Sites and plot selection and vegetation surveys
Sampling was conducted in the Westland region of the South Island of New Zealand between 42°29′ S and 43°21′ S and between 169°52′ E and 171°23′ E. The region has a wet temperate climate with an annual average temperature of 11 °C (Hessell 1982) and an average annual rainfall ranging from 3.5 m to 8.0 m. There is considerable spatial variability of basal substrate in this system; the main substrate is schist derived from glacial outwashes, other basal substrates also commonly occur including greywacke, granite and limestone. Landform or topography is also highly variable and is a major driver of soil nutrient status in these types of forests (Richardson, Allen & Doherty 2008). Vegetation was sampled from 32 unbounded plots, each between 0.25 and 0.5 ha, and located within a uniform vegetation type, geological substrate and landform. All plots were situated within broadleaf-conifer temperate rain forests that are characterized by a mixed canopy of evergreen angiosperms and emergent podocarps (Wardle 1991), with plots selected to represent the range of dominant basal soil substrates and landforms occurring in the region. For each of four basal soil substrates, namely greywacke, schist, granite and limestone, two plots were selected for each of four landforms as follows: ridge crests, slope faces, gullies and terraces (i.e. 4 basal substrates × 4 landforms × 2 plots). Previous work has demonstrated that these soil toposequences represent strong fertility gradients, from low fertility on ridges to high fertility in gullies (Richardson, Allen & Doherty 2008). Significantly disturbed vegetation was avoided. We measured the diameter at breast height (dbh, 135 cm) of all stems > 2 cm dbh and identified each stem to species within a 20 × 10 m subplot centred within each unbounded plot. In general, tree and understorey species in these forests are characterized as annual evergreens (i.e. having leaves persisting about 1 year but senescence occurs throughout the year) and are relatively slow growing (Wardle 2012). Few direct measurements of tree or leaf longevity have been made on these species, but the evidence from a previous study in the region indicates that leaf turnover is not strongly linked to soil fertility (Richardson et al. 2010). Reliable data on other life-history traits for most of these species are lacking (but see McGlone, Richardson & Jordan 2010 for tree and sapling heights).
Soil sampling was conducted from November 2009 to March 2010. To characterize the differences in soil chemical properties across plots, 3–5 soil cores, each 10 cm in diameter and of sufficient depth to ensure that the top 10 cm of mineral soil was included, were taken randomly from within each 10 × 20 m subplot. From each core, the organic and mineral soil layers were separated, and the top 10 cm of each layer was retained and pooled across cores; the litter layer and mosses and lichens were excluded. Where the organic layer was < 10 cm thick, the entire organic layer was retained, and for one gully plot, the organic layer was entirely absent, so only mineral soil was sampled. The mineral and organic soils were sieved to 4 mm before subsampling for analysis. Organic and mineral soil samples from each plot were analysed separately following standard procedures through the Landcare Research laboratory for environmental chemistry in Palmerston North, New Zealand (Blakemore, Searle & Daly 1987). Soils were analysed for pH (in water), total C and N (FP2000 CN analyzer; LECO Corp., St Joseph, MI, USA), mineral N (NO3− and NH4+ extracted in 100 mL 2 M KCl; quantified by colorimetry), total P (Kjeldahl acid digestion; quantified by colorimetry), available P (Olsen extraction; quantified by colorimetry), exchangeable bases (Ca, Mg K and Na; atomic absorbtion spectrometry from a 1M ammonium acetate extract), cation exchange capacity and base saturation (NH4+ ions displaced by 1M molar NaCl; quantified by colorimetry).
Leaf and litter sampling and trait measurements
All plant material was collected between November 2008 and March 2009. For each plot, leaf and litter material were collected from at least 5 mature individuals for each of 16 commonly occurring plant species in the region (Table S1 in Supporting Information). These species included 13 angiosperm trees and three fern species, two of which are tree ferns. From each individual, fully emerged leaves were collected; sunlight leaves for canopy species and the leaves of subcanopy and ground layer species were collected from the highest light environments in which they occurred. The primary photosynthetic units of the fern species were selected as structures equivalent to leaves, that is, pinnules from the tree ferns and pinnae from the fern Blechnum discolor. Leaf material was collected using an orchid pruner or, where necessary, a shotgun. To avoid water loss from leaf material during transport, whole branchlets, fronds or pinnae were sampled with leaves or leaf equivalents still attached. Leaf material was immediately sealed in plastic bags and pooled by species within each plot. All leaf material was transported in ice-cooled boxes and stored in the laboratory at 4 °C prior to analysis. Fresh senescent leaf litter (i.e. the most recently fallen litter on the soil surface in a physically intact form and without signs of fragmentation) for each target species within each plot was collected from underneath the same individuals as those from which the foliage was collected. In total, 345 leaf and litter samples were taken, with each species occurring on between 9 and 30 of the 32 plots (median = 22 plots).
The leaf and litter traits we measured were selected because they are known to be part of, or closely linked to, the LES; these include foliar and litter C, N and P, specific leaf area (SLA) and leaf dry matter content (LDMC). As such, we expect that they reflect the trade-off between resource acquisition and conservation, with concentrations of N and P, and SLA increasing, and concentrations of C and LDMC decreasing, with increasing soil fertility (see Diaz et al. 2004; Wright et al. 2004). Once in the laboratory, leaves, pinnae and pinnules were excised from their parent material and representative subsamples were composed, consisting of a minimum of 10 leaves. Green leaf and litter subsamples from each species from each plot were measured for total C, N and P. Subsamples were oven-dried at 65 °C for 48 h, and the concentrations of C were quantified by dry combustion; those of N and P were measured by the Kjeldahl acid digestion method. Further, for each green leaf sample (petioles included), leaf area and fresh weight were determined for 20–35 fresh leaves; these samples were then oven-dried to constant mass for 48 h and weighed again. These values were used to determine LDMC as the ratio of dry weight to fresh weight (mg g−1) and SLA as the ratio of leaf area to dry weight (m2 kg−1) as described by Cornelissen et al. (2003).
The decomposability of leaf litter for each species from each plot was determined using a standardized laboratory bioassay (Wardle, Bonner & Barker 2002; Wardle et al. 2009; Jackson, Peltzer & Wardle 2013). For each species from each plot, 3.9-cm-diameter Petri dishes were each two-thirds filled with a standardized humus substrate (1.97% N, pH 3.7; collected from a Metrosideros umbellata–Weinmannia racemosa-dominated forest near Otira, New Zealand; 42°50′ S, 171°37′ E) and amended to 300% moisture content (dry-weight basis); a disc of nylon mesh with 1 mm holes was placed on the humus surface. A subsample of leaf litter (1 g, air-dried), cut into 2 cm2 fragments (when individual leaves were larger than that), was placed on the surface of the mesh of each Petri dish; the dish was then sealed with tape to minimize water loss and incubated for 4 months at ~22 °C. Following this incubation period, all remaining leaf litter was removed from the Petri dish, picked or brushed clean of soil particles and fungal hyphae and oven-dried at 65 °C for 48 h before measurement of the remaining dry mass. A subsample of the remaining (undecomposed) material for each litter was then analysed for N and P concentration as described above. Decomposition was determined as the percentage of the initial mass (corrected for water content) lost during incubation. Net loss of N and P from the litter was calculated as (the total mass × nutrient concentration prior to incubation) − (the remaining mass × nutrient concentration after incubation) (Wardle, Bonner & Barker 2002). The proportion of total initial N and P lost from the litter during decomposition was calculated from these values.
Coefficients of variation (i.e. SD/mean; CVs) were calculated to quantify the variability of leaf and litter traits and litter mass loss both within and between species (Albert et al. 2010a; Fajardo & Piper 2011). For each trait or litter mass loss variable, within-species CVs were determined across plots. Between-species CVs for these variables were calculated using the mean trait or mass loss values for each species across all plots. For N and P loss from leaf litter, some negative values (indicating net immobilization of N or P) were observed, and thus, CVs were calculated from gross and not net values, that is, relative to an initial concentration of 100%. The coordination in the variation of traits across species was examined by principle component analysis (PCA) using the mean trait values across all plots for each species so that each species was represented by a single value (Diaz et al. 2004; Wright et al. 2004). Furthermore, to explore whether within-species changes in leaf trait values were coordinated, a separate PCA was performed for each species using the trait values for all plots in which it was present.
For each species, regression analyses were performed to examine the relationships of leaf and litter traits across plots as the response variables, with two measures of soil nutrient status of the plots as the explanatory variables, that is, the soil C/N and C/P ratios (Richardson et al. 2005; Richardson, Allen & Doherty 2008). Although several additional measures of soil fertility showed some relation to trait variation (data not shown), the soil C/N and C/P ratios were the most consistently related to the leaf and litter traits both within and between species. Within species, regression analyses were also used to assess the relationship of litter mass loss and N and P release with leaf and litter trait values across plots.
To assess the relationships at the whole plot level of the leaf and litter traits with the soil C/N and C/P ratios and with the loss of litter mass, N and P during decomposition, we used a weighted averages approach (Garnier et al. 2007; Fortunel et al. 2009), weighting species by their relative basal area in each plot:
- (eqn 1)
where traitagg is the aggregate (or weighted average) value of that trait (or decomposition variable) for all tree species collected in that plot, pi is the basal area of a tree species i as a proportion of the total basal area for all tree species collected in that plot and traiti is the value of the trait (or decomposition variable) for tree species i. To ensure that the weighted trait values were representative of the whole plant communities, those plots for which the 16 species we measured collectively accounted for < 70% of the total basal area were excluded from community-level analyses, consistent with the recommendations of Garnier et al. (2004); here, 24 of the 32 plots were retained.
We first determined the relationship of aggregated trait values with soil C/N and C/P ratios across the 24 plots using linear regression. In addition, the relative contributions of the turnover of species among plots, intraspecific variation of species, and their covariation, to the relationships between community trait values and soil nutrient status were estimated following the sum of squares decomposition procedure of Lepš et al. (2011). In this method, species turnover refers to trait variation attributable to variation in tree species' basal areas among plots, whereas intraspecific variation refers to trait variation within species among plots, both of which contribute to differences in overall aggregate trait values between plots. Aggregate mean trait values for the tree species community at each plot were calculated by firstly using the trait values specific to each plot (Specific) and secondly using the fixed species means from across all plots (Fixed). The difference between the specific mean and the fixed mean represents the effect of intraspecific variability (Intrasp.) and was retained as a third community parameter. Using linear regression, these three community parameters were then modelled as a function of the measures of soil fertility across the plots. By decomposition of the sums of squares (SS) across the three linear models, the total SS of each parameter represents how much of the variability is accounted for by each component: SSSpecific = total variance (V); SSFixed = species turnover (T); and SSIntrasp = intraspecific variation (I). Further, if V = T + I (eqn. 2), then turnover and intraspecific variability effects are independent, whereas if the total V is larger or smaller than the sum of T and I, then there is positive or negative covariation (C) between the species turnover and intraspecific effects (for details on SS decomposition procedure see Lepš et al. 2011).
Using the same variance partitioning approach, the contributions of WSV, BSV and their covariation to the total observed variation in the aggregated measures of decomposability across the plots were calculated. We then assessed the power of the aggregated leaf and litter trait measures to predict the aggregated measures of the loss of litter mass, N and P during decomposition across the 24 plots, using linear regression. Both aggregated leaf and litter trait measures inclusive of WSV and exclusive of WSV (calculated as above) were regressed against loss of mass, N and P, and the goodness-of-fit of each pair of models was compared using Akaike's information criteria (AIC). We recognize that, like other studies that have determined aggregated measures of litter decomposability (Wardle et al. 2009; Sundqvist, Giesler & Wardle 2011), our methodology does not take into account the effects of litter mixing on decomposition. However, mixing effects are likely to be negligible; it has been shown both for the type of forest and species considered in the present study (Wardle et al. 2006) as well as more widely (Srivastava et al. 2009) that these mixing effects are generally small and non-directional compared with the substantial overriding effects of plant species identity and plant traits (Cornwell et al. 2008).
All statistical analyses were performed in r version 2.12.2 (R Foundation for Statistical Computing, Vienna, Austria). Sum of squares decomposition was performed using the ‘trait.flex.anova’ macro (Lepš et al. 2011).