This dataset shows mature tree crown polygons across Canberra, Australian Capital Territory (ACT) derived from 2020 LiDAR Canopy Height Model. Mature Tree crowns are defined as those larger than 100m2 canopy footprint, and higher than 8 metres tall. Includes native and non-native tree representation and estimates height. Also see Mature Tree Centre Points.
Methods:The canopy dataset is derived from LiDAR (Light Detection and Ranging) data captured by Aerometrex and the ACT Government (available here) at 12ppm over the ACT and region in April 2020. Tree crowns were delineated using a one metre 2020 LiDAR Canopy Height Model (CHM) via a marker-controlled watershed delineation method (Popescu & Wynn 2004; Plowright & Roussel 2020).A 3X3 low pass smoothing algorithm was applied to the CHM raster. The CHM was then loaded into R and manipulated using the ForestTools R-package (Plowright & Roussel 2020). A variable window filter, which was used in parallel with watershed segmentation, was applied to determine locations of treetops. This algorithm employs a moving window for the CHM to filter pixels, the highest value within the window is regarded as a treetop. The size of the moving window differs and depends on the positive relationship between crown width and tree height. The following metrics was used for the variable window:function(x){((x^2)* 0.009 + 2.5)}. A minimum height of 3 metres was set for the treetop detection algorithm. The resulting dataset, which contained attributes for unique tree identifiers, tree height and moving window size, was saved as a point shapefile. The watershed segmentation method was implemented within the ForestTools R-package to outline tree crowns, using the previously identified treetops as markers. The resulting segmented raster was converted to a simplified polygon layer, and joined with the treetops point layer. Polygons smaller than 3 square metres were removed. Tree crown polygons larger than 100m2 and higher than 8 metres were filtered as "mature trees". Dataset represents native and non-native species.Tree Crown Delineation MethodThe inverse watershed delineation (IWD) method is the popular technique for segmenting individual tree crowns using LIDAR data (Marcu, et al., 2017) (Wannassiri, et al., 2013). LiDAR-based IWD methods have been successfully applied in conifer-dominated systems, while broadleaf-dominated systems have proven to be more challenging (Hastings, et al., 2020). A common problem with the IWD method is the overestimation of tree crowns, especially in dense tree stands. This problem can be partially overcome by a) aggressively filtering the data, and b) using a variable window filter (VWF) algorithm to detect dominant treetops, which then feeds into the IWD.In this project, 2015 and 2020 tree crowns were delineated from the respective CHM’s, via a marker-controlled watershed delineation method (Popescu & Wynne, 2004) implemented via the ForestTools R-package (Plowright & Roussel, 2021). The R-code used in this project is included in metadata lineage.A variable window filter was applied within the ForestTools package to determine locations of treetops. This algorithm employs a variable moving window for the CHM to filter pixels, and the highest value within the window is regarded as a treetop. The size of the moving window differs and depends on the positive relationship between crown width and tree height. The derivation of the appropriate window size to search for treetops is based on the assumption that there is a relationship between the height of the trees and their crown size. The higher the trees, the larger the crown size.Various functions were tested to define the window search radius, including simple linear functions. The function described by Popescu & Wynne was eventually chosen to inform the window search radius of the tree top detection algorithm, as it achieved the best results for mature trees.A minimum height of three metres was set for the treetop detection algorithm. The resulting dataset, which contained attributes for unique tree identifiers, tree height and moving window size, was saved as a point shapefile. The IWD method was then implemented within the ForestTools R-package to outline tree crowns, using the previously identified treetops as markers. The resulting raster dataset contained unique identifies corresponding with the tree tops point dataset. The ensuing segmented rasters were converted to polygon layers and joined with the treetops point layers.Delineated Tree Crown Model ValidationMap products produced by remote sensing analysis typically uses ‘ground truth’ to determine accuracy. In practice, accuracy assessments evaluate the remote sensing-derived map product relative to some higher quality determination of the map. These higher quality data, referred to as ‘reference data’, is compared to the map product using statistical sampling (Stehman, 2009).Validation of the 2015 mature tree crown polygons was done by randomly selecting 40 points from the dataset and checking them against manually digitized crown area polygons (digitized from 2015 aerial photography). The validation showed that the model tended to overestimate the tree canopy area for 32 of the reference polygons, while underestimating the tree canopy area for 8 of the reference polygons. The overestimation of the canopy projective area by the delineation model can be partly attributed to the low pass filter that was used to smooth the canopy height model during the delineation process, which effectively added pixels to the perimeter of tree crowns in the CHM.A similar process was followed to validate the 2020 mature tree crown model, using the same 40 randomly selected points and checking them against manually digitized crown area polygons from 2020 aerial photography. The validation showed that the model tended to overestimate the tree canopy area for 33 of the reference polygons, while underestimating the tree canopy area for 7 of the reference polygons.Also seeACTGOV Canopy Cover 2020 1m - Districts - Overview (arcgis.com)ACTGOV Canopy Cover 2020 Statistics - Overview (arcgis.com)References:Plowright, A & J Roussel. ForestTools: Analyzing Remotely Sensed Forest Data. R Package Version 0.2.4. 2021. https://cran.r-project.org/web/packages/ForestTools/index.htmlPopescu SC & Wynne RH 2004. Seeing the Trees in the Forest: Using Lidar and Multispectral Data Fusion with Local Filtering and Variable Window Size for Estimating Tree Height. Photogrammetric Engineering & Remote Sensing Vol. 70, No. 5, pp. 589–604.
