Quantifying tree cover in the forest-grassland ecotone of British Columbia using crown delineation and pattern detection

Tree covers in the ecotone between forest and grasslands of British Columbia (BC) vary from less that 15% to over 35% and have increased in the last century. Recent developments in remote sensing focus on high resolution, multi-spectral digital images. For the detection of historic changes in tree cover, however, black-and-white aerial photographs are the only available images in most regions. In addition, most tree delineation techniques are better suited for dense forests than for scattered trees. We tested the usefulness of black-and-white aerial photographs in image analysis for studying tree encroachment using commercially available remote sensing and GIS software within a 100 ha area of grasslands and forests. Tree crown, tree shadow, and grassland were first separated sing the maximum likelihood classifier. Tree-shadow contacts were used to model circular crown masks. A moving sample window was applied to calculate tree cover and the resulting image was classified into open grassland, treed grassland, open forest, and closed forest. Results showed that approximately 27% of areas occupied by open grassland was lost to tree encroachment between 1966 and 1995. Compared to field measurements, tree cover from image analysis was 0.8, 2.6, and 1.0% lower for open grassland, treed grasslands, and open forests, respectively, but the error increased significantly for dense forest (12.4%). The under-estimation of tree cover was possibly caused by limited resolution of the black-and-white aerial photographs, and the omission of small trees and generalization of tree groups during image classification. The accuracy of this technique was acceptable for the concerned grasslands and open forests, but improvement is needed for dense forest areas.