Ice ridge density signatures in high resolution SAR images

• Main Authors: Lensu, Mikko Johannes, Similä, Markku Henrik
• Format: Text
• Language: English
• Published: 2021
• Subjects: Sea ice
• Online Access: https://doi.org/10.5194/tc-2021-346; https://tc.copernicus.org/preprints/tc-2021-346/

The statistics of ice ridging signatures was studied using a high (1.25 m) and a medium (20 m) resolution SAR image over the Baltic sea ice cover, acquired in 2016 and 2011, respectively. Ice surface profiles measured by a 2011 Baltic campaign was used as ground truth data for both. The images did not delineate well individual ridges as linear features. This was assigned to the random, intermittent occurrence of ridge rubble block arrangements with bright SAR return. Instead, the ridging signature was approached in terms of the density of bright pixels and relations with the corresponding surface profile quantity, ice ridge density, were studied. In order to apply discrete statistics, these densities were quantified by counting bright pixel numbers (BPN) in pixel blocks of side length L , and by counting ridge sail numbers (RSN) in profile segments of length L . The scale L is a variable parameter of the approach. The other variable parameter is the pixel intensity threshold defining bright pixels, equivalently bright pixel percentage (BPP), or the ridge sail height threshold used to select ridges from surface profiles, respectively. As a sliding image operation the BPN count resulted in enhanced ridging signature and better applicability of SAR in ice information production. A distribution model for BPN statistics was derived by considering how the BPN values change in BPP changes. The model was found to apply over wide range of values for BPP and L . The same distribution model was found to apply to RSN statistics. This reduces the problem of correspondence between the two density concepts to connections between the parameters of the respective distribution models. The correspondence was studied for the medium resolution image for which the 2011 surface data set has close temporal match. The comparison was done by estimating ridge rubble coverage in 1 km 2 squares from surface profile data and, on the other hand, assuming that the bright pixel density can be used as a proxy for ridge rubble coverage. Apart from a scaling factor, both were found to follow the presented distribution model.