基于决策树—山体阴影模型的植被信息提取研究

植被是土地覆被分类的重要内容,植被分类对研究流域生态具有重要参考价值。鄂西犟河流域是南水北调中线工程的重要水源地,该区以中低山和丘陵为主,地形起伏剧烈,导致遥感影像存在大量阴影,制约了植被分类的精度。基于LandSAT OLI影像,使用ArcGIS的Hillshade模块,输入DEM数据和传感器具体参数,计算得到影像成像时刻阴影的准确分布;统计野外采集地物样本点在MNDWI、NDVI和RVI等指数上的差异;结合决策树分类法,分别设定阴影和非阴影下6类样本的阈值进行分类。结果表明:(1)该方法总分类精度能够达到92.93%,Kappa系数为0.912;(2)阴影和非阴影区植被具备明显的同物异谱和异物同谱特征,表现为阳面的植被指数整体高于阴面;3类林地的RVI值由高往低依次为:灌木,混交林和针叶林。(3)传统经验模型在不同纬度的适用性不同,无法精确提取阴影的范围,降低了分类精度;决策树—山体阴影模型作为一种智能分类方法,能够还原Landsat OLI影像准确的阴影分布,提高山地和丘陵等阴影面积大、形状复杂区域的植被分类精度。

Using Decision Tree and Hillshade Method to Improve the Accuracy of Vegetation Classification

Vegetation plays an important role in land use and land cover. Vegetation classification has a great reference value for watershed ecology research. Jianghe river basin in western Hubei is an important water source for the Middle Route of the South-to-North Water Transfer Project. Because of its large altitude range, complicated terrain, a large number of complex shadows exist in remote sensing images. It is difficult to classify the vegetation types in the shadow. This paper constructed a stratified analysis for vegetation classification. The stratified classification is a method based on the idea of division of layers step by step and different criteria and methods in each layer. The criteria included the vegetation indices, water index and shadow index, which were calculated from Landsat OLI image of 2017 and DEM.The vegetation indices included NDVI (Normalized Difference Vegetation Index) and RVI (Ratio Vegetation Index).The study selected Modified Normalized Difference Water Index (NMDWI) to indicate water. The topographic shadow was calculated from the hill-shade module based on ArcGIS. The input parameters of hill-shade model consisted of solar azimuth angle, solar zenith angle and ASTER GDEM. The results showed that: (1)Using the intelligent decision tree algorithm, the overall accuracy of the classification reached 92.93%, with a Kappa coefficient of 0.912; (2) For the same vegetation, there were higher NDVI and RVI in exposed halves of hills than that in shaded halves. It was consistent with the phenomenon of same object with different spectra. The value of RVI from high to low was: deciduous shrub, thedeciduous shrub and evergreen coniferous mixed forest and evergreen coniferous forest. (3) Because of different solar azimuth angle and solar zenith angle changing the distribution of shadow, each remote sensing image had a certain shadow shape. The accuracy of vegetation classification is largely reduced. Overall, this vegetation classification method combined the decision tree and hill-shade model to categorize the Jianghe river basin into six classes. As an intelligent decision tree classification algorithm, the method has the advantages of identifying vegetation categories in shaded halves and is of high application value, especially in low mountains and hills. There is still space to improve the classification accuracy, i.e., an optimized vegetation index, the increasing of vegetation samples in high altitude and large slope regions, and the use of higher spatial and temporal resolution images.