Long-term abundance trends of insect taxa are only weakly correlated
Abstract
Changes in the abundances of animals, such as with the ongoing concern about insect declines, are often assumed to be general across taxa. However, this assumption is largely untested. Here, we used a database of assemblage-wide long-term insect and arachnid monitoring to compare abundance trends among co-occurring pairs of taxa. We show that 60% of co-occurring taxa qualitatively showed long-term trends in the same direction—either both increasing or both decreasing. However, in terms of magnitude, temporal trends were only weakly correlated (mean freshwater r = 0.05 (±0.03), mean terrestrial r = 0.12 (±0.09)). The strongest correlation was between trends of beetles and those of moths/butterflies (r = 0.26). Overall, even though there is some support for directional similarity in temporal trends, we find that changes in the abundance of one taxon provide little information on the changes of other taxa. No clear candidate for umbrella or indicator taxa emerged from our analysis. We conclude that obtaining a better picture of changes in insect abundances will require monitoring of multiple taxa, which remains uncommon, especially in the terrestrial realm.
1. Introduction
Despite evidence for dramatic declines in insect biomass and abundance in some parts of the world [1,2], there is considerable variation in trends across different locations, taxa and realms [3–7]. What remains unknown, however, is whether trends of different taxonomic groups (e.g. orders or families) behave in a similar way at a given site, or whether taxa differ in their trends. Similarity of trends at the site-level (i.e. different taxa decreasing or increasing together) would suggest that these insect groups respond similarly to the same environmental pressures. Furthermore, it would support the utility of monitoring only a subset of taxa in a given location to provide indicators of trends for other, unmeasured taxa. Alternatively, if trends in abundances differ among taxa, we would need to monitor all taxa to understand biodiversity trends.
The use of indicator taxa, also known as surrogate or umbrella taxa, has long been advocated as a solution for the many challenges of long-term monitoring (e.g. effort, funding and expertise) [8,9], and the idea is still commonplace (e.g. [10,11]). Several terrestrial taxa have been proposed as indicators of terrestrial insect biodiversity, including butterflies [2], moths [12], true bugs [13] and parasitoid wasps [14]. Similarly, in the freshwater realm, mayflies, stoneflies and caddisflies are often used as indicators of steam-water quality [10,15]. However, in terms of temporal changes, none of these groups have been directly compared with other co-occurring taxa, which would be necessary for a taxon to be a useful indicator.
In this report, we explicitly test for correlations in long-term trends of co-occurring insect and arachnid taxa (for brevity referred to as ‘insects’). Strong positive correlations (concomitant increases or decreases of taxa across sites) would indicate that taxa respond consistently to natural or anthropogenic drivers and that one taxon can be used as an indicator of other taxa. Negative correlations would indicate opposite responses to drivers. No correlations between trends would instead indicate a balance between winner and loser taxa [16]. We tested these hypotheses using the InsectChange database [17], which contains long-term (10+ years) data of insect assemblage-level monitoring data, assessed using consistent methodology over time. We used 35 studies (968 individual sites) that measured abundances of multiple orders and/or families of insects, springtails and arachnids. We determined whether any groups showed particularly highly correlations with other taxa and, thus, could serve as potential indicator taxon.
2. Methods
We tested the correlations among abundance trends at two levels of taxonomic aggregation: (1) taxonomic order, and (2) common groupings often assessed and compared, consisting of a mix of orders, suborders and families (see [13,18]).
From the InsectChange database [17], we selected datasets and locations that reported on at least two orders or common groups of insects, springtails and/or arachnids. For each sampling location, we first added zeroes for all taxa in years that they were not observed (this was necessary as most original databases record only presences, but the absences can be inferred as the sampling was standardized), and then aggregated all individuals per order or common grouping. We only analysed the associations between pairs of taxa that were jointly observed in at least five datasets and 20 plots, where the two taxa under comparison were present in at least half of the sampling years. Finally, we only included plots and datasets that provided at least three years of data over the time series. The number of datasets we could use for freshwater insects (22 datasets with 867 plots) was much higher than for terrestrial insects (13 datasets with 101 plots; figure 1). Because of unequal global data availability, the majority of datasets came from Europe and North America (i.e. 86% of the freshwater, and 66% of the terrestrial datasets; electronic supplementary material, figure S1).
Figure 1. Number of datasets available for each pair of taxa. Pairs of orders were only assessed if they were jointly observed in at least five datasets and in 20 individual plots, and each taxon was present in at least half of all years.
We designed our analysis to incorporate the underlying uncertainty of the species’ trend estimates, using a two-stage approach: for each taxon, we first calculated the temporal trends of abundance (log10 transformed) in each plot using a Bayesian regression model with log10 abundance+1 as the dependent variable and ‘year’ as the independent variable. When there were multiple samplings per year, we included a random intercept on month. We chose priors for the slope estimate: mean = 0 and s.d. = 1, expecting no trend, and for all other priors the default settings. To propagate the uncertainty surrounding the estimated trends, we sampled 1000 values from each of the two posterior distributions of the slope estimates. In the second stage, we calculated 1000 correlation coefficients based on the 1000 samples from the posterior distributions, across all plots where two taxa co-occurred (figure 1). We checked whether weighted correlations would give similar results; for the procedure and results see electronic supplementary material, figure S3 and table S3. We performed χ2 tests on the paired mean slope estimates to test whether the proportions of positive and negative associations deviated from an expectation of equal distribution. All analyses were done in R v. 4.1.1 using the brms package [19] for the Bayesian analysis.
We inferred the strength of evidence for a correlation between two taxa using the distribution of the sampled correlation coefficients: strong evidence was inferred if 95% of the sampled correlation coefficients were larger or smaller than zero (i.e. positively or negatively correlated). Likewise, we inferred moderate evidence when 90%, and weak evidence when 80% were larger or smaller than zero, respectively. Finally, if less than 80% of the correlation coefficients were larger or smaller than zero, we interpreted this as no evidence for a correlation.
3. Results
In all, we calculated 15 825 paired temporal trends across all plots. Among the compared orders, the trends went in the same direction in 60% of the cases in freshwater (χ2 = 422.65, p < 0.001) and 58% of the cases in the terrestrial realm (χ2 = 10.19, p = 0.001), which was greater than expected by chance. Restricting only to pairs where at least one taxon showed strong evidence for a temporal trend, we found that 70% (freshwater) and 69% (terrestrial) of trends changed in the same direction.
In terms of correlative strength, the trend estimates were weak but positively correlated, indicating some congruence in the magnitude of the trends of the taxa. Across all pairs of orders, the mean correlation coefficient was 0.05 ± 0.03 (s.d.) for the freshwater realm, and 0.12 ± 0.09 for the terrestrial realm. For the common groupings, the mean correlation coefficient was 0.04 ± 0.02 and 0.10 ± 0.08 for the two realms, respectively. Moreover, for the orders, seven out of 46 pairwise comparisons showed weak to strong evidence for a positive correlation (figure 2b), but for the common groupings only two out of 23 comparisons. We found no evidence for any negative correlations for either the orders or the common groupings. The highest median correlation coefficient in the terrestrial realm was 0.26, for Lepidoptera (butterflies and moths) and Coleoptera (beetles), and for Coleoptera and Araneae (spiders) (r = 0.26), but neither provided strong evidence. In the freshwater realm, the highest correlation was 0.12 (moderate evidence) for Plecoptera (stoneflies) and Coleoptera (water beetles). Among the common groupings, we found the highest correlation of 0.20 between Carabidae (ground beetles) and Staphylinidae (rove beetles). However, the only taxon pair with evidence for a trend differing from zero was Ephemeroptera (mayflies) and Simuliidae (black flies) (electronic supplementary material, figure S1). When we used weighted correlation, we found more correlations differing from zero, as the more extreme slopes were downweighted (electronic supplementary material, figure S3a,b), but only small differences in the mean correlation coefficients (electronic supplementary material, figure S3c and table S2).
Figure 2. (a) Relations between the temporal abundance trends of all pairs of orders. This shows a positive but very weak overall relationship. For plotting purposes, slopes that were part of multiple pairwise comparisons are only depicted once. (b) Pairwise correlation coefficients among all assessed orders (for a matrix with the exact correlation coefficients see electronic supplementary material, table S1). ○ denotes a pair was not assessed; * denotes weak evidence for a correlation (80% of the sampled correlation coefficients were larger or smaller than zero); ** denotes moderate evidence (90% larger or smaller than zero); *** denotes strong evidence (95% larger or smaller than zero). No signifier indicates that more than 10% of the correlations fell to either side of zero. (c) Density plots of the correlation coefficients for all comparisons for each assessed order, based on the 1000 correlation coefficients calculated from the sampled posterior distributions of each taxon pair.
The taxa with the highest overall correlation with all other taxa were Lepidoptera and Coleoptera, with both a mean r = 0.18 (figure 2c). Using weighted correlation, Lepidoptera also showed the highest mean correlation (r = 0.19).
4. Discussion
Our synthetic analysis provides support for consistently positive, but weak correlations among temporal trends of insect and arachnid groups. The majority of pairwise comparisons showed that abundances changed in the same direction, either both increasing or both decreasing. This result indicates that there is no overall balance between ‘winner’ and ‘loser’ taxa within insect communities at the plot scale, as may have been expected given the overall directional trends observed previously [3]. The highest correlation coefficient among any pair of taxa, between Lepidoptera (butterflies and moths) and Coleoptera (beetles), was still quite weak (0.26). Hence, the magnitude of the trend of one taxon is unlikely to provide strong information on magnitude of the trends of other taxa, and thus temporal trends of potential indicator taxa are likely to be too weak to be of use in conservation planning or monitoring. This result is particularly important in the context of recent scientific and popular articles addressing the issue of insect declines [20], and in the context of improving future monitoring. Often, studies on one or few taxa showing strong trends are extrapolated as evidence for (or against) such strong declines more generally [21], whereas our study provides little support for such extrapolations to be meaningful.
The long-term monitoring data we used only allowed the comparison of order- or family-level abundance trends across sites and was not suitable to test congruence of trends in species richness or other biodiversity metrics. There is ongoing debate about which is the most appropriate taxonomic rank for indicator groups [22,23]. However, abundance trends of higher taxa are a particularly sensitive metric of temporal changes, as significant abundance trends prevail in both the terrestrial and freshwater realms [3], and ecological patterns are generally well represented at higher taxonomic levels [23].
Our results examining correlations between taxa in time series can also be compared to studies of taxon congruence done in a spatial context (reviewed in [24–26]). Although most of this body of work focused on trends in diversity or species composition in space, similarly to our abundance trends through time, most conclude that correlations and predictability among taxa are too weak to be practical in biodiversity assessments or conservation planning [25–27], despite some studies reporting high spatial congruence among some insect taxa [13,14]. The overall weak correlations we found among temporal trends of co-occurring taxa may not be surprising for several reasons: First, temporal environmental changes are typically weak compared to spatial gradients (e.g, land-use intensity gradients), at least over the spatial and temporal scales of most monitoring studies. Secondly, in cases of rapid change (e.g. land-use change or natural catastrophes), monitoring is usually stopped [28]. Finally, even if the abundances of two taxa are highly correlated, sampling effects will bias the correlation that is estimated from monitoring data towards zero (so-called correlation disattenuation or dilution [29]).
Although most of the correlations we observed among taxa through time were weak, we did find some intriguing trends with two well-studied groups in the terrestrial realm: Lepidoptera and Coleoptera. First, the two groups tended to show positive correlations in their trends. Second, the two groups were the most likely to show at least some correlations with other taxa, suggesting they may serve as somewhat of an indicator of insect trends, lending some support to the assertions of Thomas [2] and New [12] that butterflies and moths may serve as a possible indicator taxon. By contrast, we found overall lower correlations and higher uncertainty in trend estimates in the freshwater realm, which is probably due to differences in methodology. Freshwater monitoring programmes typically have a lower sampling effort than terrestrial programmes. The latter typically use continuous or repeated sampling throughout the year, probably owing to their historic focus on population dynamics [30,31]. Freshwater habitats, by contrast, are typically sampled once or twice per year over a relatively small area as part of water quality monitoring programmes. Contributing may be the often lower number of years sampled in freshwater monitoring, as well as human error [32]. Nevertheless, when we put increased restrictions on the number of years of data provided, the variability in the freshwater data, although attenuated, remained higher than in the terrestrial habitats, and the correlations remained weaker. More intensive sampling of freshwater systems may improve trend estimates but, of course, has added costs.
Overall, our results suggest that the use of indicator groups to evaluate insect abundance trends across taxa and for insect monitoring is not likely to be successful. To monitor changes in biodiversity, it will thus be necessary to assess multiple taxonomic groups simultaneously. Fortunately, with the onset of modern technologies such as automated computer vision [33] and genetic methods based on eDNA [34] or metabarcoding [35], multi-taxon monitoring will become more accessible, and the need for indicator taxa is likely to fade.
Data accessibility
The full reproducible dataset and all code are available at Zenodo [36]. The raw data and meta-data are available at Knowledge Network Biocomplexity [37] with more details on the original datasets used in [17].
Authors' contributions
R.v.K.: conceptualization, data curation, formal analysis, investigation, methodology, project administration, writing—original draft; D.E.B.: formal analysis, investigation, methodology, writing—review and editing; K.B.G.: data curation, writing—review and editing; J.M.C.: conceptualization, investigation, supervision, writing—review and editing.
All authors gave final approval for publication and agreed to be held accountable for the work performed therein.
Competing interests
We declare we have no competing interests.
Funding
R.v.K., J.M.C. and D.E.B. were supported by the German Centre for Integrative Biodiversity Research (iDiv) Halle-Jena-Leipzig, funded by the German Research Foundation; FZT 118), and K.B.G. was supported by the Russian Foundation for Basic Research (19-05-00245). Some of the data analysed here were collected using NSF grants to the LTER Network (DEB-0423704, DEB-2025982).
Acknowledgements
Data from the Greenland Ecosystem Monitoring Programme were provided by the Department of Bioscience, Aarhus University, Denmark. We thank Abdualmaghed Al-Hemiary and Nina Naderi for help extracting the data, and all researchers who made their data accessible. References to all original data sources used in this paper: [38–77].