基于遥感影像的峻河流域高寒灌丛决策树提取方法

本文选择青海湖流域内一个具有代表性的小流域—峻河流域为研究对象，通过该区域 IKNOS-2 高分辨率影像的分析，发现该区域高寒灌丛的分布与海拔、坡度、坡向等因素密切相关。 据此，将研究区1:5 万DEM 数据融入TM 影像植被分类过程中，建立一种新的决策树分类方法， 结果将分类总体精度提高到89.37%，Kappa 系数提高到0.7875，达到一般分类结果的精度要求。 这说明，加入多源数据，尤其是地形数据，能够显著提高高寒灌丛植被的分类精度。

Decision tree interpretation method based on remote sensing data of alpine shrubs in Jun River watershed, Lake Qinghai, China

Remote sensing data were interpreted to mapping shrubs in Jun River watershed, which is a semiarid alpine sub-watershed of Lake Qinghai basin in northeastern Tibetan Plateau. At fi rst, traditional unsupervised classification method (ISODATA) was applied to extract shrubs information from the Landsat image and the yielding overall classifi cation accuracy is only 67.11% and a Kappa coeffi cient 0.3419. This is mainly because of the mixture of the spectrum associated with the complicated t opography in the study area. Previous studies and IKNOS-2 high-resolution image suggested that distribution of shrubs in Jun River watershed is dominated by topographic variables, such as altitude, slope, and aspect. Therefore, we set up a decision trees together with DEM datum to classify the Landsat image for the whole Jun River Watershed and obtained an overall classifi cation accuracy of 89.37% and a Kappa coeffi cient 0.7875. It suggests that this method can effectively improve the accuracy of shrubs classifi cation and can be applied in the whole Lake Qinghai basin and even the Tibetan Plateau. The vegetation changes recontraucted by the remote sensing data would help us better evaluate the potential impacts of huma n activities and climate variability on vegetation in the Lake Qinghai basin.