Sea Ice classification from Synthetic Aperture Radar (SAR) is a popular subject of ongoing scientific discussion. The opportunity of obtaining automatically generated ice charts at high resolution, regardless of season or atmospheric conditions draws a large interest from the scientific community. To that end a large number of retrieval algorithms have already been presented and more are in development still, as data driven models are in the forefront of today’s scientific community and industry. In this state of heightened interest, development of new viable techniques is rapid and the situation is highly dynamic. In contrast to this ongoing drive for better and better automated solutions, the data being used to develop these models is largely unchanged. Most ice type retrieval algorithms are based on manually labelled data, like operational ice charts. One thing that all data driven solutions have in common however, is that an algorithm can never outperform the data it was trained on. This is precisely why we want to train on real measured physical properties – rather than manually labelled data, which contains errors and biases introduced by the person annotating the data. Due to the hostile environmental conditions, gathering in-situ ground truth measurements from the arctic pack ice masses presents a great challenge. This challenge was recently tackled by the MOSAiC campaign, wherein the RV Polarstern spent close to a year in the arctic pack ice, with a vast array of measurement instruments on board and on the ice. In this abstract we present a preliminary study, wherein ice types were extracted from Airborne Laser Scanner (ALS) measurements of the sea ice topography.
The simplest parameter to separate ice classes is their thickness. Whilst the ice thickness was directly measured by drilling out ice cores throughout the campaign, this data is not comprehensive enough to train an automated retrieval algorithm. However, the freeboard of the ice is linearly related to the ice thickness. Hence, topographical measurements of the ice are the closest we can get to the thickness information at a larger spatial scope. The freeboard was in this case be obtained from helicopter-borne ALS that operated frequently throughout the mission.
Because the measurements were acquired at different times, the drift of arctic pack ice needs to be adjusted for. To do so, we are using the GPS information of two sensors, located at the bow and the helm of RV Polarstern. Each point of ALS data is thus mapped to a corresponding pixel in the TerraSAR-X acquisition. The helicopter data has a resolution of half a meter, whilst the high-resolution SAR acquisitions have a three-and-a-half-meter pixel spacing. Accordingly, on average around forty-nine points of freeboard measurement will be mapped to each SAR pixel. This allows an insight into the surface roughness and can also serve as a measure of certainty regarding the freeboard in this pixel.
From this matched data labels can be generated via thresholding. The one thing left to account for is the depth of the snow, which obscures the ice thickness information. One possibility to work around this, can be found in the distribution of the freeboard measurements. In the available scenes a multimodal distribution in the data can be identified. Because the snow is almost uniform across different ice classes, splitting according to these modes allows separation of classes, that is hardly affected by snow depth.
Using these techniques, we can build a comprehensive multi-seasonal ice type map over the course of the mosaic mission. Critically the labels are directly derived from measured physical properties of the sea ice. We will use this data to train an algorithm capable of extrapolating these measured types to the full scope of the TerraSAR-X data.
In contrast to manually labelled data, where labelled ice types are at least different enough for a human to distinguish them, we might find ice classes, that are not able to be differentiated by humans. Here it will be particularly interesting to see how the algorithm performs.

Accurate observations of sea ice are crucial for improving the understanding of Arctic climate change and the predictive skills of coupled climate models. For sea ice related parameters like sea ice thickness, sea ice concentration, or snow on sea ice, satellites are the only tools to obtain consistent, long-term and Arctic-wide observations. Most of the used satellite sensors operate in the microwave domain of the electromagnetic spectrum because microwaves can penetrate clouds and observations are independent of sunlight. Due to the complexity of the snow-sea ice system, processes like, e.g., warm-air intrusions, melt-refreeze cycles and surface melt, snow metamorphism, or flooding of ice floes have a strong impact on the satellite observations and increase the uncertainty of the derived sea ice parameters.
Here, an analysis of the year-long ground-based microwave radiometer observations during the MOSAiC expedition is presented. The research icebreaker Polarstern drifted from October 2019 to September 2020 with the sea ice in the Central Arctic. Several microwave radiometers pointing towards the snow and ice were operated on the ice floe next to Polarstern. Here, we show results from different instruments covering microwave frequencies ranging from 1.4 GHz to 89 GHz. These frequencies are typically used in satellite-based sea ice remote sensing algorithms. During the measuring period, several atmospheric events like e.g., storms or warm air intrusions caused changes in the snow and ice cover observed by the instrument, altering its microwave signature. Some of these events are also visible in the satellite observations and impact satellite-based sea ice products, despite their large footprints of several square kilometers.
We analyze, how the seasonal changes of the snow and ice properties impact the satellite retrievals for sea ice concentration, sea ice thickness, and snow depth on sea ice and how the ground-based radiometer observations can be used to improve the quality of the retrievals and better estimate the uncertainty of satellite-based sea ice products.

Arctic rain-on-snow (ROS) can increase liquid water in the snowpack and upon refreezing, form icy crusts at the surface or within the snowpack. By altering radar backscatter and microwave emissivity, ROS over sea ice can influence the accuracy of sea ice variables retrieved from satellite radar altimetry, scatterometers, and passive microwave radiometers. While ROS detection algorithms have been developed for land, sea ice areas have been largely ignored. During MOSAiC, there was a unique opportunity to observe a ROS event using in situ instruments similar to those deployed on satellite platforms. This event occurred in September 2020. During liquid water accumulation in the snowpack, there was a four-fold decrease in radar energy returned at Ku- and Ka-bands. After the snowpack refroze and ice layers formed, this decrease was followed by a six-fold increase in returned energy. Microwave emissivity at 19 and 89 GHz increased with increasing liquid water content and decreased as the snowpack refroze, yet subsequent ice layers increased the polarization difference. Our analysis provides the first detailed in situ analysis of the impacts of both ROS and subsequent refreezing on both active and passive microwave observations, providing important baseline knowledge for detecting ROS over sea ice and understanding their impacts on satellite-derived sea ice variables.

In the polar oceans in winter, fractures and leads are the hotspots of exchange between the ocean and atmosphere which are otherwise well separated by sea ice. By altering the heat, gas, and momentum fluxes they play a crucial role in atmospheric, ecological, and oceanic processes. At the same time, leads represent a part of the present state of strain of the ice cover, opening up the possibility to study ice rheology. The transient nature of leads and their narrow appearance has set limits to the detection of leads from satellites. Different approaches using active and passive sensors from the microwave and infrared spectrum are employed so far to observe leads by means of satellite data. They make use of the strong contrast between leads and the surrounding ice pack in (i) surface temperature, (ii) microwave backscatter, (iii) emission or (iv) a change in ice drift speed.
With the increasing availability of high-resolution SAR data for the Arctic, we explored the potential to use SAR derived sea ice deformation to estimate lead fractions. We calculated sea ice drift and divergence with a spatial resolution of 1.4 km from daily Sentinel-1 scenes. We obtained the divergence-based lead fraction of a region by summing up all positive divergence pixels multiplied by the respective time step length. We derived a second lead fraction product from the deformation fields that calculates the position of linear kinematic features (LKFs) first. The advantage is a skilled noise reduction, and a tracking algorithm of the deformation zones.
We compared divergence- and LKF-based lead fractions to several other established lead fraction products in the Transpolar Drift along the drift track of the Multidisciplinary drifting Observatory for the Study of Arctic Climate (MOSAiC) between October 2019 to April 2020. We used lead fractions from helicopter-borne infrared surveys at a grid resolution of 5 m, classified Sentinel-1 (SAR) scenes at 80 m, MODIS (thermal infrared) at 1 km, AMSR2 (passive microwaves) at 3.25 km, and CryoSat-2 (altimeter in Ku-band) at 12.5 km. Since the methods rely on different physical properties of the water and ice in leads and are affected by different constraints, derived mean lead fractions vary by 1-2 magnitudes between the products. For example, infrared, SAR and microwave radiometer-based algorithms do not only detect open-water leads but also leads with thin ice up to a certain thickness, which differs between the products. Common lead events were identified across products. The time series mostly indicated a phase of increased lead activity during freeze-up in autumn 2019 and spring 2020. We used the different lead fraction time series to estimate new ice formation in the leads and compared the results to ice thickness and oceanographic measurements obtained during the MOSAiC campaign. Results yield lower and upper bounds for ice formation and brine expulsion in and from leads.
Due to the wide range of lead fractions obtained from different methods, we conclude that the specific lead fraction product must be chosen depending on research question. Divergence- and LKF-based lead fractions provide valuable information in addition to established lead fraction products at high spatial resolution and independent of cloud coverage.

We conducted thermal infrared imaging during 35 helicopter survey flights during the Multidisciplinary drifting Observatory for the Study of Arctic Climate (MOSAiC) from October 2019 to April 2020, during which the icebreaker Polarstern drifted with the sea ice in the Central Arctic. Surface temperature maps at a high resolution of 1 m are derived for a local, 5 km, and regional, 30 km, scale by merging more than 150,000 thermal infrared images. We analyze the temporal and spatial variability of the surface temperatures throughout the whole winter. While in winter the snow-covered sea ice appears very cold, thin ice and open water are significantly warmer and dominate the heat exchange between the ocean, ice, and atmosphere. Thus, in winter small cracks and leads act as windows for increased heat exchange from the warmer ocean to the colder atmosphere. These features are becoming increasingly present in a warming Arctic when the ice is thinning, moves faster, and breaks up more easily. With thermal infrared observations, we can characterize surface types based on the surface temperature of snow, sea ice, and ocean water surfaces. In our study, we developed a lead classification scheme based on an adaptive surface temperature threshold method. This allows us to identify single leads, determine the lead area fraction and lead properties like width and orientation. Results show that the lead width distribution follows a power law in agreement with previous studies; i.e. there are many narrow leads and a decreasing number of wider leads. This supports our assumption that there are many features on a small scale which can not be fully resolved by thermal infrared satellites with a spatial resolution of 1 km or lower. We use a surface temperatures dataset from the Moderate Resolution Imaging Spectroradiometer (MODIS) and investigate the sub-footprint scale variability and validate the satellite surface temperature based on our high resolution data. From this, we want to learn how well the small scale surface heterogeneity is represented in the satellite signal. For example, a continuous area of thin ice yields the same average surface temperature within the satellite footprint as a situation of dominantly thick ice with a few open water leads. However, the localization of ocean-ice-atmosphere exchange processes and amount of heat and gas fluxes would be very different for the two situations. The high resolution helicopter borne surface temperature data can support regional process studies to better analyze the influence of sea ice surface heterogeneity on polar climate processes.

The occurrence of leads represents a key feature of the sea-ice covered oceans. Leads promote the flux of sensible and latent heat from the ocean to the cold winter atmosphere and are thereby crucial for the air-sea-ice-ocean interaction. We use the warm signature of leads in thermal-infrared satellite data during winter months to infer daily lead maps for the sea ice areas in both hemispheres, 2002-2021. Based on this data set we compute a climatology of wintertime leads that indicates distinct spatial patterns of the predominant fracture zones. This opens up a completely new observational perspective on the influence of the bathymetry and associated ocean currents on the spatial variability of the sea-ice compactness. We showcase the most dominant spatial features in the computed lead climatologies for the Arctic and Antarctic and discuss the findings in context with model studies of ocean current velocities and eddy kinetic energies.