Evaluating collinearity effects on species distribution models: An approach based on virtual species simulation

Area of Science:

• Ecology
• Computational Biology
• Environmental Science

Background:

• Species distribution modeling (SDM) is widely used but faces concerns about inaccuracies and misuse.
• Collinearity among environmental predictors is a known source of model uncertainty, yet lacks detailed recent evaluation.

Purpose of the Study:

• To evaluate the impact of collinearity on SDM accuracy and uncertainty.
• To compare models using PCA-derived variables versus original correlated predictors.
• To assess how modeling algorithms and species data characteristics influence sensitivity to collinearity.

Main Methods:

• Employed a virtual species approach for controlled experimentation.
• Compared SDMs built with PCA-derived variables against those using highly correlated original climate variables.
• Investigated the influence of different modeling algorithms and species data traits on collinearity effects.

Main Results:

• Collinearity among predictors demonstrably decreases SDM efficiency and heightens model uncertainty.
• More complex algorithms (e.g., Maxent) showed better performance and potentially utilized collinearity effectively.
• Species characteristics, particularly prevalence, influenced model sensitivity to collinearity, impacting efficiency.

Conclusions:

• PCA-derived variables are recommended to mitigate collinearity's negative impacts and improve objective variable selection.
• Even less accurate cell-level predictions can yield valuable insights into macroecological traits like range-size.
• Understanding collinearity's effects is crucial for reliable SDM application in ecological research.