A classification-occupancy model based on automatically identified species data

Occupancy models estimate a species' occupancy probability while accounting for imperfect detection, but often overlook the issue of false-positive detections. This problem of false positives has gained attention recently with the rapid advancement of automated species detection tools using artificial intelligence (AI), which generate continuous confidence scores for each species detection. Novel occupancy models have been introduced that integrate these confidence scores to identify false positives, but these models require thorough assessments of diagnosis and validation. Here, we propose a new occupancy model based solely on AI-detected species data. We conducted simulations to examine the inferential and predictive accuracies with known true parameters and analyzed AI-detected species data to test the practical usefulness through goodness-of-fit tests and evaluation with external data. Our proposed model mostly outperformed alternative models that ignore imperfect detection or false-positive error probabilities in terms of accuracy in simulation analyses and goodness-of-fit tests in the case study, but not in terms of discrimination metrics based on external data. The proposed occupancy model aids in understanding species-habitat relationships and developing automated biodiversity monitoring workflows by accounting for both false-negative and false-positive errors.