Selecting predictors to maximize the transferability of species distribution models: lessons from cross-continental plant invasions

Aim Niche-based species distribution models (SDMs) are commonly used to predict impacts of global change on biodiversity, but the reliability of these predictions in space and time depends on their transferability. We tested how the strategy used to choose predictors impacts the transferability of SDMs at a cross-continental scale. Location North America, Eurasia and Australia. Method We used a systematic approach including 50 Holarctic plant invaders and 27 initial predictor variables, considering 10 different strategies for variable selection, accounting for the proximality, multicollinearity and climate analogy of predictors. We compared the average performance of each strategy, some of which used a large number of predictor combinations. Next, we looked for the single best model for each species across all the predictor combinations retained in the analysis. Transferability was considered as the predictive success of SDMs calibrated in the native range and projected onto the invaded range. Results Two strategies showed better SDM transferability on average: a set of predictors known for their ecologically meaningful effects on plant distribution, and the two first axes of a principal component analysis calibrated on all predictor variables (S-pc2). From the more than 2000 combinations of predictors per species across strategies, the best set of predictors yielded SDMs with good transferability for 45 species (90%). These best combinations consisted of eight randomly assembled (39 species) or uncorrelated predictors (6 species) and S-pc2 (5 species). We also found that internal cross-validation was not sufficient to give full information about the transferability of a SDM to a distinct range. Main conclusion Transferring SDMs at the macroclimatic scale, and thus anticipating invasions, is possible for the large majority of invasive plants considered in this study, but the accuracy of the predictions relies strongly on the choice of predictors. From our results, we recommend including either proximal and state-of-the-art variables or a reduced and orthogonalized set to obtain robust SDM projections.