Spatial Prediction of Soil Organic Carbon Stock in the Moroccan High Atlas Using Machine Learning

Soil organic carbon (SOC) is an essential component, which soil quality depends on. Thus, understanding the spatial distribution and controlling factors of SOC paramount to achieving sustainable management. In this study, prediction for Ourika watershed in Morocco was done using four machine learning (ML) algorithms: Cubist, random forest (RF), support vector (SVM), gradient boosting (GBM). A total 420 samples were collected at three different depths (0–10 cm, 10–20 20–30 cm) from concentration bulk density (BD) measured, consequently stock (SOCS) determined. Modeling data included 88 variables incorporating environmental covariates, including properties, climate, topography, remote sensing used as predictors. The results showed that RF (R2 = 0.79, RMSE 1.2%) Cubist 0.77, most accurate models predicting SOC, while none satisfactory BD across watershed. As with 0.86, 11.62 t/ha) 13.26 exhibited highest predictive power SOCS. Land use/land cover (LU/LC) critical factor SOCS, followed by properties bioclimatic variables. Both combinations bioclimatic–topographic properties–remote shown improve performance. Our findings show ML algorithms can be a viable tool modeling mountainous Mediterranean regions, such study area.