Effects of sheep grazing and temporal variability on population dynamics of the clonal herb Geranium sylvaticum in an alpine habitat
- First Online:
- Cite this article as:
- Evju, M., Halvorsen, R., Rydgren, K. et al. Plant Ecol (2011) 212: 1299. doi:10.1007/s11258-011-9906-2
An improved understanding of population-level consequences of grazing on plants can be facilitated by an assessment of grazing effects on all stages in the life-cycle. In this study, 6 years of demographic data for three populations of the perennial herb Geranium sylvaticum were analysed. We examined the effects of sheep grazing (high sheep density, low sheep density and no sheep) and interannual climatic variability on vital rates and population growth rates (λ). Grazing did not affect survival or flowering rates, but reduced rates of growth and increased rates of clonal reproduction. At the population level, high contributions from retrogression and clonal reproduction buffered reduced rates of growth and stasis, and no consistent differences in λ between populations exposed to different sheep densities were found. Instead, large between-year variability in λ, independent of sheep density, was detected, related to variation in the local summer climate. The results indicated, however, that grazing effects on λ were more severe in unfavourable than in normal years. Our study highlights that increased clonal reproduction rates functioned as a tolerance mechanism towards grazing in this herb, which forms a mechanism to explain how moderate population responses to grazing in some herbs can arise.
KeywordsHerbivoryInterannual variabilityLTRE analysisMatrix projection modelsPlant tolerance
Vertebrate grazers act directly and indirectly upon individual plants by removing biomass, altering resource availability, changing the physical environment and modifying intra- and interspecific interactions (Mulder 1999). For a given plant species, however, stages in its life-cycle may be affected by grazers in different ways. Vertebrate grazers seem, e.g., to prefer large ramets over small (Ehrlén 1995a), and grazing can reduce survival (Ehrlén 1995a), growth (Lennartsson and Oostermeijer 2001) and seed production (Bastrenta 1991; Knight 2004) of individual plants. At the same time, the grazers can facilitate seedling recruitment (Lennartsson and Oostermeijer 2001) and increased survival of juvenile plants (Brys et al. 2004; Ehrlén et al. 2005) through prevention of litter accumulation and creation of vegetation gaps. In order to evaluate population-level consequences of grazing, grazing effects need to be assessed for all stages in the life-cycle.
Disturbance, such as trampling and grazing, can increase rates of clonal reproduction in vascular plants (Diemer and Schmid 2001; Lamoureaux et al. 2003) and bryophytes (Rydgren et al. 2001). The plants’ capacity for clonal reproduction has been predicted to buffer plant populations from negative effects of herbivores (Maron and Crone 2006), and Tolvanen et al. (2001) found higher rates of clonal reproduction in grazed than in ungrazed sedge populations. However, to our knowledge, no studies exist of vertebrate grazing effects on the population dynamics of clonal herbs (see also review by Maron and Crone (2006)).
The magnitude of population-level effects of grazing will depend not only on how grazing as such affects vital rates of the plant, but also on grazing pressure (Hunt 2001; Rydgren et al. 2007), and Knight (2003) found a negative correlation between the proportion of plants being grazed and the population growth rate (λ) of Trillium grandiflorum. Grazing pressure, i.e., the proportion of plants in the population being grazed, depends on herbivore density, but also on the abundance and quality of available forage plants (Augustine and McNaughton 1998; Kausrud et al. 2006). In alpine areas, a short growing season makes vegetation development strongly dependent on weather conditions (Lenart et al. 2002), and consequently, grazing pressure can vary among years, even at constant herbivore densities (Evju et al. 2006; Jefferies 1999). Furthermore, the plants’ ability to compensate for lost biomass (plant tolerance) depends on environmental factors such as light, nutrient and water supply (Cronin and Lodge 2003; Maschinski and Whitham 1989). Plant tolerance may thus vary both in space and time (Lennartsson et al. 1998), and interannual variability in local climate may interact with grazing to affect plant population dynamics (Evju et al. 2010), e.g. through a larger reduction in population growth rate in unfavorable years in grazed versus ungrazed populations (Bastrenta et al. 1995; Martorell 2007).
In this study, we used 6 years of demographic data for a perennial, clonal herb, Geranium sylvaticum. The species is reported to be eaten regularly, but at low intensity (i.e. that only a small proportion of each plant was grazed) by sheep (Hæggström 1990), and is found to benefit from exclusion of vertebrate herbivores in low-productive alpine habitats (Moen and Oksanen 1998; Olofsson 2001). Three populations, located in areas with high sheep density (80 sheep km−2), low sheep density (25 sheep km−2) and no sheep within a landscape-scale sheep grazing experiment, were monitored from 2002 through 2007. Using a combination of stage-based matrix projection models and linear mixed-effect models, we test the following predictions in this study: (1) grazing reduces rates of growth, survival and flowering of Geranium sylvaticum whereas rates of clonal reproduction are increased by grazing, (2) these effects of grazing on vital rates translate to the population level as a lowered population growth rate (λ) at high sheep density, whereas cessation of grazing affects λ positively, (3) effects of grazing vary between years dependent on interannual climatic variability.
Materials and methods
Geranium sylvaticum L. (Geraniaceae) is a tall (20–80 cm) and long-lived, iteroparous rhizomatous herb with wide ecological amplitude, including a broad range of different forest and meadow types, from the nemoral to the low alpine zone (Lid and Lid 2005). G. sylvaticum has a short, unbranched rhizome (3–10 cm), a rosette of basal leaves with long petioles, and produces one to several flowering shoots per ramet. Meristems on the rhizome can be activated, giving rise to clonal offspring (Salomonson et al. 1994; Klimeš and Klimešová 1999). The species may flower at the age of 7–10 years (Salomonson et al. 1994). Fruits contain up to five seeds, which mature ca. 3 weeks after flowering (Ågren and Willson 1994). Seeds are dispersed through explosive dehiscence. The species has a transient seed bank; seeds survive in the soil for less than 1 year (Thompson et al. 1997; Molau and Larsson 2000).
Experimental design and sampling procedure
Mean July temperature and total July precipitation each year during the study period. Temperature and precipitation data (including the average temperature and precipitation for the period 1961–90) are provided for the study area by The Norwegian Meteorological Institute
July temp (°C)
July precipitation (mm)
Snow cover (%)
Twenty permanent plots for vegetation monitoring (0.5 × 0.5 m2 each) were distributed in each sub-enclosure at the start of the experiment in 2001 following a random stratified design (see Austrheim et al. (2008) for details; Fig. 1). These plots form the basis for the present study, in which Geranium sylvaticum ramets were monitored at annual censuses (late July/early August) from 2002 through 2007. One additional census was performed in late June each year except 2006 to detect newly emerged seedlings.
In 2002, 200 ramets (aboveground parts) were collected outside the permanent plots, measured, dried to constant weight at 80°C and weighed. We constructed a multiple regression model of dry mass (DM, in mg) as a function of five morphological variables; plant height, number of leaves, length, and width of the largest leaf and stem diameter at the base, with Radj2 = 0.944 (see Online Resource 1 for details). All ramets located within the 180 permanent plots were non-destructively tagged in 2002. At each census, the five morphological variables were recorded on each ramet, and we subsequently used the regression model to estimate plant size (DM). In addition, at each census, grazing marks on ramets were recorded and the number of capsules was counted on reproductive ramets.
In plots with high densities of G. sylvaticum ramets, a pre-defined subset of the plot which included a minimum of 20 ramets was used. This subset was kept constant for the rest of the study. New plants that appeared in the plots were also tagged and measured, and their origin judged (from seeds, clonal offspring or by immigration). Seedlings were defined as all small plants (for classification of life-stages, see below) that were <1 year old. Mother ramet and clonal offspring relations could not be established non-destructively. Based on excavations of underground structures outside the plots, we assigned new, non-seedling plants to the nearest plant if one was found within a 6 cm distance. New, non-seedling plants appearing close to the plot borders were defined as immigrants.
Six permanent plots were excluded as their corners (and, hence, ramets) could not be exactly re-located (four in high and two in low sheep density).
Life-stage classification and matrix parameterisation
Definitions of life-stage classes and important demographic characteristics of these classes; average mortality (% of plants in that stage class at time t recorded as dead at t + 1), fecundity (% of plants in that stage class setting seeds), clonal reproduction (% of plants in that stage class producing clonal offspring), and conditional lifespan (average time to death for individuals that enter that life-stage class), pooled over populations and years
Clonal reproduction (%)
Conditional lifespan (mean ± SD)
5.4 ± 7.0
Produced through clonal reproduction
8.6 ± 8.6
Log2DM < 3
7.0 ± 7.6
3 ≤ log2DM < 5
8.7 ± 8.4
5 ≤ log2DM < 7
10.1 ± 8.8
Log2DM ≥ 7
10.7 ± 9.0
9.7 ± 8.4
There were not enough ramets present in all sub-enclosures to construct transition matrices for each year and sub-enclosure separately. We therefore pooled data for all sub-enclosures exposed to the same sheep density and treated the data as one population.
To estimate seed production, 100 reproducing ramets were collected in 2007 (evenly distributed among the nine sub-enclosures). The number of capsules and seeds were counted on each ramet in the laboratory. The number of intact seeds per capsule was not related to individual ramet size or to the number of capsules per individual, and did not vary significantly among the sheep density treatments (M. Evju, unpublished data). The data were therefore pooled to give a species-specific mean of 2.35 seeds per capsule, which was treated as constant over all years. To our knowledge, there are no studies of variation in seed set with local climate in G. sylvaticum, although the species has been found to have a higher seed number per capsule in high- than in low-light habitats (Korhonen et al. 2004). Thus, the between-year variation in seed production could be underestimated in our study.
Fecundity was expressed as the average number of seedlings (number of seeds × estimated seedling emergence rate) produced by a ramet in a given stage class. We calculated seedling emergence rates separately for each population and year as the number of seedlings in t + 1 divided by the number of seeds produced in t (recorded number of capsules × 2.35 seeds per capsule). Seeds germinated throughout the season (M. Evju, personal observation), and new plants at the late June census had sometimes passed the cotyledon stage. We therefore defined all small plants <1-year-old as seedlings. In the first year, non-cotyledon seedlings could not be distinguished from small plants, and we therefore used the population-specific mean of observations from 2003–2007 as estimates for transition probabilities from seedlings to all other stages for 2002–2003. Similarly, origin by clonal branching could not be decided in 2002, and we therefore used the population-specific mean of observations for 2003–2007 as estimates for transition probabilities from clonal offspring in 2002–2003.
Prolonged dormancy, i.e., failure to produce above-ground parts for one or several growing seasons (Lesica and Steele 1994) was observed in 102 of the 3117 transitions from stage to fate recorded in the study. Of these, only five plants went dormant for two consecutive growing seasons. Mortality rates for dormant ramets cannot be observed directly, but can be estimated (Kéry et al. 2005). We estimated mortality rates based on the assumption that ramets entering dormancy had the same mortality rates as non-dormant ramets, and thus calculated a weighted mean mortality as the life-stage specific mortality rate × the proportion of dormant plants from this life-stage class, summed over all life-stage classes. For transitions from the dormant stage at time t to all other stages at t + 1, we assumed that all transition rates were affected equally by mortality in the dormant stage and reduced them by subtracting from observed transition rates the estimated mortality rate (Evju et al. 2010). As transitions to dormancy could not be observed in the last transition period (2006–2007), pooled data from 2002–2006 were used to estimate transition rates from each life-stage class to dormancy for each population.
In addition to pooling data for estimating seedling and clonal offspring transitions for 2002–2003 (see above), we used population-specific data pooled over years to estimate transition probabilities for stage classes containing less than seven ramets at time t (Online Resource 2). This was done for transitions from dormancy in one transition period in the high sheep density population, and in three transition periods in the low sheep density and no sheep populations. In addition, in the no sheep population the low number of clonal offspring prevented calculation of year-specific transition probabilities from this life-stage class (see Online Resource 2).
Statistical analyses and transition matrix modelling
We used generalised linear mixed-effect models (GLMM; Pinheiro and Bates 2000) to analyse the effects of sheep density, sheep grazing and local climate on survival, growth, flowering (probability of flowering and number of fruits produced) and clonal reproduction. As all vital rates were strongly size dependent (Table 2), we included plant size (log2DM) as a covariate in the analyses. Plant ID, plot and sub-enclosure were included as random factors in the models to account for repeated measurements on individual plants as well as the spatial dependency of plants within plots and plots within sub-enclosures. To describe local weather conditions, we used mean July temperature and total July precipitation as proxies for overall favourableness of growth during the growing season, representing mid-summer conditions during a short alpine growing season (cf., Callaghan et al. 1989). We tested for possible interactions between local climate and sheep density. Models were evaluated using AIC. The effect of sheep grazing on plant growth was analysed comparing size in t − 1 and t + 1 for plants that were grazed and ungrazed in t. In this analysis, we only included plants for which no grazing was recorded in t − 1 and t + 1, and for which we had reliable size estimates in all 3 years.
We constructed 15 stage-based transition matrices with one-year projection intervals (three populations × five transition periods; Online Resource 2). For each matrix, the population growth rate λ (maximum eigenvalue of the projection matrix) and elasticities (de Kroon et al. 2000; Caswell 2001) were calculated. Elasticity values were summed for each life-history component. We calculated Pearson’s correlation coefficient between the summed elasticities and λ. We used one-way ANOVA to analyse the difference in λ between years and between sheep density treatments, and to investigate differences in summed elasticities of life-history components between treatments. ANOVA was followed by Tukey’s test and adjusted p-values were used (Crawley 2007).
We used bootstrapping to construct confidence intervals around λ. For each combination of treatment and transition period, the plants were resampled with replacement to construct bootstrap samples. The size of the bootstrap sample was set equal to the size of the original data set. Three thousand bootstrap samples were constructed for each transition matrix. The percentile method (Caswell 2001) was used to construct 95% confidence intervals around the median λ. For transition probabilities from seedlings and clonal offspring in the first year (non-observable), we used pooled values from 2003–2007 for each population.
All statistical analyses were carried out using R 2.10.1 (R Development Core Team 2009). Analyses involving matrix computations were performed with the R package popbio (Stubben et al. 2008), and mixed-effect models were run with the R packages nlme and lme4 (Bates and Maechler 2010; Pinheiro et al. 2008).
Effects of local weather and grazing on vital rates
Generalised linear mixed-effect models (GLMM) of survival (response binomial, no. of observations = 2670, no. of plant identities (IDs) = 901), plant size (response continuous, no. of obs. = 1374, no. of IDs = 591) and flowering probability (response binomial, no. of obs. = 2507, no. of IDs = 800) as a function of summer temperature, precipitation and grazing. Parameter estimates for random factors (plant ID, plot and enclosure) are not shown
Plant size (t)
Plant size (t − 1)
Grazing × July precipitation
July temperature (t − 1)
July temperature (t)
Population growth rates and elasticities
Life-table response experiments
The year effect on variation in λ was on average 3.5 times larger than the effect of sheep density (Fig. 7b; mean of absolute values ± SD: 0.059 ± 0.050, n = 5). A positive correlation between net contributions from growth to variation in λ (r = 0.900, df = 3, P = 0.038) was found, as well as a tendency for negative correlation between net contributions from retrogression (r = −0.869, df = 3, P = 0.056). The largest positive effect on λ was observed for 2004–2005, whereas the largest negative effect was observed for 2005–2006.
The net interaction effects were intermediate between the year and treatment effects (Fig. 7c; average of absolute values ± SD; 0.032 ± 0.019, n = 15). Interaction effects were positive in a minimum of two transition periods in all populations, confirming that no population consistently had higher λ than the others.
Tolerance to grazing, i.e., the ability to regrow and/or reproduce after a grazing event (Strauss and Agrawal 1999), is dependent on both frequency and intensity of grazing (del-Val and Crawley 2005). Grazing intensity on Geranium sylvaticum, expressed in terms of proportion of biomass removed, is low in this study; normally only 1–2 leaves are removed (M. Evju, unpublished data). Our results demonstrate that growth of individual ramets is reduced with grazing, but at the same time, increased rates of clonal reproduction following grazing suggest that this species is indeed tolerant to the grazing intensity observed in this study. The grazing pressure on G. sylvaticum in the study area is relatively low compared to other herbs (Evju et al. 2006), and our results are in line with results of other studies showing that herbs are able to tolerate low levels of grazing (Huhta et al. 2003; Moser and Schutz 2006).
These effects of grazing on individual ramets are visible at the population level as higher rates of retrogression (i.e. survival to a smaller life-stage class) and clonal reproduction in the high sheep density treatment. These effects partly buffer negative effects of reduced rates of growth and stasis (survival to a larger or the same life-stage class, respectively). Nonetheless, even at high sheep densities, the proportion of plants being grazed is low (the maximum in 1 year is 16%). Consequently, no consistent differences in population growth rate (λ) occur between populations exposed to high, low or zero sheep densities. At the time-scale included in this study, high sheep density does not impact the population in the high sheep density treatment negatively. This accords with the finding of Doak (1992) that frequent, but low-intensity insect attacks on Epilobium latifolium have small effects on population growth.
Repeated defoliation is expected to deplete the plants’ resource pools, and thereby to reduce the plants’ ability to express tolerance to grazing (del-Val and Crawley 2005). Reduction of individual size, and thereby a shift in stage distribution towards smaller plants under high sheep densities could, therefore, be expected to reduce λ (Knight 2004; Ehrlén 1995b), as survival, fecundity and clonal reproduction rates all increase with size. However, the long lifespan of individual plants, as well as life-history traits such as clonal reproduction, tend to enhance population persistence (Eriksson 1996), and long-lived plants generally show more stable population dynamics than short-lived plants (García et al. 2008; Silvertown et al. 1993). Thus, negative effects of sub-optimal environmental conditions could be expected to be delayed (Morris et al. 2008).
Previous studies have found a strong positive effect on biomass of G. sylvaticum after protecting it from grazers in alpine snow beds (Moen and Oksanen 1998), but that the magnitude of the effect depends on habitat productivity (Olofsson 2001). In our study area, G. sylvaticum is most abundant in productive habitats, which are also favoured by sheep (Mobæk et al. 2009), and our results do not support a general positive effect of excluding sheep. On the contrary, we find that at the population level, the relative importance of growth is less and that of stasis is larger, in the no sheep treatment compared to low sheep density. Accumulation of biomass in G. sylvaticum is strongly linked to light conditions (Salomonson et al. 1994). If sheep grazing reduces the abundance of tall neighbours, this could favour G. sylvaticum plants in small life-stage classes in terms of increasing light availability. Consequently, low grazing pressure combined with persistence in ungrazed patches could be hypothesised to buffer the population from impacts of changes in grazing regime (Maron and Kauffman 2006).
Temporal variation in population dynamics
In alpine environments, the climate is unfavourable for sexual reproduction in a high proportion of years (Bell and Bliss 1980), and large interannual variation is typical in seed production (Chambers 1995), seedling recruitment rates (Weppler et al. 2006) and plant growth (Callaghan et al. 1989). In line with this, we find that the λ of G. sylvaticum varies among years. Warm summers affect survival, growth, and flowering rates positively; thus, the temporal variation in λ can be linked to variations in local summer climate. Particularly low λ is found in the 2005–2006 transition, the summer of 2006 being warm and dry, and with early snowmelt (Table 1). Plant growth is positively related to July precipitation, which can explain the low λ in the dry year. Furthermore, grazed plants are comparably smaller during dry years (significant precipitation × grazing interaction; Table 3), which supports earlier findings of grazing effects being more severe in unfavourable years (Bastrenta et al. 1995). Ehrlén (1995b) found that λ of Lathyrus vernus was lower in a grazed compared to an ungrazed population in one of 3 years only, but few studies of vertebrate grazing have lasted long enough to address temporal variability in demographic rates (Evju et al. 2010; Rydgren et al. 2007).
In this study, sheep are stocked at densities that are representative of densities in Norwegian mountain pastures. Observable effects of grazing pressure on Geranium sylvaticum are low to moderate and, consequently, sheep grazing treatment is not a major determinant of variation in population growth rate. The increase in relative importance of clonal reproduction with enhanced grazing pressure (from no via low to high sheep density), does, however, suggest that increased clonal reproduction rates function as a tolerance mechanism towards grazing for this herb. However, in this alpine study area, variable climatic conditions seem in general to overrule effects of grazing on population growth rate.
This study was funded by the Research Council of Norway (Project 153601/432 and 134361/720). We are grateful to Synnøve Lindgren for help with field work in 2003, and to Dagrun Vikhamar Schuler at the Norwegian Meteorological Institute for providing interpolated climate data.
This article is distributed under the terms of the Creative Commons Attribution Noncommercial License which permits any noncommercial use, distribution, and reproduction in any medium, provided the original author(s) and source are credited.