Comparison of different approaches for estimating profile soil organic carbon stocks from topsoil samples

Abstract

Changes in land use, management, and climate can significantly influence soil organic carbon (SOC) dynamics, prompting domestic and international initiatives to prioritise monitoring, reporting, and verifying SOC stock changes. Based on the greater availability of surface soil carbon samples, this study aims to assess the feasibility of estimating SOC stocks at deeper depths using topsoil samples. Both parametric and non-parametric techniques (Bayesian regression), quantile regression forests, and generalised additive models were tested to predict SOC stocks from topsoil samples (defined for either fixed or variable depths) and for different profile depths (0–30 cm and 0–60 cm). The best predictions were obtained using generalised additive models that accommodate different topsoil depths. This approach eliminates the need to harmonise topsoil depths prior to modelling. Predictions for 0–30 cm (RMSE = 14.29 Mg·ha−1, MEC = 0.72, CCC = 0.83) were more accurate than those for deeper soil depths (0–60 cm; RMSE = 24.87 Mg·ha−1, MEC = 0.60, CCC = 0.74). While those uncertainties are still substantial, these results demonstrated the potential of using surface soil samples to complement existing SOC datasets.

Keywords

• Soil organic carbon stocks
• Soil carbon inventory
• Soil carbon profile
• Regression
• Empirical modelling