Different roles of concurring climate and regional land-use changes in past 40 years' insect trends

Abstract

Climate and land-use changes are main drivers of insect declines, but their combined effects have not yet been quantified over large spatiotemporal scales. We analysed changes in the distribution (mean occupancy of squares) of 390 insect species (butterflies, grasshoppers, dragonflies), using 1.45 million records from across bioclimatic gradients of Switzerland between 1980 and 2020. We found no overall decline, but strong increases and decreases in the distributions of different species. For species that showed strongest increases (25% quantile), the average proportion of occupied squares increased in 40 years by 0.128 (95% credible interval: 0.123-0.132), which equals an average increase in mean occupancy of 71.3% (95% CI: 67.4-75.1%) relative to their 40-year mean occupancy. For species that showed strongest declines (25% quantile), the average proportion decreased by 0.0660 (95% CI: 0.0613-0.0709), equalling an average decrease in mean occupancy of 58.3% (95% CI: 52.2-64.4%). Decreases were strongest for narrow-ranged, specialised, and cold-adapted species. Short-term distribution changes were associated to both climate changes and regional land-use changes. Moreover, interactive effects between climate and regional land-use changes confirm that the various drivers of global change can have even greater impacts on biodiversity in combination than alone. In contrast, 40-year distribution changes were not clearly related to regional land-use changes, potentially reflecting mixed changes in local land use after 1980. Climate warming however was strongly linked to 40-year changes, indicating its key role in driving insect trends of temperate regions in recent decades.

Fig. 1. Study species, study region, and climate and land-use changes.

a In total, 215 butterfly species, 103 grasshopper species and 72 dragonfly species were analysed, which covered a gradient in habitat specialisation (0: lowest specialisation; 1: highest specialisation; based on literature-derived habitat preferences) and in preferred temperature niches (average annual temperatures of Europe-wide distribution). Curves show marginal density distributions per group and trait; dashed lines indicate means. b Switzerland, the study country situated along the Alps in Central Europe, was divided into five biogeographic regions indicated by colours (legend in (c)) and two elevation classes (above and below 1000 m asl.; high elevation not distinguished for the low-elevational Plateau region) indicated by shadings (strong colour for low elevation), resulting in nine bioclimatic zones. Dark squares in the map show squares for which data of at least one insect group were analysed (darker colours indicate more records, Supplementary Fig. S2a). c Changes of six potential driver variables were assessed for eight focal 5-year intervals in the study period (arranged from top to bottom for each variable) for the nine bioclimatic zones (combinations of biogeographic region and elevation class). For all variables, change was standardised to standard deviation 1.

Fig. 2. Trend estimates of mean occupancy across 40 years (1980–2020).

40-year trend estimates are shown for the (a) 215 butterfly species, the (b) 103 grasshopper species and the (c) 72 dragonfly species. Species are ordered along the point estimate of their trend (mean of posterior distribution), vertical segments show the 95% credible intervals. Trend estimates reflect a 40-year change in mean occupancy, which is illustrated in the inset plot at the right starting from the overall mean occupancy (mean across all species and years). The vertical dashed lines show the median of the number of species, the vertical solid lines show where negative trend estimates change to positive trend estimates along with a bootstrap 95% confidence interval (n = 9999). Bars in the bottom panels show 40-year mean occupancy estimates for the whole of Switzerland for each species.

Fig. 3. Regression model linking climate / land-use changes and species traits to short-term species trends.

a Schematic model representation. The response variable was 5-year species trends of regional mean occupancy (n_tot =  20,048). Explanatory variables included changes in climate (annual mean temperature, temperature seasonality, summer precipitation) and land use (total agricultural area, grassland-use intensity, crop-use intensity), two trait variables, elevation (low or high) and interactions (indicated with ×) (cf. Fig. 1). In the restricted model version underlying the results presented in panel (b), parameter estimates for change in total agricultural area and grassland-use intensity (in italic) were only based on species of agriculturally influenced habitats (183 butterfly species, 93 grasshopper species; n = 13,968). Non-independence within insect groups, time intervals, bioclimatic zones and species was accounted for. b Model results along the same arrangement as in (a). Curves show posterior distributions of model estimates; fill colours indicating effect sizes (positive values in blue, negative values red); dashed vertical lines indicate zero; numbers are posterior distribution means. Two-way interactions of change and trait variables with elevation are included such that upwards-facing curves show model estimates for high elevation and downwards-facing curves for low elevation (too low amounts of crop fields at high elevation to include interaction with crop-use intensity). Overlapping areas show other two-way interactions. The bottom centre panel shows how to interpret effect sizes. Starting at the overall mean occupancy, expected mean occupancy changes in 5 years when an explanatory variable is increased by one standard deviation are shown. All explanatory variables were standardised prior to analysis.

Fig. 4. Explained variance of long-term species trends for different climate and land-use scenarios.

Regression model results (Fig. 3) just reflect relationships of short-term trends to short-term changes, which can be of different relevance to explain long-term trends depending on the long-term change of drivers. Thus, based on the results of the regression model only including species of agriculturally influenced habitats (Supplementary Table S3), 40-year trends in mean occupancy (across the whole of Switzerland) were predicted for all 276 species for scenarios of no climate and land-use change (i.e., change assumed to be zero across all zones and time intervals) and for different combinations of single measured trajectories of climate and land-use variables. The R² values (squared Pearson correlations) indicate how well the predictions align with the observed long-term species trends. The green bar on the right shows the match for the predictions with all climate and land-use variables following their measured trajectories (analogous to the R² of the model presented in Fig. 3a). Bars show means of the posterior distribution; vertical lines show 80%- and 95% credible intervals.

Fig. 5. Forty-year trends dependent on species traits.

Based on the regression model on 5-year trends (Fig. 3), average predicted change in occupancy within a bioclimatic zone across the whole 40-year study period (1980–2020) is shown in colour across the temperature niche as well as the habitat specialisation gradient of the 390 study species (points show the trait range of all species). In relation to mean 40-year occupancy across all species, a change in the occupancy of 0.01 corresponds to a decrease or an increase of the distribution of a species in a bioclimatic zone of 4.5%.