Video
Abstract
A number of meandering rivers in the Arctic, especially in Siberia, have point bars with strikingly structured permafrost polygon patterns. These seem to mimic not just the scroll ‘bar’ lines that are often visible in permafrost-free point bars, but the radial growth lines as well that tend to be perpendicular to the scrolls and have not been observed before. We use a Unet-type convolutional neural network to map the polygons. The network is trained with image tiles derived from manually mapped scroll- and radial lines on nine point bars. Post-processing of the semantic segmentation results allows the identification of individual polygons; a Python ‘networkx’ graph is used to study the relations between the polygons. A simple model of meandering suggests a preliminary explanation of how these beautiful patterns might form.
Event
Permafrost Discovery Gateway Monthly Webinar Series, November 9, 2023.