Using artificial intelligence to detect permafrost thawing
By Matthew Fielding, Julia Barrott and Annika Flensburg
The amount of soil carbon locked in northern permafrost is about double the amount of carbon currently in the atmosphere. If the permafrost carbon pool was to (even partially) thaw, the chances of meeting the targets in the Paris Agreement would be significantly reduced, or even lost completely.
These issues are laid to bare in the 2019 IPCC Special Report on the Ocean and Cryosphere in a Changing Climate. The Special Report highlights permafrost regions as large, climate sensitive reservoirs of organic carbon that have the potential to contribute significant GHG emissions to the atmosphere over short timescales (months to years). As global warming proceeds and permafrost thaws, the rate of GHG (largely carbon dioxide and methane) emissions from permafrost is expected to increase. This positive feedback loop of warming driving further GHG emissions could severely impact our collective ability to limit global warming to 1.5 degrees Celsius.
Of particular concern is the extent and impact of abrupt – or rapid – permafrost thaw. Recent research shows that abrupt permafrost thaw results in dramatic landform changes (including thaw slumps and lake development) over hours to months that release significant amounts of carbon. The IPCC Special Report estimates that around 20% of permafrost in the northern latitudes – an area larger than the whole of India – may be vulnerable to abrupt thaw.
Understanding climate impacts on, and quantifying GHG emissions from the abrupt permafrost thaw is essential to our understanding of how fast climate change will proceed. These efforts are inhibited by the sheer challenge of undertaking large-scale permafrost assessments due to challenges in monitoring, reporting and verifying changes to carbon stocks at scale and over time. This data is desperately needed. Current large-scale climate models currently only account for gradual permafrost thaw. This is problematic because there is increasing evidence that significant and disproportionate amounts of carbon are rapidly released from abrupt thaw. Research on carbon emissions from abrupt thaw is well underway, but scaling this across such an expansive space as the northern permafrost region is a huge challenge.
A new cross-sector collaboration between SEI, Stockholm University, Alfred Wegener Institute and Accenture volunteers aims to fill these knowledge gaps by offering more accessible data and tracking trends over time rather than isolated spot measurements, and by including the speed and magnitude of the thawing and associated vegetation and soil carbon changes across regions.
The IPCC Special Report on Oceans and the Cryosphere (2019) indicates that there is only ‘medium evidence’ and ‘low agreement’ that record high temperatures in permafrost are causing additional carbon dioxide and methane release from northern permafrost regions. As a result of this limited evidence and lack of agreement, emissions from permafrost thaw receive less attention and are lower on the agenda than well quantified carbon emissions sources. This is despite the potentially catastrophic climate impact of rapid and large-scale permafrost thaw.
This collaboration, part of the newly launched project “AI to detect environmental changes and management options affecting soil carbon storage and GHG emissions”, explores new ways to fill knowledge gaps about the speed, extent and impacts of abruptly thawing permafrost in the Arctic by scaling up the use of AI and creating new partnerships across sectors.
Today, scientists are conducting assessments of soil organic carbon in person. This creates huge challenges in remote and vast land masses such as the Arctic.
This solution, co-created with Accenture volunteers, uses machine learning, AI and earth observation to develop systems of measuring and monitoring processes affecting permafrost and associated soil organic carbon stocks. Thaw slumps and lake dynamics will be used as proxies for permafrost thaw and carbon dynamics using the latest satellite image analysis techniques. “New satellite datasets, like the Sentinel missions, and processing techniques, such as AI, are a real game-changer, which have a great potential for mapping landscape changes across the Arctic and to better understand the complex interactions of permafrost and the climate system,” said Ingmar Nitze from the Alfred Wegener Institute.
These time-based datasets will be analyzed to understand current trends and to predict future ones in permafrost thaw and associated impacts, including CO2 emissions. Such an approach would give insight into the scale of rapid thaw processes that are currently invisible in conventional research.
“We have found a middle ground to measure trends at scale by proxies. This method doesn’t provide the same accuracy as someone physically on the ground, but we hope that our focus on visible trends over time will give an indication over a much wider area than previously studied of the scale of rapid permafrost thaw and make this information more accessible for the public and policymakers. If successful, we will create an online dashboard to provide data to contribute to climate negotiations and climate policy,” said project lead Matthew Fielding, SEI.
“Accenture is proud to support this important research by helping to provide the scientific community with access to specialized data science resources and cloud-based technology solutions,” said Jeffrey Nogosek, Technical Architecture Senior Manager with Accenture. Nogosek recently embarked on a skills-based volunteering project with SEI, and Accenture currently has 30 AI experts volunteering their expertise to support this vital work.