Figure 1.
Effect of occurrence thinning on geographical sample bias.
The colors on the map represent the regional sampling density, warmer colors indicating higher sample densities. Occurrence thinning substantially reduces the geographic sampling bias, as illustrated by the disappearance of the red blob along the French Riviera and closeby localities (panel A → B). There is less geographic sampling bias in the native range, so occurrence thinning does not have a big influence on the kernel density maps of that region (panel C → D). Note that the slightly elevated density close to the Spanish-French border in the Bay of Biscay (panels A and B) is caused by samples in the Mediterranean of which the kernel extends across land; there are no occurrences of C. cylindracea known from that area.
Figure 2.
Results of the surveying procedure to identify the predictors present in top-scoring models.
Each box contains the results of the survey for occurrence records from the native range, the invaded European range, or both ranges combined. Each column within a box represents a single survey carried out on one set of thinned coordinates. The circle diameter represents how often the variable in question occurred in the top 10 highest-scoring models (test AUC) for that set of thinned occurrences. The representation of each predictor in the top 10 is also summarized across columns (percentage indicates how many of the top 10 models had the predictor), and the consensus predictor set across ranges is indicated in the box on the right.
Figure 3.
Impact of modeling choices on the transferability of SDMs.
The transferability of models is approximated by test AUC (panel A) and the global niche overlap (Schoener’s D, panel B). Columns and rows represent the combinations of the four factors that were varied in our experimental design and are identical in both panels. The values are also plotted as colors along a color gradient to permit rapid visual assessment of the important factors, with warmer colors indicating higher values. Each AUC value in panel A represent the average of the AUCnative-invaded and AUCinvaded-native for the corresponding condition.
Figure 4.
Species distribution model for Caulerpa cylindracea.
Panel A shows global areas predicted to have suitable macroecological conditions for the species. This map uses a threshold for Maxent’s logistic suitability corresponding to the 10% training presences (threshold = 0.053) and predictions are plotted only for coastal areas (less than 7 pixels from shore), with predictions in the open ocean masked. Panels B and C show the continuous logistic model output for the native and invaded ranges, respectively. The corresponding multivariate environmental similarity surface (MESS) maps are shown in panels D and E, and the most dissimilar (MoD) variables in those areas that require extrapolation are shown in panels F and G.