Abstract This study assessed the influence of land use on soil physical and chemical properties and evaluated machine learning (ML) models for predicting soil organic carbon (SOC) content and stocks in the top 0–30 cm layer across selected sites in the Brazilian Amazon. A total of 649 georeferenced samples of surface soil were analyzed for soil texture, pH, total nitrogen (TN), organic carbon, cation exchange capacity (CEC), soil texture, and bulk density (BD). ML models, including random forest (RF), support vector machine, multiple linear regression, and artificial neural network, were used to improve the predictability and accuracy of the model's outcome. Validation was performed using the 10‐fold cross‐validation technique with R 2 and root mean squared error (RMSE) metrics. Land use significantly affected soil properties at p < 0.01. Pearson correlation analysis revealed strong positive relationships between SOC and clay, silt, and TN, and negative correlations with pH. Variable importance analysis identified clay, CEC, and BD as the most influential predictors. Of the models, the RF resulted in R 2 of 0.9997 and an RMSE of 0.00231. The predicted values of SOC stock over the research area varied between 137.16 and 186.88 Mg C ha − 1 in forest land, 78.6 and 81.2 Mg C ha − 1 in pasture land, and 17.95 and 46.11 Mg C ha − 1 in cultivated land at a limited upper 0‐ to 30‐cm soil depth. These findings highlight the reliability of ML for SOC prediction in developing climate‐resilient land management strategies and targeted conservation planning across the basin.
Tiruneh et al. (Thu,) studied this question.