Fusion of AVHRR and TM data for vegetation classification based on unmixing technique

Vegetation classification using satellite data requires various aspects, e.g. high data acquisition frequency or high spatial resolution. The authors propose a data fusion method for vegetation classification fusing those different types of satellite data. In the method, an unmixing technique is applied for the classification of temporal low-spatial resolution images. It clarifies areas completely classified with high reliabilities, and simultaneously it leaves areas ambiguously classified with low reliabilities. Each ambiguous area has possibilities of pure land cover or mixed land covers, and has candidates of those land covers. The classified result from low-spatial resolution data is used to extract training data for the classification on high-spatial resolution data level. The proposed method was applied for Indo-China peninsula by use of AVHRR and TM data. Classification results were verified compared with visually interpreted images, and show high classification accuracy both forest and agricultural land covers