Improved global-scale predictions of soil carbon stocks with Millennial Version 2

Soil carbon (C) models are used to predict C sequestration responses to climate and land use change. Yet, the soil models embedded in Earth system models typically do not represent processes that reflect our current understanding of soil C cycling, such as microbial decomposition, mineral association, and aggregation. Rather, they rely on conceptual pools with turnover times that are fit to bulk C stocks and/or fluxes. As measurements of soil fractions become increasingly available, it is necessary for soil C models to represent these measurable quantities so that model processes can be evaluated more accurately. Here we present Version 2 (V2) of the Millennial model, a soil model developed to simulate C pools that can be measured by extraction or fractionation, including particulate organic C, mineral-associated organic C, aggregate C, microbial biomass, and low molecular weight C. Model processes have been updated to reflect the current understanding of mineral-association, temperature sensitivity and reaction kinetics, and different model structures were tested within an open-source framework. We evaluated the ability of Millennial V2 to simulate total soil organic C (SOC), as well as the mineral-associated and particulate fractions, using three independent data sets of soil fractionation measurements spanning a range of climate and geochemistry in Australia (N = 495), Europe (N = 175), and across the globe (N = 659). When using all the data together (N = 1329), the Millennial V2 model predicted SOC (RMSE = 3.3 kg C m−2, AIC = 675, Rin2 = 0.31, Rout2 = 0.26) better than the widely-used first-order decomposition model Century (RMSE = 3.4 kg C m−2, AIC = 696, Rin2 = 0.21, Rout2 = 0.18) across sites, despite the fact that Millennial V2 has an increase in process complexity and number of parameters compared to Century. Millennial V2 also reproduced the observed fraction of C in MAOM and larger particle size fractions for most latitudes and biomes, and allows for a more detailed understanding of the pools and processes that affect model performance. It is important to note that this study evaluates the spatial variation in C stock only, and that the temporal dynamics of Millennial V2 remain to be tested. The Millennial V2 model updates the conceptual Century model pools and processes and represents our current understanding of the roles that microbial activity, mineral association and aggregation play in soil C sequestration.