Climate change and the need for agricultural production are driving the increasing demand for accurate spatial information about soil organic carbon (SOC). While remote sensing (RS) data is effective for SOC modeling in croplands, the integration of additional covariates is often neglected. In the study, we evaluated the performance of a Sentinel-2 mosaic-based approach for digital mapping of SOC content in a semi-arid region and evaluated the inclusion of additional soil-forming variables. Several scenarios were examined where crop intensity, climate, and terrain covariates were incorporated into the bare soil mosaic using a recursive feature elimination (RFE) analysis and machine learning approach. The models were evaluated with repeated cross-validation and the prediction interval coverage probability (PICP) to assess the SOC predictions and associated uncertainties. The contribution of environmental covariates was assessed by permutation feature importance and Shapley values methods. The results showed that the model with only temporal mosaic of bare soil as explanatory variables resulted in RMSE = 0.75 %, R2 = 0.15 and RPIQ = 1.11, whereas the scenario with all covariates significantly improved the model performance (RMSE = 0.56 %, R2 = 0.52 and RPIQ = 1.55) and reduced the associated uncertainties. Among the spatial covariates, climate (precipitation, land surface temperature) and elevation emerged as key factors influencing the prediction of SOC content. Interpretation of such complex models through Shapley values revealed that the decrease in temperature, solar radiation and precipitation seasonality and the increase in elevation had a strong positive contribution to the SOC predictions. In steppe drier zones, these variables had a negative contribution to the predictions. While temporal mosaics of bare soil are useful predictors, we highlighted the importance of incorporating a more diverse range of environmental covariates for SOC modeling across cultivated lands. Our results suggest that integrating climate and elevation data with RS information is essential for achieving robust and accurate SOC mapping, especially across heterogeneous semi-arid landscapes.