What Does Data Tell Us About Soil Carbon Model Structure?.

Many types of models have been developed to describe soil carbon cycling.  Most of these models are based on kinetic and stoichiometric principles to constrain elemental cycling within the soil and carbon exchange with vegetation and the atmosphere. In particular, various models have been developed to account for different assumptions of microbial roles in soil biogeochemistry.  Not many of them, however, have been rigorously examined against data.  We have used the data assimilation approach to test various types of soil carbon models against data from soil incubation, soil carbon data at the ecosystem plot level, and regional and global soil carbon databases.  For example, we used data assimilation approach to separate soil carbon (C) efflux from a long-term (385 days) incubation experiment with two types of soil into different source pools. We used first-order linear kinetic models with one, two or three pools in data assimilation and used probability density functions as a criterion to judge the best model to fit the datasets. Our results indicated that soil C release trajectories over the 385 days of the incubation study were best modeled with a two-pool C model.  This analysis was based on information contained in incubation data to separate carbon pools. Our results indicate that soil organic matter (SOM) is heterogeneous in structure and consist of various pools with different intrinsic turnover rates. We have applied this approach to hundreds of soil incubation data. All analyses indicate that two- or three-pool, linear kinetic models can adequately represent each of the soil incubation data.  Nevertheless, estimated coefficients of carbon fractions for each pools and mineralization in each pool nonlinearly vary with soil types and environmental conditions, likely due to microbial mediation.  Similar results were obtained when we analyze models at ecosystem, regional and global scales