Seasonal evolution of melt pond characteristics derived from Worldview and Quickbird imagery

Abstract:

Melt pond formation on the surface of sea ice is an integral component to the redistribution of solar energy driving summer sea ice loss. Libraries with a time series of sub-meter scale optical satellite imagery capture the summer evolution of these melt ponds. Using this imagery, we apply a semi-supervised algorithm to extract information on melt pond formation and evolution. The algorithm combines K-Nearest Neighbors segmentation and a machine learning random forest classification scheme to segment and classify Worldview and Quickbird imagery into categories of open water, snow and bare ice, and melt ponds. The partitioned imagery is processed to quantify melt pond coverage, size, and connectivity. We analyze the growing library of processed images to quantify seasonal progression of melt pond characteristics at many repeatedly imaged sites. We estimate the evolution of solar partitioning at these sites. These high resolution satellite images are used to derive melt pond statistics and solar partitioning, which are crucial for determining the strength of the sea-ice albedo feedback.