Simulating trends in soil organic carbon in long-term experiments using the soil-plant-atmosphere model DAISY

DAISY, a mechanistic model for the simulation of crop production, soil water and N dynamics in agro-ecosystems, was used for simulating trends in total soil carbon (C) for various long-term experiments as part of a model evaluation exercise. Of the seven core datasets in the global SOMNET database, DAISY could be applied to the four arable sites and, with certain critical assumptions, to the single grassland site. The model was, however, not applicable to the two forest sites, because currently DAISY cannot simulate soil water and temperature dynamics in a forest system and does not incorporate leaching of dissolved organic substances, faunal litter incorporation or the effect of low pH on decomposition. The DAISY crop submodel produced some overestimation of crop yields, especially at low N input levels, and simulated root C input to the soil may thus represent a slight overestimation. This may have been balanced, however, by the fact that no account is taken of rhizodeposition or root turnover in the current version of the DAISY model. For this reason too, plant-derived C input in perennial crops (pastures) had to be estimated and input separately. Under these critical assumptions, the model was capable of simulating long-term trends in total soil C quite well for all included sites and treatments (overall correlation, R2= 0.88), except for the Bad Lauchstadt arable site. It was evident that sufficient experimental data on belowground, plant-derived C inputs was not available, especially with perennial crops. Future research in this area should thus have high priority, because such data are a prerequisite if soil organic matter models are to incorporate and improve the simulation of this very important mechanism of soil C input. The partitioning and parameterisation of C between pools in the DAISY model has been discussed and indications were found that the parameterisation of the microbial biomass pools in the model are somewhat inadequate.