遥感影像降尺度分类主要用于在亚像元尺度下精确地表达土地覆被分类结果。目前，大多降尺度分类方法是基于空间相关性的系列方法。尽管这些方法能有效处理大于像元尺度的面状地物，但线状和点状地物处理结果不理想。尤其对同时包含面状,线状和点状地物的影像仅采用一种基于空间相关性的方法来处理，会导致一些具有离散分布的点状地物聚集，也难以保证线状地物的连通性。此外，目前降尺度分类主要集中在算法研制方面，缺乏其在应用中的探讨，尤其是降尺度分类结果对高分影像的替代性评价。鉴于此，本项目拟首先提出一种基于地物空间分布模式的降尺度分类策略，将地物分布模式划分为面状、线状和点状，逐一采用相应的方法处理，克服仅利用一种方法来处理多种地物分布模式策略上的不足，提高点状和线状地物的精度。然后，探测遥感影像降尺度分类方法在不同典型试验区的效果，并探讨降尺度分类结果与同等高分结果的比较分析与适应性评价。

SLand cover downscaled classification, also termed sub-pixel mapping, is mainly used to obtain a more accurate land cover classification results at a sub-pixel scale. Currently, the majority of sub-pixel mapping methods are based on the characteristics of spatial auto-correlation of surface objects. These methods have been proved to be an effective way to process the area feature greater than the pixel scale. However, the results are less satisfactory for the linear and point feature. Especially, for an image contains both area, linear, and point features, spatial autocorrelation based sub-pixel method will lead to the opposite result that the spatial discrete point features are gathering together. It is also difficult to guarantee the spatial connectivity of linear feature. Meanwhile, the research of sub-pixel mapping focuses mainly on the development of sub-pixel mapping algorithms and fails to explore its applications in geosciences. For example, the accuracy differences between sub-pixel mapping results and hard classification results from high resolution remote sensing images in some representative study areas play an important role in its further application in other fields, like ecology and hydrometeorology. To overcome these weaknesses of spatial auto-correlation-based sub-pixel mapping, this project plans to develop a sub-pixel mapping strategy based on the spatial pattern of surface object. First, the objects in the image are divided into area feature, linear feature, and point feature. Then, these features are processed by the corresponding sub-pixel method. Therefore, this kind of method will overcome the weaknesses arise from a single sub-pixel mapping methodology to deal with a variety of feature distribution patterns, in which to improve point and linear feature sub-pixel mapping accuracy. Furthermore, the performance of sub-pixel mapping methods in some representative study areas would be investigated and the accuracy differences between sub-pixel mapping results and hard classification results from high resolution remote sensing images would be evaluated.