Habitat modelling: a multi‐models approach to map the potential distribution of alpine vegetation assemblages in France

International audience; This paper presents a method to assess the potential distribution of alpine species assemblages in the crystalline Belledonne-Ecrins Mountains (France), at a relative fine scale. Using ecological variables and vegetation monitoring plots, we computed 8 species distribution models (SDMs) to predict the potential distribution of 6 alpine species assemblages. These vegetation communities were first constructed using graph theory approaches based on species' lists elaborated by the Alpine National Botanical Conservatory (CBNA) from field work on monitoring plots. The goal is to elaborate species assemblages (or modules) that are not constrained by phyto-sociological principles but based on the species co-occurrence at the monitoring plots. Species inside a module are thus linked with each other by their ecological affinities and not by botanical characteristics, sustaining the use of ecological dataset to predict their potential distribution. From expert knowledge, the 6 assemblages were selected for their wide representation on the field, their ecological dissimilarities (contrasted niches) and their botanical consistency. They are essentially distributed in sub-alpine and alpine vegetation levels. Based on the ecological niche theory, species distribution models (SDMs) were used to identify areas that are ecologically suitable for the presence of these 6 assemblages. In all, we used 7 environmental variables, mainly derived from a 25m Digital Elevation Model. Hence, alpine species are greatly influenced by the presence and duration of snow cover and consequently by topography and solar radiation. Moreover, the coarse resolution of climatic data did not match the prerogative of the potential distributions' maps. The BIOMOD platform was used to compute 8 SDMs (ANN, CTA, FDA, GAM, GBM, GLM, RF, and Maxent). Mean models and coefficient of variations were then calculated based on the best model performances (evaluated based upon expert knowledge). This approach, called "ensemble modelling", gives more consistent results and allows a spatial analysis of the level of agreement between models. The use of ensemble modelling, using simple ecological datasets, has shown great potential to provide reliable species ecological niches, having important implications in vegetation mapping and thus on management decisions regarding biodiversity conservation. Actually, the predictions show good correlations with field data, few false overlaps between modules and good correlations between transition communities and modules overlaps, revealing the power of these techniques for vegetation mapping in relative complex and inaccessible areas.