Enhanced soil organic carbon mapping in Gannan’s alpine meadows: A comparative analysis of machine learning models and satellite data

• Developed high-accuracy spatial maps of soil organic carbon density in Gannan across multiple scenarios. • Gradient Boosted Regression Trees with Sentinel-1 and DEM data showed optimal SOC prediction. • SHAP analysis quantified variable contributions to machine learning predictions. • Elevation was the best predictor for our soil organic carbon denisty estimation models. Soil organic carbon (SOC) is a critical climate change indicator. This study targets the Gannan region, a key livestock area in China, where accurate soil organic carbon density (SOCD) mapping using remote sensing and machine learning is crucial. We explored the effectiveness of different satellite sensors, including optical (Sentinel-2) and radar (Sentinel-1), in SOCD prediction models, assessing uncertainties. The performance of Extreme Gradient Boosting (XGBoost), Random Forest (RF), and Gradient Descent Boosted Regression Tree (GBDT) models was evaluated using metrics such as the determination coefficient (R 2 ), root mean square error (RMSE), mean absolute error (MAE), and Lin’s concordance correlation coefficient (LCCC). We also generated SOCD maps for the Gannan region and conducted an interpretability analysis using Shapley additive explanation (SHAP) from game theory. Findings indicate: (1) The GBDT model surpassed RF and XGBoost in predictive accuracy; (2) Integrating data from different satellite sensors improved the soil prediction models; (3) The GBDT model, incorporating data from Sentinel-1, Sentinel-2, and Digital Elevation Model (DEM), achieved the highest accuracy from the test sets (R 2 = 0.5702 RMSE = 4.1557 kg C m −2 , MAE = 3.1110 kg C m −2 , LCCC = 0.7689), with significant enhancements from Sentinel-1; (4) DEM data were crucial in predicting SOCD, followed by Sentinel-2 and Sentinel-1. Overall, this study highlights the utility of Sentinel data for long-term and rapid monitoring of SOCD in Gannan, advancing our understanding of soil carbon mapping capabilities.