基于空间抽样的上海市GlobeLand30 2020数据精度评估

Accuracy Assessment of GlobeLand30 2020 Based on Spatial Sampling:A Case Study in Shanghai

针对上海GlobeLand30 2020土地覆盖产品,应用顾及空间异质性的类别-异质性分层抽样方法进行抽样,将研究区各地图类别划分为匀质区和异质区子类,按照内曼分配方法布设验证样本,在研究区域内抽取2500个验证像素,并参考同时期多源Google Earth高分遥感影像数据进行样本判读,获得参考分类即真实类别信息,对区域内总体及各类别进行精度评估。实验结果表明GlobeLand30土地覆盖(2020)的总体精度为77.89%,基本反映了上海地区2020年的土地覆盖状况。其中匀质区精度为78.92%,异质区精度为50.65%。异质区的精度较低,其分类的改善和数据应用值得生产者和用户关注。

GlobeLand30 is the first set of global land cover product with 30m fine spatial resolution of China,and the 2020 base year data prod-uct has been updated.Studying the data characteristics and evaluating its accuracy are of guiding significance for data users.For the Globe-Land30 data of Shanghai in the 2020 base year,the Class-Heterogeneity-Stratified Sampling considering spatial heterogeneity was carried out and the study area were divided into homogeneous and heterogeneous area.2500 verification samples were set up and sampled within the region,and the Google Earth high-resolution remote sensing image of the same period were collected for sample interpretation to conduct accuracy as-sessment of the overall area and each category.The experiment results show that the overall accuracy of the GlobeLand30(2020)data in Shang-hai is 77.89%,indicating that the quality of the GlobeLand30 2020 Shanghai is fine and accurately reflect the land cover distribution in Shang-hai.The accuracy of the homogeneous area is 78.92%,and the accuracy of the heterogeneous area is 50.65%,which is lower than the homoge-neous area.The fact that heterogeneous strata tend to be classified with lower accuracy suggests that GlobeLand30 producers should strive to im-prove classification over such strata and users should be cautious when using the product.