Incorporating Data Sets With Multiple Sources of Uncertainty in Integrated Species Distribution Models.

Area of Science:

• Ecology
• Biodiversity Science
• Computational Biology

Background:

• Species distribution models (SDMs) are crucial for conservation but are limited by data uncertainties.
• Key uncertainties include variations in sampling effort and data reliability across different datasets.
• Integrating multiple datasets can improve SDM accuracy but requires careful handling of these uncertainties.

Purpose of the Study:

• To evaluate how different strategies for addressing sampling effort and data reliability uncertainties impact the predictive performance of integrated species distribution models.
• To compare the effectiveness of data filtering, observation modeling, and data integration approaches for handling uncertainty.

Main Methods:

• Modeled distributions of four bird species using three datasets with varying sampling designs.
• Examined strategies including data filtering (spatial thinning, effort-based exclusion), incorporating uncertainty functions in observation models, and varying dataset integration methods.
• Assessed approaches for handling false positive detections and differing data reliability.

Main Results:

• Conservative data filtering, including spatial thinning and excluding high-effort observations, effectively addressed variable sampling effort.
• Treating less reliable data as a covariate, directly modeling false positive rates, and excluding unreliable datasets were effective strategies for handling data reliability.
• The approach of using less reliable data as a covariate significantly sped up model fitting while maintaining good performance.

Conclusions:

• Best practices for integrated modeling involve careful consideration of sampling effort and data reliability.
• Flexible options exist within integrated modeling frameworks to effectively manage diverse data uncertainties.
• These findings offer practical guidance for improving species distribution estimates through robust data integration.