Where did they not go? Considerations for generating pseudo-absences for telemetry-based habitat models

Area of Science:

• Ecology
• Conservation Biology
• Spatial Modeling

Background:

• Habitat suitability models are crucial for understanding species distributions and informing conservation efforts.
• Telemetry data is increasingly used in these models, often requiring the generation of pseudo-absences to simulate un-occupied areas.
• The method of generating pseudo-absences can significantly influence model outcomes.

Purpose of the Study:

• To investigate how different pseudo-absence generation methods affect the performance of telemetry-based habitat models.
• To explore these effects across diverse species movement strategies, modeling approaches, and environmental contexts.

Main Methods:

• Developed habitat models for blue whales and African elephants using telemetry data.
• Tested four pseudo-absence generation techniques: background sampling, buffer sampling, correlated random walks, and reverse correlated random walks.
• Employed generalized linear mixed models, generalized additive mixed models, and boosted regression trees for habitat modeling.

Main Results:

• Environmental niche space separation between presence and pseudo-absence data was the primary driver of model performance.
• The optimal pseudo-absence method varied by species (background sampling for blue whales, reverse correlated random walks for elephants).
• Models with greater environmental separation showed improved predictive skill but not always biologically realistic spatial predictions.

Conclusions:

• Habitat model performance can be positively biased if pseudo-absences are sampled from environmentally dissimilar areas.
• Careful consideration of sampling domain extent and environmental heterogeneity is vital for pseudo-absence generation.
• Scrutinizing spatial predictions is essential to ensure habitat models are biologically realistic and meet management objectives.