A Robust Approach for Object-Based Detection and Radiometric Characterization of Cloud Shadow Using Haze Optimized Transformation

Cloud shadows in satellite imagery hinder understanding of ground surface conditions due to reduced illumination and the potential for confusion with illuminated low-reflectance objects such as water bodies. This paper extends the application of the haze optimized transform (HOT) from haze mapping to include object-oriented detection of clouds and cloud shadows. An integrated processing chain encompassing these tasks has been implemented and successfully applied to Landsat Enhanced Thematic Mapper Plus and Multispectral Scanner imagery covering a variety of land covers and landscapes. The results confirm that the HOT-based method for cloud shadow detection is robust and effective. Cloud shadows have been identified and extracted with overall accuracy of about 95.3%. Clear-sky dark pixels (e.g., small lakes) are well separated from cumulus cloud shadow pixels. The spatial distribution of HOT response in a given cloud patch can be used to estimate the extent and variation of incoming visible radiation reduction in its corresponding shadow patch. This information, in turn, has been used to apply a radiometric gain to compensate for the shadowing effect on the land. The HOT response has been tested for radiometric characterization of cloud shadows and subsequent shadow illumination compensation.