Answers are grounded in this paper and cite their sources.
Compares extrapolation strategies in ecological niche models to improve predictions in non-analog environments.
Ecological niche models (ENMs) are commonly used to predict species' potential distributions across space and time, allowing researchers and practitioners to assess invasion risk, estimate the impacts of future climate change, and reconstruct past distributions. These predictions, however, often require extrapolation into environmental conditions that fall outside the range of values used for model training (i.e. ‘non‐analog environments'). Two primary strategies exist for extrapolating into new conditions for some modeling algorithms: unconstrained extrapolation (allowing the modeled response curve to maintain its trajectory) and clamping (holding the response constant at the values observed at the limits of the training data). As these can lead to considerable differences in model predictions, it is important to inspect the behavior of modeled responses under different extrapolation strategies, which is rarely done in ENM studies. Such examination can help promote biologically realistic model transfers, for example by avoiding unconstrained extrapolation when the relevant tail of the modeled response has a sharply increasing trend. Here, we present a proposed workflow and new visualization tools to guide researchers in choosing appropriate extrapolation strategies, including plots for marginal response curves and non‐analog pixel densities. As an illustration, we used this workflow and tools to compare extrapolation strategies for two hindcast scenarios at the Last Glacial Maximum (~ 21 kya) for the black‐eared mouse Peromyscus melanotis in Mexico, which currently occurs in montane forests. Making extrapolation choices individually for each tail per variable (‘custom clamping') helped avoid unrealistically high suitability predictions for non‐analog environments. This study emphasizes the importance of both inspecting response curves and assessing the total area with non‐analog conditions. By making these choices by variable and tail – a first for ENMs – this study and associated code empower researchers to understand how individual extrapolation strategies affect results and to make decisions that should lead to more realistic predictions.
No takes yet. Share an insight, caveat, or question.
Pinilla‐Buitrago et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: