Prospective modelling of environmental dynamics. A methodological comparison applied to mountain land cover changes

During the last 10 years, scientists performed significant advances in modelling environmental dynamics. A wide range of new methodological approaches in geomatics - such as neural networks, multi-agent systems or fuzzy logics - was developed. Despite these progresses, the modelling softwares available have to be considered as experimental tools rather than as improved procedures able to work for environmental management or decision support. Particularly, the authors consider that a large number of publications suffer from lakes in the validation of the model results. This contribution describes three different modelling approaches applied to prospective land cover prediction. The first one, a combined geomatic method, uses Markov chains for temporal transition prediction while their spatial assignment is supervised manually by the construction of suitability maps. Compared to this directed method, the two others may be considered as semi automatic because both the polychotomous regression and the multilayer perceptron only need to be optimized during a training step - the algorithms detect themselves the spatial-temporal changes in land cover. The authors describe the three methodological approaches and their practical applications to two mountain studied areas: one in French Pyrenees, the second including a large part of Sierra Nevada, Spain. The article focuses on the comparison of results. The main result is that prediction scores are on the more high that land cover is persistent. They also underline that the geomatic model is complementary to the statistical ones which perform higher overall prediction rate but produce worse simulations when land cover changes are numerous