Figure 1.
Examples of the three approaches to construction and validation of climate envelope models.
All models are calibrated with occurrence data and climate conditions from 1967–1971 (left-hand panels), and validated with occurrence data from 1998–2002 (right hand panels). Black circles indicate species presence and white dotes indicate species absence. The prediction map against which occurrences are validated using the dynamic approach represent the climate envelope under 1998–2002 conditions and for the static approach the prediction map represents the 1967–1971 climate envelope. Under the hybrid approach, areas of range expansion between 1967–1971 and 1998–2002 are merged with the 1967–1971 climate envelope to create a third prediction map.
Figure 2.
Example map illustrating a ‘hybrid’ approach to climate envelope model construction.
Areas indicated in black are included in the 1998–2002 projection, but were not part of the initial climate envelope. These areas range expansion between 1967–1971 and 1998–2002 are merged with the initial 1967–1971 climate envelope to create a hybrid prediction map.
Table 1.
Summary statistics for climate envelope models for twelve species of resident North American breeding birds.
Figure 3.
Box plots illustrating differences in climate envelope model sensitivity between models constructed with the maximum entropy and random forest algorithms (A), differences in sensitivity between models using three approaches that differ in the way they treat range expansion and contraction (B), differences in specificity between algorithms (C) and differences in specificity among approaches (D).
Figure 4.
Relationships between differences in model sensitivity as a function of the proportion of presences in 1998–2002 that occurred in areas of range expansion (A and C for random forest and maximum entropy models, respectively) and differences in model specificity as a function of the proportion of absences in 1998–2002 that occurred in areas of range expansion (B and D for random forest and maximum entropy models, respectively).
Figure 5.
The improvement in sensitivity using a hybrid approach to climate envelope model evaluation that assumed no range contraction relative to a dynamic model in which ranges expanded and contracted in response to climate change was greatest for species that have been reported from relatively more habitat types (A), whereas differences in specificity between the dynamic approach and a static approach assuming no climate change were not significantly associated with body size (B).
Table 2.
Body size and habitat niche breadth for 12 species of resident North American breeding birds.