Combining Theory, Deep Learning, and Lidar to Test Climate and Slope Controls on Tree Throw Production on Hillslopes

Indiana University, Bloomington IN

Investigators

Abstract

When trees fall over and uproot, they suddenly heave soil and rock from deep in the soil mantle to the surface. This process, called tree throw, is an important contributor to sediment transport on hills and influences soils, water, carbon, and ecology in forested landscapes, yet quantifying the frequency of such events is challenging because events are infrequent. However, tree throw leaves a topographic signature: a pit in the location of the fallen tree and a mound of “thrown” sediment immediately downslope. The topographic signature of tree throw persists for many decades or centuries so that the land surface represents a history of tree throw events and offers an opportunity to quantify the process. Further, because tree throw is often driven by extreme weather, the topographic signature of tree throw may serve as an archive of extreme events. Building on theory that describes the roughness of the land surface due to the periodic creation of pits and mounds, the investigators will leverage topographic signatures from high resolution topographic data, theory, and machine learning to map tree throw instances across large areas. The team will engage K-12 teachers through the Indiana University’s Education for Environmental Change program that consists of a week-long workshop that focuses on experiential learning and curriculum development. Finally, by combining Earth science and deep learning, graduate students working on this grant will be trained in cross-disciplinary methods and will be able to address problems in science, industry, and informatics-related fields. Tree throw occurs when extreme atmospheric events exert forces on forest canopies that can exceed soil and root strengths. The uprooting creates a topographic signature in forest floors, which creep-like processes rework and degrade. Thus, the spatial patterns of topographic roughness contain process information of tree throw rates and the events that drive them. This project will establish new methods for automated mapping of pit-mound couplets in topographic data and theory to interpret roughness in process-based terms. The researchers will use lidar data at select sites in Indiana, West Virginia, Pennsylvania, South Carolina, and Tennessee to identify pit-mound couplets. They will also augment publicly available data with higher resolution lidar datasets that they will collect using an unmanned aerial vehicle equipped with a lidar unit. To map pit-mound couplets across large areas, the researchers will develop and train deep learning algorithms that automatically map the locations of pit-mound couplets. They anticipate mapping several million features across southern Indiana where they have already demonstrated a high density of tree throw pit-mound couplets. The research team will combine the automatically mapped inventory of tree throw events with existing theory to provide new insights on what controls the rates and spatial patterns of tree throw. This project is co-funded by a collaboration between the Directorate for Geosciences and Office of Advanced Cyberinfrastructure to support AI/ML and open science activities 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.