中文摘要：

植被类型是反映气候变化和生态环境变化最直接、最敏感的指示器。草地作为青藏高原最主要的植被类型,对其进行分类研究可以为高原草地退化、荒漠化等研究提供直接的信息,也可以为全球变暖、冻土退化等提供间接的分析数据,同时还可以为陆面模型的模拟提供重要的参数。大面积区域的植被分类可以通过将研究区内的位置、纹理和地形属性等信息综合考虑来完成,面向对象的植被分类方法可以通过对研究区多重信息进行分割和合并,来生成植被类型图。而且,该方法可以克服传统决策树分类方法成图时像元分散的缺点。利用TM假彩色合成影像、ASTER DEM数据和MODIS的EVI和LST产品,对玉树地区的高寒草地类型进行了面向对象的划分,总体精度为49.32%。虽然相比仅用单独决策树分类方法的总体精度略微偏低,不过该方法可以在保持环境、地域等因子同植被类型统计关系的基础上,克服决策树方法带来的＂椒盐效应＂。此外,植被模型的物理过程、参数计算和获取环境因子等都比较复杂,本研究的方法也可以为植被分类制图提供一种简单,快速的方法。

英文摘要：

The classification of vegetation has attracted much attention for study of ecological effects on the Qinghai--Tibetan Plateau. Previous studies have mostly focused on decision tree classifiers, and much research has been done to test this classification on a small scale. In this study, we consider a large scale method （ob- ject-oriented classification）, which can also be integrated with a conventional decision tree classifier. However, the rules of classification have only utilized the information from decision tree classifiers. This approach com- prehensively considered information of position, terrain and texture from TM （thematic mapper）, DEM （digit- al elevation model）, EVI （enhanced vegetation index） and LST （land surface temperature） in Yushu, and then segmented or merged the type of steppe. The overall accuracy is 49.32~, and Kappa coefficient is 0. 373 5. Our study suggested that this method could overcome the disadvantages of scattered pixels when division is by the type of vegetation. Compared to the conventional decision tree classifier, the overall precision of our meth- od is low. However our method maintained the statistical relationship between factors derived from the envi- ronment and geography, and vegetation types to reduce the salt and pepper effects. In addition, the physical process, parameter calculations and environmental factor collection of vegetation models are complicated. In this paper, a simple and quick way of division of vegetation types is provided by our method.