We expect trends in insect diversity and abundance to be complex due to spatiotemporal variation in the intensity of anthropogenic pressures and because species’ responses to anthropogenic threats vary among and within taxonomic and functional groups. In order to develop a predictive synthesis, it is vital for this variation to be quantified using all available evidence sources and considering not just high profile taxa (e.g. butterflies and bees) but also poorly studied groups (e.g. soil fauna).
We will prioritise three sorts of comparative analysis for the TRM: (a) the dose-response relationship between the intensity of a threat and biodiversity, (b) the relative impacts of different threats on a group, and (c) the relative impacts of a given threat and intensity on different groups. In T2.1 we will construct a database capable of storing diverse evidence streams gathered in T2.2-2.4. We see T2.2-2.4 as delivering complementary (and potentially contradictory) lines of evidence: each task is expected to deliver a series of papers. In T2.5 we will synthesise the available evidence into an explicit threat-response model (TRM) - technically a Bayesian belief network - resulting in a high-profile synthesis paper.
Task 2.1: Construct a flexible database to hold the underlying evidence. For each group of species, we need to be able to capture information about (a) response to each threat separately, for a range of biodiversity metrics; (b) variation in response among species and/or spatially; (c) direction and, if possible, strength of interactions between threats; (d) direct consequences for ecosystem functioning and ecosystem service provision of changes in diversity and abundance (asset-benefit relationships); (e) predictable dependencies and/or consequences changes in the focal group on other groups; (f) responses of ecological networks. Flexibility is essential because the evidence will vary in precision and robustness, and in geographic and taxonomic. Indeed, subdivisions of taxonomic groups (e.g. by ecosystem function, figure 2) will not all be known in advance and some insects have multiple functions. The database will need to capture all the quantitative information required for meta-analysis, and to capture estimates of response variability as well as the mean. It will be designed to permit queries such as “select all quantitative data on threat-response relationships for pollinating insects in Africa”, as well as summaries such as “what is the relationship between invasive species and beetles in Australia?” Its functionality will include automated quantitative meta-analysis and other rule-based syntheses, and highlighting in real-time the greatest gaps where additional information is most needed. Note this task is limited to the design, implementation and testing of the database: the database will be populated in T2.2-2.4.
Task 2.2: Statistical analyses to populate cells where possible. Effects of many threats (those relating to land use, transport corridors, logging, disturbance and climate change) are estimable from data that are already largely available. The PREDICTS database was designed to quantify biodiversity responses to all the above threats apart from climate change, and already holds data on over 20,000 insect species. It has already been used to explore regional and trait-based variation in how these threats and their interactions affect bees. Targeted filling of geographic and taxonomic gaps will allow similar analysis for a wide range of insect groups in most regions (pooling data across regions or groups whose models do not differ significantly), with functionally important taxa - such as pest controllers, decomposers and disease carriers - is a priority. For climate change, species distribution models will be fitted to a sample of species within each group (representing the range of functional diversity) using cleaned available occurrence records and used to quantify the threat-response relationship between diversity and radiative forcing or fire frequency. We will also explore the importance of interactions between drivers, focussing on land use and climate change and space-for-time models have already been shown to be able to capture the effects of these interactions, including for insects. We will devote special attention to analyses of trophically-structured datasets, to test for a) covariance between the trends and responses of different functional groups (e.g. detritivores, herbivores, pollinators and predators), b) variation in the response-effect trait relationship and c) trends in functional redundancy, all of which are key determinants of resilience.
Task 2.3: Meta-analyses to populate the model. Meta-analysis is increasingly used as a powerful way to synthesise information into quantitative statements about threat-response relationships and many recent examples are relevant to the TRM. Threat-response relationships are often context-dependent so we will, where possible, use the evidence base from relevant meta-analyses rather than their synthetic results. We will encode geographic, taxonomic and functional meta-data using the categories in our database, thus enabling our syntheses to reflect the geographic and taxonomic variation in the weight of evidence. Results from additional published studies will be added to the database to where possible, fill taxonomic, geographic and threat gaps, using targeted literature searches assisted by machine learning.
Task 2.4: Survey taxon- and threat-specific experts to populate all cells in the model. We do not expect that T2.2-2.3 will provide estimates of all threat responses - and especially interactions among threats or with other species groups - for all groups on all continents. We, therefore, propose to involve relevant experts from around the world ensuring representation from each continent, in a structured expert-elicitation approach to estimate each cell in the model. Structured expert-elicitation approaches are increasingly used to support decision-making in conservation and to fill gaps in published scientific knowledge using expert judgement. Two stages are envisaged, building on our previous experience (see Track Record): 1) structured gathering of information using online surveys of identified experts with taxon- and threat-specific knowledge, following a formal anonymised, iterative Delphi process; 2) workshops to achieve consensus on the outcomes with representation from all continents and major biomes. Confidence in the information will be documented throughout. Experts for the Delphi process will be identified via international networks of entomological societies, via literature searches and through snowball sampling. Surveys and comments will be translated into up to 12 languages as required. This process will allow us to predict the causes of insect decline for poorly studied taxa and regions linking into the IUCN threat classification.
Task 2.5: Synthesis: the creation of a Bayesian belief network. The lines of evidence gathered in T2.2-2.4 all indirectly estimate the threat-response relationships and, though each has strengths and weaknesses none can deliver the robust inference that one might get from adequately replicated large-scale long-term BACI experimental designs. Within the context of the TRM, we can describe the evidence from T2.2-T2.4 as a embodying our prior beliefs about threat response relationships in insects. In general terms, this network estimates the probability distribution for the trend in insect biodiversity in a location, conditional on how it is measured, the geographic region being considered, the nature and intensity of the threats it was exposed to, the taxonomic group to which it belongs, and the type of evidence that has been brought to bear:
P(Trend | Metric, Region, Threats & intensity, Taxon, Location, EvidenceSource)
The degree to which our evidence base fills the cells figure 2 will form the basis of a synthesis overview highlighting current evidence gaps.
