Herbivory and nutrient limitation protect warming tundra from lowland species’ invasion and diversity loss

Study system and experimental site

Our experiment was carried out in Kilpisjärvi, NW Finnish Lapland (69.055°N, 20.887°E), on the south-western slope of Mt. Iso-Jehkats at an altitude of 750 m (Kaarlejärvi et al., 2013). Our study site of roughly 50 × 100 m is a species-rich (up to 13 vascular plant species per 25 × 25 cm² area) treeless tundra meadow, dominated by grasses and forbs (e.g., Deschampsia flexuosa, Festuca ovina, Potenilla crantzii, Viola biflora, Thalictrum alpinum, Astragalus alpinus). In this area, treeline lies at 600–650 m a.s.l. and is formed by mountain birch [Betula pubescens Ehrh. subsp. czerepanovii (N.I. Orlova) Hämet-Ahti]. This area is an important summer grazing area for semidomesticated reindeer (Rangifer tarandus tarandus), which are released to the area by local reindeer herders at the end of June and return to the winter grazing area at the beginning of August (Heikkinen et al., 2005). In summers 2010 and 2011, ca. 1500 reindeer grazed within a summer grazing area of ca. 90 km² (corresponding to a density of roughly 17 animals per km²) (Kaarlejärvi et al., 2013). During the past decades, the amount of reindeer in our study area has considerably varied (Eskelinen & Oksanen, 2006). Reindeer is a generalist herbivore reducing vegetation biomass; however, it can also strongly select for palatable plants and affect species composition (e.g., Olofsson et al., 2002; Kaarlejärvi et al., 2013; Eskelinen et al., 2016). Other mammalian herbivores present in the area are Norwegian lemmings (Lemmus lemmus L.) and gray-sided voles (Clethrionomus rufocanus Sund.), which can cut stems of vascular plants and create small open patches, having considerable impact on vegetation during their peak years (Virtanen et al., 1997). Mountain hares (Lepus timidus L.) are encountered occasionally.

Experimental design

In August 2009, we established 56 plots of 0.8 × 0.7 m at our experimental site. Plots were randomly assigned to the following treatments in a full-factorial design: (i) mammalian herbivore exclusion, (ii) fertilization, (iii) warming, or as controls, resulting in seven replicates per treatment combination. Each of the 56 main plots was further split into two 0.25 × 0.25 m subplots, one of which was assigned to a biomass removal treatment and the other was left intact.

For the herbivore exclusion treatment, we erected circular exclosures 160 cm in diameter, 100 cm high, dug to a depth of 15 cm into the soil and made of galvanized net with a mesh size of 10 × 10 mm around the grazer exclusion plots. Our herbivore exclusion treatment excluded all mammalian herbivores including reindeer, hares, voles and lemmings and tests the role of mammalian herbivory in tundra areas in general. However, it also mimics a possible scenario that reindeer (and other mammals) are absent or exhibit very low densities for a limited time period. For the fertilization treatment, we applied fast-dissolving NPK fertilizer (16-9-22) mixed with 1 l of water from a nearby stream over the fertilized plots twice per growing season (mid-June and end of July), resulting in an addition of 9.6 g N, 5.4 g P and 13.2 g K m⁻² on the fertilized plots annually. The same amount of water (without fertilizer) was also applied to the unfertilized plots. Our nutrient addition treatment was testing the importance of nutrient limitation in general. It also mimicked variation in soil nutrient concentrations between habitats and between tundra areas of low vs. high anthropogenic nutrient enrichment (e.g., some alpine tundra areas in heavily impacted regions; Galloway et al., 2004; Dentener et al., 2006). In addition, herbivores and climate warming can also increase soil nutrient availability (Olofsson et al., 2001; Rustad et al., 2001; Natali et al., 2012).

For the warming treatment, we used ITEX (International Tundra Experiment) hexagonal open-top chambers (hereafter OTC) with a maximum basal diameter of 146 cm. While voles and lemmings can move in and out of the chambers, OTCs can at least partly act as grazing barriers for reindeer and prevent their grazing (Moise & Henry, 2010; E.K. and J.O. pers. obs.). Thus, to achieve full-factorial experiment where all mammalian herbivores could equally access all grazed (unfenced) plots, we removed the OTCs during the 1-month period when reindeer were present in the area (i.e., in July). To ensure that reindeer potentially straying from the main herd did not confound our design, we established a temporary reindeer fence around the whole experimental area every year when the OTCs were placed back (i.e., in the end of July). This temporary fence was removed every year when reindeer were released back to the area and our OTCs were removed (i.e., in the end of June). Our warming treatment therefore simulated spring and autumn warming (i.e., April–June and August–October) and extended the length of the growing season at both ends; this is when lower temperatures are likely to constrain growth the most and warming should have the biggest impact. The OTCs increased air temperature on average by 1.92 °C in June (mean ± SE in controls and in OTCs 11.20 ± 0.59, n = 4, and 13.12 ± 0.25, n = 4, respectively) and by 1.23 °C in August 2011 (mean ± SE in controls and in OTCs 9.68 ± 0.21, n = 4, and 10.91 ± 0.49, n = 2, respectively). Greater temperature increase by OTCs at both ends of growing season than in the middle of the growing season corresponds well with the predictions for our study region (Kivinen et al., 2012).

For the biomass removal treatment (simulating competition reducing disturbance), we hand-clipped all above-ground live biomass and litter from the biomass removal subplots in August/September 2009. We repeated the biomass removal treatment in 2010 and 2011; however, care was taken not to disturb any of the seeded species occurring in the plots (and therefore biomass removal treatment was only partial in these years). After 2011, seeded species were fully grown and it was not possible to clip any unseeded residents without disturbing the seeded species. Although our biomass removal treatment was designed to test the impact of competition at the time of seed arrival and germination, it also mimicked natural heavy disturbance created by lemmings during their peak years or intensive trampling by reindeer.

Seed addition

We chose 25 common plant species in our study region of which 12 were common and typical for above-treeline communities (called ‘tundra species’) and 13 species had their main altitudinal distribution below the treeline (called ‘lowland species’). The division of the species into lowland and tundra species was done based on their current occurrence in the field in our study region. Tundra species represent species that are presently common and abundant above treeline in Fennoscandia (Benum, 1958; Oksanen & Virtanen, 1995; Virtanen et al., 1999) and also occur at our study site. Many of these species can, however, also be found at lower altitudes. Lowland species represent species that are not found at our study site, are mainly found in the below-treeline forests and meadows and only occasionally enter above-treeline tundra (Benum, 1958; Olofsson et al., 2002; Virtanen et al., 2010). Single sporadic individuals of two of the lowland species were, however, found at our study plots (Geranium sylvaticum, Trollius europeus), but both at very low abundances. These two species have their altitudinal optimum in below-treeline communities where they belong to dominant species pool (Benum, 1958; Olofsson et al., 2002; Virtanen et al., 2010). Tundra species thus represent resident species that are already present at the site, and lowland species represent species that could potentially migrate from lower altitudes, establish at the site and become dominant under warmer climate.

The seeded species represented different growth forms (grasses, forbs, N-fixers, dwarf shrubs) and traits (see Table S1). Seeds of all species were collected in fall 2009 (for the 2009 sowing) and in 2010 (for the 2010 sowing) from Kilpisjärvi area. All species germinated either in laboratory or in the field (Table S1). Seed mixtures of 50 seeds per species were individually prepared to all subplots, and each subplot therefore received the same number of seeds per species. The mixture of seeds of 25 species was evenly distributed over all subplots, both with and without biomass removal treatment, in mid-September 2009 just before the first snowfall of the year. We repeated the seed addition in mid-September 2010 because germination was relatively low after the first sowing and to ensure that the seeds experienced the full range of altered environmental conditions brought by the experimental treatments. During sowing, we prevented accidental seed spread using a cardboard box around each seeded subplot. After seed addition, vegetation was gently shaken to facilitate settling of seeds to the ground.

Traits of the seeded species

We measured plant height (cm), specific leaf area (SLA, leaf area in mm² per g dry mass) and foliar carbon-to-nitrogen ratio (C : N; based on % C and N in plant leaves) for the seeded species (Table S1) following the standard protocol by Pérez-Harguindeguy et al. (2013). Trait data were collected from 10 naturally occurring individuals per species from the study region in summer 2014. We used community-weighted mean trait values (CWM traits; Garnier et al., 2004) to describe how seeded community trait composition changed in response to the experimental treatments. For each trait, each species’ relative abundance in the seeded community in 2014 was multiplied by that species’ trait value and summed across species to give a community-weighted mean value in each plot. A high CWM trait value therefore indicates a community that is dominated by species with high values of that trait.

Community measurements

We sampled subplots for seeded species community composition at the end of July/the beginning of August in 2014. The cover of all seeded species in all subplots was estimated visually by the same trained and experienced person with a minimum estimate threshold of 0.1%. We were especially careful to detect and identify even small seedlings of the seeded species. We also used these cover values to calculate community-weighted mean trait values for the communities.

To get biomass estimates for the main plots (within which the biomass removal treatment was nested), we used a modified point intercept method (Jonasson, 1988) with 108 evenly spaced sampling points in 25 × 50 cm subplots (separate from but next to the seeded plots). At each point, a pin was lowered to the ground and the number of contacts with all vascular plant species was recorded. We used the sum of all contacts of all vascular plant species as a proxy for total community biomass. This method correlates well with true biomass (Jonasson, 1988) and is widely used as a nondestructive method to estimate biomass (see, e.g., Post & Pedersen, 2008; Eskelinen et al., 2012; Kaarlejärvi et al., 2015).

Statistical analyses

To test the effects of herbivore exclusion, fertilization, warming and biomass removal on the success of seeded lowland and tundra species, we applied linear mixed-effects models (LME; Pinheiro & Bates, 2000). The cover of lowland and tundra species, total species richness of seeded species and species richness of seeded lowland and tundra species were response variables (each in their own separate models), and herbivory, fertilization, warming, biomass removal and their interactions were fixed explanatory variables. We included the hierarchical experimental design in each model by nested random effects where biomass removal treatment (subplot) was nested within the other treatments (plot). We used F-tests to assess the significance of the experimental treatments and their interactions (Pinheiro et al., 2015). The heteroskedasticity of variances and normality of errors were checked using model diagnostic plots (Crawley, 2007), and the response variables were square-root-transformed or log-transformed as necessary (see Table 1). Transformed data met homogeneity and normality assumptions.

To test whether plant functional traits mediated seeded community responses to herbivore exclusion, fertilization, warming and biomass removal, we first applied nonmetric multidimensional scaling (NMDS) with Bray–Curtis dissimilarity measure and square-root-transformation with Wisconsin double standardization (Oksanen et al., 2016) for the seeded communities. We then assessed the significance of the correlations between seeded communities (NMDS ordination) and plant functional traits (calculated using community-weighted mean trait values for the seeded species in 2014) using 999 permutations. Significances of experimental treatments on the seeded communities were assessed using permutational multivariate analysis of variance (PERMANOVA; Anderson, 2001) with Bray–Curtis dissimilarity measure and 999 permutations.

We also used correlations to look at relationships between cover of seeded tundra and lowland species, between species richness of the seeded species and total community biomass, and between the ratio of tundra-to-lowland species cover and species richness of the seeded species.

We used the function ‘lme’ in package nlme (Pinheiro et al., 2015) for LME and the functions ‘vegdist’ for Bray–Curtis dissimilarity, ‘metaMDS’ for NMDS, ‘envfit’ for fitting and testing significance of vectors/factors in NMDS ordination, and the function ‘adonis’ for PERMANOVA, all in package vegan (Oksanen et al., 2016), in R statistical software (R Development Core Team 2015 version 3.2.3).

Species’ cover and traits

Seeded lowland and tundra species’ cover showed opposite responses to many treatments (Fig. 1), and we also identified a strong negative correlation between lowland and tundra species cover (r = −0.731, t₁₄ = −4.0, P = 0.0013, Fig. 2). Overall, lowland species cover was lowest in grazed-only, and grazed and warmed plots, where tundra species had high cover and high in fertilized-exclosed and fertilized-warmed plots, where the cover of tundra species was lowest (Fig. 1). These contrasting responses were also associated with clear trait differences (Fig. 3).

Warming alone did not increase lowland species cover; however, warming and fertilization together exhibited a synergistic, positive impact on lowland species (in line with our prediction), indicating colimitation by temperature and soil nutrients (W × F interaction, Table 1, Fig. 1). Both fertilization and herbivore exclusion alone also increased lowland cover, suggesting that nutrient scarcity and grazing hamper lowland species establishment (in line with our prediction; positive main effects of F and E, Table 1, Fig. 1). Surprisingly, warming effects on lowland cover also depended on herbivore exclusion; we saw no effect inside exclosures but found a positive effect in grazed plots (W × E interaction, Table 1, Fig. 1). In line with our prediction, the cover of lowland species was positively associated with community-weighted traits representing fast-growing syndrome, especially height but also SLA and low C : N. Both lowland species’ cover and these traits were also associated with the combined herbivore exclusion, fertilization and warming treatment (NMDS and PERMANOVA results; Fig. 3).

In contrast to our second prediction, warming alone had a positive impact on tundra species cover, while, in accordance with our prediction, both fertilization and herbivore exclusion had negative effects on their cover (Table 1, Fig. 1). The cover of tundra plants was highest in warmed-only plots (in both grazed and ambient nutrient conditions, Fig. 1), suggesting that tundra species benefit from warmer conditions especially when grazers are present and when soil nutrient supply remains low (although W × E × F interaction was not significant; P = 0.0992, Table 1). In our trait analyses, tundra species cover was negatively associated with height and positively with the C :N ratio and warming (NMDS and PERMANOVA results; Fig. 3).

Species richness

Species richness of seeded lowland and tundra plants was relatively similar to each other and corresponded to total seeded richness responses to the treatments (see Table 1, Fig. S1), and we therefore only report total species richness response here. As we predicted (Prediction 3), warming increased total seeded species richness in the absence of competitors (i.e., in biomass removal plots); however, fertilization and herbivore exclusion canceled this positive impact (significant W × BR × E × F interaction, Table 1, Fig. 4). Species richness peaked in warmed and grazed plots from which biomass had been removed and which were not fertilized, while the lowest species richness was encountered in exclosed and fertilized plots without biomass removal, irrespective of warming (Fig. 4). In particular, tundra species were nearly absent from fertilized and exclosed nonbiomass removal plots (Fig. S1).

The ratio of tundra-to-lowland species cover exhibited a strong positive relationship to species richness (r = 0.58, t₁₁₀ = 7.6, P < 0.0001, Fig. 5): the dominance of tundra species (i.e., high tundra: lowland ratio) was associated with high species richness, whereas the dominance of lowland species was associated with low species richness. There was also a negative correlation between total seeded richness and total community biomass (r = −0.555, t₅₄ = −4.9, P < 0.0001, Fig. S2).