Accurate estimation of Soil Organic Carbon (SOC) is essential for sustainable land management, agricultural productivity, and climate change mitigation. This study presents a novel SOC estimation framework integrating machine learning with multispectral bands, vegetation and soil indices, topographical attributes, soil texture variables, and Hue, Saturation, and Value (HSV)-derived soil color proxies. SOC data from 180 samples collected between 2007 and 2020 across 21 agricultural fields in Manitoba, Canada, were used for model training and validation. Landsat 5, 7, and 8 data were used to derive spectral and soil indices, while SoilGrids and Shuttle Radar Topographic Mission Digital Elevation Model (SRTM-DEM) provided soil texture and topographical features. Random Forest (RF), Extreme Gradient Boosting (XGB), and a Bias Corrected–VarianceWeighted (BC-VW) ensemble model were evaluated across five feature scenarios. The ensemble model achieved the best performance, with a coefficient of determination (R²) of 0.57, Root Mean Square Error (RMSE) of 0.25, and Root Mean Square Percentage Error (RMSPE) of 7.87%, outperforming individual models. SHapley Additive exPlanations (SHAP)-based feature selection identified Clay %, Short Wave Infrared 1 (SWIR1), and Value as the most influential predictors. Independent validation using 2021 and 2023 data confirmed model robustness, with RMSPE values of 10.93% and 12.83%, respectively. The results demonstrate the importance of integrating soil-specific indices, texture, and color features with ensemble modeling for scalable and reliable large-scale SOC monitoring and carbon sequestration applications.