Automatic template-guided classification of remnant trees

Spectral features within satellite images change so frequently and unpredictably that spectral definitions of land cover are often only accurate for a single image. Consequently, land-cover maps are expensive, because the superior pattern recognition skills of human analysts are required to manually tune spectral definitions of land cover to individual images. To reduce mapping costs, this study developed the Template-Guided Classification (TGC) algorithm, which classifies land cover automatically by reusing class information embedded in freely available large-area land-cover maps. TGC was applied to map remnant forest within six 10-m resolution SPOT images of the Vermilion River watershed in Alberta, Canada. Although the accuracy of the resulting forest maps was low (58% forest user's accuracy and 67% forest producer's accuracy), there were 25% and 8% fewer errors of omission and commission than the original maps, respectively. This improvement would be very useful if it could be obtained automatically over large-areas.

Author

• Kennedy, Peter

• automatic classification
• land-cover classification
• map reuse

Identifier

hdl:10133/3616

OAI identifier oai:identifier

oai:opus.uleth.ca:10133/3616