Abstract ID: 3.10267 | Accepted as Talk | Talk/Oral | TBA | TBA

Mountain forest ecosystems are sensitive to global change. Especially at the ecotones we often expect high sensitivity to changes in climate, disturbance regimes or land use. The advent of machine learning, specifically computer vision, provides powerful tools to investigate ecotones across extended spatiotemporal extents using remote sensing data. Here, we focused on the spatiotemporal development of the treeline and montane-subalpine forest ecotones in a protected area in the European Alps. First, we aimed to identify trees and shrubs on aerial images, with special attention to integrating multiple sensor types into one computer vision framework. Second, we mapped a) the montane forest zone, b) the subalpine forest zone, and c) the krummholz zone, as well as d) the ecotones in between. Third, we investigated the spatiotemporal changes occurring in the vegetation zones and the ecotones. We based our analysis on aerial images of Berchtesgaden National Park covering nine time steps from 1953 to 2020. The images were captured through analog (panchromatic, color infrared) and digital (color infrared, RGB) cameras. To generate training data, we manually interpreted randomly distributed 0.5 ha scenes across all time steps, resulting in >110,000 annotations of trees, shrubs, and standing dead trees. We tested a set of instance segmentation frameworks and compared individual models for each image type with models integrating all image types. We used the inference of the best performing model to generate wall-to-wall tree maps. Using structure and composition of the tree maps, we delineated zones and ecotones and tracked changes over time. We found that a combined computer vision model for all image types performed better than individual models for each image type. While the extent of montane and subalpine forest zones changed over time, krummholz was notably stable. Conversely, crown cover increased more strongly for krummholz and the subalpine zone than for the montane zone. Although we did not find a general pattern of ecotones shifting upward, we observed remarkable local upward shifts. On average, the upward shift of the montane-subalpine ecotone was roughly two times faster than the subalpine-alpine ecotone, decreasing the subalpine forest area by approximately 25% over 67 years.