Determining the distribution of soil organic carbon (SOC) in subsoil (20–100 cm depth) is important with respect to the global C cycle and warming mitigation. However, significant knowledge gaps remain regarding the spatiotemporal dynamics of SOC within this layer. By integrating traditional depth functions with machine learning approaches, we quantified soil β values, which represent the relative rate of decline in SOC density with depth, and provided high-resolution assessments of SOC dynamics across global ecosystems, including cropland, grassland, and forestland. The estimated subsoil SOC densities were 62 Mg ha⁻¹ (95% CI: 52-73) for cropland, 70 Mg ha⁻¹ (95% CI: 57-83) for grassland, and 97 Mg ha⁻¹ (95% CI: 80-117) for forestland. SOC density exhibited a consistent decline with depth, ranging from 30 Mg ha⁻¹ to 5 Mg ha⁻¹ in cropland, 32 Mg ha⁻¹ to 7 Mg ha⁻¹ in grassland, and 40 Mg ha⁻¹ to 13 Mg ha⁻¹ in forestland, across 20 cm depth increments from 20 to 100 cm. The estimated global subsoil SOC stock was 803 Pg C, with cropland, grassland, and forestland contributing 74 Pg C, 181 Pg C, and 547 Pg C, respectively. On average, 57% of this carbon was stored within the top 0-100 cm of the soil profile. This study provides information on the vertical distribution and spatial patterns of SOC density at a 10 km resolution across global ecosystems, providing a scientific basis for future studies pertaining to Earth system models.

 The file named “Rawdata.xlsx” contains data sourced from the literature. The file name is “GE_β.tif”, GE represents global ecosystems, which including cropland (CL), grassland (GL), and forestland (FL). “FL_β.tif” represents the spatial distribution of β for forestland at 20-100 cm depth. The file name is “GE_d_SOCD.tif”, where SOCD represents soil organic carbon density, d represents soil depth, for example, “FL_20-100_SOCD.tif” represents the spatial distribution of SOCD for forestland at 20-100 cm depth.