Accuracy Assessment of Multi-source Land Use Products in the Loess Hilly and Gully Region

Abstract

[Objective] The study was aimed to assess the accuracy of multi-source land use products in the loess hilly and gully regions. [Method] Using the Guanchuan River Basin, a secondary tributary of the Yellow River, as a case study, this research evaluated the accuracy of six land use products with spatial resolutions of 10 and 30 m (WorldCover 10 m, ESRI 10 m, GLC_FCS30-2020, GlobleLand30, CNLUCC, and CLCD). The evaluation was based on high-precision GCLUCC land use data and focused on four aspects: spatial distribution, area characteristics, sample point accuracy, and spatial consistency. The GCLUCC data, with a spatial resolution of 1 m and an overall accuracy exceeding 95%, was derived from GF-2 (0.8 m), DEM (5 m), 6 400 field sampling points, and 2 320 Google Earth sample points, using an object-oriented approach and manual visual interpretation. [Results] (1) In terms of classification features, most products could extract major land categories, but significant differences in extraction efficiency and accuracy were evident, especially in the spatial distribution of arable land, forest land, and construction land. (2) Regarding area consistency, significant variations were found between each product and GCLUCC in terms of area of different land categories. For instance, the grassland area in some products was more than double that of GCLUCC, while forest and water areas were only 0.13%~12.11% and 1.41%~11.27% of GCLUCC, respectively. (3) In terms of overall accuracy, GlobleLand30 and WorldCover 10 m showed relatively higher accuracy, with spatial map accuracy reaching 58.21% and 50.19%, respectively. In contrast, CLCD and CNLUCC had relatively lower accuracy, with significant classification confusion between forest and grassland, arable and grassland, and construction and arable land. (4) Significant spatial discrepancies existed between all products and the actual ground cover, particularly in the classification of trees, shrubs, bare land, and water bodies. [Conclusion] Current land use products face challenges in accurately characterizing surface cover features in the loess hilly and gully regions. Future product development should consider topographical and geographical features more thoroughly and enhance the identification of specific land use types. These results can provide a theoretical reference for ecological conservation and land resource information acquisition in the Loess Plateau.

Keywords

• loess hilly and gully regions
• land use/land cover products
• accuracy assessment
• object-oriented classification