Abstract: The spatial distribution of calcium oxide (CaO) content in surface sediments of the eastern China marginal seas is controlled by multiple factors, including terrigenous input, biogenic carbonate contribution, and sedimentary dynamic condition. Its variability can indicate regional carbonate production-dissolution processes and sedimentary environmental evolution, which is important to the understanding of material cycling on continental shelves. However, due to the difficulty of marine survey and limited sampling density, traditional spatial interpolation methods often exhibit high uncertainty when predicting CaO distributions in sample-sparse areas or environmentally heterogeneous zones. Based on 414 measured CaO data points from surface sediments in the eastern China seas, we integrated multi-sourced environmental variables, including oceanographic, hydrological, biological, and geological factors, applied the random forest algorithm to predict the spatial distribution of CaO content, and compared the results with those obtained from the Kriging interpolation. Results show that the random forest model exhibited good predictive performance under both tenfold cross-validation and spatial block cross-validation, and the predicted values were generally consistent with the observations (R² = 0.63 for tenfold cross-validation; R² = 0.45 for spatial block validation). Under spatial extrapolation conditions, the random forest model outperformed the spatial autocorrelation-based Kriging method (R² = 0.19 for spatial block validation), reflecting its reasonable predictive capability in complex sedimentary environments. The predicted distribution indicated that CaO content in surface sediments of the eastern China seas showed generally lower values nearshore and higher values on the continental shelf and its outer margin. Compared with the Kriging approach, the random forest model captured more clearly local high-value zones and spatial heterogeneity. In addition, the spatial distribution of standard deviations predicted by the random forest model further revealed differences in prediction uncertainty across different sedimentary environments, providing guidance for optimizing future sampling strategies. Overall, the results demonstrate that the random forest algorithm has good applicability for predicting CaO content in surface sediments in complex shelf sedimentary systems, and can provide a reliable technical support to the studies of carbonate material cycling and spatial prediction of sediment geochemical elements in the eastern China marginal seas.