TLS-based forest canopy reconstruction for 3D radiative transfer: An approach for a tall Eucalypt forest in SE Australia

0570655 - ÚVGZ 2024 eng A3 - Přednáška/prezentace nepublikovaná
Janoutová, Růžena - Devereux, T. - Woodgate, W. - Lamsal, K. - Homolová, Lucie - Malenovský, Z.
TLS-based forest canopy reconstruction for 3D radiative transfer: An approach for a tall Eucalypt forest in SE Australia.
[ForestSAT 2022. Berlín, 29.08.2022-03.09.2022]
Způsob prezentace: Prezentace
URL akce: https://www.forestsat2022.com/ 
Grant CEP: GA MŠMT(CZ) LTC20055
Institucionální podpora: RVO:86652079
Klíčová slova: terrestrial laser scanning * 3d tree reconstruction * radiative transfer * eucalypt forest
Obor OECD: Remote sensing
https://www.forestsat2022.com/

Reconstruction of large 3D forest scenes is important for analyzing remotely sensed data and for detailed sensitivity studies at tree level using radiative transfer models. Fastly developing terrestrial laser scanning (TLS) and computing technology allow to acquire and reconstruct highly detailed large forest sites at the level of individual leaves and trunks with branches.

Here, we present a workflow for the 3D reconstruction of the Terrestrial Ecosystem Network (TERN) Tumbarumba SuperSite stand (Karan et al. 2016) (south-eastern Australia, −35.6565°N, 148.1517°E). The site contains 402 trees of two dominant eucalypt species: alpine ash (Eucalyptus delegatensis, 134 individuals) and mountain gum (Eucalyptus dalrympleana, 264 individuals). Few individual leaves were scanned to characterize and create in the Blender software small, medium and large 3D representations of leaves for both species. In addition to optical properties of the leaves, reflectance of ground, bark, and optical properties of understorey leaves were measured in an optical integration sphere coupled with the ASD FieldSpec-4 spectrometer (Lamsal et al. 2022).

We developed a buffering technique for individual tree extraction of overlapping crowns from the TLS point cloud acquired in 2019. Our method for 3D reconstruction of single trees is based on segmented TLS data and field measurements of leaf angle distribution. The method comprises of three steps: (i) leaf-wood segmentation of a TLS tree point cloud, (ii) reconstruction of wooden parts (trunks and branches, Fan et al. 2020), and (iii) biologically realistic distribution of foliage within a tree crown (Janoutová et al. 2019 & 2021).

In total 19 reference trees (9 of E. delegatensis and 10 of E. dalrympleana), representing the existing range of tree height and DBH, were selected from the TLS dataset. The trees’ point clouds were used for wooden part reconstruction. Remaining trees were clustered according to their height around those 19 individuals (± 2.5 m) and their total leaf area was estimated from the TLS points corresponding to foliage. Scaling factors were computed for every cluster, resulting in 85 individual tree models of unique dimensions and total leaf areas. The reconstructed trees were then used to build the 1 ha virtual forest scene in the Discrete Anisotropic Radiative Transfer (DART) model (Gastellu-Etchegorry et al. 2017, Malenovský et al. 2021) for the purpose of biochemical retrievals from hyperspectral drone imagery. Challenges of the reconstruction process and subsequent radiative transfer modeling will be presented and further discussed.
Trvalý link: https://hdl.handle.net/11104/0341977