A classification-occupancy model based on automatically identified species data.

Area of Science:

• Ecology
• Computational Biology
• Wildlife Conservation

Background:

• Traditional occupancy models estimate species presence but often ignore false-positive detections.
• Advancements in AI-driven species detection generate confidence scores, necessitating new modeling approaches.
• Existing models integrating AI confidence scores require rigorous validation.

Purpose of the Study:

• To propose and evaluate a novel occupancy model specifically designed for AI-detected species data.
• To assess the model's performance in handling both false-negative and false-positive detection errors.
• To improve species-habitat relationship inference and automated biodiversity monitoring.

Main Methods:

• Developed a new occupancy model utilizing AI-detected species data and confidence scores.
• Conducted simulation studies to evaluate inferential and predictive accuracy against known parameters.
• Applied goodness-of-fit tests and external data evaluation to a case study using AI-detected species data.

Main Results:

• The proposed model generally outperformed alternatives in accuracy during simulations and goodness-of-fit tests.
• The model effectively accounted for both false-negative and false-positive detection errors.
• Performance in discrimination metrics based on external data was not superior to alternatives.

Conclusions:

• The novel occupancy model offers a robust approach for analyzing AI-detected species data.
• This model enhances the understanding of species-habitat relationships in automated monitoring.
• It provides a valuable tool for developing more accurate biodiversity monitoring workflows.