Accommodating species identification errors in transect surveys.

Area of Science:

• Ecology
• Wildlife population estimation
• Statistical modeling

Background:

• Transect surveys are crucial for estimating animal population density and abundance.
• Species misclassification errors are common in ecological surveys but often unaddressed by statistical methods.
• Ignoring misidentification can lead to significant biases in population estimates.

Purpose of the Study:

• To examine biases caused by uncorrected species misidentification in ecological surveys.
• To develop statistical models for unbiased density estimation in the presence of classification errors.
• To provide a framework for integrating misclassification data into population estimation models.

Main Methods:

• Treating true species identity as a latent variable.
• Utilizing auxiliary information on the misclassification process (e.g., informative priors, known-species experiments, double-observer protocols).
• Applying a model-based distance-sampling framework integrated with misclassification modeling.

Main Results:

• Simulated data showed reliable density estimation when misclassification rates were informed by experimental data.
• Double-observer protocols yielded robust inference, especially with "unknown" observations or symmetric misclassification.
• The models provided reasonable seal densities despite identification imprecision, with improved reliability when accounting for spatial variation.

Conclusions:

• Ecologists should consider spatially explicit models to address species misidentification and distribution differences.
• Double-observer sampling protocols are recommended to mitigate species misclassification, particularly by recording uncertain observations as "unknown."
• The developed statistical approach offers a robust method for obtaining unbiased population density estimates in ecological surveys.