Density-dependent growth and survival in a natural population of the facultative biennial Digitalis purpurea

Authors

  • NINA SLETVOLD

    Corresponding author
    1. Department of Biology, Division of Botany and Plant Physiology, University of Oslo, PO Box 1066 Blindern, N-0316 Oslo, Norway
    • Present address and correspondence: Nina Sletvold, Natural History Museum, University of Oslo, PO Box 1172 Blindern, N-0318 Oslo, Norway (tel. +47 22 85 16 12, fax +47 22 85 18 35; e-mail: nina.sletvold@nhm.uio.no).

    Search for more papers by this author

Summary

  • 1Density-dependent effects on vital rates may vary in both magnitude and direction at different stages of the life cycle. In monocarpic perennials, however, it is often assumed that recruitment is the stage most affected by density.
  • 2The spatial pattern of newly emerged individuals of the facultative biennial Digitalis purpurea was recorded and followed in five 0.5-m2 plots censused twice during each of three seasons.
  • 3To examine effects of local density on growth and survival, a plant's neighbourhood was defined as the number of individuals inside a circle of fixed radius around it. Species identity of the nearest neighbour was established to see whether the effects of intra- and interspecific interactions differed.
  • 4Relative growth rate (RGR) was negatively related to local density during both of the first two summers, but not during winter. In 1998 the negative effects were stronger when the closest neighbour was conspecific.
  • 5Contrasting patterns of size-dependent growth were found during summer and winter, with summer RGR an increasing, and winter RGR a decreasing, function of size.
  • 6Summer survival was generally high, and was positively related to spring size. There was a significant negative effect of increasing local density on survival probability in 1998. No effect of neighbour identity was detected.
  • 7Winter survival was low, and mainly dependent on autumn size, with no significant effects of local density in any season. During the first winter, individuals with a conspecific as the nearest neighbour had lower survival probabilities.
  • 8There was no evidence of a tradeoff between summer RGR and survival probability.
  • 9Density-dependent effects may be significant beyond the recruitment stage in monocarpic perennials such as D. purpurea. Density dependence was strongest at early life stages although, because effects on growth persisted into the second year, it is likely that local density may influence timing of reproduction in this population. The use of detailed studies investigating the timing and magnitude of density dependence across the entire life cycle may provide new insight into life history evolution.

Introduction

Monocarpic perennials have in recent years been favourite systems in the modelling of plant life history evolution (de Jong et al. 1989, 2000; Rees et al. 1999; Rees & Rose 2002; Rose et al. 2002; Childs et al. 2004). Most monocarpic perennials can be described as facultative biennials, as they can complete their life cycle within 2 years, but quite commonly delay reproduction until the third or later year (e.g. de Jong & Klinkhamer 1988; Klinkhamer et al. 1987a,b), and timing of reproduction has been modelled with high accuracy for several species (e.g. Kachi & Hirose 1985; Rees & Rose 2002). Some of these plants share a ‘seed-bank-oriented’ life history (Kachi 1990), with highly synchronous germination following disturbance. High initial seedling densities suggest that the onset of intraspecific competition will be almost immediate in these species, and there will therefore be high mortality in early phases of the life cycle. Most modelling attempts have indeed assumed a density-dependent recruitment phase (Metcalf et al. 2003) and, in the few studies in which the effect of density dependence actually has been quantified, it has been found to act primarily on smaller/younger plants (Rees et al. 1999; de Jong et al. 2000; Rose et al. 2002).

The degree, and even direction, of density dependence may vary across the life cycle (Howard & Goldberg 2001). Such differences in timing and/or direction may have large effects when addressing questions concerning both population dynamics and life history evolution (Mylius & Diekmann 1995; McPeek & Peckarsky 1998). In addition, effects of density may depend on the actual demographic rate measured. Rose et al. (2002) found no significant effects of density on mortality of the monocarpic perennial Carlina vulgaris, but there was a significant negative effect on growth. Several studies confirm that effects of local density on growth are negative, while effects on survival are more often absent or even positive (reviewed in Goldberg & Novoplansky 1997). Many studies of density-dependent demography in plants have only considered intraspecific density (references in Goldberg et al. 2001), thereby ignoring the fact that there may be species-specific differences in direction as well as magnitude of the effects of density (e.g. Holmgren et al. 1997). This limitation applies to most studies performed on monocarpic perennials (Rees et al. 1999; de Jong et al. 2000; Rose et al. 2002), and the omission of the effects of the other species present may have underestimated interference.

As plants are sessile organisms they primarily experience the environment in their immediate fixed neighbourhoods (Antonovics & Levin 1980) and, when quantifying density dependence, local density will consequently be more informative than mean population density. Focusing on effects at the individual level also permits examination of the evolutionary consequences of interactions. Plant neighbourhoods have been defined in various ways (see, for example, Sletvold & Hestmark 1999), and the studies with most success in relating individual performance to some measure of local density have been conducted in experimental populations of annuals (e.g. Silander & Pacala 1985; Pacala & Silander 1987; Oosthuizen et al. 1996). The usefulness of this approach in natural populations has, however, been questioned, particularly where perennial species are concerned (Waller 1981). Natural densities may be too low to see any pronounced effects, and variation in time of emergence (Ross & Harper 1972) or among microsites (Harper et al. 1965) could override the importance of initial plant position. The neighbourhood approach is thus most likely to be successful in predicting subsequent performance in species with synchronous germination at high densities, typically from a large seed bank. Very few studies on density dependence in monocarpic perennials have used detailed information on individual plant positions, but Rees et al. (1999) found weak negative effects on growth and mortality of Onopordum illyricum using a neighbourhood approach, and Suzuki et al. (2003) showed that early mortality in Lysimachia rubida was positively related to individual local density.

A density-dependent recruitment phase should suggest selection for faster growth and higher survival in small individuals. As allocation to faster growth may be at the expense of allocation to storage and resistance, a tradeoff between relative growth rate and survival might be expected. Metcalf et al. (2003) reanalysed data on three monocarpic perennial species and found such a tradeoff for the smaller plants.

In this study I examine the effects of local density of plants on vital rates in Digitalis purpurea L., a facultative biennial with a ‘seed-bank-oriented’ life cycle (Kachi 1990). Detailed studies of how density dependence plays out during the life cycle of monocarpic plants are very few, and my main objective is to evaluate if density dependence is restricted to the seedling/first-year stage, a key assumption in current life history modelling (Rose et al. 2002; Metcalf et al. 2003). I include both intra- and interspecific density, and use an individual-based approach to address: (i) how variation in local density influences growth and survival at different life stages, (ii) whether density effects on growth and survival are of the same direction and magnitude, (iii) whether effects on focal plants depend on neighbour identity, size, proximity and/or dispersion, and (iv) if there is a tradeoff between first-year growth and survival.

Materials and methods

study species

Digitalis purpurea L. is a facultative biennial herb, widespread in coastal areas of south-western Norway. Large populations are found in areas where the vegetation has been disturbed (e.g. roadcuts, woodland clearings and pastures). The species has a persistent seed bank from which abundant germination, flowering and seed production takes place after soil disturbance (van Baalen 1982). D. purpurea spreads its seeds ballistically, and the majority of seeds are likely to fall within a few metres of the maternal plant. Seeds mainly germinate in spring, reaching very high initial seedling densities (van Baalen 1982). Individuals form basal rosettes that may produce a flowering stem during the second summer, but often delay flowering for several seasons (van Baalen & Prins 1983). There is large variation in population development among habitats: established populations in pastures may persist with apparently stable dynamics for several years (personal observations) whereas, in successional habitats, the absence of repeated disturbance frequently leads to local population extinction (van Baalen & Prins 1983).

fieldwork

This study was conducted in an established population in a pasture at Ulvik (Hordaland County), in south-western Norway. Disturbance had exposed an area of bare soil in late 1997, and five plots of 1 × 0.5 m were established on a SE-facing slope within this area in spring 1998. All individuals within each plot were tagged with small numbered flags, and their positions were registered as x- and y-coordinates in a grid system to a precision of 1 mm. The plots were censused twice a year from 1998 to 2000 (spring census in May and autumn census in August/September). In each survey the number of leaves and the length of the longest leaf from the base to the apex were recorded on each individual of D. purpurea. This simple parameter (leaf length) was the field estimate that correlated best with dry weight in a test study on D. purpurea in 1997 (rs = 0.97, P < 0.0001, n = 200). The size of individuals of other species was measured as rosette size, leaf size or stem height, depending on the growth form and morphology of the species concerned. Relative growth rate (RGR) was calculated as RGR = log(sizet+1) − log (sizet), where size was the length of the longest leaf.

local density

An individual's neighbourhood was defined as a circle of fixed radius around it, and an index of interference (W) was calculated for the neighbourhood area (Silander & Pacala 1985). Only individuals with complete neighbourhoods were included as focal individuals, i.e. neighbourhood circles that intersected the plot boundary were excluded from the analyses. In its simplest form, the interference index W is equal to the local density, i.e. the number of neighbours within the chosen radius, W = N. The optimal neighbourhood radius was defined as that which maximized the amount of variation accounted for in a given model, and it was determined in all analyses by varying r systematically in 1-cm increments from 3 to 12 cm.

To determine whether intra- and interspecific effects of local density differed, the species identity of the focal individual's nearest neighbour was determined. Owing to the low frequency of most species present, species identity had only two levels, i.e. neighbours were classified as conspecific or not. To determine whether within-neighbourhood proximity had any effect, an index that weighted each neighbour by its distance from the focal plant was calculated as: inline image, where di is the distance of neighbour i from the focal plant, r is the neighbourhood radius and n the number of neighbours (Silander & Pacala 1985). Assuming competitive strength to be proportional to individual size [Weiner's (1990) ‘relative size symmetry’] an index summing all sizes within the neighbourhood was calculated as: inline image, where si is the size of neighbour i and n the number of neighbours. To see whether neighbour interference was more intense when the focal individual was surrounded by neighbours than when the neighbours were clustered to one side, all interference indices were weighted by angular dispersion as W * z, where inline imagei is the angle to neighbour i and n is the number of neighbours in the neighbourhood) (Mack & Harper 1977). Angular dispersion varies from 0 (complete aggregation) to 1 (maximal dispersion).

analyses

Data were pooled across plots within season to increase sample size. Neighbourhoods and neighbour species identities were defined in a Pascal program developed for this purpose (unpublished program, John M. Grindeland), which also calculated the estimates of neighbour interactions. Individual RGRs met the assumptions of normal distribution and constant variance and RGR was therefore analysed by linear models (Sokal & Rohlf 1995). Survival was analysed by logit models, i.e. with a binomial response distribution and logit link function (Agresti 1996). In all models, species identity was included as a fixed categorical factor and neighbourhood interference (W) as a continuous covariate. In survival analyses, spring maximum leaf length/autumn maximum leaf length was included as a continuous covariate. Type III tests were used for all analyses, and the interaction term between species identity and neighbourhood interference was included to determine whether effects of local density depended on whether the closest neighbour was a conspecific or not. The goodness of fit of logit models were assessed by Pearson chi-squared statistics and was always found to be appropriate. All analyses were performed with the Insight module in SAS 8.02 (SAS Institute 2001). Although the interference indices tested included additional information, local density always accounted for as much, or even more, of the variation than either index. The results from the latter analyses are therefore not reported.

Results

descriptives

The number of species present in the plots at the time of establishment varied from six to nine, and the total number of individuals was 1083. Other than D. purpurea (n = 816), the only frequently occurring species was Omalotheca sylvatica L. (n = 195). Total density varied 1.8-fold (from 159 to 287 individuals per plot), and number of D. purpurea twofold (from 111 to 229) between plots. Only two new D. purpurea individuals emerged during the whole study, and both died within the same summer. No individuals of other species emerged. Total mortality of D. purpurea during the study was 92.4%. Only 13 individuals flowered (1.6%). At the last survey only 62 individuals of D. purpurea were present in the plots, and only two other species were present.

Mean size and size variability of D. purpurea increased during the study (Fig. 1). The size distribution was initially skewed towards small sizes, but from the second spring maximum leaf length became normally distributed (Fig. 1). All species present in the plots at the time of establishment were nearest neighbours of focal plants. In 1998 77% of the D. purpurea individuals had a conspecific as nearest neighbour, but in 1999 this was reduced to 32% and in 2000 to 21%. The optimal radius (which maximized the amount of variation accounted for in the models) was found to be 7 cm, and mean number of neighbours within this area was nine at the onset of the study.

Figure 1.

Size distributions of D. purpurea during the study. Size categories corresponded to 1-cm intervals: 1 = 0–0.9 cm, 2 = 1–1.9 cm, etc., with the last category, 11 ≥ 10 cm. Dotted areas represent the number of individuals dying during the next time interval. Arrows indicate mean size.

relative growth rate

The relationship between size in spring and autumn was well described by a line of the form log (sizet+1) =b log (sizet) − a, with a negative intercept and a slope greater than 1.0 (Fig. 2a–c). The relationship between autumn size and spring size next year was less accurately described, with more variation in growth between individuals of the same size (more scatter around the regression line), and with a positive intercept and a slope below 1.0 (Fig. 2d,e). This implies that summer RGR is an increasing, and winter RGR a decreasing, function of size.

Figure 2.

The relationship between spring and autumn size in 1998 (a), 1999 (b) and 2000 (c), and between autumn and spring size from 1998 to 1999 (d) and 1999 to 2000 (e), all on a log–log scale. The thin line is the curve of zero growth during the time interval; individuals above this line increase in size, and individuals below this line decrease in size. Linear regressions (thick line): (a) y = 1.15x − 0.0317, r2 = 0.77, P < 0.0001, n = 536; (b) y = 1.15x − 0.0105, r2 = 0.70, P < 0.0001, n = 83; (c) y = 1.29x − 0.104, r2 = 0.83, P < 0.0001, n = 43; (d) y = 0.601x + 0.187, r2 = 0.36, P < 0.0001, n = 86; (e) y = 0.657x + 0.233, r2 = 0.43, P < 0.0001, n = 59.

In both 1998 and 1999 there was a significant effect of local density on summer growth rates (Table 1), with a linear decrease in summer RGR with increasing local density. A similar relationship was apparent in summer 2000, but was not significant (Table 1). There was also a significant effect of neighbour identity in 1998 (Table 1), when individuals with a conspecific as the closest neighbour had somewhat lower RGRs than those with any other species. No significant effects of identity were found on summer RGRs in either 1999 or 2000, and the interaction between neighbour identity and local density was never significant (Table 1). There was no significant effect of either local density or neighbour identity on winter RGR or yearly RGR in any year (Table 1).

Table 1. ancova of the effects of local density and species identity of the closest neighbour on individual relative growth rates (RGRs). No interactions were significant
Source of variationSSd.f. F P
RGR summer 1998
 species identity0.14  16.060.014
 local density0.18  17.820.005
 residual9.47408  
RGR winter 1998–99
 species identity0.0050  10.1440.706
 local density0.073  12.1210.151
 residual2.01 59  
RGR summer 1999
 species identity0.0003  10.0220.883
 local density0.14  18.980.004
 residual0.94 58  
RGR winter 1999–2000
 species identity0.0465  11.340.254
 local density0.0886  12.540.118
 residual1.534 44  
RGR summer 2000
 species identity0.018  11.430.240
 local density0.014  11.170.289
 residual0.42 34  
RGR yearly 1998–99
 species identity0.053  10.930.339
 local density0.088  11.540.220
 residual3.36 59  
RGR yearly 1999–2000
 species identity0.052  10.820.373
 local density0.027  10.410.525
 residual2.18 34  

survival

Summer survival was highly dependent on spring maximum leaf length, and winter survival on autumn maximum leaf length (Table 2, Fig. 3a–c). Survival during the first summer (1998) was relatively high, with a mean survival probability of 0.77. Only one individual died during each of the two next summers, and survival probability was therefore not analysed. Survival during the first winter (1998–99) was low, with a mean survival probability of 0.15, but the value for the next winter (0.71) was considerably higher. The survivorship curve for the population demonstrated an overall exponential decay of plant number, despite a strongly seasonal rhythm of mortality (Fig. 4a,b).

Table 2. Generalized linear model analyses of the effects of size, local density and species identity of the closest neighbour on summer and winter survival (binomial errors, logit link). No interactions were significant
Source of variationd.f.χ2 P
Survival summer 1998
 spring size150.41< 0.0001
 local density119.61< 0.0001
 species identity1 0.97  0.3244
Survival winter 1998–99
 autumn size152.39< 0.0001
 local density1 0.17    0.6812
 species identity110.97    0.0009
Survival winter 1999–2000
 autumn size113.74    0.0002
 local density1 3.08    0.0793
 species identity1 3.69    0.0547
Figure 3.

Predicted survival probability as a function of size during summer 1998 (a), winter 1998–1999 (b) and winter 1999–2000 (c). Models: (a) logit (survival) = −0.161 + 1.06 * size + 0.285 * species identity − 0.0957 * local density, n = 536; (b) logit (survival) = −5.57 + 0.771 * size + 0.806 * species identity, n = 410; (c) logit (survival) = −4.64 + 0.724 * size + 1.45 * species identity, n = 75. intrasp. = D. purpurea nearest neighbour, intersp. = different species nearest neighbour.

Figure 4.

(a) Survivorship of D. purpurea, time in weeks after first observation. (b) Total seasonal mortality rates of D. purpurea during the three years, s = summer, w = winter.

Summer survival in 1998 was negatively affected by local density, with a linear decrease in survival probability with increasing density (Table 2, Fig. 5a), but was not affected by neighbour identity (Table 2). A high proportion of the individuals within the densest areas died (Fig. 5b). There was no significant effect of local density on winter survival in any season (Table 2). There was, however, a significant effect of neighbour identity on winter survival in 1998, with focal individuals with a conspecific as the closest neighbour having lower survival probabilities (Fig. 3b). This relationship was also nearly significant in winter 1999 (Table 2, Fig. 3c). There was no significant interaction between neighbour identity and local density in any year. Neither was there any evidence of a tradeoff between first-year summer RGR and survival probability (Fig. 6).

Figure 5.

(a) The relationship between local density and predicted survival probability during summer 1998. Model: logit (survival) = 0.149 + 1.07 * size − 0.0944 * local density, n = 536. (b) Proportion of deaths within different density groups during summer 1998; dotted area = dead individuals, black area = living individuals.

Figure 6.

Predicted survival probability during first winter in relation to RGR during first summer. Model: logit (survival) =−5.57 + 0.771 * size + 0.806 * species identity, n = 410.

Discussion

Density-dependent effects lasted well beyond the seedling stage, with growth of Digitalis purpurea during the second summer still negatively density-dependent. In view of the high degree of autumn size-dependency of both winter survival and flowering probability, local density may have considerable impact on individual life expectancy. Density effects were also strikingly seasonal, and depended on plant life stage and the actual demographic rate measured.

growth

Growth during summer was invariably an increasing function of plant size, with the largest plants having the highest growth rates during all three summers, whereas the opposite was true for winter or yearly RGR. In accordance with the vast majority of earlier studies on monocarpic perennials (reviewed in Metcalf et al. 2003), RGR was lower for large plants. As most individuals of D. purpurea decrease in size during winter, this means that large individuals both gain and lose size at a faster rate than small individuals, and experience higher variation in growth throughout a year. Individuals of the same size varied considerably in growth, and this variation was much higher during winter. This scatter about the growth functions demonstrates that age and size are weakly related in D. purpurea. The slope of the growth functions also varied more between winters than between summers, suggesting that temporal heterogeneity at this season plays an important role in regulating demographic rates in this species.

Several earlier studies on monocarpic perennials have suggested that competition operates during a fixed period early in life (Rees et al. 1999; de Jong et al. 2000; Rose et al. 2002). In this population of D. purpurea, growth was negatively density-dependent during the first two summers, with local density accounting for 10% of the variation in growth in both years. Population size was dramatically reduced during the first winter and local summer densities were lower in 1999 than in 1998. Nevertheless, negative effects of increased local density on RGR were just as strong in the second summer, suggesting that competitive influence increased with plant size. It is possible that growth rates in the earliest phase of plant establishment reflected genetic composition (Bazzaz et al. 1982) and/or maternal effects (Gutterman 1992) rather than initial spatial pattern. An earlier study found a positive effect of increased maternal size on both emergence and early growth rates of offspring in D. purpurea (Sletvold 2002). This effect disappeared between 8 and 14 weeks after germination, suggesting that positive maternal size effects on offspring fitness may have obscured negative effects of local resource competition during the first 2–3 months of the present study. During the second winter local densities were further reduced, and there was no evidence of local competition in the third summer. This is similar to the results of de Jong et al. (2000) for Carlina vulgaris, where no significant relation was found between density and growth for rosettes older than 2 years. In contrast to the patterns found during summer, growth in winter was density-independent during both years. As little active growth actually takes place during a normal winter, it is likely that local resource competition is considerably relaxed during this season.

The fact that individuals whose closest neighbour was a conspecific had somewhat lower RGRs during the first summer indicates that intraspecific competition was stronger than interspecific competition in the recruitment phase in this population. This was not due to simple size effects, as the most frequent interspecific competitor, Omalotheca sylvatica, had a similar size distribution and a slightly larger mean size than D. purpurea in the first summer. I did not address the explicit mechanisms of interactions but, because O. sylvatica has rosettes growing closer to the ground at this stage in the life cycle, shading is unlikely. The population was situated in a dry and rather unproductive habitat, and this suggests competition for below-ground resources.

survival

There was a significant negative effect of increasing local density on first summer survival probability in D. purpurea, and mortality was concentrated in high density areas within the plots. Around 10% of the variation in survival was accounted for by local density. No mortality occurred in the second summer, indicating that effects of competition were then less severe, although growth was still affected. Most mortality occurred during winter in a density-independent manner, and was most likely due to substrate instability. Small earthslides swept away whole clusters of plants, and a positive effect of density on survival might have been expected, with dense patches stabilizing the substrate. The large difference in mortality rate between the two winters reflects the fact that individuals are larger by the second autumn, but considerable differences in the patterns of size-dependency emphasize the importance of stochastic variation at this season in populations of D. purpurea.

Autumn size was clearly the best determinant of fate during both winters. Survival probability during the first winter increased from about 20% to 80% between size categories 4 and 8 (maximum leaf length 3.0–3.9 and 7.0–7.9 cm, respectively). This may be due to higher levels of stored resources in large plants, but can more likely be attributed to stochastic mortality factors having less impact, e.g. because large plants will not be uprooted as easily as small plants. In addition, individuals with an interspecific nearest neighbour had a slightly higher survival probability than those closest to another D. purpurea. This was probably due to the fact that the most common interspecific neighbour, O. sylvatica, stabilized the substrate to a higher degree with its denser, more low-growing rosettes.

The seasonal shift in the effects of both size and local density was observed throughout the study, and seemed independent of plant life stage/age. The negative density effects observed during the first summer confirm that the onset of competition is immediate in D. purpurea, and strong enough to override possible positive correlations between microhabitat quality and local plant density. This contrasts with the seasonal variation described from a population of Silene dioica (Matlack & Harper 1986), in which the correlation between local density and performance was positive at the seedling stage, but negative at later stages. Suzuki et al. (2003) recently studied mortality patterns in the short-lived biennial Lysimachia rubida, and found similar seasonal patterns to those reported here for D. purpurea, with significant effects of density observed only during some months within a year. However, because L. rubida usually behaves as a strict biennial (Suzuki et al. 2003), few individuals in a cohort will experience seasonal shifts more than once, and the effects of age and season are confounded. Temporal variation in both growth and mortality has been found to be of critical importance in the modelling of flowering strategies in recent studies of monocarpic perennials (Rees et al. 1999; Rose et al. 2002), and the present results suggest that variation between winters will have considerable effect in populations of D. purpurea.

There was no evidence of any tradeoff between RGR and survival probability in this study. Given the high, density-independent winter mortality this was not surprising; 76% of the first-year mortality occurred during winter in a stochastic manner, as did virtually 100% of the second-year mortality. If the study years are typical for this population, there should be directional selection for high growth rates during summer, enabling individuals to reach the large autumn size needed to resist earthslides and burials, particularly during the first winter. The behaviour of this population thus strongly contrasts the pattern found by Sarukhán & Harper (1973), although they also demonstrated a strong seasonal rhythm of mortality in four species of Ranunculus. In their study, mortality rates peaked during the most active period of growth in early summer, and were low during winter months.

In contrast to several earlier studies (e.g. Mack & Harper 1977; Silander & Pacala 1985), the inclusion of information on the proximity, size or angular dispersion of neighbours did not increase the amount of variation accounted for in the models of variation in vital rates. This may simply be due to limited plasticity of growth in D. purpurea, as a rosette plant is, for instance, less likely to be able to respond to one-sided crowding by growing in the opposite direction.

Concluding remarks

Density dependence is often assumed to be operating only at the recruitment stage in monocarpic plants, simplifying evolutionary calculations and life history modelling (Metcalf et al. 2003). The present results suggest that this may be unjustified. Although many confounding factors present in a natural population were ignored, there was repeated evidence for competitive interactions in this population of D. purpurea. The results support the notion that competition was most severe during the first summer, when local density directly influenced both growth and survival, but crowding apparently still restricted summer growth the second year. As reproduction in monocarpic perennials is highly size-dependent (e.g. Lacey 1986; Klinkhamer et al. 1987a,b, 1996), competition may influence timing of flowering by reducing summer growth rates. Autumn size is a good predictor of flowering probability in D. purpurea, and around 40% of the individuals had passed the minimum threshold size for vernalization observed in this population by the autumn of the second year (unpublished data). Local competition at this stage clearly has the potential to delay flowering in this species, and these results highlight the importance of explicitly addressing this matter in life history studies of monocarpic perennials.

Acknowledgements

I thank J. Ågren for helpful comments on the manuscript and J.M. Grindeland for writing the Pascal program to parameterize neighbourhood interactions. Constructive suggestions from J. Weiner, L. Haddon and an anonymous referee also improved the manuscript. This research was financially supported by the Norwegian Research Council (grant no. 123845/410), the Nansen Foundation of the Norwegian Academy of Sciences (grant no. 73/99) and the University of Oslo (Foundation of H. Kallevig & F. Petersen).

Ancillary