Misidentifications in citizen science bias the phenological estimates of two hard‐to‐identify Elaenia flycatchers

Citizen science initiatives contain a large volume of observations that can be useful to address ecological questions for a wide array of organisms. However, one limitation of citizen science data is the potential for species misidentification. Although recent studies have shown that citizen science data are relatively accurate for many taxa, the effect of misidentification errors in hard‐to‐identify species remains poorly explored. If misidentification events occur at large scales, ecological estimates can be compromised. Here, we show that misidentifications contained in citizen science databases biased phenological estimates in a pair of migratory and partially overlapping Neotropical flycatchers: the Chilean Elaenia Elaenia chilensis and the Small‐billed Elaenia Elaenia parvirostris. We reviewed and re‐classified 4399 photos of these species from Argentina, Chile and Uruguay. We found that overall identification accuracy was high (c. 90%) for both species when they were allopatric, but dramatically low for Small‐billed Elaenia during autumn migration (from 28.6% to 84.6%) because migrating individuals of Chilean Elaenia were systematically reported as Small‐billed. The phenological estimates for both Elaenias were biased due to the large number of misidentifications concentrated towards the autumn migration period. These errors caused a 1‐week advancement in the estimated arrival, and a 2‐week delay in the estimated departure for Small‐billed Elaenia. For Chilean Elaenia, errors caused a 1‐week delay for the estimated spring peak passage and underestimation of the magnitude of the autumn passage. Our results highlight the importance of performing critical assessments of records when using citizen science databases to describe ecological patterns in species that are hard to identify. The large volume of information provided by citizen science initiatives is useful to describe spatio‐temporal patterns in birds, particularly in those of poorly known regions. However, to further enhance the usefulness of such databases, it is imperative actively to post‐process and contrast patterns derived from documented (and undocumented) records, with a special focus on misidentifications. This will only be possible through a thorough review of the documented data, together with an intimate understanding of the natural history of the study species.