Integrating structured and unstructured citizen science data to improve wildlife population monitoring

Accurate and robust population trend assessments are key to successful biodiversity conservation. Citizen science surveys have provided good evidence of biodiversity declines whilst engaging people with them. Citizen scientists are also collecting opportunistic biodiversity records at unprecedented scales, vastly outnumbering records gathered through structured surveys. Opportunistic records exhibit spatio-temporal biases and heterogeneity in observer effort and skill, but their quantity offers a rich source of information. Data integration, the combination of multiple information sources in a common analytical framework, can potentially improve inferences about populations compared to analysing either in isolation. We combine count data from a structured citizen science survey and detection-nondetection data from an opportunistic citizen science programme. Population trends were modelled using dynamic N-mixture models to integrate both data sources. We applied this approach to two different inferential challenges arising from sparse data: (i) the estimation of population trends for an area smaller than a structured survey stratum, and (ii) the estimation of national population trends for a rare but widespread species. In both cases, data integration yielded population trajectories similar to those estimated from structured survey data alone but had higher precision when the density of opportunistic records was high. In some cases this allowed inferences about population trends where indices derived from single data sources were too uncertain to assess change. However, there were differences in the trend magnitude between the integrated and the standard survey model. We show that data integration of large-scale structured and unstructured data is feasible and offers potential to improve national and regional wildlife trend estimates, although a need to independently validate trends remains. Smaller gains are achieved in areas where uptake of opportunistic recording is low. The integration of opportunistic records from volunteer-selected locations alone may therefore not adequately address monitoring gaps for management and policy applications. To achieve the latter, scheme organisers should consider providing incentives for achieving representative coverage of target areas in both structured and unstructured recording schemes.