Predicting detection probabilities to estimate species' population sizes

1. Information on species' population sizes is essential for setting conservation priorities, but it is notoriously difficult to obtain and usually involves intense monitoring efforts. A core step towards estimating a species' population size is first estimating its detection probability. Here, we test the value of possible shortcuts in deriving species' detection probabilities and the implications of using these shortcuts for population size estimates. 2. Using a distance‐sampled dataset for breeding birds in Denmark, we tested the ability of species' traits to predict their detection probabilities and contrasted the population size error from using trait‐based estimates versus direct estimates of detection probabilities. We also assessed the sample size requirements for direct estimates of detection probabilities to surpass the performance of trait‐based estimates. 3. Species' detection probabilities were highly predictable—across species and across years. Across species, detection probabilities were predicted by a combination of morphological traits and ecological niche metrics. Body size alone explained 40% of the variation among species. Phylogeny was unimportant once key traits were taken into account. 4. Error in population size estimates from using a trait‐based probability rather than a direct estimate of detection probability was within the 95% confidence intervals of the population size estimates for most species. However, the error in population size was lower using a previous direct estimate of species detection probability (i.e. using data from the previous year) than using a trait‐based estimate. Moreover, direct estimates outperformed trait‐based estimates even when only a relatively small number (median of 40) of distance‐sampled observations was available. 5. Practical implication : Our findings indicate that information on species' detection probability can be borrowed across species, using traits and across years. This raises the potential of mixed study designs that collect detection probability information in only some sampling units (sites, years and even species), reducing the demands for intensive monitoring and enhancing the scalability of approaches for estimating species' population sizes.