基于VGI的土地覆被遥感产品精度验证

以GEO_WIKI和Degree Confluence Project获取的自发地理信息为参考,通过误差矩阵分析FROM-GLC-agg 500 m升尺度数据和MODIS COLLECTION5两种相同空间分辨率土地覆被遥感产品的类别精度及类别混淆,采用Inverse Distance Weighting模型计算FROM-GLCagg 500 m数据的类别精度空间图谱,结果表明:无论是制图精度还是用户精度,FROM-GLCagg 500 m数据与MODIS数据相比,并没有表现出显著的优势,两种数据在林地与灌木、草地与裸地、耕地与灌木间存在一定程度的类别错分现象;FROM-GLC-agg的类别精度图表明在中国的西北部、东北及华北南部等裸地、林地及耕地聚集区,类别精度较高,该区域约占研究区总面积的36.77%,而在青藏高寒区、中国西南部等地表景观异质性显著区域,类别精度相对较低,约占研究区总面积的20%。

Analysis of land cover category accuracy based on volunteered geographic information

The information on land cover at national scales is critical for addressing climate change, biodiversity conservation, ecosystem assessment and environmental modeling. The Volunteered Geographic Information derived from the GEO_WIKI and Degree-Confluence Project was used as referenced data. The category accuracy and the category confusion about FORM-GLC-agg 500m and MODIS COLLECTION5 were compared and the spatial distribution of FORM-GLC-agg''s category accuracy was analyzed. The results show that:there is no significant difference on producer and user accuracy between FORM-GLC-agg and MODIS data. They both have a serious confusion between Forest/shrub, cropland/shrub, especially the grassland and bare land. The category accuracy has an uneven distribution, for example, its higher value mainly distributes in northwest and north China, and the lower value more mainly distributes in Tibet Plateau, and southwest China. The results can provide reference for land cover research.