Soil carbon has long been central to climate mitigation discussions. However, while total soil organic carbon is widely mapped and reported, the more persistent fraction of soil carbon has received far less global attention. This gap motivated our study. We wanted to know not only where soil carbon is stored, but where it is stabilized and therefore more likely to contribute to long-term climate regulation.
The project started from a methodological challenge. Across the literature, stabilized soil carbon has been described using different terms, including recalcitrant, persistent, passive, or stable carbon. These terms are often used interchangeably, but they may reflect different measurement approaches. To make a global synthesis possible, we first needed to define what could be compared. We therefore adopted a unified operational definition: stabilized soil organic carbon was represented by hydrolysis-resistant organic carbon measured by acid hydrolysis, or, when direct measurements were unavailable, by total organic carbon minus explicitly reported labile carbon fractions. This step was essential because any global pattern would be meaningful only if the underlying measurements were analytically comparable.
After screening the literature, we built a georeferenced database covering croplands, forests, grasslands, and wetlands. The process was laborious because each record required not only carbon measurements, but also coordinates, soil depth, ecosystem type, and associated environmental information. Many potentially useful studies could not be included because they lacked coordinates, used incompatible methods, or did not report the necessary soil layers. This was one of the most important lessons from the project: global carbon science still depends heavily on careful field measurements and transparent reporting.
With the database established, we used a machine-learning framework to predict stabilized soil organic carbon across the globe. Among several candidate algorithms, XGBoost showed the strongest predictive performance and was used to generate global maps at fine spatial resolution. We also used SHAP analysis and moving-window partial correlations to identify the dominant controls of stabilized carbon across ecosystems. This combination allowed us to move beyond a single global average and explore how drivers varied spatially and among land-use types.
One of the clearest findings was the importance of wetlands. Wetlands stored much more stabilized soil carbon than forests, grasslands, or croplands. This pattern reflects the unique biogeochemical environment of wetland soils, where water saturation limits oxygen availability, slows decomposition, and promotes long-term carbon preservation. Cold-temperate regions also emerged as important hotspots, consistent with the role of low temperature in slowing microbial decomposition and supporting carbon persistence.
Another key result was that soil properties, rather than climate or human activities alone, explained most of the spatial variation in stabilized carbon. This does not mean that climate and management are unimportant. Instead, it suggests that their effects are often mediated through soil conditions such as organic carbon content, pH, texture, bulk density, and nutrient status. In croplands, for example, fertilizer use, straw return, tillage, irrigation, and cropping intensity all influenced stabilized carbon, but their effects depended on the soil context. This finding reinforces the idea that soil carbon management must be ecosystem-specific and regionally targeted.
A central contribution of the study is the concept of soil negative carbon potential, or SNCP. We define SNCP as the proportion of stabilized soil organic carbon within total soil organic carbon. This metric was designed to capture not just how much carbon is stored, but how much of it is likely to persist. Regions with higher SNCP may have greater potential to contribute to durable climate mitigation. Importantly, we found that higher SNCP was associated with lower greenhouse gas emissions and showed positive relationships with economic indicators, while crop yield trade-offs appeared limited. These relationships do not prove causality, but they suggest that soil carbon stabilization could be connected to broader climate and socioeconomic benefits.
For us, the broader message is that soil-based climate mitigation should not be assessed only by the size of the carbon pool. Durability matters. A soil that accumulates carbon rapidly but loses it easily may provide less long-term climate benefit than a soil that stores a larger fraction in stabilized forms. By mapping stabilized soil organic carbon and proposing SNCP, our study provides a framework for evaluating soil carbon through the lens of persistence.
This work also points to several future directions. First, more field measurements are needed in underrepresented ecosystems and regions, especially where global datasets remain sparse. Second, mechanistic studies are required to better understand how microbial processes, mineral protection, aggregation, and land management interact to control carbon stabilization. Third, Earth system models should better represent stabilized carbon pools rather than treating soil carbon as a single homogeneous reservoir. Finally, policy frameworks for carbon neutrality could benefit from indicators that distinguish short-term carbon accumulation from long-term carbon persistence.
In the end, this paper is about making the hidden part of soil carbon more visible. Stabilized soil carbon is not always the most dynamic or easily measured fraction, but it may be one of the most important for long-term climate mitigation. By identifying where this carbon is stored and what controls it, we hope this work provides a benchmark for future soil carbon research, Earth system modeling, and land-based climate strategies.