Diagnosing Discrepancies in Soil Carbon Stocks between Upscaled Global Data Products

Soil organic matter stores more carbon than all terrestrial vegetation and the atmosphere combined. However, global estimates of soil organic carbon (SOC) and its spatial variability still vary drastically across upscaled data products, with important implications on downstream predictions and model benchmarking. In this study, we systematically compared several widely used soil carbon data products and ground-truthed them with soil profile observations from the World Soil Information Service (WoSIS). Upscaled data products included the Harmonized World Soil Database (HWSD), the Northern Circumpolar Soil Carbon Database (NCSCD), and SoilGrids, among others. We explored the role of environmental covariates on the residuals between soil profile measurements and the corresponding values from each data product, and found clear biases across biomes and soil depths. While biases among SOC stock estimates in organic soils were substantial, they were generally consistent for a given data product. In contrast, biases among tropical to subtropical mineral soils were greater than expected, but also lacked structure and predictability. Our results depict geographic regions and soil depths for targeted improvements of upscaled SOC stock estimates, and further inform the selection of soil carbon data products in future studies depending on the application and region of interest.