Researchers developed a method that uses high-resolution satellite imagery and deep machine learning to double the mapped infrastructure of Alaska and more accurately project economic risks associated with permafrost thaw
In the Arctic, permafrost plays a crucial role in building infrastructure. However, as the region warms and permafrost thaws, infrastructure is threatened as the ground shifts beneath the built environment. Unfortunately, the full extent of the risks associated with this process is not yet realized, but researchers are working to address this knowledge gap.
UConn Department of Natural Resources and the Environment researchers including PhD student Elias Manos and Assistant Professor Chandi Witharana, and Anna Liljedahl from the Woodwell Climate Research Center developed a method that uses high-resolution satellite imagery and deep machine learning to double the mapped infrastructure of Alaska and more accurately project economic risks associated with permafrost thaw. Their findings are published in Nature Communications Earth and Environment.
Witharana says this is the latest in his research group’s long-term study of how satellites can help monitor changes in the Arctic landscape over time, in this case, the largely unaccounted for risks of thawing permafrost for communities and their vital infrastructure like buildings and roads.
“The main focus here is there was a visual gap for infrastructure, and we need to have more detail to create critical information layers for downstream analysis like economic risk. We didn’t have that for Alaska,” says Witharana.
The motivation behind this research stems from the need to understand hazards in a changing world, says Manos. However, those assessments cannot happen without a clear understanding of what is in harm’s way.
“We know that local temperatures are rising and there is change in the frequency, intensity, and timing of extreme weather and hazardous events. Whether they are rapid onset events like hurricanes, flooding, wildfires, or slow onset hazards like droughts, permafrost thaw in this case, we need to understand the potential harm these events pose,” says Manos.
Manos says that permafrost serves as a structural foundation where piles are secured through it and buildings are designed to help maintain its thermal integrity. It is, therefore, essential that the pile foundation remains stably anchored into the permafrost, but the structural integrity is compromised as this layer thaws.
“When the temperature of permafrost starts to increase, piles start to shift out of place, and that’s what we call bearing capacity loss, or decrease in bearing capacity. That was the main hazard that we looked at which impacts buildings,” says Manos. “Then there’s also transportation infrastructure that’s primarily impacted by ground subsidence. When ice-rich permafrost thaws, the ground will cave in and that was the hazard we used to assess the disaster risk for roads.”
Previous studies made risk estimates based on data from OpenStreetMap (OSM), which is one of the most widely used geospatial data sets available, says Manos. OSM is available for every nation across the globe, and information is updated by volunteers who manually input local data, like buildings, trails, roads, or other kinds of infrastructure, from high-resolution imagery on a global scale.
For some regions, like Europe and parts of the United States, the data is accurate says Manos, but that is not true for all locations. Unfortunately for the Arctic, OSM data is lacking.
“There are several previous risk studies that relied on this incomplete infrastructure data. It all goes back to the fact that infrastructure across the Arctic is not completely mapped, and that’s problematic if you want to understand disasters because you must have the full picture to understand the scale of what is or could potentially be exposed,” says Manos.
One of the objectives of Witharana’s research group is to improve methods to analyze large sets of satellite images quickly and accurately. Here, they developed a method to accurately map infrastructure and permafrost thaw risk called High-resolution Arctic Built Infrastructure and Terrain Analysis Tool (HABITAT). The model uses machine learning and AI to extract road and building information from high-resolution satellite images from the years 2018-2023. They compared the HABITAT data with OSM data to evaluate the new model’s quality and to look for potential misclassifications. Then they added the new information to OSM, nearly doubling the previous amount of information available for Alaska.
“The sheer amount of infrastructure and buildings that were missing from Open Street Map was, really shocking to me, 47% missing,” says Manos. “Though OpenStreetMap is a powerful volunteer-based resource, it has limitations and that is not a surprise.”
Owing to the large amount of data previously not considered, the researchers estimate that the costs of permafrost damage to infrastructure will double under low and medium emissions scenarios by 2050.
“Damages to infrastructure caused by permafrost thaw is on par with the average yearly cost of all natural disasters in the country, yet permafrost thaw is not recognized by the federal government as a natural hazard making it harder for people in Alaska to obtain disaster relief funding. In addition, Alaska is decades behind the rest of the country in terms of geospatial data readiness. Maps are key for assessments and planning and I think the research community can help with some of that,” says Liljedahl.
Witharana’s research group and collaborators are working to fill these knowledge gaps to create data that can be used to help prepare communities for the future. Manos plans to expand this analysis to account for the entire Arctic region to assess economic losses using a comprehensive infrastructure map.
Witharana adds that by combining OSM data with the thousands of sub-meter resolution satellite images provided by the National Science Foundation, along with access to NSF supercomputing infrastructure, it was possible for the researchers to enhance the completeness of these datasets.
“We can see that impact and do better assessments of economic disturbances and risk so we can prepare for whatever policy actions or downstream efforts that are needed,” says Witharana “That’s a major outcome. Overall, the integration of AI and big data sets within our application has helped make useful, actionable products that researchers and communities can use right now.”
The combined HABTAT and OSM dataset is available for anyone to explore on the Permafrost Discovery Gateway. This work is funded by the U.S. National Science Foundation’s Office of Polar Programs (NSF-OPP) (grant No. 1927723 and 2052107) and Google.org’s Impact Challenge on Climate Innovation. The image in Fig. 1b was acquired and provided through NSF RISE-1928237. Furthermore, this work used the Delta supercomputer at the National Center for Supercomputing Applications at the University of Illinois Urbana-Champaign through allocation #EES220055 from the Advanced Cyberinfrastructure Coordination Ecosystem: Services & Support (ACCESS) program, which is supported by National Science Foundation grants #2138259, #2138286, #2138307, #2137603, and #2138296. Geospatial support for this work was provided by the Polar Geospatial Center under NSF-OPP awards 1043681, 1559691, and 2129685.
Developing a clearer understanding of permafrost thaw risk in Alaska added by Elaina Hancock | University of Connecticut on
View all posts by Elaina Hancock | University of Connecticut →