多尺度高分辨率全球土地覆被遥感产品相对一致性比较

    Comparison of relative uniformity between GLOBCOVER andMODIS land cover data sets

    • 摘要: 国家及区域尺度的土地覆被信息对于解决环境演变、生物多样性保护、生态系统评价及环境建模等一系列问题起着至关重要的作用。该文从国家及区域尺度分别比较了当前全球2种最高分辨率的土地覆被遥感产品,以便解释两者在空间及专题上的一致性和异质性。结果表明,GLOBCOVER和MODIS 2种产品在国家尺度上具有较好的一致性,但在区域尺度上仍有较明显的差异,特别是在中国东北和西南地区;2种产品在林地、灌木、耕地和草地间存在严重的混淆现象;总体精度和Kappa系数在全国尺度和区域尺度有着较大的差异,总体精度值从全国尺度的56.35%降为区域尺度的27.01%,Kappa系数值从全国尺度的47.09%降为区域尺度的16.57%。在全国尺度上东北区域有着最好的一致性,四川盆地的总体一致性最差;MODIS产品的类别均质性百分比明显优于GLOBCOVER,类别一致性百分比与空间一致性之间存在显著的相关性,其决定系数R2为0.724,说明产品间的类别均质性百分比差异越小,两者的空间一致性越高。该研究可为遥感产品的使用者提供准确可信的信息分布, 为产品生产者有针对性改进分类算法提供科学合理依据。

       

      Abstract: The information on land cover at national scales is critical for addressing a range of problems, including climate change, biodiversity conservation, ecosystem assessment and environmental modeling. In this study, two of the most highly resolution global land cover products were compared: GLOBCOVER and MODIS Collection5, with resolution 300 m and 500 m, respectively, by using the relative comparison analysis to identify areas of spatial agreement and disagreement in national, regional and category levels. The result show that: there is a consistency in national scale, but there remain substantial inconsistencies and discrepancies in subarea, especially in Northeast and Southwest zone; Two products have a serious confusion between Forest/mixed forest, Woodland/shrub land, Cropland and Grassland, especially the Crop land between other categories; the gap for overall accuracy and kappa coefficient form national scale to subarea scale is conspicuous, the value vary from 27.01% to 56.35% for overall accuracy; the range vary from 16.57% to 47.09% for kappa coefficient. Northeast zone has the best consistency with the national scale in general, but Sichuan Basin has the worst consistency with the nation scale. MODIS has a well homogeneous category than GLOBCOVER, the relation between the difference value of homogeneous category and category consistency present an obvious negative correlation, the R2 is 0.724. The results can provide a reference for land cover research.

       

    /

    返回文章
    返回