Permafrost (ground that has been permanently frozen for more than two years) occupies most of the globe and occupies about 15% of the Northern Hemisphere.
Permafrost is important to our climate, contains large amounts of biomass stored as methane and carbon dioxide, and uses tundra soil as a carbon sink. However, the inherent and changing properties of permafrost are not widely understood.
As global warming Heating the Earth to cause soil thawing is expected to accelerate the carbon cycle of the permafrost layer, releasing greenhouse gases contained in the soil into the atmosphere and creating a feedback loop that exacerbates climate change. ..
Remote sensing is one way to understand the width, dynamics, and changes of the permafrost layer. Candi Witarana, an assistant professor of natural resources and the environment at the University of Connecticut, said: “Satellite imagery helps us monitor remote landscapes in unprecedented detail.”
Over the last two decades, much of the Arctic Circle has been very accurately mapped by commercial satellites. These maps are a treasure trove of data about this largely unexplored area. However, the data is so large and unwieldy that scholarships can be difficult, says Withalana.
With funding and support from the National Science Foundation (NSF), Witharana, Kenton McHenry of the National Center for Supercomputing Applications, and Anna Liljedahl Center, an Arctic researcher at Woodwell Climate Research, are permanent in the Arctic as part of the “Navigating the New Arctic” program. It makes the data on permafrost much more accessible.
The team had free access to an archive of over 1 million image scenes taken in the Arctic Circle. This is a large amount of data, and traditional analysis and feature extraction methods have failed. “Therefore, we introduced AI-based deep learning techniques to process and analyze this large amount of data,” says Witharana.
One of the most distinctive and comprehensible features of the permafrost is the ice wedge, which produces polygons recognizable in satellite imagery.
“The ice wedge is formed from the freezing and thawing of the tundra soil,” said Liljedahl. “Some of them are tens of thousands of years old.”
The shape and dimensions of the ice wedge polygon can provide important information about the situation and pace of change in the region. However, they short-circuit traditional analysis.
“I noticed that I was using Facebook a few years ago and started using facial recognition software for my photos,” recalls Liljedahl. “I was wondering if this could be applied to the Arctic ice wedge polygon.”
She contacted Witharana and McHenry, whom she met during a panel review in Washington, DC, and invited her to participate in her project ideas. They each provided complementary skills in domain expertise, code development, and big data management.
Starting in 2018, Witharana began using neural networks to extract polygons from thousands of Arctic satellite images rather than the faces of friends. To do this, Witharana and his team had to first annotate 50,000 individual polygons, then hand-draw their contours and classify them as either low-center or high-center.
The low center ice wedge polygon forms a pool in the center of the raised outer part. According to Rishdar, the high ice wedge in the center looks like a muffin, proof that the ice wedge is melting. The two types have different structural hydrological characteristics and it is important to understand them in terms of their role in climate change and to plan the future infrastructure of the Arctic community.
“Permafrost is not characterized by these spatial scales of the climate model,” says Liljedahl. “This study helps us to derive a baseline and see how changes are occurring over time.”
We trained the model with annotated images, fed satellite images into a neural network, and tested them with unannotated data. There was the first challenge. For example, images trained for Canada were less effective in Russia, where the ice wedges are old and different in shape. However, after three years, the team’s accuracy rate will be 80-90%.
They have the results of this study ISPRS Journal for Photogrammetry and Remote Sensing (2020), Journal of Imaging (2020) and Remote sensing (2021).
After showing that their deep learning methods worked, they Longhorn supercomputerTexas Advanced Computing Center (TACC), a GPU-based IBM system that can perform AI inference tasks quickly, Bridge-2 The Pittsburgh Supercomputing Center system assigned through the NSF-funded Extreme Science and Engineering Discovery Environment (XSEDE) to analyze the data.
As of the end of 2021, the team identified and mapped 1.2 billion ice wedge polygons in satellite data. They estimate that they are in about half of the complete dataset.
Individual image analysis includes pre-processing (to improve image sharpness and remove features of non-land areas such as lakes), processing (where polygons are detected and characterized), and post-processing (where the data is detected and characterized). (Reduce to manageable scale and upload) is included. To the data archive of the permafrost layer). This method obtains information about wedge size, valley size, and other features, as well as identification and classification of ice wedge polygons.
Individual analysis can be performed within an hour. However, the number of them is so large that it is impossible to run them anywhere other than a large supercomputer that can calculate in parallel.
Recently, Witharana and co-workers benchmarked workflows to find the optimal configuration to run efficiently on supercomputers. Written in Photogrammetry Engineering and Remote Sensing (PE & RS) in 2022, he evaluated four workflow designs on two different high-performance computing systems and found the optimal setup for high-speed analysis. Another 2022 study at PE & RS demonstrated the effectiveness of various image enhancement methods (such as changing hue and saturation) when applied to a deep learning convolutional neural net algorithm that recognizes ice wedge polygons from commercial satellite images. I investigated. (Both projects were presented at the American Geophysical Union Autumn Conference in December 2021.)
“Every year, we get a near real-time pulse rate monitor in the form of sea ice spreads in the Arctic,” Rishdar said. “We want to do the same with permafrost. There are a lot of rapid changes. We need to really understand and communicate what is happening with permafrost.”
Ice wedge data will be available for rapid analysis of new things Permafrost Discovery Gateway“We will make information about the Arctic Circle more accessible to more people,” Rishdar said. “You don’t have to wait 10 years to learn something, they can quickly learn about it and explore it directly through their own experience.”
Another important phase of the research project occurs when researchers analyze satellite images representing different years and times. Comparing the states of the ice wedge polygons can show trends and trajectories, such as the speed of change in the landscape and where those changes intersect villages and infrastructure.
“This is a perfect example of how previous investments in computing infrastructure and a new understanding of deep learning technology are building resources to help key issues in the Arctic,” said NSF Program Director. Kendra Klauklan says.
“Plato says,’A man must rise above the earth, above the atmosphere, and beyond. Only then will he fully understand the world in which he lives.'” I said, “said Withalana. “How do we with earth observation technology? Climate change What’s happening, and how even the land is changing. It is the primary tool for monitoring, monitoring, forecasting and making decisions to prevent adverse effects on vulnerable areas. ”
Optimal GeoAI workflow for detecting features of the Arctic permafrost from high-resolution satellite images Photogrammetry engineering and remote sensing (2022)
M. Udawalpola et al, Operational-scale GeoEye for feature detection of pan-Arctic permafrost from high-resolution satellite images, International archive of photogrammetry, remote sensing, and spatial information science (2021). DOI: 10.5194 / isprs-archives-XLIV-M-3-2021-175-2021
Chandi Witharana et al, an object-based approach for mapping tundra ice wedge polygon troughs from very high spatial resolution optical satellite images, Remote sensing (2021). DOI: 10.3390 / rs13040558
University of Texas at Austin
Quote: Monitoring of the Arctic permafrost by satellite, supercomputer, and deep learning (February 22, 2022) is available at https://phys.org/news/2022-02-arctic-permafrost-satellites-supercomputers-deep. Obtained from html on February 22, 2022.
This document is subject to copyright. No part may be reproduced without written permission, except for fair transactions for personal investigation or research purposes. Content is provided for informational purposes only.
Monitoring the Arctic permafrost with satellites, supercomputers and deep learning
Source link Monitoring the Arctic permafrost with satellites, supercomputers and deep learning