However, only temperature, precipitation and wind speed were retained for analysis, as the other variables were significantly correlated with the selected variables [R(temperature, cover) = −43.2%, R(temperature, sunshine) = 53.4%, R(precipitation, moisture) = −47.3%] and because we wanted to keep the number of covariates as low as possible. However, the number of studies on insect trends with sufficient replication and spatial coverage are limited [10, 23–25] and restricted to certain well-studied taxa. https://doi.org/10.1371/journal.pone.0185809.g001. The expected residual variance of each sample , is expressed as the sum of variances of daily biomass values (). Land use variables were measured at a coarse temporal resolution, but fortunately cover the temporal span of insect sampling. https://doi.org/10.1371/journal.pone.0185809.g004. Project administration, Models fitted independently for each habitat location. They operated continuously (day and night), and catches were emptied at regular intervals, on average every 11.2 days (sd = 6.3). Moreover, and contrary to expectation, trends in herb species richness were weakly negatively correlated with trends in insect biomass, when compared on per plot basis for plots sampled more than once. In light of previously suggested driving mechanisms, our analysis renders two of the prime suspects, i.e. As such, a seasonally weighted estimate (covering the period 1-April to 30-October; see methods) results in an overall 76.7% [74.8–78.5%] decline over a 27 year period. (6) Supervision, Although the fit is not perfect in the case of herb richness, we believe our trend adequately describes direction of change over time. Malaise traps are known to collect a much wider diversity of insect species (e.g. Increased agricultural intensification may have aggravated this reduction in insect abundance in the protected areas over the last few decades. PLoS ONE 12(10): In that study, 12.2m high suction traps were deployed at four locations in the UK over the time period 1973–2002, and showed a biomass decline at one of the four sites only. Ebmer, R. Eckelboom, B. Franzen, M. Grigo, J. Günneberg, J. Gusenleitner, K. Hamacher, F. Hartfeld, M. Hellenthal, J. Hembach, A. Hemmersbach, W. Hock, V. Huisman-Fiegen, J. Illmer, E. Jansen, U. Jäckel, F. Koch, M. Kreuels, P. Leideritz, I. Loksa, F. B. Ludescher, F. J. Mehring, G. Milbert, N. Mohr, P. Randazzo, K. Reissmann, S. Risch, B. Robert, J. de Rond, U. Sandmann, S. Scharf, P. Scherz, J. Schiffer, C. Schmidt, O. Additionally, variable exposure intervals between trap samples is expected to induce variation in trapped biomass between samples, and hence induce heteroscedasticity. Annual means as well as mean trends are depicted in the corresponding colors. Investigation, TracesOfWar.com tells you more! Writing – review & editing, Roles In order to obtain biomass per sample with sufficient accuracy and comparability, the measuring process was fixed using a standardized protocol [34]. University of Saskatchewan, CANADA, Received: July 28, 2017; Accepted: September 19, 2017; Published: October 18, 2017. Software, Finally, all significant variables were combined into our final model (Table 4), which included effects of an annual trend coefficient, season (linear and quadratic effect of day number), weather (temperature, precipitation, number of frost days), land use (cover of grassland and water, as well as interaction between grassland cover and trend), and habitat (number of herb and tree species as well as Ellenberg temperature). Writing – review & editing. (B), Seasonal phenology of insect biomass (seasonal quantiles of biomass at 5% intervals) across all locations revealing substantial annual variation in peak biomass (solid line) but no direction trend, suggesting no phenological changes have occurred with respect to temporal distribution of insect biomass. Firstly, our basic model (including an annual rate of decline) outperformed the null-model (without an annual rate of decline; ΔDIC = 822.62 units; Table 3), while at the same time, between-plot variation (i.s. Writing – review & editing, Roles On average, cover of arable land decreased, coverage of forests increased, while grassland and surface water did not change much in extent over the last three decades (S3 Fig). We decomposed the daily values of each weather variable into a long-term average trend (between years), a mean seasonal trend, and a yearly seasonal anomaly component (S2 Fig), modeled using regression splines [42] while controlling for altitude of weather stations. Seasonal profiles of daily biomass values are depicted in S4 Fig. Secondly, using only data from sites at which malaise traps were operating in at least two years, we estimated a rate of decline similar to using the full dataset (Fig 4), with the pattern of decline being congruent across locations (S4 Fig). https://doi.org/10.1371/journal.pone.0185809.s011, https://doi.org/10.1371/journal.pone.0185809.s012, https://doi.org/10.1371/journal.pone.0185809.s013. Next we developed 3 models each consisting of either weather variables (S1 Table), land use variables (S2 Table), or habitat variables. Do you want to create your own battlefield tour to sights of wars from the past? (2017) More than 75 percent decline over 27 years in total flying insect biomass in protected areas. Grants and permits that have made this work possible are listed below: Bezirksregierungen Düsseldorf & Köln, BfN - Bundesamt für Naturschutz, Land Nordrhein-Westfalen - Europäische Gemeinschaft ELER, Landesamt für Agrarordnung Nordrhein-Westfalen, Landesamt für Natur, Umwelt und Verbraucherschutz Nordrhein-Westfalen, Landesamt für Umwelt Brandenburg, Landesamt für Umwelt Rheinland-Pfalz, LVR - Landschaftsverband Rheinland, Naturschutzbund Deutschland, Nordrhein-Westfalen Stiftung, RBN Bergischer Naturschutzverein, RVR - Regionalverband Ruhr, SGD Nord Rheinland-Pfalz, Universitäten Bonn, Duisburg-Essen & Köln, Untere Landschaftsbehörden: Kreis Düren, Kreis Heinsberg, Kreis Kleve, Kreis Viersen, Kreis Wesel & AGLW, Stadt Düsseldorf, Stadt Köln, Stadt Krefeld, Rheinisch Bergischer Kreis, Rhein Kreis Neuss & Rhein-Sieg-Kreis. Our final model, based on including all significant variables from previous models, revealed a higher trend coefficient as compared to our basic model (log(λ) = −0.081, sd = 0.006, Table 4), suggesting that temporal developments in the considered explanatory variables counteracted biomass decline to some degree, leading to an even more negative coefficient for the annual trend. Yes (B) Distribution of mean annual rate of decline as estimated based on plot specific log-linear models (annual trend coefficient = −0.053, sd = 0.002, i.e. Trend coefficients of richness over time between a Poisson mixed effects model and a negative binomial model were comparable but differed in magnitude (Poisson GLMM: −0.034 (se = 0.003), vs NB GLMM −0.027 (se = 0.006)). Privacy statement, cookies, disclaimer and copyright, Belgium (1830-present, Constitutional Monarchy), Canada (1931-present, Constitutional Monarchy), Privacy statement, cookies, disclaimer and copyright. University of Sussex, School of Life Sciences, Falmer, Brighton BN1 9QG, United Kingdom, Roles For each location, different colors represent different years, with time color-coded from green (1989) to red (2016). Project administration, Methodology, Funding acquisition, Insect biomass was positively related to temperature and negatively to precipitation (S1 Table). Climate change is a well-known factor responsible for insect declines [15, 18, 21, 37]. PLOS ONE promises fair, rigorous peer review, Including lagged effects of weather revealed no effect of either number of frost days, or winter precipitation, on the biomass in the next season (S1 Table). Mean changes in plant species richness are depicted in S3C Fig. We used spatio-temporal geostatistical models [39, 40] to predict daily values for each weather variable to each trap location. Over the course of the study period, some temporal changes occurred in the means of the weather variables (S2 Fig), most notably an increase by 0.5°C in mean temperature and a decline 0.2 m/sec in mean wind speed. To verify that this is not the case, we fitted our basic model (but excluding the day number and year interaction to avoid overparameterization) to the subset of our data that includes only locations that were sampled in more than one year. Color gradient in all panels range from 1989 (blue) to 2016 (orange). Our data provide repetitions across years for only a subset of locations (n = 26 out of 63). Yet, the annual rate of decline was similar, suggesting that the decline is not specific to certain habitat types (S5 Fig). While presence of surface water appeared to significantly lower insect biomass, none of the other variables were significantly related to biomass. However, including interactions between the annual trend coefficient and land use variables increased the model fit slightly (Table 3), and revealed significant interactions for all variables except coverage of surface water (S2 Table). The widespread insect biomass decline is alarming, ever more so as all traps were placed in protected areas that are meant to preserve ecosystem functions and biodiversity. Note, zero values for tree and shrub species not depicted. where c is a global intercept, X a design matrix of dimensions n×p (number of samples × number of covariates; see Model analysis below), βx the corresponding vector of coefficients that measure the weather, habitat and land use effects, and log(λ) a mean annual population growth rate parameter. Prolonged trapping across years is in the present context (protected areas) deemed undesirable, as the sampling process itself can negatively impact local insect stocks. net effect) of the explanatory variables to the observed decline, both combined and independently. The investigations of the Entomological Society Krefeld and its members are spread over numerous individual projects at different locations and in different years. Each Ellenberg indicator (we considered nitrogen, pH, light, temperature and moisture) was averaged over all species for each location-year combination. σsite) and residual variation (v) decreased by 44.3 and 9.7% respectively, after incorporating an annual rate of decline into the models. Insect trap locations (yellow points) in Nordrhein-Westfalen (n = 57), Rheinland-Pfalz (n = 1) and Brandenburg (n = 5), as well as weather stations (crosses) used in the present analysis. Validation, Roles The reserves in which the traps were placed are of limited size in this typical fragmented West-European landscape, and almost all locations (94%) are enclosed by agricultural fields. Parameter convergence was assessed by the potential scale reduction factor [54] (commonly ), that measures the ratio of posterior distributions between independent MCM chains (in all models, all parameters attained values below 1.02). Weather effects explored were daily temperature, precipitation and wind speed, as well as the number of frost days and sum of precipitation in the preceding winter. Members of the Entomological Society Krefeld and cooperating botanists and entomologists that were involved in the empirical investigations are greatly acknowledged: U.W. Discover a faster, simpler path to publishing in a high-quality journal. Additionally, we are interested in being able to compare the residual variance with the random effects variance, and this requires them to be on the same scale. Roles Validation, As we were interested in whether the declines interact with local productivity, traps locations were pooled into 3 distinct habitat clusters, namely: nutrient-poor heathlands, sandy grassland, and dunes (habitat cluster 1, n = 19 locations, Fig 1A), nutrient-rich grasslands, margins and wasteland (habitat cluster 2, n = 41 locations, Fig 1B) and a third habitat cluster that included pioneer and shrub communities (n = 3 locations). Using the method of moments: To further characterize trap locations, we used past (1989–1994) and present (2012–2015) aerial photographs and quantified land use cover within 200m around the trap locations. Here, we investigate total aerial insect biomass between 1989 and 2016 across 96 unique location-year combinations in Germany, representative of Western European low-altitude nature protection areas embedded in a human-dominated landscape (S1 Fig). However, the data do permit an analysis at a higher spatial level, i.e. Yes Abts, F. Bahr, A. Bäumler, D. & H. Beutler, P. Birnbrich, U. Bosch, J. Buchner, F. Cassese, K. Cölln, A.W. https://doi.org/10.1371/journal.pone.0185809.s010.