An automated procedure to map breaking river ice with C-band HH SAR data

The development of effective strategies to manage the river ice breakup process or the associated risks is hindered by a lack of understanding and information. Radar earth observation satellites offer excellent potential for collecting up-to-date information on the conditions of and changes in river ice cover during the breakup period. This text describes the development, performance and limitations of an automated procedure to map breaking river ice by means of C-band, HH-polarized Synthetic Aperture Radar (SAR) images. An original two-step supervised classification model (IceBC), which uses backscatter intensities, lies at the core of the procedure. First, IceBC identifies three primary classes: water, sheet ice and rubble ice. Next, each primary ice class is divided in three secondary classes that denote top surface roughness scale differences. Input images must have incidence angles from similar to 27 degrees to similar to 60 degrees. Below similar to 36 degrees, IceBC may assign a class labelled "unclassified" to water or sheet ice pixels. The primary classification model yields overall accuracies of similar to 86% and similar to 93% for independent test pixels with incidence angles <= similar to 49 degrees and >= similar to 29 degrees or >= similar to 36 degrees, respectively. The associated class accuracies for water, sheet ice and rubble ice are similar to 97% & 96%, similar to 69% & 85% and similar to 97% & 99%. Given its connection to ice jam flood events, the classification accuracy achieved for rubble ice is particularly important. Maps produced by means of IceBC comprise detailed spatial information regarding ice cover conditions and the development of the breakup process. Their quality may be affected by: freezing conditions, wet snow cover, meltwater pools, infrastructure, rapids or high winds. Monitoring is the key to managing the impacts of most of these challenges. An IceBC prototype has been used operationally since 2015.