Finding the Trees in the Forest
High resolution datasets acquired by aircraft and UAVs can facilitate the automated identification of individual tree crowns across large areas. The ability to analyse remotely sensed data at the individual tree crown scale can provide significant advantages for the management and monitoring of forest environments.
Dr Chloe Barnes
Head of Remote Sensing
Chloe joined the 2Excel geo team in 2017, following the completion of her PhD in Remote Sensing. She is a domain expert in tree disease detection using spectral imaging and LiDAR techniques from airborne platforms and is an experienced data analyst.
Many remotely sensed datasets provide useful information about trees such as height, species and health. One of the great advantages of high resolution datasets is the ability to distinguish individual treetops and tree crowns within the forest canopy. This information facilities the targeted analysis of individual trees to generate parameters specific to that particular individual.
Manual delineation of individual trees is an inefficient means of distinguishing tree crowns over large areas. However, data captured via remote sensing methods can be used to automatically detect treetops and crown boundaries. One approach to identify individual trees requires canopy height models (CHM), which represent the surface height of the forest canopy. CHM’s can be generated from LiDAR sensors or derived from stereo imaging. Treetops are located on the CHM by identifying local maxima or the tallest points across the canopy surface. In order to isolate or segment the individual tree crowns, the CHM is then turned upside down so that tree crowns represent a series of basins. Each basin (tree crown) is then flooded from the bottom, crown boundaries are subsequently drawn at the points where ‘water’ floods into the next basin (an application of the watershed segmentation technique). Outputs from the segmentation typically include points representing the location and height of treetops and polygons representing the crown boundaries.
Figure 1a
Automated tree crowns overlaying an RGB image.
Figure 1b
Automated tree crowns overlaying a canopy height model.
Figure 1a
Automated tree crowns overlaying an RGB image.
Figure 1b
Automated tree crowns overlaying an canopy height model.
Additional processing steps using structural and spectral information can also be applied to remove non-forest structures from the data prior to the segmentation, to ensure that only areas of tree cover are used as inputs for the automated delineation of tree crowns.
Figure 2
Inverted canopy model demonstrating the appearance of tree crowns as basins.
Figure 2
Inverted canopy model demonstrating the appearance of tree crowns as basins.
This particular technique of automated tree detection to best suited to homogenous environments such as commercial plantations. In environments with more variable canopy structures such as mixed broadleaved woodlands or urban areas, the identification of individual trees is still achievable but requires a more complex approach and a finer tuning of multiple input parameters.
Automating the identification of tree crowns in this way provides several benefits to the assessment and management of forests. It provides a means of determining the height and crown of each tree within the region of interest. It is this ability to conduct analysis at the individual tree crown scale which creates the greatest opportunity for forest management.
Figure 3
Different indices
