Mangrove ecosystems, often referred to as “blue carbon”, play a significant role in storing vast amounts of carbon dioxide, thereby mitigating the effects of climate change. However, human-induced activities are endangering their carbon sequestration potential, particularly along the Pacific coast of Colombia. This study aims to quantify carbon stocks in mangrove soils along the Colombian Pacific coast. This quantification will serve as a foundation for monitoring potential future changes in these stocks due to regional transformations. Data from multispectral sensors, including Landsat 8 and Sentinel 2A, from 2014 to 2021 were integrated using a machine learning (ML) methodology. The efficacy of the model was assessed using the coefficient of determination (R²) and root mean square error (RMSE). The extreme gradient-boosted regression model (XGBoost) applied to the Landsat 8 dataset yielded optimal values of R² = 0.825 and RMSE = 1.748 Mg C ha-1 for soil organic carbon (SOC). According to the model, the estimated SOC content varied from 0.524 Mg C ha-1 at a depth of 0–15 cm to 263.2 Mg C ha-1 at depths ranging from 50–100 cm within the mangrove forests. The results underscore the importance of machine learning and remote sensing as effective tools for establishing a rapid and reliable reference base. This will enable the prioritization of conservation efforts related to soil resources in mangrove ecosystems.