Very-high resolution aerial imagery and deep learning uncover the fine-scale patterns of elevational treelines

Treelines are sensitive indicators of global change, as their position, composition and pattern directly respond to ecological and anthropogenic factors. Treelines worldwide exhibit a great variability even within single landscapes, which limits the reliability and generalizability of locally measured patterns. Advancing methods to accurately map fine-scale treeline spatial patterns over large extents is crucial to overcome this limitation. Innovative approaches integrating remote sensing with uncrewed aerial vehicles (UAV) and deep learning offer a promising way to bridge the gap between field-based observations of fine-scale patterns and their large-scale implications, ultimately informing and supporting practices for the conservation of forest ecosystems in the face of ongoing and future ecological challenges. In this study, we combined field data and UAV-based remote sensing with a deep learning model to retrieve individual tree-scale information across 90 ha in 10 study sites in the Italian Alps. Using the proposed methodology, we were able to correctly detect individual tree crowns of conifers taller than 50 cm with a detection rate of 70 % and an F1 score of 0.76. Accuracy increased with tree height, reaching 86 % for trees taller than 2 m. Canopy delineation was robust overall (Intersection over Union, IoU Combining double low line 0.76) and excellent for tall trees (IoU Combining double low line 0.85). Tree position and height estimates achieved RMSEs of 59 cm and 92 cm, respectively. Our results demonstrated that the proposed methodology effectively detects, delineates, georeferences, and measures the height of most trees across diverse Alpine treeline ecotones. The proposed methodology enables the analysis of fine-scale patterns in order to achieve an interpretation of underlying ecological processes over large ecotonal extents. The inclusion of heterogeneous study areas facilitates the transferability of the segmentation model to other mountain regions and offers a benchmark for developing a global network of fine-scale mapped treeline spatial patterns, bearing a great potential in monitoring the effects of global change on ecotone dynamics.