Collinearity in ecological niche modeling: Confusions and challenges

Area of Science:

• Ecology
• Biogeography
• Computational Biology

Background:

• Ecological niche models (ENMs) are crucial tools in ecology and biogeography.
• Maxent is a widely adopted software for ENM, yet predictor collinearity treatment remains debated.
• Quantitative studies on collinearity's impact, especially in model transfer, are scarce.

Purpose of the Study:

• To quantify the effects of predictor collinearity on Maxent model performance.
• To investigate collinearity shifts and environmental novelty impacts during model transfer.
• To provide recommendations for improving Maxent model transferability.

Main Methods:

• Maxent modeling was employed to assess predictor collinearity effects.
• Scenarios included varying predictor collinearity during training and projection.
• Model performance was evaluated under collinearity shifts and environmental novelty.

Main Results:

• Maxent models showed robustness to predictor collinearity during the training phase.
• Excluding highly correlated variables had minimal impact on model performance.
• Collinearity shifts and environmental novelty significantly reduced model transfer performance.

Conclusions:

• Maxent is resilient to predictor collinearity within the training dataset.
• Maxent inherently handles redundant variables, making exclusion strategies less critical.
• Collinearity shifts and environmental novelty pose significant challenges to Maxent model transferability, necessitating their evaluation.