Accuracy, error and bias in species identification

Reliable, robust data is fundamental to effective decision-making. Species observations are used as evidence in a range of areas that work towards conserving biodiversity. Decisions made on these data are only well informed if the species have been accurately identified. Moreover, the misidentification of species can have widespread socio-economic impacts. Despite these important applications of species data, the possibility of accuracy, error, and bias in species identification remains largely unexplored. Both volunteers and professionals conduct species identification, and in its simplest form, this process is a judgement made by reference to identification aids, or from prior knowledge. This thesis aims to fill an essential knowledge gap by investigating accuracy in species identification between individuals, across levels of expertise, and the levels of agreement between individuals with similar experience. Applying methods from forensic face recognition research, individuals with varying levels of expertise, and interest in biodiversity, participated in a series of simple image-based tasks. These tasks involved online, pairwise matching tasks under optimised conditions, and sorting tasks with images downloaded from Internet sources. This study shows that decisions on species identification are highly variable between individuals, and high levels of accuracy are achievable by experts and non-experts. Moreover, experience is no guarantee of accuracy, and inter-specific disparity does not always exceed intra-specific variation. There is a need for a simple, principled method for assessing identification accuracy, which can be performed by experts and non-experts alike. This method also needs to be sensitive enough to capture individual differences. Improvements in technology have led to an increase in data being collected from previously inaccessible areas, and citizen science has widened participation. However, as data collection adapts to incorporate changes in how species observations are collected and by whom, methods for assessing and evaluating the reliability of those data must evolve.