You must have a javascript-enabled browser and javacript and stylesheets must be enabled to use some of the functions on this site.
 Earthnet Home  
Sessions and Session Summaries
First Annoucement
Scientific Committee
ESA Specific Links
Organising Committee
Round Table Discussion Questions
Conference Photos
List of Participants
All papers


SAR surface ice cover discrimination using distribution matching

Rashpal S. Gill (1)

(1) Danish Meteorological Institute (DMI), Lyngbyvej 100, DK - 2100 Copenhagen , Denmark


Discrimination between open water and sea ice in SAR imagery can still pose a problem to the ice analysts in their daily task of charting the sea ice for safe navigation. To help them in this task, a new algorithm have been tested that rely on a user first manually identifying a particular region in a SAR image (e.g., open water area or sea ice of certain concentration or ice type) then the algorithm will automatically determine similar regions in the remainder of an image. The algorithm is based on matching the statistics of the known and unknown regions using either (a) Kolmogorov-Smirnov (KS), and (b) Chi-Square (CS) distribution matching tests. The main advantage in using these distribution matching tests is that the probability distribution function (pdf) of the known region does not need to be known. Both KS and CS tests determine whether the two data sets belong to the same or different, yet undetermined, distributions. The main difference between KS and CS tests is that they are valid for un-binned and binned data respectively. In this note the relative performance of the KS and CS tests is presented. The tests were carried out using the amplitude SAR image and the image products: (a) Power-to-Mean Ratio (PMR), and (b) Gamma-pdf which are computed from it. The results presented in this report shows that the KS test is reasonably successful at identifying similar surface types. It performed best with the amplitude data and Gamma-pdf while results using the Gamma-pdf and PMR images were prone to ambiguities. The CS test did not perform as well as the KS test. This is because the data first has to be arbitrarily binned which results in some information being lost. It was also found to be many times slower to run on the computer. The information obtained using the KS tests can be considered as the "best statistical guess" during situations when the ice analysts have difficulty in interpreting parts of a SAR image. Keyword: Sea ice, RADARSAT, image interpretation, distribution matching, Kolmogorov-Smirnov test, Chi-Square test, Greenland, Ice Charting.


Full paper


  Higher level                 Last modified: