Spruce Tree Crown Delineation using a Novel Algorithm


Spruce trees dominate boreal and temperate forest biomes and are critical to maintaining the global climate. Accurate counting of individual spruce trees and estimation of their crown span have many applications including forest inventorying, estimating the carbon captured as biomass, assessing tree health, and precision forestry practices. Very-high resolution (VHR) multispectral remote sensing image data of trees are acquired using low-flying unmanned aerial vehicle (UAVs) platforms, whereafter the data are processed to yield the desired tree-level metrics/parameters in a cost-effective manner. Delineation errors occur at crown boundaries of overlapping trees due to a) lack of evident spectral or spatial difference between the neighboring crowns, b) intra-crown spectral variability, c) non-uniform illumination/shadowing, and d) under exploitation of the spectral, spatial contextual, and structural information in VHR multispectral data.                         



Researchers at the University of Toronto have developed an algorithm to delineate spruce tree crowns more accurately from VHR multispectral data. In this method, spectral, spatial contextual and structural information is jointly leveraged to reduce the effects of intra-crown spectral variance and non-uniform illumination/shadowing in crowns. In the novel approach, tree crowns are delineated by integrating the structural and spatial-contextual information modelled capability on a Fuzzy C-means spectral classifier. The framework of the process is illustrated in Figure 1. We refer to the algorithm as the Spectral, Spatial-contextual, and Structural Information based Individual Tree Crown Delineation (S3-ITD).

Graphical user interface, application, Teams</p>
<p>Description automatically generated

Figure 1.  Block scheme of the Spectral, Spatial-contextual, and Structural Information based Individual Tree Crown Delineation method (S3-ITD) crown detection and delineation method for VHR multispectral data. The geometric and radiometric corrections of the VHR data are performed using the photogrammetrically derived Digital Surface Model (DSM), and the known reflectance panel parameters, respectively. The set of treetops t correspond to local maxima in the Canopy Height Model (CHM) are detected using the Local Maxima detection (LM) algorithm. The crown-fractional-map (uc) generated using the Markov-Random-Field based spatial-contextual model (FCM-MRF) classifier is integrated with the ridge map (ur) obtained using the marker-controlled watershed segmentation in order to derive the ridge-integrated fractional map (urc). Spruce crown delineation is achieved by performing region-growing on the ridge-integrated fractional map (urc) using the Gradient Vector Field (GVF) Snake algorithm.



  • More accurate delineation of spruce crowns
    • An improvement in the average Intersection over Union (IoU) of 0.1 and 0.05, and the lower average absolute Crown-Area Difference (CAD) of 0.8 m2 and 0.2 m2 respectively, compared to the state-of-the-art Marker-controlled Watershed Segmentation (WS-ITD) and the Bias-Field Segmentation Algorithm (BF-ITD) methods.
  • Cost-effective collection and processing of canopy level data
    • Labor reduced due to the use of optical UAV data, fully-automatic end-to end processing, and scalability.



  • Tree crown delineation for forestry and agricultural applications
    • Biomass and carbon capture determination
    • Effect of abiotic factors on tree growth



  • Provisional patent application filed (April, 2021)



The S3-ITD delineation of tree crowns has been tested in the field.  The study area is in Saint-Casimir (Quebec) and contains both managed and unmanaged forests. A Micasense Red-Edge multispectral camera mounted on a DJI Matrice M600 UAV/drone were used to acquire the VHR images of eight circular plots (10 m radius).  Tree crowns were manually identified by an expert operator and compared to those delineated using the S3-ITD algorithm and two other benchmark state-of-the-art (SoA) algorithms (i.e. the WS-ITD and the BF-ITD algorithms).




Related Resources

VPRI Contact

Jae-Ho Yoo

Innovations and Entrepreneurship Manager
Innovations & Partnerships Office (IPO)