Automating Plot-Level Stem Analysis from Terrestrial Laser Scanning

• Main Authors: Zhouxin Xi, Christopher Hopkinson, Laura Chasmer
• Format: Article
• Language: English
• Published: MDPI AG 2016-10-01
• Series: Forests
• Subjects: terrestrial laser scanning; forestry; DTM; DBH; stem detection; stem form; automatic; plot scale; TLS; point cloud segmentation
• Online Access: http://www.mdpi.com/1999-4907/7/11/252

Terrestrial laser scanning (TLS) provides an accurate means of analyzing individual tree attributes, and can be extended to plots using multiple TLS scans. However, multiple TLS scans may reduce the effectiveness of individual tree structure quantification, often due to understory occupation, mutual tree occlusion, and other influences. The procedure to delineate accurate tree attributes from plot scans involves onerous steps and automated integration is challenging in the literature. This study proposes a fully automatic approach composed of ground filtering, stem detection, and stem form extraction algorithms, with emphasis on accuracy and feasibility. The delineated attributes can be useful to analyze terrain, tree biomass and fiber quality. The automation was experimented on a mature pine plot in Finland with both single scan (SS) and multiple scans (MS) datasets. With mensuration as reference, digital terrain models (DTM), stem locations, diameters at breast height (DBHs), stem heights, and stem forms of the whole plot were extracted and validated. Results of this study were best using the multiple scans (MS) dataset, where 76% of stems were detected (n = 49). Height extraction accuracy was 0.68 (r2) and 1.7 m (RMSE), and DBH extraction accuracy was 0.97 (r2) and 0.90 cm (RMSE). Height-wise stem diameter extraction accuracy was 0.76 (r2) and 2.4 cm (RMSE).