Inferring species richness using multispecies occupancy modeling: Estimation performance and interpretation.

Area of Science:

• Ecology
• Statistical Modeling
• Conservation Biology

Background:

• Multispecies occupancy models estimate species richness using detection/non-detection data, accounting for imperfect detection.
• Data augmentation extends these models to infer total community size, including undetected species.

Purpose of the Study:

• To investigate the robustness of total species richness estimates derived from multispecies occupancy models.
• To evaluate model sensitivity to assumptions, priors, and data characteristics through simulations and real-world application.

Main Methods:

• Reviewed key assumptions of multispecies occupancy models with data augmentation.
• Conducted simulations across various species traits and sampling regimes, testing assumption violations and Bayesian prior sensitivity.
• Applied the model to a real dataset, comparing estimates with and without predictors and with different data subsets.

Main Results:

• Total species richness estimation can be poor when a high proportion of species (>15%-20%) are missed, even when model assumptions are met.
• Commonly used Bayesian priors can exacerbate overestimation; while models may tolerate some assumption violations, lower-tail deviations cause bias.
• Omitting relevant predictors in real-data analyses led to significant underestimation of species richness.

Conclusions:

• Estimates of total species richness are inherently sensitive to model structure and uncertain, especially when many species are undetected.
• Careful selection of priors, rigorous assumption testing, and model refinement are crucial for improving estimator performance.
• Realistic expectations regarding accuracy are necessary; maximizing survey effort to minimize undetected species is vital for critical management decisions.