A dataset of surface soil organic matter in grasslands of northern China

Soil organic matter (SOM) plays a pivotal role in global ecosystem properties and functions, especially in the grassland soils of northern China, where substantial carbon stocks and nutrient reserves are stored. However, current datasets on soil organic matter content (SOMC) lack temporal information, which is crucial for understanding soil dynamics in response to climate change and human activity. The spatial resolution and simulation accuracy of some existing datasets are insufficient to meet the requirements for the accurate evaluation of soil carbon stocks and dynamics. To address these limitations, this study used the relationships between soil properties and environmental factors to develop a CatBoost model. This was performed by combining multi-source soil sample data and environmental covariate data representing climate, terrain, vegetation, and soil factors associated with grassland soil development. By incorporating the local spatial autocorrelation information of SOMC, we developed a new model, CatBoost-Gi*, whose performance was superior to that of other gradient boosting decision tree methods. This new model was used to generate spatial maps of SOMC in the topsoil (A-horizon) of grasslands in northern China for 1981–2000 and 2001–2020 at spatial resolution of 90 m × 90 m, with more accurate and reliable estimations than the existing main datasets.