The International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences
Publications Copernicus
Articles | Volume XLII-3/W3
20 Oct 2017
 | 20 Oct 2017


N. Saarinen, M. Vastaranta, R. Näsi, T. Rosnell, T. Hakala, E. Honkavaara, M. A. Wulder, V. Luoma, A. M. G. Tommaselli, N. N. Imai, E. A. W. Ribeiro, R. B. Guimarães, M. Holopainen, and J. Hyyppä

Keywords: Spectral Imaging, Remote Sensing, Forest Ecology, Forest Mensuration, Forest Inventory, UAS, Photogrammetry

Abstract. Biodiversity is commonly referred to as species diversity but in forest ecosystems variability in structural and functional characteristics can also be treated as measures of biodiversity. Small unmanned aerial vehicles (UAVs) provide a means for characterizing forest ecosystem with high spatial resolution, permitting measuring physical characteristics of a forest ecosystem from a viewpoint of biodiversity. The objective of this study is to examine the applicability of photogrammetric point clouds and hyperspectral imaging acquired with a small UAV helicopter in mapping biodiversity indicators, such as structural complexity as well as the amount of deciduous and dead trees at plot level in southern boreal forests. Standard deviation of tree heights within a sample plot, used as a proxy for structural complexity, was the most accurately derived biodiversity indicator resulting in a mean error of 0.5 m, with a standard deviation of 0.9 m. The volume predictions for deciduous and dead trees were underestimated by 32.4 m3/ha and 1.7 m3/ha, respectively, with standard deviation of 50.2 m3/ha for deciduous and 3.2 m3/ha for dead trees. The spectral features describing brightness (i.e. higher reflectance values) were prevailing in feature selection but several wavelengths were represented. Thus, it can be concluded that structural complexity can be predicted reliably but at the same time can be expected to be underestimated with photogrammetric point clouds obtained with a small UAV. Additionally, plot-level volume of dead trees can be predicted with small mean error whereas identifying deciduous species was more challenging at plot level.