Insect diversity in heterogeneous agro-environments of Central Europe

Insect diversity has been decreasing significantly during the past decades. A main driver causing this negative trend is agricultural intensification, which causes habitat destruction and a deterioration of habitat quality. We caught flying arthropods with 28 Malaise traps and two light traps across a heterogeneous Central European agro-environment in southern Germany over a period of four years. The arthropods captured with Malaise traps were assigned to barcode index numbers (BINs) based on metabarcoding. Lepidopterans caught with light traps and light towers were identified based on morphological characters. Within the four years study, a total of 11,984 insect operational taxonomic units (OTUs) related to unique BINs on BOLD (> 95% similarity) were recorded across this agro-environment. 7835 of them could be assigned to insect order and the vast majority of them represent different species. Different estimates of the total species richness ranged between 8188 and 11,512 OTUs (a proxy for species). This number corresponds to about 26–36% of the total German insect fauna. Light trapping also yielded a high species diversity and returned 502 Lepidoptera species (compared to 784 species captured with Malaise traps). The overlap between light and Malaise traps was less than 45%. The Malaise trap catches showed that insects are more diverse along forest edges than in the centre of meadows. Our findings underline the high value of heterogeneous agro-environments for biodiversity, and the high value when combining data collected with different sampling techniques.


Introduction
Agricultural intensification is a main driver causing global biodiversity loss (Maxwell et al. 2016).Demographic pressure and rising life-standards have led to an ever-increasing demand for resources and land.As a result, natural and semi-natural habitats have been converted into settlements and intensively used monocultures for the cultivation of food and energy crops.This trend caused the vanishing of biodiversity and negatively affects ecosystem functioning (Emmerson et al. 2016).Numerous studies evidenced that agricultural intensification causes the loss of insect diversity (Wagner 2020; Raven and Wagner 2021).Particularly landscape homogenization and pesticide spraying were found to impact insect diversity and insect abundance at the landscape level (Brühl and Zaller 2019;Sánchez-Bayo and Wyckhuys 2019;Seibold et al. 2019).The influx of pesticides and atmospheric nitrogen decrease habitat quality and has led to the gradual disappearance of arthropods (Habel et al. 2016;Hallmann et al. 2017).
However, heterogeneous agro-environments provide valuable habitats to many plant and animal species (Fahrig et al. 2011).Parts of the agricultural landscapes of Central and particularly Eastern Europe are characterized by comparatively high structural heterogeneity and extensively used semi-natural habitats (Culbert et al. 2017).Therefore, agro-environments may also harbour a large number of ecological niches and thus species (cf.Loos et al. 2014).
To study the diversity of arthropod species across a heterogeneous Central European agro-environment, we caught insects with Malaise traps and light traps over a period of four years.The collection of insects was performed across a habitat mosaic in southern Germany, consisting of conventional and organic farmland, forest, hedges and settlement areas.The traps were set in the centre of grassland sites and along forest margins.Arthropods caught with Malaise traps were subsequently identified with metabarcoding.Lepidopterans caught with light taps and light towers were identified based on morphological characters.Based on the data obtained, we will answer the following research questions: 1. Which proportion of the German insect fauna can be found in a heterogeneous agroenvironment of Central Europe? 2. Do species diversity caught with Malaise traps differ between sites, the centre of grasslands vs. the forest margin?3. Does lepidoptera diversity caught with Malaise traps coincide with lepidoptera diversity caught with light traps and light towers?4. Which conclusions can be drawn with respect to insect conservation in agro-environments and the use of different sampling techniques?

Study area
The study area is located near the city of Pfaffenhofen (southern Germany) and represents a heterogeneous landscape consisting of conventionally and organically farmed meadows and fields (ca.50%), forest (ca.35%), shrubs (ca.5%) and settlement areas (ca.10%).The landscape configuration is characterized by a comparatively high level of heterogeneity also due to its topographical diversity.Cattle grazing is practiced on the organic farmland sites.
Meadows are mowed twice a year without any application of pesticides (but with the use of organic fertilizers).Conventionally managed farmland mainly produce hay, silage and hops.These meadows are mowen several times a year and treated with artificial fertilizers.
Pesticides are applied to fields in the conventional farmland.

Sampling
We run in total 56 standard Malaise traps (B&S Entomological services) from April to October from 2018 to 2021 with in total 11 sites on conventional farmland, 19 sites on organic farmland, and 4 sites on meadows recently converted from conventional to organic treatment (here considered as `recently converted´).In 2018, we run two traps each on the organic and the conventional grasslands, one in the middle of a meadow and one near the forest edge.In 2019 and 2020 we run 8 and 10 traps, respectively on organic farmland, 4 and 4 traps on conventional, and 8 and 8 traps on recently converted farmland.In 2021 we run 4 traps on organic farmland, 3 traps on conventional farmland, and 3 traps on recently converted farmland.The traps were always set at least 150 m apart to minimize potential effects of autocorrelation (see also Steinke et al. 2021).
Due to the fact that all traps were placed in the same agricultural region within an area of less than 10 km 2 , non-independence of sampling (pseudoreplication) might be an issue.However, we were mainly interested in the estimation of richness.Observed richness differences between sites of different agricultural practices might be levelled out to a certain degree making the observed differences between farmland sites even more compelling.Furthermore, Steinke et al. (2021) reported that similarity in species composition between malaise traps on meadows dropped below 25% similarity above 30 m trap distance.Although similarities at higher distances were not reported, we can assume an exponential decrease of similarity (P) with sample distance (D) which results in the following decay function: P = e − 0.048D .The shortest distance between two malaise traps used in this study was 150 m.At this distance, the compositional similarity between traps would be (well) below 0.5%, thus making the traps spatially nearly independent.Furthermore, considering a trapping radius of half of this minimal distance (75 m), each malaise trap sampled an area of less than 0.02 km², which resulted in a total trapping area of 0.99 km 2 for all traps during the 4 years given assuming a faunal overlap of P = e − 0.048*75 =2.7%.Below we use this rough estimate of sampled area (1 km 2 ) as well as tenfold lower and upper margins of 0.1 km 2 and 10 km 2 to compare observed and estimated species richness with the expected richness extrapolated from the total insect richness of Germany.
Openings of the traps were facing southwards.The 600 ml sampling bottles were filled with 80% ethanol.All Malaise traps were emptied simultaneously on the same day.The material was stored in pure ethanol until DNA sequencing.We excluded non-insect arthropods as these were underrepresented in Malaise traps.A complete list of the raw data for each trap is given in Appendix A, Table A1.
Automated light traps (nonlethal traps without killing agent and a 12 V black light bulb) were activated from sunset to sunrise, 15 times in 2018, 9 times in 2019, 8 times in 2020, and 17 times in 2021 (between April and September), simultaneously at each of the Malaise trap sites.These light traps were activated from sunset to 01:00 am.In addition, we collected nocturnal moths actively at two light towers set in the study area.This was repeatedly conducted, 15 times in 2018, 9 times in 2019, 8 times in 2020, and 17 times in 2021, from sunset to 01:00 am.Each of the light towers was equipped with two 12 V/15 W superactinic bulbs.As killing agent, we used ethyl acetate.Collected vouchers were identified by T. Greifenstein, A. Hausmann and A.H. Segerer.If a classical species identification was not possible, we applied DNA barcoding (Canadian Centre of DNA Barcoding, CCDB).Data from both, light traps and the active collecting at the light towers were pooled afterwards.All raw data are contained in Appendix A, Table A2.

Metabarcoding
After drying the organic material of the Malaise traps, species identification was performed using DNA metabarcoding following the methodology described in Hausmann et al. (2020b).Each single dried sample was homogenized in a FastPrep96 machine (MP Biomedicals) using sterile steal beads in order to generate a homogeneous mixture of arthropods.Homogenized tissue samples were subsequently sent out for metabarcoding (conducted by AIM GmbH, Leipzig, Germany).Prior to DNA extraction, 1 mg of each homogenisate was weighed into sample vials and processed using adapted volumes of lysis buffer with the DNeasy 96 blood & tissue kit (Qiagen, Hilden, Germany) according to the manufacturer's protocol.The mitochondrial DNA barcode CO1-5P target region was amplified using a 313 bp long mini-barcode by PCR (Leray et al. 2013;Morinière et al. 2016), using forward and reverse HTS primers, equipped with complementary sites for the Illumina sequencing tails.In a subsequent PCR reaction, index primers with unique i5 and i7 inline tags and sequencing tails were used for amplification of indexed amplicons.Afterwards, equimolar amplicon pools were created and size selected using preparative gel electrophoresis.
Raw FASTQ files from Illumina were bioinformatically pre-processed using VSEARCH v2.9.1 (Rognes et al. 2016) (further details see Hausmann et al. 2020b).Briefly, paired-end reads were merged, and forward and reverse adapter sequences removed from each read.Reads that did not contain the appropriate adapter sequences were discarded.The resulting reads were dereplicated and those of short length and/or low quality were filtered out.Chimeric sequences were removed using the de novo algorithm.Finally, reads were clustered into operational taxonomic units (OTUs) using global pairwise alignment followed by de novo distance-based greedy clustering (at 98% pairwise identity) to the closest centroid sequence.Centroids, defined initially as the most abundant reads at the level of the entire dataset, were kept as the representatives of OTUs, and the resulting OTU FASTA file used as a reference database to create an OTU table of read counts in each sample.OTUs were blasted using Geneious (v.10.2.5 -Biomatters, Auckland -New Zealand) against (1) a custom, taxonomically annotated Animalia database downloaded from BOLD and (2) a local copy of the NCBI nucleotide database downloaded from ftp://ftp.ncbi.nlm.nih.gov/blast/db/(both downloaded on September 25, 2020).
Top BLAST hits for each OTU were exported from Geneious, combined with the OTU table produced by the pre-processing pipeline, and noise-filtered as described in Hausmann et al. (2020b).Interactive Krona charts were produced from the taxonomic information using KronaTools v1.3 (Ondov et al. 2011).Species identification in the Malaise trap samples was based on High Throughput Sequencing (HTS) data grouped to genetic clusters (OTUs), blasted and assigned to barcode index numbers ('BINs': Ratnasingham and Hebert 2013) which are considered to be a good proxy for species numbers.In our case, the detailed analysis of the Lepidoptera data revealed that the frequency of 'false positives' (0.5%) and BIN-sharing (1.5%) obstructing species discrimination (but nevertheless still pointing to species complexes) played a negligible role (see results for details).

Statistics
For each of the Malaise trap sample data, we used the numbers of different OTUs and total numbers of OTU reads as proxies for species richness and biomass (numbers of reads are proportional to abundance and to species body weight and population biomass is the product of abundance and average body weight), respectively (total of 13,230 OTUs).In the comparisons of species richness, we excluded the order Orthoptera (1246 OTUs), as some genera in this order are known to have increased intraspecific OTU diversity (Hawlitschek et al. 2016), which might bias the results when compared to classical taxonomic richness.
For the remaining 11,984 OTUs (referred to as S T ) we first used a conservative estimate to insect species richness and used only "BINs s.str.",i.e. all OTU-clusters exactly matching Barcode Index Numbers (BINs) on BOLD or referring to a nearest-neighboring BIN with a sequence similarity of p i >0.95 (below referred to as S A , total of 7835 "BINs s.str.",Table 1).From the S A values we subsequently obtained the expected richness (S E ) using first order jackknife estimates.
We next related observed (S A ) and estimated (S E ) BIN numbers with those expected from power function species (S) -area (A) relationships (SARs; S = S 0 A z with z being the slope and S 0 being the average richness at unit of area) assuming 0.01 km 2 , 0.1 km 2 , and 1 km 2 sample area and slopes of z = 0.05, z = 01, and z = 0.15, a range most often encounters at regional scales.From these assumptions, we obtained predictions of the range of expected local diversities (S 0 ) with which we compared our data.In turn, the observed local taxon richness in combination with the estimated German richness (Table 1) provided estimates for taxon SAR slopes, with which we compared the assumed slope range (0.05-0.15) for the whole German insect fauna Compositional differences between study sites and sampling years were assessed from read number (biomass) -rank order distributions (SADs).
The vast majority of the OTUs not assigned to species will have their own BINs when the reference library will be completed and the vast majority of these "BINs s.l." (further referred to as "BINs") should represent independent species.As a less conservative but pos-

Table 1
Estimated numbers of insect species in Germany (S G ) according to Völkl and Blick (2004) sibly more realistic estimate, we added the sequence similarities of all OTUs for each insect order to obtain a weighted estimate of possible order richness S P = i p i (all p i > 0.75) that was always higher than or equal to S A and S E and lower than the total OTU number of a focal insect order.We compared the observed (S A ) and expected (S E ) numbers of BINs between study sites and with the German insect species pool size as estimated by Völkl and Blick (2004).We compared light and Malaise trap samples using BINs of known species identity.We note that these values might not be fully comparable as BINs might also contain cryptic species not included in the compilations of classical morphospecies as provided by Völkl and Blick (2004) for the German insect fauna.Below we discuss this problem in detail.

Results
Of the S P = 11,512 insect OTUs, 3,387 (25.6%) could be assigned to taxonomic species (with Linnean species names on BOLD) and 6,985 (52.8%) to a genus (Appendix A, Table A1).We identified S A = 7,835 BINs that could be placed within an insect order.Of these most species were identified within the order of Diptera Diptera (S A = 3,547), within the order of Hymenoptera (2,245), Lepidoptera (784), and Coleoptera (727) (Table 1).First order Jackknife estimates pointed to 3,736 Diptera, 2,389 Hymenoptera, 789 Lepidoptera, and 744 Coleoptera S A OTUs (Table 1).
A comparison of the S A and S E numbers with the German insect fauna revealed that during four study years the present meadows contained at least 26% of the German insect fauna of more than 32,000 species (Table 1).Under the assumption that BINs are an appropriate proxy to taxonomic richness we found that of the larger orders, Diptera were represented by 41%, Neuroptera by 29%, Lepidoptera by 22%, and Hymenoptera by 26% of the German fauna.Comparatively underrepresented were Coleoptera (11%) and Hemiptera (16%) (Table 1).Possible species numbers S P on these meadow sites were much higher (S P = 11,512), at least for the larger orders and would represent 36% of the German fauna (Table 1).Diptera would have been represented by 60%, Hymenoptera by 38%, Neuroptera by 30% (Table 1).
During the four sampling years, the Malaise traps caught 784 species of Lepidoptera (Table 1).The light trap catches revealed 512 species.Altogether, we collected 899 species of Lepidoptera (see Appendix A, Table A2) covering 25% of the German fauna.Of these species 397 species were collected by both trap types.Our results imply that at least 502 species (55.8%) were trapped by only one method.
Calculations of power function SARs (species-area-relationships) indicated insect slope values between z = 0.01 (Raphidoptera) and z = 0.20 (Ephemeroptera), whereas slopes of the larger taxa varied between z = 0.06 (Diptera) and z = 0.14 (Coleoptera) (Fig. 1a).These SAR slope estimates were only marginally dependent on the assumed sample area (Fig. 1a).Richness of the majority of orders, particularly of the larger orders, and of the whole insect fauna did not significantly differ from the predicted German SAR, assuming SAR slopes between z = 0.05 and z = 0.15 (Fig. 1b).Of the smaller orders, particularly the water associated Ephemeroptera, Odonata, and Trichoptera, as well as the Thysanoptera and Psocoptera appeared to be less species rich than predicted (Fig. 1b; Table 1).Mecoptera and Rhaphidioptera occurred more species rich than expected by the SAR (Fig. 1b; Table 1).Of the larger orders, the Diptera appeared to be more specious and the Coleoptera less specious than expected (Fig. 1b; Table 1).
SADs at all study sites followed log-series sample distributions (Fig. 2).The three farmland types did not markedly differ in SAD shape and total richness, irrespective of insect order (Fig. 2a).Recorded diversity on the conventional meadows achieved between 71.6% (2021) and 93.9% (2019) of the organic meadows (Fig. 2a).From 2019 to 2021, 11.8%, 11.6%, and 11.4%, respectively, of the S A BINs were recorded on the conventional meadows only.In line with this finding, average Spearman correlation coefficients of species read number rank orders between study sites and years were low (r < 0.3) to intermediate (r < 0.5) (Fig. 2b) and explained on average less than 25% of variance (Fig. 2b).Rank order correlations within farmland types were always higher than between farmland types (Fig. 2b).Numbers of BINs in traps placed on the conventional farming sites were highest in only three of 21 taxon × study year combinations (Fig. 2c, d, e).Trapping on the organic farming sites scored highest 14 and on the recently transformed sites (from conventional into organic farming treatment) 6 times (Fig. 2c, d, e).Finally, the 2018 samples revealed that forest edges (S A = 2756) were more diverse than the meadows (S A = 2179, Table 1).Interestingly, the abundance patterns were reversed, being higher on the meadow (631,260 reads) than at the forest edge (479,406 reads).

Discussion
We detected a high number of insect species within a comparatively short period of time (4 vegetation periods) and for a rather restricted region.This insect diversity represents about 26 -36% of the total German insect fauna.This diversity is found in a comparatively smallscale and heterogeneous landscape, consisting of different ecosystems.Various studies showed that high landscape heterogeneity promotes high plant and animal diversity (Clough et al. 2005;Kovács-Hostyánszki et al. 2011;Rischen et al. 2021).Our study underlines, that heterogeneous agro-environments with also extensively used meadows and fields supports a large array of insect groups.
Our study area is dominated by agricultural use, whereby the field sizes are comparatively small, and thus the level of landscape homogenisation is still comparatively little.Numerous studies now show that agricultural intensification with large field plots and intense cultivation with artificial fertilisers and the use of pesticides cause rapid losses of biodiversity (Tilman et al. 2001;Raven and Wagner 2021).Thus, landscape homogenisation and the reduction of habitat quality are the main drivers of biodiversity loss (Benton et al. 2003;Fahrig et al. 2011), which takes place at the landscape level and is not restricted to single sites (see Seibold et al. 2019).In conclusion, despite the numerous disturbances in agroecosystems, these landscapes provide numerous ecological niches for a variety of species.
Our reference standard was the German insect fauna that is all taxonomic species irrespective of habitat.This fauna includes therefore all specialists of habitats, also species which occur in very specific habitats, such as bogs, heathlands, calcareous grassland or the alpine realm.And, OTUs cover the whole genetic diversity, including potential cryptic species.The fact that we could assign only 25.6% of the OTUs to taxonomic species and 52.8% to genera is an indication of the high hidden insect diversity.On the other hand we have to consider that the coverage of BOLD is still limited with about 50% of all known German insect species (Hausmann et al. 2020a).A higher coverage would surely reveal that additional OTUs can be assigned to taxonomically defined species.At present we are unable to assess how many of the 8,125 OTUs not assigned to species belong to this hidden diversity.Of course we also do not know how many cryptic species hide within the more than 32,000 German taxonomically defined insect species.Anyway, even if we would add cryptic species to this German species pool, our insect data collected reflect the situation of a heterogeneous agro-environment.
Species -area relationships are commonly applied to estimate and to compare species richness (Fattorini et al. 2021).Here, we used species -area relationships as auxiliary standard to assess whether the observed local richness was higher or lower than expected.Of course, such an approach can only give a rough impression, as species -area relationships are taxon specific, and generally do not have constant slopes over the whole range of area (Guilhaumon et al. 2008;Tjørve et al. 2021).To overcome these problems we used upper and lower boundaries of observed slope values for regional power functions (Rosenzweig 1995) to assess whether the observed local richness values are compatible with these models.Typical mainland regional insect species -area relationships follow power functions with slopes between 0.1 < z < 0.2 (e.g.Connor and McCoy 1979;Rosenzweig 1995).Indeed, the observed total insect richness was compatible with a comparatively low power function slope of z = 0.1 (Fig. 2).This result was largely independent of the assumed sample area of the traps.Any higher true slope would imply that the agricultural meadows under study were exceptionally species rich.Further, for Diptera, Hymenoptera, and Neuroptera, observed richness is compatible only with very low species -area relationship slopes of z < 0.1.Such low SAR slopes are not common among invertebrates.Only orders such as Trichoptera, Thysanoptera, Ephemeroptera, Odonata, and Psocoptera were less species rich than expected.Trichoptera, and Ephemeroptera, and to a lesser extent Odonata, are hygrophilous and were not expected to occur in high numbers in the Malaise traps set in meadows.
Our results do not allow for a precise assessment of insect abundances.Numbers of BIN reads are only an approximation to local population biomass, which correlates positively with abundance.Recently, Hallmann et al. (2017) reported strong declines in European insect abundances by as much as 2/3 during the past decades.Given that this decline affects all taxa proportionally we might expect species extinctions due to too low abundances.Such extinctions are expected from species accumulation curves (Scheiner 2003).Assuming a power function species accumulation (S = S 0 N z , where N is the number of individuals, S the associated richness and S 0 and z are the parameters).A reduction of individuals by 50% and a canonical slope value of z = 0.25 implies a reduction in diversity by p = 1 − S 1 S 2 = 1 − 0.5N N 0.25 = 16%.A slope of z = 0.10 implies a decrease in diversity of 7%.Unfortunately, the precision of our richness estimates does not allow for an assessment of species loss due to loss in abundance.Future studies have to link abundance and richness for a precise assessment using truly quantitative insect surveys.
We did not find a clear diversity trend with respect to agricultural management intensity of the respective sampling site.Prior work reported comparatively low biodiversity in conventionally managed agro-environments (Raven and Wagner 2021).However, this becomes mainly detectable when comparing two landscapes covered with either conventional or with organic farming.Our study area represents a patchwork of both.Differences in insect diversity are particularly dependent on local habitat structures.However, a site-specific descrip-tion of the species community would be better achieved with types of traps, such as pitfall traps.Malaise traps are known to primarily collect airborne plankton and thus catch mobile organisms that move on a landscape scale and thus strongly intermix.However, we found slight signals depending on the landscape and vegetation structure: High insect diversity was found in traps placed at the forest edge, regardless of whether the site was conventionally or organically managed (Table 1).This could be due to the linear structure along which winged insects often orient and move, and thus are more likely to accumulate there (Steinke et al. 2021;Hausmann et al. 2022).
Apart from the high diversity of lepidopterans (784 species) obtained from Malaise traps, we also caught high lepidopteran diversity (502 species) with light trapping.The overlap of species caught with light and Malaise traps was about 66%.This little similarity is because Malaise traps capture mainly diurnal arthropods like butterflies.In contrast, light traps and light towers actively attract species from far away and from other types of ecosystems (Habel et al. 2019); these techniques might also filter for specific which are getting attracted by light.Therefore, the light-trapped species are even more representative of the overall landscape situation, rather than the particular habitat on which the light traps and light towers are placed (Habel et al. 2019).The little overlap underlines the high relevance and value of using various sampling methods (here Malaise traps, light traps, light towers) in parallel for creating a holistic and realistic picture of the biodiversity of a specific landscape.
There are several weaknesses in the data collected that need to be taken into account when interpreting these results.(1) Sampling intensity and sites vary over the period considered.This makes it impossible to derive temporal trends from the results or to rely on a large sample size for individual sites.(2) The study area is characterised by high landscape heterogeneity and a strong mosaicking of organic and conventional agriculture.Thus, it can be assumed that insect diversity is largely mixed in the area with differences in land use intensity (e.g.organic vs. conventional), which are therefore not detectable.The selected trap technique is aimed at catching flight-active insects, which usually move in comparatively large spatial scales.As a result, potential signals of species diversity are also weak or fall below the detection limit.Despite these limitations, we can clearly show with the data collected that a heterogeneous agro-environments may provide a high level of diversity of species.Thus, Central European agricultural landscapes do have great potential for biodiversity conservation.
, columns 2-6: cumulative samples from all traps (meadows and forest edge, 2018-2021): total numbers of BINs (S T ), numbers of "BINs s.l." with OTU-cluster-hits > 0.95 (S A ) and total numbers of BINs s.l.(S P ) in the samples, first order jackknife estimates of BINs (S E ), and the proportions of S E / S G , S P / S G , and S E / S SAR (assuming a sample area of 1 km 2 and a SAR slope of 0.1).Column 7: samples from the 2018 trap placed at the forest edge.NN: OTUs with unassigned order on BOLD.Order

Fig. 2
Fig. 2 (a) Read number -rank order distributions (all years combined) on organic (green), transformed (blue), and conventional (red) fields.(b) respective mean Spearman correlations (r S ) of rank orders between (brown bars) and within (green bars) the three agricultural sites (conventional, organic, transformed).Errors denote standard errors, parametric significances of differences between within site and between site r s values: **: P(t-test) < 0.01.Total numbers of BINs on organic, transformed, and conventional farmland recorded in 2019 (c), 2020 (d),and 2021 (e).Colours as in (a)