Supplementary material to "Mapping global distributions, environmental controls, and uncertainties of apparent top- and subsoil organic carbon turnover times"

Lei Zhang, Lin Yang, Thomas W. Crowther, Constantin M. Zohner, Sebastian Döetterl, G.b.m. Heuvelink, Alexandre M.j.‐c. Wadoux, A‐Xing Zhu, Y.h. Pu, Feixue Shen, Haozhi Ma, Yibiao Zou, Chenghu Zhou2025 · DOI

This supplementary material provides the technical backbone for a major study on Soil Organic Carbon (SOC) turnover times ($\tau$). By integrating massive global datasets with machine learning, the authors have mapped how long carbon stays in the soil across different depths and biomes, revealing critical gaps in our current climate models.

TL;DR

The researchers produced high-resolution global maps of soil carbon turnover times for both topsoil and subsoil, using Random Forest models to synthesize data from thousands of soil profiles. Their findings show that current Earth System Models (CMIP6) systematically underestimate how long carbon remains in the soil, potentially miscalculating future climate feedbacks.

What Problem Does This Paper Address?

Soil is the largest terrestrial carbon pool, but the rate at which it "turns over"—releasing carbon back into the atmosphere—is poorly quantified at a global scale. Most existing models simplify soil into a single layer or rely on theoretical equations that don't match real-world observations.

Specifically, there is a lack of data-driven maps that distinguish between topsoil (0–30 cm) and subsoil (30–100 cm). Since subsoil carbon is often older and influenced by different environmental factors (like mineralogy rather than just temperature), failing to separate these layers leads to significant uncertainties in climate predictions.

Methods

The authors employed a multi-step geospatial predictive mapping workflow:

  • Data Harmonization: They used equal-area smoothing splines to standardize soil organic carbon content across varying profile depths from multiple databases.
  • Filling the Gaps: Missing bulk density (BD) values—crucial for calculating carbon stocks—were estimated using pedotransfer functions (exponential and log-transformed functions based on SOC content).
  • Carbon Input Modeling: $\tau$ is calculated as the ratio of SOC stock to carbon input. The authors meticulously estimated carbon inputs by combining Net Primary Production (NPP) with root distribution models and "litterfall fractions" (the portion of plant matter that actually enters the soil).
  • Machine Learning: A Random Forest (RF) model was trained on environmental covariates (climate, soil chemistry, topography) to predict $\tau$ globally.
  • Variable Importance: They used permutation-based feature importance to identify which environmental factors (e.g., temperature vs. clay content) dominate carbon cycling in different regions.

Datasets

The study draws from an impressive array of global sources:

  • Soil Profiles: WoSIS (World Soil Information Service), NCSCD (Northern Circumpolar Soil Carbon Database), and the China Soil Survey.
  • Vegetation & Roots: MODIS NPP data, the RSIP (Root Systems of Individual Plants) database, and the GRooT (Global Root Traits) database.
  • Environmental Covariates: WorldClim (climate), SoilGrids (soil properties), and GTOPO30 (topography).
  • Benchmarks: Results were compared against four major CMIP6 Earth System Models (e.g., ACCESS-ESM1-5, NorESM2-LM).

Key Insights and Ideas

  • The Depth Divide: Environmental controls shift dramatically with depth. While climate (temperature and precipitation) is the primary driver of topsoil turnover, soil physical and chemical properties (like the C:N ratio and pH) become much more influential in the subsoil.
  • The Tropical Anomaly: In tropical ecosystems, the relationship between temperature, NPP, and SOC stocks follows a unique 2nd-order polynomial curve. Unlike extratropical regions where higher temps generally increase productivity, tropical plants are often near their thermal optimum; further warming may increase respiration faster than growth, reducing net carbon gain.
  • Model Underestimation: Perhaps the most striking "aha moment" is that 92% of global land area shows a discrepancy where Earth System Models predict faster turnover than the data suggests. Models are essentially "leakier" with carbon than the real world.

Key Results

  • Model Accuracy: The Random Forest model performed exceptionally well, with validation $R^2$ values reaching 0.84–0.94 for topsoil $\tau$ in most biomes.
  • Global Drivers: Climate factors account for 46.7% of the importance for topsoil $\tau$ globally. However, in the subsoil, soil chemical properties (like C:N ratio and Cation Exchange Capacity) rise in importance.
  • Quantifying the Gap: The average CMIP6 model has a median "difference ratio" of only 0.39 compared to observations. This suggests that models may be overestimating the speed of the soil carbon cycle by more than double in many regions.

Limitations and Future Work

  • Human Influence: The authors note that human-induced impacts like grazing, deforestation, and non-cropland manure application were not fully included due to a lack of accurate global datasets.
  • Subsoil Inputs: The study assumes the fraction of aboveground NPP transferring to subsoil is zero—a simplification that future research could refine.
  • Data Sparsity: While the dataset is massive, sample density is still low in the tropics and high-latitude permafrost regions, leading to higher uncertainty in those critical areas.

Summary updated

BibTeX
@misc{zhang2025supplementary,
  author = {Lei Zhang and Lin Yang and Thomas W. Crowther and Constantin M. Zohner and Sebastian Döetterl and G.B.M. Heuvelink and Alexandre M.J.‐C. Wadoux and A‐Xing Zhu and Y.H. Pu and Feixue Shen and Haozhi Ma and Yibiao Zou and Chenghu Zhou},
  title = {{Supplementary material to "Mapping global distributions, environmental controls, and uncertainties of apparent top- and subsoil organic carbon turnover times"}},
  year = {2025},
  doi = {10.5194/essd-2024-618-supplement},
}
Figure