Upscaling Lichen Coverage from cm-Scale Drone to m-Scale Satellite RGB Imagery

Sara Luedke

Program: Master’s Program in Ecology
Date: Wednesday, May 6, 2026
Time: 1:30pm
Location: Donald P. Shiley BioScience Center (BSCI)

Abstract

The Arctic is warming at a fast pace causing changes in vegetation dominance. Lichen, an important player in microclimates across the Arctic, are at risk of being lost in these shifting times. A crucial step to knowing how to incorporate lichen into ecosystem models, is being able to identify its cover change over time, but most importantly, being able to identify lichen at all. In this research we explored the accuracy of using k-means classification and convolution neural networks on RGB imagery and sought to find if modern satellite imagery is fine-scale enough to pick up lichen.

At drone fine-scale resolution, both resulted in adequate outputs, though slightly overestimated due to the difficulty of separating lichen from soil and green vegetation. At satellite coarse-scale resolution, k-means failed to identify anything other than broad categories and the CNN misclassified things at a larger scale as well.

These results show that the typically coarse imagery used for vegetation shifts is not good enough for smaller scale lichen shifts and that we need to identify better methodology as well as find a middle ground resolution that is not as hard to acquire as drone imagery, but better than satellite imagery, so that we may interpret more accurate results of such small organisms moving forward.