Description of study site
This study was performed in 2010 as part of the “GrassMan”-Experiment[13] near Neuhaus (Solling) in the Solling Mountains in Northern Germany (51°440 N, 9°320 E, 490 m a.s.l.).
Prior to the start of the experiment the study site was a nutrient poor, moderately wet Lolio-Cynosuretum grassland with high abundances of Agrostis capillaris (L.), Festuca rubra (L.), Rumex acetosa (L.), Veronica chamaedrys (L.) and Ranunculus repens (L.)[13].
Mean annual precipitation is 1028 mm and mean annual temperature is 6.9°C (Deutscher Wetterdienst, 1961–1990, station Holzminden-Silberborn, 440 m a.s.l.). In 2010, the year of the study, mean annual temperature was 8.0°C and annual precipitation was 1110 mm. The dominant soil type in the experimental area is a shallow (40–60 cm), stony Haplic Cambisol, developed on sediments of loess on the Middle Bunter (Triassic sandstone) formation with a loamy silt texture[41].
Study design
The experiment was established in 2008 in a permanent, formerly extensively used, cattle-grazed grassland. It was laid out as a three-factorial Latin rectangle[42] with the following factors (Additional file5: Figure S1): (i) plant functional group manipulation (three levels) using herbicides, (ii) fertilizer application (two levels) and (iii) cutting frequency (two levels), resulting in twelve treatment combinations.
To manipulate plant functional group presence, we applied (i) a combination of the forb-specific herbicides Fluroxypyr (Starane; Dow AgroSciences, Munich, Germany; 3 L ha-1) and Mecoprop-P (Duplosan; KV, Du Pont de Nemours, Neu-Isenburg, Germany; 3 L ha-1) or (ii) the graminoid-specific herbicide Select 240 EC (Stähler Int., Stade, Germany; 0.5 L ha-1), resulting in three levels of plant diversity: (i) forb-reduced (=graminoid-rich), (ii) graminoid-reduced (=forb-rich) and (iii) control. Herbicides were applied once in June 2008 (a “pulse” experiment sensu Bender et al.[43]).
In 2009 and 2010, plots were fertilized with N (Calcium ammonium nitrate N27: 13.5% NH4-N, 13.5% NO3-N, 4% MgO, 6% Ca) at two equal doses (2 × 90 kg ha-1) in April and May/June; in addition, fertilized plots received 30 kg P ha-1 and 100 kg K ha-1 in early June (Thomaskali®, 8% P2O5, 15% K2O, 20% CaO).
Control plots were not fertilized. Plots were cut either once (in July) or three times a year (May, July, September) using a Haldrup® forage combine harvester (INOTEC Engineering GmbH, Ilshofen, Germany) at a cutting height of 7 cm. The resulting twelve treatment combinations (equalling one block of the Latin rectangle; see Everwand et al.[10], Figure 4 and Additional file5: Figure S1) were arranged randomly and replicated six times, resulting in 72 plots. Each plot was a 15 × 15 m square surrounded by at least 3 m of frequently cut grass between plots, and 5 m between blocks.
Plant functional groups were not entirely removed, but target plant species were strongly reduced in abundance. Plant functional groups slowly recovered following herbicide application, but all FG manipulation treatments significantly affected vegetation parameters, such as compressed vegetation height, harvested biomass, functional group composition and plant species richness. More details on the experimental design, setup and treatment effects on vegetation can be found in Petersen et al.[13, 44] and Rose et al.[23, 45, 46].
Leafhopper sampling
Leafhoppers were sampled using two methods: i) by sweep netting (Heavy Duty Sweep Net, 7215HS, BioQuip, diameter: 38 cm), while walking a circular transect with a diameter of 8 m around the centre of each plot (30 sweeps each), in dry weather on two occasions (at the beginning of July and at the end of August 2010). Transects length was approximately 20 m, and there was a distance of at least 4 m to the edge of each plot. In addition, ii) we sampled leafhoppers by placing two transparent pan traps, containing an ethylene glycol/water mixture (1:3), 1 m apart, near to the centre of each plot. Pan traps were about 5 cm above vegetation height and were active for one week in five time intervals in 2010 (end of June, mid-July, early August, mid-August, end of August).
The specimens caught with both methods were transferred into ethanol (70% vol.) separately and identified to species level in the laboratory using Biedermann & Niedringhaus[47] and Kunz et al.[48]. One species with woody host plants was excluded, as we assumed that it had been swept off its host tree by wind and was not a true member of the grassland fauna. Species whose larvae used herbs or grasses as host plants and whose imagines fed on trees were also included in the analysis.
For female specimens of several genera, identification to species level is not possible (e.g. Psammotettix)[47, 48]. Thus, if male specimens were present, female specimens were assumed to belong to the same species. If not, they were only identified to genus level. If males of more than one species of a genus were present, the proportion of females was assumed to mirror that of males.
We found no interaction effects of the two sampling methods with the management variables (cutting frequency, fertilizer application) on leafhopper species richness (see e.g. Figure 4, Additional file6: Table S1 and Additional file7: Figure S2). In addition, vegetation height (a proxy for vegetation density) did not affect the number of leafhoppers caught by sweep netting.
We therefore pooled the data of both methods, which allowed us to cover the growing season of 2010 from early May until late September. For all diversity assessments, we used species richness, Shannon’s diversity index (H’) or its numbers equivalent exp(H’)[49].
Assessment of vegetation parameters
Because our treatments are likely to have affected plant productivity and vegetation structure, possibly indirectly affecting leafhopper species richness, we additionally measured a series of vegetation parameters:
(i) We conducted vegetation surveys on two quadrates, each of 1 m2 size, twice (in May before the first harvest and also again in August) on each plot. We recorded the percentage of cover, proportional yield of each species [50], plant species richness, functional group composition and presence-absence data of the functional groups (graminoids and forbs).
(ii) Plant aboveground biomass (AGB) was estimated as follows: First, during harvest, fresh weight of two 1.50 × 15 m strips per plot was measured using the harvester’s built-in scale.
To determine the water content of this sample, we took four subsamples that were homogenized, weighed, dried for 48 h at 65°C and subsequently weighed again. We then multiplied fresh weight by water content to obtain the total aboveground dry biomass (t ha-1) for every plot.
(iii) Proportions of graminoids and forbs (%) were determined from the vegetation surveys (derived as described above). Harvest was performed on all plots once a year in the end of June and additionally in mid-May and mid-September for the 3-cut treatment [13].
(iv)Compressed sward height (cm) was measured using a rising plate meter according to Castle [51] and the average value of 25 measures per plot was calculated. This was performed every 2-3 weeks, resulting in eleven time points throughout the growing season of 2010.
Statistical analysis
Data were analysed using the statistical software package R (version 2.15.2)[52]. In addition, we performed structural equation modelling using AMOS 20.0 (SPSS, Inc.). Treatment effects on vegetation and leafhoppers were assessed using generalized linear models (GLMs;[53]).
Models contained row- and column effects (fitted as factors, column was nested within block), sward composition (factor with three levels), cutting frequency and nutrient input (two levels each) with up to two-way-interactions. For abundance data (e.g. Table 4) we used quasipoisson GLMs, for proportion data LMs with a logit link[54, 55] and for exp(H’) we used GLMs with Gamma errors and an inverse link. Corresponding alternative models (e.g. quasipoisson or Gamma with log link) had higher residual deviance and were therefore not considered.
Continuous response variables (e.g. biomass or vegetation height) were log-transformed and analyzed using GLMs with an identity link. For each response variable in turn, maximal models containing all possible terms were manually simplified into models containing fewer explanatory variables. We compared the resulting nested models using F-tests (and Chi2 for quasipoisson models), until a minimal adequate model that only contained significant effects was obtained.
Significance of terms was assessed in two ways: (i) each parameter estimate from linear models was compared to zero using marginal t-tests; and (ii) terms in the models were additionally tested by sequential addition to a null model (sequential analysis of deviance tables; Additional file8: Table S3).
In addition to traditional GLM-based analyses, we employed structural equation models (SEMs), allowing us to test more complex hypotheses on indirect effects of treatments, plant productivity and plant diversity on leafhoppers[56–58]. SEMs are particularly well suited for experimental contexts, i.e. where some variables are deliberately manipulated experimentally[59]. Furthermore, SEMs “can be used to develop accurate and meaningful final multiple regression models when collinearities among explanatory variables are thought to be present”[60], which was clearly the case for the vegetation properties measured here.
SEMs contained all three treatment variables, as well as latent variables[56] for plant productivity and plant diversity. For the SEMs we specified our design variables as numeric variables as follows:
Fertilizer treatment: no fertilizer = 0; NPK-fertilizer application = 1
Cutting frequency: one cut/year = 0; three cuts/year = 1
FG manipulation: FG graminoid-reduced = -1; FG control = 0; FG forb-reduced = 1
The sorting of FG manipulation was according to its effect on plant diversity and proportion of graminoids (see Figure 1). Plant productivity had two indicator variables: harvested aboveground biomass in July (AGB, t ha-1), and average compressed sward height. Plant diversity had the indicator variables “forbs” and “graminoids”; since only four legumes species (Lotus corniculatus, numeric variables as followsL. pedunculatus , Trifolium repens, Lathyrus pratensis) were present in a very low cover on 61 plots only, and none of the leafhopper species found had been categorized as preferentially feeding on legumes, we did not take legumes into account separately for the SEMs. Leafhopper abundance and species richness were taken separately (instead of (eH`) Shannon diversity) for the SEM to identify effects of design variables and vegetation parameters on leafhoppers.