Since the launch on 3 April 2014 of the first satellite of the Sentinel-1 family, Sentinel-1A, the Sentinel-1 Mission has provided a wealth of Synthetic Aperture Radar (SAR) imagery that has contributed to the development of an immense number of operational applications, and has helped scientist in Europe and across the world better understand a wide number of geophysical processes. Completing the constellation, Sentinel-1B, the second satellite in the series, was launched on 25 April 2016, and has provided since then SAR data of equivalent quality as its companion satellite. The constellation offers global coverage with a revisit time of six hours and provides an almost complete coverage of the polar regions every 2 days.

Besides, the introduction of focused-SAR altimetry (Egido & Smith, 2017) demonstrated that nadir-looking altimeter data can be processed with conventional SAR imagery data processing techniques, and could actually lead to improvements in the performance of SAR altimeters.

During the commissioning phase of Sentinel-1B, we conducted an experiment to determine the capabilities of the radar on-board the Sentinel-1 satellites to acquire the nadir echoes through the nadir side-lobe of the SAR antenna radiation pattern, with the sufficient signal-to-noise ratio to obtain altimetric estimates. That being the case would enable the acquisition of both the nadir returns and the side-looking echoes, used to form the SAR image, without compromising or affecting the mission’s routine operations.

For this experiment, we acquired two tracks over the Antarctic ocean in July 2016, covering a significant portion of sea-ice and ice-shelves. In these two acquisitions we extended the reception window and advanced the sampling window start time, which allowed us to receive simultaneously the nadir echoes and the nominal side-looking echoes, from echo ranks 9 and 10, respectively. Both side-looking and nadir returns were processed in the same manner, i.e. with a focused SAR processor. The SAR images show the typical backscatter data together with a fine and bright line across the image corresponding to the nadir return. The high contrast between the nadir return and the backscatter information, particularly over the sea-ice, makes us believe that the signal-to-clutter ratio is high enough to obtain useful altimetric information. In this paper we explore the capabilities of this technique to measure sea-ice freeboard, which could be an excellent complementary source of information for other altimeter polar missions.

(Egido & Smith, 2017): A. Egido and W. H. F. Smith, "Fully Focused SAR Altimetry: Theory and Applications," in IEEE Transactions on Geoscience and Remote Sensing, vol. 55, no. 1, pp. 392-406, Jan. 2017

Area D - Board 436

Sea ice drift is forced by winds and ocean currents and is an essential element in the dynamics of the polar oceans. Sea ice extent, concentration, and thickness are heavily influenced by ice dynamics. Satellite observations of sea ice drift provide valuable information about the governing dynamical processes of sea ice and its role within the Antarctic climate system. In this study, we examine the sea ice drift from high-resolution satellite images and the percentage influence of wind (speed reduction factor) on sea ice motion in the western Ross Sea region, Antarctica. The Ross Sea region has experienced a significant increase to the extent of the sea ice in recent decades. The study area consist of three areas of open water - polynyas; McMurdo Sound Polynya (MSP), Terra Nova Bay Polynya (TNBP) and the Ross Sea Polynya (RSP).
Due to the vast area of the Antarctic sea ice cover and the timescale over which such investigations are required, satellite missions provide the only feasible mechanism for large-scale assessment of the sea ice motion. We used sequential satellite images for the estimation of sea ice motion. We utilized the high spatial and temporal resolution Envisat satellite Advanced Synthetic Aperture Radar (ASAR) images in Wide Swath (WS) mode. The drift velocity is calculated in centimeter per second using phase correlation technique. The images are downsampled from 75m to 150m spatial resolution. The outcome is validated from manually drawn vectors. For wind velocities, Antarctic Mesoscale Prediction System (AMPS) model data set is used. This 10m and three hourly wind data has 5km spatial resolution. The validated sea ice drift vectors are correlated with the wind velocity, which is one of the main forces responsible for driving sea ice motion. Here we present the results of a case study for a one month period.

Area D - Board 437

The NISAR mission, planned to be launched in 2022, will carry both L-band and S-band SARs in a sun-synchronous orbit with a 12-day orbit repeat. The mission is a collaboration between NASA and the Indian Space Research Organization (ISRO). The planned very high data rate will provide complete mapping of the global land cover and polar regions to provide SAR imagery for the key science focus areas of solid earth, ecosystems, and cryosphere. Planned coverage includes complete mapping of the sea ice extent for both the Arctic and Southern Oceans. This talk will describe the NISAR sea ice science requirements, data coverage, imaging mode characteristics, and planned studies focused on the use of SAR imagery for sea ice research.
We will provide a brief overview of the NISAR mission and sensor characteristics, followed by the sea ice coverage maps and sea ice products. The L-band-derived products include maps of sea ice motion and deformation for both poles to be derived on routine basis, in conjunction with Sentinel-1 C-band data, as well as sea ice type and wind speed products for ice chart mapping. The S-band SAR has more limited coverage and primarily supports ISRO-supported science. However, certain polar regions will include both coverage by both sensors.

It has been shown by earlier missions and airborne campaigns that L-band provides different imaging characteristics than C-band for sea ice detection, arising from increased penetration and less sensitivity to upper zone air bubbles to provide enhanced detection of deformed ice but reduced detection of multiyear and first year ice. A study on ice motion using Radarsat-2 and ALOS PALSAR coincident imagery has shown that both frequencies perform similarly in the winter months but more motion vectors are obtained during the melt season by L-band arising from continuing detection of deformed ice, while the C-band signatures are strongly reduced by surface melt. These same differences in imaging characteristics between C- and L-band will likely improve ice type classification methods as well, particularly if both sensors are utilized.

Area D - Board 438

This study aims to exploit wide-swath SAR data to generate new sea ice motion fields based on combined use of offset tracking (pattern matching and feature tracking) and range Doppler shift with spatial resolution on the order of 1 km. The conventional sea ice drift retrieval relies on the offset tracking between consecutive image pairs. The combined pattern matching and feature tracking algorithm has high accuracy (ice displacement errors are below 300 m) but does not perform well in the marginal ice zone or in strong deformation zones where the appearance of sea ice on SAR imagery is not persistent. The SAR Doppler centroid shift can be used to extract instantaneous line-of-sight velocity fields at no extra cost, in all kinds of existing SAR systems. The main challenges in using the Doppler centroid shift method for sea ice motion retrieval are the low speed of the sea ice drift, which ranges from 7-23 cm/s in the Fram Strait, and several uncertainties in the geophysical Doppler shift estimation procedure. Considering the sensitivity of the Doppler shift to actual motion in the line-of-sight direction, which is 0.2 Hz/cm/s at 35 degrees incidence angle, the signal uncertainty must be lower than 3 Hz. We have developed an efficient Doppler calibration scheme that suppress the errors comes from orbit and attitude anomaly and antenna mis-pointing in a level of 2.5 Hz. Sea ice drift in the Fram Strait was derived using both the offset tracking and the Doppler shift from 22 ENVISAT ASAR ScanSAR images acquired in January 2010. The correlation between instantaneous line-of-sight drift velocity derived from the individual Doppler scenes and drift speed derived from pairs of SAR scenes is generally low due to the noise in the Doppler signal and the differences in the temporal scales. However, an intercomparison of the products averaged over one month showed overall high consistency of the two products (RMSE: 15 cm/s, slope: 1.15). The difference is the largest for the first sub-swath (RMSE: 36 cm/s, slope: 0.9) and decays with incidence angle (RMSE: 20 cm/s, slope: 0.99). Despite the low signal-to-noise ratio, low efficient resolution, and only one component of the drift vector, the Doppler algorithm proves to be very useful for complementing the offset tracking method in the marginal ice zones and ice deformation zones. An approach for assimilation of the Doppler instantaneous velocity into ice drift mean speed is proposed and its efficiency is illustrated in a case study.

Area D - Board 439

The mandatory automatic identification system (AIS) was developed to support the safety of marine traffic. An AIS message includes the ship identifier, location, speed and heading. In ice-covered seas, the ship speeds extracted from AIS data vary with ice conditions that are simultaneously reflected by features in synthetic aperture radar (SAR) images. In this study, the speed variation was related to the SAR features and the results were applied to generate a chart of expected speeds from the SAR image. The study was done in the Gulf of Bothnia in March 2013 for ships with ice class IA Super that are able to navigate without icebreaker assistance. The ships generally have different icebreaking capabilities and ice transit speeds. To establish comparability between ships we normalized their speed relative to their maximum recorded speed during the study period. Hence the speed was expressed as a dimensionless quantity called unit.
The SAR information was extracted from a SAR mosaic covering the whole Baltic Sea. The matching between AIS and SAR was complicated by ice drift during the time gap (from hours to two days) between the SAR image acquisition and the AIS message. Hence we calculated a set of local statistical SAR statistics over several scales, from 0.5 to 3.5 km scale. In addition to the SAR statistics we included into a feature vector ice thickness category which was extracted from the Finnish ice chart. The ice thickness information improved the results compared to feature vectors consisting only SAR statistics. A feature vector was assigned to each SAR pixel (resolution of 500 m) to enable the speed estimation. It functioned as the predictor vector and the ship speed as the response variable in the regression estimation.

Random forest tree regression (RFTR) was used to estimate the speed. One of the benefits of RFTR is that the feature vector can consist of many kinds of variables (e.g. continuous, discrete or categorical). About 25 % of the AIS measurements were used in the training of RFTR. Using the ensemble regression RFTR we converted the SAR statistics augmented with ice thickness to an estimated ship speed chart where to each pixel was estimated a ship speed. The accuracy was quantified by the mean squared error and by the fraction of estimates close to the actual speeds. These depended strongly on the route and the day. The mean squared error varied from 0.4 to 2.7 unit2 for daily routes. Sixty-five percent of the estimates deviated by less than one speed unit and 82\% by less than 1.5 speed units from the AIS speeds. The estimated daily mean speeds were close to the observations. The largest speed decreases were provided by the estimator in a dampened form or not at all. We also analyzed the error statistics in different ice thickness categories.

Area D - Board 440

Finnish Meteorological Institute (FMI) carries out a demonstration of a new coastal downstream service for the Baltic Sea: Baltic Sea landfast ice extent and thickness (acronym BALFI). This new service includes derived products (landfast ice extent and thickness, snow thickness, and sea ice deformation) based on three existing CMEMS (Copernicus Marine Environment Monitoring Service) Baltic Sea products, and additionally, on SENTINEL-1 SAR imagery, and a sea ice thermodynamic model run at FMI. These new Baltic Sea landfast ice products were developed and are generated operationally by FMI, likewise the service demonstration (web-service) is developed, maintained and hosted by FMI.
The Baltic Sea is partly covered by ice in every winter season. Every winter landfast ice occurs along the coasts of Finland, Sweden and Estonia (Bay of Bothnia, Sea of Bothnia, Gulfs of Finland and Riga). In a severe winter it is also present in the Southern Baltic Sea (e.g. Danish straits). Currently, Baltic Sea landfast ice extent and thickness information in fine scale (~1 km) is not directly available from any product or source. New BALFI service gives daily information on the landfast ice extent and thickness, snow thickness and degree of ice deformation (a qualitative figure) at spatial scale of 1 km.
Targeted end-users include people living on the Baltic Sea coast and islands who require information on landfast ice properties for recreational activities such as skiing, skating, snowmobiling and ice fishing, and for transporting people and goods along ice roads to/from islands; and governmental and local authorities/institutions like national Ice Services, icebreaker management, harbour authorities, search and rescue operators, and commercial activities like tourism and marine wind power. The targeted end-users cover two CMEMS areas of benefit: Maritime Safety and Marine and Coastal Environment.
Starting point for the BALFI products is extraction of the landfast ice extent along the Baltic Sea coast. It is estimated based on daily S-1/RS-2 SAR mosaics and the CMEMS SAR Baltic Sea ice drift product. Fast ice extent can be extracted from the ice drift data by locating the areas which have remained static, i.e. the ice drift integrated in time is small for a period of days before the current date. The landfast ice regions are also estimated by thresholding the temporal median of cross-correlation of S-1/RS-2 SAR imagery mosaics. In the product processing we combine estimates of fast ice area from the sea ice drift and SAR cross-correlation analyses for an more reliable overall estimate. Additionally, some morphological filtering can be applied to smooth the landfast ice boundary. The identified landfast ice area is used as an input for its sea ice thickness and snow thickness estimation. The BALFI service is started with snow and ice thickness data from the FMI’s HIGHTSI model, but later it is easy to switch to data from a CMEMS forecast model, or combine the data from the two models. Finally, sea ice deformation (here a qualitative figure) is estimated based on the cross-correlation between the HH and HV SAR channels. The more deformed ice, the more distinguishable features (above the noise level) the SAR imagery has at both channels, resulting to an increasing inter-channel cross-correlation as a function of increasing deformation. A schematic production chain for the new Baltic Sea landfast ice extent and thickness product is shown in Figure 1.
The BALFI products are distributed via web-portal at Sodankylä National Satellite Data Center (NSDC) of FMI, see http://balfi.nsdc.fmi.fi/. At the web-portal the products are visualized over a map layer, and users can zoom in and out in the map, and see product data values at a wanted geographical location. The web-portal gives also short overview of the BALFI service, e.g. product properties and conditions of product use. The products can also be downloaded as GeoTiff-files. More detailed information about the BALFI service for the public is in its dedicated webpage (http://nsdc.fmi.fi/services/ Copernicus_Marine_Service/BALFI).
The BALFI project was funded by Mercator Ocean within CMEMS User Uptake programme. The BALFI service demonstration will start in Dec 2018, and continues at least through three Baltic Sea ice seasons to May 2021.

Area D - Board 441

At a time when general trend shows a reduction of overall volume of Arctic sea ice, ice conditions vary dramatically geographically and from year to year. This leads to big uncertainties in prediction, it also demonstrates high importance of local real-time monitoring for safety purposes. Along with that, increasing shipping and oil and gas operations in the Arctic additionally accentuates the necessity of development of more efficient system.

In subpolar regions, where the polar night takes place throughout the winter period, the main source of sea ice information is satellite radar data from active microwave sensors, particularly synthetic aperture radar (SAR). Such data is generally unaffected by the atmosphere, solar illumination, or precipitation which make them essential for sea ice analysis.

The majority of sea ice information services in the world provide ice charts as one of the main products of sea ice analysis which to great extent depends on satellite radar data especially in areas with a lack of in-situ measurements. Though, being important assistance for marine operations, ice charts are characterized by range of disadvantages such as following:

-Loss of information about sea ice due to generalization (which includes loss of information on hazards);

-Uncertainties caused by retrieving the ice thickness information based on correlation of a stage of development and morphometric properties of the sea ice (especially in areas with lack of in-situ data);

-Delay in providing real-time sea ice information due to time-consuming process of ice chart composition.

In addition to the above points, the growing experience of ship captains in reading satellite images and the development of automated methods for large-scale ice mapping have determined the shift of the focus in the role of human analysis to small-scale sea ice analysis for hazards monitoring. It implies the necessity of deeper understanding of how hazardous conditions formation and development are detected and identified on SAR images.

As a case study heavy ice conditions in Kara Sea in spring of 2018 were examined. First, seven hazardous situations were detected in the third decade of March with use of Automates Identification System (AIS). Hazardous situations were defined as a vessel dropping considerably its speed due to ice conditions, stopping, and returning back to an initial route, or, in the extreme cases, the vessel grounding in packed ice. In total four hazardous areas were detected which were drifting under the wind-force across shipping lanes and where all seven events happened in different dates from 20th March to 31 March.

Then, using Sentinel-1 SAR for the whole period of interest and preceding one, a post-event analysis was accomplished which helped to detect factors led to development of hazardous conditions and their main characteristics. Size, shape, tone, and texture of hazardous areas, as well as their changes in time, were inspected. Also, a Landsat 8 image for 30th of March was used for the examination of the ice ridges situation and evaluation of their approximate highs.

High concentration of ridges with heights of around 10 meters and more were detected which correlates with relatively high brightness on SAR images. All areas corresponded to compression and shear zones which were formed in 19-20th of March and accompanied by intense deformations. Slip lines are clearly detected on radar images as newly-formed lineaments with sharp shift in sea ice drift velocities on both sides. Also, a dramatic change in shape of separate ice fields which were drifting along the slip lines was traced. Such alteration zones, being a distinctive indicator for severe deformations, determined potentially hazardous zones where vessels lately experienced difficulties.

In the result, based on the post-event analysis for Kara sea 2017-2018 winter navigation, the potential of combination of SAR and AIS data for real-time hazardous areas monitoring were shown. Taking into account that 5 out of 7 vessels experienced difficulties in two areas, it is also important to use radar data for tracing the hazardous zones drift after happened events. The further analysis of other alike events in different Arctic seas would help to improve SAR images interpretation in purpose of hazardous ice conditions detection.

Area D - Board 442

Introduction: To improve our understanding of the deformation processes of sea ice, monitoring the temporal evolution of deformed ice area is quite important. For this purpose, SAR is expected to be a useful tool because of its high spatial resolution. While C-band SAR has been used most frequently for the polar sea ice research, it was pointed out that L-band SAR is more suitable for discriminating ridges from level ice than C-band SAR from the comparative analysis of SAR images (e.g., Dierking and Busche, 2006). However, the difference between L-band and C-bad SAR in the wide range using satellite SAR images has not been fully investigated in the seasonal ice zone yet, and moreover, in the previous studies for the polar sea ice, validation from the in-situ observation was insufficient. Since in-situ observations have been conducted in the southern Sea of Okhotsk using PV “Soya” every winter in collaboration with Japan Coast Guard, we have available data sets for validation. Besides, since there is no multi-year in the Sea of Okhotsk, the classification would become simpler than in the polar regions. The purpose of this study is to examine the validity of L-band SAR for detecting the deformed ice area by comparing with a C-band SAR image in the southern Sea of Okhotsk, to develop an algorithm to extract deformed ice area, and then to find the properties of deformed ice area to clarify the dynamical thickening processes in the seasonal ice zone.

Data: PALSAR images obtained at a ScanSAR mode in the southern Sea of Okhotsk in January, February, and March from 2009 to 2011 were used in this study. The total number of the images amounted to 35. For comparison, a Radarsat-2 image obtained nearly at the same time was used. The photos and ice thickness data obtained on board the PV “Soya” were used for validation. PALSAR-2 images at a ScanSAR mode were also used to see the applicability of our algorithm.

Analytical methods: In analysis, we took four steps as follows: 1) to develop an algorithm to detect deformed ice with L-band SAR, based on in-situ observations, 2) compare between L-band and C-band SARs, 3) map deformed ice area with PALSAR images, and 4) examine the applicability of this algorithm to PALSAR-2 images. With photos, sea ice was categorized into three ice types: nilas (thin level ice), pancake (thin ice with rough surface), and
deformed ice (thick ice with rough surface).

Results: To develop an algorithm, the backscatter coefficients of PALSAR images were plotted as a function of incidence angle. As a result, the following two threshold curves were derived by the least square method to classify into three ice types.
σ0 = -6.0 log θi + 6.8 (dB) (deformed - pancake ice) for 20 < θi < 40 deg.
σ0 = -6.1 log θi + 3.2 (dB) (pancake ice - nilas) for 20 < θi < 40 deg.
where σ0 and θi denote a backscatter coefficient and the incidence angle, respectively. In the comparative analysis between Radarsat-2 and PALSAR images, ice types were determined by these thresholds. The results show that 1) L-band SAR is more suitable for detecting deformed ice than C-band SAR also in the southern Sea of Okhotsk, 2) the deformed ice area appears in relatively linear alignments mainly near the ice edge and with a width of a few tens of kilometers in the interior region (Figure), 3) affected by winds, ocean waves, and floe interaction, deformed ice area is quite variable with time and space, and 4) when applied to ALOS-2/PALSAR-2 images for the same region in February 2018, reasonable results were obtained. It was found that the temporal evolution in deformed ice area corresponded approximately to the convergence derived from the AMSR-derived ice drift pattern, showing the applicability of this method to some extent. To quantify these properties, further studies are needed.

Acknowledgments: ALOS/PALSAR images were provided by JAXA through the ALOS Research Project (PI No.573).

References
Dierking, W. and T. Busch, Sea ice monitoring by L-band SAR: An assessment based on literature and comparisons of JERS-1 and ERS-1 imagery, IEEE Trans. Geosci. Remote Sens., 44(2), 957-970, 2006.

Area D - Board 443

Sea ice mapping has been significantly improved during the last years by the advanced abilities of all-weather synthetic aperture radars (SAR), where high-resolution data from RADARSAT-2 and Sentinel-1 can perform the sea ice monitoring with higher resolution and better quality. Reliable sea ice information providing the operational activities in the polar region plays an important role for support the navigation, in particular, to develop the recommendations for the planning of ship routes taking into account long-term trends, seasonal and interannual variations of the ice conditions. Since identification of large-scale ice structures (e.g., floes) and deformation features (e.g., ridges, fractures) in SAR image are considered for accurate discrimination of ice stages of development, the delineation of these sea ice cover features and accurate detection of their size and location is also very useful for practical applications.
Discrimination between stages of development is hampered by the large SAR signal variations and significant overlapping of the backscatter signatures of various sea ice types entire the scenes. Due to the high wind and the incidence angle effect, some areas of the open water (OW) have similar properties of ice at HH. Additional information about sea ice and OW can thus be gained by using dual-polarized data and image texture property.
The algorithm has been created to automatically locate the ice edge and retrieve the sea ice types in the Kara Sea. The classification technique is based on the support vector machine (SVM) applying Sentinel-1 Extra Wide (EW) swath dual-polarization mode data, i.e. HH (horizontally transmitted and horizontally received) and HV (horizontally transmitted, vertically received). Based on the backscatter characteristics for both polarization channels, their ratio, and texture features extracted through gray level co-occurrence matrix (GLCM) measures in different window sizes, SAR images are classified into different ice types zones as well as ice free areas. SVM is a supervised learning method, and hence a training dataset with correct classes is needed. Analyzing of a number of SAR images taken over region of interest in winter months (November – May) the polygons of the ice types, which have to be classified, are delineated by experienced ice experts. Contribution of the thermal noise at HV band affects the quality of the image. Image pre-processing is the first step consisted of reducing of the backscatter dependency on incidence angles variations and thermal noise reduction at HV band using well developed approach. Next step is a calculation of texture features using combinations of sliding windows and interpixel distances, and then efficient set with optimal parameters is selected. Finally, the SVM is trained based on training datasets, obtained after collocation of expert visual analysis images with features images. Fully automated algorithm contains the pre-processing and texture features computation steps, and implementation of prepared SVM for automatic sea ice classification to divide the SAR scene into the predefined classes.
Our automated high resolution SAR image classification can provide clear ice edge location in all-year period that is very important especially during the onset of the melt or freezing seasons for making decisions to start/stop/continue the navigation operations. The algorithm can recognize ice floes inside the sea ice areas. The extended knowledge about the ability to retrieve the ice types in the region of interest received by the algorithm tests in last year allows to increase the classification quality. The current ability to correct the thermal noise effect at HV band (coming with IPF 2.9) is incorporated into the algorithm and let us prevent classification errors. The false separation between calm OW areas and floes of level first-year ice caused by the SAR image noise is improved.
The operational application of our algorithm according to the user’s requirements includes SAR data requests and collection, near-real-time processing with ice parameter extraction from the SAR data, integration of the obtained information into a certain data product, and providing the resulting ice charts to the user. Algorithm is able to classify types of the surface in the Kara Sea region: thin first-year ice, medium and thick first-year ice, young ice, ice free areas and nilas.

Area D - Board 444

Arctic sea ice extent and thickness have exhibited unprecedented minima in recent years, with observations indicating that significant declining trends in sea ice extent and thickness will continue at an alarming rate. Monitoring and identifying sea ice is crucial for assessing its seasonal and inter-annual variability, as well as Arctic climate change as a whole. Active microwave space-borne sensors like Synthetic Aperture Radar (SAR) have been widely used for sea ice type detection and classification. Advanced SAR systems allow for fully polarized data, which is characterized by high information content related to scattering mechanisms, and provides acute knowledge for sea ice classification. However, these datasets are limited to narrow spatial swaths due to power system constraints, restricting the possibility of operational exploitation of the data. Compact polarimetric SAR is the compromise between dual-polarized and fully polarized SAR, reducing the complexity of system design and maintenance, and enlarging the imaging breadth of the system (500 km in some imaging modes). Thus, information content and spatial resolution of compact polarized data is highly effective for sea ice type detection. Compact polarimetric features associated with complex sea ice backscatter signatures present the ability, under variable sea ice phenology, to discriminate between smooth newly-formed ice and calm open water. Statistical analysis of compact polarimetric parameters provides useful discriminatory criteria between the different ice types, but the discrimination capabilities vary strongly with sea ice conditions, sea ice phenological regimes, and incidence angle.
The present study focuses on assessing simulated compact polarimetric data for sea ice type detection and classification, under different sea ice conditions. Several synthetic aperture radar scenes, acquired in quick succession over the same area with shallow and steep incidence angles, are compared. Radarsat-2 scenes over the Canadian Archipelago were simulated using the Radarsat Constellation Mission (RCM) simulator, and 32 compact polarimetric parameters were calculated for each scene. During post processing, an average Refined Lee filter was applied as a compromise between the number of pixels needed for parameter estimation and resolution degradation for the medium resolution mode (MR 50m). Subsequently, homogeneous regions of interest (ROIs) were selected over the sea ice extent and statistically analyzed. Immobile landfast ice predominated the acquired scenes, hence the selected ROIs were separated into two categories, rough and smooth sea ice. Boxplots representing the statistical variability of the samples for each category were created. Analysis of statistical separation between sea ice types and open water pairings was performed using a two-sample Kolmogorov-Smirnov test. Quantitative selection of the best parameters for each sea ice type and open water pairing was assessed for smooth and steep incidence angle ranges. Moreover, data reduction and correlation ranking methods were examined to determine which parameters provide salient information for sea ice categorization schemas. The examined compact polarimetric parameters demonstrated significant variability between the selected samples. Shallow incidence angle data provided a stronger degree of separability between the sea ice categories for the majority of the compact polarimetric parameters. Highly correlated decomposition parameters were observed; thus a refined group of compact polarimetric parameters for sea ice classification and analysis was created. Data comparison for variable incidence angle ranges presented valuable insights for compact polarimetric feature analysis and is a missing link for determining effective classification schemes for sea ice.

Area D - Board 446

Sea ice monitoring has attracted increasing attention over the last few decades. Besides the scientific interest in sea ice, the operational aspect of ice charting is becoming more important due to the growing navigational possibilities in an increasingly ice-free Arctic, especially over the North East and North West passages. For this purpose, space-borne SAR imagery has become an invaluable tool, in particular for its higher resolution compared to traditional passive sensors. Previously, mostly single polarimetric datasets were investigated in supervised and unsupervised classification schemes for sea ice investigation. Despite proven sea ice classification achievements on single polarimetric data, a fully automatic, general purpose classifier for single-pol data has not been established due to the large variation of sea ice manifestations, weather and incidence angle impacts. Recently, through the advent of polarimetric SAR sensors, polarimetric features have moved into the focus of ice classification research. The higher information content of two or four
channels promises to offer greater insight into sea ice details and overcome some of the shortcomings of single-polarimetric classifiers.
In parallel, deep learning techniques for image processing have developed rapidly. Convolutional Neural Networks (CNN) have now achieved human-level image recognition ability with minimal preprocessing. Once trained, CNN can automatically extract and weight the complex spatio-polarimetric features needed for SAR image classification. Instead of providing a single label for each input image, the problem of semantic segmentation is to classify images pixel-wise, to identify regions belonging to each predefined class. Here too, CNN based network architectures have proved very successful. Therefore, this work is intended to explore the potential of deep learning techniques and polarimetric SAR for the application of automated sea ice charting in Near Real Time (NRT).

Densely Connected Convolutional Networks (DenseNets) are a recent development in network architecture designed to improve accuracy and ease of training by introducing additional connections between otherwise non-consecutive layers. Skip connections between layers with common resolution, allow the network to analyse the image at multiple scales simultaneously. Reducing the burden of training is particularly relevant, given the very limited amount of labelled data available for this task. Ancillary data from national ice services, it in situ measurements from various Arctic campaigns and expert judgement are utilized to identify and verify the governing ice regimes.

Area D - Board 447

Northern Canadian residents are interested in having access to remotely sensed image data, enhanced image products, and ancillary information to help plan snow mobile travel and subsistence activities on sea ice. Sea ice surface roughness regimes contribute to on-ice trafficability ease and safety. We investigate a semi-empirical approach for quantifying information about winter sea ice surface roughness regimes utilizing data from the Sentinel-1 synthetic aperture radar (SAR) missions. We selected Sentinel-1 data, processed to minimize incidence angle effects, because microwave scattering over sea ice is responsive to sea ice surface roughness; images are available regardless of weather, cloud cover, and darkness; and the data are freely available. Sentinel-1-based roughness was assessed using two other spatiotemporally coincident roughness indices: 1) root-mean-square roughness (from elevation measurements) captured by high-resolution airborne LIDAR terrain data, collected in Victoria Strait in April 2017, and 2) a normalized difference angular index (NDAI) sensitive to surface roughness, collected by the optical Multi-angle Imaging SpectroRadiometer (MISR) on the North American Space Agency’s (NASA) Terra platform. Areas where the three roughness indices correlated and decorrelated were compared quantitatively. Sentinel-1-based roughness products were also assessed by local community sea ice users and experts for spatial relationships to trafficability. We found that community-scale roughness extracted from SAR was dependent on ice type, with surface scattering dominantly influencing smooth and deformed first year ice (FYI) regimes, and volumetric scattering significantly impacting results over multi-year sea ice (MYI) and land-based freshwater input areas. While Sentinel-1 backscatter had a stronger relationship (R2= 0.62) with the LIDAR root-mean-square roughness than the MISR NDAI (R2= 0.26), Sentinel-1 overpredicted sea ice roughness in areas of MYI more than the MISR NDAI. Areas with deformed FYI viewed by Sentinel-1 and MISR NDAI were highly correlated (R2= 0.88), while smooth FYI areas were not correlated but had low roughness values. Local sea ice experts found Sentinel-1 SAR images extremely accurate for assessing surface roughness regimes, although they anoted that snow modifications to roughness also decreased the accuracy Sentinel-1-based roughness. We are working to make Sentinel-1-based roughness products available to community sea ice users using Google’s Earth Engine platform.

Area D - Board 448

Active microwave remote sensing utilizing Synthetic Aperture Radar (SAR) has demonstrated its utility in detection of sea ice melt onset (MO) in the Arctic. Since the 1990s, C-band spaceborne SAR sensors have been used extensively for sea ice operational dynamic and thermodynamic monitoring including MO detection. The choice of C-band as the preferred frequency for sea ice monitoring was made in the 1980s when the Arctic sea ice regime was different (ie. MYI) from what it is today (ie. predominantly FYI). Given the longer wavelength of L-band SAR compared to C-band SAR, L-band can better separate FYI and MYI types given its longer penetration depth Thus, L-band SAR is considered as an optimal choice for sea ice monitoring in the new Arctic sea ice regime. However, L-band sea ice application research is limited, thereby providing a significant opportunity to use the novelty of L-band SAR for the development of enhanced sea ice monitoring strategies.

This study aims to detect sea ice MO in the Queen Elizabeth Islands (QEI) of the Canadian Arctic Archipelago region for both the first-year (FYI) and multi-year ice (MYI). Time series wide beam L-band SAR imagery from ALOS PALSAR from 2009 were utilized in this study. Moreover, C-band SAR imagery from RADARSAT-2 and Ku-band data from QuikSCAT were also used to compare MO timing among multi-frequency sensors. 30 first-year and 30 multi-year ice homogeneous sites (2x2 km) were chosen randomly to develop the MO detection algorithm. The Canadian ice service digital archive was used to get ice type information during site selection. A threshold approach from time series winter backscatter was used to detect MO in the QEI region. The algorithm detects MO on the day where the backscatter changes (more than ~3 dB) from the mean winter backscatter and the air temperature was more than -5⁰C. The air temperature data were acquired from APP-x (Extended AVHRR Polar Pathfinder) datasets. The seasonal evolution of microwave backscatter for L-band was found be opposite to the Ku- and C-band for MYI. The mean MO date for 2009 was on YD 162 (for FYI) and YD 164 (for MYI). L-band detected MO were within ±3 days of Ku- and C-band detected MO. The snow thickness on sea ice and progression of air temperature dictate the timing of MO detection from all sensors. However, data availability of L-band SAR imagery during winter to melt transition can also have an effect on the MO detection variation as daily L-band SAR imagery was not available during winter to spring transition.

This research provides much-needed baseline scientific information of L-band signatures and MO detection algorithm of the new Arctic sea ice regime, which will be invaluable with imminent the launch of new L-band sensors such as Satellites for Observation and Communications (SAOCOM) and NASA-ISRO Synthetic Aperture Radar (NISAR).

Area D - Board 449

We developed a robust method of deriving sea ice extent in Arctic with high spatial resolution (~2 km), which is now fully operationally working for C-band spaceborne SAR, including the Chinese Gaofen-3 and Sentinel-1A/1B.
The method is basically based on a machine learning method, i.e., Support Vector Machine (SVM). The key point of the SVM method of discriminating open water and sea ice from SAR images is that appropriate samples should be chosen to train it. We noticed that sea ice presenting in spaceborne SAR images are highly variable, due to different types of sea ice have different radar backscatter characteristics. On the other hand, along with changing of sea surface wind speed and direction, sea surface also present highly variable radar backscatter in SAR images. Therefore, it is not possible to choose fixed samples of sea ice and open water in advance and use them to train a "fixed" SVM. Based on the radar backscatter of sea ice and open water have distinct characteristics in the cross-polarized (HV or VH) SAR images, we developed an automatic method of selecting samples of sea ice and open water for each SAR images, which are used for training the SVM, and then is consequently applied for the whole SAR image to discriminate sea ice and open water. Seven textures are derived from the Gray Level Co-occurrence Matrix (GLCM) of the cross-polarized SAR data are used in the SVM. Among the seven textures of GLCM, entropy and energy of sea ice and open water naturally have opposite characteristics, and therefore, are used to select corresponding samples for training the SVM.
While the method is applied for the Sentinel-1A/1B Extra-Wide (EW) Swath data, we encountered a few problems. Scalloping effect still exist in the Sentinel-1 EW data, particularly in the HV/VH polarized data, though the TOPSAR technique is applied for processing of the IW and EW data. Second is the in-homogeneous feature of radar backscatter in the EW HV-polarized images, particularly in the first beam. Step-by-step, we partially solved the problems. So far, we are starting to process all the Sentinel-1A/1B EW data acquired in the Arctic to get sea-ice extent products for each summer season (July - September) starting from 2014.

Area D - Board 450

The Arctic marine environment is rapidly transitioning from thick multiyear sea ice coverage to an ocean that is primarily covered with first year sea ice types. Consequently, sea ice mobility and regional variability have changed dramatically in recent years. The transition comes with challenges and consequences that microwave remote sensing techniques can be used to understand. The increased mobility of sea ice in the Canadian Arctic suggests that the formation of leads may increase. Accordingly, transient phenomenon, such as frost flowers, would become more prevalent. Moreover, with longer summer shipping seasons resulting in increased shipping traffic, the risk of oil spills has garnered the attention of the public, government, industry, and local communities. Changes in sea ice conditions and properties are readily observable using active and passive microwave sensors, driving our research into this area.

Herein, we present an overview of progress in sea ice remote sensing experiments at the University of Manitoba and the connection to sea ice thermodynamic state, physical state, and formation processes. Recent work includes (i) studies on frost flowers and their physical and thermodynamic evolution, (ii) evaluating the effects of the introduction of oil and petroleum products to emulate oil-spill scenarios in cold laboratory experiments, (iii) investigation of the impact of sediments associated with freshwater marine fluxes, and (iv) novel electromagnetic model development for microwave scattering from sea ice and the linkage to experimental polarimetric radar measurements.

We perform annual experiments at an outdoor mesocosm, called the Sea-ice Environmental Research Facility (SERF). The facility includes a pool filled with artificial seawater, which is formulated to mimic surface Arctic Ocean water, and is outfitted with a suite of environmental sensors and instruments. We have conducted experiments on sea ice processes and growth under a variety of winter conditions. Recent experiments have included frost flower growth and the associated C-band radar response. In addition, the cold laboratory facilities within the Centre for Earth Observation Science (CEOS) allow us to conduct small-scale studies on how the introduction of oil into a sea ice system influences the thermodynamic, physical, and dielectric properties of sea ice. We also conduct extensive field studies of the electromagnetic properties of snow covered sea ice, freshwater sea ice, freshwater-marine coupling, and physical-biological coupling using various remote sensing tools. These projects are conducted onboard our research icebreaker (NGCC Amundsen), our new Coastal research vessel (MV William Kennedy), and at field sites in the Canadian Arctic, Greenland, and the Russian Arctic.

In conjunction with the experimental measurements of sea ice scattering and properties, we perform research in electromagnetic modeling studies. Recent developments include multilayer scattering models employing integral equation methods and numerical methods. The scattering models are developed to describe how sea ice physical properties, including interface roughness, brine content, and inhomogeneities, give rise to microwave scattering at L-, C-, and X-band. The developed models can be run on computing clusters for large-scale problems. The goal of this research is to develop an operational, user-friendly, electromagnetic scattering model that can be used for a variety of young sea ice conditions. Furthermore, we are working to incorporate the forward scattering models into inverse models to extract sea ice parameters.

Area D - Board 451

Young sea ice types, namely grease-pancake ice (GPI), and thin ice floes as well, mainly compose the marginal ice zone (MIZ) during the freezing period. Such types of sea ice are considered the primary source of the sea ice fringing the Antarctica continent during winter season through the process called the "pancake cycle", and are massively produced in the Arctic as a result of the sea ice extent and volume decline. The properties of such sea ice types are exceedingly difficult to measure autonomously, due to their size, transient nature, and the hostile environment where they form and grow. However, some internal properties of the ice cover, such as the thickness, can in principle be inferred by measuring the modifications induced by the presence of the sea ice cover on the propagation of gravity waves coming from the open ocean. In this regard, the available spaceborne SAR imagery can be operated as a synoptic wave buoy and measure the attenuation and the shortening of incoming ocean waves inside GPI fields.
The SAR inversion procedure is based on knowledge of the open sea wind-generated waves spectrum. It is essential that such waves come from the open ocean before traveling inside the GPI field. The mapping between ocean wave spectrum and SAR image spectrum is provided by the closed nonlinear integral transformation by Hasselmann and Hasselmann (Hasselmann and Hasselmann 1991) and later developments, which account for the SAR image cross-spectrum (Engen and Johnsen 1995). The wind wave spectrum is thus changed according to the rheological properties and to the thickness of the ice layer crossed in order to achieve the best fit between the observed SAR image spectrum and the simulated one. Finally, swell waves are estimated directly from the observed SAR image spectrum regardless of their provenience.
The ability of different viscous layer models to describe the attenuation of gravity waves propagating in GPI covered ocean has been already investigated (De Santi et al. 2018). In particular, the Keller’s model (Keller, 1998), the two-layer viscous model (De Carolis and Desiderio 2002) and the close-packing model (De Santi and Olla 2017) have been extensively validated by using wave attenuation data collected during two different field campaigns (Weddell Sea, Antarctica, April 2000; Western Arctic Ocean, autumn 2015). For thin GPI, good retrievals of ice thickness have been obtained by considering the ice layer as the only source affecting the wave dynamics, so that the wind input can be disregarded. Validation of these viscous layer models is now extended by inverting different ERS and Sentinel 1 SAR images acquired in the Arctic.

References
De Carolis G. and D. Desiderio (2002), Dispersion and attenuation of gravity waves in ice: A two-layer viscous fluid model with experimental data validation, Phys. Lett. A, 305, 399–412.
De Santi F., De Carolis G., Olla P., Doble M., Cheng S., Shen H.H., Wadhams P., Thomson J (2018) On the ocean wave attenuation rate in grease‐pancake ice, a comparison of viscous layer propagation models with field data, Journal of Geophysical Research: Oceans, 123, 5933-5948.
De Santi F. and Olla P. (2017), Effect of small floating disks on the propagation of gravity waves, fluid Dynamic Research, 49, 025512
Hasselmann K. and S. Hasselmann (1991), On the nonlinear mapping of an ocean wave spectrum into a synthetic aperture radar image spectrum and its inversion, J. Geophys. Res., vol. 96, no. C6, pp. 10713-10729, 1991.
Engen G. and H. Johnsen (1995), SAR-ocean wave inversion using image cross spectra, IEEE Trans. Geosci. Remote Sensing, vol. 33, pp. 1047–1056, 1995.

Area D - Board 452

The Curonian Lagoon is a shallow estuarine freshwater system in the southeastern part of the Baltic Sea. This water body is prone to freeze during the winter time, however it is lacking detailed exploration of spatial distribution of the ice cover throughout the whole lagoon area. To address it, here we use high-resolution synthetic aperture radar (SAR) measurements from three Earth observation missions: Envisat ASAR, RADARSAT-2, and Sentinel-1A and 1B.
In this study we analyze spatial and temporal variability of ice cover in the Curonian Lagoon over 15 winter seasons from 2002 to 2017. Dates of freeze-up, full freezing, final break-up, and last observation of ice were defined upon analysis of satellite observations. As found, only the final melt onset has a statistically significant decreasing trend. In general, all analyzed winters seasons exhibit full freezing of the lagoon, which usually occurs very fast. The average ice cover period in the Curonian Lagoon is 88 days (the shortest 43, the longest 134 days).
The ice cover starts to form all along the eastern shore of the lagoon and retreats from the western shore to the eastern. The break-up of the ice cover starts in the northern part due to the turbid nature of this area and saline water intrusion from the Baltic Sea, and resides longest in the eastern, south-eastern part of the lagoon where it drifts under prevailing westerly winds.
Overall, the satellite data corresponds well to in situ measurements during the long periods of cold air temperatures, and most of the inconsistencies arise during the melting periods. Freeze-up and break-up events correspond well to the fluctuations of air temperature. Initial ice cover formation is observed firstly on the ground and then a couple of days later in the SAR images. Due to the SAR image resolution, the full freezing is firstly apparent in satellite images, then on average a day later in the coastal stations. As ice break-up occurs further away from the coast it is firstly observed in the SAR images, which also show that the last ice traces survive on average 1 day longer than it is recorded at coastal stations.
The shortest ice cover seasons in the Curonian Lagoon, observed in winter 2006-2007 and 2007-2008, correspond to the first record-low minimum ice cover extent observed in the Arctic Ocean, which was observed on 18 September 2007. The analysis of the duration of the ice cover season in the Curonian Lagoon and North Atlantic Oscillation (NAO) winter index show a high negative correlation R=-0.74 between these two parameters. The decreasing of ice period corresponds to the increase in NAO winter index.
There were several occasions when several consecutive SAR images were available on the same day providing the opportunity to evaluate drift velocity of ice floes during the melting period.
The authors would like to thank Marine Research Department of the Environment Protection Agency of Lithuania for providing the data of ice cover ground observations, and European Space Agency for openly sharing the Sentinel-1A & 1B images, and DWH_MG1_CORE_11 dataset with RADARSAT images. Igor Kozlov acknowledges the support from Russian Science Foundation Grant no.17 77 30019.

Area D - Board 453

Arctic sea ice concentration (SIC) is a primary information for the prediction of climate change and the development of sea route in polar oceans. Passive microwave (PM) sensors such as Advanced Microwave Scanning Radiometer-2 (AMSR2) have observed microwave radiation properties of sea ice and open water and provided the SIC by using various algorithms since the 1970s. However, the SIC retrieval algorithms adopted for the PM observations show low accuracy in the Arctic summer. This is because the algorithms only use the PM observation data that could not reflect the unique microwave radiation characteristics of sea ice and open water in summer due to the effects of atmospheric water contents and ice surface melting by high temperature. In this study, SIC estimation models for the Arctic summer considering the atmospheric effects on the PM observations were developed by using the AMSR2 brightness temperatures, numerical weather prediction (NWP) data from the European Centre for Medium-Range Weather Forecasts (ECMWF) ERA-Interim reanalysis and Korea Multi-Purpose Satellite-5 (KOMPSAT-5) synthetic aperture radar (SAR) images. Four machine learning approaches―Decision Tree (DT), Random Forest (RF), Multi Layer Perceptron (MLP), and Convolution Neural Network (CNN)― were adopted for the development of the SIC estimation models. A total of 339 KOMPSAT-5 Enhanced Wide-swath SAR images in HH-polarization were acquired in the Pacific Arctic Ocean (East Siberian Sea, Chukchi Sea and Beaufort Sea) during the summer (July to September) from 2015 to 2017. We discriminated sea ice from open water in the KOMPSAT-5 SAR images and computed 42,480 values (samples) of SIC within 10 km × 10 km window. Eighty percent of the KOMPSAT-5 SIC values were used as training dataset for the SIC estimation models and the remaining values were used as test dataset. The brightness temperatures measured at each channel of AMSR2 and their combinations and the atmospheric parameters (atmospheric water vapor, wind speed, sea level pressure, 2-m temperature, and 925 hPa-temperature) predicted by the ERA-Interim reanalysis were used as input variables for the SIC estimation. The SICs estimated from the machine learning approaches showed the value of bias less than 1% and the value of root mean square error (RMSE) less than 9% compared to the KOMPSAT-5 SAR SIC values. The values of bias and RMSE from the machine learning models are much smaller than those from the existing SIC retrieval algorithms for the AMSR2 such as Bootstrap algorithm (bias of 8.8% and RMSE of 21.7%) and ARTIST Sea Ice algorithm (bias of 6.7% and RMSE of 20.3%).

Area D - Board 454

Satellites at polar orbit enable daily and global coverage of the polar oceans at both hemispheres, providing an unique monitoring capability of monitoring sea ice.

Microwave sensors onboard satellites are commonly used for sea ice monitoring at high latitude : radiometers are routinely used for sea ice concentration estimate, and scatterometers have also shown they can contribute significantly to it. Since 1991, numerous scatterometers data at C and Ku-bands are available since ERS-1.
Backscatter data can be use for at least three applications : they enable to discriminate sea ice from open ocean areas, in particular scatterometers are useful to detect new ice, even at the early stage of growth, they can also be used for sea ice type detection (first year from multi-year sea ice in the Arctic), moreover, sea ice displacement maps, useful to follow dynamics of the sea ice during the winter, can be built.

The benefit of the combination of sensors and datasets for a long-term observation of the polar areas have been shown. Such data are developped and processed at IFREMER/CERSAT with easy data access and format, available for the scientific community. They provide an exceptional basis for analysis of long-term variations of the sea ice in both polar areas. They are used for different applications and are available through dedicated projects such as Copernicus CMEMS programs. They also will be part of the integrated Arctic Observation System of the INTAROS EU project : the overall objective is to build an efficient integrated system
by extending, improving and unifying existing systems in the different regions of the Arctic using databases of atmosphere, ocean, and cryosphere.

If possible, we will show new data from the recent ASCAT on MetOp-C and CFOSAT sensors, both launched late 2018, and how they can contribute to continue the long-term observation data time series for sea ice monitoring.

Area D - Board 455

Sea ice age (SIA) is one of the components of the essential climate variable (ECV) for sea ice as defined by the Global Climate Observing System (GCOS) (WMO, 2015) TS5 . It is an important climate change indicator in polar regions, which describes the sea ice cover state in addition to its concentration and thickness. More generally, the changes in SIA can be seen as a proxy of sea ice decline in Arctic.

A new algorithm for estimating sea ice age (SIA) distribution based on the Eulerian advection scheme was developed in the framework of the ESA Sea Ice Climate Change Initiative (SICCI) project. The advection scheme accounts for the observed divergence or convergence and freezing or melting of sea ice and predicts consequent generation or loss of new ice. The algorithm uses daily gridded sea ice drift and sea ice concentration products from the Ocean and Sea Ice Satellite Application Facility. The major advantage of the new algorithm is the ability to generate individual ice age fractions in each pixel of the output product or, in other words, to provide a frequency distribution of the ice age allowing to apply mean, median, weighted average or other statistical measures. Comparison with the National Snow and Ice Data Center SIA product revealed several improvements of the new SIA maps and time series. First, the application of the Eulerian scheme provides smooth distribution of the ice age parameters and prevents product undersampling which may occur when a Lagrangian tracking approach is used. Second, the utilization of the new sea ice drift product void of artifacts from EUMETSAT OSI SAF resulted in more accurate and reliable spatial distribution of ice age fractions. Third, the algorithm is also constrained by the observed sea ice concentration from OSI SAF, which reduces fractions of old ice and, consequently, ice age by 20–30 %. It was applied to generate time series of daily sea ice age fraction product from October 2012 to October 2017. Comparisons with the NSIDC SIA time series indicate that the fractions of MYI in the new product melt faster during the year and after a spin-off CE2 time of 3 years the area of older ice in the SICCI product is almost 20 % lower than in the NSIDC product. MYI concentration is computed as a sum of all MYI fractions and compares well to the MYI products based on passive and active microwave and SAR products.

The figure shows step-wise improvements of the NSIDC algorithms:
A) NSIDC ice drift, NSIDC advection algorithm
B) SICCI sea ice drift, NSIDC advection algorithm
C) SICCI sea ice drift, SICCI advection algorithm
D) SICCI sea ice drift, SICCI advection algorithm, accounting for sea ice concentration

Area D - Board 456

Much of the published CryoSat-2 (CS-2) sea ice thickness retrievals concentrate on average thicknesses over a certain area and period. For most of the current products, this means monthly averages of a 625 km^2 grid cell. This is in stark contrast with the spatial resolution requirement of 0.1 m - 100 m if thickness estimates are to be used in studying ice induced loads on ships. Basically, sea ice processes relevant for ice loads happen in ship scale. In this study we take steps towards providing estimates of ice conditions using CryoSat-2 along track products.

Average sea ice seldom poses a risk to ice navigation. Average conditions are observed by the crew which is well capable of adjusting the operations, usually by decreasing speed, to control risk. However, thick ice floes within an area of comparably low sea ice concentration and thickness are potentially dangerous. Thus here we concentrate on the capability of CS-2 to detect the thickest ice present in a certain region.

We compare the along track Level 2 SIT product from the ESA Sea Ice CCI and daily distributions of draft measurements from the Beaufort Gyre ULS moorings in the Beaufort sea. The sea ice draft is calculated from the thickness and freeboard in the CCI product. We find good agreement between the thickest 5% (and 1%) percentiles of daily CS-2 and BGEP drafts. This agreement builds confidence that single tracks of CS-2 already available are meaningful for detection of potentially dangerous ice. We also present Arctic -wide maps of thick ice detections and trends over the CS-2 period from 2010 to present.

Our study lays foundation for future use of satellite radar altimetry within the ice navigation community. We shall discuss the current challenges, and offer potential solutions to meet the user requirements in the future.

Area D - Board 457

Surface temperature is one of the most important variables in characterizing the energy balance between surface and atmosphere. Land- and sea surface temperature (LST/SST) estimates from satellite-borne sensors have already received much attention, and they are commonly applied as boundary conditions in numerical weather prediction and climate models. Retrieval of surface temperatures over sea and land ice (IST), on the other hand, is a relatively new field with special challenges, one of which being discontinuities in temperature retrievals across surface types (ice and water), which complicates direct inclusion into models. In recognition of its importance and the difficulties in characterizing surface temperature across all surfaces of the Earth using satellite data, Copernicus has initiated a new project at EUMETSAT: Sea-ice surface temperature retrieval and validation for Copernicus Sentinel-3 Sea and Land Surface Temperature Radiometer.

This project develops and recommends retrieval algorithms for sea-ice and marginal ice zone temperatures for use with the Copernicus Sentinel3 Sea and Land Surface Temperature Radiometer (SLSTR). The SLSTR is a dual view conically scanning imaging radiometer with heritage from ATSR on ERS and AATSR on ENVISAT. There is special focus on bridging high-latitude SST and IST estimates to ensure minimal nonphysical discontinuities in the Marginal Ice Zone (MIZ) in retrieved surface temperature across water and ice surface types. There is also focus on exploiting the SLSTR dual view configuration and the advanced SLSTR technology in general, to achieve an IST product superior to existing IST product from e.g. Metop AVHRR, VIIRS and MODIS Infrared sensors. Data from half a year of Sentinel3-A and -B will be inter-compared and validated at the end of the project

One of the largest sources of uncertainty when mapping the sea ice surface temperature using infrared radiometers is the ability to detect clouds. Therefore, the study will apply the outputs and recommendations from a 2016 study on “Sea-ice cloud screening for Copernicus Sentinel3 SLSTR”, in order to provide optimal discrimination of cloud over sea-ice and marginal ice zone regions. In addition to the official cloud screening algorithm, a probabilistic cloud mask developed for existing Land Surface Temperature retrievals will be tested and inter-compared with the official cloud screening algorithm. Initial results will be shown at the conference.

Area D - Board 458

The Arctic sea-ice is a fundamental actor and witness to the on-going climate changes. The ice coverage acts as an insulating layer between the atmosphere and the underneath ocean as well as a shield to incident solar radiations that limits the warming of the ocean. Long term satellite optical series (1973-) show that the Arctic sea-ice extent is shrinking fast, triggering a positive feedback of ice melt. More recently, altimetric missions have shown that the sea-ice is also thinning, getting even more fragile to atmospheric and oceanic forcing. Moreover, it had been recently demonstrated that the assimilation of sea ice thickness improves considerably the sea ice concentration forecasts. These considerations motivate robust and reliable monitoring the sea ice thickness.
The SIRAL instrument aboard the polar mission CryoSat-2, endowed with the delay-Doppler mode, has enabled to produce several truthful estimations of the Arctic sea ice freeboard. Nevertheless, the ice thickness estimation using altimetry still holds several sources of uncertainties.
One major source of error is coming the algorithm that analyses the radar echoes to retrieve the heights: the retracker. Indeed, over sea ice, the interlinking of strongly specular echoes, over leads, and diffuse echoes, over floes, prevents to apply the well-known Brown model that allows to robustly measure the open ocean surfaces.
To overcome these difficulties several solutions have been developed. These solutions can be categorized within two mains class of retrackers: the heuristic threshold retrackers (ice1, OCOG, TFMRA), and the physical model based retrackers (SAMOSA+, Kurtz, IceNew). The heuristic retrackers are the most represented because they are simple to implement and are able to find out a solution what ever is the shape of the waveform. The physical retrackers are enhanced Brown models that can handle specular echoes thanks to an additional Mean Square Slope parameter that characterizes the specularity of the surface. Sea-ice products based on these physical retrackers are still few, as they require heavy fitting processing or simplifying assumptions. Above all, these various solutions present up to 100% of sea ice freeboard discrepancies [H.Sallila et al. TC-2018, D.Yi et al. IEEE-TGRS-2018].
Another source of error that is independent of the retracker solution is a consequence of two delay-Doppler aliasing effects that are encountered over surfaces with mixed up specular and diffusive echoes. The first one, that impacts both LRM and SAR altimetry, is the ‘off-nadir’ effect that occurred when a lateral scattering much stronger than the nadir scattering hooks the retracker aside. The second one is the ‘side-lobe’ effect, which is a detrimental side-effect inner to the SAR processing that considers undesired specular targets from synthetic antennas side-lobes leakage along the satellite track. These aliasing effects on sea-ice are responsible for substantial errors for all retrackers.
The very new two dimensional retracker (2D-retracker) presented here is able to overcome the aliasing effects in a very efficient and robust way. It relies on the flatness characteristic of sea-ice (+-50cm) to detect and extract a “ground reference line” from each along-track 2 dimensional waveform image. Side-lobe and off-nadir peaks have characteristic hyperbolic trademarks identifiable on the aligned along-track image of waveforms.
This very fast and efficient method allows to eliminate side-lobe and off-nadir effects and to provide an accurate classification of leads and floes. Its drawbacks which is common to all heuristic retrackers is its high sensibility on the waveform peak position [Laforge & Fleury 2018] and on the choice of the empirical percentage of waveform for the retracking. This limitation can be overcome in conjunction with a physical retracker such as SAMOSA+: the first guest obtained by the 2D retracker can feed SAMOSA+ and help to converge more rapidly toward the correct solution.
This presentation will show the results and the performances of the 2D retracker alone, and some first demonstrations in combination with SAMOSA+.

. S. Dinardo, L. Fenoglio-Marc, C. Buchhaupt, M. Becker, R. Scharroo, M. J. Fernandes, and J.Benveniste, “Coastal sar and plrm altimetry in german bight and west baltic sea,” Advances in Space Research, 2017.
K. Guerreiro, S. Fleury, E. Zakharova, A. Kouraev, F. Re ´my, and P. Maisongrande, “Comparison of cryosat-2 and envisat radar freeboard over arctic sea ice: toward an improved envisat freeboard retrieval,” The Cryosphere, vol. 11, no. 5, pp. 2059–2073, 2017.
V. Helm, A. Humbert, and H. Miller, “Elevation and elevation change of greenland and antarctica derived from cryosat-2,” The Cryosphere, vol. 8, no. 4, pp. 1539–1559, 2014.
A.Laforge, S.Fleury, S. Dinardo, K. Guerreiro, F. Remy, F. Garnier, J. Verley. Inter-comparison of SAR processing options for sea-ice freeboard retrieval. Submitted ti IEEE GSRS 2018.
C. Ray, C. Martin-Puig, M. P. Clarizia, G. Ruffini, S. Dinardo, C. Gom- menginger, and J. Benveniste, “Sar altimeter backscattered waveform model,” IEEE Transactions on Geoscience and Remote Sensing, vol. 53, no. 2, pp. 911–919, 2015.
H. Sallila, J. McCurry, S. L. Farrell and E. Rinne. Assessment of Contemporary Satellite Sea Ice Thickness Products for Arctic Sea Ice. TC-Discuss 2018.
D. Yi , N. Kurtz, J. Harbeck, R. Kwok, S. Hendricks, and R. Ricker. Comparing Coincident Elevation and Freeboard From IceBridge and Five Different CryoSat-2 Retrackers. IEEE-TGRS-2018.

Area D - Board 459

Due to the remoteness and inaccessibility of the Arctic Ocean, satellite remote sensing is the primary tool for observations in this region. Although optical remote sensing is temporally limited by the seasonal availability of sunlight and frequent cloud cover, it is the only method that enables the retrieval of some geophysical parameters such as surface albedo or chlorophyll concentration. Coarse resolution satellite missions such as AVHRR and MODIS are already used in sea ice research; higher spatial resolution missions, however, still play a minor role. The reasons are the limited spatial and temporal coverage and sensor radiometric resolution of previous missions, e.g. in case of Landsat 7 ETM+. In our paper, we aim to show that ESA’s Sentinel-2 (S-2) mission has the potential to support sea ice research. The spatial coverage reaches up to ~ 83° north, covering wide parts of the Arctic Ocean and adjacent seas. Due to the operation of two satellites shifted by 180° on the same orbit, revisit time reduces to five days. Due to the overlapping of swaths in the far North, however, up to one image per day over a certain area of interest may be available.

The spatial, spectral and radiometric setting of the S-2 mission enables observations of sea ice parameters on a new level. The high spatial resolution allows detailed mapping of sea ice features such as melt ponds, ridges and leads, which can exist on the scale of meters to tens of meters. It has the potential to bridge spatial scales from in situ sampling and very-high-resolution imagery to coarser resolution satellite missions such as Sentinel-3.
Currently, geographical coverage is limited to areas close to the shoreline, leaving large areas of the Arctic Ocean unrecorded. Yet, geographical coverage may be extended on request. With this paper, we emphasize the potential of the MultiSpectral Instrument on board of S-2A and S-2B to support Arctic research. One major challenge in the processing of the data, however, is a proper atmospheric correction (AC). We therefore present an inter-comparison of AC processors based on S-2 data and in situ measurements acquired during RV Polarstern cruise PS106 cruise in June/July 2017.

Area D - Board 460

Melt pond in the Arctic plays a significant role in controlling Earth’s radiation balance. The fraction of melt pond is increasing, stimulating ice-albedo feedback. In order to understand the role of melt pond in the Earth climate system, spatiotemporally continuous Pan-Arctic melt pond data is needed. Here, we retrieved Pan-Arctic melt pond using spatiotemporally coincident MODIS and high resolution World view satellite data. The world view is obtained from March to October 2015, which considers the various spatiotemporal spectral characteristics of melt pond in the Arctic. We use world view based melt pond classification results according to Nicholas and Polashenski (2018). The melt pond and sea ice class for the MODIS were determined based on world view classification results using a majority voting. As the MODIS is an optical sensor, the reflectance is considerably affected by the geometry of the sun and a sensor. Especially, the sun elevation is low in the Arctic. The band ratio is used to minimize this effect. The band ratio of melt pond and sea ice is consistent regardless of the geometry of the sun and a sensor. Currently, MOD09 product that previous studies have used for retrieving melt pond is not properly atmospherically corrected in the Arctic atmosphere. We used atmospheric profile around Arctic coastal from radiosonde data for the atmospheric correction. The MODIS L2 data are atmospherically corrected via a 6s program in this study. The melt pond and sea ice class from MODIS based on world view classification in 2015 are used for the training set. For the machine learning approaches such as random forest, deep learning, and support vector machine are used to retrieve melt pond and sea ice in melting season from 2000 to 2017. The remaining 20% of world view data are used as a validation.

Area D - Board 461

Sea ice plays a significant role in energy and water balance of climate systems. In addition, sea ice is an important factor that measure global warming and on-going climate changes. Thus, the prediction of sea ice concentration (SIC) is necessary for various applications such as climate forecasting, polar ecosystem change prediction, and development of arctic shipping routes. However, many sea ice prediction approaches based on either numerical or statistical models show weakness to estimate extreme (anomalous) cases, especially in the sea ice melting season. Since such anomalies are significant near sea ice edge areas, this study tried to develop a deep learning based model in order to improve prediction accuracy. The objectives of this study is: (1) to develop a more accurate monthly SIC prediction model, (2) to modify the prediction errors near sea ice edge area considering spatial patterns as well as inter-connection of estimators, and (3) to conduct quantitative and qualitative analysis between two approaches: deep learning and machine learning approaches. This study used monthly mean SIC and other predictors from NOAA Optimum Interpolation (OI) Dataset, Moderate Resolution Imaging Spectroradiometer (MODIS)/Terra, and European Centre for Medium-Range Weather Forecasts (ECMWF) ERA-interim re-analysis data. The spatial coverage of this study is 60° - 90° (S – N) / -180° - 180° (W - E) with 10km x 10km grids, and the temporal coverage is June – September in 2002 – 2017 (sea ice melting season). A total of seven variables—(1) Sea ice concentration (SIC, %), (2) ice surface temperature (IST, K), (3) forecast albedo (FAL, %), (4) 2-m air temperature (T2M, Celsius degree), (5) u-wind vector (u10, m/s), (6) v-wind vector (v10, m/s), and (7) sea surface temperature (SST, K) —were selected as input variables to predict monthly SIC. This study developed the 3D Convolutional Neural Networks (3D CNN) model, which can consider spatial pattern as well as inter-connection of variables, and Random Forest (RF) model, which is widely used machine learning model, in order to build a baseline model. Results show that the CNN model produced good performance resulting in average RMSE = 2.43% in compare with RF model (9.87%) during melting season (June-September). Each month shows RMSE by follows: June (3.58%), July (2.45), August (2.02%), and September (1.66%). In addition, the model produces relatively good prediction accuracy in case of anomalous cases (Figure): June 2006 (3.62%) and August 2012 (2.05%) even near sea ice edge area. This study is expected to contribute in many fields such as extends of knowledge and understanding regarding arctic sea ice changes and provide a good reference for scholars in polar ecology and climate.

Area D - Board 462

At the vicinity of the ice edge dark absorbing open ocean waters and highly reflective sea-ice and snow result in high spatial and temporal heterogeneity at the surface. Continuous estimations of the surface albedo in such rapidly changing environments are thus essential to quantify how much solar radiation is reflected back into the atmosphere from the surface. The Earth-observation community often derives surface albedo from single or multiple directional quantities measured by nadir and off-nadir field, airborne, or satellite sensors. Therefore, field measurements of the anisotropy of the reflectance of the targeted surface are essential to aid in improving algorithms to estimate the surface albedo from directional reflectance measurements. In situ surface reflectance anisotropy measurements are also of great interest for vicarious calibration of airborne and satellite sensors and atmospheric correction algorithms, among others, to better quantity the total solar flux reflected by the area surrounding the target and diffusely transmitted to the sensor, the so-called adjacency effect. Adjacency effects are particularly acute for satellite ocean color remote sensing in polar regions, where the concurrence of bright scattering ice and dark absorbing waters generate significant inaccuracies in the retrieved ocean color satellite products.

The detailed reflectance properties of a surface are generally described by the bidirectional reflectance distribution function (BRDF), defined as the reflectance of an incident light flux from an infinitesimal narrow incoming illumination beam at one direction in the hemisphere into another direction in the hemisphere. However BRDF cannot be measured on the field and is often approximated by the Hemispherical-Directional Reflectance Factor (HDRF). Here, we introduce a new method to determine the HDRF of sea ice and snow at spatial scales from 1 m^2 to 80 m^2 using a multispectral circular fish-eye radiance camera (CE600). The CE600 allows measuring radiance simultaneously in all directions of a hemisphere at a 1° angular resolution. The spectral characteristics of the reflectance and its dependency on illumination conditions obtained from the camera are compared to those obtained with a hyperspectral field spectroradiometer manufactured by Analytical Spectral Device, Inc. (ASD). Results confirm the potential of the CE600, with the suggested measurement setup and data processing, to measure commensurable sea ice and snow HDRF values. Compared to the ASD, the reflectance anisotropy measured with the CE600 provides much higher resolution in terms of directional reflectance (N =16, 020). The hyperangular resolution allows detecting features that were overlooked using the ASD due to its limited number of measurement angles ( N = 25). This data set of HDRF further documents variations in the anisotropy of the reflectance of snow and ice with the geometry of observation and illumination conditions and its spectral and spatial scale dependency. Finally, in order to reproduce the hyperangular CE600 reflectance measurements over the entire 400–900 nm spectral range, a regression-based method is proposed to combine the ASD and CE600 measurements. Results confirm that both instruments may be used in synergy to construct a hyperangular and hyperspectral snow and ice reflectance anisotropy data set.

Area D - Board 463

The 2019-2020 Multidisciplinary drifting Observatory for the Study of Arctic Climate (MOSAiC) year-round expedition into the Central Arctic will provide a unique opportunity to critically investigate the sensitivity of ESA's CryoSat-2 (Ku-band) and CNES-ISRO's SARAL/AltiKa (Ka-band) radar penetration to seasonally varying snow geophysical and thermodynamic properties. This is essential in order to improve current sea ice thickness retrievals using these two satellite platforms. Current assumptions are that CryoSat-2 radar returns originate from the snow-ice interface at Ku-band, and from the snow-air interface at Ka-band. However, the complex nature of snow cover properties on sea ice has been found to critically impact on Ka- and Ku-band radar penetration through snow cover. To address these issues, we will acquire a benchmark in situ dataset year-long during the MOSAiC expedition. The project will involve acquiring high-resolution surface-based radar observations from snow-covered seasonal and perennial sea ice together with near-coincident detailed measurements of snow and sea ice geophysical properties. This will help us better characterize and validate how the snow and sea ice properties from both seasonal and older types influence Ka- and Ku-band radar penetration, frequencies currently used in currently operational satellite-based radar altimeter missions. These observations will be complemented by radiative transfer modeling to link snow properties, such as snow temperature, salinity, density and grain microstructure to the radar returns, with the aim to improve pan-Arctic sea ice thickness retrievals from CryoSat-2 and AltiKa. A secondary objective will be to investigate how blending two radar altimeters (e.g. Ku- and Ka-band), or blending laser and radar altimeters (e.g. CryoSat-2 and ICESat-2) can be used to improve our understanding of snow depth variability over sea ice.

Area D - Board 464

We present simple algorithms to derive key sea ice parameters such as the snow depth on sea ice, the snow-ice interface temperature and the effective temperature of sea ice at microwave window frequencies typically used by spaceborne passive sensors like AMSR-E and AMSR2. These parameters are important for understanding the sea ice variability, to improve the sea ice concentration mapping, and for polar atmosphere sounding applications. The methods to estimate these parameters are so far very limited.
Passive microwave observations are currently used for sea ice monitoring (e.g. sea ice concentration, sea ice thickness, snow depth on sea ice). At low microwave frequencies, the sensitivity to atmospheric parameters like water vapor or cloud liquid water is low, and it is possible to derive sea ice parameters due to the penetration of microwaves into the snow and ice layers. Sea ice thermodynamic is controlled by the regional heat budget. The snow on sea ice strongly affects the sea ice energy and radiation balance, with its high heat insulation and reflectivity of solar radiation. In winter, snow reduces the heat transfer from the ocean to the cold atmosphere and thus reduces the ice growth, and in summer, its high albedo limits the sea ice melting rate. The temperature gradient in the snow and ice also modulates the snow metamorphism and the sea ice growth and melt. Improved estimates of snow depth, as well as snow-ice interface temperature from satellite observations would provide valuable information on the vertical thermodynamic processes in the snow and ice, allowing to improve current sea ice circulation models and therefore the prediction of sea ice variability.
For an accurate estimation of the temperature profile in the lower atmosphere, quantifying the surface contribution to satellite-observed microwave signal is required. The surface contribution depends on the frequency, and it is the product of the surface emissivity and the surface effective temperature, the latter being defined as the temperature integrated over a layer corresponding to the penetration depth of the given frequency. The relationship between the effective temperature and the physical temperature profile is complicated, but it has been shown that from 6 to 50 GHz, it is possible to derive the effective temperature from the snow-ice interface temperature.
The algorithms to retrieve snow depth and the snow-ice interface temperature are developed using the Round Robin Data Package of the ESA sea ice Climate Change Initiative project. It contains brightness temperatures observed by the Advanced Microwave Scanning Radiometer 2 (AMSR2) collocated with measurements from Ice Mass Balance (IMB) buoys and the NASA Operation Ice Bridge (OIB) airborne campaigns over the closed Arctic sea ice pack. For the estimation of the effective temperature, we use simulations from the sea ice version of the Microwave Emission Model of Layered Snowpacks (MEMLS) and a thermodynamic model to relate physical snow and ice properties to brightness temperatures and effective temperatures of sea ice.
Brightness temperatures of several channel combination have been used for linear regressions to estimate the different parameters. The snow depth on sea ice is estimated with an root mean square error of 5 cm using a multilinear regression with the brightness temperatures at 6.9, 18.7, and 36.5 GHz in vertical polarization. The snow-ice interface temperature is retrieved using a linear regression as a function of the snow depth and the brightness temperatures at 10.65 or 6.9 GHz in vertical polarization, with a root mean square error of 1.7 and 1.9 K respectively. Finally, the effective temperatures between 6 and 89 GHz are expressed as a function of the snow-ice interface temperature with a root mean square errors of less than 1 K.
Maps of the snow depth and of the snow-ice interface temperature during winter are produced with the algorithms and compared to multiyear ice concentration maps from the University of Bremen. The results are spatially and temporally consistent, with snow depth larger north of Greenland and with estimates compatible with the seasonal cycle. Our snow-ice interface temperature estimates show a high correlation with multiyear ice concentration estimates obtained with independent algorithms.
These algorithms are easy to use. They can be used to create snow depth and snow-ice interface temperature data products. They can improve the study of sea ice variability by constraining the sea ice temperature gradient and the thermodynamical ice growth in sea ice models. The effective temperature estimations may be useful for atmospheric radiative transfer calculations and to reduce noise in sea ice concentration retrieval algorithms. So far, these retrieval algorithms are applied to AMSR2 observations. They could be used for the future observations by the Copernicus Imaging Microwave Radiometer (CIMR) which is one of the high priority candidate missions for the extension of the Sentinel satellite series.

Area D - Board 465

The recent proliferation of next-generation dual-polarimetric and fully-polarimetric Synthetic Aperture Radars (SAR) such as Sentinel-1, Radarsat-2, ALOS-PALSAR-2, and COSMO-SkyMed, has stimulated research into expanding its utility as a tool for providing reliable estimates of Antarctic sea ice extent, sea ice type identification and discrimination, ice-open water discrimination, ice floe tracking, demarcating ridged versus level ice areas, and as an information source for geophysical, climatological, and sea ice biogeochemical studies. However, characterizing the temporal variations snow geophysical and thermodynamic state on Antarctic sea ice from SAR remote sensing remains poorly understood owing to dynamic snow thermophysical changes and related atmosphere-snow/sea ice interactions at the high-resolution in-situ or plot-scale. Plot-scale studies are essential for understanding the intricate ‘detailed’ spatiotemporal behaviour of various thermophysical properties and processes controlling the microwave backscatter, crucial to accurately estimate critical snow and sea ice state variables such as snow and sea ice thickness. With the abovementioned satellite-based SAR sensors exhibiting a wide range of spatial and coarse temporal resolutions and swath width configurations, it is difficult to implement satellite-based SAR datasets to provide detailed and characteristic microwave signature variations with dynamic fluctuations in snow thermophysical properties over diurnal timescales.

This study presents diurnal-scale time series of high-resolution, surface-based microwave backscatter measurements acquired at C-band (5.52 GHz), to examine the thermophysical and atmospheric controls on C-band microwave backscatter from two environmentally distinct snow-covered sea ice regions in Antarctica, during the transition phase from late winter to melt onset season. The unique surface-based C-band polarimetric microwave scatterometer acquired measurements from homogenous sea ice targets in the Weddell and the Bellinghausen Seas, during the 2010 IceBell field campaign. This specific study is focused on using the diurnal scale microwave measurements from the scatterometer to investigate the influence of snow cover layering and sea water flooding on sea ice microwave signatures.

Diurnal scale fully-polarimetric measurements from the Weddell Sea were acquired on 15th November 2010 (between 1200 and 2345 UTC) and from the Bellinghausen Sea on 27th December 2010 (1500 and 2230 UTC). Sea ice in the Weddell Sea was characterized by relatively thin snow covers on thick sea ice, with the air temperatures exhibiting diurnal cycling throughout the day, above 0⁰C and snow temperatures close to the melting point. In contrast, sea ice in the Bellinghausen Sea was characterized by thicker snow covers on thinner sea ice, with the thicker snow loading depressing the sea ice freeboards. Snow covers were found to have thick saline slush layers due to seawater flooding. Air temperatures did not exhibit diurnal cycling and remained consistent around -4⁰C.

Scatterometer observations clearly demonstrate distinctive differences in microwave signatures from both Weddell and Bellinghausen Seas. C-band signatures from the Weddell Sea shows distinct cycling in microwave scattering mechanisms, near-coincident with cyclic diurnal changes in atmospheric forcing. At steeper incidence angles, melt and refreeze cycles within the snow surface, show clear fluctuations in microwave signatures, with surface scattering from the moist snow surface assumed to be the dominant scattering mechanism. On the other hand, microwave signature from thick snow-covered sea ice in the Bellinghausen Sea do not show any cyclic behaviour in microwave interactions, and predominantly exhibits the majority of the backscatter from within the cold volume, which does not seem to exhibit any thermodynamic changes during the diurnal time period. With support from in situ measured snow properties, near coincident with scatterometer measurements, our study clearly identifies distinctive variations in diurnal-scale microwave signatures from two different snow-covered sea ice regimes.

Preliminary results highlight the sensitivity of C-band microwave backscatter to subtle diurnal changes in air temperatures and snow thermophysical properties, which is difficult to investigate using satellite-based SARs. Our results provide an indication of the potential of using this surface-based scatterometer system to determine the thermophysical and electrical state of snow-covered Antarctic sea ice types.

Area D - Board 466

Currently operating satellite altimeter datasets from Ku-band altimeters (such as ERS-1/2, ENVISAT RA-2 and CryoSat-2), has demonstrated its ability to provide extensive basin-scale temporal and spatial measurements of Arctic sea ice freeboard and thickness. It has been theorized that Ku-band microwaves attain complete penetration through dry, cold and homogeneous snow, unless the snow cover exhibits highly dense compacted snow layers and/or ice lenses. Based on this theory, the main radar scattering horizon of Ku-band altimetric returns are assumed to originate from the snow/sea ice interface. However, recent studies has proven that, snow salinity, which occurs on first-year sea ice, reduces the penetration depth of Ku-band microwaves, resulting in a vertical upward shift in the main scattering horizon of Ku-band altimetric returns by several centimeters above the snow/sea ice interface. This vertical upward shift leads to an overestimate of Ku-band altimeter derived sea ice freeboard retrievals, resulting in significant overestimates of sea ice thickness.

In this study, we developed a semi-empirical approach to quantify the snow salinity, snow temperature and snow density induced uncertainty on Ku-band altimeter derived Arctic sea ice freeboard and thickness measurements. Using early- to late-winter high-resolution field measurements of snow salinity, temperature and density from several sites and environmental conditions in the Canadian and the Norwegian Arctic from 1993 to 2018, we illustrate the critical impact of snow brine volume (a modeled dependent function of snow salinity, temperature and density) on Ku-band radar scattering horizon and penetration depth. We utilize 10 naturally occurring saline/non-saline, warm/cold and simple/compleexly-layered snow covers ranging between 1 cm to 40 cm from the Canadian and the Norwegian Arctic, overlying sea ice covers with thicknesses ranging between 40 cm to 1.8 m. These snow covered sea ice case studies used in this study will replicate current and future early- to late-winter snow covered Arctic sea ice conditions, and their impact on Ku-band altimeter derived sea ice freeboard and thickness estimates.

Through our simulations, we find that the highest sea ice thickness retrieval errors arise from warm, highly saline snow covers (greater brine volume) overlaying thinner ice conditions; these errors decrease with an increase in sea ice thickness and when snow covers are colder and fresh/less-saline (lower brine volume). Our results also demonstrate the impact of metamorphosed, complexly-layered snow cover types, with their highly dense, compacted wind slabs, saline ice lenses and crusts, significantly affecting the average Ku-band signal velocity through the snow cover, resulting in enhanced Ku-band altimeter range retrieval errors, critically impacting the accuracy of sea ice freeboard and thickness estimates.

With a warming Arctic experiencing diverse spatial and temporal patterns of snow accumulation on sea ice such as frequent rain-on-snow events, larger snow accumulation over thinner ice types etc; resulting impact of complex snow properties on accurate estimates of sea ice thickness measurements from radar altimeters will be greater and more crucial. Our study, therefore, recommends to the radar altimetry research and operational community to incorporate brine volume as a function of snow density, salinity and temperature as a critical error sources along with error sources previously documented, affecting Ku-band altimeter derived Arctic sea ice freeboard and thickness measurements.

Area D - Board 467

Much effort has been dedicated to retrieving surface snow and sea ice characteristics from radar altimeter returns over the last decades. In particular the development of SAR altimetry with CryoSat-2 has resulted in new sea ice thickness products all the way up to 88 degrees South and North. More recently, combining radar freeboard retrievals at different frequencies has also shown promise for snow thickness retrievals. Beyond snow and ice thickness analysing the radar echo shapes can provide information on the surface characteristics at the sub-footprint resolution (i.e. significant wave height over open ocean or surface roughness over sea ice).

Traditionally sea ice processing from radar altimetry has relied on empirical algorithms (such as OCOG and TFMRA) that retrack the range between the satellite and dominant scattering surface. While over open ocean physical retrackers that are fitting a model of the radar echoe to the observation have been widely used, their application over sea ice surfaces has proven difficult because of the mixed surface characteristics (partially ice, partially lead) and because of the complex surface topography and backscatter over sea ice covered oceans.

Here we explore new techniques to improve on both empirical and physical retrackers of sea ice (and snow) surface characteristics. First, by combining radar freeboard estimates from satellite altimeters operating at different frequencies it is possible to write a system of equation that can be inverted for the retrieval of snow and sea ice thickness jointly. We describe our inverse modelling methodology and compare our results to other existing estimates of snow and sea ice thickness both from satellite processing and airborne measurements.

Second, we build on a recently developed numerical model designed to simulate delay-Doppler SAR (Synthetic Aperture Radar) altimeter echoes from snow-covered sea ice, such as those detected by Cryosat-2, to create sea-ice topographies based upon simulated ones with this model. Using a deep learning methodology, a recurrent neural network will be used, in order to encapsulate dynamic behaviour within subparts of echoes to ultimately simulate real sea-ice topography.

Area D - Board 468

Snow on sea ice represents a large source of uncertainty in altimetric retrievals of sea ice properties. In particular, the snow depth and backscatter properties of the snow layer can contribute to errors in retrievals of sea ice freeboard and thickness from both laser and radar altimeters. Knowledge of the pan-Arctic snow depth distribution could help to improve sea ice retrievals and enhance regional and global climate models. Here, we utilize a CryoSat-2 waveform fitting algorithm to retrieve the snow depth on Arctic sea ice. This waveform fitting method employs a physical model that accounts for scattering from the snow layer atop sea ice, in addition to scattering from the sea ice, and returns both the air-snow and snow-ice interfaces as a function of radar return time. Snow depth can then be computed by subtracting the elevation of the snow-ice interface from that of the air-snow interface. This method shows strong agreement with snow depth measurements made by the NASA Operation IceBridge FMCW Snow Radar over the Arctic.

We are able to calculate sea ice thickness and draft by combining these CryoSat-2 snow depth values with measurements of ice freeboard and estimates of snow, sea ice, and sea water density. Additionally, we combine these CryoSat-2 snow depth data with ICESat-2 freeboard data to test a joint satellite laser-radar altimeter retrieval of sea ice thickness. Draft and thickness calculations made here are compared with mooring data from the Beaufort Gyre Exploration Project (BGEP) and airborne data from the AWI EM-Bird system, respectively. The comparison with BGEP data allows for seasonal trends in sea ice draft from 2011-present to be assessed at three point locations in the Beaufort Sea, while the comparison with EM Bird data allows for an evaluation of the derived sea ice thickness distribution along various flight lines in the Arctic.

Area D - Board 469