Distribution and habitat suitability maps for Central European steppe plants

In each habitat suitability map, three model evaluation metrics resulting from 5-fold cross-validation are given: 1) The Area Under the ROC Curve (cvAUC) for assessing the ability of the model to discriminate occurrences in test data from background locations. 2) Overfitting, i.e. the difference between training and test AUC. It is expected to be high for models that overfit the training data. 3) Boyce index (cvBoyce) for measuring how much the model predictions deviate from a random distribution of observed presences across the prediction gradients. All these metrics were averaged over five cross-validation runs. The percentage contribution of each predictor is also shown. The source area maps show the results of generalized linear models that tested whether the simulated accessibility of each climatically suitable area (in addition to environmental factors) contributes significantly to explaining the current species distribution in Europe. We considered "source area" to be any area whose accessibility under the most efficient migration rate had a significant (p < 0.05) positive effect (β > 0) on the current species distribution. For each climatically suitable area, we report the percentage decrease in the residual deviance of the model when accessibility from that area was included (Ddiff). The coefficient estimates (β) and associated p-values are also shown.