基于MODIS影像的土地覆被分类研究——以京津冀地区为例

在全球变化研究中,如何快速、准确获取土地覆被信息对该项研究有着至关重要的作用.随着遥感科学的不断发展和应用领域的深入,研究者可以利用遥感影像进行土地覆被分类研究,并且具有准确、快速、自动化等优点.本文利用MODIS数据具有的多光谱、多时相特点,以京津冀地区为例,选取2013 年全年16-day 的MOD13Q1/EVI时间序列数据、2013 年5 月份一期的MOD09Q1(1、2 波段)和MOD09A1(3-7 波段)产品,并运用时间序列谐波分析法对全年MOD13Q1/EVI 时间序列数据进行去云、去噪的平滑重建处理,使其数据更能反映物候周期性变化规律.选择谐波分析后的全年MOD13Q1/EVI 时间序列数据、MODIS数据的1-7 波段地表反射率和NDWI(归一化差异水体指数)、MNDWI(改进归一化差异水体指数)和NDSI(土壤亮度指数),构建了3 种特征变量组合方案的CART决策树,分别进行京津冀地区的土地覆被分类研究.结果表明:方案一(全年EVI 的23 个时相)、方案二(方案一+MOD09 的1-7 波段地表反射率)和方案三(方案二+MNDWI+NDSI+NDWI)的总体分类精度分别达到86.70%、89.98%、91.34%,Kappa系数分别为84.94%、88.66%、90.20%.研究表明,仅利用MODIS遥感影像自身多种分类特征和决策树方法对宏观土地覆被分类就可达到较高精度,显示了本文分类方法在实践中的可行性及MODIS数据在区域尺度土地覆被分类研究方面的优势与潜力.

Land cover classification based on MODIS images:taking the Beijing-Tianjin-Hebei region as an example

With intensifying human activities global ecological and environmental problems have become increasingly pressing. Therefore the study of global change has become more prominent. Obtaining accurate land cover information in a timely fashion is critically important for such research. With the development of remote sensing science and technology and application, numerous studies have the issue of land cover classification using remote sensing image. In this study, the Beijing-Tianjin-Hebei region was selected as a study area for land cover classification using MODIS data. The 16-day MOD13Q1/EVI data in 2013, MOD09Q1 (Band1, 2) and MOD09A1(Band3- 7) products in May 2013 were used as the basic data. Harmonic analysis method was employed to remove clouds and noises of the whole year's EVI, so that it can better reflect plant phenology. Then, the processed MOD13Q1/EVI data, surface albedo of MODIS/Ref1-7, modified normalized difference water index (MNDWI), normalized difference soil brightness index (NDSI), and normalized difference water index (NDWI) were integrated to construct three schemes of CART decision tree to investigate the land cover classification of the Beijing-Tianjin-Hebei region. The three band combinations are scheme1: 23 phases of EVI of 2013; scheme2: 23 phases of EVI of 2013 plus B1-B7 of MOD09; and scheme3: scheme2+MNDWI+NDSI+NDWI. The results show that the overall classification accuracy of the three schemes reaches 86.70%, 89.98%, and 89.98% respectively. Their Kappa coefficients are 84.94%, 88.66%, and 90.20% respectively. Therefore, MODIS images with various classification characteristics, combined with the decision trees can achieve higher precision land cover classification. This proves the feasibility of the proposed method for land cover classification at the regional scale.