A land cover distribution composite image from coarse spatial resolution images using an unmixing method

A method of sub-pixel land cover estimation including an algorithm for minimizing missing data due to cloud cover was proposed for the purpose of evaluating and monitoring the environment of wide areas. A pair of Landsat Thematic Mapper (TM) scenes over coincident multitemporal National Oceanic and Atmospheric Administration (NOAA) Advanced Very High Resolution Radiometer (AVHRR) time-series of directional hemispherical reflectance were used to develop a fine-scale land cover map using either eight or three categories and to estimate the endmembers of the AVHRR image using a positive constrained linear least-squares method. Furthermore, three approaches were evaluated for compositing sub-pixel estimates over cloudy areas in the AVHRR image. Finally, from validation tests made for unmixing and compositing methods, the results suggest that these methods may be generally useful for comparing multispectral images in space and time.