New OSI SAF Southern Hemisphere sea-ice type product

Last year, we released for the first time a new product mapping the sea-ice type in the Southern Hemisphere. The ice type classification is provided daily in the Antarctic winter period, and March 1st is the first classification of a new winter season.

"Antarctic sea-ice type March 1, and sea-ice concentration at minimum February 24"

Figure 1: [Left] Sea-ice type in Southern Hemisphere, March 1, 2022. [Right] Sea-ice concentration around the minimum ice extent, February 24, 2022.

Below is the News story first published through EUMETSAT OSI SAF, December 20, 2021, when presenting the new product release. See also the related story on this year's minimum record of the Antarctic sea-ice extent in February


We present here a new product mapping the sea-ice type in the Southern Hemisphere. The product distinguishes between seasonal sea ice (first-year ice, or FYI) and sea ice that has survived at least one summer melt (multiyear ice, or MYI). The retrieval is based on the satellite passive microwave data from AMSR-2. In addition, a new correction scheme, based on sea-ice drift information is used to correct for erroneously classified sea ice. 

The new sea-ice type product is available at, and for the Southern Hemisphere, the data goes back to June 2021. If you would like to have a quick look at the sea-ice type product, you can visit the OSI SAF High-Latitude “Quicklooks” page and have a look at our sea-ice products at any (available) date.

"Sea-ice type in the Southern Hemisphere from March to August 2021"

Figure 2: Sea-ice type in the Southern Hemisphere from March to August 2021.


OSI SAF has since 2005 provided a Northern Hemisphere sea-ice type product, whereas a similar product for the Southern Hemisphere has been missing up until now. The Northern Hemisphere has the advantage (from a retrieval perspective) that the oldest sea ice is found off the northern coast of Greenland and the Canadian Archipelago. This relatively permanent ice cover of multiyear ice opens up the possibility to collect training data for each ice class from predefined target regions throughout the year. In the Southern Hemisphere, the situation is different. Here, the sea ice is located around the great continent of Antarctica and is affected by off-shore winds and strong circumpolar currents. The ice is forced to break up, moves dynamically, and changes location. 

In addition, Southern Hemisphere sea ice is typically younger than sea ice in the Northern Hemisphere, and so there is less differentiation in signature between the two ice types, especially later in the winter season. The effects of both younger and more dynamically moving multiyear ice make it more challenging to do an automatic classification.

The product 

The animation above shows the Antarctic winter season 2021 of the freezing of sea ice. At the beginning of March, the remaining ice since the yearly minimum is seen as multiyear ice (black), and most of this is found in the Weddell Sea along the Antarctic peninsula. While the ice cover is expanding as first-year ice (lighter blue), the area of multiyear ice is seen to be drifting north and eastward. As for the Northern Hemisphere, no classification is done during the summer months due to uncertainties arising from melting at the ice surface.

The daily ice type product is provided with an uncertainty field based on the classification probabilities, and a status flag giving information on the processing steps and which filters are applied. See an example of the different fields in the figure below from July 29, 2021. Between the categories of first-year ice (FYI) and multiyear ice (MYI) in the map, there is also an “ambiguous” ice category (mid-blue). The class of ambiguous is sea ice which does not have significant classification. Ambiguous ice is usually a mixture of classes, but can also be pure FYI or MYI.  The uncertainty is a value between 0 and 1, with 0 indicating high trust and 1 low trust in the statistical classification. The status flag provides information including regions of land, lakes, areas that never have sea ice in a climatological perspective, masking due to different correction schemes, or missing and unclassified pixels. The product files also contain a layer showing the number of satellite orbits used in the calculations. For more details on the product and its different fields, please consult the Product User Manual v3.1. The figure below includes for comparison the navigational ice chart, showing the Stage of Development, from Arctic and Antarctic Research Institute (AARI) for the same date.

Figure 3: Example of Southern Hemisphere Ice Type product. Top left: ice type field; Top right: status flag; Bottom left: uncertainty; Bottom right: AARI ice chart for comparison from the AARI-NIC-NMI pilot project on integrated sea ice analysis for Antarctic waters (brown is MYI, the yellow/green/purple colours represent different stages of young and FYI). 


The different physical properties of the different sea-ice types are utilized in the classification. For passive microwave radiometer, the signature dependency of the frequency differs between multiyear and first-year ice. Therefore, a common variable to use for ice type is the spectral gradient ratio “GR3719V”. In the Southern Hemisphere, static functions found by fitting this spectral gradient ratio with Gaussians (see figure below) for each month in a typical reference year are applied using the Bayesian method to obtain probability maps of sea-ice type.

Figure 4: Gaussian distribution fitted to the monthly histogram of GR3719V for the MYI (red) and FYI (blue) ice type (solid lines). The dashed line shows the probability distribution function used for FYI with the standard deviation set to match the larger standard deviation of the MYI to avoid weighting in favour of MYI.

For the Southern Hemisphere ice type retrieval a correction scheme based on sea-ice drift is included to mask unrealistic multiyear ice. By tracking an ice parcel back to the minimum ice extent at the end of February we see if the ice parcel survived the summer melt or if it has been created since then. This backtracking filter is crucial for the Southern Hemisphere where huge areas of pancake ice (rough first-year ice) forms and is misinterpreted as multiyear ice by the Bayesian approach. The area where this filter is activated is shown in the flag field, labelled in the product figure as “drift mask”.


The amount of in situ sea-ice type observations is very modest and not sufficient to be used for regular validations. Instead, the OSI SAF ice type is quality checked by monitoring the areal coverage of multiyear ice. By assuming that the multiyear ice does not have large and rapid changes over the winter season, the requirement is based on keeping the monthly variability low. This assumption of a slowly changing multiyear ice coverage is more applicable for the Northern Hemisphere than for the Southern Hemisphere where the sea ice - including the multiyear ice - is much more dynamic. Another difference we observe between the Northern and Southern Hemisphere is the behaviour of multiyear ice extent (MIE).

The Arctic MIE during winter mainly reduces by export out of the Arctic Ocean and therefore MIE is expected to decrease over the winter months. In the Southern Hemisphere, MIE is seen to increase for the recent winter seasons, see figure below showing the MIE in 2021. This can be explained by its more dynamic and divergent behaviour which causes ice-freezing in between multiyear ice floes which is not captured by the relative coarse AMSR-2 data.

The figure below shows the rolling average of the multiyear ice extent for 2021 (red line) with the rolling mean standard deviation (orange shaded area).

Figure 5: Multiyear ice extent, daily values during March-August 2021 (dots). Rolling average of the MYI extent (red line) with standard deviation (orange shaded area).

For more information see the Validation Report for the Global Sea-Ice Edge and Type Product .

Future Work

This is the first version of our Southern Hemisphere sea-ice type product. For upcoming versions we will investigate several potential improvements. Firstly, we will include more sensors (like scatterometer and other passive microwave radiometers) to make it a multi-sensor product as for the Northern Hemisphere products. Secondly, the utilization of ice drift information will be upgraded for a better masking of erroneously classified ice. With an improved backtracking methodology of sea-ice parcels, we also hope to reduce the summer gap in the product. Finally, we will investigate better ways to validate the product.

We welcome any feedback on the product from users. You can either send a request from the central OSI SAF web portal (registered users), or send a direct email to one of our dedicated contact points, see contact information here.


This story was first produced and released through EUMETSAT OSI SAF (20 December 2021). Visit the website for the latest news about OSI SAF :

You are welcome to register to access the catalogue of OSI-SAF products, get the latest info and be informed of live product availability.