Projected changes in global soil organic carbon stocks and their vulnerability to drought

Projected changes in global soil organic carbon (SOC) stocks under intensifying drought remain incompletely characterized, limiting assessments of terrestrial carbon vulnerability under climate change. We combined bias-corrected climate drivers with a skill- and independence-weighted ensemble of seven CMIP6 Earth system models. We then used standardized drought indices, multiscale partial-Spearman analyses, and land-cover comparisons to examine historical and projected changes in SOC stocks and heterotrophic respiration (R h ) fluxes, which represent complementary stock and respiratory-flux indicators. Under SSP5-8.5, the multi-model weighted mean projected a small decrease in late-century SOC relative to the 1985–2014 baseline (−0.26%), whereas changes among the seven individual models ranged from −4.68% to +3.32%, with five models projecting increases. In contrast, R h increased in all seven models, and the weighted-ensemble mean for 2070–2099 was 43.54% higher than the historical baseline. Projected SOC decreases in the weighted ensemble were concentrated in parts of the Northern Hemisphere mid- and high latitudes, whereas changes across lower-latitude and subtropical regions were spatially heterogeneous. At the global scale, SSMI showed the largest selected association magnitude with SOC, whereas the selected association magnitudes for Rh were generally similar for SPEI and SSMI. Across the 1–120-month candidate range, the selected accumulation scales varied across grid cells and drought indices. We interpret these values as descriptive summaries of the accumulation windows associated with the largest retained partial-Spearman coefficients. These association results are based on MMEM-derived drought indices paired with MMWM SOC and R h products, and their model dependence remains unquantified. The analysis maps scenario- and region-dependent patterns of drought-associated SOC vulnerability. Differences in CMIP6 soil‑carbon structure and among benchmark products limit quantitative interpretation.