基于MODIS NDVI时序数列的柴达木盆地土地覆盖分类研究  

Land-cover classification of the Qaidam Basin based on MODIS NDVI time series data

作　　者：马俊飞[1] 吕昌河[1] 王茜[1] 

机构地区：[1]中国科学院地理科学与资源研究所

出　　处：《干旱区地理》2008年第3期442-448,共7页Arid Land Geography

基　　金：自然科学基金项目(40671007)资助

摘　　要：以2003年柴达木盆地的MODIS卫星遥感影像为基础数据,结合野外实地考察资料,综合分析了2003年柴达木盆地的植被指数时间变化与空间分布特征,提取了该年度该区域各主要土地覆盖类型NDVI的时序数列曲线,采用NDVI时序数列变化曲线形状匹配方法对柴达木盆地进行了土地覆盖类型分类。在此基础上使用了将以月为单位的变化曲线转换为十二维空间中的单位向量,比较向量夹角以决定其相似度的先进算法,取得了良好的分类效果,获得了2003年柴达木盆地比较精确的土地覆盖分类图。提出了利用MODIS卫星遥感影像进行土地覆盖分类的新方法。The Normalized Difference Vegetation Index （NDVI） can well reflect the status of vegetation and landcover. Its time series has great significance for land-cover classification since they are very sensitive to vegetation＇ s temporal changes and can more precisely distinguish the vegetation type, which is a very important characteristic of the land-cover. In this paper, using MODIS remote sensing image of the Qaidam basin in the year of 2003 as the source data, we calculate the NDVI and compile the NDVI time series. To avoid the impact of the cloud and guarantee the accuracy of the calculated NDVI time series, a maximum value composing （MVC） process is implemented after the geometrical and radiometer calibration for the MODIS data. In the year of 2003, we invested the Qaidam basin and made samplings of its major vegetation types. Abundant field survey data were got. Based on the NDVI time series and field survey data, this paper analyzed the temporal changes and spatial distribution of the NDVI in the Qaidam Basin. It gave the NDVI time variation curves and the coefficient of variation（COV） of major landcover types in the Qaidam Basin for the year of 2003 7ahich were then employed as the base for the land-cover clas- sification using a newly innovative method namely the NDVI time variation curve＇ s shape matching method. The method compares the NDVI time variation curve of an unclassified region to the curve of each of 9 given land-cover types. The land-cover type whose curve most matches the curve of the unclassified region determines the region＇ s classification type. To make the shape match more accurate, the monthly NDVI variation curves were transferred in- to a twelve-dimension space where one vector with a particular angle represents the original curve with the aid of ENVI software. The extent of similarity and the classes the vectors belong to were determined by comparing the angle of vectors. With the approach, a relatively precise land-cover classification result whose precision was vali

关 键 词：土地覆盖分类 MODIS NDVI 

分 类 号：TP79[自动化与计算机技术—检测技术与自动化装置]