基于混合像元分解的天山典型地区冰雪变化监测

Monitoring of Snow Cover Changes in Tianshan Mountains Based on Mixed Pixel Decomposition

针对中低分辨率遥感图像中存在大量混合像元,而传统的图像分类方法存在只能将某个像元归到某一类中,不能正确反映混合像元实际情况的问题。以新疆天山典型冰川覆盖区为例,根据TM/ETM+遥感图像的光谱特征,结合天山地区地表覆盖特点,在线性混合像元分解方法基础上,设计一种符合冰川地区特点的"冰雪-植被-裸露山体-阴影"端元组分模型。通过选择合适的端元并将其反射率值代入改进后的且满足约束条件的线性混合像元分解模型,得到各端元组分丰度图,进而精确提取出冰雪信息并计算其面积。1989年TM和2000年ETM+遥感图像冰雪信息提取结果表明,运用线性混合像元分解模型能很好地监测实验区的冰雪覆盖变化情况。

Mixed pixels are abundant in medium -low resolution images, but the traditional methods for image classification could only assign pixels to one class,with the ignorance of the mixed pixels. To tackle this problem, the authors selected the typical area in Tianshan Glacier of Xinjiang as an experimental area. Based on the theory of mixed pixel decomposition and the principle of the linear model and taking into account the spectral characteristics of TM/ETM ＋ image as well as the land cover characteristics of Tianshan area, the authors developed an end - member composition model suitable for the glacier area,i, e. , Snow - Vegetation - Rock - Shade model. After the appropriate end - members were selected and the reflectance values were substituted into the improved linear mixed pixel decomposition model, which satisfied the constraints, the abundance image of individual end - member was calculated and the snow cover information was easily and precisely extracted. The extraction results of snow cover in 1989 and 2000 demonstrate that the mixed pixel decomposition and the linear model could be used to monitor the snow cover changes in the glacier area.