基于EOS／MODIS资料的中国黄土高原西部土地覆盖分类

Land Cover Classification in Western Loess Plateau of China with MODIS Imagery Data

土地利用和覆盖信息不仅是全球及区域气候模式中所需的重要信息,也是描述生态系统的重要基础数据;获取其现状和变化信息,对于揭示土地覆盖的特征和变化规律,探讨变化的驱动因子,分析评价区域生态环境具有重要现实意义。采用多时相和多光谱遥感资料进行土地覆盖分类已成为一种行之有效的方法。植被类型的差异除了可表现为光谱差异外,还可表现为植被生长规律的差异;植被生长以年为周期,在这个周期内不同植被类型有着各自的生繁衰枯的物候节律,表现出不同的生长规律,其规律性极强,这种规律性也可以作为植被分类的一个出发点。于是,NDVI的时间序列分析成为基于生物气候特征的地表覆盖分类的一种手段。基于MODIS资料,首先对2003年植被生长季内共9个月的NDVI时间序列数据进行离散傅立叶变换,取得该时间序列的均值和前4个频率分量的振幅和相位后;采用ISODATA算法对中国黄土高原西部及其周边地区进行了分层土地覆盖分类,并采用现有的土地覆盖数据集,进行了类别归并和精度评估,分类结果的总体精度达到了81.3%,Kappa系数达到了75.4%,结果证明了这种方法的可行性。

Land cover and use are not only the important information for climate model of globe and region, but also the base character data of ecosystem. It has vital meaning for obtaining the spatial state and variety information of land surface, investigating the driving factor and analyzing and evaluating the regional ecosystem. Land cover classification using the multi-temporal.and multi-spectral remote sensing observation data has been an effective technique. In addition to spectrum difference, the vegetation divergence is displayed by the variation of growing pattern. The plant growing period is one year and in this period the different plant types have different phenological successions and different growing patterns that go round and round regularly and this can be used as a point of departure to classify plant. Hence, the NDVI time sequence analysis became the basic means of land cover classification according to biometeorology characteristic. In this paper, based on MODIS data, we perform the discrete Fourier transformation for NDVI of nine months in the plant growing season of 2003 and gets the mean value and the amplitude and phase of the former four harmonics. According to these data, the land cover classification in the West Loess Plateau of China and its surrounding area is finished by ISODATA method and the classification result merging and accuracy assessment are accomplished by using the current land cover maps. The calculatedoverallaccuracywas 81.3%, while Kappa was equal to 75.4%. It shows that these methods arevalua ble for practical use.