Based upon thermal-infrared satellite imagery in combination with ERA-Interim
atmospheric reanalysis data, we derive long-term polynya characteristics such as
polynya area, thin-ice thickness distribution, and ice-production rates for a
13-year investigation period (2002–2014) for the austral winter
(1 April to 30 September) in the Antarctic southern Weddell Sea. All polynya
parameters are derived from daily cloud-cover corrected thin-ice thickness composites.
The focus lies on coastal polynyas which are important hot spots for new-ice formation,
bottom-water formation, and heat/moisture release into the atmosphere. MODIS
has the capability to resolve even very narrow coastal polynyas. Its major
disadvantage is the sensor limitation due to cloud cover. We make use of a
newly developed and adapted spatial feature reconstruction scheme to account
for cloud-covered areas. We find the sea-ice areas in front of the Ronne and
Brunt ice shelves to be the most active with an annual average polynya area of
3018
Coastal polynyas are recurring areas of thin ice and open
water which are generally formed by divergent ice motion, e.g., by strong
offshore winds or ocean currents. During winter, polynyas are important hot spots for ice
production (IP), deep-water formation and gas ventilation of the ocean (e.g.,
Due to the dependency of ocean-to-atmosphere heat flux on the ice thickness
In this study, the geographical focus lies on the southern Antarctic Weddell
Sea region between 69–78
Sketch of the study area in the southern Weddell Sea with all six
investigated subregions along the coast: Antarctic Peninsula (AP), Ronne Ice
Shelf (RO), the area around the grounded iceberg A23A (IB), Filchner Ice
Shelf (FI), Coats Land (CL), and the Brunt Ice Shelf (BR). Color shadings show
the bathymetry based on
The majority of recent studies dealing with polynya dynamics in this region make use of passive-microwave
sensors. In comparison to thermal-infrared sensors, passive-microwave sensors retrieve
data with no limitations due to day-/nighttime or clouds.
In this study, we present long-term results of coastal-polynya dynamics in the
southern Weddell Sea that are derived from Moderate-Resolution Imaging
Spectroradiometer (MODIS) thermal-infrared imagery. Remote sensing of sea ice
using thermal-infrared data yields the opportunity to monitor thin-ice
thicknesses (TITs) and distribution on a regular basis
In this study we make use of the MODIS Sea Ice product (MOD/MYD29,
The overall IST accuracy of the MOD/MYD29
sea-ice product under ideal (i.e., clear-sky) conditions is 1–3 K
For comparison, we use daily sea-ice concentration data derived from the Advanced Microwave
Scanning Radiometer 2 (AMSR2) provided by the University of Bremen
The set of input data is complemented by the ECMWF ERA-Interim atmospheric reanalysis data
For our analysis we used the 2 m air temperature, the 10 m wind-speed components, the mean sea-level pressure, and the 2 m dew-point temperature. These data sets supply all necessary input to calculate the energy balance for the thin-ice retrieval.
In addition to these meteorological variables, cloud-cover information was
also taken from the ERA-Interim reanalysis data. In our case, we used the
medium-level cloud-cover fraction data, which were found to show the in
general best agreement with the MODIS cloud mask
The derived TIT of up to 0.5 m is calculated by using a
surface energy balance model
The turbulent fluxes of sensible and latent heat
as well as the long-wave radiation balance are calculated following
Studies of the ERA-Interim downward long-wave radiation data
revealed that ERA-Interim overestimates the cloud-cover fraction of low clouds
The exact procedure of the thin-ice thickness retrieval as well as all
equations are thoroughly described in
For a better comparison to other studies, the southern Weddell Sea coastal area
was divided into six subregions (Fig.
All MODIS swaths were processed in several steps as visualized in Fig.
First, the swaths were projected onto a common equirectangular grid with an average spatial
resolution of 2 km
Work flow based on the described input data and thin-ice retrieval.
Established processing steps that were already used in other studies are
marked in blue, while new additions are marked in red. These additions
comprise a persistence index (PIX), the newly defined combined cloud-cover
information of MODIS cloud mask and ERA-Interim medium-level cloud cover.
Missing data due to cloud cover are accounted for by a two-step procedure that
comprises spatial feature reconstruction
Effect of different ERA-Interim medium-level cloud cover (mcc)
thresholds on the retrieved daily thin-ice thickness of 28 June 2008
Based on these adjusted swath-based data sets, TIT is calculated pixel-wise and daily TIT composites are calculated based on the median thin-ice thickness and corresponding ice-surface temperature of all available swaths per pixel. The resulting composites comprise TIT and IST data together with the daily swath-based median energy-balance components of each thin-ice pixel. In the next step, cloud-covered data were identified and flagged as will be described in the following subsections.
As already mentioned, the MODIS cloud mask shows deficiencies in the nighttime
cloud-cover detection. We therefore use ERA-Interim medium-level cloud-cover information to reduce the
influence of cloud-affected pixels in the aggregation process of the daily
TIT composites. We apply a simple cloud-cover threshold (75 %, Fig.
On average, about five swaths cover each pixel in a daily composite. This number of swaths can vary mainly due to cloud cover with up to 20 different swaths covering a single pixel or region.
In order to reduce the inherent misclassification problems of the MODIS cloud
mask with clouds taken for either thin or thick ice, we introduced an
additional procedure that assigns a persistence index (PIX;
Figs.
Both the number of swaths and the persistence index are then used to
further reduce the effect of undetected cloud cover on later processing steps (Fig.
Employing the above mentioned two procedures of cloud-cover dependent
classification and persistence index calculation, we are able to identify
cloud-contaminated MODIS data in each daily TIT composite. These are then complemented by a two-step
procedure utilizing a combination of spatial feature reconstruction (SFR;
In the SFR approach, the information of a 7-day interval (doi
The PE approach on the other hand assigns thin ice to cloud-covered areas in the same proportion as it is detected in the cloud-free area. For example, if a region is 80 % cloud free and 50 % of the cloud-free area features a thin-ice signal, then 50 % of the cloud-covered region is considered as thin ice.
In the two-step procedure, we first apply the SFR approach to all cloud-free pixels (i.e., pixels
in the ccs class and mcp class where the majority of pixels show clear-sky conditions) that also feature
a PIX value greater than 0.5 (i.e., pixels with a thin-ice thickness
Based on Eq. (
Subsequently, cloud-contaminated pixels with a probability above the threshold (0.34) are
assigned a pixel-wise weighted average ice thickness and
ice-surface temperature value based on the 6 days surrounding the initial doi
(TIT
The remaining coverage gaps that could not be corrected
for by this approach, e.g., due to temporal gaps longer than 3 consecutive
days, are filled by the PE scheme
Comparison between frequency of low sea-ice concentration occurrences
(
In the case that after the application of the SFR approach more than 50 % of the investigated subregion is cloud-contaminated, daily estimates of polynya area and ice production will be interpolated between neighboring days with sufficient (i.e., above 50 %) cloud-free coverage.
From our cloud-cover corrected daily thin-ice thickness composites, we then
derive daily polynya area (defined as area with open water and thin-ice
between 0.0 and 0.2 m thickness) as well as the accumulated
wintertime ice production from heat loss for each POLA pixel (e.g.,
Furthermore, thin-ice thickness distributions of daily POLA as well as frequency distributions of thin-ice occurrence are calculated for each subregion. The results are then put in context with other recent remote sensing and model studies.
A threshold of 70 % was used for the definition of a polynya pixel from
AMSR2. The comparison of frequency of polynya pixel occurrence from AMSR2
(Fig.
The difference in frequencies is shown in Fig.
The spatial coverage of cloud-free MODIS data without a cloud-cover
correction is roughly between 40 and 100 % due to the large amount of
daily swaths that at least partially cover a large region like the southern
Weddell Sea. The use of the spatial feature reconstruction approach (
An average of 78 % in all MODIS composites per day over all six subregions is classified as cloud free. After the application of the SFR approach, we achieve an average coverage of 97 % per day, where the remaining 3 % get corrected for by PE.
Frequency distribution of thin-ice occurrences in percent for the years
2002 to 2014
When comparing our two-step cloud-cover correction procedure to the sole use
of proportional extrapolation (Table
Whenever no coverage with cloud-free information above 50 % after the use of SFR can be
achieved, the POLA information of that day will be interpolated between days
with sufficient (i.e., above 50 %) coverage. Under the same premise,
the PE approach is used for comparison to the two-step procedure
(SFR
Almost the complete coastal area in the south and east of the investigated
southern Weddell Sea features a recurrent thin-ice signal for the years from
2002 to 2014 (around 25 % and above with extremes of up to 80 % for some years, Fig.
The relative frequency of thin-ice occurrences in the AP
region is spatially focused around smaller grounded icebergs and rather
low compared to the other subregions. The interannual contribution also decreases
during our investigation period in the years 2007 to 2010 (Fig.
Comparison between uncorrected MODIS data, sole use of
proportional extrapolation (PE;
The very light blue areas in the overall thin-ice frequency distribution (Fig.
There is a sharp separation between a zone with present activity in the north and northeast to a zone with almost no
activity in the south and southwest closer to the Ronne and Filchner ice
shelves in the total frequency distribution (Fig.
In comparison to results found by
The overall distribution of thin-ice thicknesses for the southern Weddell Sea
(Fig.
Results for POLA and IP for the years from
2002 to 2014 between 1 April and 30 September are presented in
Figs.
Averaged distribution of thin-ice thicknesses in the southern
Weddell Sea over the whole 13-year time interval between April and
September. The class width is 2 cm; i.e., the 1 cm class
includes thicknesses between 0 and 2 cm. Error bars represent
Generally, by means of multi-year trends over the
course of 13 years in POLA and IP (Figs.
Another important fact that is shown by our data set is the importance of the
FI and CL region. Together with the area
around the grounded IB, each shows a higher average POLA and
IP than the AP region. However, the interannual
variability is very high, especially in the IB and FI region. The combined
average POLA of IB, FI, and CL is in the same order of magnitude as the larger
contributors of the Ronne and Brunt ice shelves (Figs.
The frequency of days exceeding a certain polynya area threshold is shown in
Fig.
A threshold of 8000 and 9000 km
Annual average polynya days (Fig.
The interannual winter average POLA and the wintertime
accumulated IP are compared to the results from
Mean wintertime polynya area (POLA, km
Accumulated wintertime ice production (IP, km
Frequency distribution (%) of days passing the corresponding
polynya area threshold (km
Number of days (polynya days) separated into three polynya area (POLA)
classes (km
Average wintertime (May to September) polynya area
(POLA, km
Accumulated wintertime (May to September) ice-production rate
(IP, km
For the AP region, the model estimates are consistently
higher than our results (Figs.
Focusing on polynya-area comparison for the two subregions RO and BR
(Fig.
By means of ice production, estimates for the RO as well as
the BR region are in the same order of magnitude and
oftentimes very similar (with the exception of the years 2002 and 2003 for
RO; Fig.
One can only speculate about the reasons for the increased average annual
POLA and IP in the years 2002 and 2003 as there is no statement in the study
of
Assuming that the satellite-based estimates represent the upper limit for ice
production, the relatively high model estimates
In the study of
Summary of average polynya-area estimates (km
A comparison between different studies based on satellite observations as
well as models is difficult due to different investigation periods, diversity
in spatial extent of used POLA masks as well as varying methods and changes
in atmospheric forcing. For the following comparison of our results with
several recent studies and investigations of the southern Weddell Sea
(Tables
Based on average wintertime polynya area (Table
Our multi-year averages exceed the results from
Summary of accumulated volume ice-production estimates (km
Average wintertime (May to September) energy fluxes of
sensible/latent heat (
The comparison of ice production between ours and other recent studies in the
southern Weddell Sea (Table
Similar to the POLA results, the most recent study by
The mean wintertime area-weighted atmospheric fluxes are shown in
Fig.
As expected, we find the sensible heat flux to be the largest contributor to
the total atmospheric flux (e.g.,
We stated earlier that the model estimates seem to achieve a higher ice
production per polynya area compared to the MODIS-derived results. This
is potentially related to large differences between the used atmospheric
reanalysis data of NCEP
In this study, we present a data set of MODIS-derived polynya-area estimates and ice-production rates as well as thin-ice frequency distribution, thin-ice thickness distribution, and energy-balance components for the southern Weddell Sea, Antarctica. This was done for a 13-year time interval (2002 to 2014) during the austral winter period from April to September for the complete coastal area separated into six subregions. For that, we utilized the higher spatial resolution of MODIS compared to passive-microwave sensors such as SSM/I and AMSR-E/AMSR2. The addition of a more strict exclusion of cloud-covered data using ERA-Interim data to the established thin-ice retrieval as well as the adaptation of a new approach to compensate for cloud-covered areas in daily MODIS composites is presented and discussed.
The data set is unique in a way that it is the first long-term investigation
of polynya dynamics based on cloud-cover corrected thermal-infrared data that
covers the complete southern Weddell Sea coastal area. The results were
discussed in comparison to recently published studies using a variety of
different methods and approaches (satellite sensors and models). On average
over 13 years, we find the sea-ice areas in front of the Ronne and Brunt
ice shelves to be the most active with an annual average polynya area of
3018
Given these results and also the presented thin-ice frequency distribution,
the neglect of certain regions in other studies, namely the area around the
grounded iceberg A23A as well as the area off the coast of the Filchner Ice
Shelf and Coats Land, showed up to drastically underestimate the total
average polynya area and ice production in the southern Weddell Sea.
Together, all three regions contribute comparably to the most active regions
in front of Ronne and Brunt ice shelves. These regions should be further
investigated by upcoming studies as they also contribute to the bottom-water
formation. Supplementary data are available at
The study was funded by the Deutsche Forschungsgemeinschaft in the framework of the priority program SPP1158 “Antarctic Research with comparative investigations in Arctic ice areas” by grant HE2740/12. The authors want to thank the National Snow and Ice Data Center and the European Centre for Medium-Range Weather Forecasts for the provision of the here-used data. The authors appreciate the help of Verena Haid and Ralph Timmermann by providing FESOM model results. The authors want to express their gratitude to an anonymous reviewer and Stefan Kern who helped to greatly improve this manuscript during the peer-review process alongside editor Daniel Feltham. Edited by: D. Feltham