A spatial clustering approach to automated tree crown delineation

Forest resource information is increasingly needed at fine spatial scales for use in operational to strategic management programs. Applications include multiple-use management, planning harvesting operations and silvicultural prescriptions, and ensuring the maintenance of biodiversity and ecological sustainability. High resolution remotely sensed imagery is one data source that has demonstrated, and continues to demonstrate, great promise. The benefits of high spatial resolution data include the potential to apply algorithms capable of automatically delineating individual tree crowns in the imagery. These algorithms commonly search for distinct spectral patterns in the forest scene and use specific image features for the automated delineation of individual tree crowns. These include the spectral maxima and minima, being indicative of crown centroids and boundaries respectively. This paper describes a threshold-based spatial clustering approach to tree crown delineation. The algorithm is designed for application in Australian native forests, where the dominant genus, Eucalyptus, typically exhibits low foliage density and complex crown structure. Algorithm features designed to minimise crown segmentation were therefore key considerations. The effects of distortions in high resolution imagery is also discussed, particularly those variables that will influence the spectral 'topography' of the forest canopy, notably sun angle and viewing geometry. To achieve this, a 3-Dimensional simulation model has been developed which allows bi-directional reflectance and off-nadir viewing angle effects to be investigated.