Identifying species at coextinction risk when detection is imperfect: Model evaluation and case study

Area of Science:

• Ecology
• Conservation Biology
• Network Analysis

Background:

• Species loss can lead to cascading extinctions (coextinctions).
• Estimating coextinction risk often relies on host-breadth derived from observational interaction data.
• Observational data is prone to imperfect detection, compromising coextinction risk assessments.

Purpose of the Study:

• To develop a robust method for estimating coextinction risk by accounting for data uncertainty.
• To predict the number of interaction partners for each species and assess community sampling completeness.
• To evaluate the influence of sampling effort, interaction probability, and abundance on model accuracy.

Main Methods:

• Fitting a hierarchical N-mixture model to individual-level interaction data.
• Utilizing simulated interaction data to test model performance under varying conditions.
• Applying the model to an empirical dataset of plant-insect mutualistic interactions.

Main Results:

• The model accurately predicted interaction partners in high-abundance scenarios but showed high uncertainty with rare species.
• Model predictions were generally closer to true values than direct observations.
• Empirical data suggested only 14-59% of insect visitor species were detected, indicating inadequate sampling in common pollinator study designs.

Conclusions:

• Imperfect detection significantly biases inferences from interaction networks, potentially misrepresenting host specificity and coextinction risks.
• Hierarchical models offer a way to estimate coextinction risk more reliably.
• Improved data collection, including intensified sampling and individual-level observations, is crucial for reducing uncertainty in ecological network analyses.