PhD Defence Jing Liu

retrieving the 3d distribution of leaves in forests using lidar

Jing Liu is a PhD student in the department of Natural Resources. His supervisors are prof.dr. A.K. Skidmore from the Faculty of Geo-information Science and Earth Observation and prof.dr. S. Jones from RMIT University.

In forests, leaves are the interface between the biosphere and the atmosphere, where most of the energy fluxes exchange. The three dimensional (3D) distribution of leaves strongly influence the interception and distribution of radiation, carbon and water. Leaf area index (LAI) and the vertical LAI profile are important metrics describing the amount and distribution of leaves. LAI has been identified as an essential climate variable and an essential biodiversity variable. The vertical LAI profile is a more detailed description of the 3D distribution of leaves inside the canopy. Both metrics have been used in ecology, hydrology and biodiversity modelling.

Remote sensing techniques provide a non-destructive, rapid and economic way for estimating LAI and vertical LAI profile across a wide range of spatial and temporal scales. However, passive optical sensors suffer from limitations including signal saturation and inability to resolve the vertical distribution. As a result, LiDAR plays an indispensable role in mapping the 3D distribution of leaves in forests. Nevertheless, accurate 3D mapping of leaves is a challenging task for forest ecosystem, owing to the complexity of canopy structure, underlying topography, and LiDAR settings.

This thesis evaluated several key factors in the LAI and vertical LAI profile retrieval using LiDAR data at local and regional scale. These factors include the leaf angle distribution (LAD), gap fraction, LiDAR scan angle, and uneven topography, all of which were parameters in the physically based gap fraction model to estimate LAI and vertical LAI profile. At local scale, the thesis first examined which in-situ technique produced more accurate LAD estimate. Using field-based and simulation dataset, terrestrial LiDAR was proved more accurate than DHP when estimating LAD in broadleaf forests. Then the thesis employed the proposed LAD method to examine whether the spherical LAD assumption was valid for natural European beech forests. Using terrestrial LiDAR, large LAD variation was demonstrated both in different stands and in different canopy layers. A uniform distribution rather than a spherical distribution was a more valid LAD assumption. At regional scale, the thesis evaluated the effect of airborne LiDAR flight settings, in particular the scan angle, on the retrieval of gap fraction and vertical gap fraction profile. The results proved that the underestimation of gap fraction amplified at large o-nadir scan angle. It implied that large of-nadir scan angle LiDAR data should be avoided to ensure a more accurate gap fraction and LAI retrieval. Finally, the thesis assessed the effect of uneven topography and topographic normalization in the vertical LAI profile retrieval. The findings demonstrated that topographic normalization undermined the complexity of the vertical LAI profile. For ecological applications, such as biodiversity modeling, topographic normalization was suggested not to be applied.

The new methodologies and findings in the study can be extended to other forests on different sites or of different species. Further studies are recommended to explore the application of LiDAR derived LAI and vertical LAI profile product in modelling carbon stock, forest dynamics and biodiversity.