Developing a forest inventory approach using airborne single photon lidar data: from ground plot selection to forest attribute prediction. (8th December 2021)

• Record Type: Journal Article
• Authors: Queinnec, Martin; Coops, Nicholas C; White, Joanne C; McCartney, Grant; Sinclair, Ian
• Is Part Of: Forestry. Volume 95:Number 3(2022)
• Page Start: 347
• Page End: 362
• Publication Date: 2021-12-08
• Subjects: Forests and forestry -- Periodicals; Forests and forestry -- Great Britain -- Periodicals; 634.9
• DOI: 10.1093/forestry/cpab051
• Languages: English
• ISSNs: 0015-752X

Abstract

Abstract: An increasing number of jurisdictions are integrating airborne laser scanning (ALS) into forest inventory programs to produce spatially explicit and accurate inventories of forest resources. However, wall-to-wall ALS coverage relative to the total area of managed forest remains limited in large forest nations such as Canada, wherein logistics, cost and acquisition capacity can be limiting factors. Technologies such as single photon light detection and ranging (SPL) have emerged commercially, which have the capacity to provide efficient ALS acquisitions over large areas and with a greater point density than conventional linear-mode ALS. However, the large-scale operational application of SPL in a forest inventory still needs to be effectively demonstrated. In this study, we used wall-to-wall SPL data (collected with a Leica SPL100) across a 630 000 ha boreal forest in Ontario, Canada to develop a forest inventory. Specifically, we used a structurally guided sampling approach enabled via a principal component analysis of the SPL100 data to establish a network of 250 ground plots. Random forest models were then used to produce area-based estimates of forest attributes of interest. Results demonstrated that the sampling approach enabled the optimization and enhancement of the existing plot network by extending the range of sampled structural types and reducing the number of plots in oversampled forest types. Moreover, Lorey's height, basal area, quadratic mean diameter at breast height, stem density, gross and merchantable volume and above-ground biomass were estimated with a relative root mean square error of 8.5, 19.76, 13.97, 30.82, 21.53, 23.79 and 22.87 per cent, respectively, and relative bias <1 per cent. Model accuracies achieved using the SPL100 were comparable with those obtained using linear-mode ALS in a previous forest inventory. This study demonstrates the utility of the SPL100 for the complete development of a forest inventory over large forest areas, from ground plot establishment through to the production of forest attribute estimates.