February 20, 2020

New Map Reveals Heterogeneous Permafrost Degradation in Ice Wedge Polygons

Machine learning employed to map and measure the microtopography of over one million ice wedge polygons near Prudhoe Bay, Alaska.

The Science

A machine-learning based approach was used to map the occurrence of tundra landforms known as ice wedge polygons across a ~1,200 km2 landscape in northern Alaska. Microtopographic relief was measured in over one million polygons, revealing complex spatial patterns in permafrost degradation with unprecedented detail.

The Impact

The new map visualizes the spatial distributions of low-centered polygons (LCPs)—which are associated with pristine conditions—and high-centered polygons (HCPs)—which are associated with thawing permafrost, and emit elevated amounts of carbon dioxide to the atmosphere—in ultra-high resolution. The map can be used to estimate landscape-scale carbon fluxes and to monitor contemporary rates of permafrost degradation, by measuring increases in the spatial coverage of HCPs in the future.

Summary

Ice wedge polygons are tundra landforms that cover an estimated 2.5 million square kilometers in the circumpolar Arctic. Most polygons fit between two geomorphic endmembers: low centered polygons (LCPs), which are characterized by rims of soil at the edges; and high-centered polygons, which resemble mounds surrounded by a network of troughs, and usually reflect thaw in the underlying permafrost. Understanding the spatial distributions of LCPs and HCPs is important, because the two morphologies are associated with pronounced differences in runoff generation, soil moisture, and greenhouse gas emissions. However, high-resolution mapping of ice wedge polygons is difficult, as several thousand polygons may occupy as single square kilometer of terrain, and the microtopographic features distinguishing LCPs and HCPs commonly represent only a few tens of centimeters of relief. Researchers from NGEE-Arctic employed a novel machine learning-based approach, built around a cutting-edge algorithm known as a convolutional neural network, to map the boundaries of more than one million ice wedge polygons across a ~1,200 km2 landscape near Prudhoe Bay, Alaska, using a high-resolution digital elevation model generated through an airborne lidar survey. We then measured the relief at the center of each ice wedge polygon, to place it on a spectrum between LCP and HCP. Their map reveals complex trends in ice wedge polygon form, on spatial scales varying from meters to tens of kilometers, with unprecedented detail. This high-resolution quantification of ice wedge polygon form provides rich spatial context for extrapolating ground-based measurements of carbon emissions from tundra soils, and parameterizing microtopography within earth system models. It also represents an extensive baseline dataset for quantifying how contemporary rates of permafrost degradation vary across a landscape, by observing where new high-centered polygons form as air temperatures in the Arctic continue to rise.

Principal Investigator

Charles Abolt
Los Alamos National Laboratory
[email protected]

Program Manager

Daniel Stover
U.S. Department of Energy, Biological and Environmental Research (SC-33)
Environmental System Science
[email protected]

Funding

  • Next Generation Ecosystems Experiments Arctic (NGEE-Arctic) project (DOE ERKP757), funded by the Office of Biological and Environmental Research in the U.S. Department of Energy Office of Science
  • NASA Earth and Space Science Fellowship program, for an award (80NSSC17K0376) to the lead author.

References

Abolt, C. J., and M. H. Young. "High-resolution mapping of spatial heterogeneity in ice wedge polygon geomorphology near Prudhoe Bay, Alaska." Scientific Data 7 87  (2020). https://doi.org/10.1038/s41597-020-0423-9.