Mixture models for distance sampling detection functions.

Area of Science:

• Ecology
• Wildlife Biology
• Statistical Modeling

Background:

• Distance sampling is a key method for wildlife population estimation.
• Current models, like key function plus series adjustment (K+A), have limitations in ensuring detection function assumptions.
• There is a need for more robust and flexible detection function models.

Purpose of the Study:

• To introduce a novel class of finite mixture models for distance sampling detection functions.
• To compare the performance of these mixture models against the traditional K+A approach.
• To assess the applicability of mixture models in challenging survey scenarios.

Main Methods:

• Development of finite mixture models using parametric key functions (e.g., half-normal).
• Comparison with K+A models through simulations under various conditions.
• Re-analysis of four problematic real-world distance sampling case studies.

Main Results:

• Mixture models demonstrate superior performance over K+A methods in many simulations.
• Mixture models are particularly effective for 'spiked' line transect data and larger sample sizes.
• The new models inherently satisfy distance sampling assumptions without requiring constrained optimization.

Conclusions:

• Finite mixture models represent a promising advancement for distance sampling detection functions.
• These models offer greater flexibility and robustness compared to existing methods.
• It is recommended to include mixture models in standard model selection procedures for distance sampling analysis.