Appendix B. Details on site selection, arthropod sampling, vegetation structure descriptors, environmental measurements, and data analysis.
B-1 Investigated sites: selection criteria
Preferably, study sites were selected well away from the road surface (verge width ranged up to more than 100 m in some sites; median width was 13 m). To qualify as a study site, we required plant species composition, vegetation structure and flower richness to appear visually uniform (unchanging) over a minimum length of 50 m parallel to the road and a width of at least 3–4 m. Sampling of local factors was confined to a 25 × 2 m plot at the centre of the selected uniform area.
B-2 Arthropod sampling techniques
As the 47 study sites were scattered over the country, all sites could not be visited on the same day. A single sampling round typically took 6–10 days, depending on sampling method and weather conditions. Below, each of the various techniques used for sampling arthropods is described in detail. We also discuss the possible impact sampling may have had.
Pitfall traps
As pitfall traps we used plastic containers with a diameter of 9 cm and a height of 12 cm. The containers were placed in the ground with their rim level to the ground surface. To avoid detection and disturbance by public, a green metal cap was placed over each pitfall, 1–1.5 cm above ground level. These caps also acted as a rain shelter.
The containers were filled up to one quarter with a 4% formaldehyde-solution as a preservative and a drop of detergent added as defractant. Because of the preservative and the rain shelter the traps only had to be emptied every five or six weeks. When emptied the formaldehyde solution was replaced.
At each site we used five plastic containers positioned along a straight line at 5-m intervals. The pitfalls were applied continuously from March to November 1998, and June to November 1999 (for spiders only the catches from 1999 were used).
Sweep nets
To be able to sample efficiently also close to the soil surface, we used sweep nets with a pentagonal shape (39.5 cm wide at the lower side and 35 cm high). These dimensions correspond to the opening of a circular net with diameter 37.5 cm.
We swept each plot at four different times of year in both 1998 and 1999. We sampled on clear days between 10:00 and 17:00 h when the vegetation was dry. Sampling consisted of sweeping a transect 25 m long and 2 m wide, while slowly walking the length of the transect. The method was mainly applied to collect weevils and hoppers, yet all ground beetles, bees and hoverflies captured were identified as well.
Sweep net sampling can severely damage tall vegetations and might subsequently influence its species composition. In managed roadside grassland, however, this problem only occurs in summer. In spring, and also in autumn after mowing, the vegetation is too short to be significantly damaged. In our study we further minimized effects by timing the summer sampling with regard to the regular mowing regime. Productive vegetations in roadside verges are generally mown twice, usually no later than the end of June for the first cut and early September for the second. Our summer sampling therefore took place in mid June and near the end of August, about one or two weeks before the vegetation would be mown anyway.
In less productive roadside vegetation the June cutting is generally omitted but because of the low productivity, vegetation is low and disturbance by sampling is limited also in June. Pioneer and heath vegetation are often not managed or mown only once every few years, but since pioneer vegetation is very low and heath vegetation very strong, sampling disturbance is always small in these cases.
Sound observations
The Dutch Orthoptera fauna in natural and semi-natural habitats only comprises 41 species. Males of 32 species can easily be identified by their song and these species all sing at least partly during daytime. As morphological identification is often very difficult, sound observations are an accurate way to record the presence and abundance of most Dutch Orthoptera species (Kleukers et al. 1997). Species abundances were estimated in 5 density classes, approximately corresponding to 0, 1–10, 10–25, 25–100 and >100 individuals.
This method was applied during various visits throughout the year in both 1998 and 1999. Sound observations took place for about 20–30 min. between 10:00 and 17:00 h on days with less than 50% cloud cover and a minimal temperature of 17˚C. For non-singing species (Tetrigidae, Oedipoda) and more or less inaudible species (some Tettigoniidae) visual sampling was applied.
Visual sampling and hand-catches
Bees, hoverflies, and some grasshoppers (Tetrigidae, Oedipoda, and a few Tettigoniidae) were recorded after identification by sight or collected with an insect net, killed in a plastic container with ethylacetate and identified later using a binocular microscope. This method was applied during various visits throughout the year in both 1999 and 2000. Each site was sampled for about 20–30 min per visit, between 10:00 and 17:00 h on days with less than 50% cloud cover and a minimal temperature of 12˚C in spring and 17˚C in summer. The observation rounds were combined with the sound sampling of grasshoppers as much as possible.
By catching and recording as many species as possible this method was initially intended to be qualitative. Yet a fair proportion of the bee and hoverfly species could not be identified in the field and were collected for later identification, thus yielding quantitative information. Additional quantitative information was obtained from yellow pans (see below) and sweep net catches, also for the species that were not hand-caught. Nevertheless, the data table for bees and hoverflies still remains somewhat unbalanced between species. For some species the data represents the actual number of individuals collected at a site, for others it tends to be biased towards the number of visits a species was observed (visually) at a site.
As any arthropod group knows differences in sampling efficiency between species and only bias between sites would harm our analyses, this partial unbalance was not considered a problem. Explorative analyses for bees and hoverflies also showed that using only presence/ absence data (qualitative approach) yielded lower predictabilities than when utilizing the quantitative information available. This further confirmed the quantitative data to contain analyzable and interpretable information.
Yellow pans
Colored pan traps attract bees and hoverflies because of their bright color. We used this method in addition to visual sampling and hand-catches. The latter were applied during short visits only and appeared to be less effective at sites with low bee and hoverfly densities. So far, there is no general consensus about the most effective trap color (e.g., Toler et al. 2005). We chose white as this was expected to attract the least public attention.
As pan traps we used cups (white) with a diameter of 9 cm and a depth of 12 cm. In May/June and again in August 2000 three cups were set out at each site when fine weather was predicted for a prolonged period. The cups were partly filled with a NaCl-solution as a preservative and a drop of detergent as defractant. After two weeks the traps were emptied and removed.
Possible effects on vegetation and soil
Repetitive sampling inevitably leads to some disturbance of the investigated sites. Trampling and the sampling itself can damage the vegetation and can lead to soil compaction. By not exaggerating our sampling frequency, combining visits for different sampling techniques, and careful timing of sweep net sampling (the most damaging method), we tried to minimize these effects. According to our observations the only obvious, more or less permanent damage occurred near pitfall traps in plant communities with a low management frequency (heathland, Spergulo-Corynephoretum). However, the areas involved were so small that structural shifts in species composition are unlikely. Also supplementary inspection of our data for possible increased appearance of atypical species (using ecological species knowledge) gave no indication that shifts in species composition have occurred in these or any of the other plant communities.
B-3 Vegetation structure: descriptors
From the digitized analysis of vertical vegetation profiles at peak standing crop (see main text) we derived the following descriptors of vegetation structure (averaged over the two profiles per plot): maximum plant height (labeled: Peak Height), total vertical cover over all 10 cm height layers (Peak Density; assumed proportional to total biomass, see Bobbink 1991), the fraction of vegetation height below which 50% of the biomass is located (HalfDensity Level), and an index representing the general shape of the height-density relationship (from convex to almost linear to concave: Vert. DensityShape).
The data from the drop-disk technique (100 measurements in each of two sub-plots measuring 1 × 4 m using a 20 × 20 cm grid) was treated as follows. Groups of four adjacent grid measurements were combined to represent "patches" of vegetation, with the average height for a patch referred to as its characteristic height. The characteristic height for the site as a whole was calculated as the average of patch-heights. The spatial variation in height was calculated as the coefficient of variation in patch-heights (averaged over the two sub-plots).
From the drop-disk data obtained during five visits (from April to November, every five to eight weeks) the following variables on horizontal and temporal variation in vegetation height were derived: maximum seasonal height (Temp. MaxHeight: largest site-height encountered during the five visits), mean seasonal height (Temp. MeanHeight: average site-height over the five visits), the temporal variation (Temp. VarHeight: coefficient of variation of site-heights over the visits), the ratio between mean and maximum seasonal height (Temp. RatioHeight: roughly indicating whether vegetation height is near its peak value during most of the season or during a short period only), and the maximum spatial variation (Spatial VarHeight: maximum over the 5 visits of the coefficient of variation in patch-height).
Horizontal cover percentages were also estimated, not only for individual vegetation layers: Herb Cover, Moss Cover, Litter Cover, and Open Cover (amount of open ground), but also for the total vegetation (Herb+Moss Cover), for open ground with litter included (Open+Litter Cover), and for open ground with both litter and mosses included (Open+Litter+Moss).
B-4 Environmental measurements
Soil sampling took place in March 2001, using bulked samples consisting of five subsamples (Ø 25 mm) taken in a random pattern. All samples were taken from the top 10 cm of the soil. Prior to sampling the loose litter layer was removed. Samples were kept cool during transport, dried at 40 °C, and sieved using a 1-mm mesh. Material remaining in the sieve was weighed so that results could be adjusted to the complete sample-size. Bulk densities were measured by taking five random samples of exactly 0.2 l undisturbed soil using a specially designed auger 10-cm long. Chemical results were expressed on a volume basis (amount ha-1 [0–10 cm depth]).
Dried samples were extracted using 0.01M CaCl2, which is a weak extraction solution approaching the average concentration of many soil solutions (Houba et al. 1986, 1990, 1996). Extraction was performed using 3 g soil in 30 mL of extraction solution by shaking for two hours, followed by centrifuging for 10 minutes at 3000 g.
The centrifugate was used for spectrophotometric determination of NO3-N, NH4-N, PO4-P by a Segmented Flow Analyzer (Skalar, Breda, the Netherlands), and for determination of K using an Atomic Absorption Spectrophotometer (AAS). Mineral N was taken to be the sum of NO3-N and NH4-N. The fraction of mineral N occurring as NO3 was labeled the nitrification degree (not to be confused with the actual rate of the nitrification process).
Soil pH measurements took place in the settling suspension of the CaCl2 extracts, before centrifuging (Houba et al. 1994, Schofield and Taylor 1955). The resulting pH-CaCl2 usually takes values approximately halfway between pH-H20 and pH-KCl (Gupta and Rorison 1975).
Sand content was determined using a 65µ sieve and was expressed as the percentage of all material larger than 65µ, relative to the to total weight of the mineral soil fraction (including any possible CaCO3). Organic matter content was measured by weight-loss on ignition at 550 °C.
Soil moisture content was determined not only in March (just before the start of the growing season) but again in July (mid-summer). It was expressed gravimetrically after drying at 40 °C.
Two non-edaphic site conditions were used in this study, reflecting consequences of site exposition and inclination. We derived a relative index of the amount of total radiation in summer and a relative index of site "temperature" in summer. Estimations were based on a slope-aspect diagram for July (Grime and Lloyd 1973) with highest radiation values necessarily occurring on south sloping sites (180°). The temperature index was estimated in a similar fashion, but assumed highest temperatures to occur at expositions of 215° (indicating slopes directly facing the sun 2 to 2½ hours after midday).
B-5 Data analysis
Variable selection in predictive CCA
Whenever computationally possible (16 predictor variables or less) variable selection was performed by checking all possible subsets of variables. For larger predictor sets variable selection was implemented using a recursive algorithm, calculating the fit for all possible subsets with one variable removed and those with one variable added, and repeating this process from the best subset encountered until a better performing set could not be found. By running this recursive procedure twice, once starting from an empty subset and once starting with all variables, we further improved the chance of finding the best predicting subset of variables and so maximized prediction levels of the predictor data under analysis.
Randomization test for prediction differences
To test whether the difference in cross-validatory fit between two models is significant, we used a simple randomization test (van der Voet 1994). This procedure uses as a test statistic the difference between the mean squared prediction errors of the two models (for each model this mean is directly proportional to the cross-validatory fit). For a large number of evaluation cases (in our study n = 999), site prediction errors are rearranged by swapping the prediction errors for random sites (a different random set each time) between the two models. By calculating the proportion of cases where the test statistic turns out to be higher than or equal to the actually observed value (after including the actual value itself) we obtain the probability that the observed difference between the two models could also have been be due to random variation.
As a consequence of this method, the same amount of difference in prediction that is significant for one pair of models, need not turn out to be significant for another pair. Much also depends on the combined distributions (often skewed) of the prediction errors of the models.
LITERATURE CITED
Bobbink, R. 1991. Effects of nutrient enrichment in Dutch chalk grassland. Journal of Applied Ecology 28:28–41.
Grime, J. P., and P. S. Lloyd. 1973. An Ecological Atlas of Grassland Plants. Arnold, London, UK.
Gupta, P. L., and I. H. Rorison. 1975. Seasonal differences in the availability of nutrients down a podzolic profile. Journal of Ecology 63:521–534.
Houba, V. J. G., Th. M. Lexmond, I. Novozamsky, and J. J. van der Lee. 1996. State of the art and future developments of soil analysis for bioavailability assessment. The Science of the Total Environment 178:21–28.
Houba, V. J. G., I. Novozamsky, A.W. M. Huybregts, and J. J. van der Lee. 1986. Comparison of soil extractions by 0.01 M CaCl2, by EUF and by some conventional extraction procedures. Plant and Soil 96:433–437.
Houba, V. J. G., I. Novozamsky, Th. M. Lexmond, and J .J. van der Lee. 1990. Applicability of 0.01 M CaCl2 as a single extraction solution for the assessment of the nutrient status of soils and other diagnostic purposes. Communications in Soil Science and Plant Analysis 21:2281–2290.
Kleukers, R. M. J. C., E. J. van Nieukerken, B. Odé, L. P. M. Willemse, and W. K. R. E. van Wingerden. 1997. De sprinkhanen en krekels van Nederland (Orthoptera). (Grasshoppers and Crickets of the Netherlands). Nationaal Natuurhistorisch Museum, KNNV-uitgeverij & EIS-Nederland, Leiden, The Netherlands.
Schofield, R. K., and A.W. Taylor. 1955. The measurement of soil pH. Soil Science Society of America Proceedings 19:164–167.
Toler, T. R., E. W. Evans, and V. J. Tepedino. 2005. Pan-trapping for bees (Hymenoptera: Apiformes) in Utah's west desert: the importance of color diversity. Pan Pacific Entomology 81:103–113.
van der Voet, H. 1994. Comparing the predictive accuracy of models using a simple randomization test. Chemometrics and Intelligent Laboratory Systems 25:313–323.