Many studies have documented (or predicted) large-scale declines in biodiversity and population size due to changes in land use and climate (e.g., Outhwaite et al., 2022; Warren et al., 2018). However, these studies can be confounded if they combine data from different populations and taxonomic groups without controlling for geographic differences in phenology (Duchenne et al., 2022), habitat selection (Bladon et al., 2020), and life history (Belitz et al., 2021; Kingsolver et al., 2011). There is ample evidence that the responses of plant and animal populations to temperature vary geographically (e.g., Louthan et al., 2021; Primack et al., 2009), but climate change predictions based on “ecological niche” or “species distribution” models rarely account for this variation (DeMarche et al., 2019; Hallfors et al., 2016; Zhang & Kubota, 2021). These models usually assume that the effects of climate change (and other factors) on abundance and distribution are consistent across space and time (see Sinclair et al., 2016 for discussion). In the context of climate change studies, this “space-for-time” assumption allows researchers to extrapolate the thermal responses of plants and animals (from different climates) to warmer (or cooler) times in the past or future. The evidence in support of this assumption is mixed, so some researchers have advocated studying spatial variation in population responses to understand better the causes of long-term changes, particularly climate change (Blüthgen et al., 2022).
Ectotherms, such as insects and other arthropods, are particularly sensitive to temperature and, thus, should be good indicators of the effects of climate change (Buckley, 2022). However, the large-scale effects of climate and land use change across insect populations are poorly known and have been widely debated (e.g., Dornelas & Daskalova, 2020; Wagner et al., 2021). Much of the controversy about the extent of declines in insect abundance has involved problems that arise when combining geographically and temporally heterogeneous data from different populations and taxa in meta-analyses (e.g., Desquilbet et al., 2020; Didham et al., 2020; Duchenne et al., 2022). This heterogeneity is illustrated by the Rothamsted Insect Survey in the UK, which is one of the longest-running studies of insect abundance (since 1965). In this survey, there was a decline in the total biomass of all species at one site, but not at three other sites, and the decline was dominated by one species of fly (Shortall et al., 2009). Moths also showed a significant, but fluctuating, decline across the UK, particularly in certain types of habitats (Bell et al., 2020), but aphids showed no temporal decline. These heterogeneous results emphasize that we should not extrapolate population trends across regions and taxa unless we have an understanding of the causes, and, for climate change, the responses of taxa to temperature across their range.
We still know relatively little about how climate change affects the abundance of insects in different populations across entire continents (but see; Crossley et al., 2021, 2022; Outhwaite et al., 2022; Soroye et al., 2020; Warren et al., 2018). To date, most large-scale insect surveys have been designed primarily to examine diversity rather than biomass or abundance (e.g., Steinke et al., 2017), or have focused on specific taxa, such as bumblebees (Kerr et al., 2015; Soroye et al., 2020; Weaver & Mallinger, 2022) or butterflies (e.g., Crone et al., 2019; Forister et al., 2021). Predictions about the influence of climate change on insects depend critically on knowledge of the effect of temperature on biomass or abundance, and, perhaps most importantly, whether those relationships can be extrapolated across large regions and different taxa (e.g., see Hallfors et al., 2016; Zhang & Kubota, 2021). To address this need, we established a network of Malaise traps to test the assumption that insect biomass responds to temperature similarly across North America. A second goal was to use our data to design more efficient sampling schemes and provide a baseline for future long-term studies. There has been relatively little discussion of study design for insect monitoring in terms of the number of samples needed to detect population changes (Lebuhn et al., 2013) or the optimal spacing of traps (Steinke et al., 2021), so we also examined the spatial correlation between samples and estimated sample sizes needed to detect significant evidence of insect declines.
- University of Wisconsin-Milwaukee
- Siena Heights University
- University of North Carolina - Chapel Hill
- Baylor University
- Université de Sherbrooke
- University of Illinois
- University of Saskatchewan
- Trinity College Hartford
- Salisbury University
- Amherst College
- University of Montana
- University of Northern British Columbia
- University of Iowa
- University of Manitoba
- George Washington University
- Cornell University
- Dalhousie University
- Florida State University
- University of Maryland
- Luther College
- University of Guelph
- University of Waterloo
- Queen‘s University
- University of Wisconsin - Parkside
- Catawba College
- University of Lynchburg
- University of the District of Columbia
- Mississippi State University
- University of Florida
- Westfield State University
- University of Alberta
- Central Washington University
- Appalachian State University
- University of Alaska Fairbanks
- Davidson College
- University of Jamestown
Summary
Climate change models often assume similar responses to temperatures acrossthe range of a species, but local adaptation or phenotypic plasticity can lead vplants and animals to respond differently to temperature in different parts oftheir range. To date, there have been few tests of this assumption at the scaleof continents, so it is unclear if this is a large-scale problem. Here, we exam-ined the assumption that insect taxa show similar responses to temperature at 96 sites in grassy habitats across North America. We sampled insects withMalaise traps during 2019–2021 (N=1041 samples) and examined the bio-mass of insects in relation to temperature and time of season. Our samplesmostly contained Diptera (33%), Lepidoptera (19%), Hymenoptera (18%), andColeoptera (10%). We found strong regional differences in the phenology ofinsects and their response to temperature, even within the same taxonomicgroup, habitat type, and time of season. For example, the biomass ofnematoceran flies increased across the season in the central part of the conti-nent, but it only showed a small increase in the Northeast and a seasonal decline in the Southeast and West. At a smaller scale, insect biomass at differ-ent traps operating on the same days was correlated up to ~75 km apart.Large-scale geographic and phenological variation in insect biomass and abun-dance has not been studied well, and it is a major source of controversy in pre-vious analyses of insect declines that have aggregated studies from differentlocations and time periods. Our study illustrates that large-scale predictionsabout changes in insect populations, and their causes, will need to incorporateregional and taxonomic differences in the response to temperature.
Methodology
We sampled insect biomass using Malaise traps at 96 sites across Canada and the USA from 2019 to 2021 (Figure 1; Appendix S1: Table S1). Malaise traps primarily sample flying insects that intercept and climb up the mesh walls (Skvarla et al., 2021). We were also interested in studying flying insects because they form the food supply for aerial insectivores, such as swallows, which have shown some of the greatest declines in population size of North American birds (Spiller & Dettmers, 2019). Traps were located in open areas (>20 m from trees) with mostly grassy vegetation to standardize land cover type (Appendix S1: Figure S1). Collaborators were recruited throughout Canada and the USA in 2018–2020 through email messages to the ECOLOG-L (ECOLOG-L@community.esa.org) and ENTOMO-L (ECOLOG-L@listserv.uoguelph.ca) list-servs, as well as networks of researchers studying aerial insectivores, especially tree swallows (Tachycineta bicolor). Aerial insectivores have been declining sharply, especially in eastern North America (Spiller & Dettmers, 2019), and there is concern that this could be caused by a decline in their food supply (flying insects). Thus, one of the reasons for organizing this study was to examine the food supply of tree swallows across their range, although a few of our samples (in Georgia and Florida) came from areas outside their breeding range. We attempted to maximize participation among researchers by using a simple protocol with a small number of sampling periods (three) and minimal sorting and processing of samples. Insect biomass typically increases over the season (mainly May–July) at our sampling locations (see Results), so we primarily sampled at three phenologically defined time periods to reduce sampling error from seasonal changes (see below). To simplify post-collection processing, we primarily sorted insects to order, rather than finer taxonomic classifications.
Malaise traps and sampling procedures
We sampled insect biomass using Malaise traps at 96 sites across Canada and the USA from 2019 to 2021 (Figure 1; Appendix S1: Table S1). Malaise traps primarily sample flying insects that intercept and climb up the mesh walls (Skvarla et al., 2021). We were also interested in studying flying insects because they form the food supply for aerial insectivores, such as swallows, which have shown some of the greatest declines in population size of North American birds (Spiller & Dettmers, 2019). Traps were located in open areas (>20 m from trees) with mostly grassy vegetation to standardize land cover type (Appendix S1: Figure S1). Collaborators were recruited throughout Canada and the USA in 2018–2020 through email messages to the ECOLOG-L (ECOLOG-L@community.esa.org) and ENTOMO-L (ECOLOG-L@listserv.uoguelph.ca) list-servs, as well as networks of researchers studying aerial insectivores, especially tree swallows (Tachycineta bicolor). Aerial insectivores have been declining sharply, especially in eastern North America (Spiller & Dettmers, 2019), and there is concern that this could be caused by a decline in their food supply (flying insects). Thus, one of the reasons for organizing this study was to examine the food supply of tree swallows across their range, although a few of our samples (in Georgia and Florida) came from areas outside their breeding range. We attempted to maximize participation among researchers by using a simple protocol with a small number of sampling periods (three) and minimal sorting and processing of samples. Insect biomass typically increases over the season (mainly May–July) at our sampling locations (see Results), so we primarily sampled at three phenologically defined time periods to reduce sampling error from seasonal changes (see below). To simplify post-collection processing, we primarily sorted insects to order, rather than finer taxonomic classifications.
Insect Biomass
o estimate insect biomass, we first sorted most samples by order of insect, and then to the suborder Nematocera (including mosquitoes, midges, and blackflies) for dipterans. We were interested in nematocerans because many of them contain high levels of long-chain omega-3 polyunsaturated fatty acids, which are important to the breeding success of tree swallows and other aerial insectivores (Twining et al., 2016). We also sorted and weighed spiders (order Araneae), but they were a relatively small portion of the total biomass, so we do not analyze them here (see Appendix S1: Table S2 for a taxonomic summary). After 1 h of sorting and drying at room temperature, we weighed the samples to 1 mg on electronic balances. This protocol was chosen to allow time to sort samples to order prior to weighing. Biomass was measured separately for each order (suborder for Nematocera) and then summed to calculate total biomass. Some sites did not have separate data for each order, so we only estimated total biomass (13%, 139/1041 samples). The original insect data are available online at figshare (Dunn et al., 2023).
Weather Variables
To examine the potential effects of weather on insect biomass, we obtained daily average data for temperature (mean, maximum and minimum), precipitation (rainfall), and wind speed. We used weather stations at each trap site, if available, or from the nearest weather station with complete data in the Global Historical Climatology Network daily (https://www.ncei.noaa.gov/data/global-historical-climatology-network-daily/access/) at the National Center for Environmental Information (Menne et al., 2012). All weather data are provided at figshare (Dunn et al., 2023), and the weather stations used in this study are summarized in Appendix S1: Table S8.
The correlations between average (r = 0.948, 95% confidence interval (CI): 0.935–0.958, N = 317) and maximum (r = 0.920, 95% CI: 0.901–0.936, N = 311) temperature, total precipitation (r = 0.613, 95% CI: 0.539–0.677, N = 315), and average wind speed (r = 0.642, 95% CI: 0.510–0.745, N = 99) at the trap locations and these GHCN stations were all significant (median distance between traps and stations = 40.3 km). Weather variables were averaged (temperature, wind speed) or summed (precipitation) for the days of each sampling period (usually 3 days). We first examined total biomass in relation to these weather variables to produce a parsimonious set of predictor variables for the inclusion in subsequent analyses (based on corrected Akaike information criterion [AICc] values; Appendix S1: Table S3). The initial regression models for total biomass included temperature (mean and maximum), precipitation, wind speed, sampling (calendar) date, latitude, longitude, and elevation as predictors. Mean and maximum temperatures were analyzed separately because they were highly correlated. These initial analyses revealed that maximum temperature (Tmax) was the best predictor of total biomass (Appendix S1: Table S3; β = 0.031 ± 0.001 [SE], F1,1039 = 249.5, p < 0.001), so we focused on it in our main analyses below.
Statistical analyses
We were primarily interested in geographic and seasonal variation in arthropod biomass, especially in response to temperature, so our main analyses included interactions between: (1) location and date, and (2) location and temperature to test for variation in the effects of temperature in different locations and at different times of the season. We first analyzed biomass data from the three main sampling periods (laying date, hatch date, nestling day 12) to control for the inclusion of data from different times of the season, which has been a source of controversy in recent studies (e.g., Welti et al., 2021). Then we examined data from throughout the sampling season, which allowed a larger sample size. Here we used sampling date as a predictor to control for phenological changes across the entire season. Prior to inferential analyses, we examined the distribution of biomass and then applied log10 transformations to improve the normality of residuals.
Mixed models of insect biomass during the three phenological sampling periods included the fixed effects of the phenological period of the sample, the maximum temperature during the sample period, geographic region (see below) and the interactions between region and both phenological period and maximum temperature. Random effects included sampling site and year nested within site (Appendix S1: Tables S3 and S4). In these mixed models we used site to account for the lack of spatial independence of data from nearby traps; however, we also conducted separate analyses of the spatial correlation between traps using the Euclidean distance between traps in km. We included the geographic region in the mixed models (Figure 1), which was coded as four categories: West (west of the 100th meridian), Central (87–100° W), Northeast (east of the 87th meridian and north of 38° N), and Southeast (east of the 87th meridian and south of 38° N). These four regions were chosen because most of our sites were located in a much wider longitudinal (64 to 157° W), than latitudinal (29.6 to 64.9° N) band, except in the east where we had numerous sites in the south, so we divided the eastern region into Northeast and Southeast (Figure 1). We used these regions in analyses of biomass, rather than latitude and longitude, because they explained similar amounts of variation, had lower AICc values, and were simpler to interpret than models with latitude and longitude (Appendix S1: Table S3). We conducted all analyses in JMP v. 16 (SAS Institute, 2021), unless indicated otherwise. We estimated the proportion of variance explained by fixed effects (marginal R2, �m2) and by both fixed and random effects (conditional R2, �c2) using the r.squaredGLMM command (Nakagawa et al., 2017) in the MuMIn package (cran.r-project.org/web/packages/MuMIn) in R v. 4.1 (R Core Team, 2021).
Designing future studies
To help design future studies, we conducted repeatability analyses of total biomass using the rptR package (Stoffel et al., 2017), and power analyses using the simR package (Green & MacLeod, 2016) in R v. 4.1 (R Core Team, 2021). Previous studies of insect biomass have reported declines of up to 6% annually in Germany (Hallmann et al., 2017), although the average from a meta-analysis was 1% annually for terrestrial insects (van Klink et al., 2020). Therefore, we estimated the minimum necessary sample sizes to achieve a significant effect of year based on annual declines of 1% and 5%. Here we started with the observed biomass values in our phenological data set (N = 394 samples) and imposed a constant 1% or 5% annual decline in biomass in future years. On average, we collected 131 samples per year during the three main phenological periods of this study (394 total/3 years = 131 per year), so we assumed 131 samples would be collected each year in the simulated future. Using simR we estimated power in mixed models similar to those above but used the lme4 package (Bates & Sarkar, 2007) in R.