Calibrating images from different dates to 'like-value' digital counts

Sequences of satellite images are increasingly being used to monitor our environment. An important step in the development of methods for detecting, measuring, and monitoring change through time is the ability to compare images from different dates and sites in different scenes. These comparisons require the digital counts from each scene to be calibrated to common reference values. A method of relative calibration using robust regression and large numbers of likely invariant targets has been developed and implemented operationally. One or more images are calibrated to the digital numbers of a common reference scene. The technique has been used to calibrate over 100 Landsat Thematic Mapper (TM) scenes of the Western Australian wheatbelt. The method and its application are described. The repeatability of the calibrations over a sequence of images of the same scene from different dates is examined. (C) 2001 Elsevier Science Inc. All rights reserved.