Merel Soons, Utrecht University, Sorbonnelaan 16, 3584 CA Utrecht, the Netherlands (e-mail email@example.com).
1Habitat fragmentation as a result of intensification of agricultural practices decreases the population size and increases the site productivity of remnant populations of many plant species native to nutrient-poor, species-rich grasslands. Little is known about how this affects the colonization capacity of populations, which is highly important for regional species survival. We studied the effects on four wind-dispersed forbs that represent two major dispersal strategies in grasslands: Cirsium dissectum and Hypochaeris radicata, which have plumed seeds and are adapted to long-distance dispersal by wind, and Centaurea jacea and Succisa pratensis, which have plumeless seeds and are adapted to only short-distance dispersal by wind.
2Colonization capacity decreased with decreasing population size. This was due to lower seed germination ability in all species, and a lower seed production and a narrower range of seed dispersal distances in the species with plumed seeds. Inbreeding depression is the most likely cause of this. We found no evidence for a stronger selection for reduced dispersal in smaller populations.
3Increasing site productivity changed the colonization capacity of all species. The capacity for colonization of nearby sites increased, due to higher seed production and seed germination ability, but the capacity for colonization of distant sites decreased, due to a lower long-distance dispersal ability.
4Seed dispersal ability and germination ability were negatively correlated in the species with plumeless seeds, but not in the species with plumed seeds. The dispersal ability of individual plumed seeds remained constant under changes in population size and site productivity. This indicates a strong selection pressure for long-distance dispersal ability in these species.
5When habitat fragmentation results in a simultaneous decrease in population size and increase in site productivity, both the local survival probability and the colonization capacity of remnant populations of wind-dispersed grassland forbs are likely to be severely reduced. This increases regional extinction risks of the species.
Nutrient-poor, species-rich grasslands in north-west Europe are becoming highly fragmented due to the intensification of agricultural practices (Vos & Zonneveld 1993; Bakker & Berendse 1999). As a result, populations of plant species that are restricted to this habitat are becoming ever smaller and more isolated and consequently their extinction probabilities rise (Ellstrand & Elam 1993; Ouborg 1993). Habitat fragmentation also increases the extinction probabilities of these populations indirectly by decreasing the quality of the remaining habitat patches: as the patches become smaller and more isolated, they become more vulnerable to influences from the surrounding agricultural landscape (e.g. Neitzke 2001). This primarily results in a higher productivity due to an increased inflow of nutrients (Bakker & Berendse 1999), often accompanied by acidification and lowering of the water table (Vos & Zonneveld 1993).
Despite the increasing extinction risks of local populations, a species may survive regionally if there is sufficient colonization of unoccupied habitat patches (Van der Meijden et al. 1992; Ouborg 1993; Husband & Barrett 1996). Little is known, however, about the effects of decreasing population size and increasing site productivity on the colonization capacity of isolated remnant populations. We define colonization capacity of a population as the capacity to establish seedlings at suitable sites not occupied by individuals from that population. Whilst re-colonization from the seed bank also plays an important role in this capacity (Bakker et al. 1996; Strykstra et al. 1998a), this study focuses only on colonization through spatial seed dispersal. This colonization capacity is determined by seed production, dispersal ability and germination ability.
The effects on dispersal ability are, however, less clear. Wind is the main dispersal agent in early successional vegetation types such as grasslands (Van der Pijl 1982; Fenner 1985). Because long-distance dispersal is a very important aspect of dispersal ability, but almost impossible to measure (Bullock & Clarke 2000; Cain et al. 2000), various mechanistic models have been developed that predict wind dispersal from physical laws (e.g. Okubo & Levin 1989; Andersen 1991; Nathan et al. 2001, 2002; Tackenberg 2002). These models describe the flight of seeds based on seed, plant, vegetation and wind characteristics. The plant-controlled characteristics that determine wind dispersal distance are seed terminal velocity (VT) and seed release height. Release height is only relevant, however, when the seed is released above the directly surrounding vegetation (Sheldon & Burrows 1973) and above the height at which horizontal wind speed is zero. We therefore use relative seed release height (Hrel, the difference between seed release height and the height at which wind speed is zero) as the second plant-controlled dispersal parameter.
Decreased population size and increased site productivity might be expected to have a negative impact on dispersal ability. First, it has been suggested that isolated populations experience selection for reduced dispersal ability, as only seeds that disperse nearby contribute to the genetic pool (Carlquist 1966). In small populations such selection would be stronger than in large populations. Cody & Overton (1996) found measurable effects of selection for reduced dispersal in small isolated populations after less than six plant generations; these effects consisted of changes in seed traits that determine VT. Secondly, in small fragmented populations genetic drift may reduce the variation in plant traits (Ellstrand & Elam 1993; Booy et al. 2000). If the variation in traits that determine VT is reduced, this results in a narrower range of dispersal distances (Augspurger & Franson 1993), and thus lower dispersal ability. Thirdly, if plants at more productive sites produce heavier seeds, their wind dispersal ability may be lower, as heavier seeds are likely to have a higher VT (e.g. Greene & Johnson 1993). In addition, wind dispersal may be hampered by tall vegetation at a productive site if the flowering stalks are shorter than the surrounding vegetation, i.e. Hrel is low.
Effects on seed production, dispersal ability and germination ability are not independent of each other. In particular, changes in dispersal and germination ability may be correlated: lighter seeds are better dispersed by wind but have a lower germination ability in Arnica montana (Strykstra et al. 1998b). So far, the mechanisms by which changes in seed production, dispersal and germination ability influence each other remain unknown, as the three have not been studied together. Furthermore, the effects of decreasing population size and increasing site productivity on seed dispersal ability have not been studied, and differences between species with different dispersal strategies are unknown. The purpose of this study is to fill in these gaps and to clarify the effects on the colonization capacity of fragmented populations. We studied isolated remnant populations of grassland forbs with two different wind-dispersal strategies, i.e. long-distance and short-distance seed dispersal. Specifically, we addressed the following questions:
• Does a decrease in population size reduce colonization capacity by decreasing the number of viable seeds produced per plant, variation in VT of the seeds, and/or germination ability of the seeds?
• Does a decrease in population size result in stronger selection for reduced dispersal, so that smaller populations have lower Hrel and/or higher seed VT?
• Does an increase in site productivity change colonization capacity by increasing the number of viable seeds produced per plant, decreasing Hrel, and/or increasing VT and germination ability of the seeds?
• Do seeds with a higher ability for long-distance dispersal due to a lower VT have a lower germination ability?
Four wind-dispersed grassland forbs confined to nutrient-poor, species-rich grasslands were selected as study species. All species are outcrossing (Grime et al. 1988; Smulders et al. 2000) and were common in the Netherlands before the rapid fragmentation and destruction of their habitat. Two of the study species, Cirsium dissectum (L.) Hill and Hypochaeris radicata L. (both Asteraceae), have plumed seeds and are adapted to long-distance dispersal by wind (Fig. 1). The other two study species, Centaurea jacea L. (Asteraceae) and Succisa pratensis Moench (Dipsacaceae), are also classified as wind-dispersed, but have plumeless seeds that are dispersed by wind only over short distances. The plumeless, smooth-surfaced seeds of C. jacea (Fig. 1) are ejected from the seed head, and transported further by wind, when strong winds fling the seed head back and forth (Bouman et al. 2000). Seeds of S. pratensis are surrounded by a persistent calyx (Fig. 1), which is dry and hairy and increases the surface area of the seed without adding much weight (Bouman et al. 2000).
All four species are hemicryptophytes that overwinter as leaf rosettes (Grime et al. 1988; Van der Meijden 1990). In summer they produce long flowering stalks that (at least under nutrient-poor conditions) protrude from the surrounding vegetation. Rosettes of C. dissectum produce one flowering stalk that carries one, rarely two, flower heads. Rosettes of the other species may produce multiple branched stalks, with various numbers of flower heads. All species propagate both sexually via seeds and asexually by side-rosettes and clonal extension (especially C. dissectum). C. dissectum, S. pratensis and C. jacea are long-lived species, while individuals of H. radicata live for a decade at most. All species have transient seed banks (Thompson et al. 1997).
collection of field data and sampling
Ten fragmented populations, representative of the full range of population sizes and site productivities in the Pleistocene soil areas in the Netherlands, were selected for each species. Effective population size was determined by counting the number of flowering rosettes. Productivity was assessed by clipping three vegetation plots of 20 × 20 cm in or, in the case of very small populations, close to the population. The vegetation samples were dried in a stove for 3 days at 75 °C, and dry weight was averaged for each site. Although a decrease in population size and an increase in site productivity co-occur in many populations, we took care to select populations so that size and productivity were not correlated (Fig. 2). For S. pratensis site productivity of two populations could only be estimated and these populations were discarded from statistical analyses of site productivity effects; the remaining eight populations, however, showed a correlation between size and productivity (Fig. 2). This has been taken into account during interpretation of the data on productivity effects. All selected populations have the same management regime of mowing once per year in autumn. The selected C. dissectum and S. pratensis populations have been isolated from populations of conspecifics by more than 1 km for over 50 and 25 years, respectively; populations of C. jacea and H. radicata have been isolated for c. 25 years by more than 500 m and 100 m, respectively (Soons et al., unpublished data). All selected H. radicata populations are also isolated by forest landscape elements that act as a barrier to seed and pollen transport. As this species is still quite common in the study area, however, only seven isolated populations of different sizes were found.
In each population, 10 plants with intact seed heads containing mature seeds were selected randomly for measuring seed production, Hrel and VT. In populations with less than 10 individuals, all individuals were measured. Seed production was measured by counting the number of seed heads per flowering stem and the number of viable seeds in the first produced (top) seed head. All filled seeds were assumed to be viable. In case of doubt, seed weight was used to determine viability: a previous germination experiment showed that filled seeds of all weights of H. radicata and S. pratensis could be viable, but partially filled seeds of C. jacea and C. dissectum with seed weights below 1 and 2 mg, respectively, were never viable. Seeds of these species that appeared (partially) filled but weighed less than this were discarded. The number of seed heads per stem and the number of viable seeds in the first seed head provide good estimates to compare seed production between individuals within a species. Total seed production per individual can, however, not be calculated from this in H. radicata, C. jacea or S. pratensis, because seed heads lower along the stem produce fewer seeds and stem number is also important. Stem number was not measured as it was impossible to determine non-destructively which rosettes (and thus stems) belong to one individual. Hrel was calculated as the height above the soil surface of the first produced seed head, minus the height at which horizontal wind speed is zero (approximately 0.74 times the vegetation height; Monteith 1973). Average vegetation height (excluding emerging flowering stalks) was estimated visually and measured with a ruler.
determining terminal velocities
Experimental determination of VT of many seeds is time-consuming and therefore VT of the seeds was calculated using the mechanistic relationship (Burrows 1973):
where M is seed mass, D is seed diameter, g is the gravitational acceleration, ρa is the density of air, and Cd is the aerodynamic shape constant of the seed. Although seed shape within species is somewhat variable, it is assumed that under average conditions it varies relatively little. Therefore, Cd may be considered a species-specific constant (Burrows 1975), and the second term of equation 1 becomes a constant that was determined experimentally by measuring VT, M and D of a subset of the seeds. VT values of the other seeds were calculated from their M, D, and this constant. Although beak length of the seeds of H. radicata is also variable, it does not affect VT (Cody & Overton 1996).
Eight seeds per collected seed head (i.e. 800 per species) were selected randomly for measurements of M and D. The selected seeds were weighed using Sartorius Ultramicro scales. For the plumed seeds D was measured as the maximal diameter of the completely opened plume. For C. jacea the maximum diameter of the seed coat was measured, and for S. pratensis the maximum diameter of the seed including the calyx was measured. Sliding callipers (accuracy 0.05 mm) were used for all diameter measurements.
One hundred seeds per species (25 for H. radicata) were selected randomly for VT measurements; in the case of plumed seeds, only seeds with open and intact plumes were selected. VT of the plumed seeds was measured by dropping the seeds over 2 m in still air inside a hardboard and Plexiglas shaft (cf. Andersen 1992; Cody & Overton 1996). Drop time was measured manually using a digital stopwatch (accuracy 0.01 s) and averaged over three falls seed−1. Plumeless seeds were dropped down a 15.83-m shaft with mechanical drop time measurements (Jongejans & Schippers 1999). Because this shaft was made of thick plastic foil it was not entirely resistant to air movements and the air inside the shaft was not perfectly still. This reduced the high precision of the measurements, mainly because horizontal air movement caused some S. pratensis seeds to touch the inner side of the tower, thereby disrupting their fall. VT values were calculated from drop time (T) and drop height (H) using a numerical solution of the equation describing T as a function of h and VT, including the acceleration phase of the fall (cf. Burrows 1975):
VT values of all eight seeds selected per seed head were calculated using equation 1. Both per plant and per population the variation in VT was calculated as the coefficient of variation of VT values, with CV = (SD/mean) × 100. The CVs calculated per plant were averaged per population, so for each population the averaged individual CV and the overall population CV are known.
the germination experiment
Four seeds per individual (i.e. 400 per species) were randomly selected for the greenhouse germination experiment. Seeds of C. jacea, C. dissectum and H. radicata were air dried and stored in plastic pots in the dark at 4 °C for 4 months. Seeds of S. pratensis were stored similarly but at room temperature (c. 18 °C), as storage at low temperatures has been found to reduce germination ability (Kotorová & Leps 1999). Seeds were put to germinate on steamed, moist sand at 21 °C, under a light/dark regime of 16/8 h and high air humidity (> 90%). A seed was classified as germinated when the radicle was visible. Germination was followed for 10 weeks and scored twice per week.
All statistical analyses were conducted using SPSS 8.0. Significances of the species-specific relationships between seed VT and M and D were tested using linear regression analysis. Log-transformed population size was used in all analyses on population size effects. The effects of population size and site productivity on the number of viable seeds, Hrel, VT, the variation in VT, and the average time to germination of the seeds, were also determined by linear regression analyses. We tested whether effects were equal for both species representing the same dispersal strategy (long-distance or short-distance wind dispersal) by calculating the significance of a regression model that includes both species:
y = β0 + (β1 × species) + (β2 × x) + (β3 × species × x)( eqn 3)
where y is the dependent variable, species is a dummy variable, and x is the independent variable. When β3 is not significant, the two species do not have significantly different regression coefficients. In these cases, a regression analysis excluding the last term in equation 3 was carried out to determine the regression coefficient β2 and model statistics P and R2. If there was no significant effect of population size or productivity for both species, but the data indicated a species-specific effect in one of the species, a linear regression analyses was carried out to test significance of that effect. Effects of log-transformed population size and site productivity on seed germination were tested for significance with a logistic regression analysis with Nagelkerke's R2 value as indicator of the explained variation (Nagelkerke 1991).
determining terminal velocities
For the plumed seeds, the relationship between VT and √M/D is exactly as described in equation 1, i.e. positive, linear and passing through the origin (Fig. 3). For the plumeless seeds, however, VT is explained better by √M than by √M/D (C. jacea R2 = 0.574, P < 0.001 vs. R2 = 0.325, P < 0.001; S. pratensis R2 = 0.264, P < 0.001 vs. R2 = 0.247, P < 0.001) and √M was therefore used to estimate VT for these seeds. In spite of a positive intercept at the y-axis, √M predicts VT of the plumeless seeds adequately (Fig. 3), even for S. pratensis, where the relationship is much weaker due to lower accuracy of VT measurements (see Methods). For all species, the relationship that predicts VT best is linear. This shows that Cd can be considered a species-specific constant and justifies the assumption that VT can be estimated as a linear function of √M/D or √M.
Average seed production statistics per species are presented in Table 1. The expected decrease in seed production with decreasing population size was found in the species with plumed seeds, due to lower numbers of viable seeds per seed head (Table 2). The species with plumeless seeds did not produce fewer seeds, and smaller populations of S. pratensis even had more heads per stem (Table 2).
Table 1. Average values of colonization capacity parameters (species mean ± SD). Nheads= number of seed heads per stem, Nviable= number of viable seeds in the first seed head, Hrel= relative seed release height, VT= calculated seed terminal velocity, Pop CV VT= variation in VT per population, Ind CV VT= variation in VT per individual, Germ = germination percentage
Pop CV VT
Ind CV VT
With plumed seeds
24.69 ± 16.00
9.31 ± 21.49
0.380 ± 0.036
7.23 ± 1.72
5.19 ± 1.33
2.88 ± 2.31
54.59 ± 22.42
5.20 ± 16.23
0.335 ± 0.039
10.08 ± 2.65
6.97 ± 2.90
With plumeless seeds
5.69 ± 7.57
31.72 ± 10.91
5.98 ± 19.26
4.33 ± 0.44
8.38 ± 2.07
5.47 ± 3.25
4.59 ± 3.45
44.40 ± 23.04
24.87 ± 21.08
2.14 ± 0.18
7.15 ± 1.44
3.77 ± 2.22
Table 2. Regression coefficients (β) and model statistics (P and R2) of the regressions of plant variables on log-transformed population size and site productivity. Relationships valid for both species sharing the same dispersal strategy are shown, as well as species-specific relationships when present (denoted by Cj for C. jacea, Hr for H. radicata, Sp for S. pratensis). Nheads= number of seed heads per flowering stalk, Nviable= number of viable seeds in the first seed head, Hrel= relative seed release height (cm), VT= seed terminal velocity (m/s), M= seed mass (mg). Pop CV is the coefficient of variation per population, Ind CV is per individual. The regression analysis of germination percentage on population size showed no differences between the two dispersal strategies and regression statistics were calculated for all species together.  indicates a marginally significant relationship (0.050 < P < 0.070)
β log-transformed population size
β site productivity
Sp: −1.85 (P = 0.004; R2 = 0.719)
[Cj: 2.19·10−2 (P = 0.054; R2 = 0.353)]
6.68 (P = 0.007; R2 = 0.505)
Sp: 8.63·10−2 (P = 0.011; R2 = 0.690)
Hr: −8.57 (P = 0.007; R2 = 0.794)
−8.57·10−2 (P = 0.002; R2 = 0.583)
5.96·10−2 (P = 0.000; R2 = 0.976)
8.29·10−4 (P = 0.000; R2 = 0.983)
1.12 (P = 0.013; R2 = 0.366)
−0.804 (P = 0.030; R2 = 0.236)
Hr: 1.42 (P = 0.014; R2 = 0.811)
−1.03 (P = 0.007; R2 = 0.339)
9.56·10−2 (P = 0.000; R2 = 0.703)
1.35·10−3 (P = 0.000; R2 = 0.805)
[3.46 (P = 0.062; R2 = 0.277)]
−1.44 (P = 0.046; R2 = 0.204)
[0.694 (P = 0.063; R2 = 0.226)]
−1.78 (P = 0.003; R2 = 0.497)
7.17·10−3 (P = 0.030; R2 = 0.417)
4.22 (P = 0.000; R2 = 0.767)
4.22 (P = 0.000; R2 = 0.767)
−5.42·10−2 (P = 0.000; R2 = 0.725)
2.34·10−2 (P = 0.000; R2 = 0.725)
The expected increase in seed production with increasing site productivity was significant in the species with plumeless seeds (Table 2), while the species with plumed seeds only showed a non-significant trend in this direction. Populations at more productive sites produced more seeds in all species, but due to the very slight increases in number of seed heads per stem and number of seeds per head in C. dissectum, overall relationships with increasing productivity were not significant for the species with plumed seeds. In the species with plumeless seeds, plants at more productive sites produced significantly more heads per stem in C. jacea and significantly more seeds per head in S. pratensis (Table 2).
Average values of Hrel, VT and the variation in VT are presented in Table 1 (VT values calculated using equation 1). Average seed VT shows the greatest variation between species: almost tenfold instead of two- to fivefold for other characteristics. The averaged individual CV of VT was always lower than its population CV.
We found no evidence for stronger selection for reduced dispersal ability in smaller populations than in larger populations. Hrel did not decrease with decreasing population size in any of the species (and, in H. radicata, Hrel was even significantly higher in smaller populations; Table 2), nor did VT increase. In the species with plumeless seeds there was even a significant positive relationship between VT and population size, due to a decrease in seed mass with decreasing population size (Table 2). The plumed seeds also decreased in seed mass with decreasing population size, but this relationship was not significant (β = 5.77·10−2, P = 0.142, R2 = 0.972) and it did not affect VT as it was balanced by increased seed diameters. The variation in VT decreased with decreasing population size in the species with plumed seeds, but increased in the species with plumeless seeds (Table 2). Thus, in contrast to our expectations, it appears that in the species with plumed seeds dispersal ability is lower in smaller populations, due only to a reduced variation in VT, whereas in the species with plumeless seeds dispersal ability is higher in smaller populations, due to both a lower VT and a higher variation in VT.
Populations at more productive sites had a lower Hrel, as expected, though this was only significant in the species with plumed seeds (Table 2). In the species with plumed seeds no relationship between VT and site productivity was found, but in the species with plumeless seeds VT increased with increasing site productivity, mainly due to the significant increase in their seed mass (Table 2). The plumed seeds increased in seed mass as well, but this again was not significant (β = 4.08·10−4, P = 0.182, R2 = 0.971) and did not result in an increase in VT, as it was compensated for by increased plume diameters. Site productivity did not affect variation in VT (Table 2). Thus, an increase in site productivity only decreased the VT of plumeless seeds, not of plumed seeds.
As expected, seed germination decreased with decreasing population size in all species (Table 2). Germination of the plumeless seeds increased with increasing site productivity, as expected, but germination of the plumed seeds was lower at more productive sites (Table 2). The latter was due entirely to the strong decrease in germination in C. dissectum, however; H. radicata showed a slight increase, just like the species with plumeless seeds. Time to germination was not affected by population size or site productivity.
For the plumeless seeds a positive relationship between germination and individual seed √M, and thus VT, was found (Table 3). No relationship between germination and √M/D, and thus VT, was found for the plumed seeds. The hypothesis that seeds with a higher long-distance dispersal ability due to a lower VT value have a lower germination ability, was thus supported by the data on the plumeless seeds only.
Table 3. Statistics of the logistic regression of germination status (germinated or not germinated) on the seed variable √M/D (for the species with plumed seeds) or √M (for the species with plumeless seeds)
Species with plumed seeds
Species with plumeless seeds
determining terminal velocities
The mechanistic model used in this study to calculate VT from √M/D provides a simple and accurate method for determining VT values of large quantities of plumed seeds of the same species. For plumeless seeds, the method is even simpler as just√M predicts VT accurately, though the fit between √M and VT is low in S. pratensis, which is at least partly explained by the lower precision of measurements of VT. The model used in this study relates seed mass, diameter and terminal velocity to each other in a mechanistic way and is therefore not only a useful tool to derive terminal velocity values of large quantities of seeds, but also explains the relationships between seed characteristics.
a decrease in population size reduces the colonization capacity
A decrease in population size affects the species with plumed seeds and the species with plumeless seeds differently. The lower values for seed production, dispersal ability (due to lower variation in VT) and germination ability in smaller populations of species with plumed seeds are as expected from the literature (Table 4). The lower seed production may have been caused by inbreeding depression and/or pollination limitation as both have been demonstrated to occur in fragmented forb populations (Fischer & Matthies 1998b; Groom 2001; Moody-Weis & Heywood 2001). Inbreeding may likely have contributed to this, as it may also explain the lower seed germination ability of all species. For H. radicata and S. pratensis it has been demonstrated that inbreeding results in lower seed germinability (C. Mix, unpublished data). The selected species are all outcrossing and have become severely fragmented over a relatively short time period, so they may be especially susceptible to inbreeding depression (Husband & Schemske 1996; Booy et al. 2000). The lower variation in VT in the species with plumed seeds may be due to genetic drift (Ellstrand & Elam 1993; Booy et al. 2000). This is in agreement with another study on the S. pratensis populations that documents lower allozyme variation in smaller populations (P. Vergeer, unpublished data). These results suggest that the isolation levels of the selected populations are sufficient to affect their genetic composition, even though low rates of pollen and seed flow may still occur.
Table 4. Expected and observed relationships between colonization capacity variables and population size or site productivity. Species-specific relationships are indicated by Hr (H. radicata) or Sp (S. pratensis). O = no relationship; += positive relationship (P ≤ 0.050); –= negative relationship (P ≤ 0.050)
Number of viable seeds
Relative release height (Hrel)
Terminal velocity (VT)
Variation in terminal velocity (CV VT)
In the species with plumeless seeds, smaller populations had a lower seed germinability but, in contrast to expectation, smaller populations of S. pratensis produced more seeds and smaller populations of both species had seeds with a lower average VT and a larger variation in VT (Table 4). The increase in variation in VT and the decrease of average VT values are both due to the production of increasing numbers of low quality seeds with low seed mass and, thus, low VT. This is likely to be another effect of inbreeding depression, and may be the direct cause of the lower germinability of the seeds. Such a change in VT is not observed in the species with plumed seeds, as plumed seeds with a lower seed mass also have a smaller plume diameter, and VT remains constant. Thus, the long-distance dispersal ability of individual plumed seeds remains unaffected. The data suggest that plumeless seeds from smaller populations have a higher dispersal ability. However, in the species with plumeless seeds, a strong positive correlation between VT and germinability of individual seeds was found. Because of this relationship, the decrease in VT and the increase in variation in VT hardly increase the dispersal ability of the seeds, as the seeds that disperse furthest are the ones least likely to germinate.
Overall therefore the colonization capacity of all four wind-dispersed grassland forbs is reduced by a decrease in population size, with the lower seed production and seed germinability also reducing local recruitment to the population and the (transient) seed bank.
no short-term evolution of reduced long-distance dispersal
No support for the hypothesis that selection pressure for reduced dispersal is more effective in smaller populations was found for any species (Table 4). Cody & Overton (1996) found evidence of selection for reduced dispersal in wind-dispersed species after just up to six generations of isolation, but in this study no indication of such selection was found at all. For C. dissectum, S. pratensis and C. jacea the time of isolation spanned only a few plant generations, and may be too short for such a selection pressure to have had a measurable effect, but populations of H. radicata have been isolated for up to 12 generations. Possibly, the isolation levels of the remnant populations may not have been high enough to prevent gene flow sufficiently, particularly in H. radicata. Although both seed dispersal by wind and pollen dispersal by insects over distances of more than 100 m are rare, the occurrence of low rates of gene flow may explain the lack of evidence for selection against dispersal. The isolation levels did, however, appear to be sufficient to affect the genetic composition of small populations.
an increase in site productivity changes the colonization capacity
The effects of an increase in site productivity were mostly as expected (Table 4). In the species with plumed seeds, however, the strong decrease in germination ability of C. dissectum seeds resulted in an overall significant negative relationship between site productivity and seed germination ability, despite H. radicata showing the expected positive relationship. Other studies on effects of maternal site productivity on germinability yielded ambiguous results. In most cases, an increase in site productivity resulted in larger plants with more reproductive output (Bazzaz et al. 2000) and higher seed germinability (Roach & Wulff 1987). Nevertheless, plants may also produce seeds with a lower germinability when their nutrient availability is increased (Wulff & Bazzaz 1992; Galloway 2001). This has been attributed to a relatively larger allocation to the seed coat than to endosperm and embryo, but if this is the case, also time to germination is generally longer (Galloway 2001). This was not found in C. dissectum, where lower seed quality is a more likely cause for the decrease in germination.
In one important aspect the effect of an increase in site productivity is not as expected: VT increased with increasing productivity for plumeless but not for plumed seeds (Table 4). The significant increase in VT of the plumeless seeds is due to the increase in seed mass, which is also reflected in the increase in germination ability. The VT of the plumed seeds is not affected, because the increase in seed mass is proportional to the increase in plume diameter.
Overall, an increase in site productivity changes the colonization capacity of all four wind-dispersed grassland forbs: more seeds are produced, and these seeds germinate better (except in C. dissectum). All seeds are dispersed over shorter distances, as their dispersal ability is lower. Higher site productivity thus appears to result in a higher colonization capacity at shorter distances, but a lower colonization capacity at longer distances. This may result in an overall lower recruitment, because the seedlings that emerge closer to the parent plant may suffer from increased density-dependent mortality (Willson & Traveset 2000). Also, the colonization capacity over time may be reduced, as larger seeds tend to form shorter-lived seed banks (Thompson et al. 1998).
dispersal ability and germination ability are negatively correlated in plumeless seeds only
In contrast to the findings of Strykstra et al. (1998b), no relationship between dispersal ability as determined by VT and germination ability was found in the species with plumed seeds. In the plumeless seeds, however, lower VT values were caused by lower seed mass and strongly correlated to lower germination ability. We found that both smaller population size and lower site productivity correlate with lower seed mass of plumed seeds, but this never affected VT of these seeds due to compensating adjustments in plume diameter. This is in agreement with the findings of Matlack (1987), who measured a variation in seed weight by a factor of 82.3, but a variation in wing loading by only a factor 2.8 in plumed seeds of Asteraceae species. In species with plumed seeds, there appears to have been a strong selection for constant low VT values, ensuring a high long-distance dispersal ability of the seeds. This is in agreement with the results of recent model analyses (Tackenberg 2002; Tackenberg et al. 2002) that show species VT to be the most important plant-controlled variable in determining long-distance seed dispersal.
The results of this study show that a decrease in population size reduces the colonization capacity of fragmented populations of wind-dispersed grassland forbs, and that an increase in site productivity shifts their colonization capacity to a higher capacity to colonize nearby sites, but a lower capacity to colonize distant sites. Although these effects are not dependent on the wind-dispersal strategy of the species, the mechanisms that determine them are. The main difference between the dispersal strategies is that the long-distance dispersal ability of the species with plumed seeds is decreased only by a reduction in relative release height of the seeds, whereas for the species with plumeless seeds it is also reduced by an increase in their terminal velocity. This indicates a selection for constant low seed terminal velocity in the species adapted to long-distance dispersal.
Habitat fragmentation is a major cause of decreasing population size and increasing site productivity of remnant populations of wind-dispersed grassland forbs. In these populations a decrease in population size and an increase in site productivity often co-occur and their effects will reinforce each other. The resulting reduced colonization capacity, even when selection for reduced dispersal does not occur, has important consequences for species survival. Lower local recruitment contributes to a higher extinction risk of fragmented populations, while lower regional recruitment and colonization of new sites increase the regional extinction risks of species. For wind-dispersed species in increasingly fragmented landscapes, the connectivity between habitat patches not only decreases because distances between habitat patches increase, but also because the colonization capacity of the remaining populations decreases. Also, the total (re)colonization capacity of the species is reduced as even recruitment to the seed bank is reduced. Together, these factors pose yet another threat to the regional survival of wind-dispersed forbs restricted to nutrient-poor grasslands.
This research was supported by the Research Council for Earth and Life Sciences (ALW), with financial aid from the Netherlands Organization for Scientific Research (NWO). The park rangers of Staatsbosbeheer kindly permitted us to carry out field measurements in many of their nature reserves. Marinus Werger, Jan van Groenendael, Heinjo During and two anonymous referees made constructive comments on earlier versions of the manuscript, and Paul Stoy made linguistic corrections. Eelke Jongejans, Carolin Mix, Felix Knauer and Philippine Vergeer provided many helpful discussions. Cas Kruitwagen and Roger Donders of the Centre for Biostatistics, Utrecht University, advised on statistical analyses.