Field site and experimental design
The Swiss site of the European BIODEPTH project is a former arable field overlaying calcareous nutrient-rich soil, situated at Lupsingen (47°27′ N, 7°41′ E, 439 m a.s.l.) in the Jura Mountains near Basel (Diemer et al. 1997). Soil analyses were carried out before establishment of the experiment (as described in Diemer et al. 1997) and yielded the following values (mean ± standard error): texture, loam; pH, 7.20 ± 0.04 (0.01-M CaCl2; pH meter 761, Knick, Berlin, Germany); total carbon, 3.74 ± 0.04 mg g−1 (CHNS-analyser; LECO-932, St. Joseph, Michigan, USA); extractable nitrate, 81.05 ± 0.77 µg g−1 (CaCl-solution 1 : 4); extractable phosphate, 1.61 ± 0.05 µg g−1 (saturated CO2-solution); extractable potassium, 5.83 ± 0.24 µg g−1 (saturated CO2-solution); extractable magnesium, 14.02 ± 0.51 µg g−1 (CO2-solution 1 : 10).
Two replicate blocks, each consisting of 32 plots, 8 × 2 m, separated by at least 1 m, were established in April 1995. A pool of 48 local grassland species belonging to 13 plant families was selected and used to assemble 32 mixtures of plant species at five diversity levels (Table 1).
Table 1. Species pool (a) and the 32 communities assembled from them (b). PSR = log2-transformed species number, each community is identified by a number and a lower-case letter. Nomenclature follows Binz & Heitz (1990)
|Agropyron repens||AgR||Anthyllis vulneraria||AV||Achillea millefolium||AM|
|Agrostis stolonifera||AgS||Lathyrus pratensis||LaP||Ajuga reptans||AjR|
|Agrostis tenuis||AT||Lotus corniculatus||LC||Anthriscus sylvestris||AnS|
|Alopecurus pratensis||AP||Medicago sativa||MS||Bellis perennis||BP|
|Anthoxanthum odoratum||AO||Onobrychis viciifolia||OV||Campanula patula||CP|
|Arrhenatherum elatius||AE||Trifolium pratense||TP||Centaurea jacea||CJ|
|Cynosurus cristatus||CC||Trifolium repens||TR||Centaurea scabiosa||CS|
|Dactylis glomerata||DG||Vicia cracca||VC||Crepis biennis||CB|
|Festuca ovina||FO|| || ||Daucus carota||DC|
|Festuca pratensis||FP|| || ||Galium verum||GV|
|Festuca rubra||FR|| || ||Geranium pratense||GP|
|Holcus lanatus||HL|| || ||Heracleum sphondylium||HS|
|Lolium perenne||LoP|| || ||Knautia arvensis||KA|
|Phleum pratense||PhP|| || ||Leucanthemum vulgare||LV|
|Poa pratensis||PoP|| || ||Pimpinella major||PM|
|Trisetum flavescens||TF|| || ||Plantago lanceolata Potentilla erecta Prunella vulgaris Ranunculus acris Salvia pratensis Sanguisorba officinalis Scabiosa columbaria Silene vulgaris Taraxacum officinale||PL PE PV RA SP SO SC SV TO|
Plant species number increased in geometric series from monocultures to mixtures of 32 species (Table 1). The log2-transformed plant species number is hereafter referred to as plant species richness or PSR. A number of different communities was assembled at each PSR level according to a restricted random sampling procedure from the species pool (Diemer et al. 1997; Joshi et al. 2000). All mixtures contained at least one grass, and within each level there were roughly equal numbers of mixtures containing only grasses, mixtures with one other functional group legumes or non-leguminous herbs, hereafter referred to as forbs, and where appropriate mixtures of all three functional groups. The number of functional groups (hereafter referred to as plant functional diversity, PFD) therefore ranged from 1 (in 1 to 8 species mixtures) to 3 (in 4–32 species mixtures) and was used as an additional treatment variable.
Each mixture was sown into one of the plots in each block in May 1995. Pilot germination studies were carried out and sowing densities adjusted so that for each community total seedling density was 500 m−2 and seedlings of all component species were initially present at equal frequencies.
Each plot was further subdivided into four subplots of 2 × 2 m of which two were subsequently sampled. Although one of these was subjected to trampling, disturbance had little effect and the two types of subplots were not therefore separated in the final analyses. All communities were mown twice a year as is typical for this type of extensively managed permanent grassland.
We wanted to simulate the consequences of biodiversity loss due to plant species extinctions, and therefore removed seedlings of any non-sown species as they appeared (Spehn, Joshi, Alphei, Schmid & Körner 2000; Spehn, Joshi, Schmid, Diemer & Körner 2000). Except at the highest diversity all initially sown species survived in all plots at the Swiss site, and PSR therefore remained constant. Although a few species did die out in the analysed portions of some 32-species plots, the numbers always remained closer to the initial value than to any other PSR level within the experiment (between 24 ± 1.8 in 1996 and 27 ± 1.7 species in 1997 per 2 × 2-m plot). An analysis of the effective species number (the exponential of the Shannon diversity index, see below, calculated from above-ground plant biomass proportions) within small 0.2 × 0.5 m subplots in the second year of the experiment showed a very strong congruence between designed and realized values (linear regression: R2 = 0.89, n = 64).
Plant diversity treatments affected both biomass allocation patterns (Spehn, Joshi, Schmid, Diemer & Körner 2000) and soil characteristics (Spehn, Joshi, Alphei, Schmid & Körner 2000). Above-ground biomass increased by 143% from the lowest to the highest diversity in 1997 and, although total below-ground plant biomass was not affected, that of fine roots also increased significantly. Soil moisture during the growing season was not influenced by diversity treatments, but soil temperature decreased slightly with increasing diversity. The slight increase in substrate-induced respiration suggested that soil microbial biomass may increase with increasing plant diversity, and although in-situ decomposition of cellulose and birch-wood was not affected, both numbers and biomass of earthworms were strongly positively correlated with diversity. The presence of legumes in the experimental plant communities often had significant effects on the activity of soil fauna.
Over a 4-day period in the third week of August in each of 1997 and 1998, we took two soil samples of 100 ± 20 mg bulk soil (no rhizosphere, no living or dead root parts) at a depth of 3 cm in each of the 64 plots (except that in 1997, accidentally, four plots were not sampled and four other plots were sampled twice). Preliminary sampling in three of the 64 plots had shown that the highest activity of culturable soil bacteria occurred at a depth of 2 cm and the second highest at a depth of 4 cm. In 1997, the two samples were taken in the same (undisturbed) subplot at points separated by 2 m, whereas in 1998, one sample was taken in each subplot. The soil samples were placed directly into 2.5-mL Eppendorf tubes in the field and were thereafter treated blindly.
Extraction and incubation of bacteria
Soil bacteria were extracted within 6 h after sampling. After shaking (Vortex, full speed) for 20 min in 1 mL 0.2% Tetrasodium-pyrophosphate, the samples were allowed to settle for 3 min. An aliquot (150 µL) of the supernatant was then diluted 100-fold with 0.9% NaCl before 100 µL were transferred to each well of the BIOLOG Ecoplate (BIOLOG Inc., Hayward, CA, USA).
Each of the 96 wells of an Ecoplate contained dried nutrient solutions, containing a single carbon compound, and a redox-colourant (tetrazolium violet) (three replicate wells for each of 31 C sources and no-carbon control). When culturable bacteria grow, the oxidation of the C source forms NADH which can be quantified by its reduction of the colourant. The microtiter plates were incubated at 22 °C and read with a spectrophotometer (Microplate Reader 3550, BIO-RAD, Hercules, CA, USA) at 590 nm. Absorbance values were recorded after 72 and 120 h in 1997, but as some of the latter exceeded the maximum recordable, we decided to shorten the incubation times in 1998 to 48, 72 and 96 h. We present only data after 72 h: analyses using other incubation times did not give different or further information.
We took the mean of the three replicate wells per plate containing a particular substrate and subtracted the absorbance value for the control on the same individual plate to calculate the catabolic activity in the use of that C source (Ai). Negative Ai values were set to zero. Overall catabolic activity was calculated as the sum of activities for all 31 C sources and, if divided by the number of C sources, is equivalent to the AWCD (average well colour development) calculated by Garland (1996). For every sample, Ai values were used to calculate richness S, the Simpson index D′ (the reciprocal of the dominance index), Shannon index of diversity H′, equitability of D′ and equitability of H′. We used the formulae given in Begon et al. (1990):
Richness simply represents a numerical measure of the diversity of C sources that are utilized by a sample. The Simpson and Shannon indices are combined measures of the ‘number’ and ‘abundance’ aspects of this diversity, but ‘number’ can be removed by dividing by richness or the natural logarithm of richness, respectively. Equitabilities therefore reflect only the ‘abundance’ aspect of the diversity in C source utilization: equitability reaches a maximum if C sources that can be used have the same Ai values, whereas low values indicate that one or a few C sources have much higher Ai values than any others.
The general linear model (GLM) approach to analysis of variance (anova) was used to analyse the data by means of the Genstat 5 software (release 3; Payne et al. 1993). The split-plot design involved a block and plot factor and treatment factors as in Table 2 (a similar approach to that used by Meyer & Schmid 1999), to allow calculation of F-values for significance tests. Block and plot effects were used to eliminate variation caused by spatial differences within the experimental site. Sample mass was used as a covariate in both years; fine root length per plot and fine root biomass per plot (data from Spehn, Joshi, Alphei, Schmid & Körner 2000) were included as further covariates in a separate anova of 1997 data (results not presented).
Table 2. Dummy or skeleton analysis of variance for all characters measured (see text for further details)
|Source of variation||d.f.||Mean square||Variance-ratio|
|Covariate(s) (cov)||1 (n)1||MScov||MScov/MSr|
| Sample mass (w)||1||MSw||MSw/MSr|
|Plot total (pt)||63||MSpt||MSpt/MSr|
| Block (b)||1||MSb||MSb/MSp|
| Mixtures (m)||31||MSm||MSm/MSp|
| Diversity treatments (d)||2||MSd||MSd/MSmi|
| Species richness loglinear (s)||1||MSs||MSs/MSmi|
| No. functional groups linear (f)||1||MSf||MSf/MSmi|
| Mixture identities (mi)||29||MSmi||MSmi/MSp|
| Taxon 1 (t1)||1||MSt1||MSt1/MSp|
| Taxon 2 (t2)||12||MSt2||MSt2/MSp|
| Deviation (mid)||272||MSmid||MSmid/MSp|
| Plot (p)||31||MSp||MSp/MSr|
|Year × plot total||573||MSy.pt||MSy.pt/MSr|
|Residual (r)||1263||MSr|| |
|Total (t)||2483||MSt|| |
Each level of PSR was represented by several communities, and the effect of PSR itself therefore could be tested against the variation among communities within PSR levels, to determine the effect of species number per se, unconfounded by any particular species composition occurring at a particular PSR level (see Hector et al. 1999). PSR levels could be ordered along a continuous axis, with a logarithmic scale giving better fits than the untransformed plant species number, allowing us to test for significant linear contrasts of PSR and the deviation from linearity. Parsimony, the small deviations from linearity and previous reports that biodiversity effects are generally linear at the logarithmic scale (Hector et al. 1999; Schmid et al. 2001) led us to prefer this method to polynomial contrasts using untransformed species number. Deviation from linearity was non-significant and small enough for it to be omitted in all final analyses of the effects of both PSR and functional diversity (PFD) within PSR (i.e. ‘eliminating PSR’, see Payne et al. 1993).
Despite allocating large amounts of time and labour to setting up and running the experiment, it was only possible to include 32 plant communities. P < 0.1 was therefore regarded as a (‘marginal’) significance level in testing the effects of PSR and PFD to reduce the risk of making type-II errors (i.e. not rejecting the null hypothesis if in fact it is false). In addition to the P-values for PSR and PFD, we report partial coefficients of determination (R2) as a measure of effect size (Cohen 1977). These coefficients measure the proportion of the variation in a response that is explained by an independent variable in a GLM. They were calculated according to Rosenthal & Rosnow (1985) as
The variation among communities within a particular PSR and PFD level could be tested for significance at the plot level, because each particular community occurred exactly once in each of the two blocks. The factor year was used to test for temporal variation in the response variables.
The full anova was done for all response variables. Analyses of individual Ai values are not presented as these values were well reflected by the overall catabolic activity. The effect of PSR, the factor in which we are particularly interested, is presented as a linear regression of the Ai adjusted for the covariate sample mass. The analyses of individual and overall activity and of diversity values used untransformed data because residuals were normally distributed and homoscedasticity was not improved by transformation.