- Top of page
Species distributions are determined both by habitat quality and by the ability of species to colonize suitable habitats. Knowledge of the relative importance of these factors is crucial for predicting the impact of habitat fragmentation and climate change on vegetation. Although the role of seed availability for plant distributions and abundance has been increasingly emphasized in recent years (e.g. Turnbull et al. 2000; Rees et al. 2001; Bolker et al. 2003), there is still little empirical evidence underpinning this notion. Seed limitation may be inferred from patterns of seedling emergence relative to the parent plants, from contrasting distributions of species with different seed dispersal vectors or from the existence of suitable but unoccupied patches. The abundance of suitable unoccupied habitat indicates the relative importance of seed limitation vs. limitation by other factors as determinants of species distributions. The exact size and location of occupied and unoccupied patches can, assuming equilibrium conditions, be used to infer extinction and dispersal functions (Hanski 1994).
Although long-term experiments are often needed to assess seed limitation unequivocally, such experiments are difficult and time consuming. Patch suitability is therefore often assessed indirectly, using correlations between occurrence and habitat characteristics (e.g. Burke & Grime 1996; Hanski & Simberloff 1997; Stevens et al. 2004). Two main types of evidence can be used for this. First, information on the correlation between environmental variables and species occurrence could be used to identify unoccupied suitable habitats. Secondly, we could predict habitat suitability using data on the co-occurrence of the target species with other species (Beals 1984; Münzbergová & Herben 2004). However, an important question is how accurately such correlative evidence assesses habitat suitability.
In a previous 3-year study (Ehrlén & Eriksson 2000), we performed seed-sowing experiments at 48 different patches and showed that the availability of seeds was limiting recruitment in six of seven deciduous forest herbs. In the present study, we use 11 years of data for these six species to address the following questions. (i) Are regional distributions of perennial plants limited by seed availability, as judged from long-term recordings of sowing experiments? (ii) Do bottlenecks at post-seedling stages make short-term assessments poor predictors of the recruitment success over longer time periods? (iii) Do sowing experiments and correlations with environmental factors or vegetation composition identify the same habitat patches as suitable?
- Top of page
The results of the present study support the notion that the distributions of the investigated plant species are limited by seed availability in the study area. Because all studied species occur and reproduce in the study area, the most likely cause of seed limitation is that seeds have not been deposited at potentially suitable sites. Thus, these results imply that dispersal is a key process in the regional dynamics of these plants. This conclusion is true even when evaluated 11 years after transplantation. Yet the results indicate that there are several problems associated with detection of seed limitation.
These results for a set of forest herbs with relatively low rates of seed production, large seed size and long generation periods, in a relatively fragmented habitat, agree well with previous studies that have suggested that the distribution of such species may be limited by the availability of seeds (Eriksson & Ehrlén 1992; Ehrlén & Eriksson 2000; Clark et al. 1998; Graae et al. 2004; Verheyen & Hermy 2004). Four of the investigated species have fleshy fruits, i.e. a trait most likely associated with endozoochorous seed dispersal, and are thus potentially well dispersed. However, the only observed dispersal vectors of these fruits in the area are rodents that hoard fruits and consume the seeds (J. Ehrlén, O. Eriksson & H. von Zeipel, unpublished data), and dispersal may therefore occur only over relatively short distances.
It has been suggested that seed limitation, defined as the reduction in abundance and distribution due to a reduced number of available seeds, can be broken down into two components: a limited number of seeds at source populations and a limited dispersal of available seeds (Clark et al. 1998; Nathan & Muller-Landau 2000; Muller-Landau et al. 2002). The main objective of this study was to assess the extent to which the occurrence of respective species within a forest fragment was due to local habitat conditions or due to a lack of incoming seeds. We did not specifically address the issue of whether the lack of seeds was because the seeds were present in such low numbers in the study area or whether it was the result of seeds being available in the area but not able to disperse to the focal patches. Given that several of the investigated species have a considerable rate of seed production in the study area, it is reasonable to assume that seed limitation of recruitment at the patch level largely represents dispersal limitation.
The proportion of unoccupied sites with at least one surviving seedling, and the average number of seedlings within plots, was much lower when evaluated after 11 years than when evaluated after 3 years for all species except for Convallaria and Lathyrus. For Convallaria, the proportion of unoccupied patches with surviving individuals was high and relatively constant. This might be attributed to the fact that Convallaria has the largest realized niche of all the species included in this study (Table 1). Unlike the other species, Convallaria is capable of growing in all plant communities of deciduous forest in this study area. The disappearance of seedlings in many plots from year 3 to year 11 for the remaining species is in agreement with Turnbull et al. (2000), who found that although 64% of seed introductions resulted in seedlings, adults were established only at 23% of sites (although the time of recording was not provided). It has been suggested that monitoring of seed sowing experiments should continue at least until the plants reach reproductive age (Turnbull et al. 2000; Zobel & Kalamees 2005). Given that even 11 years after sowing the experimental species had only started to reproduce in single plots, it seems probable that individuals in several other plots will still die before they start to reproduce. Against the background of the long prereproductive life spans of the study species, one might argue that our 11-year study is also short-term and preliminary. However, demographic studies of two of the study species, Lathyrus and Actaea (Ehrlén 1995; Fröborg & Eriksson 2003), show that mortality decreases with size. Given also that surviving non-flowering individuals in sowing plots had attained a relatively large size, it is likely that we would have arrived at a somewhat, but not dramatically, lower estimated proportion of unoccupied suitable habitats if survival up to the age of reproduction had been used as a criterion for patch suitability.
The implications of these results for the assessment of seed limitation are also problematic for another reason. It was only for Dentaria that we recorded a greater long-term survival at occupied than at unoccupied patches, suggesting that occupied patches were on average more suitable for this species. For the remaining five species, the number of successful established populations decreased at a similar or for some species even higher rate for cohorts sown at occupied, and hence by definition suitable, patches. These patterns imply that the selected set of unoccupied patches were on average at least as suitable as the occupied patches. This may mean that recruitment conditions may have changed over time in some patches so that a fraction of formerly suitable sites, now hosting adult individuals, for reasons of disturbance or succession have become unsuitable for recruitment (cf. Augspurger 1983; Cody 1991; Battaglia et al. 2000; Eriksson 2002; Collins & Carson 2004). The apparent similarity between occupied and unoccupied patches may also have been influenced by spatial and temporal heterogeneity. The suitability for recruitment in unoccupied as well as occupied patches may hold only for the actual sowing plots, not for the patch as a whole (Münzbergová 2004). In a similar way, plots may have been suitable for establishment in some but not all years. Hence, a proportion of patches where transplantation failed are likely to be suitable for the focal species in the sense that populations could be established in some places in some years. Failed establishment may also have been the result of demographic stochasticity and higher sowing densities might have resulted in successful recruitment in some additional patches.
Another factor that possibly could affect these results is that the probability of establishment of the four species with fleshy fruits may under natural conditions benefit from the ‘treatment’ of these fruits by vertebrates, whilst we sowed the seeds without any particular treatment. Given that the only observed dispersal vectors of these fruits in this area are rodents (J. Ehrlén, O. Eriksson & H. von Zeipel, unpublished data), in which the seeds do not pass through the gut, this may not be a major problem. It is also true that this and other factors, such as the density of sown seeds, which may affect the incidence and number of established seedlings, did not differ between sowing experiments at unoccupied and occupied patches.
We expected to find a positive and increasingly over time strong correlation between the presence of seedlings and number of seedlings in the sowing experiment, vegetation composition (in terms of Beals index) and environmental factors. We also expected to find a strong correlation between the occurrence of adults, vegetation composition and environmental variables. However, we found generally relatively weak correlations between Beals index and the presence of seedlings in the experimental plots. The relationships with the number of seedlings were slightly stronger and there was a trend of increasing correlation over time for Paris, Lathyrus, Convallaria and Dentaria. The correlation between presence of adults in the plots and Beals index was significant for all species except Paris, but only for Dentaria could the majority of this variation be explained. Environmental variables were correlated with the number of surviving seedlings in five out of the six study species during the first year following sowing. However, by the end of the study, only in Convallaria and Dentaria were they significantly related to seedling number. Similarly, Graae et al. (2004) found that soil variables had little influence on recruitment of herbs in new and ancient forests.
The accuracy of prediction of Beals index and of the environmental variables appears to be related to the size of the particular species’ niche (Table 1), i.e. whether a species is capable of occupying a large fraction of the environmental ‘space’ or not. The presence of adults is explained by the predictable variables in Dentaria and Polygonatum with comparatively narrow niches, while it is only weakly related to these variables in Convallaria with a large niche. The fact that survival was significantly higher at occupied than at unoccupied sites only for Dentaria suggests that it is only in this species that the niche requirements were sufficiently narrow to detect significant differences between occupied and unoccupied patches (cf. Gustafsson et al. 2002). That is, the range of environmental conditions encompassed by the plots may have been too narrow (for example, the pH ranged only from 4.6 to 7.1) to expect clear differences between occupied and unoccupied patches in all species. Another possible reason for the lack of clear differences is that the environmental factors measured do not contain sufficient or relevant information about habitat suitability for these species.
In spite of potential methodological problems, it is important to bear in mind that our results could equally well reflect the true weakness of species composition and prior establishment as predictors of recruitment success. Such weak relationships may be the result of changes over time in the environmental conditions important for recruitment, while the survival of already established individuals is still possible. In addition, correlations between species regarding how they perceive environmental conditions may differ between different phases of their life cycle. If habitat suitability changes over time, then the presence of the target species at the locality, as well as Beals index (calculated from the presence of adult plants) and environmental variables (using the correlation with adult plants to assess suitability) would indicate past habitat suitability, whereas sowing experiments would indicate present habitat suitability. As a result, assessment of habitat suitability using environmental correlations is likely to prove particularly difficult in species with long life spans compared with the rate of change in habitat suitability (Turnbull et al. 2000). A life span of up to 50–100 years is not unlikely for the investigated species (Ehrlén & Lehtilä 2002), and during such a time period our study habitats most likely undergo relatively large changes.
Thus, several lessons have been learnt from this investigation and many aspects need to be considered in designing future experiments to assess the relative importance of seed limitation. First, it is important that experiments are run over such a period of time that they encompass the bottlenecks associated with establishment and growth up to the stage of reproduction. Our results indicate that the time required in forest herbs could be greater than 11 years. Secondly, most available studies demonstrate that variation in probability of seedling establishment between sites and years is important. This underlines the need to carry out spatially and temporally replicated sowing experiments and to include experimental sowings at reference sites with adult plants, as well as non-sown controls to estimate recruitment from natural seed deposits. Thirdly, demographic stochasticity can be important for probability of establishment and thus there is a high risk that establishment fails at suitable sites if sowing densities are too low. Finally, it is important to mimic, as closely as possible, the natural processes by which seeds are released and dispersed.
Our results also highlight some problems associated with the criteria usually used to define unoccupied, suitable habitats and seed limitation. These criteria are more or less fixed in terms of the type of establishment that we regard as evidence of seed limitation, ranging from successful germination, via reproducing individuals to populations with positive growth rates. It is important to recognize that very different quantitative results can be obtained depending on which criteria are used. Given that conditions suitable for recruitment may vary between years and over small spatial scales within patches, habitat suitability might be best described in terms of probability of successful establishment.
In conclusion, our results strongly suggest that the distribution of long-lived forest herbs is limited by seed availability. They also show that bottlenecks at post-seedling stages make short-term assessments poor predictors of recruitment success over longer time frames. Lastly, our results show that environmental variables and vegetation composition sometimes have only a limited capacity to identify unoccupied suitable habitats.