Skip to Main Content
This letter presents a technique to derive cloud- and shadow-masks from satellite images. The technique is based on a mutual confirmation algorithm that uses all available solar bands, but no thermal infrared bands. The process of mutual confirmation consists of first searching and then identifying partial geometrical overlap between first-guess (preliminary) bright (cloud candidates) and associated dark (cloud shadow candidates) pixels in the image. The thresholds needed for the algorithm to operate are adjusted dynamically and are scene dependent, but need no additional information about radiometric calibration or solar geometry. The algorithm's success rate (percentage of correct classification compared to a manual/visual mask) when applied on a handful of Landsat-5 scenes is found to exceed 95% for cloud pixels nd 90% for shadow pixels.