Canadian Forest Service Publications
Hierarchical image classification and extraction of forest species composition and crown closure from airborne multispectral images. 1998. Gerylo, G.; Hall, R.J.; Franklin, S.E.; Roberts, A.; Milton, E.J. Canadian Journal of Remote Sensing 24(3): 219-232.
Issued by: Northern Forestry Centre
Catalog ID: 18757
High spatial resolution digital multispectral (DMS) images were acquired at a pixel resolution of 32 by 25 cm to determine the extent forest stand and vegetation detail could be derived from existing per-pixel and spatial feature-based image analysis methods. The image data were acquired from approximately 150 m above a mature forest ecosystem near Barrier Lake in Kananaskis Country, southwestern Alberta. Alberta Vegetation Inventory (AVI) data, including species composition and crown closure, were collected at 22 plots scattered throughout several pure and deciduous and coniferous dominant mixedwood stands. Image classification accuracy was determined for DMS data using maximum likelihood classification techniques applied in a hierarchical fashion, and to AVI class labels at decreasing levels of detail. Initial accuracy was low, but the hierarchical decision process by which image classes were merged, eliminated, or accepted, increased average accuracy to over 65% for a limited selection of stand types. Spatial feature-based methods of image analysis were employed using a series of filtering, classification and spatial operations to separate individual attributes such as tree crowns, understory, and shadows resolved in the image data. There were no statistical differences between crown areas measured at the plot level compared to similar measurements derived from the digital image when delineated by a Laplacian filter. Species composition classification accuracy was higher for trembling aspen (89%) than for lodgepole pine (80%) and white spruce (84%). A contextual classifier was subsequently used to construct a forest composition label for species composition and crown closure.