ChinaSoybeanYield1km: a 1-km annual soybean yield dataset from 2001 to 2020 in China

Dataset Name: ChinaSoybeanYield1kmTemporal Coverage: Annual, 2001 to 2020Geographic Region: Soybean Production Areas in ChinaSpatial Resolution: 1kmData Format: TIFData unit: kg/haReference Coordinate System: Projected Coordinate System (WGS_1984_UTM_Zone_49N)Dataset Composition: The dataset consists of 20 files of annual soybean yield raster data from 2001 to 2020 in China, with "Soybean_year.tif" representing the yearly soybean yield raster data.Data Generation Process: First, we collected climate, remote sensing, management, soil and municipal-level statistical soybean yield data as modelling data. Second, we divided China into three main soybean production regions. For each region, 20 individual machine learning-based yield estimation models were developed according to the modeling process of a data-driven model. In this step, well-performed models were further selected as base models and meta-models. Third, the number of base variables and the types of input variables were determined to develop the final ensemble learning model for each region .Finally, this study utilized ensemble learning models to generate annual simulated soybean yield maps. These maps were employed as weights for spatial disaggregation of municipal-level statistical data, resulting in the generation of a yearly rasterized soybean yield dataset for China from 2001 to 2020 with a spatial resolution of 1km. The results indicate that at the county level, ChinaSoybeanYield1km has an R² of approximately 0.50 and an RMSE of 0.61t/ha, while EarthStat, MapSPAM, and GDHY have RMSE values of 0.75 t/ha, 0.80 t/ha, and 1.19t/ha, respectively. In comparison, the accuracy of the soybean yield dataset generated in this study at 1km resolution is higher.