Modeling temperature sensitivity of soil organic matter decomposition: Splitting the pools

The direction and magnitude of change of soil organic carbon (SOC) stocks due to global warming depend strongly on the temperature sensitivity (e.g., Q(10)) of carbon mineralization. To date, most multi-pool SOC models assume a general Q(10) of 2 despite experimental evidence suggesting different Q(10) for different carbon fractions. The aim of this study was to test if the use of experimentally derived pool specific Q(10) values improves the performance of SOC models. Five contrasting data sets from three field experiments and two laboratory incubations were used to study the link between carbon pool recalcitrance and Q(10) using two different approaches: a) Bayesian calibration of the Daisy SOC model parameters to infer Q(10) of SOC and crop-litter pools, and b) using measured Q(10) values of carbon degrading enzymes as proxies for Q(10) of different Daisy pools. Namely beta-glucosidase (median Q(10) of 1.82) was assigned to metabolic litter and phenol/peroxidase (1.35) to structural litter and both SOC pools. To partition litter-carbon and SOC into model pools, the lignin-to-nitrogen ratio and the ratio of aliphatic/aromatic-carboxylate carbon were used, respectively. Measurements included soil microbial biomass, soil carbon dioxide (CO2) evolution and remaining carbon in soils and crop-litter. In the Bayesian calibration, strong differences in inferred Q(10) values of the same pools between experiments suggested that intrinsic substrate recalcitrance was not the main driver of temperature sensitivity. For field experiment simulations, both the Q(10) values derived by Bayesian calibration and measured enzyme Q(10) were centered around values below 2, contrasting with high Q(10) values for mineralization under laboratory incubations (close to 3). Furthermore, assigning measured phenol/peroxidase Q(10) values to the slow crop-litter as well as both SOC pools and (beta-glucosidase to the fast crop-litter pool (approach b), could significantly improve model performance compared to using the default Q(10) value of 2 for all pools. Root-mean-squared-deviation reductions were between 3 and 10% for field experiments, with no change in the laboratory experiments. Thus, site specific Q(10) values of soil enzymes show potential as proxies for pool specific Q(10). We present a new conceptual framework to explain the observed differences in temperature sensitivities between experiments as a result of two fundamental driving factors classified in a) state variables, that fluctuate in time, and b) soil properties, that are constant over decades. Measured enzyme Q(10) values were interpreted as a proxy incorporating both factors. More than intrinsic substrate recalcitrance, the state variables such as physical protection, substrate abundance and unfavorable conditions for microorganisms control temperature sensitivity of mineralization. To reduce the uncertainty in global SOC simulations under a changing climate, their relative contributions should be disentangled and then implemented into SOC models.