise
means that PD is preferable for the discrimination. Simulated
noise floors were generated by adding white Gaussian noise to
the UAVSAR data, with the results indicating that discrimination
between the two low backscatter phenomena was possible until
the new noise floor reached −24 dB. The pixel resolution had
limited effect on the separability.
ACKNOWLEDGMENT
The authors would like to thank NOFO for including our
experiment in their exercise in 2015 and for providing ground
truth data in 2011–2013 and 2015, scientists at the Norwegian
Meteorological Institute for collecting metocean data, and KSAT
for providing detection reports. Observations were provided for
N-ICE2015 by the Norwegian Polar Institutes Centre for Ice,
Climate and Ecosystems (ICE), FS15 by the Fram Strait Arctic
Ocean Observatory (http://www.npolar.no/framstrait) and the
Nansen Legacy cruise by the Nansen Legacy project (RCN
project number 276730). The authors were grateful to all those
who participated in the different fieldwork experiments.
The RS-2 image overlapping the MC-20 area was provided by
MDA Geospatial Services (2016) and the RS-2 image covering
the sea ice and OOW exercises was provided by NSC/KSAT
under the Norwegian–Canadian RADARSAT agreement 2011,
2012, 2013, and 2019. UAVSAR data can be downloaded
from uavsar.jpl.nasa.gov or from the Alaska Satellite Facility
(www.asf.alaska.edu). The publication charges for this article
have been funded by a grant from the publication fund of UiT
The Arctic University of Norway.
REFERENCES
[1] J. Stroeve, T. Markus, W. Meier, and J. Miller, “The Arctic’s rapidly
shrinking sea ice cover: A research synthesis,” Climate Change, vol. 110,
pp. 1005–1027, 2012.
[2] L. Pizzolato, S. Howell, C. Derksen, J. Dawson, and L. Copland, “Chang-
ing sea ice conditions and marine transportation activity in Canadian Arctic
waters between 1990 and 2012,” Climate Change, vol. 123, pp. 161–173,
2014.
[3] L. Pizzolato, S. Howell, J. Dawson, F. Lalibertê, and L. Copland, “The
influence of declining sea ice on shipping activity in the Canadian Arctic,”
Geophysical Res. Letters, vol. 43, no. 23, pp. 12 146–12 154, 2016.
[4] J. Wilkinson et al., “Oil spill response capabilities and technologies for
ice-covered Arctic marine waters: A review of recent developments and
established practices,” Ambio, vol. 46, pp. 423–441, 2017.
[5] L. C. Smith and S. R. Stephenson, “New trans-Arctic shipping route
navigable by midcentury,” PNAS, vol. 110, no. 13, pp. E1191–E1195,
2013.
[6] M. Babiker, K. Kloster, S. Sandven, and R. Hall, “The utilisation of
satellite images for the oil in ice experiment in the Barents Sea,” SINTEF,
Trondheim, Norway, Oil in Ice - JIP, Tech. Rep. 29, 44p, 2010.
[7] H. Hühnerfuss, W. Alpers, and K. Richter, “Discrimination between crude-
oil spills and monomolecular sea slicks by airborne radar and infrared
radiometer-possibilites and limitations,” Int. J. Remote Sens., vol. 7, no. 8,
pp. 1001–1013, 1986.
[8] W. Alpers and H. Hühnerfuss, “Radar signatures of oil films floating on the
sea surface and the Marangoni effect,” J. Geophysical Res. Atmospheres,
vol. 93, no. C4, pp. 3642–3648, 1988.
[9] K. Singh, A. Gray, B. P. Hawkins, and R. O’Neil, “The influence of surface
oil on C- and Ku-band ocean backscatter,” IEEE Trans. Geosci. Remote
Sens., vol. GE-24, no. 5, pp. 738–744, 1986.
[10] M. Gade, W. Alpers, H. Hühnerfuss, V. R. Wismann, and P. A. Lange,
“On the reduction of the radar backscatter by oceanic surface films:
Scatterometer measurements and their theoretical interpretation,” Remote
Sens. Environ., vol. 66, pp. 52–70, 1998.
[11] N. Firoozy et al., “An electromagnetic detection case study on crude oil
injection in a young sea ice environment,” IEEE Trans. Geosci. Remote
Sens., vol. 55, no. 8, pp. 4465–4475, 2017.
[12] D. Desmond et al., “Examining the physical processes of corn oil (medium
crude oil surrogate) in sea ice and its resultant effect on complex permit-
tivity and normalized radar cross-section,” Marine Pollut. Bull., vol. 142,
pp. 484–493, 2019.
[13] M. Oggier, H. Eicken, J. Wilkinson, C. Petrich, and M. O’Sadnick, “Crude
oil migration in sea-ice: Laboratory studies of constraints on oil mobiliza-
tion and seasonal evolution,” Cold Regional Sci. Technol., vol. 174, 2020,
Art. no. 102924.
[14] D. Saltymakova et al., “Effect of dissolution, evaporation, and photoox-
idation on crude oil chemical composition, dielectric properties and its
radar signature in the Arctic environment,” Mar. Pollut. Bull., vol. 151,
no. 110629, 2020.
[15] C. Brekke, B. Holt, C. Jones, and S. Skrunes, “Discrimination of oil spills
from newly formed sea ice by synthetic aperture radar,” Remote Sens.
Environ., vol. 145, pp. 1–14, 2014.
[16] W. Dierking, “Mapping of different sea ice regimes using images from
Sentinel-1 and ALOS synthetic aperture radar,” IEEE Trans. Geoscience
Remote Sens., vol. 48, no. 3, pp. 1045–1058, Mar. 2010.
[17] R. Kwok, E. Rignot, B. Holt, and R. Onstott, “Identification of sea ice
types in spaceborne synthetic aperture radar data,” J. Geophysical Res.:
Oceans, vol. 97, no. C2, pp. 2391–2402, 1992.
[18] R.G. Onstott, “SAR and scatterometer signatures o f sea ice,” in Microwave
Remote Sensing of Sea Ice. Washington, DC, USA: Amer. Geophysical
Union, 1992, pp. 73–104.
[19] D. Isleifson, B. Hwang, D. G. Barber, R. K. Scharien, and L. Shafai,
“C-Band polarimetric backscattering signatures of newly formed sea ice
during fall freeze-up,” IEEE Trans. Geosci. Remote Sens. Geoscience
Remote Sens., vol. 48, no. 8, pp. 3256–3267, Aug. 2010.
JOHANSSON et al.: CAN MINERAL OIL SLICKS BE DISTINGUISHED FROM NEWLY FORMED SEA ICE USING SAR? 5009
[20] W. Alpers and H. Hühnerfuss, “The damping of ocean waves by surface
films: A new look at an old problem,” J. Geophysical Rese., vol. 94,
pp. 6251–6265, 1989.
[21] I. Sergievskaya, S. Ermakov, T. Lazareva, and J. Guo, “Damping of surface
waves due to crude oil/oil emulsion films on water,” Mar. Pollut. Bull.,
vol. 146, pp. 206–214, 2019.
[22] B.Minchew, C. E. Jones, and B. Holt, “Polarimetric analysis of backscatter
from the deepwater horizon oil spill using L-band synthetic aperture
radar,” IEEE Trans. Geosci. Remote Sens., vol. 50, no. 10, pp. 3812–3830,
Oct. 2012.
[23] L. Klein and C. Swift, “An improved model for the dielectric constant
of sea water in microwave frequencies,” IEEE J. Ocean. Eng.,vol.2,
pp. 104–111, Jan. 1977.
[24] K. Folgerø, “Bilinear calibration of coaxial transmission/reflection cells
for permittivity measurement of low-loss liquids,” Meas. Sci. Technol.,
vol. 7, no. 9, pp. 1260–1269, 1996.
[25] T. Friisø, Y. Schildberg, O. Rambeau, T. Tjomsland, H. Føredal, and J.
Sjøblom, “Complex permittivity of crude oils and solutions of heavy crude
oil fraction,” J. Dispersion Sci. Technol., vol. 19, no. 1, pp. 93–126, 1998.
[26] N. Firoozy et al., “A controlled experiment on oil release beneath thin sea
ice and its electromagnetic detection,” IEEE Trans. Geoscience Remote
Sens., vol. 56, no. 8, pp. 4406–4419, Aug. 2018.
[27] T. Neusitzer et al., “Examining the impact of a crude oil spill on the
permittivity profile and normalized radar cross section of young sea ice,”
IEEE Trans. Geosci. Remote Sens., vol. 56, no. 2, pp. 921–936, 2018.
[28] S. Angelliaume et al., “SAR imagery for detecting sea surface slicks:
Performance assessment of polarization-dependent parameters,” IEEE
Trans. Geosci. Remote Sens., vol. 56, no. 8, pp. 4237–4257, Aug. 2018.
[29] HELCOM, “Joint final report on II audit of implementation of provi-
sions of the convention on the protection of the marine environment
of the Baltic Sea Area (Helsinki Convention) - Pollution from ships in
the Baltic Sea,” HELSINKI Commission, Riga, Latvia, Jan. 2005. [On-
line]. Available: https://portal.helcom.fi/Archive/archive2/RESPONSE%
205-2005 10- 1.pdf
[30] M. Konik and K. Bradtke, “Object-oriented approach to oil spill detection
using ENVISAT ASAR images,” ISPRS J. PhotogrammetryRemote Sens.,
vol. 118, no. 2016, pp. 37–52, 2016.
[31] M. Migliaccio, A. Gambardella, and M. Tranfaglia, “SAR polarimetry to
observe oil spills,” IEEE Trans. Geoscience Remote Sens., vol. 45, no. 2,
pp. 506–511, 2007.
[32] T. Geldsetzer and J. J. Yackel, “Sea ice type and open water discrimina-
tion using dual co-polarized C-band SAR,” Canadian J. Remote Sens.,
vol. 35, no. 1, pp. 73–84, 2009. [Online]. Available: http://dx.doi.org/10.
5589/m08-075
[33] M. Espeseth, C. Brekke, C. E. Jones, B. Holt, and A. Freeman, “The
impact of system noise in polarimetric SAR imagery on oil spill observa-
tions,” IEEE Trans. Geosci. Remote Sens., vol. 58, no. 6, pp. 4194–4214,
Jun. 2020.
[34] S. Skrunes, C. Brekke, C. Jones, M. Espeseth, and B. Holt, “Effect of
wind direction and incidence angle and polarimetric SAR observations
of slicked and unslicked surfaces,” Remote Sens. Environ., vol. 213,
pp. 73–91, 2018.
[35] A. M. Johansson, C. Brekke, G. Spreen, and J. A. King, “X-, C-, and
L-band SAR signatures of newly formed sea ice in Arctic leads during
winter and spring,” Remote Sens. Environ., vol. 204, pp. 162–180, 2018.
[36] G. Valenzuela, “Theories for the interaction of electromagnetic and
oceanic waves—A review,” Boundary-Layer Meteorology, vol. 12, no. 1-4,
pp. 61–85, 1978.
[37] I. Hajnsek, E. Pottier, and S. R. Cloude, “Inversion of surface parameters
from polarimetric SAR,” IEEE Trans. Geosci. Remote Sens., vol. 41, no. 4,
pp. 727–744, Apr. 2003.
[38] V. Kudryavtsev,B. Chapron, A. Myasoedov, F.Collard, and J. Johannessen,
“On dual co-polarized SAR measurements of the ocean surface,” IEEE
Geosci. Remote Sens. Lett., vol. 10, no. 4, pp. 761–765, Jul. 2013.
[39] I. H. Wasbotten, Oil budget and oil analysis, NOFOs, OPV2015, (in
Norwegian), Akvaplan-niva AS, Troms, Norway, Tech. Rep. 7471-03,
2015.
[40] C. Jones and B. Holt, “Experimental L-band Airborne SAR for oil spill
response at sea and in coastal waters,” Sensors, vol. 18, no. 2, 2018, Art.
no. 641.
[41] J. Thomson et al., “Overview of the Arctic sea state and boundary layer
physics program,” J. Geophysical Res.: Oceans, vol. 123, pp. 8674–8687,
2018.
[42] A.G. Fore et al., “UAVSAR polarimetric calibration,” IEEE Trans. Geosci.
Remote Sens., vol. 53, no. 6, pp. 3481–3491, Jun. 2015.
[43] S. Sun, C. Hu, O. Garcia-Pineda, V. Kourafalou, M. Le Hénaff, and Y.
Androulidakis, “Remote sensing assessment of oil spills near a damaged
platform in the Gulf of Mexico,” Mar. Pollut. Bull., vol. 136, pp. 141–151,
2018.
[44] M. A. Granskog, I. Fer, A. Rinke, and H. Steen, “Atmosphere-ice-ocean-
ecosystem processes in a thinner Arctic sea ice regime: The Norwegian
Young Sea ICE (N-ICE2015) expedition,” J. Geophysical Res.: Oceans,
vol. 123, no. 3, pp. 1586–1594, 2018.
[45] S. Skrunes, C. Brekke, and T. Eltoft, “Characterization of marine surface
slicks by Radarsat-2 multipolarization features,” IEEE Trans. Geosci.
Remote Sens., vol. 52, no. 9, pp. 5302–5319, Sep. 2014.
[46] S. Skrunes, C. Brekke, T. Eltoft, and V. Kudryavtsev, “Comparing
near-coincident C- and X-band SAR acquisitions of marine oil spills,”
IEEE Trans. Geosci. Remote Sens., vol. 53, no. 4, pp. 1958–1975,
Apr. 2015.
[47] S. Skrunes, C. Brekke, and A. Doulgeris, “Characterization of low-
backscatter ocean features in dual-copolarization SAR using log-
cumulants,” IEEE Geosci. Remote Sens. Lett., vol. 12, no. 4, pp. 836–840,
Apr. 2015.
[48] O. Garcia-Pineda et al., “Classification of oil spill by thicknesses using
multiple remote sensors,” Remote Sens. Environ., vol. 236, 2020, Art. no.
111421.
[49] P. Persson et al., “Shipboard observations of meteorology and near-surface
environment during autumn freezeup in the Beaufort/Chukchi seas,” J.
Geophysical Res.: Oceans, vol. 123, pp. 4930–4969, 2018.
[50] M. A. Granskog et al., “Snow contribution to first-year and second-year
Arctic sea ice mass balance north of Svalbard,” J. Geophysical Res.:
Oceans, vol. 122, no. 3, pp. 2539–2549, 2017.
[51] F. Massry, “The Kolmogrov–Smirnow test for goodness of fit,” J. Amer.
Stat. Assoc., vol. 46, pp. 68–78, 1951.
[52] M. M. Espeseth, C. E. Jones, B. Holt, C. Brekke,and S. Skrunes, “Oil-spill-
response-oriented information products derived from a rapid-repeat time
series of SAR images,” IEEE J. Sel. Topics Appl. Earth Observ. Remote
Sens., vol. 13, pp. 3448–3461, 2020.
[53] M. Younis, S. Huber, A. Patyuchenko, F. Bordoni, and G. Krieger,
“Performance comparison of reflector- and planar-antenna based digital
beam-forming sar,” Int. J. Antennas Propagation, 2009, Art. no. 614931.
[54] W. Alpers, B. Holt, and K. Zeng, “Oil spill detection by imaging radars:
challenges and pitfalls,” Remote Sens. Environ., vol. 201, pp. 133–147,
2017.
[55] RADARSAT-2 Product Description. Richmond, BC, Canada: Maxar Tech-
nologies, RN-SP-52-1238, Issue 1/13, Mar. 2016.
[56] W. Press, S. Teukolsky, W. Vetterling, and B. Flannery, Numerical recipes:
The art of scientific computing. New York, NY, USA: Cambridge Univ.
Press, 2007.
[57] T. Geldsetzer, M. Arkett, T. Zagon, F. Charbonneau, J. J. Yackel, and R.
K. Scharien, “All-season compact-polarimetry C-band SAR observations
of sea ice,” Can. J. Remote Sens., vol. 41, no. 5, pp. 485–504, 2015.
[58] M. Dabboor, B. Montpetit, S. Howell, and C. Haas, “Improving sea ice
characterization in dry ice winter conditions using polarimetric parameters
from C- and L-Band SAR Data,” Remote Sens., vol. 9, 2017, Art. no. 1270.
[59] H. Wakabayashi, T. Matsuoka, K. Nakamura, and F. Nishio, “Polarimetric
characteristics of sea ice in the Sea of Okhotsk observed by airborne L-band
SAR,” IEEE Trans. Geosci. Remote Sens., vol. 42, no. 11, pp. 2412–2425,
Nov. 2004.
[60] M. Espeseth, S. Skrunes, C. Jones, C. Brekke, B. Holt, and A. Doulgeris,
“Analysis of evolving oil spills in full-polarimetric and hybrid-polarity
SAR,” IEEE Trans. Geosci. Remote Sens., vol. 55, no. 7, pp. 4190–4210,
Jul. 2017.
[61] D. Perovich and J. Richter-Menge, “Surface characteristics of lead ice,” J.
Geophysical Res., vol. 99, no. C8, pp. 16 341–16 350, 1994.
[62] D. Isleifson, R. Galley, N. Firoozy, J. Landy, and D. Barber, “Investiga-
tions into frost flower physical characteristics and the C-band scattering
response,” Remote Sens., vol. 10, 2018, Art. no. 991.
[63] M.Brath, S. Kern, and D. Stammer, “Sea ice classification during freeze-up
conditions with multifrequency scatterometer data,” IEEE Trans. Geosci.
Remote Sens., vol. 51, no. 6, pp. 3336–3353, Jun. 2013.
[64] M. R. Drinkwater, R. Kwok, D. P. Winebrenner, and E. Rignot, “Multi-
frequency polarimetric synthetic aperture radar observations of sea ice,”
J. Geophysical Res.: Oceans, vol. 96, no. C11, pp. 20 679–20 698,
1991.
[65] D. P. Winebrenner, L. D. Farmer, and I. R. Joughin, “On the response of
polarimetric synthetic aperture radar signatures at 24-cm wavelength to
sea ice thickness in Arctic leads,” Radio Sci., vol. 30, no. 2, pp. 373–402,
Mar./Apr. 1995.
5010 IEEE JOURNAL OF SELECTED TOPICS IN APPLIED EARTH OBSERVATIONS AND REMOTE SENSING, VOL. 13, 2020
[66] T. Fritz and M. Einender, “TerraSAR-X ground segment basic product
specification document,” DLR, Cologne, Germany, Tech. Rep. TX-GS-
DD-3302, 2013.
[67] M. Shimada, M. Watanabe, T. Motohka, Y. Kankaku, and S. Suzuki,
“Calibration and validation of PALSAR-2,” IEEE Int. Geosci. Remote
Sens. Symp., 2015.
[68] J. Karvonen, E. Rinne, H. Sallila, and M. Mäkynen, “On suitability of
ALOS-2/PALSAR-2dual-polarized SAR data for Arctic sea ice parameter
estimation,” IEEE Trans. Geosci. Remote Sens., 2020, to be published, doi:
10.1109/TGRS.2020.2985696.
[69] S. Singha, A. M. Johansson, N. Hughes, S. M. Munk Hvidegaard, and
H. Skourup, “Arctic sea ice characterization using spaceborne fully po-
larimetric L-, C- and X-band SAR with validation by airborne measure-
ments,” IEEE Trans. Geosci. Remote Sens., vol. 56, no. 7, pp. 3715–3734,
Jul. 2018.
[70] NASA-ISRO SAR (NISAR) Mission Science Users’ Handbook. NASA Jet
Propulsion Lab., Pasadena, CA, USA, Tech. Rep. 261, 2018.
A. Malin Johansson (Member, IEEE) received the
M.Sc. degree in physical oceanography from Gothen-
burg University, Sweden, in 2005, and the Ph.D.
degree in remote sensing from Stockholm University,
Sweden, in 2012.
She is a Research Scientist with the Department of
Physics and Technology, UiT The Arctic University
of Norway, Troms, Norway, which she jointed in
2014. She was a Postdoctoral Researcher with the
Radar Remote Sensing group with the Department
of Earth and Space Sciences, Chalmers University of
Technology, Gothenburg, Sweden. Her research interests include multisensor
remote sensing of sea ice, oil spills, and harmful algae blooms.
Martine M. Espeseth (Member, IEEE) received the
M.Sc. and Ph.D. degrees in remote sensing from the
Department of Physics and Technology, UiT The
Arctic University of Norway, Tromsø, Norway, in
2015 and 2019, respectively.
She is currently a Postdoc with the same depart-
ment, with the Centre for Integrated Remote Sensing
and Forecasting for Arctic Operations (CIRFA).From
February to April 2016 and August to December
2018, she was a Visiting Ph.D. student with the Jet
Propulsion Laboratory, California Institute of Tech-
nology, Pasadena, CA, USA. Her current research interests include remote
sensing of polarimetric SAR and with a focus on compact polarimetry within
both marine oil pollution and sea ice applications.
Camilla Brekke (Member, IEEE) received the Cand.
Mag., Cand. Scient., and Ph.D. degrees from the
Department of Informatics, University of Oslo, Oslo,
Norway, in 1998, 2001, and 2008, respectively.
She is currently the Vice-Dean Research with the
Faculty of Science and Technology, the Deputy Cen-
tre Leader with the Centre for Integrated Remote
Sensing and Forecasting for Arctic Operations and
full Professor at Department of Physics and Technol-
ogy, UiT The Arctic University of Norway, Tromsø,
Norway. Her current research interests include syn-
thetic aperture radar and ocean color remote sensing for Arctic and marine
applications.
Benjamin Holt (Member, IEEE) received the B.A.
degree in human biology from Stanford University,
Stanford, CA, USA, in 1972, and the M.S. degree in
physical oceanography from the University of South-
ern California, Los Angeles, CA, USA, in 1988.
He is a Research Scientist with the Ocean Circula-
tion group within the Earth Science Section at the Jet
Propulsion Laboratory, California Institute of Tech-
nology, Pasadena,CA, USA, since 1978. His research
interests include using multisensor remote sensing
data to examine the geophysical state of polar sea
ice and snow, coastal oceanography and circulation, and the detection of marine
pollutants. In addition, he is also involved with new instrument development and
techniques for microwave measurement of sea ice thickness.
