Table 3.
Evaluation of the performance of the SDMs: goodness of fit was evaluated as the percentage of variance explained by the models (PVE)
| Species | PVE | r | Goodness of fit | NA transferability | EU transferability | Number of cases | ||||
|---|---|---|---|---|---|---|---|---|---|---|
| TSS | SI | TSS | SI | Se | SI | Pres. (%) | Tot. | |||
| SDM‐NAEU | ||||||||||
| Acer negundo | 90.95 | 0.97* | 0.94 | 0.83 | 0.94 | 0.79 | 1.00 | 0.52 | 22.6 | 30,1551 |
| Prunus serotina | 92.19 | 0.96* | 0.94 | 0.84 | 0.94 | 0.86 | 1.00 | 0.56 | 18.8 | 30,2082 |
| Quercus rubra | 96.42 | 0.98* | 0.96 | 0.89 | 0.97 | 0.88 | 1.00 | 0.51 | 16.5 | 30,4365 |
| Robinia pseudoacacia | 97.00 | 0.99* | 0.97 | 0.99 | 0.98 | 0.99 | 1.00 | 0.51 | 4.9 | 30,8405 |
| SDM‐NA | ||||||||||
| Acer negundo | 94.72 | 0.98* | 0.98 | 0.88 | – | – | 0.07 | 0.5 | 22 | 29,9138 |
| Prunus serotina | 95.21 | 0.98* | 0.98 | 0.91 | – | – | 0.02 | 0.5 | 18 | 29,9138 |
| Quercus rubra | 97.97 | 0.99* | 0.99 | 0.94 | – | – | 0.01 | 0.5 | 15 | 29,9138 |
| Robinia pseudoacacia | 96.60 | 0.98* | 0.96 | 1.00 | – | – | 0.00 | 0.51 | 2 | 29,9138 |
| SDM‐EU | ||||||||||
| Acer negundo | 80.83 | 0.95* | 0.97 | 0.5 | 0.26 | 0.67 | – | – | 41 | 5958 |
| Prunus serotina | 92.32 | 0.97* | 0.99 | 0.5 | −0.16 | 0.8 | – | – | 24 | 12,369 |
| Quercus rubra | 89.60 | 0.98* | 0.99 | 0.5 | 0.10 | 0.5 | – | – | 53 | 9881 |
| Robinia pseudoacacia | 92.69 | 0.96* | 0.97 | 0.5 | 0.09 | 0.93 | – | – | 90 | 10,285 |
Generalization power was assessed by Pearson's correlation r between observed P/A values and predicted values using an independent dataset (*, P < 0.05). The capacity of the SDM to predict the presence of the species was assessed using true skill statistics (“TSS”), ranging from −1 to 1 (from the worst possible to a perfect match between the habitat suitability prediction and the occurrence of the species). The suitability index threshold allows the conversion from SI continuous maps into binary maps. Maps are presented in Figure 2 and Figures S2 and S3. The values shown correspond to the SI threshold chosen to optimize TSS values. Values are averaged on the results of 10 SDM runs. SDMs‐NAEU were calibrated on presence data from Western Europe (mainly France, Spain, the United Kingdom, Flanders, Switzerland, and Germany) and presence/absence data from North America; SDMs‐NA were calibrated on P/A data from North America, and SDMs‐EU were calibrated on P/A from Western Europe. Total number of points used and percentage of plots where the species is present are indicated. PVE is the percentage of the variance explained by the models; r is the Pearson coefficient. TSS is the true skill statistics and SI the suitability index coming for the predictions. Se is the sensitivity. Pres., presences; Tot., total of presences + absences.