A new combined classification method for land covering using AVHRR data supervised by TM data

High resolution satellite images with wide area coverage are usually expensive and rarely obtained, but the precision of classification basically depends on the resolution of the satellite images. In this paper, a new combined classification method using low-resolution satellite data (NOAA AVHRR data) supervised by high-resolution satellite data (LANDSAT TM data) was proposed. In this method, some sample areas of TM data were classified by unsupervised cluster at first, and then the likelihood functions were estimated for each class according to the "multi-single" spatial correspondence between TM and AVHRR pixels. Finally, the whole AVHRR image can be classified with these likelihood functions by maximum likelihood classification method. The result of our experiments shown that, the accuracy of the classification greatly depended on where and which the sample areas were selected, the precision of the classification would be equal to those classified by high resolution satellite images directly if the sample areas were selected properly.