Where are the wild things? Why we need better data on species distribution

Aim The effects of ongoing global change are causing increasing concern about the ability of species or biomes to shift or adapt. Tremendous efforts have been made to develop ever more sophisticated species distribution models to provide forecasts for the future of biodiversity. All these models rely on species occurrence data, either for calibration or validation. Here we evaluate (i) whether distribution data diverge among widely used sources, for supposedly well-known taxa, and (ii) to what extent these divergences affect species distribution models. Location Europe (as an example). Methods We compared the distribution maps of 21 of the most common European trees, according to four large-scale, putatively reliable sources of distribution data. For each species, we compared the outputs of correlative species distribution models built using occurrence data from each of these sources of data. We also investigated how discrepancies in large-scale occurrence data affected the validation scores of two process-based tree distribution models. Results Maps of tree occurrence diverged in 8-74% of the forested area, depending on species. These discrepancies affected projections of niche models: for example, 22-75% of the area projected as suitable by at least one model generated using one source of data was not projected as such by all other models. For most species, this proportion increased under scenarios of climate change, whatever the model used. To a lesser extent, uncertainties on current species distributions also affect the validation score of process-based distribution models. Main conclusions Reliable, widely used sources of occurrence data strongly diverge even for well-known taxa - the most common European trees. Scientists and stakeholders should acknowledge this gap in knowledge, since accurate data are a prerequisite to providing stakeholders with robust forecasts on biodiversity. Participatory science programmes and remote sensing techniques are promising tools for rapidly gathering such data.