Stratifying Forest Overstory and Understory for 3-D Segmentation Using Terrestrial Laser Scanning Data
Zengxin Yun, Guang Zheng
Abstract
AbstractAccurately and rapidly segmenting tree crowns from a three-dimensional (3-D) perspective is of great significance to precision forest management, and better understand the carbon and water cycles between the soil-plant-atmosphere system. However, it remains challenging to group points into individual trees from a 3-D perspective in the forest stand with highly overlapped tree crowns and abundant understory. The objective of this paper was to extract the overstory and understory of individual trees from terrestrial laser scanning (TLS) data considering the vertical forest structure and overlapped tree crowns processing strategy suitable for various crown shapes and sizes. Our results showed that: (1) the proposed algorithm had better performance in the low overlapping rate coniferous (F1-score: 0.96) and broadleaf (F1-score: 0.91) forest stands, while the F1-score decreased down to 0.89 and 0.65 in the high overlapping rate for coniferous and broadleaf forest stand, respectively. (2) A multi-station TLS data produced better (F1-scores: 0.85-1) segmentation results than those obtained from single-station TLS data (F1-scores: 0.67-0.83) in coniferous forest stands. (3) The vertical forest structure profiles affected the final forest 3-D segmentation accuracy. Our work provides a solid foundation for precision forestry and natural resources management.