Accuracy and applicability assessment of vegetation types of the land cover products produced in China

• Evaluating the long-term accuracy and spatial applicability of CLCD, GlobeLand30, and GLC_FCS30D products. • Proposing a spatial accuracy method based on the multiscale geographically weighted regression model. • CLCD optimal for forest/grassland monitoring; GlobeLand30 limited temporally; GLC_FCS30D stable for shrublands. • Providing a basis for Chinese users to select or integrate land cover products according to application needs. Highly accurate land use/land cover (LULC) information is crucial for monitoring current environmental changes. The existing LULC products generated based on Landsat, MODIS, Sentinel, and other remote sensing images have obvious differences in classification accuracy, spatial and temporal resolution, the applicability, and the stability of the time series products in different regions. The accuracy and applicability assessment of LULC products in China is predominantly concerned with the comparative analysis of area and spatial distribution characteristics of single temporal data, while the assessment of vegetation types in the spatial and temporal dimensions is seldom involved, resulting in the applications of these products in forest resource monitoring and ecological value evaluation at national and regional scales remaining unclear. Therefore, we analyzed the applicability of three land cover products produced by Chinese institutions, including 35 issues of CLCD, 3 issues of GlobeLand30 and 10 issues of GLC_FCS30D, in terms of temporal stability and spatial consistency, and developed a spatial accuracy assessment method to realize the spatial accuracy mapping of different land cover products using the multiscale geographically weighted regression model (MGWR). The results show that with respect to temporal stability, the CLCD product has high stability in forests and grasslands and is suitable for monitoring long-term changes; the GLC_FCS30D product has obvious advantages in the stability of shrubland types and local feature recognition; the vegetation type stability of the GlobeLand30 product reaches more than 0.85, but is limited by the short time series data. With respect to spatial consistency, the GLC_FCS30D product shows better overall spatial consistency than the CLCD and GlobeLand30 products, the fineness of vegetation type delineation is correlated with the level of spatial consistency, and there are obvious geographic differences in spatial expression between different products. The three products have different levels of ambiguity in distinguishing the boundaries of forests, shrublands and grasslands. The CLCD product has the highest overall spatial accuracy, with local spatial accuracy variations in the north, southeast, and southwest in China, and users who are concerned about these regions should be aware of the differences between the products. This study provides a reasonable basis for Chinese users to make appropriate selections of land cover products or to integrate different products according to application scenarios.