Integrating expert opinion, spatial comparisons, and meta-analyses for the robust prediction of global insect biodiversity change

Mitigating biodiversity loss depends on understanding how anthropogenic drivers reshape ecological communities. Ideally such understanding would come from manipulative experiments alone, but these are insufficient in both time and space to provide a robust picture. Instead, spatial comparisons and meta-analyses are often used to estimate the response of biodiversity to drivers. Where neither of these exist, gaps are supplemented through eliciting expert opinion. Although each of these three evidence types has inherent value, combining them meaningfully is challenging. Given recent reports on global insect declines, building evidence-type ensemble models of insect biodiversity change represents a powerful case-study for demonstrating how evidence types can be combined.

 

Here, we demonstrate how Bayesian regression can be used to integrate meta-analyses, spatial comparisons, and the opinion of experts, to estimate the relationship between fertiliser application rate and site-level insect abundance. Our meta-analytic effect sizes come from a newly published set of data quantifying the effect of a set of anthropogenic threats on insect biodiversity. Our spatial comparisons are drawn from the PREDICTS database, subset for relevant insect groups. Our expert elicitation data is drawn from two sets of workshops eliciting the opinion of experts: first, a set of globally threats ranked by experts as globally important; and second, dose-response curves drawn by experts for the response of specific metrics of biodiversity to a given change in some quantifiable threat. We first describe how each of our evidence types were collected such that we could eventually combine among them, highlighting how such data collection requires teamwork and solving difficult engineering problems. Using Bayesian regression, we then show that the average predictions from both empirical forms of evidence are more similar to each than the opinion of experts. We conclude by suggesting how data should be collected such that dose-response integrations are possible for other taxonomic groups.