Depth-solved modeling of soil organic carbon stocks and uncertainty in a tropical agricultural landscape

Soil organic carbon (SOC) stocks in tropical agricultural systems exhibit strong variability across space and depth, reflecting interacting biogeochemical and landscape controls. The objective of this study is to model and quantify SOC stocks and their associated uncertainty implementing a depth-solved digital soil mapping methodology in a tropical agricultural landscape of the Colombian dominated by highly weathered Oxisols and Ultisols. A total of 636 soil samples from 159 locations were collected at four depth intervals.

SOC stocks were modeled using environmental covariates representing key drivers of soil organic matter dynamics, including vegetation activity, soil moisture, topographic controls on drainage, and depth to solve SOC stocks across predefined soil layers. Recursive Feature Elimination was applied to identify the most relevant predictors, and Quantile Regression Forests were used to generate median predictions together with prediction intervals as a measure of uncertainty.

Our results indicate that while depth is the main factor influencing SOC stocks, horizontal variability within depth layers was associated with moisture-sensitive spectral indices and terrain attributes, highlighting the role of hydrological regulation in SOM accumulation. Predicted SOC stocks showed values between 8 and 33 Mg.ha-1 with a mean of 21.7 Mg.ha-1. The model showed moderate performance (RMSE ≈ 6.02 Mg/ha, R² ≈ 0.44 and MAE ≈ 4.21 Mg/ha) and generally low to moderate uncertainty.

By modeling depth and spatial uncertainty, this framework aligns with key principles of MRV protocols for soil carbon and can inform baseline definition and monitoring strategies for soil carbon accounting in tropical agricultural systems globally applicable.