Towards automated lake ice classification using dual polarization RADARSAT SAR imagery

Great Lakes;Lake ice;Remote sensing;Synthetic aperture radar (SAR);RADARSAT;Classification

Wang, Junqianaffiliation1:Faculty of Environment ; advisor:Duguay, Claude ; Duguay, Claude ;

University of Waterloo

关键词: Lake ice;    Master Thesis;    Classification;    RADARSAT;    Synthetic aperture radar (SAR);    Remote sensing;    Great Lakes;   

Others  :  https://uwspace.uwaterloo.ca/bitstream/10012/12880/3/Wang_Junqian.pdf

瑞士|英语

来源: UWSPACE Waterloo Institutional Repository

【 摘 要 】

Lake ice, as one of the most important component of the cryosphere, is a valuable indicator of climate change and variability. The Laurentian Great Lakes are the world’s largest supply of freshwater and their ice cover has a major impact on regional weather and climate, ship navigation, and public safety. Monitoring detailed ice conditions on large lakes requires the use of satellite-borne synthetic aperture radar (SAR) data that provide all-weather sensing capabilities, high resolution, and large spatial coverage. Ice experts at the Canadian Ice Service (CIS) have been manually producing operational Great Lakes image analysis charts based on visual interpretation of the SAR images. Ice services such as the CIS would greatly benefit from the availability of an automated or semi-automated SAR ice classification algorithm. We investigated the performance of the unsupervised segmentation algorithm ;;glocal” iterative region growing with semantics (IRGS) for lake ice classification using dual polarized RADARSAT-2 imagery. Here, the segmented classes with arbitrary labels are manually labelled based on visual interpretation. IRGS was tested on 26 RADARSAT-2 scenes acquired over Lake Erie during winter 2014, and the results were validated against the manually produced CIS image analysis charts. Analysis of various case studies indicated that the ;;glocal” IRGS algorithm can provide a reliable ice-water classification using dual polarized images with a high overall accuracy of 90.2%. The improvement of using dual-pol as opposed to single-pol images for ice-water discrimination was also demonstrated. For lake ice type classification, most thin ice types were effectively identified but thick and very thick lake ice were often confused due to the ambiguous relation between backscatter and ice types. Texture features could be included for further improvement. Overall, our ;;glocal” IRGS classification results are close to visual interpretation by ice analysts and would have expected to be closer if they could draw ice charts at a more detailed level.