多种遥感土地覆盖产品一致性分析与精度评价——以淮河流域为例

土地覆盖(Land Cover)数据是人们进行气候变化探究、生态系统评估与地理国情监测等研究的重要数据源。随着研究的深入,国内外产生了众多不同时空尺度、不同分类体系的土地覆被分类产品集,但这些产品在局部尺度应用时的一致性与精度还有待分析。本研究以淮河流域为研究区域,基于5种土地覆被分类产品(CLCD、ESACCI-LC、GLC_FCS30、GlobeLand30、MCD12Q1),将各土地覆被类型重分类为农用地、建设用地、未利用地三大类,从面积和空间两方面进行了一致性分析与精度评价。结果表明:5种产品对于淮河流域土地覆被的类型特征具有较强的一致性,面积估算的相关性系数均大于0.9; 在3种30m分辨率土地覆被分类产品中,CLCD的识别精度最高,以此作为参考数据;其他四种产品总体精度在92.37%至95.53%之间,Kappa系数在0.523至0.695之间,GlobeLand30产品与GLC_FCS30产品精度较高、且各有优势,ESACCI-LC产品和MCD12Q1产品精度较低。研究成果为不同时空尺度上的土地覆被研究提供参考。

Consistency analysis and accuracy evaluation of various remote sensing land cover products -- a case study of Huaihe River Basin

_{Land Cover is the distribution of matter on the earth"s surface. Land Cover dataset is a critical data source for people to conduct research on climate change, ecosystem assessment and geographical conditions monitoring.As the developing of research, many sets of Land Cover classification products with different spatiotemporal scales and different classification systems have been produced by domestic and foreignresearchers, but the consistency and accuracy of these products in the application of local scales still need to be analyzed. In this paper, taking the Huai River Basin as the research area, five land cover classification products (CLCD, ESACCI-LC, GLC_FCS30, GlobeLand30, MCD12Q1) were reclassified into three categories: agricultural land, construction land and unused land, and the consistency analysis and accuracy evaluation were carried out from the aspects of area and space. Results showed that 5 kinds of products for the Huai River Basin of land cover types and the features are almost the same, has the strong consistency, the correlation coefficient of area estimation were greater than 0.9; Among the three 30m resolution land cover classification products, CLCD has the highest recognition accuracy, which is used as reference data.The overall accuracy of the other four products is between 92.37% and 95.53%, and the Kappa coefficient is between 0.523 and 0.695. The accuracy of GlobeLand30 and GLC_FCS30 are higher and have their respective advantages, while the accuracy of ESacci-LC and MCD12Q1 are lower. The research results provide references for land cover research at different spatial and temporal scales.}