Over the past two decades, much of the Arctic has been mapped with extreme precision by commercial satellites. These maps are a treasure trove of data about this largely underexplored region. But the data is so large and unwieldy, it makes scholarship difficult, Witharana says.<\/p>\n With funding<\/a> and support from the U.S. National Science Foundation (NSF) as part of the \u201cNavigating the New Arctic\u201d program<\/a>, Witharana, as well as Kenton McHenry<\/a> from the National Center for Supercomputing Applications, and Arctic researcher Anna Liljedahl<\/a> of the Woodwell Climate Research Center, are making data about Arctic permafrost much more accessible.<\/p>\n The team was given free access to archives of over 1 million image scenes taken in the Arctic. That\u2019s a lot of data \u2014 so much that traditional analysis and features extraction methods failed. \u201cThat\u2019s where we brought in AI-based deep learning methods to process and analyze this large amount of data,\u201d Witharana says.<\/p>\n Shapes of the Arctic<\/strong><\/p>\n One of the most distinctive, and telling, features of permafrost are ice wedges, which produce recognizable polygons in satellite imagery. The shape and dimensions of ice wedge polygons can provide important information about the status and pace of change in the region. But they short-circuit conventional analysis.<\/p>\n \u201cI was on Facebook some years ago and noted that they were starting to use facial recognition software on photos,\u201d recalled Liljedahl. \u201cI wondered whether this could be applied to ice wedge polygons in the Arctic.\u201d<\/p>\n She contacted Witharana and McHenry, whom she had met at a panel review in Washington, D.C., and invited them to join her project idea. They each offered complimentary skills in domain expertise, code development, and big data management.<\/p>\n The team trained their model with annotated satellite imagery into a neural network and tested it on unclassified data. There were initial challenges, but three years later the team is seeing accuracy rates between 80 and 90%.<\/p>\n They have described the results of this research in the ISPRS Journal of Photogrammetry and Remote Sensing<\/em><\/a> (2020), the Journal of Imaging<\/em><\/a> (2020) and Remote Sensing<\/em><\/a> (2021).<\/p>\n Counting ice wedges with supercomputing<\/strong><\/p>\n After showing that their deep learning method worked, they turned to the Longhorn supercomputer, operated by the Texas Advanced Computing Center (TACC) \u2014 a GPU-based IBM system that can perform AI inference tasks rapidly \u2014 as well as the Bridges-2 system at Pittsburgh Supercomputing Center, allocated through the NSF-funded Extreme Science and Engineering Discovery Environment<\/a> (XSEDE), to analyze the data.<\/p>\n As of the end of 2021, the team had identified and mapped 1.2 billion ice wedge polygons in the satellite data. They estimate they are about halfway through the full dataset.<\/p>\n Each individual image analysis involves pre-processing, processing, and post-processing. In addition to identifying and classifying ice-wedge polygons, the method derives information about the size of the wedge, the size of the troughs, and other features.<\/p>\n Each individual analysis can be performed in less than an hour. But the sheer number makes it unfeasible to run anywhere but on a large supercomputer, where they can be computed in parallel.<\/p>\n![\"\"](\"https:\/\/today.uconn.edu\/wp-content\/uploads\/2022\/02\/To-Aaron_03-300x200.jpeg\")