Imperfect detection biases occupancy estimates in plants: A review and case study.

The distributions of rare species are often estimated from presence-absence data, but undetected occurrences (imperfect detection) can lead to consequential underestimates of occupancy. Despite its standard use in animal ecology, detection probability (p) is rarely incorporated into plant occupancy estimates. Plants with cryptic life stages or traits (e.g., seed banks, dormancy) may be especially prone to periods of unobservability, resulting in imperfect detection even when a species is present. We tested the hypothesis that cryptic traits lower detection by reviewing published estimates of p in plant systems. While taxa with cryptic traits, particularly annual plants with seed banks and deciduous geophytes, exhibited lower p, there was wide variation across genera and publications. We then conducted a case study of the rare, annual, aquatic plant, Gratiola amphiantha (snorkelwort), endemic to ephemerally rain-filled granite rock outcrop pools, to estimate occupancy (Ψ), p, and their environmental drivers. We found that Ψ increased with deeper soils and larger pool areas. Snorkelwort exhibited a high but imperfect p (mean p = 0.87) that varied with competing vegetation cover and census date. Modeling Ψ across varying levels of survey effort showed that ignoring imperfect detection can underestimate total occupancy estimates by up to 43%; however, our empirical p estimates indicate a minimum of just two surveys can reach 95% detection for snorkelwort. Our findings demonstrate that imperfect detection is an important and underappreciated limitation in plant occupancy studies, even for species with high detection probability. For rare plants, explicitly incorporating taxon-specific p estimates and marginally increasing survey effort can substantially improve the accuracy of occupancy models.