Functional ecology and imperfect detection of species

Functional traits are increasingly being used to understand the response of species to environmental change and their effects on ecosystem functioning. However, some ecologically important traits, such as plant height, influence the probability of species detection during field surveys. Imperfect detection of species could therefore bias measures of functional trait composition and diversity, leading to incorrect estimates of trait–environment relationships due to a process of “detection filtering.” The importance of detection filtering for functional ecological studies remains unknown. We used hierarchical models that account for detection filtering to analyse data on 1,296 vascular plant species sampled in 362 1-km2 plots, distributed along a 2,460-m elevational gradient in Central Europe. We examined how detection filtering altered measures of functional diversity (multivariate functional richness and packing) and composition (community means of three traits). We also determined whether the strength of detection filtering varied over the gradient, to determine whether detection filtering biased trait–environment relationships. Species detectability was correlated with all three functional traits tested in this study, meaning that short species with small seeds and high specific leaf area values were less likely to be detected. This suggests that imperfect detection has the potential to bias measures of functional composition. Generally, measures of functional composition were not strongly affected by detection filtering, but functional packing was underestimated when detection filtering was not accounted for. In addition to the traits, distributional characteristics were important; rare species and species occurring mainly at low elevations tended to have lower detection probabilities. Overall, detection filtering did not strongly bias trait–environment relationships because the effects of the environment on functional composition and diversity were larger than the effects of detection. Our results suggest that many measures of functional composition and diversity are robust to detection filtering, but some are likely biased. Functional ecologists should consider correcting for imperfect detection, and our approach provides a simple method to do so for a wide range of datasets