Bulletin No. 1, 2021

Prof. Liu Lin Earth System Science Programme Keeping a Sharp Eye on Retreating Ice An AI model can now automate the delicate, costly task of delineating the fronts of retreating glaciers from satellite images, helping scientists to monitor changes in the cryosphere under climate change. Trained and tested on satellite photos of three glaciers in Greenland taken between 2002 and 2019, the deep learning model delivers promising results with deviations from manual delineations being as small as three pixels. (Background photo courtesy of Prof. Nicolaj Krog Larsen of the University of Copenhagen) 1. Cold data Greenlandic glaciers have been retreating much faster over the past 20 years. To understand how they react to different climatological and glaciological changes, scientists examine their satellite images and meticulously trace and track their calving fronts, the edges along which they collapse. 2. Preparations For the AI system to take up the task, the satellite images must first be cropped and augmented. Since they were captured by different satellites, which, for one, work on different colour bands, they also had to be standardized with some having to be denoised. calving front ice mélange 20