Sea ice kinematics and deformation play an important role in determining the sea ice thickness distribution through the opening of leads and growth of new ice and by the formation of thick ice in pressure ridges. Leads generated by ice divergence or shear are responsible for a significant ocean-atmosphere heat flux, and brine rejection into the ocean mixed layer. Hence, observations of sea ice motion at time and spatial scales suitable for studying deformation are important to understand the evolution of the sea ice pack. SAR imagery collected by the Sentinel-1A and -1B satellites are processed using the Radarsat Geophysical Processor System (RGPS) to produce daily maps of sea ice motion (displacement) in the Central Arctic Ocean. The data are posted on a 10km stereographic grid, and in the area surveyed 50% of the region has valid ice motion data 75% of the time, and 30% of the region has valid ice motion data 90% of the time. We compute daily fields of ice deformation (divergence, shear, vorticity), and examine its seasonal cycle and variability over the course of a full annual growth/melt cycle. The quick repeat times of the Sentinel-1A/B imagery is ideal for studies of Arctic sea ice motion and deformation, and we anticipate they will play an important role in the remote sensing component of the upcoming MOSAiC cruise, due to embark during freeze-up in 2019.

Shipping and commercial activities in the Arctic are inevitably increasing as a results of the climate change, opening of the Northern sea route and oil exploitation. As a consequence, demand for better operational sea ice forecast in the coastal areas becomes crucial both for the climate/marine researchers and for the commercial users.
A new general circulation sea ice model neXtSIM (Rampal et al, 2016) has been developed at NERSC on the basis of the Maxwell elasto-brittle rheology framework (Dansereau et al, 2016) in order to simulate realistic sea ice conditions for regions as large as the whole Arctic and over time scales up to several years. The model equations are discretized with the finite-element method and a pure Lagrangian scheme has been implemented to preserve the discontinuities (highly localized features like cracks, leads and ridges) simulated by the model. Sea ice deformation simulated with this model spontaneously localizes along linear-like faults separating essentially undamaged ice plates/floes. The new sea ice model accounts for the local level of mechanically-induced damage of sea ice. Initialization of sea ice damage using observations would likely increase the model skill in terms of both sea ice drift and deformation forecasting. Satellite derived ice deformation can be used to estimate the local damage of the ice. The main idea is that ice damage in the model could be initialized with a higher value at locations where the sea ice cover has deformed during the previous couple of days.
Sea ice drift is derived from sequences of Sentinel-1 SAR images using the combined feature tracking and pattern matching algorithm (Korosov and Rampal, 2017) at unprecedented 5 km spatial resolution. For the first time drift is derived not on a regular grid but on a Lagrangian mesh of the neXtSIM model. Deformation is derived from the spatial derivatives of the sea ice drift at high spatial resolution. Anisotropic filtering is applied (Bouillon & Rampal, 2015) for suppressing small scale noise on the calculated sea ice deformation. The temporal scale of ice deformation (duration of ice drift integration) is varied for identification of persistent deformation zones. Ice damage is calculated as a function of sea ice deformation.
Improvements of the model skills for 5 days forecasts are estimated by comparison with satellite observations. Preliminary results show that accounting for sea ice damage in the model initialization may significantly improve the skill of neXtSIM not only to predict ice fracturing and leads/ridges location, but also to improve prediction of local and large scale sea ice drift.
The figure show sea ice deformation (shear) in the Lincoln sea calculated from sea ice drift at 2 km spatial resolution, derived from Sentinel-1 data acquired on 13 Feb 2018. Green arrow show wind speed from ECMWF forecast.

Safe passage in ice infested areas is important for the maritime industry. L-band synthetic aperture radar (SAR) images are not regularly used operationally though multiple studies have shown that L-band SAR can detect and separate deformed sea ice from the surrounding sea ice and provide easier separation between first year ice (FYI) and multi-year ice (MYI) in the early and advanced melt season than C-band SAR. Here we want to evaluate the suitability of L-band sensors with regard to Near Real Time (NRT) operational services in ice infested water and quantify the two major challenges, effects of incidence angle and seasonal variation.

We use 108 fully polarimetric ALOS-2 Palsar-2 images acquired during freezing and early melt conditions during the Norwegian Young Sea Ice Cruise 2015 (N-ICE2015) drift study and the Polarstern's (PS92) transitions in the Arctic Seasonal Sea Ice Zone. We use a supervised sea ice classification (neural network) to classify open water (OW), newly formed sea ice (NI), smooth FYI and deformed FYI and MYI within these scenes.

For training and validation of the neural network we use overlapping in-situ data from N-ICE2015 and PS92 to identify different sea ice areas and from this generate a training and validation dataset. The in-situ data consists of airborne electromagnetic (AEM) sea ice thickness measurements combined with photographs. The AEM has an accuracy of ±0.10m (Haas et. al., (2009), enabling easy separation of NI from surrounding thicker sea ice, though there are limitations in separating NI from OW. The photographs taken during the flights are therefore an essential part of the training area identification. The study area used here is located in the Arctic Ocean north of Svalbard, between 80°N to 84°N and 8°W to 25°E. The sea ice within the study areas was primarily a mixture of FYI and second year ice (SYI) though areas of thinner sea ice types such as nilas, young grey ice and young white ice were observed. Overlapping optical images from Landsat-8 show stages of new ice formation, grey ice and grey-white ice.

The SAR scenes were acquired with 5 different incidence angles from late March until early June where approximately 20 scenes acquired for each incidence angle. Each different incidence angle was used for a fixed number of days and are season specific. One training and validation dataset was therefore extracted for each incidence angles. Out of the 5 different incidence angles 3, where acquired during the freezing season and 2 of them during early melt season. The freezing season was defined as when the air temperatures are below -5°C and the snow cover is dry. Temperature records from N-ICE2015 show a range from -25°C to +2°C at the time of satellite image acquisitions. Until mid-May the temperatures were below -5°C, thereby ensuring that approximately half of our satellite images were acquired during the freezing season and half during the early melt season. This enables us to study not only incidence angle dependency but also seasonal changes for sea ice classifications using L-band SAR data.

The L-band SAR images used here are fully polarimetric and polarimetry has been shown to improve sea ice classification in C-band SAR (Dabboor and Geldsetzer (2014) and Geldsetzer et. al. (2015)). In addition, has studies by e.g. Wakabayashi et. al. (2004), Dabboor et. al. (2017) and Singha et. al. (2018) shown that for L-band SAR polarimetric parameters, such as entropy and anisotropy, are useful for separation of open water, thin sea ice and thicker sea ice. We extract 18 polarimetric features for each pixel and investigate the importance of different polarimetric parameters for sea ice classification and how the incidence angle affects those parameters. Earlier work by Mahmud et. al. (2018) and Wakabayashi et. al. (2004) among others have investigated the incidence angle dependency for different polarimetric parameters for different sea ice types for L-band SAR though not simultaneously with the seasonal changes. Here we classify the sea ice in two different ways, firstly with an incidence angle dependent training and validation dataset (self-training) and secondly, we train the algorithm with one incidence angle dataset and validate it using another incidence angle dataset (cross-training). This enables us to quantify the effects of incidence angle and seasonal variation.

For the self-training we found an accuracy of 92% or higher. In the cross-training results we observed seasonal variations, where cross-seasonal cross-training data achieves 51% to 82% accuracy. The highest values are observed when freezing season data is used to classify early melt season images (70% to 82%). When cross-training is used within season the accuracy is between 60% to 90%, where the early melt season cross-training has the smallest range of between 69% to 80% accuracy. From this we observe that for an automatic sea ice classification we need to address both the seasonality and the incidence angle.

Earlier studies have shown that L-band SAR has an advantage over C-band SAR for detection of deformed sea ice areas as well as is less sensitive to the onset of the melt season. L-band SAR can therefore be considered a useful complement to the ongoing C-band SAR missions and with the ongoing mission of ALOS-2 Palsar-2 and planned new missions such as SAOCOM (2018), ALOS-4 PALSAR-3 (2020) and NASA-ISRP's NISAR (2021) a continued availability of L-band SAR is foreseen.

References
Dabboor M., and T. Geldsetzer, Towards sea ice classification using simulated RADARSAT Constellation Mission compact polarimetric SAR imagery, Remote Sensing of Environment, 140, 189–195, 2014.

Dabboor M, B. Montpetit, S. Howell, and C. Haas, Improving sea ice Characterization in Dry Ice Winter Conditions Using Polarimetric Parameters from C- and L-Band SAR Data, Remote Sensing, 9(1270), 2017.

Geldsetzer T., M. Arkett, T. Zagon, F. Charbonneau, J. J. Yackel, and R. K. Scharien, All-Season Compact-Polarimetry C-band SAR Observations of Sea Ice, Canadian Journal of Remote Sensing, 41(5), 485–504, 2015.

Haas C., J. Lobach, S. Hendricks, L. Rabenstein, and A. Pfaffling, Helicopter-borne measurements of sea ice thickness, using a small and lightweight, digital EM system, J. Appl. Geophys., Airborne Geophys., 67, 234–241, 2009.

Mahmud M.S., T. Geldsetzer, S. E. L. Howell, J. J. Yackel, V. Nandan, and R. K. Scharien, Incidence Angle Dependence of HH-Polarized C- and L-Band Wintertime Backscatter over Arctic Sea Ice, IEEE Trans. Geosci. Remote Sens, 56(11), 2018.

Singha S., A. Johansson, N. Hughes, S. Munk Hvidegaard, and H. Skourup, Arctic Sea Ice Characterization using Spaceborne Fully Polarimetric L-, C- and X-band SAR with Validation by Airborne Measurements, IEEE T. Geoscience and Remote Sensing, 56(7), 3715–3734, 2018.

Wakabayashi H., T. Matsuoka, K. Nakamura, and F. Nishio, Polarimetric characteristics of sea ice in the Sea of Okhotsk observed by airborne L-band SAR, Geoscience and Remote Sensing, IEEE Transactions on, 42(11), 2412–2425, 2004.

Spaceborne synthetic aperture radar (SAR) high resolution (at or below 50 m) images from the ongoing satellite missions such as Canadian RADARSAT-2 are extensively used for routine monitoring of sea ice conditions. The upcoming Canadian RADARSAT constellation mission (RCM) equipped with three SAR platforms will further increase the amount of SAR data over the Arctic region. One of the key applications of the large amount of SAR images is assimilation of observations over sea ice into coupled ocean-sea ice numerical models, such as the Regional Ice-Ocean Prediction System (RIOPS) at the Environment and Climate Change Canada (ECCC). Therefore, there is a strong interest in making use of the large quantities of SAR observation in the sea ice analysis component of RIOPS.

Komarov and Buehner recently introduced a new approach for automated extraction of ice and open water from RADARSAT-2 HH-HV images with very high accuracy required for data assimilation [1], [2]. However, in situations where the algorithm detects ice, it is further desired to determine if this is first-year (FY) or multi-year (MY) ice. Despite the fact that the ice type information cannot be currently assimilated directly in RIOPS, this information is very important for assimilation of ice thickness provided by Cryosat-2 and SMOS satellites. This is associated with the fact that Cryosat-2 tends to underestimate freeboard over winter MY ice due to increased penetration of Ku-band signal which leads to significant underestimation of MY ice thickness. SMOS is able to detect ice thickness of thin ice only (up to ~50 cm); therefore, SMOS observations over MY ice need to be rejected using ice type information.

The main objective of this study is to develop a new approach for detecting FY and MY ice from RADARSAT-2 HH-HV images.

We collected more than 223,000 of MY ice samples and more than 1.6 million FY ice samples from all available RADARSAT-2 images and the corresponding Canadian Ice Service (CIS) Image Analysis products acquired over Canadian and adjacent waters monitored by CIS over the period from November 2010 to May 2016. Only ice samples identified as ice by the ice/water detection algorithm [1], [2] were selected. It is well known that FY/MY ice separation from C-band SAR is most effective under cold conditions due to strong volume scattering from MY ice [3], which was the case in our study as well. Therefore, the ice samples were collected under cold conditions only defined by a model-based melting-degree hour value calculated over past three days. Our analysis of the collected dataset revealed that FY/MY separation in winter can be effectively performed in the space of two selected SAR-based variables. A MY ice probability empirical model as a function of the selected predictor variables was built in a form of logistic regression, based on the training subset (all data with year 2013 excluded for testing). The developed approach showed very good performance on the independent testing subset (year 2013): 50% of the testing MY ice samples were classified with the accuracy of 98.1%, and 50% of the testing open water samples were classified with the accuracy of 98.3%. We also found that incidence angle correction improves FY/MY ice separability in a 1D space (for a given SAR parameter); however, such correction does not improve the FY/MY ice detection performance in the 2D space.

The developed FY/MY ice detection approach will be used to support assimilation of Cryosat-2 and SMOS ice thickness in ECCC RIOPS.

[1] A. S. Komarov, M. Buehner, “Automated detection of ice and open water from dual-polarization RADARSAT-2 images for data assimilation,” IEEE Trans. Geosci. Remote Sens., vol. 55, no. 10, pp. 5755-5769, October 2017.
[2] A. S. Komarov, M. Buehner, “Adaptive probability thresholding in automated ice and open water detection from RADARSAT-2 images,” IEEE Geosci. Remote Sens. Lett., vol. 15, no. 4, pp. 552-556, April 2018.
[3] R. G. Onstott, “5. SAR and scatterometer signatures of sea ice,” in Microwave Remote Sensing of Sea Ice, F. D. Carsey, Ed. Washington, DC, USA: AGU, pp. 73–104, 1992..

Sentinel-1A and 1B operate in Extra Wide swath dual-polarization mode over the Arctic Seas, and the two-satellite constellation provides the most frequent SAR observation of the Arctic sea ice ever. The cross-polarization is known to be more sensitive to difference in the scatter of sea ice and open water than the co-polarization, and the combination of HH- and HV-polarizations has been widely used for ice edge detection and ice type classification. However, the majority of the ice classification algorithms were developed using RADARSAT-2 ScanSAR images which has different sensor characteristics from Sentinel-1 TOPSAR, and the use of Sentinel-1 for the same purpose is very limited in literature. The main drawback of applying existing algorithms to Sentinel-1 data is the relatively higher level of thermal noise contamination and its propagation to image textures. We developed an efficient noise correction method by adopting a novel concept of noise equivalent standard deviation and compensating for the multiplicative textural noise along with the additive noise equivalent sigma nought. The denoised images are processed into Haralick texture features and a machine learning-based classifier is trained by feeding digitized ice charts, produced by various operational sea ice service centers. By using the existing ice charts as reference for training classifier, it is possible to automate producing SAR-based ice chart with less human effort and bias. The resulting classifier for five classes (nilas, young ice, first-year ice, old ice, and water) is tested over the Fram Strait area for an extensive dataset of Sentinel-1 acquired from October 2017 to May 2018. The classification results will be shown in comparison with the ice charts, and the feasibility of the ice chart-feeded automated classifier will be discussed. The implication of the proposed ice classification algorithm is to make automated SAR-based ice charts which can be utilized for generation of near real-time ice navigation map service, assimilation into a numerical ice model, and validation of the passive microwave radiometer-derived ice concentration and ice types.

Information on the timing of melt onset over Arctic sea ice is important for understanding the Arctic climate system. The daily temporal resolution of passive microwave brightness temperatures provide the most widely utilized observations to detect melt onset but are limited to a spatial resolution of ~25 km. Synthetic aperture radar (SAR) imagery provides a much higher spatial resolution (~50 m) but melt onset detection remains challenging because of i) insufficient temporal resolution to facilitate accurate melt onset detection, ii) incidence angle variability and iii) limited image availability across the Arctic. Here, we construct high temporal resolution multi-sensor backscatter products (1-day, 2-day and 4-day) using Sentinel-1 and RADARSAT-2 over a close-to-seamless revisit region located in northern Canadian Arctic and Greenland for estimating melt onset over Arctic sea ice in 2016 and 2017. We employ radiometric terrain flattening and local resolution weighting to generate normalised gamma nought (γ°) backscatter over the entire study region, not restricted to a single sensor or track's swath width limitations, integrating both ascending and descending passes into multi-sensor γ° products. Results indicate that higher temporal resolution γ° products that make use of the most imagery are less noisy and enhance the temporal stability of γ°, especially over first-year ice during the winter period. This in turn allows for more representative estimates of melt onset as it is easier to separate the melt onset threshold from winter γ° variability which, is a considerable challenge for SAR based melt onset algorithms because of temporal resolution inconsistency. Multi-sensor γ° melt onset detection is in good agreement with ASCAT and the Passive Microwave (PMW) algorithm over homogenous sea ice regions but very noticeable improvement was found within narrow channels and regions with more heterogeneous sea ice. In anticipation of the availability of data from even more SAR satellites with the launch of the RADARSAT Constellation Mission, the multi-sensor γ° approach presented here may offer the most robust approach to estimate the timing melt onset over sea ice across the Arctic.