Biomass estimates derived from sector subsampling of 360° spherical images

• Published in: Forestry: An International Journal of Forest Research
• Main Authors: Dai, Xiao, Ducey, Mark J, Wang, Haozhou, Yang, Ting-Ru, Hsu, Yung-Han, Ogilvie, Jae, Kershaw, John A
• Other Authors: New Brunswick Innovation Foundation, Natural Sciences and Engineering Research Council of Canada, Department of Natural Resources, Government of Newfoundland and Labrador
• Format: Article in Journal/Newspaper
• Language: English
• Published: Oxford University Press (OUP) 2021
• Subjects: Newfoundland
• Online Access: http://dx.doi.org/10.1093/forestry/cpab023; http://academic.oup.com/forestry/article-pdf/94/4/565/39611197/cpab023.pdf

Abstract Efficient subsampling designs reduce forest inventory costs by focusing sampling efforts on more variable forest attributes. Sector subsampling is an efficient and accurate alternative to big basal area factor (big BAF) sampling to estimate the mean basal area to biomass ratio. In this study, we apply sector subsampling of spherical images to estimate aboveground biomass and compare our image-based estimates with field data collected from three early spacing trials on western Newfoundland Island in eastern Canada. The results show that sector subsampling of spherical images produced increased sampling errors of 0.3–3.4 per cent with only about 60 trees measured across 30 spherical images compared with about 4000 trees measured in the field. Photo-derived basal area was underestimated because of occluded trees; however, we implemented an additional level of subsampling, collecting field-based basal area counts, to correct for bias due to occluded trees. We applied Bruce’s formula for standard error estimation to our three-level hierarchical subsampling scheme and showed that Bruce’s formula is generalizable to any dimension of hierarchical subsampling. Spherical images are easily and quickly captured in the field using a consumer-grade 360° camera and sector subsampling, including all individual tree measurements, were obtained using a custom-developed python software package. The system is an efficient and accurate photo-based alternative to field-based big BAF subsampling.