Better estimates of soil carbon from geographical data: a revised global approach

Soils hold the largest pool of organic carbon (C) on Earth; yet, soil organic carbon (SOC) reservoirs are not well represented in climate change mitigation strategies because our database for ecosystems where human impacts are minimal is still fragmentary. Here, we provide a tool for generating a global baseline of SOC stocks. We used partial least square (PLS) regression and available geographic datasets that describe SOC,climate, organisms, relief, parent material and time. The accuracy of the model was determined by the root mean square deviation (RMSD) of predicted SOC against 100 independent measurements. The best predictors were related toprimary productivity, climate, topography, biome classification, and soil type. The largest C stocks for the top 1 mwere found in boreal forests (254 +/- 14.3 t ha(-1)) and tundra(310 +/- 15.3 t ha(-1)). Deserts had the lowest C stocks (53.2 +/- 6.3tha(-1))and statistically similar C stocks were found for temperate and Mediterranean forests (142 - 221 t ha-1), tropical and subtropical forests (94 - 143 t ha(-1)) and grasslands (99-104 t ha(-1)). Solar radiation, evapotranspiration, and annual mean temperature were negatively correlated with SOC, whereas soil water content was positively correlated with SOC. Our model explained 49% of SOC variability, withRMSD (0.68) representing approximately 14% of observed C stock variance, overestimating extremely low and underestimating extremely high stocks, respectively. Our baseline PLS predictions of SOC stocks can be used for estimating the maximum amount of C that may be sequestered in soilsacross biomes.