Forest Clearing with SPOT-5 [article]

Neil Flood, Tim Danaher, Tony Gill, Peter Scarth
The government of the Australian state of New South Wales (NSW) has a program to monitor the clearing of trees across the whole state, using remotely sensed imagery, originally based around Landsat. In order to extend this program to use the higher resolution SPOT-5 sensor, an index was fitted to a large set of training data, mapped over previous years, to detect clearing in an automated fashion from bi-temporal imagery, separated by approximately one year. A number of different forms for this
more » ... ndex were tested, and the best performing one was selected. The resulting index is used to produce an initial classification image showing areas likely to be clearing, which is then manually edited to produce a high accuracy map of clearing over a given year. This initial classification has been found to greatly ease the process of manual classification. The initial classification has a number of probability levels, which have user's and producer's accuracy for the clearing class of around 90%, while the user's and producer's accuracy for the non-clearing class are in excess of 99%. The index has been trained using data in eastern Australia, but would probably be of use in other parts of Australia.
doi:10.6084/m9.figshare.12980897.v1 fatcat:ozkdw2wwpvbx3golc54hwdqwsy