Reproductive limits of a late-flowering high-mountain Mediterranean plant along an elevational climate gradient


Author for correspondence: Luis Giménez-Benavides Tel: +34 91 488 81 01 Fax: +34 91 664 74 90 Email:


  • • Mountain plants are particularly sensitive to climate warming because snowmelt timing exerts a direct control on their reproduction. Current warming is leading to earlier snowmelt dates and longer snow-free periods. Our hypothesis is that high-mountain Mediterranean plants are not able to take advantage of a lengthened snow-free period because this leads to longer drought that truncates the growing season. However, reproductive timing may somewhat mitigate these negative effects through temporal shifts.
  • • We assessed the effects of flowering phenology on the reproductive success of Silene ciliata, a Mediterranean high-mountain plant, across an altitudinal gradient during two climatically contrasting years.
  • • The species showed a late-flowering pattern hampering the use of snowmelt water. Plant fitness was largely explained by the elapsed time from snowmelt to onset of flowering, suggesting a selective pressure towards early flowering caused by soil moisture depletion. The proportion of flowering plants decreased at the lowest population, especially in the drier year. Plants produced more flowers, fruits and seeds at the highest population and in the mild year.
  • • Our results indicate that water deficit in dry years could threaten the lowland populations of this mountainous species, while high-altitude environments are more stable over time.


In alpine and arctic regions, snowpack and snowmelt timing, which are directly correlated to snowfall regime and ambient temperature (Groisman et al., 1994; Dye, 2002), are the two major factors determining soil moisture and (indirectly) the onset of the growing season, plant growth and annual rates of primary production (Walker & Webber, 1994; Kudo, 2003). In reproductive terms, they have been widely reported as drivers of flowering phenology and reproductive output (Billings & Mooney, 1968; Galen & Stanton, 1995; Stinson, 2004; Molau et al., 2005). Current global warming is affecting snowmelt timing and, subsequently, the reproductive output of high-mountain plants (Inouye & McGuire, 1991; Price & Waser, 1998; Inouye et al., 2002; Kudo & Hirao, 2006). However, latitudinal poleward shifts in the distribution range of plants are not always feasible because of orographic isolation, and upward shifts to higher summits are irremediably associated with a decrease in habitat area (Pauli et al., 2003). Moreover, lowland species are expanding their upper distribution limits, mainly because of water availability (Pigott & Pigott, 1993) and higher temperatures (Conolly & Dahl, 1970), and thus may push cryophilic plant populations to the verge of extinction (Grabherr et al., 1994; Pauli et al., 2003).

Analysis of the limitations on reproduction of high-mountain plants may provide us with valuable information on plants’ response ability and potential to counteract rapid climate change. The reproductive assessment of high-mountain plants should therefore consider environmental variability at several scales: altitudinal gradient, which controls the local climate; small-scale heterogeneity, which may mitigate adverse conditions or exacerbate restrictions; and temporal heterogeneity, which may be especially important in stochastic environments such as the Mediterranean. This environmental variability may cause changes in flowering phenology (Molau et al., 2005), pollinator frequency (Tøtland, 2001 and references therein), predator frequency (Galen, 1990; Freeman et al., 2003) and intensity of plant–plant interactions (Klanderund, 2004), affecting the overall performance of plants. Additionally, conditions for the reproduction and survival of plants should vary at broader scales. Regeneration processes are theoretically more suitable in the centre of the species’ distribution area than in the periphery (Lawton, 1993), thus viability assessments of peripheral plant populations in global warming scenarios have become a conservation concern. Several reproduction and regeneration studies have been carried out on marginal populations (McCarty, 2001), however, most have focused on northern plant limits (Arft et al., 1999; Castro et al., 2002; Castro et al., 2004; Kluth & Bruelheide, 2005; Norton et al., 2005), while they have rarely been conducted on southern boundaries (but cf. García et al., 1999; Peñuelas & Boada, 2003; Hampe, 2005). When latitudinal gradients run parallel to climatic gradients, special adverse conditions for regeneration often occur at the species’ southernmost distribution margin (in the northern hemisphere), mainly constrained by water deficit (Lesica & McCune, 2004; Hampe, 2005; Hampe & Petit, 2005).

Mediterranean-type mountains hold a great number of endemic species, many of which are restricted to the higher summits (Giménez et al., 2004). Many arctic and alpine species also find their southernmost limit in these mountains, and thus represent a global conservation priority (e.g. Sierra Nevada; Pauli et al., 2003). Although these ecosystems are among those predicted to be the most vulnerable to global warming in Europe (Thuiller et al., 2005), our knowledge is scarce about the potential threats of climate warming to Mediterranean habitats (Peñuelas et al., 2002, 2004; Peñuelas & Boada, 2003), and no studies have specifically focused on Mediterranean high mountains in above-timberline habitats. Extrapolation of arctic and alpine reproductive mechanisms and causal factors to Mediterranean high-mountain environments should be conducted with caution because specific constraints, such as the severe water shortage during the growing season, must be taken into account (Castro et al., 2004).

We examined the reproductive performance of Silene ciliata, a high-mountain Mediterranean plant, in contrasting environments along an altitudinal gradient in its southern margin. Our aim was to identify the main limitations on the reproductive process of its populations. Specifically, we aimed to quantify the phenological variation of the flowering event within populations and among sites to assess to what extent phenological shifts can cope with changing conditions. For this purpose, we modelled the overall reproductive response of the species using a large set of phenological, biotic and abiotic variables at contrasting spatial and temporal scales. We hypothesize that a lengthening of the snow-free period as a result of increasing temperatures may become a critical problem for plants occurring in Mediterranean mountains because of harsh summer drought. Furthermore, we expect warming conditions to be adverse for reproduction at the lowest-margin populations, especially in years with low precipitation and high temperatures. Survival under such conditions may be feasible only if reproductive timing is plastic enough to cope with short-term environmental changes, and/or evolutionary adaptation has time to act (Jump & Peñuelas, 2005).

Materials and Methods

Species and study area

Silene ciliata Poiret (Caryophyllaceae) is a perennial cushion plant, inhabiting main mountain ranges in the northern half of the Iberian Peninsula, the Massif Central in France, the Appenines and the Balkan Peninsula (Tutin et al., 1995). The plant reaches its southern latitudinal limit in central Spain. The species typically forms cushion rosettes of up to 2 cm high and 15 cm diameter. Flowering stems reach 15 cm high and bear one to five flowers. Hand-crossing experiments indicate that S. ciliata is a self-compatible species. Both fruit set and seed set are high under xenogamy and geitonogamy treatments. However, passive autogamy is totally restricted by a pronounced protandry. Fruit capsules open at the top when ripe, and up to 100 seeds per fruit are dispersed by stalk vibration. Seeds are relatively small (1.53 ± 0.49 mm diameter, 0.59 ± 0.06 mg weight). Although a small fraction is able to germinate immediately after dispersal (< 25% at 20°C), nearly 100% of dispersed seeds germinate after a cold-stratification period or contrasting high temperatures (Giménez-Benavides et al., 2005).

Three altitudes covering the local range of S. ciliata were chosen in Sierra de Guadarrama (Circo de Peñalara Natural Park), a south-west to north-east-running mountain range located 50 km north of Madrid (40°N, 3°W). The lowest altitude (1976 m, hereafter ‘L’) was in a moraine deposit at Hoya de Peñalara, a Pleistocene glacial cirque situated in the timberline zone. Vegetation is dominated by stunted pines (Pinus sylvestris) interspersed in a shrub matrix of Cytisus oromediterraneus and Juniperus communis ssp. alpina in a Festuca curvifolia pasture. This site is the lower altitudinal limit of the species. The intermediate altitude (2256 m, ‘I’) was located on the Dos Hermanas summit approx. 3 km from altitude L. At this site, a patchy xerophytic pasture dominated by F. curvifolia co-occurs with the high altitudinal limit of the Cytisus–Juniperus shrub formation. The highest altitude (2428 m, ‘H’) was located at Pico Peñalara, the highest peak of Sierra de Guadarrama approx. 3 km from altitude I. The summit flat areas and crests are covered by F. curvifolia fellfields, occasionally displaced by a moister Nardus stricta-dominated community where snow cover remains for longer. This fellfield community, which experiences extreme winds and is characterized by a high diversity of cushion plants, constitutes the main habitat of S. ciliata (for further details see Gavilán et al., 2002; Escudero et al., 2005).

Sierra de Guadarrama is characterized by a Mediterranean-type climate. Snowfall generally begins in October, and snowmelt concludes in May–June. Mean annual precipitation at the Navacerrada Pass weather station (40°46′ N, 4°19′ W; 1860 m), located 8 km south-west of Pico Peñalara, is 1350 mm and is concentrated from early October to late May. A pronounced drought season (< 10% of total annual rainfall) occurs from June to September (Palacios et al., 2003). For annual climate trends we used data from the above-mentioned station, as it had the only long climatic series available for the area (1960–2004, Instituto Nacional de Meteorología). In 2002–03, climatic data at each study site were provided by a network of portable weather stations of the Peñalara Natural Park.

Lacking direct measures of soil water content during our study, we calculated the daily reference evapotranspiration rate (ETo) as an estimation of water demand. We applied the FAO 56 Penman–Monteith equation (Allen et al., 1998), which uses standard climatic records of solar radiation, minimum and maximum air temperature, humidity and wind speed. As direct measurements of solar radiation were unavailable, we used the Angstrom formula to estimate it from the observed duration of sunshine hours (Allen et al., 1998). Calculations were carried out using CropWat 4 for windows (Clarke et al., 1998).

Plant monitoring

In 2002, three plots were randomly established at each altitudinal level. Plot location was selected to minimize differences in exposure (north-east), slope (0–10°) and human management (almost null) among sites. Plot size varied between 10 and 15 m2. At each plot, 55–60 individuals > 2 cm in diameter (therefore excluding seedlings) were randomly selected and tagged for a total sample size of 166 plants at L, 165 plants at I and 171 plants at H. Plant cushions located at a minimum distance of 2 cm from each other were considered different individuals. As the species does not propagate vegetatively, each labelled plant was considered a genet. A detailed measure of plant size, calculated as the horizontally projected area of the cushion in cm2, was obtained from digital images taken of each plant using ImageTool 3.0 (University of Texas). To estimate microsite conditions, a 25 × 25-cm quadrat was centred on each plant. Cover percentage of bare soil, rocks, other S. ciliata individuals, and any other plant species was calculated by digital image analysis. Each year the snowmelt date was estimated for each altitude from digital images taken at weekly intervals and was assigned to the first date when the picture displayed over 80% bare ground.

Flowering and fruiting phenology were monitored at weekly intervals for a total of 13 and 14 census days in 2002 and 2003, respectively. At each census, we recorded the number of open flowers and mature fruits on each plant, as well as the number of flowers and fruits with external signs of predation. To estimate seed production, mature and predated fruits were collected before opening to avoid loss of seeds by dispersal.

Flowering phenology was characterized for each plant by measuring six variables: (a) prefloration interval – number of days from snowmelt date to first open flower; (b) first flowering date – number of days from 1 January to first open flower; (c) maximum flowering moment – number of days from first open flower to day of maximum flower count on each plant; (d) duration – number of days the plant remained in bloom; (e) flowering synchrony – number of days when the flowering of one individual overlaps with the flowering of every other plant in the population. We applied the following function modified from Augspurger (1981):

image(Eqn 1)

where n is the number of plants, aij is the number of days individuals i and j are simultaneously in bloom, and bij is the number of days at least one of them is in bloom. Si ranges between 1, when flowering completely overlaps, and 0, when there is no synchrony.

Female reproductive success was assessed for each plant by means of five variables: flower number, fruit number and seed number (expressed as total number per plant), fruit set (proportion of flowers setting fruit), and seed set (average proportion of ovules setting seed in each ripe fruit). Additionally, a measure of total yield per unit of plant area was calculated for each plot (n = 9) by determining the ratio between total seed number and total plant area (in cm2). This allowed us to assess differences in seed production among populations, irrespective of differences in plant size.

Statistical analysis

Table S1 in Supplementary material summarizes the different statistical techniques used to analyse the processes that sequentially affect reproductive success.

Flowering probability

To determine the dependence of flowering ability on plant size (Lacey, 1986) and the effect of altitude, we modelled flowering probability by fitting generalized linear mixed models (GLMMs). Models were built for each year with plant size and altitude as fixed variables and plot as a random variable nested in altitude. As the response variable was a probability ranging from 0 to 1 (not flowering or flowering), we used a binomial estimator, using a logit link function and setting the variance to ‘mean(1 − mean)’. Effects of random factor were tested using the Wald Z-statistic test, and fixed factors were tested with F-tests. The degrees of freedom in each level of variation (595 plants and three plots per altitude) were estimated by Satterthwaite's method (Littell et al., 1996).

We also estimated flowering probability curves by fitting a new GLMM for each altitude and year separately, maintaining plant size as a fixed variable and plot as a random variable. As the plot random variable was always nonsignificant, we plotted the curves using the estimated parameter for plant size and the formula:

image(Eqn 2)

where µ is the intercept with the x-axis, which is related to the threshold size for reproduction, and α is the slope of the curve, which is related to the percentage of reproductive plants (Méndez & Karlsson, 2004). Flowering probability curves were compared between years by including year as a fixed factor.

Structural equation modelling: phenology and female reproductive success

The relationships among plant size, microsite variability (intra- and interspecific neighbour cover and bare soil cover), phenological variables and reproductive success were studied with a nonstandard structural equation model (SEM) with a latent variable. This technique allows the analysis of complex multivariate causal schemes represented in a path diagram. We formulated our path diagram based on our a priori hypothesis of causal links between variables (Fig. 1). The SEM tests whether our data are supported by the underlying mechanisms that our specific model describes (Shipley, 1999). Our working model proposes that the absolute fitness of each plant (expressed as fruit number) depends ultimately on plant size, flowering phenology components and microsite conditions (Fig. 1). We hypothesized that onset of flowering would depend on reaching a resource threshold. Thus larger plants would have an earlier flowering start (shorter prefloration interval). Plant size might also determine flower number and flowering period, because larger plants may invest more resources in reproduction. We also expected the prefloration interval to affect flower number, both directly and/or through duration, as early-flowering plants can have longer flowering periods than late-flowering plants. We considered prefloration interval and duration to be important phenological traits affecting synchrony among individuals, which ultimately control fruit set enabling cross-pollination. Fruit set could also be determined by flower number (positively in terms of attractiveness to pollinators, or negatively because of resource limitation). Finally, plant microsite characteristics were added to the model as a latent variable to test their effect on reproductive output. Latent variables are not measured directly, but can be expressed in terms of one or more directly measurable variables called indicators (Loehlin, 1992). We constructed the latent variable as an estimator of ‘microsite suitability’, which comprised abundance of surrounding plants (both intra- and interspecific plant cover variables) as an estimator of potential plant–plant interactions, and available space for plant development (soil cover variable). Microsite suitability was expected to influence flower number, fruit set and fruit number, mainly through resource limitation and opportunities for cross-pollination. There is also an evident relationship between plant size and microsite characteristics, but as the direction of causality is difficult to interpret, we modelled this relationship as an unanalysed correlation (Mitchell, 1993).

Figure 1.

Path diagram representing hypothesized causal relationships among phenotypic traits, microsite characteristics and plant fitness of Silene ciliata at contrasting altitudes for 2 yr. Positive effects, solid lines; negative effects, broken lines. The effects of unexplained causes are expressed by external arrows to each dependent variable. Arrow widths are proportional to adjacent standardized path coefficients. Path coefficients nonsignificantly different from zero are omitted. Fit statistics (nonnormed fit index, NNFI; comparative fit index, CFI; P; χ2) and sample size (N) are given at the upper-left corner of each path. Microsite suitability is a latent variable.

The model was evaluated separately for each altitude and year. For each altitude, data from the three plots were pooled and only plants that flowered were included in the analyses (Table 1). Sample size, ranging from 63 to 146, was at least five times the number of variables to be estimated, as recommended by Loehlin (1992). As the estimation method in SEM depends on the sample covariance matrix, which is a poor estimator for distributions with high kurtosis (Bollen, 1989), we removed outliers when necessary. We also calculated the variance inflation factor (VIF) for each variable to check for multicollinearity. As all VIF values were < 10 (Petraitis et al., 1996), we incorporated all variables in the model. The null hypothesis of the model is that the observed and predicted covariances are equal. Standardized path coefficients (equivalent to standardized partial regression coefficients) were estimated using the maximum likelihood method. This method is considered the best option when deviations from multivariate normality are severe and the sample size is not very large (Finch et al., 1997). Microsite suitability is an unobserved variable with no unit of measurement. To solve this scale indeterminacy problem, we made it equal to the unit of measurement of the indicator variable that best represented the latent construct (that with the largest standardized coefficient in a preliminary exploratory analysis) by fixing its path coefficient to 1 (Hatcher, 1994).

Table 1.  Reproductive output, phenological traits and microsite conditions of Silene ciliata in Sierra de Guadarrama (Madrid)
Low (n = 166)Intermediate (n = 165)High (n = 171)Low (n = 166)Intermediate (n = 165)High (n = 171)
  • Values expressed as mean ± SD except flowering plants, expressed in number (percentage of total) and snowmelt date, expressed as date.

  • *

    , Variables used to construct the latent variable ‘microsite suitability’ for structural equation models.

Flowering plants108 (65.0%)126 (75.7%)146 (85.8%)63 (37.9%)120 (72.7%)148 (86.5%)
Plant size (cm2)  38.4 ± 30.4  33.0 ± 32.3  28.0 ± 23.0  38.4 ± 30.4  33.0 ± 32.3  28.0 ± 23.0
Flowering plant size (cm2)  42.5 ± 31.1  37.2 ± 35.2  30.2 ± 24.0  47.0 ± 33.9  37.6 ± 35.3  29.9 ± 23.9
Flower number  4.22 ± 6.14  4.25 ± 4.87 11.72 ± 12.93  1.82 ± 3.65  4.08 ± 4.99  8.65 ± 9.67
Fruit number  2.56 ± 3.84  1.02 ± 1.70  3.55 ± 4.55  1.19 ± 2.38  1.87 ± 2.39  3.24 ± 3.52
Seed number127.54 ± 131.97 56.16 ± 62.94138.35 ± 143.00 90.87 ± 106.89 87.27 ± 72.91119.46 ± 93.42
Fruit set  0.37 ± 0.35  0.24 ± 0.32  0.30 ± 0.31  0.24 ± 0.34  0.36 ± 0.36  0.40 ± 0.35
Seed set  0.32 ± 0.22  0.22 ± 0.21  0.19 ± 0.17  0.27 ± 0.20  0.25 ± 0.21  0.23 ± 0.17
First flowering date (d)207.83 ± 12.47215.73 ± 12.41218.09 ± 11.39191.14 ± 8.99195.53 ± 10.83196.32 ± 11.40
Prefloration interval (d)  75.8 ± 12.5  75.3 ± 12.4  71.1 ± 11.4  86.5 ± 9.7  84.4 ± 9.1  76.7 ± 8.1
Max. flowering moment (d)  34.53 ± 10.76 41.56 ± 14.45 48.44 ± 13.70 14.30 ± 9.07 16.98 ± 12.90 16.34 ± 14.89
Duration (d) 10.40 ± 10.45 13.37 ± 13.61 19.02 ± 15.00  7.42 ± 9.11  9.31 ± 9.73 11.18 ± 9.70
Synchrony  0.26 ± 0.10  0.25 ± 0.11  0.29 ± 0.10  0.21 ± 0.10  0.29 ± 0.11  0.32 ± 0.10
Neighbour cover (%)*
 Intraspecific   3.8 ± 4.1   2.2 ± 2.9   1.6 ± 2.6   3.5 ± 4.2   2.2 ± 3.0   1.5 ± 2.5
 Interspecific  57.1 ± 18.2  46.6 ± 21.1  41.8 ± 17.0  55.8 ± 17.3  46.9 ± 20.5  41.7 ± 16.9
Bare soil cover (%)*  20.4 ± 18.1  12.4 ± 10.2  13.6 ± 9.9  21.5 ± 18.2  12.1 ± 10.0  14.0 ± 9.9
Snowmelt date13 May21 May28 May19 April28 April9 May

The degree of fit between observed and predicted covariance structures was first assessed by a χ2 goodness-of-fit test. A significant χ2 indicates that the model does not properly fit the data. However, this test may show inadequate statistical power when data depart from multivariate normality and sample sizes are small (Loehlin, 1992). A χ2 : df ratio < 2.0 is usually considered an acceptable fit to data (Hatcher, 1994). We used two alternative fit indices that provide an accurate fit regardless of sample size: the comparative fit index (CFI) (Bentler, 1989) and the nonnormed fit index (NNFI) (Bentler & Bonet, 1980). Values > 0.90 and 0.81, respectively, indicate a good fit compared with a null model that assumes independence among all variables (Hatcher, 1994). The significance of each individual path coefficient was subsequently assessed by a multivariate Wald test (P < 0.05). This test locates the path coefficients that can be eliminated without significantly increasing the χ2 value of the model. The effect of unexplained causes on each variable is measured as (1 − R2)1/2, with R2 being the coefficient of determination (the proportion of observed variance explained by each equation). Model comparisons between altitudes and years were systematically conducted by overimposing path diagrams. Computation of the structural equations was performed using the CALIS estimation procedure with the LINEQS statement (sas Statistical Package 1990, SAS Institute, Cary, NC, USA). For a complete description of structural equation models in ecology see Mitchell (1993); Shipley (1999); Iriondo et al. (2003).

Models for total seed production and seed set

Seed production and seed set were not included in the SEM because sample size (number of plants bearing seeds) was below the recommended threshold (Loehlin, 1992). Instead, we fitted GLMMs for these components of plant fecundity. Seed production (expressed per unit of plant area to avoid plant-size bias) was fitted using the gamma estimation, as Poisson resulted in a high dispersion. A ‘log’ link function was used, setting the variance to ‘mean’ (Venables & Ripley, 1998). Year, altitude, year × altitude, fruit number, seed set and the indicator variables of microsite suitability (intra- and interspecific neighbour cover and soil cover) were included as fixed variables, and plot as a random variable nested in altitude. For seed set, we fitted the mixed model using the binomial estimation (probability from 0 to 1), a ‘logit’ link function, and setting the variance to ‘1 − mean’. We included the following fixed variables: year, altitude, year × altitude, plant size, fruit number, synchrony, and the indicator variables of microsite suitability. Plot was again included as a random variable nested in altitude. Effects of random and fixed factors were tested as described above.


Annual trends and microclimate at study sites

In the past 45 yr, mean annual temperature increased approx. 1.8°C in the study region (r2 = 0.41, P < 0.001). This warming was especially pronounced in the past 35 yr, as stated for the whole planet (IPCC, 2001). Temperature increased 2°C in the winter months (October–March) and 1.5°C in the growing season (April–September). Seasonal rainfall revealed no significant downward decline (Fig. 2a,b). Duration of snow cover, expressed as the accumulated number of days that snow covered the ground during the thaw season (from April to June), decreased 19.7 d in the past 45 yr (Fig. 2c). This decrease amounted to 31.3 d when only the past 35-yr data series was used in the regression (r2 = 0.32, P < 0.001).

Figure 2.

Linear trends in climatic variables over the period 1960–94 at Puerto de Navacerrada weather station (1860 m). Mean total precipitation and mean temperature over (a) winter season (October–March) and (b) growing season (April–September). (c) Mean days of snow cover during snowmelt (April–June). The last two values on each plot (○) correspond to the observed years (2002 and 2003).

Temperature, precipitation and snowmelt date varied greatly between the two years of this study (Table S2). The year 2002 had average temperatures and summer rainfall, whereas 2003 was relatively cold and rainy in winter, but very hot and dry in the growing season due to the most severe known European heatwave (Beniston & Stephenson, 2004; Schär et al., 2004). Maximum temperatures of the complete climate series were reached in summer (Fig. 2b). No rainy days were recorded in July, and only one in August. As a consequence, plants emerged from snow approx. 22 d earlier in 2003. Along the altitudinal gradient, the lower site was an average 1.8°C warmer than the intermediate site, which was 1.1°C warmer than the higher site. In both years, snow disappeared from low altitude plots approx. 9 d earlier than from intermediate plots, which emerged approx. 9 d earlier than high-altitude plots (Table S2). Monthly ETo was comparatively higher in 2003 than in 2002, mainly due to higher temperatures and lower air humidity (Fig. 3). These interannual differences were especially evident in June and September rather than in July, when annual maximum ETo was reached in both years.

Figure 3.

Precipitation and daily evapotranspiration rates (ETo) of the study sites in 2002 and 2003. ETo is usually expressed in mm d−1, but here is presented in mm month−1 to allow comparison with monthly precipitation records.

Flowering probability

In 2002, minimum size at reproduction coincided with the minimum monitored size (plant size ranged from 0.8 to 216 cm2) at all altitudes. Thus we considered every plant a potential monitored as a reproductive individual, and the percentage of flowering plants was calculated with respect to all individuals. Silene ciliata showed important altitudinal and interannual differences in the percentage of flowering plants. In altitudes I and H, S. ciliata had a high rate of flowering plants in both years, whereas the proportion of flowering plants at altitude L was very low, especially during the second year (Table 1; Fig. 6a).

Figure 6.

Altitudinal differences in reproductive traits of Silene ciliata in 2002 (solid line) and 2003 (dashed line). All variables are mean values per plant except percentage of flowering plants and seeds cm−2, which are expressed as mean values per plot (n = 3 per altitude). Error bars, SD.

Results of the complete GLMM showed a lower probability of flowering in altitude L than in H and I (Table 2). Furthermore, flowering probability was size-dependent in both years (Table 2). Altitudinal and interannual differences can be easily explored in the associated flowering probability curves (Fig. 4). It is interesting to note the significant drop in flowering probability experienced at altitude L between years (F = 24.07, P < 0.0001), while I and H remained similar (F = 0.58, P = 0.448 and F = 0.14, P = 0.708, respectively). On the other hand, an important shift took place in the flowering probability of smaller individuals from one year to the next (see differences in the intercept point of curves). In 2002, all populations had a flowering probability of approx. 0.6 for the smallest monitored plant, while in 2003 flowering probability dropped to almost 0.25 in altitude L, remained nearly the same in I, and increased to 0.9 in H.

Table 2.  Generalized linear mixed model for flowering probability of Silene ciliata in 2002 and 2003
EffectSolution for fixed effectsDeviance change
CoefficientSDdf t P F P
  1. Altitude was added as a categorical variable (2 df). The random variable plot, nested in altitude, was nonsignificant for both models (0.0087 ± 0.0726, Z = 0.12, prob. Z = 0.4519 in 2002; 1.0272 ± 0.7769, Z = 1.32, prob. Z = 0.0931 in 2003).

Alt L−1.44310.2995  6.61−4.82 0.0023
Alt I−0.68410.3041  7.19−2.25 0.0583
Alt H
Altitude     11.88 0.0099
Plant size 0.02740.0062  349 4.41< 0.000119.45< 0.0001
Alt L∠3.38610.9092  4.92−3.72 0.0141
Alt I−1.60050.9056  4.84−1.77 0.1393
Alt H
Altitude      6.99 0.0412
Plant size 0.02600.0051496 5.04< 0.000125.39< 0.0001
Figure 4.

Size-dependent flowering probability in Silene ciliata. Curves drawn according to intercept and slope parameters estimated by generalized linear mixed models, as described in the text.

Flowering phenology

Populations showed marked altitudinal and interannual differences for all the phenological variables. The flowering season started at the end of June and extended until the end of September in 2002, but shortened substantially to early September in 2003. In 2002, populations at altitudes I and H showed a considerable delay in the first flowering day (8–10 d) compared with L, but this difference disappeared in the extremely dry and hot 2003 (Fig. 5). Flowering duration increased progressively with altitude in 2002, but shortened at all sites in 2003. Prefloration interval was approx. 2.5 months and did not differ among altitudes within years, but was slightly longer in the second year. Maximum flowering moment took place approximately in the middle of the mean flowering period at each altitude in 2002, but occurred either earlier or at the same time in 2003 (Fig. 5). Flowering synchrony was relatively low at all altitudes. Plants of altitude H had the highest flowering synchrony each year (Table 1).

Figure 5.

Altitudinal variation in flowering phenology among populations of Silene ciliata in 2002 and 2003. Horizontal black bars refer to start, duration and finish dates of flowering period per altitude and year. Dotted lines represent absolute values; solid lines, range of mean values; diamonds, mean flowering moment. Black arrows, snowmelt date.

Female reproductive success

In general, plants performed better at the upper distributional limit. Flower number, fruit number and seed number reached maximum values at altitude H in both years (Table 1; Fig. 6). Although all populations had a similar phenological pattern, plants at altitude I had comparatively lower values than at altitude H for all reproductive traits, especially in 2002. Significant interannual differences were also found. While plants at altitude L had the highest fruit set in 2002, more fruits per flower were produced at altitudes I and H in 2003, the harsh year (Fig. 6c). Results of the mixed model for seed production reinforced the differences observed (Table 3). Seed production at the L and I sites was approximately half that observed at H. As differences among altitudes did not remain consistent between years (Fig. 6e), the model added significance to the year × altitude interaction term. Fruit number, seed set, and all the microsite suitability variables except intraspecific neighbour cover also explained a fraction of variability in seed production (Table 3). On the other hand, seed set was affected significantly by the year × altitude interaction, and also by plant size, fruit number and synchrony (Table 4; Fig. 6d).

Table 3.  Generalized linear mixed model for seed production of Silene ciliata (seed number per cm2 plant) in 2002 and 2003
EffectSolution for fixed effectsDeviance change
CoefficientSDdf t P F P
  1. The random variable plot, nested in altitude, was nonsignificant (0.0334 ± 0.0361, Z = 0.92, prob. Z = 0.1775).

Year (2002) 0.03420.1422  1 0.24  0.8495
Year (2003)
Year  0.29  0.6844
Alt L−1.18560.2475  2−4.79  0.0409
Alt I−0.38210.2126  2−1.80  0.0214
Alt H
Altitude 15.41  0.0509
Year × altitude  3.17  0.0432
Fruit number 0.13230.011038911.96< 0.0001143.09< 0.0001
Seed set 2.52810.2599389 9.73< 0.0001 94.62< 0.0001
Soil cover 0.02470.0047389 5.17< 0.0001 26.71< 0.0001
Neighbour cover
 Intraspecific13.74687.1807389 1.91  0.0563  3.66  0.0563
 Interspecific 0.01630.0032389 5.05< 0.0001 25.50< 0.0001
Table 4.  Generalized linear mixed model for seed set of Silene ciliata in 2002 and 2003
EffectSolution for fixed effectsDeviance change
CoefficientSDdf t P F P
  1. The random variable plot, nested in altitude, was nonsignificant (0.0239 ± 0.0399, Z = 0.60, prob. Z = 0.2744).

Year (2002)−0.27390.1470  1−1.86  0.3136
Year (2003) 
Year 0.37  0.6509
Alt L 0.09980.2621  2 0.38  0.7399
Alt I 0.20560.1672  2 1.23  0.3438
Alt H
Altitude 2.68  0.2721
Year × altitude 3.02  0.0500
Plant size−0.07670.0019389−4.04< 0.000116.30< 0.0001
Fruit number 0.04950.01415389 3.50  0.000512.25  0.0005
Synchrony−3.94510.5218389−7.56< 0.000157.16< 0.0001
Soil cover−0.00740.00541389−1.37  0.1702 1.89  0.1702
Neighbour cover
 Intraspecific−8.76517.1927389−1.22  0.2237 1.49  0.2237
 Interspecific−0.00190.0036389−0.53  0.5995 0.28  0.5995

Within-site variation

The SEM provided a good overall fit for all altitudes and both years studied (Fig. 1). Five of the six path models had an excellent fit with observed data, as indicated by their nonsignificant χ2 (P > 0.05) and χ2 : df ratio < 2 (range 0.9–1.3), and by goodness-of-fit indices (NNFI and CFI) > 0.90. However, the model for population H in 2003 was rejected by both the χ2 test (χ2 = 44.24, df = 26, P = 0.014, χ2 : df = 1.7) and by NNFI and CFI (0.79 and 0.88, respectively), although the χ2 : df ratio was < 2 (Hatcher, 1994). The strength of all six models rested mainly on the sequence of causal relationships from plant size and prefloration interval to duration, from duration to flower number, and from flower number to fruit number. Prefloration interval had a negative effect on duration, which largely determined flower number (coefficients > 0.43 in all cases, P < 0.001) and synchrony. Although direct effects of prefloration interval on flower number were nonsignificant in all models but one, indirect effects through duration were negative and relatively high in some models (−0.35, −0.11, −0.18 in 2002 and −0.41, −0.31, −0.20 in 2003, at altitudes L, I and H, respectively). On the other hand, plant size also influenced duration in all cases, but surprisingly did not affect flower number directly in some models. In 2002, plant size negatively affected prefloration interval in altitudes I and H.

Fruit number was significantly affected by flower number and fruit set in all models. However, neither synchrony nor microsite suitability influenced fruit set, except in the second year at altitude H, where variables accounted for equal but opposite effects. Finally, no causal relationships were found between microsite suitability and other reproductive traits. Nevertheless, the indicator variables of the latent construct gave us some valuable information about microenvironmental differences at each site. The effect of soil cover was highly positive at altitude L, but negative at I and H, and the opposite took place with interspecific neighbour cover. As noted in Table 1, mean values of both variables were comparatively higher at L.


Recent climate trends in this mountainous region suggest a very rapid warming (Fig. 2), as reported worldwide (IPCC, 2001). As shown by Agustí-Panareda & Thompson (2002) and Granados & Toro (2000) by estimates from chironomid fossil records, sharp climate change during the past century has been the norm in this part of the Iberian Peninsula. However, the recent increase in temperature is unprecedented in these mountains, and the ecological consequences are extremely difficult to forecast. The first field evidence of this recent warming in this region was reported by Sanz-Elorza et al. (2003), who pointed out the ongoing replacement of the typical high-mountain grassland F. curvifolia by C. oromediterraneus–J. alpina, a scrubland from lower altitudes. Wilson et al. (2005) found a marked contraction of lower elevational limits for 16 butterfly species in the same area. Similar altitudinal shifts in species distributions linked to climate warming have been reported in other Mediterranean mountains at a lower altitude (Peñuelas & Boada, 2003), and in mountains around the globe (Grabherr et al., 1994; Parmesan et al., 1999; Dullinger et al., 2003).

One of the most significant effects of climate warming in high mountains is related to snow cover duration. The onset of the growing season in mountains depends mainly on snowmelt date, which has become substantially earlier in central Spain (Fig. 2c). This seems to be related mainly to an increase in spring temperatures, as mean seasonal precipitation has not dropped significantly (Fig. 2b). In addition to mean warming trends, one of the major concerns of current climate change is the increasing frequency and intensity of extreme climate events (Easterling et al., 2000; Schär et al., 2004). During the course of our phenological study in summer 2003, the greatest heatwave registered in the past 150 yr took place in Europe (Schär & Jendritzky, 2004). In central and south-western parts of the continent, heat and drought conditions caused severe damage to cultivated and wild plant species (Beniston & Stephenson, 2004). This unexpected coincidence allowed us to evaluate the species’ reproductive response under an extreme climatic event, and gave us an opportunity better to infer the potential threats of global warming through the rising intensity and frequency of extreme events (Gutschick & BassiriRad, 2003).

Flowering phenological pattern

Silene ciliata starts flowering over 2 months after snowmelt, once the annual leaf crop has been produced (Fig. 5). According to the classification of life-history strategies made by Molau (1993) for arctic and alpine plants, S. ciliata would be a typical late-flowering species. The plant started flowering on similar dates in both years monitored, despite a much earlier snowmelt date in 2003 (Fig. 5). This indicates that in S. ciliata, as in many late-flowering alpine plants, the onset of blooming is controlled by photoperiodic triggers and not by snow conditions or soil moisture (Körner, 1999; Keller & Körner, 2003). This late-flowering pattern is unusual in Mediterranean-type climates where flowering is bimodally concentrated in spring/early summer and autumn to avoid the difficulties of severe summer drought (Petanidou et al., 1995; Thompson, 2005).

Molau (1993) stated that in late-flowering species the reproductive output may be limited by early snow events in late summer/early autumn, shortening the time available for seed ripening. Thus it has been hypothesized that late-flowering species in temperate mountains may enhance their reproductive success under a climate-warming scenario because of the extended growing season (Molau, 1993; Alatalo & Tøtland, 1997; Molau et al., 2005). However, such a potential advantage of a longer growing season may be seriously limited if it is coupled with a parallel intensification of evapotranspiration rates and topsoil desiccation (Walker et al., 1995; Starr et al., 2000). In S. ciliata, we found a markedly worse reproductive performance in 2003 at low altitude (Fig. 6). Comparing the evapotranspiration rates with the input of water (monthly precipitation) during the course of our study, we found a more negative water balance in 2003 (Fig. 3). The difference in soil water deficit between years could be even greater, taking into account the finding that in 2003, water supply from snowmelt finished approx. 1 month earlier than in 2002 (Fig. 5; Table 1). Moreover, direct measures of soil water content performed at the same sites in 2004 (an average climate year) confirmed the existence of severe topsoil water limitation during summer (< 5% soil water content in July), and water availability was positively correlated with altitude (L.G.-B., unpublished data). These results therefore suggest that summer drought in Mediterranean high mountains exists, and may impose important reproduction limitations on late-flowering plants, especially at lower altitudes (Cavieres et al., 2006). So the assumption that late-flowering arctic and alpine plants benefit under warming conditions should be reconsidered for Mediterranean mountains, where drought can truncate the extended growing seasons.

Effect of size on reproductive success

Silene ciliata showed notable size-dependence at several stages of the reproductive process. Thus plant size played a major role in flowering probability (Table 2) and indirectly affected flower and fruit production through the onset and duration of the flowering period (Fig. 1).

Size-dependence of flowering probability has been cited by many authors as an important factor controlling plant fitness (Méndez & Karlsson, 2004). However, intraspecific variation of such size-dependence among sites and years has been little studied, especially in relation to environmental factors (Méndez & Karlsson, 2004 and references therein). Our results support the idea that flowering probability in S. ciliata varies substantially on a spatiotemporal scale that is influenced by environmental conditions. A marked decrease in flowering probability occurred at its lower altitudinal limit, especially in 2003, a year with an extremely dry summer (Fig. 4). Differences in the physiological status of plants probably induced by drought stress may explain why the fraction of flowering plants varies among sites and between years. Likewise, if flowering in a given year is determined by reaching a resource threshold, then larger plants will have a higher chance of flowering.

We did not find a direct causal relationship between plant size and total flower number in 2003 at any altitude, or at population L in 2002 (Fig. 1). The existence of a strong relationship between plant size and flower production constitutes a general pattern in most plant species and habitats (Mitchell, 1994; Ollerton & Lack, 1998), including the Mediterranean (Herrera, 1991; Albert et al., 2001). However, plant size exerted a constant positive effect on duration which, in turn, had a significant positive effect on flower number and flowering synchrony. Thus, despite the lack of a direct effect, as shown in our SEM diagrams, plant size indirectly controls flower production: larger plants have a longer flowering period because of higher resource investment, and this corresponds to higher flower production.

Importance of the prefloration interval

Although flowering start in high mountains is mainly predicted by snowmelt date (Inouye & McGuire, 1991; Kudo, 1991; Inouye et al., 2002; Stinson, 2004), variation in flowering start has also been reported as a heritable trait (Zimmerman, 1988). However, this variation can also be affected by other factors such as individual plant size (Ollerton & Lack, 1998; Stinson, 2004). Our results show that plant size partially determines the prefloration interval (flowering start) in two of the six scenarios presented (Fig. 1). The absence of a direct causal relationship between plant size and prefloration interval in some models could be simply a consequence of the size-dependence of flowering probability. Plant size probably loses its predictive power on prefloration interval in those scenarios where smaller plants simply did not flower, or had a low flowering probability.

The strength of all six SEMs rested mainly on a sequence of causal relationships linking prefloration interval, duration, flower number and fruit number. A close link between prefloration interval and flower and fruit production has been found in other alpine (Stinson, 2004) and Mediterranean-type species (Volis et al., 2004). Tøtland (1999) found no evidence of selection acting on flowering time in the alpine Ranunculus acris subjected to experimental warming, but his previous studies in the same area found higher plant fitness in early-flowering plants, suggesting that selection could differ between seasons depending on temperature conditions (Tøtland, 1994, 1997). Stanton et al. (2000) concluded that phenotypic selection may promote the avoidance of stressful conditions for reproduction through earlier flowering. In an ingenious experiment that manipulated the reproductive timing of an annual species, Griffith & Watson (2006) found that evolution to early reproduction would allow plants to reproduce beyond their actual range. Similarly, our results suggest that a short prefloration interval could be an important phenotypic trait under selection, because some fitness components (number of flowers and consequently number of fruits and seeds) increase markedly in early-flowering plants.

In S. ciliata, a shorter prefloration interval also led to higher flowering synchrony among plants, but contrary to our expectations, synchrony did not contribute to improving fruit set but produced a negative effect in population H in 2003 (Fig. 1), and globally over seed set (Table 4). This could be caused by low pollinator availability and therefore competition among simultaneously flowering plants, as reported in other alpine environments (Tøtland, 2001). Our SEM analyses also show that the higher the flower number, the lower the fruit set in the two best scenarios (populations M and H in 2002; Fig. 1). Resource limitation may explain these findings. When a plant produces more flowers than it can support, flower abortion takes place even under optimal conditions (Holland et al., 2004; Volis et al., 2004). Nevertheless, despite occasional higher fruit-set and seed-set values at population L, total yield at the population level was significantly higher for population H because more individuals flowered (Table 1; Fig. 6e). This fact also has important genetic consequences, as total seed production of higher populations comes from more individuals, reinforcing the genetic variability of the population's offspring.


Our results showed that S. ciliata presents contrasting reproductive success and fitness cost among altitudes and between years at its southern marginal limit. At the lowest altitudinal edge, the species showed a comparatively marked reproductive failure in the extremely hot and dry year, whereas the intermediate and high altitudinal limits tended to provide more stable reproduction over time. Such differences were mediated mainly by the percentage of adult plants that flowered, and by the flowers produced per reproductive plant at each population in each year.

Our findings highlight the existence of different response patterns to climate warming around the globe. Although late-flowering alpine and subarctic plants may enhance their reproductive performance in warming scenarios (Molau, 1993; Alatalo & Tøtland, 1997; Molau et al., 2005), our results suggest that in high Mediterranean mountains increased summer droughts imposed by warming trends could limit plant reproductive success.

Reproductive timing strongly influences natural selection and evolutionary potential (Donohue, 2005) and, as Stinson (2004) points out, the extent to which alpine plants can endure climate change may depend on their potential for adaptive plasticity in flowering phenology under new environmental conditions. In S. ciliata, although earlier flowering improves plant fitness, the flowering start seems to be photoperiodically constrained toward the end of spring, preventing efficient use of the limited water supply. Flexibility to exploit selective pressure towards early flowering in evolutionary terms is limited, especially in marginal populations when faced with rapid warming and frequent extreme heat events. Thus the time scale may not be long enough to allow for local adaptation (Etterson & Shaw, 2001; Jump & Peñuelas, 2005).


The authors thank the staff of Parque Natural de las Cumbres, Circo y Lagunas de Peñalara for permission to work in the area; M. Gómez-Espasandín and S. González for help with seed counting; and R. García, C. Fernández, Dr A. L. Luzuriaga and Dr M. J. Albert for help with the field work, often under less than ideal conditions. They also thank Dr D. Palacios for providing the digital images for snowmelt date estimation; Dr D. Ackerly, Dr D. Inouye, Dra. M.J. Albert and two anonymous referees for helpful comments on earlier drafts of the manuscript; and L. De Hond for her linguistic assistance. This work was supported by the Spanish Government CICYT project REN2003-03366.