Nocturnal at Noon? Addressing False Positives in Activity Pattern Data.

Area of Science:

• Ecology
• Computational Biology
• Wildlife Monitoring

Background:

• False positives in wildlife detection datasets can significantly bias ecological inference, particularly diel (24-hour) activity patterns.
• The increasing reliance on automated classifiers for camera-trap images and bioacoustic recordings exacerbates this issue, as full human verification is often impractical.

Purpose of the Study:

• To develop a Bayesian framework that accounts for false positives in wildlife detection data.
• To provide a more accurate alternative to standard pooled kernel density estimation for analyzing diel activity patterns.

Main Methods:

• A Bayesian framework was developed to estimate the dataset-average true-positive rate (θ) and time-of-day distributions for true detections and false positives.
• Two error structures were implemented: a uniform false-positive model and a flexible skewed model to capture temporally clustered misclassifications.
• The model was validated using simulations and an empirical dataset.

Main Results:

• The proposed Bayesian approach successfully recovered true diel activity patterns even with limited human verification.
• The method demonstrated a reduction in bias compared to uncorrected estimates.
• Applying the model to an empirical dataset yielded different activity inferences than standard pooled methods.

Conclusions:

• This false-positive-aware Bayesian framework offers a robust method for estimating diel activity patterns from automated wildlife detection data.
• The approach can be extended to incorporate covariates and random effects, and integrated with occupancy and abundance models.
• Accurate estimation of wildlife activity patterns is crucial for reliable ecological inference.