According to a news release last week from the National Science Foundation, scientists with the NSF-led “Navigating the New Arctic” climate research program recently applied deep-learning technology to spot changes to permafrost, making use of a supercomputer from University of Texas at Austin’s Texas Advanced Computing Center to analyze archived satellite images of ice-wedge polygons.
Chandi Witharana, an assistant professor of natural resources and environment at the University of Connecticut and one of the lead researchers, said despite the large amount of data collected by commercial satellites in recent years, the information has been difficult to translate into comprehensive research. But deep learning systems have the potential to change that.
Together with Anna Liljedahl, an Arctic researcher from the Woodwell Climate Research Center in Massachusetts, and others from the National Center for Supercomputing Applications (NCSA), Witharana said scientists hope to make existing data more accessible for climate scholars.
“To better model and understand the response of the Arctic, one needs data,” he said. “The main idea is to understand the evolution of the Arctic and permafrost landscape using existing satellite imagery so Arctic researchers can understand the impact of climate change on Arctic environments. We can fill in the gaps with satellite imagery.”
According to the team, permafrost, which covers about 15 percent of the Northern Hemisphere, acts as a “biosink” for biomass that’s stored as greenhouse gases, such as methane and carbon dioxide. As the planet’s permafrost thaws, these gases will be released, exacerbating climate change and altering soil density, creating issues for anything built on top.
Witharana said researchers still need a better method for predictive data analysis, but advanced satellite imaging techniques and remote sensing technology will help them monitor changes in the soil. This could prove useful for public agencies and others tasked with maintaining and constructing vital infrastructure in the Arctic, such as roads, buildings and pipelines.
“From field observations, scientists have found that the Arctic permafrost is [facing] increasing ground temperatures ... and it’s changing the micro-topography,” he said, adding that the ground is essentially “collapsing” due to global warming.
“One critical aspect is that this difference in settlement of the ground can affect Arctic infrastructure.”
According to the team, Witharana’s research focuses largely on the use of neural networks to detect polygons — underground ice formations similar to honeycombs in shape — from thousands of satellite images. Liljedahl said examining changes to these polygons is crucial to gaining a clearer understanding of climate change’s topographical impacts.
“Every year, we get a near real-time pulse meter on the Arctic in the form of sea ice extent,” Liljedahl said. “We want to do the same with permafrost. There are so many rapid changes. We need to be able to really understand, and communicate, what’s happening in the permafrost.”
According to the foundation, researchers are focused on analyzing different years and seasons in order to predict topographical trends, as well as where future changes may affect infrastructure.
“This is a perfect example of how previous investments in computing infrastructure, combined with a new understanding of deep learning techniques, are building a resource to help with an important issue in the Arctic,” NSF Program Director Kendra McLauchlan said in a public statement.
