A Bayesian modeling framework for estimating equilibrium soil organic C sequestration in agroforestry systems

•SOC sequestration potential in agroforestry is often reported as average rates.•Average rates are a zero-order approach that produces artifacts on long time scales.•We calibrated a first-order method to model SOC sequestration in agroforestry.•The variability existing in these systems is dealt with Bayesian methods.•The model can be a speditive assessment of agroforestry SOC sequestration potential. Agroforestry is a form of productive land management that combines trees or bushes with annual crops or pasture, and it can bring benefits in terms of food security and increased carbon (C) sequestration compared with conventional agriculture. But agroforestry as a structured form of agronomic management is relatively new compared with well-established and widespread agronomic systems. Consequently, there is a lack of data and few models of soil organic carbon (SOC) have been developed specifically for agroforestry systems. Also, agroforestry SOC sequestration data measured in field experiments are often reported only as average linear sequestration rates over the study period. This approach, equivalent to zero-order kinetics, makes it difficult to compare results since, in reality, SOC sequestration rates are variable over time and change depending on the duration of measurements. Sequestration rates are also strongly dependent on former C stocks in the soil, further hampering comparisons between agroforestry systems established on different former land uses. To describe the SOC stocks variation over time, researchers often employ models considering at least first-order kinetics. This approach can take care of the two above mentioned issues, considering both the variation of the sequestration over time and the effect of previous land use. However, the variability of agroforestry systems makes applying these models more challenging compared to simpler agricultural systems. To deal with this problem we propose to use detailed uncertainty estimation methods, based on stochastic calibrations that can deal with broad probability distributions. To do so, we adapted a first-order compartmental SOC model to agroforestry systems. It was calibrated within a Bayesian framework on global agroforestry data. Compared to linear coefficients, the model (ICBMAgroforestry) estimates equilibrium SOC stocks of different agroforestry systems probabilistically and is providing uncertainty bounds. These values are independent of initial land use and time duration of the experiments