Popular modeling tools for land change simulation, especially those using Markov chains, undertake model training based only on two land use / cover (LUC) maps. This paper analyses uncertainty and potential errors caused by taking into account only two former, model known, LUC maps. This is illustrated by a simple data set of six LUC maps allowing various Markovian transition matrices; a range even larger by considering different confidence levels. Results underline the randomness in choice of

more »

... nly two training dates. Authors propose alternative methods to Markov chains integrating all available LUC maps in order to simulate forecasting scenarios. To do so, they incorporate all possible LUCC (land use / cover change) budgets to perform simple arithmetic combinations between the six training dates. Comparing Markov chain transitions based on two training dates and alternatively performed change rates taking into account all training dates results to important differences. This study underlines the importance of the choice of training dates during model calibration for path-dependent simulations. TEST AREA AND DATASET The test area is an 8 750 ha catchment located in the western part of the French Department Pyrénées Orientales called Garrotxes. The altitude ranges 350 Paegelow, M. Land Change Modeling Handling with Various Training Dates.