Abstract: A spatial estimation of soil organic carbon density in Xing'an League of the Inner Mongolia Autonomous Region was conducted for the year 2023 using remote sensing and geographic information technologies. Multiple environmental variables were extracted based on remote sensing data and environmental data (e.g., terrain, climate, soil), with soil property data obtained from field soil sampling serving as the response variable. Regression modeling and accuracy comparison were performed using Random Forest (RF), Extreme Gradient Boosting (XGBoost), and Multilayer Perceptron (MLP). Based on the accuracy evaluation results, the best-performing model was selected to ultimately complete the spatial mapping of soil organic carbon density in the study area. The results showed that the RF model performed the best (R² = 0.86, RMSE = 1.51 kg/m²), the XGBoost model performed slightly worse than the RF model but still well, while the MLP model performed significantly worse in this task, with its accuracy being much lower than the other two models. The spatial distribution of soil organic carbon density in Xing'an League showed obvious north-south differences, generally decreasing from northwest to southeast. In terms of vertical distribution, the organic carbon density first decreased and then increased with increasing sampling depth.

Key words: soil, organic carbon density, Sentinel optical, remote sensing, inversion, Xing'an League