Recognizability bias in citizen science photographs

Citizen science initiatives and automated collection methods increasingly depend on image recognition in order to provide the amounts of observational data research and management needs. Training recognition models, meanwhile, also requires large amounts of data from these sources, creating a feedback loop between the methods and the tools. Species that are harder to recognize, both for humans and machine learning algorithms, are likely to be underreported, and thus be less prevalent in the training data. As a result, the feedback loop may hamper training mostly for species that already pose the greatest challenge. In this study, we trained recognition models for various taxa, and found evidence for a "recognizability bias", where species that models struggle with are also generally underreported. This has implications for the kind of performance one can expect from future models that are trained with more data, including such challenging species. We consider identification methods that rely on more than photographs alone to be important in improving future identification tools.