Microbial carbon use efficiency promotes global soil carbon storage

Abstract

Soils store more carbon than other terrestrial ecosystems^1,2. How soil organic carbon (SOC) forms and persists remains uncertain^1,3, which makes it challenging to understand how it will respond to climatic change^3,4. It has been suggested that soil microorganisms play an important role in SOC formation, preservation and loss^5-7. Although microorganisms affect the accumulation and loss of soil organic matter through many pathways^4,6,8-11, microbial carbon use efficiency (CUE) is an integrative metric that can capture the balance of these processes^12,13. Although CUE has the potential to act as a predictor of variation in SOC storage, the role of CUE in SOC persistence remains unresolved^7,14,15. Here we examine the relationship between CUE and the preservation of SOC, and interactions with climate, vegetation and edaphic properties, using a combination of global-scale datasets, a microbial-process explicit model, data assimilation, deep learning and meta-analysis. We find that CUE is at least four times as important as other evaluated factors, such as carbon input, decomposition or vertical transport, in determining SOC storage and its spatial variation across the globe. In addition, CUE shows a positive correlation with SOC content. Our findings point to microbial CUE as a major determinant of global SOC storage. Understanding the microbial processes underlying CUE and their environmental dependence may help the prediction of SOC feedback to a changing climate.

Fig. 1. Two contrasting pathways in determining the relationship between microbial CUE and SOC storage.

a, The first pathway indicates that a high CUE favours the accumulation of SOC storage through increased microbial biomass and by-products. b, The second pathway emphasizes that a high CUE stimulates SOC losses via increased microbial biomass and subsequent extracellular enzyme production that enhances SOC decomposition.

Fig. 2. CUE–SOC relationship.

a,b, The CUE–SOC relationship that emerged from the meta-analysis of 132 measurements (a) and data assimilation using the microbial model with 57,267 globally distributed vertical SOC profiles (b). The black lines and statistics shown are the partial coefficients from mixed-effects model regressions (see Extended Data Tables 1 and 2 for details).

Fig. 3. Maps of global SOC stock and related components.

The maps were obtained from 57,267 globally distributed vertical soil profiles using the PRODA approach with the microbial model (see Methods for details). a–h, Global distributions of SOC stock (a), microbial CUE (b), non-microbial carbon transfer (c), plant carbon inputs (d), carbon input allocation (e), baseline decomposition (f), environmental modifier (g) and vertical transport rate (h). Values shown are predicted by the best-guess model calibrated using all available data (Methods; see Extended Data Fig. 7 for their uncertainties). Boxplots represent the SOC stock for the top 1 m and model components in different pre-defined climate zones (Supplementary Fig. 3). The lower, middle and upper hinges show the first, median and third quartiles of the distribution. Whiskers in the boxplot represent the 1.5 times the interquartile range from the hinges. The components and units for the maps and the boxplots are the same.

Fig. 4. Microbial CUE as the primary regulator of global SOC storage.

a, Spatially explicit model components obtained through the PRODA approach with the microbial model were substituted by their spatially invariant counterparts to assess their influence on SOC stocks (that is, the sum of absolute deviations from PRODA estimates over the globe) and distributions (that is, deviation of explained variation in observations defined by equation (2)). b, We further proportionally changed the retrieved parameter values of different components and found that global SOC stock is most sensitive to changes of CUE. Error bars and shaded areas show the 2σ confidence interval in the 200-time bootstrapping. It is noted that the axes in a are scaled by a signed square-root function.

Extended Data Fig. 1. Spatial distributions of datasets.

Spatial distributions of datasets used in meta-analysis (a) and vertical SOC profiles used in the PRODA approach (b).

Extended Data Fig. 2. Workflow of the PROcess-guided deep learning and DAta-driven modelling (PRODA) approach.

We first applied the Bayesian data assimilation to fuse observational data at each soil profile with the microbial model. Parameters that represent different components in modelling soil carbon cycle were estimated through the Markov Chain Monte Carlo (MCMC) method. A deep learning model then predicted the optimised parameter values (i.e., the mean value of the posterior distribution after MCMC) by a set of 60 environmental variables. The predicted parameter values by the deep learning model were further applied in the microbial model to calculate SOC storage and related components.

Extended Data Fig. 3. Structure of the microbial-process explicit model used in this study.

See Methods for detailed descriptions of the model.

Extended Data Fig. 4. Varying CUE-SOC relationships at the steady state under different parameter values in the microbial model.

${\tau }_{\mathit{ENZ},\mathit{decay}}$ is the turnover time for enzyme decay (unit: yr). ${\tau }_{\mathit{MIC}}$ is the turnover time for microbial mortality (unit: yr).

Extended Data Fig. 5. CUE-SOC relationships at different soil depths.

Results are from assimilating all the available soil profiles (n = 57,267) to the microbial model. Declining explanatory power of CUE to the variation in SOC with soil depths possibly indicates more interactions of organic matter with mineral particles at depth.

Extended Data Fig. 6. Influences of environmental variables on different components investigated in this study.

Results are the median values from 1000-time permutations to the best-guess model. Error bar indicates the two-sigma confidence interval. Results for the plant carbon input is not available because we directly used the simulation results by CLM5 as the carbon input instead of predicting it by the neural network.

Extended Data Fig. 7. Uncertainties of retrieved global SOC storage and related model components.

The uncertainty maps (except carbon input) showed the standard deviations in the 200-time bootstrapping. Uncertainty of carbon input derived from the CLM5 interannual simulations.