Toward Automated Ice-Water Classification on Large Northern Lakes Using RADARSAT-2 Synthetic Aperture Radar Imagery

Abstract

Changes to ice cover on lakes throughout the northern landscape has been established as an indicator of climate change and variability. These changes are expected to have implications for both human and environmental systems. Additionally, monitoring lake ice cover is required to enable more reliable weather forecasting across lake-rich northern latitudes. Currently the Canadian Ice Service (CIS) monitors lakes using RADARSAT-2 SAR (synthetic aperture radar) and optical imagery through visual interpretation, with total lake ice cover reported weekly as a fraction out of ten. An automated method of classification would allow for more detailed records to be delivered operationally. In this research, the Iterative Region Growing using Semantics (IRGS) approach has been employed to perform ice-water classification on 61 RADARSAT-2 scenes of Great Bear Lake and Great Slave Lake over a three year period. This approach first locally segments homogeneous regions in an image, then merges similar regions into classes across the entire scene. These classes are manually labelled by the user, however automated labelling capability is currently in development. An accuracy assessment has been performed on the classification results, comparing outcomes with user-generated reference data as well as the CIS fraction reported at the time of image acquisition. The overall average accuracy of the IRGS method for this dataset is 92%, demonstrating the potential of this semi-automated method to provide detailed and reliable lake ice cover information.