Simulation of carbon and nitrogen dynamics in arable soils: a comparison of approaches

Although mechanistic soil-crop models are increasingly accepted as valuable tools in analysing agronomical or environmental issues, potential users are faced with an equally increasing number of available models. In principle, model selection should be based on a rational assessment of its merit with respect to the objectives pursued. Such information may be obtained by comparing the ability of candidate models to predict given sets of experimental data. However, because the basic components of soil-crop models interact strongly in producing model outputs, little can be drawn as to the validity of the approaches used for the individual components. Here, we focused on the soil carbon and nitrogen turnover module of four soil-crop models (CERES, NCSOIL, SUNDIAL, and STICS), which were selected based on their representativity of currently used models, and the range of complexity and process approaches they offered. The C–N modules of models other than CERES were extracted and linked within CERES, so that they were all supplied with the same physical and chemical data. Inputs and outputs other than those involved the N cycle were provided with good reliability by the common CERES shell. The performance of the various modules was assessed according to two criteria: short-term response of topsoil inorganic N to climate and crop residues input, and long-term dynamics of soil organic matter (SOM). Accordingly, data sets involving net mineralization and topsoil inorganic N dynamics under contrasting bare or wheat-cropped soils, and long-term soil carbon data were used to test them. The results highlight a trade-off between the prediction of N mineralization in the short-term (day to year) and SOM dynamics in the long-term (year to decade). On a yearly basis, NCSOIL over-estimated immobilization of inorganic N associated with the decomposition of crop residues, and CERES predicted extremely low mineralization fluxes. STICS and SUNDIAL gave good predictions of soil N supply, but over-estimated the rate at which soil carbon from slow-turnover pools was degraded as a result. Comparison with a model dedicated to predicting SOM turnover (RothC) showed that the discrepancy may be attributed to a strong under-estimation of the turnover of belowground plant material by the plant modules of CERES. Crop models should thus be improved from this point of view before coupling with SOM models.