Numerical comparison of five soil microbial models, in relation to measurable soil organic matter fractions

The last decade has seen an increase of innovative soil carbon models that takes explicitly into account the microbial community interaction with the soil organic matter, and various state of protection of the soil organic matter itself. This proliferation is fuelled by (a) the recognition that microbial ecology is the main determinant of soil organic carbon mineralization, and (b) that soil organic matter can be protected from microbial degradation in various ways. Of particular interest is the interaction of the organic matter with the mineral fraction of the soil, which can lead to mineral adsorbtion and the formation of soil aggregates. However, the uncertainties about soil microbial ecology and organic matter – mineral fraction have led to the formulation of various soil microbial models, each one modelling some of the aspects of the complex net of interacting processes, but not other. These models often use different assumptions, model structures, and pool definitions. The lack of comparability among models, and the low comparability of models with measurable data, makes it hard to discriminate among them and to use them to assess the driving processes relevant for soil carbon dynamics depending on climatic, soil and vegetation conditions. A first attempt to compare some of these models has been presented in Sulman et al. (2018); however, the lack of a harmonization framework for the models, and the use of lumped model pools/flows such as soil respiration and bulk soil organic carbon, have led to the conclusion that the uncertainties are too elevated to discriminate among the models. Here, we propose a framework to harmonize five different soil microbial models among them (MEND, CORPSE, MIMIC, DEMENT, RESOM), and harmonize them with measurable soil organic matter fraction widely recognized as related to processes of interaction with the soil mineral fraction (aggregates, mineral associated organic matter, dissolve organic matter, and particulate organic matter). We reformulated the five models based on this framework, and analysed them on the same parameter space to understand in which regions of said space the models gave results that were substantially different. The results show that: (a) the model can be clearly distinguished in most regions of the parameters space, (b) it is possible to calculate an index of robustness of the models. This information can help in design specific experiments to test the models and, this way, get insights about the driving processes in certain conditions (different climates, soils, vegetations); moreover, the robustness index can give indication about their applicability to different conditions, which is of utmost importance if they are to inform Earth System Models.