ABSTRACT
To protect soils against threats, it is necessary to predict the consequences of human activities and global change on their evolution on a ten to hundred year time scale. Mechanistic modelling of soil evolution is then a useful tool. We analysed the ability of the SoilGen model to be used for projections of soil characteristics associated to various soil threats: vertical distributions of <2µm fraction, organic carbon content (OC), bulk density and pH. This analysis took the form of a functional sensitivity analysis in which we varied the initial conditions (parent material properties) and boundary conditions (co-evolution of precipitation and temperature; type and amount of fertilization and tillage as well as duration of agriculture). The simulated scenario variants comprised anthroposequences in Luvisols at two sites with one default scenario, six variants for initial conditions and 12 variants for boundary conditions. The variants reflect the uncertainties to our knowledge of parent material properties or reconstructed boundary conditions. We demonstrated a sensitivity of the model to climate and agricultural practices for all properties. We also conclude that final model results are not significantly affected by the uncertainties of boundary conditions for long simulations runs, although influenced by uncertainties on initial conditions. The best results were for organic carbon, although improvements can be reached through calibration or by incorporating a dynamic vegetation growth module in SoilGen. Results were poor for bulk density due to a fixed-volume assumption in the model, which is not easily modified. The <2µm fraction depth patterns are reasonable but the process of clay new formation needs to be added to obtain the belly shape of the Bt horizon. After calibration for organic carbon under agriculture, the model is suitable for producing soil projections due to global change.
