Wrong, but useful: regional species distribution models may not be improved by range-wide data under biased sampling

Area of Science:

• Ecology
• Conservation Biology
• Biodiversity Informatics

Background:

• Species distribution modeling (SDM) is crucial for ecology and conservation.
• National datasets for SDMs are often limited by environmental gradients, biased, and incomplete, questioning model quality.
• Evaluating the adequacy of national presence-only data for regional SDMs is essential.

Purpose of the Study:

• To assess the suitability of national presence-only data for calibrating regional species distribution models (SDMs).
• To compare the performance of regional SDMs trained solely on national data versus those incorporating global data.
• To investigate the impact of using global model predictions as 'priors' and correcting for sampling bias on regional SDM accuracy.

Main Methods:

• Trained SDMs for Egyptian bat species using Maxent and elastic net algorithms at national (Egypt) and global scales.
• Measured congruence between global and regional model predictions within Egypt.
• Incorporated global model predictions as priors and applied sampling bias correction to regional models.
• Quantified model improvement using AUC and congruence metrics.

Main Results:

• Predictions from global and regional models for Egypt showed only weak concurrence on average.
• Using global model predictions as priors did not substantially improve regional model performance (AUC and congruence).
• Sampling bias correction enhanced model performance significantly, reducing the perceived benefit of using priors.

Conclusions:

• National presence-only data in Egypt are of questionable adequacy for robust regional SDM calibration.
• Global data integration did not improve regional model performance under biased and incomplete sampling conditions.
• Future efforts should focus on acquiring bias-free regional data and utilizing global models to guide targeted surveys in data-poor regions.