Global spatially explicit land system models are important tools to gain a better scientific understanding and to investigate future development trajectories in form of scenarios. However, uncertainties e.g. of input data and model structure affect their simulation results. This article investigates the impact of model initialization on land change simulations and builds a single-model ensemble based on the global land system model LandSHIFT. Each ensemble member is initialized with a unique combination of a remote sensing land-cover dataset. A specific method to derive model parameter values is used for land suitability evaluation, to identify and quantify significant discrepancies between model outcomes obtained with different ensemble members. The discrepancies related to both the land-use maps generated as starting conditions of model runs and to the simulated land change and natural vegetation loss. The results underline the importance to account for the above uncertainties in global simulation studies and communicating them transparently.
