Extraction of Tree Level Forest Structure from Aerial Imagery and Photogrammetry Digital Surface Models

Forest inventory variables such as tree height, species distribution and biomass form the foundations for understanding of forest ecosystems. This paper presents an integrated workflow that combines data refinement, deep learning (DL) based Individual Tree Crown (ITC) segmentation and structural metric extraction to enhance the forest inventory process. The individual tree crown segmentation follows two steps. To close the domain gap from the publicly available RGB UAV dataset and single band panchromatic aerial imagery, we explore the possibility of involving a DL based colorization approach to generate synthetic RGB images. For tree height structure analysis, we integrate and compare canopy height observations from photogrammetry digital surface models in pixel and tree levels. In addition, we conduct an in-depth investigation to simulate the forest inventory process with various sampling plot sizes at a natural forest site.