Emergent temperature sensitivity of soil organic carbon driven by mineral associations

Soil carbon–climate feedbacks are a major uncertainty in predicting terrestrial biosphere responses to warming, in part due to poorly constrained temperature sensitivities of soil organic carbon (SOC) decomposition and stabilization processes. Bulk soil C responses have been widely studied, but the heterogeneity of soil organic matter—particularly distinctions between mineral-associated (protected) and particulate (unprotected) organic matter—is often ignored, despite evidence that these fractions respond differently to warming. Mineral associations can limit microbial access, leading to older, slower-cycling mineral-associated C compared with particulate C. Given limited multi-decadal observations of SOC dynamics and the dominance of bulk metrics in model benchmarking, there is a need to quantify, at global scale, the temperature sensitivities of these underlying pools and how their distributions drive the emergent climatological temperature sensitivity of bulk SOC. This study leverages a global observationally derived dataset of mineral-associated and particulate SOC to quantify pool proportions and their climatological temperature sensitivities, and benchmarks these against an ensemble of Earth system and land models to improve understanding and prediction of carbon cycle–climate feedbacks.

Prior empirical and modeling work shows that warming accelerates decomposition and often reduces bulk SOC, but results vary with climate and soil properties. Studies indicate fine-textured soils with greater clay and silt contents tend to have lower climatological temperature sensitivity due to higher mineral stabilization capacity. Laboratory and field experiments suggest mineral-associated (protected) C generally has lower temperature sensitivity and older ages than particulate (unprotected) C, consistent with mineral protection limiting microbial access. However, most experiments focus on bulk SOC and rarely report pool-specific responses. Space-for-time approaches using spatial temperature gradients can benchmark long-term climatological sensitivities but cannot fully constrain transient dynamics due to acclimation and other factors. In models, protected C is often represented explicitly in microbially explicit models or implicitly as slow/passive pools in first-order models (e.g., Century derivatives), but the interpretability of model pools relative to measurable fractions remains imperfect. Radiocarbon syntheses show many ESMs underestimate mean SOC age, implying too much fast-cycling C and motivating pool-specific benchmarks.

Data: Analyses used observationally derived global datasets gridded at 0.5° resolution. Mean annual temperature (MAT) was from CRU TS3.10 (30-year means), precipitation from GPCC, land cover from MODIS MCD12C1, and productivity (NPP) from MODIS. Total SOC to 1 m was estimated as the mean of Harmonized World Soil Database and SoilGrids products. Mineral-associated SOC and its uncertainty (90% prediction intervals) were taken from Georgiou et al. 2022, derived via a random forest trained on global fractionation data with cross-validation. Analyses focused on non-permafrost mineral soils with MAT > 0 °C; gridcells with >50% peat (Histosols/Histels), hyperarid (aridity index <0.05), or precipitation <100 mm yr−1 were excluded to avoid water-saturated or arid limitations. Models: Nine CMIP6 ESMs that reported soil C pool distributions (ACCESS-ESM1-5, BCC-CSM2-MR, CESM2, CNRM-ESM2-1, E3SM-1-1-ECA, MIROC-ES2L, MRI-ESM2-0, IPSL-CM6A-LR, NorESM2) and three offline models from a biogeochemical testbed (CASA-CNP, MIMICS, CORPSE) were analyzed. Historical-period outputs were used: SOC stocks averaged over 2005–2015; climate covariates over 1985–2015. Total SOC corresponded to cSoil; protected C was mapped to cSoilSlow (passive) in first-order models and to explicit physicochemically protected pools in MIMICS and CORPSE. TaiESM1 was excluded due to incompatible pool definitions. While not depth-resolved, cSoil was compared to 0–1 m SOC as best available correspondence. Pool interpretability and parameters: In Century-like models, the passive pool formation depends on microbial transfers modulated by clay+silt, with turnover times of protected C typically 200–450 years (1,000 years in MIROC-ES2L and MRI-ESM2-0), and particulate (slow) pools 5–10 years (25 years in MIROC-ES2L), modified by environmental scalars. Microbially explicit testbed models used shorter protected C turnover (e.g., CORPSE fixed 75 y; MIMICS 15–80 y), noted as biased low relative to radiocarbon constraints. Analytical approach: SOC stocks in each pool were log-transformed, and multiple linear regressions related log(SOC) to MAT while controlling for NPP, precipitation, and clay+silt content to reduce confounding. Climatological temperature sensitivity was summarized as the proportional decline in C stocks for each 10 °C increase in MAT (a stock-based analogue to Q10), with 95% confidence intervals derived from regression slopes. Analyses were conducted globally for MAT > 0 °C and separately for cool (0–15 °C) and warm (≥15 °C) regimes. Model outputs were compared with the observationally derived data product for pool proportions and climatological sensitivities. Spatial summaries included global means, latitudinal means, and multi-model means; future scenario analyses (ssp585/RCP8.5) were conducted but are summarized in supplementary materials.

- The proportion of SOC that is mineral-associated (protected) increases with MAT and is higher in fine-textured soils across all temperatures. - Globally, unprotected (particulate-like) C stocks exhibit a 28% (95% CI: 26, 30%) greater climatological temperature sensitivity than protected (mineral-associated-like) C stocks. In cool regions (<15 °C), unprotected C is 53% (46, 60%) more temperature sensitive; in warm regions (≥15 °C), the relative difference is 15% (11, 20%). Both pools show weak sensitivities in warm regions. - Fine-textured soils have lower climatological temperature sensitivity than coarse-textured soils, consistent with higher protected C proportions in finer soils. - The emergent climatological temperature sensitivity of bulk SOC is driven by both the distribution of C between protected and unprotected pools and the pools’ distinct temperature sensitivities; in the data product, protected C dominates bulk SOC stocks and thus bulk sensitivity. - Global models vary widely in pool distributions: the global proportion of protected C ranges from 16% to 85% across CMIP6 ESMs and offline models. About half of the ESMs underestimate the protected fraction, implying too much fast-cycling C and too-young bulk SOC ages. - While several models reproduce bulk SOC climatological sensitivity, many do so for the wrong reasons; pool-specific sensitivities and proportions often diverge from observations. Some CMIP6 models show similar sensitivities across pools due to shared temperature parameters; a few even simulate unprotected C as less temperature sensitive than protected C in certain regimes. - Across climates, both pools are more temperature sensitive in cool than warm regions, with a stronger contrast for unprotected C. Models tend to overestimate the sensitivity of protected C (especially MIROC-ES2L, CASA-CNP, MRI-ESM2-0) and underestimate that of unprotected C in cool climates. - These biases imply ESMs may overestimate productivity-driven SOC gains (due to excess unprotected C) and underestimate warming-driven losses from unprotected pools in cool climates, affecting projections of carbon cycle–climate feedbacks.

The study demonstrates that the emergent climatological temperature sensitivity of bulk SOC is fundamentally governed by the distribution of SOC between mineral-associated (protected) and particulate (unprotected) pools and their distinct temperature dependencies. Observationally derived global patterns show increasing protected C with warmer MAT and finer textures, leading to lower bulk sensitivity where protected C dominates. Unprotected C is substantially more sensitive to temperature, particularly in cool climates, highlighting that ecosystems in cooler regions are more temperature-limited and vulnerable to SOC losses with warming. Benchmarking global models against these pool-specific metrics reveals large discrepancies in both pool proportions and climatological sensitivities, even in models that match bulk behavior. Underestimation of protected C and overestimation of sensitivity of protected pools, together with underestimation of unprotected C sensitivity in cool climates, suggest potential misrepresentation of SOC age distributions and ecosystem responsiveness. These issues can cascade into biased projections, such as overestimating SOC sequestration from increased productivity and underestimating warming-induced SOC losses in cool regions. Incorporating pool-specific benchmarks, alongside radiocarbon constraints, can guide model refinements, improve realism of SOC dynamics, and reduce uncertainties in carbon cycle–climate feedback projections.

This work provides a global, observationally derived benchmark for the distributions and climatological temperature sensitivities of mineral-associated (protected) and particulate (unprotected) soil carbon pools and shows that unprotected C is, on average, 28% more temperature sensitive than protected C (up to 53% in cool climates). The proportion of protected C increases with temperature and soil fineness, driving lower emergent bulk SOC sensitivity in warmer, fine-textured regions. Global models exhibit wide spreads in pool proportions (16–85% protected) and often misrepresent pool-specific sensitivities, which can bias projections of SOC responses to climate change. The study advocates that pool distributions and their climatological temperature sensitivities are essential ecosystem properties that models should reproduce. It urges routine reporting of soil C pool distributions in future CMIP phases and calls for concurrent benchmarking against radiocarbon-inferred ages. Future research should prioritize measuring transient pool-specific responses across climates and biomes, improving model depth resolution and parameter documentation, and refining process representations (e.g., mineralogical controls and transfer coefficients) to align model pools with measurable fractions and observed age distributions.

- The benchmarking relies on spatial climate gradients (space-for-time), which capture long-term climatological sensitivities but cannot fully constrain transient responses or acclimation dynamics. - Analyses exclude permafrost and MAT ≤ 0 °C regions, peatlands, hyperarid areas, and very low-precipitation sites due to distinct controls and data limitations, limiting generalizability to those environments. - Interpretability mismatches exist between operationally defined measured fractions and model pools; mapping protected C to passive pools in first-order models is an approximation that may break down regionally (e.g., high latitudes with vertically resolved processes). - Model SOC depths are not explicitly defined and are not depth-resolved; comparisons assume cSoil approximates 0–1 m SOC. - Only CMIP6 models that reported soil pool distributions were included, potentially introducing selection bias. - Turnover times in some microbially explicit models (MIMICS, CORPSE) appear shorter than radiocarbon constraints, affecting sensitivity and future projections; radiocarbon constraints are not directly assimilated. - The mineral-associated SOC data product, while cross-validated, carries uncertainties and sparser constraints in cooler climates and sandy soils.