Ecological and modelling opportunities of semi-structured insect monitoring

Insects are vastly under-represented in biodiversity monitoring, but different forms of heterogeneous data have indicated long-term declines. Unstructured citizen science monitoring, generating occurrence records collected without a common protocol, have been particularly valuable for deriving large-scale trends at the species-level. Such species-level trend estimates are essential for defining conservation priorities and understanding implications of declines for ecosystems. However, unstructured monitoring data contain multiple challenges related to potential confounding effects of sampling variability. New monitoring approaches are needed to deliver the robust insect data needed to reverse insect loss and maintain associated ecosystem services. Semi-structured monitoring, relying on protocol reporting than protocol standardization, has emerged as a promising and more flexible alternative to structured monitoring, retaining broad participation of citizen scientists. While semi-structured monitoring has already been highly successful for birds, via apps such as eBird, its potential application to insects is less clear. In this project, we examined the potential opportunities and barriers for semi-structured monitoring of insects, using dragonfly monitoring in the UK as a case study. We explored the recording practises of citizen scientists and the modelling performance of semi-structured data. We found that many insect recorders already follow attributes of structured surveying, reporting complete species lists, collecting abundance data and revisiting the same sites, but existing platforms are not sufficiently retaining all the information to make use of the structure. We also found that most recorders are willing to report more comprehensive metadata, such as survey effort, but only a subset would follow a fully standardised protocol. In our modelling experiments, we contrasted different approaches to making use of available semi-structured data streams, either by filtering data to those of the highest quality, or by building integrated models that combines the information within different data streams. Results suggest that a bias-variance trade-off is faced when making such analytical decisions. Overall, our findings highlight the overlooked structure within citizen science monitoring data that could be leveraged for more robust analyses of insect biodiversity change.