MCA: Combining Drone Imagery and Deep Learning to Map Fine-Scale Heterogeneity in Arctic Vegetation

Colgate University, Hamilton NY

Investigators

Abstract

The Arctic is warming nearly four times faster than the rest of the planet. As a result of this warming, plants in Arctic lands are changing. In some places, plants are taking advantage of warmer conditions and growing larger or spreading across the landscape. For example, shrubs may be increasing in height and growing in areas that used to contain only grasses, or grasses may be growing taller. Arctic warming is also resulting in more wildfires, which remove plants and create different conditions for new plants to grow. The investigators will use images from satellites to tell where plant growth is increasing and where fires have occurred. However, these satellite images do not have the resolution to tell us what types of plant changes are occurring. In this project, the investigators will analyze detailed images from drones using new computer science tools to help us understand how plants are changing, and to help interpret changes we see in less detailed satellite images. Using an archive of drone images from across the Arctic, the investigators will map patches of shrubs and grasses to understand how they influence estimates of plant growth made from satellites. The investigator will use a different image archive to see how plant growth after wildfires is related to the amount of plants present before the fire. The results will provide insights into how changes in Arctic plants influence the global climate. This research will also help establish new methods for using drone data and computer science tools to map plant processes in fine detail that can be used in other regions. The project will allow a college professor to develop new research skills and new teaching materials and provide research opportunities for undergraduate college students. Results from the project will be shared through scientific journal articles, and data and class materials will be shared publicly on the internet. As the Arctic warms nearly four times faster than the rest of the planet, a wide range of ecosystem changes are occurring. Besides general increases in plant productivity, changes in the geographic distribution of different vegetation types have also been observed. For example, in tundra ecosystems shrubs may be expanding into grass-dominated areas, while elsewhere grass productivity may be increasing without expansion of shrubs. Larger and more intense wildfires with warming may be changing patterns of vegetation recovery post-fire. Satellite images are a useful tool for observing and monitoring these types of changes. However, the resolution of these images is not high enough to identify the specific types of vegetation change that are occurring, which is important because the nature of these changes will determine their impacts on global climate. In this project, the investigators will combine high resolution drone images with cutting-edge machine learning tools to map variation in Arctic vegetation at the level of individual plants. Using an archive of drone images from across the Arctic, the investigators will map the distribution of shrubs and grasses to understand their respective changes, as well as how they affect satellite estimates of ecosystem productivity. Using a second archive, we will map live and dead trees and shrubs in areas affected by wildfire to determine how vegetation regrowth after fire varies in northeastern Siberia. These maps will help increase understanding of the processes driving Arctic vegetation change, and aid interpretation of changes inferred from satellite images. The results will aid our understanding of Arctic vegetation feedbacks to global climate. The combination of drone imagery and artificial intelligence tools will develop and refine methods for studying fine scale vegetation dynamics that can be employed in other regions and ecosystems. The project will enable a mid-career college professor to develop new analytical expertise and skills, and new teaching materials to incorporate in undergraduate classes focused on geographic analysis. Results from the project will be published in peer-reviewed journals, and the resulting data sets and teaching materials will be published in freely accessible online repositories. This project is co-funded by the Directorate for Geosciences to support AI/ML advancement in the geosciences. This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.