中文摘要：

新疆北部冬季常多云天气,受地形和植被等因素影响,积雪分布不均,这导致了积雪分量遥感制图精度不高的难题。针对这一问题,本文利用MODIS影像,提出了分层变端元混合像元分解的积雪定量反演方法。该方法首先建立研究区影像和参考端元库,再对区域影像地类进行从粗到细的逐级划分,对每级包含的地类进行2端元或3端元混合像元分解。通过对上一级解混后与积雪相关的子类再实施上述变端元混合像元分解方法,得到下一级更细的地类划分。综合各级地类解混结果实现了高精度积雪分量制图。以HJ CCD影像分类数据、植被分布数据和实测数据为数据源,验证了变端元解混时模型包含端元数较少时（如2端元或3端元）研究区域积雪分类精度最高,为87%。反之,包含端元数越多,分类精度会降低。

英文摘要：

Due to cloudy days often in winter and an influence of topography and vegetation as well as uneven snow cover distribution in the area over northern Xinjiang,it is very hard to achieve a snow fraction mapping product with high accuracy using remote sensing images.In order to make snow fraction mapping more accurate,a snow quantitative inversion method based on hierarchical dynamic endmember spectral mixture analysis（DESMA）technology for snow fraction mapping using MODIS data acquired in the study areas suggested.The details of the proposed method areas follows：Image endmembers libraries and reference endmembers libraries were initially built.Then,multi-level category analysis from a coarse to a fine scale on the entire image of the study area was carried out.Two-endmember or three-endmember models for the whole MODIS data was used to unmix each pixel at each level and all pixels were classified into one of two categories：pixels containing snow endmembers and pixels without snow endmembers.Only those pixels containing snow endmembers were further unmixed using DESMA technology to achieve a more refined classification in a finer layer.Unmixing layer by layer,it produced a classification result for each layer.Finally,a snow fraction mapping product with higher precision was generated by combining classification results from each layer.The experiments showed that the suggestion of unmixing using DESMA technology when a smaller number of endmember models（e.g.two-or three-endmembers）,is selected then a snow fraction mapping product with the highest overall classification accuracy of 87%can be achieved.The accuracy of snow fraction mapping will be lower if the number of endmembers used in the unmixing model is larger than three.This suggestion has also been verified by snow-cover and vegetation distribution maps derived from HJ CCD remote sensing data and field data of snow-cover.