Correcting for measurement errors in a long-term aerial survey with auxiliary photographic data

Long-term, large-scale monitoring of wildlife populations is an integral part of conservation research and management. However, some traditional monitoring protocols lack the information needed to account for sources of measurement error in data analyses. Ignoring measurement error, such as partial availability, imperfect detection, and species misidentification, can lead to mischaracterizations of population states and processes. Accounting for measurement error is key to robust monitoring of populations, which can inform a wide variety of decisions, including harvest, habitat restoration, and determination of the legal status of species. We undertook an effort to retroactively minimize bias in a large-scale, long-term monitoring program for marine birds in the Salish Sea, Washington, USA, by conducting an auxiliary study to jointly estimate components of measurement error. We built a novel model in a Bayesian framework that simultaneously harnessed human observer and photographic data types to produce estimates necessary to correct for the effects of partial availability, imperfect detection, and species misidentification. Across all 31 species identified in photographs, both observers had instances of undercounting and overcounting birds but tended to undercount (observers undercounted totals across all species on 69.3%–78.9% of transects). We estimated species-specific correction factors that can be used to correct both historical and future counts from the Salish Sea survey, which has been running since 1992. Our novel modeling framework can be applied in other multispecies monitoring contexts where minimal photographic data can be collected for the purposes of correcting for measurement error in large-scale, long-term datasets.