Space versus time variation in the population dynamics of three co-occurring perennial herbs



    Corresponding author
    1. Nature Conservation and Plant Ecology Group, Wageningen University, Bornsesteeg 69, NL-6708 PD Wageningen, The Netherlands, and
    2. Department of Ecology, Radboud University Nijmegen, Toernooiveld 1, NL-6525 ED Nijmegen, The Netherlands
      Present address and correspondence: Eelke Jongejans, Department of Biology, Pennsylvania State University, 208 Mueller Laboratory, University Park, PA 16802, USA (tel. +1 814 865 8778; fax +1 814 865 9131; e-mail
    Search for more papers by this author

    1. Department of Ecology, Radboud University Nijmegen, Toernooiveld 1, NL-6525 ED Nijmegen, The Netherlands
    Search for more papers by this author

Present address and correspondence: Eelke Jongejans, Department of Biology, Pennsylvania State University, 208 Mueller Laboratory, University Park, PA 16802, USA (tel. +1 814 865 8778; fax +1 814 865 9131; e-mail


  • 1Many plant species are currently restricted to small and isolated populations as a result of habitat destruction and fragmentation. Lack of sufficient data often leads to spatial variation being substituted for temporal variation in models to evaluate management options, but variation in population growth rate (λ) between sites or over years may be caused by variation in different life history components.
  • 2We studied the demography of three coexisting and related perennial herbs (the long-lived perennial Succisa pratensis, the shorter-lived Hypochaeris radicata and the long-lived, but more clonal Centaurea jacea) in the same plots over 4 years (1999–2003).
  • 3Life table response experiment analysis revealed that temporal and spatial variation in the life history components of S. pratensis and H. radicata were qualitatively different: variation in fecundity contributed most to variation in λ between sites, but variation in growth contributed most to variation between years.
  • 4In years with a below average λ, most life history components of S. pratensis and H. radicata had lower transition probabilities, whereas the reasons for low population growth differed between the three poor sites. All major life history components of S. pratensis had lower probabilities in the most productive of these sites, but in the other two, negative contributions from some components (fecundity or growth) were partly compensated for by positive contributions from others (growth or stasis).
  • 5Centaurea jacea showed a different pattern: in poor years, negative contributions from some life history components were buffered by positive contributions from other components; however, this buffering did not occur in poor sites.
  • 6Our analysis shows that the population dynamics of perennial plants may respond differently to temporal and spatial variation in environmental conditions. Moreover, co-occurring species with similar life histories responded differently to the same spatiotemporal variation. We conclude that temporal and spatial dynamics cannot readily be interchanged in population viability analyses and management studies, as such substitution may lead to incorrect projections of a species’ population dynamics in time.


It has been shown that the dynamics of grassland herb populations respond to variation in site-specific environmental conditions (Oostermeijer et al. 1996; Menges & Dolan 1998; Watkinson et al. 2000). Nutrient-poor, species-rich grasslands may differ in productivity, abiotic conditions (e.g. water table), vegetation composition and management regime (Soons & Heil 2002; Soons et al. 2003; Vergeer et al. 2003). Although annual mowing and hay removal are stabilizing factors that halt biomass accumulation and maintain the vegetation in an early successional stage, differences between years exist owing to variation in mowing date and climatic factors such as rainfall.

In order to manage the conservation of threatened plant species, the contribution of different life history components to the population growth rate must be clearly understood (Oostermeijer et al. 1996). General patterns have been shown within species: good years or sites (i.e. those that have above average population growth rates) are associated with a higher contribution of sexual reproduction, whereas populations in poor years or sites rely more on survival (Oostermeijer et al. 1996; Menges & Dolan 1998; Valverde & Silvertown 1998). It is tempting to substitute spatial variation for temporal variation (Silva et al. 1991; Damman & Cain 1998; Valverde & Silvertown 1998; Bühler & Schmid 2001; Quintana-Ascencio et al. 2003), but few studies have investigated whether these patterns are actually interchangeable (Horvitz et al. 1997). Here we systematically analyse the life history components of three perennial, herbaceous species to determine whether they respond in similar ways to temporal and spatial variation.

Population matrix projection models provide a powerful tool for studying such variation. Several authors have shown that the relative importance, or elasticity, of particular matrix elements in determining the projected population growth rate (λ) is correlated with the value of λ (Silvertown et al. 1993; Oostermeijer et al. 1996; de Kroon et al. 2000). However, differences in elasticities do not need to coincide with changes in the life history components that have caused the population growth rate to vary, because elasticities are only local properties of a particular matrix (Moloney 1988; Horvitz et al. 1997). Techniques have, however, been developed to decompose the variation in λ into contributions resulting from variation in each of the underlying components (Horvitz et al. 1997; Caswell 2000). We use fixed life table response experiments (LTREs), a variance decomposition technique, to investigate whether variation in the same life history components determines the variation in λ between sites and between years. We studied the demography of three perennial herbs in the same plots in five sites for 4 years. We contrasted the spatiotemporal patterns in the life history components of the threatened, long-lived Succisa pratensis with those of the short-lived Hypochaeris radicata and another long-lived species, Centaurea jacea, that shows greater clonal growth.

Materials and methods

study system

Nutrient-poor, species-rich, moist meadows declined in abundance by more than 99% in the Netherlands in the 20th century because of drainage, cultivation and fertilization, and are now restricted to small nature reserves (Berendse et al. 1992; Soons et al. 2003). Such wet grasslands were previously solely used for haymaking and extensive cattle grazing (Pegtel 1983). Current management aims to conserve local floristic diversity using annual mowing and hay removal as well as restoration of hydrological conditions by halting the influx of nutrient-rich surface water and stimulating the upwelling of base-rich groundwater.

Succisa pratensis Moench (devil's bit scabious, Dipsacaceae) is a perennial with polycarpic rosettes that can survive for many years (Adams 1955; Hooftman & Diemer 2002). Leaves grow in alternating pairs and remain alive for almost a year. Flower stalks arise from axillary buds of old leaves, and flowers are produced from late July until October. Axillary buds occasionally form side-rosettes on short stolons that disintegrate after 1 year. Succisa pratensis is a characteristic species in Cirsio dissecti–Molinietum communities (Schaminée et al. 1996).

Hypochaeris radicata L. (cat's ear, Asteraceae) is also perennial, but lives for only about 3 years (de Kroon et al. 1987; Fone 1989). Its flower stalks and new rosettes are formed clonally in the centre of the main rosette. The ramification rate in nutrient-poor meadows is lower than in road-verge populations (de Kroon et al. 1987; van Groenendael et al. 1994). Flowering, on one to several leafless flower stalks, begins in June and continues until autumn. Compared with the two long-lived species discussed here, H. radicata allocates more (25% vs. 7.5%) of its total biomass to sexual reproduction (Hartemink et al. 2004).

Centaurea jacea L. s.l. (knapweed, Asteraceae) is a long-lived perennial with monocarpic rosettes (Tamm 1956). Flowers are borne on a single apical flower stalk from June until autumn. During and after flowering, vegetative side-rosettes are formed on the woody rootstock at the soil surface alongside the flower stem (Hartemink et al. 2004). Compared with the less clonal species, C. jacea allocates much more (55% vs. 10%) of its total biomass to side-rosettes (Hartemink et al. 2004). In the meadows studied, the number of connected rosettes of C. jacea rarely exceeds three, although it can form extensive below-ground woody branches in more productive habitats such as dykes and road verges (E. Jongejans, personal observation).

permanent plots

The three species were studied from 1999 until 2003 in five, annually mown, nutrient-poor grasslands in the centre and east of the Netherlands. Arranged from low to high productivity, these sites were: Leemputten (L) (52°17′ N, 5°44′ E), Konijnendijk (N) (52°02′ N, 6°26′ E), Bennekomse Meent (B) (52°00′ N, 5°36′ E), Koolmansdijk (O) (52°01′ N, 6°33′ E) and Veerslootlanden (V) (52°36′ N, 6°08′ E). Site L is a very species-rich grassland with heather species as well. The average vegetation height is only 10 cm and 19% of the soil surface is bare. The plots in site L were not mown during the last 2 years of the study. Site N has a typical Cirsio dissecti–Molinietum vegetation (21 cm high, 16% bare soil) with Cirsium dissectum and Molinia caerulea. The latter two species are also abundant in the plots in site B, which has taller vegetation (53 cm) and 16% bare soil. Site O (vegetation height 52 cm, 14% bare soil) is drier than the other sites because a nearby water pumping station has drained the area over the last few decades. Site V (vegetation height 50 cm, 7% bare soil) is dominated by Festuca spp. and dense M. caerulea tussocks; few other species occur (see Jongejans 2004 for complete species lists).

Succisa pratensis was studied in all five sites, but H. radicata only in sites L, N and O, and C. jacea only in sites B, N and O. Site O was mown unexpectedly early in 2001 (supplementary Table S1), causing the loss of two transition matrices for C. jacea, as all flower stalks were removed. The other two species flowered before (H. radicata) or after (S. pratensis) this early mowing and they were censused as in the other years. Thus there were 20, 12 and 10 year-to-year transition matrices for S. pratensis, H. radicata and C. jacea, respectively. Permanent plots (three to eight per site; Table S1) of 1 × 1 m2 were randomly established in April 1999. A border zone of 15 cm around these plots was left undisturbed. All rosettes of the study species were mapped to 5-mm accuracy. Their establishment, survival, growth and flowering were recorded once every year during flowering (see Table S1 for exact census dates). Plant size was quantified as the maximum length of rosette leaves. The number of flower heads was counted and their phenological stage (bud, flowering, seeded or disseminated) at the moment of census was recorded.

stage and size classes

Classification of individual rosettes was based on a combination of size, flowering status and the formation of side-rosettes. We distinguished six classes:

  • 1Seedlings (sdl) are tiny rosettes with leaves up to 2 cm (S. pratensis and H. radicata) or up to 5 cm (C. jacea) in length. Note that we use ‘seedling’ in a broad sense, and that plants assigned to this class may actually also be older, but still very small rosettes.
  • 2Small vegetative rosettes (sml) are larger than seedlings, with leaves up to 5 cm (S. pratensis and H. radicata) or up to 12.5 cm (C. jacea).
  • 3Large vegetative rosettes (lrg) have longer leaves than do small rosettes.
  • 4Flowering rosettes (flow) have at least one flower stem and are grouped irrespective of rosette size.
  • 5Vegetative side-rosettes (side.veg) are those newly formed by clonal reproduction and without flower stems.
  • 6Flowering side-rosettes (side.flow) are newly formed by clonal reproduction and have at least one flower stem.

The first four classes largely correspond to an earlier classification for S. pratensis (Bühler & Schmid 2001). No seed stage is modelled, as these species have predominantly transient seed banks (Thompson et al. 1997).

Clonal reproduction by ramification resulted in new rosettes close to old ones but, without disturbance, it was not always possible to determine from which old rosette a new one originated. Therefore we assigned all new, non-seedling rosettes as derived from the nearest old, non-seedling rosette, providing one could be found within a set, species-specific distance. Based on field experience, this distance was 25, 10 and 40 mm for S. pratensis, H. radicata and C. jacea, respectively. In those cases where it was possible to determine connections without disturbance (n = 63 for S. pratensis, n = 18 for H. radicata and n = 200 for C. jacea), more than 92% of the distances between old and new rosettes were within the species-specific distance. In C. jacea, side-rosettes were only assigned to a non-flowering parent when no flowering rosettes were present within the 40-mm radius, because 89% of the 200 side-rosettes of known origin were attached to flower stems.

matrix parameterization

With the calculated year-to-year transition probabilities we constructed 6 × 6 projection matrices in which each element (aij) represents the transition from the jth category in year t to the ith category in year t + 1. The 36 matrix elements were classified as representing fecundity (F), growth (G), clonal reproduction (C), survival of side-rosettes that were formed in the previous year (T), stasis (S) and retrogression (R) (Table 1). Growth is defined as the transition from a smaller, vegetative stage to a larger or flowering stage class, stasis as a rosette remaining in the same class and retrogression as surviving but moving to a smaller class.

Table 1.  Population matrices divided into life history components and mean matrices per species. The six stage classes are: very small vegetative rosettes or seedlings (sdl), small vegetative rosettes (sml), large vegetative rosettes (lrg), rosettes with at least one flower stem (flow), vegetative rosettes produced by clonal reproduction (side.veg), rosettes produced by clonal reproduction with at least one flower stem (side.flow). The six life history components are fecundity (Fij), growth (Gij), clonal reproduction (Cij), survival of side rosettes formed in the previous year (Tij), stasis (Sij), and retrogression (Rij). Matrix elements are means of 20 transition matrices of Succisa pratensis, 12 of Hypochaeris radicata, and 10 of Centaurea jacea. The temporal and spatial coefficients of variation (Tcv and Scv respectively) are given beside each mean. The CVs of the first column are zero because we have no data on the variation of seedling survival and growth within species
S. pratensis
 sdl0.4890.033 98136  0.994 43 92  0.99443 92
 sml0.3700.306 39 730.075 70 850.677 43 850.29522 890.75035 70
 lrg0.416 44 530.566 16 240.413 48 420.45118 480.680 0 43
 flow0.0601101400.280 33 500.471 40 360.15546 820.180 0138
 side.veg0.041 881130.038 731020.105 67 910.033191990.200 0137
 side.flow0.0022002240.0021711860.0131131110.000  0.000  
H. radicata
 sdl0.3750.033102110  1.743 25 28  1.74325 28
 sml0.1630.260 45 600.111 77 410.572 20 220.18410 870.33425 28
 lrg0.282 73 520.479 35 220.349 15 290.107141080.111 0173
 flow0.0331271360.135 81 810.190 73 510.000  0.111 0173
 side.veg0.0151881610.011 531060.025 801100.000  0.000  
 side.flow0.000  0  00.0051441730.0071671730.000  0.000  
C. jacea
 sdl0.4980.026136105  0.199 62 52  0.199 62 52
 sml0.3990.199 55 590.148 30 190.050 62 520.22220 820.050 62 52
 lrg0.240 48 550.318 18 34  0.27634 60  
 flow0.109 98 880.266 19 72  0.26825 35  
 side.veg0.109 94 920.081 25 720.390 30 300.09842 640.522 24 38
 side.flow0.0351571150.012103 750.591 34 400.01659 720.272 19 17

In the first year (1999–2000) it was not possible to observe fates of side-rosettes formed in the previous year because no observations were made in 1998. For those first year classes (side.veg and side.flow), and for stage classes containing fewer than six rosettes at time t, the transition probabilities of that class to all other classes at time t + 1 were assumed to be the mean of all observations over the years for that class and population. Insufficient (i.e. fewer than six) observations occurred only in 9% of the instances for the adult classes sml, lrg and flow, but were more frequent in the clonal offspring classes, as ramification was rare in H. radicata and S. pratensis.

A seed addition experiment was used to study seedling establishment and fate because the density of seedlings was much lower than that of adult plants. One hundred seeds were added to each of 10 rectangular plots, 5 cm wide and 50 cm long, in sites B, N and O in November 1999, and again in 10 different plots a year later. The numbers of seedlings and small rosettes were recorded prior to the start of the experiment, and 12 and 24 months after seed addition. These results were compared with the numbers of seedlings and small rosettes in 10 control plots. We assumed that seeds do not survive for more than 1 year, and thus seedling and small rosette establishment per seed (psdl and psml, respectively) were calculated as follows:

image( eqn 1 )
image( eqn 2 )

where N is the number of seeds, seedlings (sdl) or small rosettes (sml) in either the seed addition (SA) or control plots (NS) at the beginning (t = 0) of the experiment or 1 year later (t = 1). These seed establishment probabilities, p, were multiplied by the average seed production of a flowering rosette in each population, in order to obtain the sexual reproduction matrix elements, Fij. Rosette seed production was estimated as the product of the mean number of observed flower heads and the population average number of seeds per flower head. Seeds were counted in randomly selected apical flower heads outside the plots (Table S2). In S. pratensis, only open flower heads and those that had finished flowering when censused were considered, because we assumed that the remaining buds in this late flowering species would be lost by mowing. Flower buds were included in the counts for the other two species, as buds of these earlier flowering species would presumably produce ripe seeds before mowing.

Seedling fates, S11 and G21, were also derived from the seed addition experiment because few seedlings were observed in the permanent plots. Seedling stasis, S11, was calculated as follows:

image( eqn 3 )

pooling the plots at each site. Seedling growth, G21, was estimated from the increase in the number of small rosettes in the SA plots in the second year after seed addition. This increase was divided by the number of seedlings present 1 year after seed addition. The control plots were excluded as they rarely contained more than a few seedlings. The annual mortality rate of small rosettes was estimated from the permanent plot data. Seedling growth, G21, was computed as

image(eqn 4 )

in which the plots were pooled for each site and starting year, and the sum of R12, S22, G32 and G42 was the average survival probability of small rosettes.

matrix analysis

The Matlab student edition of 1996 was used for all matrix computations. For each population transition matrix we calculated the projected population growth rate, λ, which is the dominant eigenvalue of a matrix. To construct 95% confidence intervals for the projected population growth rates, we applied the bootstrapping method by resampling 3000 times the data set of observed rosette fates from which a particular matrix was constructed (Efron 1982; Kalisz & McPeek 1992). The survival (T, G, S, and R) and clonal reproduction (C) observations were re-sampled for all classes separately. New sexual reproduction (F) elements were calculated after re-sampling the accompanying data set for the number on flower heads on the observed flowering rosettes. The upper and lower limits of the 95% confidence intervals were adjusted for small deviations between the mean λ of the 3000 newly constructed matrices and the λ of the original matrix (Caswell 2001, p. 306).

To decompose the variation in λ we applied fixed LTREs rather than a random decomposition technique because our research question required a comparison between the main factors. Only the fixed design allows comparison of the effects of individual species, sites or years within an LTRE analysis. The LTRE model with two factors within a species is (Caswell 2001)

image( eqn 5 )

in which a given λ of site m and year n is written as the sum of the dominant eigenvalue of the mean of all matrices of a species, λ(··), the main effect of site m, α(m), the main effect of year n, β(n), and the residual ‘interaction’ effect (αβ)(mn) (Horvitz et al. 1997). First, all main effects are estimated separately, while ignoring the interaction term (Caswell 2001):

image( eqn 6 )
image( eqn 7 )

in which differences between each matrix element and the corresponding matrix element of the overall mean matrix, A(··), are multiplied by the sensitivity values of the matrix halfway between the matrix of interest and the overall mean matrix. The LTRE interaction effect (αβ)(mn), quantifies how much the λ of an individual matrix differed from what would be expected from the associated main effects. The interaction effect is calculated by comparing an individual matrix with the reference matrix and by subtracting the contribution values of the associated main effects:

image( eqn 8 )

Small interaction terms indicate that the main factors influenced λ independently. The main and interaction effects can then be decomposed into positive or negative contributions from the different matrix elements. Note that if a given matrix has a λ smaller than the λ of the overall mean matrix, the sum of all LTRE contributions [i.e. α(m) + β(n) + (αβ)(mn)] will be negative.

A number of summary statistics may be derived from an LTRE analysis. First, α(m) or β(n), being the sum of all LTRE contributions of a mean matrix, is a measure of the effect of that site or year on λ (Horvitz et al. 1997) and we calculate these net values for each of the main effects. However, the relative strength of the effects of the different main factors can only be evaluated within and between the LTREs by comparing the absolute values for the individual year and site effects, because the net values can be negative. The averages of the absolute values of the effects of each site or year are therefore derived for each factor within the LTREs.

Although it is useful to compare the overall impact of year and site, information on the magnitude and direction (positive or negative) of the contributions of the matrix elements is lost in these measures. Therefore, we also summed the LTRE contributions for each of F, G, C, T, S and R for each of the transition matrices, and plotted the net contribution of each life history component against the deviation of the λ of the matrix of the corresponding main effect from the λ of the overall mean matrix. Strong correlations between these LTRE contributions and deviations in λ are expected for those components for which variation contributes consistently to the variation in λ between years or sites.

If trade-offs exist between life history components, a combination of negative and positive contributions within a matrix can be expected in a variance decomposition analysis (cf. Sterck et al. 2003). Therefore, we separated the positive and negative LTRE contributions within life history components to determine whether and how positive contributions by some elements were compensated for by negative contributions of others. Similar buffering occurs by elements with positive contributions for matrices with low population growth rates. Sterck et al. (2003) showed that life history trade-offs may thus damp the variation in λ.

To evaluate the effects of species, site and year simultaneously, we also performed a three-way LTRE for the N and O populations at which all three species were studied. The reference matrix, A(··), in this case was the mean of the three species mean matrices. Both fecundity and retrogression contribute to the transition from a flowering stage to a small vegetative rosette but the LTRE contributions of these matrix elements could not be separated, and were therefore completely assigned to fecundity, which is the predominant component (82% in S. pratensis, 80% in H. radicata and 100% in C. jacea).


The projected population growth rates (λ) were higher in Succisa pratensis than in the other two species: 55% of the projected population growth rates were significantly larger than unity in S. pratensis, vs. 9% in the other species (Fig. 1). The projected population growth rates of the Centaurea jacea populations did not differ much between years, but there was a tendency for λ to decrease over time in two populations of S. pratensis and in two populations of Hypochaeris radicata.

Figure 1.

Projected population growth rates of the three species in the different sites and years. The sites are Konijnendijk (N), Koolmansdijk (O), Bennekomse meent (B), Leemputten (L) and Veerslootlanden (V). The year transitions: 9 = 1999–2000, 0 = 2000–01, 1 = 2001–02 and 2 = 2002–03. The error bars are bootstrapped 95% confidence intervals.

The temporal and spatial coefficients of variation in the values of the matrix elements were considerable for elements that had low transition probability values, such as those representing clonal reproduction (Table 1). The CVs were smaller in the elements with higher probabilities, i.e. survival and sexual reproduction. When temporal CV differed from spatial CV it tended to be smaller.

The three-way (species, site and year) decomposition of the variation in λ of populations in sites N and O fitted well, with an average difference of 1.6% between the observed and modelled population growth rates. The three-way LTRE revealed that the species effect was stronger than the year effect, which was again stronger than the site effect (Table 2, Fig. 2). The different interaction effects were on average as large as the main site effect and almost as large as the main year effect (Table 2). This indicates that there is a clear effect of species, but that there are also interactions between the factors: S. pratensis performed better than average in site N (αγ = +0.046) whereas C. jacea performed less well in that site (αγ = −0.038). The opposite pattern was found in site O (αγ = −0.077 for S. pratensis and αγ = +0.037 for C. jacea). Hypochaeris radicata deviated little from the species and site averages (αγ = +0.002 for site N and αγ = −0.004 for site O). Succisa pratensis followed the mean year effect most closely, as is shown by a small deviation from the species and year averages (mean | βγ | = 0.015 for S. pratensis, 0.081 for C. jacea and 0.055 for H. radicata). Especially in the last year C. jacea performed better (βγ = +0.128) and H. radicata worse (βγ = −0.121) than average. Unfortunately, no statistical tests for significance have been developed yet for these novel LTRE analyses.

Table 2.  Magnitude of the different effects in four variation decomposition analyses (LTRE) of variation in population growth rate (λ): all species (Succisa pratensis, Hypochaeris radicata and Centaurea jacea) together in site N and O over 4 years (three-way), and each species separately in all observed sites over 4 years. The mean and standard deviation (SD) of the absolute values of all levels within a LTRE effect are given (× 100). The overall mean λ was 1.047, and the mean population growth rates were 1.176, 0.811 and 0.910, respectively, for the three species separately
LTRE effect (× 100)Three-wayS. pratensisH. radicataC. jacea
Site (S)| αm | 4.320.4712.316.19 4.16 1.227.532.48
Year (Y)| βn | 8.085.49 5.994.3113.7013.652.701.63
Species (P)| γq |15.868.31      
S × Y| αβmn | 4.412.52 8.866.11 8.40 6.503.682.22
S × P| αγmq | 3.382.80      
Y × P| βγnq | 5.324.97      
S × Y × P| αβγmnq | 7.075.48      
Figure 2.

Positive and negative species effects (γq) grouped by life history components in the three-way LTRE analysis of the variation in population growth rate (λ) between the matrices of the three species in sites N and O over 4 years. The six life history components are fecundity (F), growth (G), clonal reproduction (C), survival of side-rosettes formed in the previous year (T), stasis (S) and retrogression (R).

The LTREs within species also fitted well, with an average difference between the observed and modelled population growth rates of less than 1% (6 out of 42 cases showed a difference of > 1%, with a maximum of 6%). Within-species variation in λ was decomposed differently into site, year and interaction effects in the three species. In S. pratensis the main site effects were stronger than year and interaction effects (Table 2). Over all sites, contributions of fecundity (F) were positively correlated with site effects [α(m)] (Fig. 3a). Analysing the components that explained the deviation in λ for each of the sites separately, it was apparent that the F, G and S components in particular displayed high positive contributions in sites with higher λ than average (αN = +0.152 and αB = +0.131; Fig. 4a), and almost all their matrix elements were higher than average. The reverse was true for site V where the very low λ (αV = −0.207) was due to overwhelming negative contributions (−0.239) with only a small positive contribution (+0.032). In contrast, the other two ‘poor’ sites had an average λ only slightly lower than the overall mean λ (αO = −0.059 and αL = −0.066) because the large negative contributions (−0.267 on average) were compensated for to a large extent by positive contributions (+0.204). In site O, low values for F elements were partly compensated for by high values for G elements; in site L, negative G contributions were buffered by high S contributions.

Figure 3.

Correlation diagrams between variation in population growth rate, λ, and the variation in different life history components. On the y-axis the deviation from the species mean population growth rate, Δλ, is plotted for each site mean (a, c, e) and for each year mean (b, d, f). On the x-axis we plotted the respective variation decomposition (LTRE) effects, summed over the matrix elements of four life history components separately: fecundity (F), growth (G), clonal reproduction and survival of side-rosettes formed in the previous year (C + T), and stasis and retrogression (S + R). Black symbols and trend lines signify regressions with R2 > 0.50: the variation in these life history components contributed most to Δλ. Open symbols signal weaker regressions. Note that the axes of different diagrams can have different scales.

Figure 4.

Site effects in the decomposition of variation (LTRE) in population growth rate (λ) in Succisa pratensis (a), Hypochaeris radicata (b) and Centaurea jacea (c). Contributions of positive and negative matrix elements are separately grouped by life history component: fecundity (F), growth (G), clonal reproduction (C), survival of side-rosettes formed in the previous year (T), stasis (S) and retrogression (R). For each site and species the mean λ, the site effect (αm) and the mean absolute value of the interaction effect (αβ)(mn) are given. The overall mean λ for each species is 1.176, 0.811 and 0.910, respectively.

The year effects (mean | β | = 0.060) in S. pratensis were not only smaller than the site effects (mean | α | = 0.123) but also qualitatively different. In contrast to the differences between sites (mean | α | = 0.096 for F, Fig. 4a), F did not contribute much to variation in λ between years (mean | β | = 0.025 for F, Fig. 5a). Growth (G) contributions, however, covaried positively with λ variation over years (Fig. 3b) but did not correlate with λ variation over sites (Fig. 3a). Especially in the last year, which had the lowest λ, G and S had large negative contributions. This indicates that rosettes grew less often to larger classes in this year. Because they remained small, or even retrogressed, some S and R elements were actually higher than average, resulting in positive LTRE contributions that buffered the predominantly negative contributions.

Figure 5.

Year effects in the decomposition of variation (LTRE) in population growth rate (λ) in Succisa pratensis (a), Hypochaeris radicata (b) and Centaurea jacea (c). Contributions of positive and negative matrix elements are separately grouped by life history component: fecundity (F), growth (G), clonal reproduction (C), survival of side-rosettes formed in the previous year (T), stasis (S) and retrogression (R). For each year and species the mean λ, the year effect (βn) and the mean absolute value of the interaction effect (αβ)(mn) are given. The overall mean λ for each species is 1.176, 0.811 and 0.910, respectively.

In the shorter-lived H. radicata, site effects were smaller than year effects (Table 2). The contributions to λ by life history components were more consistent between site and year effects than they were in S. pratensis, probably because of the very flexible life history responses of H. radicata (Hartemink et al. 2004). G again covaried strongly with λ, but other components had less effect (Fig. 3c,d). Positive and negative contributions of G and S were largely in balance in all three sites (Fig. 4b). Both G and S were both strongly positive in the first year and strongly negative in the last year (Fig. 5b).

Centaurea jacea showed little variation in λ (Fig. 1), and therefore the different effects were also smaller, with site effects being the most important (Table 2). Clonal reproduction (C) contributions had the strongest correlations with variation in λ, both between sites and between years (Fig. 3e,f). Variation in F was negatively correlated with variation in λ between sites, but not between years. Within sites, C made the largest contributions, although G was also important in site B. Contributions of all life history components tended to be either positive or negative in a given site (Fig. 4c), except in poor years, when positive and negative contributions occurred simultaneously (Fig. 5c).


differences in demography between species

The three-way LTRE showed that variation between species made up the largest contribution to the observed variation in λ, followed by year-to-year variation and then by site-to-site variation. This means that the demographic behaviours of the co-occurring study species were relatively more similar in time than in space. This occurred because one year (1999–2000) had the highest growth rates in all species, whereas the growth rate in the last year (2002–03) was below average for all species, probably because of the dry early spring and summer of 2003 (precipitation was only 265 mm from February through August, which is 51% of the average in the other four study years; data from the meteorological station De Bilt of the Royal Netherlands Meteorological Institute). However, the magnitude of the temporal variation differed between species: year-to-year fluctuations in environmental conditions were buffered best by the clonal life history of the woody Centaurea jacea and worst by the life history of the short-lived Hypochaeris radicata.

Comparing the species, Succisa pratensis had the highest population growth rates, although it has declined as a result of habitat destruction and deterioration (Soons et al. 2003; Vergeer et al. 2003). Its good performance in this study is probably because the current management in the reserves fits the species well or because the weather conditions were particularly favourable in the first years of the study.

temporal and spatial variation in population dynamics

Our results suggest that year and site variations influence population dynamics in very different ways, especially when considering the way in which low growth rates were realized. For example, in S. pratensis low values of λ were caused by below average fecundities in poor sites, but by below average growth in poor years. As in the perennial herb Lathyrus vernus (Ehrlén 1995) site effects were stronger than year effects in S. pratensis. Although the differences between temporal and spatial effects were much smaller in the flexible and short-lived H. radicata and the clonal C. jacea, fecundity again covaried with variation in λ between sites, but not between years. High fecundity has frequently been reported to cause high population growth, e.g. in the non-clonal Centaurea corymbosa (Fréville et al. 2004), but no distinction was made between spatial and temporal variation in these studies.

Large interaction effects between sites and years would mean that demographic variation over time is weakly correlated between populations (Menges 2000). In our case, however, the interaction effects were always intermediate in strength between the site and the year effect. Hypochaeris radicata showed marked population declines in most sites over time. The last year, which was also the worst year for S. pratensis, may partly be explained by drought (see above). Another explanation may be found in the higher rosette turnover rate of shorter-lived species. By selecting sites for permanent plots in which H. radicata rosettes were present at the beginning of our study we may have been biased to find population decline (Crawley 1990), but as we did not observe high colonization rates in the vicinity of the plots, this is unlikely to be the main explanation for the observed population decline.

flexibility in life history responses

In two of the three sites in which S. pratensis performance was below average, and in both sites in which H. radicata performance was below average, large negative LTRE contributions were almost completely countered by positive contributions of other life history components, resulting in a small, negative net effect. Cancelling out of negative and positive contributions of different elements of the life cycle seems to be a more general pattern. For example, negative covariances between life history components were also reported by Picóet al. (2002) and Sterck et al. (2003). These negative correlations may reflect trade-offs between life history functions. For instance, the least productive site, L, in which S. pratensis and H. radicata plants remained small, showed negative contributions of growth (G) and compensating positive contributions of stasis (S) when compared with the mean reference matrix. This can be understood from the observation that G was reduced whereas survival (G + S + R) was not. Similar trade-offs between different fates of surviving meristems were found in the tropical tree Vouacapoua americana (Sterck et al. 2003).

In another poor site for S. pratensis, site O, a negative LTRE contribution of F (caused by mowing before seed set) was partly compensated for by a positive G contribution. A demographic trade-off between F and G may be expected because Hartemink et al. (2004) experimentally found costs of seed production in S. pratensis, expressed as reduced vegetative growth. However, such a trade-off is unlikely to explain fully the high G contribution, as G may also be high as a result of favourable site-specific environmental conditions in this productive and somewhat drier site. Experimentation would be required to compare the importance of external factors compared with those of trade-offs. Life history flexibility was also found to buffer population growth rate variation in H. radicata (between S and G, in the same low-productive site L) and in C. jacea (C vs. G and T, in different poor years).

Buffering of negative contributions to variation in λ, irrespective of the underlying mechanisms, is important for population dynamics as it reduces variance in λ. Lower variance in λ leads to lower population extinction risks (Tuljapurkar & Orzack 1980; Menges 1998). However, the spatial synchrony in the dynamics of the populations of S. pratensis and especially of H. radicata makes these species regionally more vulnerable to extinction (Harrison & Quinn 1989; Heino et al. 1997; Matter 2001).


Temporal and spatial variation in the dynamics of populations may not be readily substitutable, as we have found that the effects of poor years can be different from those of poor sites and that sites may also show different reasons for being poor. Model studies on the regional dynamics of plants should take this point into consideration. Population viability analyses depend on the demographic variation that is incorporated (Menges 2000), and using spatial variation instead of temporal variation may therefore result in incorrect estimates of extinction risks. The differences between temporal and spatial variation are likely to be the result of the different factors that generate the variation, i.e. climatic fluctuations and variation in management vs. site productivity and vegetation composition.

Our LTRE analyses show that variation in very different life history components may underlie the variation in λ, even among species with similar life histories. Although S. pratensis and H. radicata showed similar responses to the spatiotemporal variation in growing conditions, the magnitude of the responses differed considerably, whereas the third species (C. jacea) barely responded when exposed to exactly the same variation in conditions. Attempts to predict the population dynamics of poorly studied species from data collected from well studied species [see Heppell et al. (2000) for an example with mammals] should therefore be cautiously undertaken in plants, as our results suggest that generalizations among plant species can be difficult even when they possess similar life histories.


We thank Frank Berendse, Johan Ehrlén, Linda Jorritsma-Wienk, Felix Knauer, Juul Limpens, Carolin Mix, Xavier Picó, Merel Soons, Jasper van Ruijven, Emily Rauschert, Michael Hutchings, Teresa Valverda, Lindsay Haddon and two anonymous referees for their helpful comments and fruitful discussions. The Netherlands Organization for Scientific Research funded this research (NWO project 805-33-452).

Supplementary material

The following material is available from

Table S1 Number of plots and census and mowing dates

Table S2 Flower head and seed counts