The use of contextual information to detect cumulus clouds and cloud shadows in Landsat data

Abstract

Abstract The use of contextual information to identify areas of cloud and cloud shadow in digital Landsat data is described. The distance from cloud to shadow is calculated from sun elevation and azimuth, cloud base height and image geometry. Pixels with a grey level above a given threshold (Tc) are considered to be potential cloud. A 3 × 3 pixel window centred at the calculated shadow position is searched. Pixels in this window with grey levels below a second threshold (Ts) are classed as cloud shadow. If no shadow pixels are found the potential cloud is rejected. Modifications to allow for overlapping cloud and shadow are included The detection procedure was found to work well for bands 5, 6, or 7 of Landsat data. Use of combinations of bands gave no advantage over the use of a single band. Small errors in distance and grey level thresholds could be tolerated Evaluation is hindered by lack of suitable ground data. However, for one scene where limited data were available, it was estimated that at least 8...