A team of researchers led by Franco Fernández-Catinot of the Max Planck Institute for Biogeochemistry in Jena, Germany, working with colleagues in Argentina and China, set out to do something surprisingly rare in soil science: rigorously test whether two-pool models built on the particulate-mineral framework can actually reproduce real measurements. Their approach combined a laboratory incubation experiment with inverse modelling, a technique that estimates unknown model parameters directly from empirical observations. The soils came from high-elevation grasslands in the Córdoba mountains of central Argentina, sampled at 2,100 meters above sea level, where Mollisols derived from granitic substrates support short grasses and forbs adapted to a harsh, frost-prone climate.

The experimental design was deliberately straightforward. The team placed 50 grams of soil in flasks and subjected it to two treatments: control soils left untouched, and soils mixed with one gram of dried litter from Muhlenbergia peruviana, a dominant and highly decomposable grass species in the region. The microcosms were held at 25 degrees Celsius and field capacity for six months. Throughout the incubation, the researchers measured carbon dioxide respiration at eight time points using sodium hydroxide traps and titration. Crucially, they also destructively harvested soils at the beginning, middle, and end of the experiment to directly measure how the particulate and mineral-associated carbon pools changed over time, using a physical fractionation method that separates fractions at a 53-micrometer sieve threshold.

This combination of pool contents and respiration fluxes is what gave the study its analytical teeth. The researchers fitted two families of compartmental models to the data. The two-pool models treated the soil as a particulate carbon compartment feeding into a mineral-associated one. The three-pool models added a separate, fast-cycling litter carbon compartment ahead of the other two. Model performance was judged using the Akaike Information Criterion and mean squared error, both of which reward goodness of fit while penalizing unnecessary complexity. Collinearity tests confirmed that combining respiration data with pool content measurements substantially constrained the parameter estimates, making the comparison statistically meaningful rather than an artifact of loose fitting.

The results were striking. In the control soils, which contained no added litter and were assumed to consist only of particulate and mineral-associated carbon, the two-pool models simply could not do both jobs at once. One set of parameters reproduced the changing pool contents but failed badly at predicting respiration; a second set captured respiration but missed the pool dynamics entirely. Only the three-pool model, which carved out roughly 15 percent of the particulate pool as a fast-cycling litter-like compartment, could simultaneously predict both variables with a single parameter set, achieving the lowest AIC and mean squared error values by wide margins. The same pattern held in the litter-addition treatments, where the three-pool model again dominated on every metric.

The implication is provocative: even in soils with no fresh litter added, particulate organic carbon behaved as a heterogeneous mixture rather than a single coherent pool. The two-pool models effectively collapsed processes operating at different timescales into one compartment, and in doing so lost the ability to capture both the fast initial flush of respiration and the slower, gradual changes in pool sizes. The three-pool structure, by contrast, distributed the work cleanly: the litter compartment handled rapid dynamics on the order of months, the particulate pool represented intermediate dynamics on the order of years, and the mineral-associated pool carried the slow dynamics of decades. The authors emphasize that they are not advocating for a specific model configuration, but their results show that the conceptual simplification of soil carbon into just two fractions can fail to capture the multiple timescale responses frequently observed in experiments.

The study then pushed further, asking how model structure and parameter choices shape predictions of carbon persistence. Using the mathematical framework of compartmental systems, the team calculated two system-level metrics: carbon age, the time elapsed since carbon atoms entered the soil, and transit time, how long atoms take to pass through the entire system. These probability distributions are sensitive diagnostics of how tortuous the pathways through the soil are. The two-pool models that best fit respiration data predicted much longer mean ages and transit times than the three-pool models, and the two-pool variants fitted to pool contents produced persistence estimates of tens of thousands of years, driven by implausibly low mineral-associated carbon decomposition rates.

In a series of simulation experiments, the researchers systematically altered model structures and parameters. Adding a direct pathway that allowed carbon to bypass the particulate pool and enter the mineral-associated pool directly increased mean carbon age by roughly 96 percent in two-pool models and 187 percent in three-pool models, with transit times rising by 44 and 161 percent respectively. Tripling transfer rates from litter to particulate carbon and from particulate to mineral-associated carbon pushed transit times up by 230 percent, while tripling only the litter decomposition rate barely moved the needle. When structural and parameter changes were combined, transit times soared to roughly seven times the baseline values, revealing strong non-additive interactions. The clearest signal was that transfer rates into the most persistent pool, not decomposition rates of fresh litter, were the dominant control on how long carbon stays locked in soil.

That finding carries real-world weight. Mineral-associated carbon formation is known to be modulated by the saturation deficit of mineral surfaces and by the presence of cations such as oxalate-extractable aluminum and iron and exchangeable calcium, meaning soils rich in these elements may effectively exhibit higher transfer rates and longer transit times. Vegetation matters too: rhizodeposition has been shown to promote mineral-associated carbon formation more efficiently than aboveground inputs, and recent evidence suggests existing mineral-associated carbon can catalyze the formation of more. Yet the authors note that research has focused heavily on quantifying carbon stocks and decomposition rates, while the transfer rates between compartments remain poorly constrained. Reducing that uncertainty, they argue, may be crucial for improving the soil biogeochemical models that inform climate projections and land management decisions.

The broader message is a call for humility and flexibility in how the scientific community conceptualizes soil carbon. The particulate-mineral framework has been enormously productive, offering measurable fractions that map onto distinct formation mechanisms and protection regimes. But this study, grounded in a carefully controlled incubation with simultaneous measurements of pools and fluxes, demonstrates that two timescales are not always enough. Soil carbon models, the authors conclude, should explicitly represent processes operating across multiple temporal scales, with structures chosen to fit the ecosystem and context at hand, and validated by jointly evaluating slow changes in pool sizes against rapid respiration responses. For a pool holding more than 2,000 petagrams of carbon in the top meter of the world’s soils, getting the number of clocks right could matter for everything from climate forecasts to the durability of soil-based carbon removal.

Subject of Research: Timescales of soil organic carbon dynamics in particulate and mineral-associated carbon pools under experimental litter manipulation

Article Title: Temporal dynamics of particulate and mineral-associated carbon reveal three timescales of response to experimental manipulation

Article References: Fernández-Catinot, F., Hu, W., Sarquis, A., Vaieretti, M. V., Pérez-Harguindeguy, N., Feng, X., & Sierra, C. A. (2026). Temporal dynamics of particulate and mineral-associated carbon reveal three timescales of response to experimental manipulation. SOIL, 12(2), 805-819. https://doi.org/10.5194/soil-12-805-2026

Image Credits: AI Generated

DOI: 10.5194/soil-12-805-2026

Keywords: soil organic carbon, particulate organic carbon, mineral-associated organic carbon, soil incubation, carbon models, carbon transit time, carbon age, inverse modelling, soil respiration, carbon persistence, biogeochemistry, grassland soils