We will confront “prior knowledge” about insect decline (T2.5) with time-series data (T1.1). Importantly, the data contributing to WP2 are largely independent of the data in T1.1. This will provide a basis for defining uncertainty not just for those taxa and regions for which we have high quality data but, by establishing the ecological forecast horizon, for the wider set of insects across diverse biomes and geographical regions (T3.3), for understanding ecosystem consequences (T4.1) and making future projections (WP4).
Task 3.1: Validation: does the threat-response model predict observed trends? We will confront the threat-response relationships (from WP2) with the time-series data on insect biodiversity and threats (from WP1), treating the probabilistic statement of T2.5 as a set of “informative priors”. This exercise is not a straightforward validation, because of the multiple lines of evidence contributing to the TRM, the multitude of biodiversity metrics, and complications due to spatial scale and taxonomic mismatch or imprecision. Rather, we propose a multidimensional optimisation procedure in order to identify evidence weights for the data collated in WP2, and to identify which aspects of insect biodiversity change are (not) predictable without long-term time-series data. We will produce a series of publications focussed on specific regions (e.g. UK), taxa (e.g. beetles), and for specific research questions:
a) Which lines of evidence best predict trends? Our TRM comprises evidence from statistical models, experiments, meta-analysis and expert judgement. We will test and compare the ability of each line of evidence to predict recent trends, both independently and as a weighted ensemble.
b) How does predictability decay with spatial, environmental and phylogenetic distance? Many of the time-series in WP1 will align to multiple lines of evidence in WP2 that differ in proximity and precision, both spatially and phylogenetically/taxonomically. For example, a dragonfly time-series from Italy might be matched with a) predictions for a congeneric species in Finland, and b) evidence for damselflies (a different suborder) in Switzerland. The first example is closer taxonomically but farther in geographical and environmental space. Across all such comparisons, we will conduct analyses to quantify the shape of these distance-decay curves, in order to understand the degree to which threat-response relationships are transferable.
c) What evidence exists for interactions between pressures? Datasets collated in WP1 are subject to more than one threatening process, yet the TRM (WP2) might lack strong evidence about synergies between threats. It is therefore capable of producing predictions for each time-series by assuming threats are additive, but we may also test whether the fit improves (on average) if we model a synergistic effect.
d) Which biodiversity metrics are most easily predicted? As noted in T1.3, there are multiple ways in which insect biodiversity trends might be measured, not all of which are likely to be equally predictable. For example, individual population trends are likely too noisy or idiosyncratic to be consistently predictable; by contrast, trends in species richness may be insensitive and show little or no signal. We anticipate that metrics of intermediate information content (e.g. evenness) are likely to show the most consistent patterns compared with simpler metrics (e.g. species richness or occupancy).
Task 3.2: What are the domains of (un)predictability? We can define the probabilistic statements of T2.5 as a set of prior beliefs about the relationship between anthropogenic threats and insect biodiversity. Having challenged these probabilistic statements with data and optimised the evidence weights (T3.1), we will be in a position to evaluate the posterior distribution of each threat-response relationship, distinguishing three broad domains:
- where indirect evidence is strongly supported by data, defined as where parameter values don’t change much but the credible interval narrows, i.e. indirect evidence (e.g. space for time substitution or expert opinion) provides effective predictive tools describing insect decline;
- where parameter value changes significantly because of data, indicating that the indirect evidence is overruled, overturning what we thought we knew (indirect evidence is inadequate); and
- undecidable, where the posterior distribution is indistinguishable from the prior. This third category identifies the evidence gaps, which can prioritise future research.
Understanding where these inadequacies lie, in particular which facets of insect biodiversity (abundance, species richness, diversity, functional diversity) are predictable, will allow us to address the critical challenge of projecting outside the range of places and taxa with time-series data, and into the future.
Task 3.3. Identifying current hotspots of insect decline. We will apply the posterior probabilities from T3.1 & T3.2 across the full spectrum of insect taxa and ecosystems, to identify hotspots (geographic and taxonomic) of decline. This will allow estimation of recent trends (quantifying uncertainty) in data-poor taxa and regions (especially the tropics), permitting a synoptic global overview of insect declines. We will also project individual models from WP2 to identify hotspots of loss associated with land-use and climate change (and other important threats), and their interactions.
Task 3.4: Resetting the baseline. We will project the models (see T3.3) onto historical estimates of important pressures, especially land use (tinyurl.com/s52euzt) and climate (tinyurl.com/vrowjpx), to estimate the historical baselines for insect biodiversity. We will focus on regions where the evidence is strong (T3.2) and on functionally-important insect groups (defined above). For some threats, projections will require proxies, e.g. we previously developed projections of land-use intensity using land use, human population density and economic region. The uncertainty resulting from the use of proxies will be propagated into the projections. Our efforts to collate or develop new projections will be focused on those pressures identified as having the greatest impacts. Where possible, we will validate these hindcast projections with independent data from quasi-time-series (T1.2).