Sea Ice Classification on Different Spatial Scales for Operational And Scientific Use
Alfred Wegener Institute for Polar and Marine Research, GERMANY
Satellite radar systems are essential for continuous monitoring of the Arctic and Antarctic sea ice cover and its spatial and temporal variations. Such data are required for providing regular information about ice conditions to users from shipping, fishery and offshore industries, from environmental protection agencies, and from the polar science community. The information is usually prepared as ice charts that show ice conditions on hemispherical and regional scales. The ice charts indicate spatial distribution and extent of distinct zones, typically characterized by a mixture of different criterions such as open water fraction, ice type, thickness, and state of deformation. For ice chart production, operational sea ice services preferably employ images acquired with synthetic aperture radar (SAR) operated in so-called wide-swath or ScanSAR-mode with swath widths between 200 and 500 km and spatial resolutions of several tens of meters. The reason is that - with recent satellite configurations- only these imaging modes allow the necessary spatial coverage and sufficiently frequent updates of changes in ice conditions.
With a spatial resolution of several tens of meters, typical ice characteristics such as single ridges, narrow brash ice belts between ice floes, narrow leads, or rafting zones cannot be directly identified in the wide-swath SAR images. These structural elements, however, change the backscattered radar intensity of the level ice areas, which indirectly indicates their existence. Nevertheless, it is not possible to quantify their areal fraction, in particular if different types of elements are located within one resolution cell of the SAR image.
In conjunction with more scientifically oriented studies also the capabilities of high-resolution images (spatial resolution better than 30 meters, in many cases even better than 10 meters) have been evaluated, using corresponding imaging modes of spaceborne SAR systems such as Envisat ASAR, TerraSAR-X, Radarsat-2, ALOS-PALSAR, or airborne data from systems such as the Danish EMISAR or the German ESAR. In some cases, complementary data (e. g. optical imagery and observation notes) are available from field campaigns. These high-resolution products make it possible to identify even small-scale structural elements on the ice and to increase the accuracy of ice type classification. One way to achieve a highly accurate classification is to combine images acquired at different polarizations and - in particular - at different frequencies. The latter case is rare since it requires data acquisitions from different satellites with only small temporal gaps.
This presentation addresses the combination of images of different spatial resolutions. Considering the fact that each pixel in a wide-swath SAR image may contain a mixture of different structural ice elements, the basic questions are: How can one quantify the information loss in coarse-resolution images compared to high-resolution reference data? What kind of information is typically lost? What are the optimal definitions for different ice classes seen at coarse spatial resolution? For an assessment of the information content of an image, different methods can be used. Selected methods will be discussed. Since the focus of the presentation is both on needs of the operational ice services and of the research community, another important question is which spatial resolution is required for a certain mapping task? The answers depend on the configuration of the SAR system (frequency, polarization, incidence angle), on the ice regime and on the season (melting versus freezing conditions). For the analysis, airborne and satellite SAR data are available that were acquired almost simultaneously, partly complemented by aerial optical imagery and airborne field observation. The data are from the region around Svalbard, Fram Strait, and Beaufort Sea (all Arctic) and the Weddell Sea (Antarctic).