Canadian Forest Service Publications
Use of high spatial resolution satellite data as a calibration data source to aid in the cluster labeling when classifying Landsat data. 2001. Wulder, M.A.; Sánchez-Azofeifa, G.A.; Hamilton, S.; Dutchak, K. Pages 27-33 (Vol. 1) in M. Bernier and C. Duguay, editors. 23rd Canadian Symposium on Remote Sensing, Proceedings: Remote sensing in the third millennium: from global to local. August 21-24, 2001, Université Laval, Sainte-Foy, Québec. Canadian Aeronautics and Space Institute, Ottawa.
Issued by: Pacific Forestry Centre
Catalog ID: 26715
Availability: PDF (download)
The robust classification of Landsat data requires data to calibrate between the reality on the ground and the corresponding image representation. Often ground data is difficult or expensive to collect. Further, in Canada where approximately 5 million km2 are considered forested, much of the forest is not monitored through industrial or provincial/territorial forest monitoring activities on a regular basis. An issue that is compounded also by the lack ground data or aerial photography to help during pre-and post classification processes.
High spatial resolution remotely sensed data, such as the approximately 6 meter Indian Remote Sensing (IRS) 1C and 1D and 1 meter IKONOS panchromatic data, may be used as calibration data source where field or inventory data is unavailable. The labelling of clusters generated on Landsat imagery may be aided by interpretation of IRS imagery fused with the Landsat multispectral data, allowing for expert knowledge and texture information as basic entry levels.
In this communication we describe the method for clustering and labelling Landsat data using IRS and IKONOS data as a complementary information sources. The high spatial resolution data is used in a similar manner to air photographs, and due to the digital nature of the data, it has additional capabilities when fused with Landsat multispectral data. The fused data provide appropriate information for the cluster labelling of a limited number of classes. Recommendations are made based upon the use of high spatial resolution data as a surrogate data source to aid in cluster labelling for classification in northern or remote Canadian locations.