The natural climate grid combines 12 sites representing four levels of mean annual precipitation [ca. 600 (1) 1200 (2), 2000 (3) and 2700 (4) mm] and three levels of summer temperature [means of the four warmest months; ca. 6.5 (ALP), 8.5 (INT) and 10.5 (LOW) °C; Fig. 1]. The grid was designed to cover a gradient of ca. 4 °C across the boreal to low-alpine zone transition. We targeted grazed intermediate-rich meadows (Potentillo-Festucetum ovinae; G8 sensu Fremstad 1997) occurring on south-facing, shallow slopes (5–20°) with relatively rich bedrock in terms of nutrient availability. Sites were selected specifically to keep grazing regime and history, bedrock, slope, aspect and vegetation types as constant as possible. Full names and geographic coordinates of each site are available in Appendix S1. All sites were fenced in spring 2009 to avoid animal disturbance. Geographical distance between sites is on average 15 km and ranges from 175 km (LOW1 and LOW4) to 650 m (LOW2 and INT2, which are also 400 m a.s.l. apart).
Figure 1. Position of each site within the SEEDCLIM climate grid. Altitude is the main driver for changes in mean summer temperature, and continentality is the main driver for changes in annual precipitation within the grid, but there are interactions between the two and sites are therefore positioned in geographical space so as to decouple the two gradients as far as possible.
Download figure to PowerPoint
We used interpolated temperature and precipitation data from the period 1961–1990 with a resolution of 100 m (Norwegian Meteorological Institute, www.met.no; see Tveito et al. 2005 for method description) for site selection and statistical analyses. The interpolated mean summer temperatures were highly correlated with on-site 2-m height temperature measurements in 2009 (Pearson correlation 0.93, n = 12). Precipitation loggers were also set up locally in 2009, but the recordings contained too many measurement errors to be used.
The two alpine–lowland species pairs, Veronica alpina–V. officinalis and Viola biflora–V. palustris, were chosen so that our sites covered the rear edge of the two alpine species temperature niche and the leading edge of the two lowland species temperature niche. We chose common species within the climatic grid to maximize the number of sites where the species occurred individually and where both species of a pair occurred simultaneously. Similar branching structure within our species pairs allows for easier comparison between alpine and lowland species in further studies including morphological species traits. All four species are clonal, and often develop long lateral rhizomes and several flowering shoots on the same genet (especially the two lowland species). We therefore use the shoot as our working unit.
Viola biflora is common in moist and relatively rich mountain habitats and is found in snowbeds and leesides, grazed upland pastures, stream banks and birch forests. Viola palustris grows on moist soils and is common in moist pastures, meadows, forests, mires and stream banks. Veronica alpina is found in upland habitats and is common in snowbeds, upland forests, grasslands and stream banks. Veronica officinalis is found on shallow well-drained soils within pastures and meadows, along road verges and in grazed forests and uplands (Lid & Lid 2005; Mossberg & Stenberg 2007).
Plant trait sampling
At each site, we selected five blocks of ca. 5 m² each within an area of ca. 30 m². Blocks were chosen to be as similar as possible in terms of vegetation structure, slope and aspect. Within each block, five 25 cm × 25 cm plots were placed systematically, with occurrence of the target community and/or one or more of our four target species (see Appendix S1 for species occurrence) as acceptance criteria. For Veronica shoots (ramets), we recorded shoot height, length and width of the largest leaf, and number of leaves, flowers, buds and capsules. For Viola shoots, we recorded length of the longest leaf stalk, length and width of the largest leaf, number of leaves, flowers, buds and capsules, and height of the highest reproductive organ. Flowering probability was calculated as the proportion of shoots in each plot that had a reproductive organ, while flower production was calculated as the sum of buds, flowers and capsules per shoot, excluding the non-flowering shoots. This trait was not considered for Viola palustris as it mostly produced a single flower. Hereafter, we refer to this data set as the ‘demography data’.
Additionally, we collected 14–23 genets of each target species outside the blocks by repeatedly dropping a 50 cm × 50 cm quadrat on the ground and harvesting all genets in the quadrat until we had collected at least ten genets of each of the focal species occurring in each site. Shoots (ramets) of these genets were measured in the same way as in the demography data. The different plant parts were weighed to estimate the vegetative biomass of those occurring within the blocks (hereafter ‘biomass data’).
Vegetative biomass (hereafter ‘biomass’), as a measure of size for the shoots in the demography data, was estimated from the biomass data using linear mixed effect models (see Pinheiro & Bates 2000 for details) and was modelled as a function of the non-reproductive traits. All models were nested on site and genet to account for repeated measurements. To assess the goodness-of-fit of these models, we calculated an R2 analogue based on likelihood ratios (Magee 1990): , where logLM is the log-likelihood of the model, logL0 is the log-likelihood of the null model with a fixed intercept and random intercepts for sites and individuals, and N is the number of observations. We then used these models (Table 1) to estimate the biomass of shoots in the demography data, while correcting for random effects of site.
Table 1. Fixed effects coefficients of mixed effects models used to estimate biomass (BM) for the four focal species with sample sizes (N) and RLR2 based on likelihood ratios (see Methods for details). The response variable is log2(BM (mg)) and shoot height, leaf length and leaf width are expressed in mm
|Species\Terms||N||RLR2||Intercept||Shoot height||Number of leaves||Leaf length||Leaf width|
| Veronica alpina ||165||0.78||0.90**||0.01***||0.09***||0.09**||0.16**|
| Veronica officinalis ||455||0.77||2.63***||0.01***||0.07***||0.07***||0.19***|
| Viola biflora ||91||0.72||2.02***||n.s.||0.21***||n.s.||0.15***|
| Viola palustris ||125||0.56||2.53***||n.s.||0.20*||0.18***||n.s.|
Generalized linear mixed effect models (see Pinheiro & Bates 2000 for details) were used to investigate direct climate effects as well as climate-driven changes in minimum size for reproduction (hereafter referred to as additive size and climate effects) and in reproductive investment (hereafter referred to as interactive size and climate effects) in the demography data. In these models, flowering probability and flower production were regressed against biomass, summer temperature, annual precipitation and potential interactions between biomass and climate variables. Hereafter, we refer to these models as ‘flowering probability’ and ‘flower production’ models, respectively.
Plant size responses to climate were investigated in the demography data set using linear mixed effect models with biomass as the response variable and summer temperature and annual precipitation as explanatory variables. Hereafter we refer to these models as ‘Biomass’ models.
Figure 2 illustrates how the investigated effects were identified. A direct climate effect is indicated by responses to at least one of the two climate variables in the flowering probability and/or flower production models (Fig. 2a). An indirect climate effect through plant size is indicated by responses of flowering probability and/or flower production to biomass, combined with a biomass response to climate in the biomass models (Fig. 2b). Climate-driven variation in minimum size for reproduction (additive size and climate effects) is indicated by a response of flowering probability to both biomass and climate (Fig. 2c), as an effect of climate shifts the size-dependent curve for flowering probability towards smaller or larger sizes. Finally, climate-driven variation in reproductive investment (interactive size and climate effects) is indicated by interactions between biomass and climate in the flowering probability and/or flower production models (Fig. 2d,e), since an interaction between biomass and climate indicates that a change in climate causes a shift in the slope describing the relationship between flowering probability or flower production and plant size.
Figure 2. Hypothetical effects of climate and plant size on flowering performance (flowering probability and flower production). Direct climate effects are indicated by a main effect of temperature or precipitation on flowering probability and/or flower production (a), while indirect climate effects through plant size are indicated by a main effect of climate (temperature and/or precipitation) on biomass and a main effect of biomass on flowering probability and/or flower production (b). A change in minimum size for reproduction with climate (additive size and climate effect) is indicated by main effects of both biomass and climate in the flowering probability models (c). Changes in reproductive investment with climate (interactive size and climate effect) are indicated by interactions between climate variables and biomass, and can be investigated for both flower probability (d) and flower production (e).
Download figure to PowerPoint
All four types of effect can act individually or in concert. For example, an indirect climate effect through plant size in addition to a direct climate effect and an additive size and climate effect would be indicated by a response of biomass to climate and responses of flowering probability to both biomass and to climate. Note that in such cases we cannot separate the additive size and climate effect from the direct climate effect.
The two alpine species were only present in the alpine sites and in one of the intermediate temperature sites (INT3, see Appendix S1). Thus, the variation in summer temperature is dependent on one site only for the two alpine species. We therefore investigated the responses to annual precipitation and summer temperature in separate models for these species. When assessing the response to annual precipitation, we removed plots from the intermediate temperature site (INT3, see Fig. 1) to avoid incorporating the variance caused by temperature differences between the two sites of the third precipitation level. Similarly, we analysed responses to mean summer temperature using the sites INT3 and ALP3 (ca. 2000 mm annual precipitation for both sites; Fig. 1). In these models, mean summer temperature was expressed as a categorical variable.
Biomass was log2 transformed for all analyses to meet the normality assumption. Annual precipitation was expressed in metres in the regressions, so that it obtained coefficients with a similar scale to the other predictors. The inclusion of quadratic terms was tested for the two climate variables in all models, if suggested by visual inspection of the data. All models were selected using likelihood ratio tests in a step-wise backward selection process. Flowering probability, flower production and biomass models were nested hierarchically on sites, blocks and plots to account for data set structure. We assumed binomial error distributions for the flowering probability model, Poisson error distributions for the flower production models, and Gaussian distributions for the biomass models.
We assessed the relative importance of the investigated responses for an increase of 2 °C in summer temperature and a 10% increase in annual precipitation using the model predictions. To do so, we calculated the absolute change in flowering probability and/or flower production due to each variable, and then assessed its relative contribution compared to the sum of the absolute changes (see Appendix S2 for detailed methods and equations). For the lowland species, we based our calculation on the coldest wettest site where the species occurred (leading edge), and for the alpine species, on the driest site (rear edge).
In addition to mixed models, we explored structural equation modelling (SEM, see Fox 1980 for detailed method) as a method for assessing direct and indirect climate effects. These analyses were inferior to mixed models, as they could not easily account for random variance components. We present one example in Appendix S3 for comparative purposes.
All analyses were carried out in R (v. 2.13.1; R Foundation for Statistical Computing, Vienna, AT), using the packages nlme (v. 3.1-97; Biomass model) and lme4 (v. 0.999375-33; all other models).