Grant Award View - GA356424

Our project will develop automated methods for monitoring vegetation structural and functional metrics at fine levels of spatial detail (tree and branch level) across local to regional scales (100s ha) by addressing a missing drone-based LiDAR processing capability. Algorithms will be developed and verified to produce accurate and precise tree and woody vegetation models down to cm accuracy from 3D point cloud data. These 3D models will allow vegetation structural and functional baseline metrics to be automatically extracted at both individual tree and entire site levels, namely: stem density, stem diameter, crown height, crown width, woody volume and aboveground biomass.

This new method may significantly reduce vegetation field inventory budgets by providing a step-change in vegetation metrics’ accuracy and scale of observation (100’s ha). This capability is required to take full advantage of the increased area coverage of drone operations that will be possible with longer flight time capabilities and beyond visual line of sight (BVLOS) drone operation. A current limitation of satellite-based vegetation structure measurements at regional to continental scales is the lack of detailed 3-dimensional calibration data across a range of vegetation communities – which is what this project will produce. 

A highly capable team from The University of Queensland (UQ) supported by our CSIRO collaborators will use recently collected data at long-term TERN monitoring sites across a range of vegetation communities for the algorithm development. A primary outcome is expected to be an operational workflow applicable to the vegetation communities across Australia, suitable for ongoing use by government and industry.  

The methodology developed in this project will build-out a capability in a data simulation environment for further benchmarking and assessing of remote sensing methods for extracting structural and functional attributes. The data simulation approach overcomes limitations of solely using real data by having a precisely known benchmark or true reference value for testing purposes. Finally, this approach will lead to recommendations as to the optimal settings for future data collection required to reach a specific target accuracy.