<p>This study found effective evidence that combining modern remotely sensed technology with algorithms of machine learning (ML) approaches can effectively predict soil organic carbon (SOC). We employed the Google Earth Engine (GEE) to prepare two comparable datasets with full-band and shared-band variables from Sentinel-2A (S2A) and MODIS images. The predictive power of four machine learning algorithms included random forest (RF), decision tree (DT), boosted decision tree (BDT) and gradient boosted regression tree (GBRT), was evaluated in a systematic manner to assess their capability for predicting the levels of soil organic carbon (SOC) in the Karvir region. All the models were trained and validated on a dataset of environmental and soil-related variables to check for the accuracy and reliability of their SOC prediction. Our findings revealed that models based on MODIS` full-band dataset generated the greatest results. There was less error in full-band predictions with GBRT and BDT, models. With this dataset, a lower “root mean square error” (RMSE equals 3.04 (g/kg) and a higher coefficient of determination, (R² equals 0.71) were recorded. In S2A-based, RF had the optimal outcome with R² valued at 0.63 and RMSE equals 2.93 (g/kg). Furthermore, models based on MODIS data demonstrated a significant correlation with actual SOC data from the full-band dataset. This suggests that MODIS data, with its superior cloud-free masking and spectral resolution, exceeded S2A data in mapping SOC in the Karvir area.</p>