Accuracy evaluation and consistency analysis of multi-source remote sensing land cover data in the Yellow River Basin

With the development of multi-source remote-sensing platforms and technologies, various land cover datasets have been developed that provide a wealth of data to support the understanding of global land cover conditions, land surface process model simulations, and socioeconomic development decisions. However, selecting appropriate data for different regions from nationally or globally available land cover datasets is challenging. In this study, six land cover products in 2020 over the Yellow River Basin, including CLCD_v01_2020, GLOBELAND30, GLC_FCS30_2020, LANDCOVER (300 m), MCD12Q1 (500 m), and CNLUCC1000 (1000 m), with resolutions ranging from 30 to 1000 m, were evaluated for regional-scale accuracy and consistency analysis. Accuracy analyses were performed on six products based on 1540 samples for seven land cover types collected by Google Earth. Data with the highest overall accuracy (OA) were used as a reference for the area consistency analysis of the other five products. Category confusion and confusion mapping analyses were performed on six types of data. Hopefully, this study will provide a scientific reference for users to select appropriate land cover data for the Yellow River Basin. The results showed that the highest classification accuracy was for CLCD_v01_2020, with an OA of 88.12%, followed by GLOBELAND30 (OA=85.32%), GLC_FCS30_2020 (OA=84.09%), LANDCOVER300 (OA=77.79%), MCD12Q1 (OA=73.38%), and CNLUCC1000 (OA=71.82%). The KAPPA coefficients of the land cover products with a resolution of 30 m were all above 0.8, and the classification accuracy decreased as the spatial resolution decreased. CLCD_v01_2020, with the highest OA, was used as the reference dataset, and the area correlations and confusion mapping were calculated separately for the remaining five validation product datasets. The relative proportions of different land cover types were generally consistent across the six products; however, there were still large differences between croplands and grasslands. GLC_FCS30_2020 had the highest correlation with the reference data CLCD_v01_2020, with an R² value of 0.9976. Category confusion analysis showed that the six data types were generally confused between croplands, forests, and grasslands. There was good consistency in the grasslands of eastern Qinghai in the upper reaches of the Yellow River and the cropland and construction land of the middle and lower reaches. The areas of poor consistency were mainly in the middle reaches of the Yellow River in northern Shaanxi and western Shanxi, which were mainly confused grasslands with forests. For the primary classification of land cover data in the Yellow River Basin, it is recommended that CLCD_v01_2020 data be selected for 30 m resolution and LANDCOVER300 for 100-m scale resolution data. In contrast, secondary classification can be chosen according to the desired classification system.