Cloud shadows are a frequent problem affecting airborne and satellite data from many parts of the world. A number of different techniques have been developed to suppress the effect of cloud shadows but these only work effectively if the image data are aligned to an hypothetical 'dark point virtual endmember' (DPVE) in the multispectral feature space. The DPVE may lie outside the cloud of points forming the image data, so it is not sufficient to simply subtract the digital number of the darkest pixel in each band. This paper identifies the DPVE on a number of images by the convergence of lines of constant hue in a 2-dimensional feature space. These lines are derived from regions-of-interest that straddle the boundary of sunlit and cloud shadow areas on bare soil and green vegetation. Aligning the data set to an accurate DPVE in this way is shown to greatly improve the ability of the simple NIR/Red ratio to suppress shadows from clouds as well as those from building and within forest canopies.
展开▼