Accurately characterizing the spatial distribution of soil organic carbon (SOC) in cropland topsoil is essential for cropland quality assessment and soil resource management in Northeast China. This study used 640 cropland topsoil samples collected at a depth of 0–20 cm during July and August 2022, together with 25 environmental covariates, to develop random forest (RF) and extreme gradient boosting (XGBoost) models and compare their predictive performance. The samples were divided into a training set (n = 512) and a test set (n = 128). Model hyperparameters were optimized through 10-fold cross-validation within the training set, and predictive performance was evaluated using the test set. The optimal cross-validated RMSE values for RF and XGBoost were 7.34 and 7.43 g kg?1, respectively. On the test set, RF achieved an R2 of 0.64 and an RMSE of 6.70 g kg?1, whereas XGBoost achieved an R2 of 0.63 and an RMSE of 6.83 g kg?1. RF therefore exhibited slightly better predictive performance and was selected for spatial mapping. Feature-importance analysis showed that annual mean temperature (BIO1) ranked as the most important predictor in both models, while MODIS nighttime land surface temperature and other temperature-related covariates also ranked highly. Finally, the RF model was used to generate a spatial prediction map of cropland SOC across Northeast China, providing a spatially explicit basis for regional cropland quality assessment and soil resource management.