Seed bank formation during early secondary succession in a temperate deciduous forest


and present address: Laura Hyatt, State University of New York, Department of Ecology and Evolution, Stony Brook, NY 11794–5245, USA (tel. 516 632–1381; fax 516 632–7526; e-mail


1 Seed banks are dynamic entities, with input occurring through dispersal and loss occurring through germination and various sources of mortality. Measures of abundance of seeds in the soil at one point in time cannot distinguish those seeds destined to germinate or die in the next growing season from those entering the long-term seed bank, and cannot therefore reveal the proportion of seeds that are carried over from year to year.

2 We studied seed bank dynamics in a 2.5-ha deciduous forest gap by following the fate of a single year’s cohort of seeds over 2 years. Seed input for 1 year, subsequent in situ germination and survivorship were measured at 60 locations. We examined how temporal changes in seed bank composition were affected by the presence of the dominant gap colonizing species Rubus allegheniensis.

3 Over the course of 2 years, both mean seed density and its variance declined. The temporal changes in density were not, however, accompanied by detectable changes in diversity.

4 Three species (R. allegheniensis, Phytolacca americana and Paulownia tomentosa) dominated the seed bank. Seed bank accumulation patterns in Rubus and Phytolacca showed contrasting responses to the presence of Rubus, with Phytolacca excluded from seed banks in such patches through reduced input and increased mortality.

5 First year germination and post-dispersal mortality interact with seed input to influence the spatial distribution of density and diversity in the long-term seed bank. Substantial long-term seed banks do form within temperate forest gaps and patterns of above-ground vegetation can have substantial effects on their dynamics.


In most ecosystems, the soil contains many viable but ungerminated seeds produced either in the most recent reproductive period or in previous years. Several ecological functions have been proposed for seeds surviving multiple years in the seed bank. First, by acting as a propagule reservoir, the seed bank may reduce the probability of population extinctions (Cohen 1966; Venable & Brown 1988). Secondly, because banked seeds contain alleles that may be present at different frequencies to those in the above-ground portion of a population, seed banks may retard the rate of evolution or may change population genetic structure (Levin 1990). Thirdly, by allowing species to specialize on environments that occur infrequently, seed banks may facilitate the coexistence of potentially competing species (Chesson 1986; Pake & Venable 1995). Finally, the seed bank is likely to be the major source of the above-ground plant community following environmental changes such as treefalls, restoration efforts or substantial rainfall events, all of which trigger the germination of buried seeds (Grime 1989; Wilson et al. 1993; Brown 1998; Hyatt 1999). Such functions depend to a large extent on seeds persisting in the soil for more than 1 year.

Between-year survival

To begin to understand the role that seed banks can play in a particular community, it is important to know what proportion of seeds in the seed bank have survived more than 1 year. Most work on seed banks either documents the size and species composition of the seed bank at a single point in time (Leck et al. 1989; Ashton et al. 1998; for example) or the temporal patterns of seed bank density for individual species (Thompson & Grime 1979). Neither of these approaches obtains information about seed survival between years, which would require following individual cohorts of seeds to reveal seed bank age structure. Other work has investigated germination requirements, using this information to infer the species-specific potential to form a seed bank (Baskin & Baskin 1994; Doucet & Cavers 1996). None of these types of approaches considers the dynamic nature of seed banks or environmental influences on seed survival and they cannot, therefore, reveal the extent of between-year seed bank persistence in the field. Two seed banks of similar sizes may have dramatically different dynamics and thus different potential functions: one may be constantly renewed over time with little between-year carryover, while another may be comprised mainly of very old seeds with few new seeds being added and/or those that are added being removed rapidly. The latter type of seed bank is more likely to influence the population dynamics and genetic structure of the above-ground population, but the necessary distinction between the two types of seed bank cannot be made using common methodologies.

Spatial components of seed bank dynamics

Seeds are added annually to the seed bank through dispersal while they are removed through germination of both newly dispersed seeds and those stored in the seed bank, and through senescence, predation and pathogenesis. The balance between these processes will determine the turnover rate of the seed bank in a particular location.

The balance between inputs and removals is likely to vary substantially in space, especially in relation to above-ground vegetation patterns. Additions to the seed bank through seed dispersal are modified both by proximity to parent plants and by disperser behaviour (Schupp & Fuentes 1995), while in situ seed pre-germination and germination loss of seed viability can vary spatially both with abiotic environmental conditions and with changes with the distribution of pathogens and litter (Hamrick & Lee 1987; Cavers et al. 1995). Losses from the seed bank due to predation and seed pathogenesis may also vary spatially (Lonsdale 1993; Verdu & Garcia-Fayos 1996; Hyatt 1998). The historical and spatial pattern of such variations will influence the distribution of older seeds in the seed bank.

Deciduous forests

Deciduous forests are a particularly interesting system for investigating the influence of above-ground vegetation on seed bank dynamics. First, the seed bank provides a substantial number of new plants that emerge after treefalls and understanding the processes that generate seed bank patterns may help predict the future role of the seed bank in such disturbances, both natural and anthropogenic (Wilson et al. 1993; Qi & Scarratt 1998; Hyatt 1999). Secondly, deciduous forest seed banks often contain large numbers of seeds of species that are native to other kinds of habitats, especially to early successional habitats such as old fields (Pickett & McDonnell 1989). It is not clear whether these high densities are due to recent inputs from adjacent habitats or are remnants of earlier successional patterns when such species were locally present, but the two causes imply very different consequences for community structure. Finally, there are several extant databases on seed bank composition and density in deciduous forests (Nakagoshi 1985; Thompson et al. 1997) and understanding the underlying dynamics may help guide our interpretation of these data sets.

The spatial patterns of seed bank density and species composition in deciduous forests appear likely to be established primarily during early secondary succession. Seed banks are mainly composed of early successional species, even under closed-canopy conditions (Pickett & McDonnell 1989). Seed banking species are conventionally thought to be quite locally dispersed because of an apparent evolutionary trade-off between dispersal ability and dormancy (Venable & Brown 1988), although much evidence suggests otherwise (Marks 1974; Thompson et al. 1998). Nevertheless, it is suspected that very few seeds are added to the seed bank after canopy closure, i.e. when early successional species are no longer present (Pickett & McDonnell 1989; Hyatt 1996). The larger seeds of dominant forest species which are produced under closed-canopy conditions rarely persist between years in the seed bank in part because they are thought to be choice foods for granivores (Thompson 1987). Many dominant tree species lack long-term seed dormancy mechanisms and instead form a seedling bank (Harper 1977).

We explored the patterns of seed bank dynamics in an early successional, temperate-zone deciduous forest by determining: (i) the relationship between initial seed input and subsequent germination and mortality, (ii) how the removal of seeds by germination and mortality affects seed density and species diversity, and (iii) how seed bank dynamics are influenced by the presence of above-ground patches of early colonizers, especially Rubus allegheniensis.


Study site

This study was conducted at the Laurels, a preserve managed by the Brandywine Conservancy in the Piedmont Uplands of south-eastern Pennsylvania, USA, that consist of a mosaic of hayfields and old growth forest dominated by beech (Fagus grandifolia Ehrh.) and several species of oak (Quercus rubra L., Q. alba L. Q. velutina Lam. and Q. coccinea Muenchh.). Acer rubrum L., Liriodendron tulipifera L., and Nyssa sylvatica Marshall. also occur in the canopy. Common understorey species include Arisaema triphyllum (L.) Schott, Aster divaricatus L., Carex pensylvania Lam., Podophyllum peltatum L. and Viburnum acerifolium L.

In the fall of 1989, a major downdraft snapped off or tipped over 50–55% of the canopy trees in a 2.5-ha portion of a 6-ha forest stand (B.B. Casper, unpublished data), the area subjected to this blowdown is shown in Fig. 1. Although shrubs such as Lindera benzoin (L.) Blume and Viburnum acerifolium and saplings of some canopy trees survived the storm, the vegetation in the blowdown quickly became dominated by the semi-woody shrub Rubus allegheniensis T.C. Porter and the herbaceous perennial Phytolacca americana L. These remained the most common species throughout this 2-year study, which began in May 1992.

Figure 1.

Aerial photograph of study site, indicating extent of forest (bordered by black line), extent of blowdown (region south of black and white line within forest), transect locations within the blowdown (white lines), and adjacent hayfields.

For this and other studies that were being conducted in the area, five parallel transects, 20 m apart, were created in the blowdown (extending into the adjacent intact forest, not shown in Fig. 1) by cutting a swath of vegetation just wide enough to allow passage through the tangled blowdown vegetation. The sections of transects used differed in length depending on the distance between the intact forest and the edge of the hayfield to the south-west, and ranged from 100 to 300 m (Fig. 1). Transect 1 was farthest north-west and was overall the closest to the border with the remaining forest, whilst transect 5 was closest to the surrounding hayfields along its length.

Sampling method

To examine the fate of 1 year’s seed rain at a particular location, we devised a sampling method that captured only that year’s net seed input, allowed in situ seedling germination during the next growing season, and enabled us to estimate the proportion of the remaining seeds that survived until the second growing season. This approach made it possible to examine changes in seed bank composition and density over time for a single cohort of seeds and to determine the relative contribution of in situ germination and seed mortality to these changes.

Sampling units consisted of solid-sided glasshouse flats with gridded bottoms (50 × 25 × 5 cm plastic trays with 1 cm2 holes; 1020 heavyweight flats, ITML Co., Waco, TX). At each location, sampling units (flats) were embedded in the surrounding soil so that their tops were level with the soil surface. These flats were filled to the rim with a seed-free soilless medium (Metromix 500, Grace-Sierra Horticultural Products Co., Milpitas, CA). Because the flats were open on the top and had wide holes on the bottom, there was free vertical water, seed and insect movement. The medium in the flats was also available for colonization by roots outside the flats, fungal hyphae and bacteria. These units were placed in the field on 17 and 18 May 1992 and visited on a monthly basis. Eleven months later on 20 April 1993, before any seedling emergence had occurred (as confirmed by our monthly observations), all soil and accumulated litter was collected from a randomly selected half (25 × 25 × 5 cm volume) of each flat and taken into the glasshouse. The empty space created in each flat was filled with sterilized, seed-free sand to provide structural support for the remaining Metromix.

Total seed bank density and composition, hereafter referred to as seed bank input, were estimated from germination under glasshouse conditions of seeds from these samples. Such seedlings represent seeds dispersed to the flats during the first growing season minus those that lost viability, were consumed by predators or pathogens, or were deeply dormant and failed to germinate. This approach was the best strategy available to estimate seed input density because all seeds were collected in spring and had been exposed to at least some overwinter conditioning.

Within 48 h of collection, each sample was individually sieved to break up large clumps and spread on top of a layer of seed-free Metromix in a standard plastic glasshouse flat with small holes (< 2 mm diameter) in the bottom for drainage. Because the soil medium used was organic, it was not possible to use traditional salt flotation techniques to extract seeds. Seedlings were tallied and removed as they emerged. At monthly intervals, the soil was stirred to bring deeply buried seeds to the surface. Any seedlings that were difficult to identify were transplanted to separate pots until they were larger. This germination assay was continued until November 1993. To correct seedling counts for any contamination by glasshouse weeds, flats of seed-free sterilized Metromix were placed among the samples. Arabadopsis thaliana was the only species to appear in these control flats and so when it appeared in flats containing field soil (fewer than 15 seedlings) it was not included in the analysis.

Germination in the field from the half flat that was left intact was also measured. Emerging seedlings were counted on a biweekly basis throughout the 1993 growing season. We assumed that seed composition in both halves of the flat were similar, but that fewer seedlings would emerge in the field than in the glasshouse where water and light were plentiful and soil was manipulated in such a way as to maximize seed exposure. We left flats in the field undisturbed because we wished to measure germination as a result of dispersal and local conditions without changing the depth of seed burial. Emerging seedlings were tallied and marked, but not removed. Emerging seedlings represent loss form the seed bank through in situ field germination.

Once the summer germination season was over (on 20 July 1993) the Metromix remaining in the flats was covered with a 2-cm layer of sterilized sand to reduce the likelihood that new seeds would be incorporated into it, yet keeping it in relatively ambient moisture and temperature conditions. Only one winter perennial species (Claytonia virginica) might have released seeds between April and July, but seedlings of this species were never observed to emerge in the glasshouse. All field-emergent seedlings were completely covered with the sand and killed. Nine months later (April 1994), the sand was carefully scraped off and the Metromix was transported to the glasshouse, where it was sieved and spread out on flats as described above. Seedlings emerging from this sample represent what we will refer to as the surviving seed bank; that is, our estimate of seeds that had been dispersed between May 1992 and July 1993, did not germinate in the field in 1993, but survived over the winters of 1992–93 and 1993–94. Like the seed input estimate, this approach does not sample seeds that remained deeply dormant for a second year. The second germination trial ended in late October 1994.

We intended to compare the seed input estimates and the surviving seed estimates, subtracting loss to in situ germination in order to determine indirectly the site-specific rates at which seeds were removed from the seed bank through mortality. However, these differences do not solely arise from seed mortality. Because the two estimates were derived from different halves of a flat and the initial distribution of seeds was not likely to be uniform across the flat, error in seed mortality estimates could have arisen from sample independence. In addition, because the presence of seeds was determined by glasshouse germination, this technique failed to estimate accurately the numbers of viable seeds that did not germinate, i.e. those that were deeply dormant. Thus, it is likely that estimates of both input and survival rates are conservative.

Sampling design

We established 60 sample units in the field but could only retrieve 59. The placement of these units enabled us to examine how seed input, field germination and survival varied spatially and interacted to determine seed bank density. Sample units were placed along transects, and differences among transects in the measured parameters reflect large-scale spatial variation in seed gains and losses. Transects differed not only in their proximity to old-growth forest and hayfield, but also in the extent to which the wind storm disturbed the vegetation. From survey data, transect 1 appeared to be more intensely disturbed than transect 5 because residual, tall-standing vegetation was far more dense on transect 5 than transect 1.

To examine how early colonizing species influence subsequent seed bank dynamics, we selected the common Rubus allegheniensis as a focal species and hypothesized that it would influence seed dynamics through its effect on seed input, field germination, and seed predation by granivores. We suspected that seed dispersal by birds would be denser in Rubus patches because it provides perches and draws frugivorous birds by its own fruit production, thus increasing both passive dispersal of Rubus and active dispersal of other bird-dispersed species. On the other hand, we also anticipated that seed loss to predation might be greater in Rubus patches due to preferential foraging by vertebrate granivores protected from their predators by its thorny growth form. Finally, we suspected that seed bank losses due to field germination would be lower in Rubus patches because leaves appear early in spring, thus casting heavy shade throughout the germination season.

To investigate the influences of transects and Rubus on seed bank dynamics, sample units were distributed roughly evenly among transects (8–11 units per transect) with approximately half the retrieved units inside patches of Rubus (n = 31) and half outside (n = 28). When possible, Rubus and non-Rubus sites alternated along each transect and were spaced approximately 10 m apart. This placement of sampling units made it possible to generate a general spatial map of seed banks and germination in the field.


Changes in the distribution of seeds were first analysed for all species together. However, because 80% of the seedlings observed were either Rubus allegheniensis, Phytolacca americana or Paulownia tomentosa, these species were also analysed individually. Although both R. allegheniensis and P. americana were abundant in the post-blowdown vegetation, P. tomentosa, known as the Empress tree, was only represented by two individuals. Individual trees produce millions of seeds each year, and although the trees were not numerically dominant in above-ground vegetation, their seeds made up a large portion of the seed bank.

Spatial distributions of seed density

To visualize spatial patterns of seed input, germination and survival, we constructed contour plots of seed or seedling densities using the splining graphics capabilities of Statistica/W (StatSoft 1994). The general influence of spatial variation in seed bank input on spatial variation in surviving seed bank density was examined with regression (StatSoft 1994). The log(numbers of seeds surviving) was regressed onto log(number of seeds initially added to the seed bank). The proportion of variation in survival explained by input alone (r) was examined for all seeds together as well as for each subgroup.

Vegetation effects on seed density

To examine the influence of above-ground vegetation patterns on seed density or field germination, we performed two-way analyses of variance. Independent factors were site type (Rubus/non-Rubus, fixed effect) and transect (1–5, random). Dependent variables (density of seeds input, seedlings germinating in situ, or seeds surviving by group) were rank-transformed to reduce violations of anova assumptions.

Germination in situ and survival are likely to covary with seed input; the more seeds that are initially added, the more seeds that are available for germinating or surviving. We used two approaches to examine the interactions between above-ground vegetation and proportional germination or survival. If the slopes of regressions of the log(germinants) or log(surviving seeds) on log(seed input) were homogeneous among groups, as indicated by non-significant F-values associated with interactions between independent variables and the covariate, we used ancova techniques to remove the effects of seed input on the response variable and analysed the effects of site types and transects on the resulting adjusted means. If these slopes were not homogeneous, significant interactions would indicate a violation of ancova assumptions. In the one case in which this occurred (Rubus seed survivorship), the relationship between the significant interacting variables was examined graphically and compared for differences in slopes.

Vegetation effects on diversity

To investigate how the process of seed removal through germination and mortality in the field led to changes in the species composition of the seed bank, we took three complementary approaches. Our first approach was to construct relative abundance curves, plotting ranked species against their proportional representation, and comparing the curves representing seed input and seed survival. Relative abundance curves readily indicate the evenness with which seeds are distributed among species; the flatter the curves, the more even the distribution (and less dominated a sample might be by individual species). Pairs of relative abundance curves representing input and survival were generated for all plots together, and then for groups of plots described by site types (Rubus/non-Rubus) and for each transect separately (transects 1–5).

While relative abundance curves allow examination of the evenness of species distributions, they are based on samples of different sizes. To examine variation in species diversity among patch types and transects while controlling for sample size differences, we used an Expected Species estimation method (ES) developed by Smith & Grassle (1977) and discussed by Heck et al. (1975), Hurlbert (1971) and Sanders (1968). The ES method directly calculates the number of species expected to be encountered in samples of a range of sizes, given the background distribution of individuals among species specified by the field data. Thus, the ES generates a family of diversity measures that has been shown to reduce the sample-size bias inherent in most measures of diversity (Smith & Grassle 1977). We calculated ES and 95% confidence intervals (Heck et al. 1975) for seed bank input and surviving seeds for all sites together, as well as for site types and transects using hypothetical seed banks ranging in size from 10 to 150 seeds at intervals of 10 seeds. We favoured this approach because our samples varied widely in size (0–1744 seeds m−2) and were dominated by a few species.

Finally, because we wished to know how and whether species composition changed after seeds were initially added, our third approach compared initial and surviving species seed distributions. Normalized Expected Species Shared (NESS) (Grassle & Smith 1976), like ES, is an index that can be calculated for a range of sample sizes, yielding a family of indices with confidence intervals. NESS makes it possible to compare species composition among samples while reducing the bias contributed to standard similarity measures by differences in sample sizes and dominance. NESS ranges from 0 to 1 and can be expressed as a percentage. We computed NESS for all sites together as well as for each site type and transect.

The normalization procedure uses division and combinatorial mathematics that limit the population size for which NESS may be calculated (dividing by zero and finding the factorial of a negative number are not possible). Thus, at maximum, NESS is only computable for population sizes less than half of the number of seeds in the smallest actual sample (or groups of samples) to be compared. To facilitate comparisons across transects and site types, we calculated NESS for the largest population size possible for all of our categories (site types or transects), which was 50. This size is not optimal for reducing the bias of species dominance, but it does normalize the effects of differences in sample sizes. Confidence intervals for NESS were generated using the method of Smith et al. (1979).


The seedlings observed throughout this experiment represented 39 species and comprised a wide variety of life forms (trees, shrubs, vines, herbaceous annuals and perennials), fruit types, seed sizes, and dispersal modes (Appendix 1, in the Journal of Ecology archive on the World Wide Web; see the cover of a recent issue of the journal for the WWW address). Fifty-nine seedlings (2% of total) died before they could be identified to genus, but were categorized as monocots, non-composite dicots, or composites. The unknown non-composite dicot category is likely to have encompassed multiple species, but the monocot group and the composite group were represented by multiple identical individuals, and were both likely to have been single species.

Overall patterns

Mean seed input density was 340.61 seeds m−2 (± 76.10, 95% confidence interval) whereas the mean number of seeds remaining in the soil after 2 years was 222.91 seeds m−2 (± 37.30), or 65% of the input density. On average, germination accounted for 79% of the 35% decline in seed numbers; the remaining 21% represent losses to predators, pathogens, or other sources of mortality, as well as to sampling error. Not only did average seed density decline over time, but the variation in density declined as well (coefficient of variance = 85% in seed input density; 64% in surviving seed density).

The spatial distribution of these seeds changed over time (Fig. 2a vs. Fig. 2c), although the percentage of seeds that survived did not vary systematically by site type (F1,58 = 1.96, P > 0.1) or transect (F4,4 = 0.66, P > 0.1). The largest seed input peak (Fig. 2a) was mainly composed of Paulownia. Virtually all the seed loss in this species was due to seed mortality, with only two germinating seedlings observed in the field, and fewer than 30% of seeds surviving over time. Areas of high seed input density had been replaced two years later by peaks in other areas(Fig. 2a vs. Fig. 2c), although the large input peak persisted, mainly composed of non-Paulownia species.

Figure 2.

Contour plots for all species: (a) spatial distribution of seeds added to the seed bank; (b) spatial distribution of field germinants; and (c) spatial distribution of seeds remaining in the seed bank. The height and darkness of the peaks indicate areas of greater seed density. The y-axis shows distance along the transects. Note that the z-axis indicates relative seed density and that the z-scale differs for all three plots: the scale of (a) is twice that of (b) and (c). Points between data sampling plots are interpolated.

Only a very small, insignificant portion of the overall variation in density of the total number of surviving seeds is explained by seed input as shown by the very low correlation coefficient (r2 = 0.03, P = 0.18). Thus, local patterns of seed mortality or germination are likely to be more important in determining seed bank density than is initial seed input. Seed bank input explained slightly more variation in final density for the dominant species (Phytolacca, r2 = 0.05, P = 0.07; Rubus, r2 = 0.20, P < 0.01; Paulownia, r2 = 0.20, P < 0.01).

Changes in seed density over time were not accompanied by detectable changes in diversity. Relative abundance curves show that the distribution of seeds among species was only slightly more equitable in the surviving seed bank than in seeds initially added to the seed bank (Fig. 3). We speculate that this might be due to reduced dominance of the three main species. This speculation is only marginally supported by the ES analysis (Fig. 3 inset) that shows the expected number of species encountered in seed banks of standardized sizes was somewhat greater in surviving seed banks than in the initial inputs to the seed bank, although the confidence intervals for these two curves overlap throughout their length. For visual clarity, we have only plotted confidence intervals for the largest seed number (n = 50) for which ES was calculated. The Normalized Expected Species Shared (NESS) between seed bank input and survivors for a seed bank of normalized size (n = 50) was 85 ± 83.3%, a range clearly no different from 100%. This suggests that there were no detectable changes in species makeup as seeds were lost to germination or mortality over time.

Figure 3.

Relative abundance and rarefaction curves for all species and plots together. Proportion of seeds attributable to each species in all seed banks pooled; ● = Input, ○ = Surviving. The x-axis indicates relative species rank. Plots for each subtype are offset from each other for visual clarity. Inset figure shows number of species expected plotted against the number of seeds with data generated by rarefaction. Solid lines indicate input, dashed lines indicate surviving, bars at end show 95% confidence intervals for curves at n = 150 only.

Patch effects

Although Rubus patches appeared to have no influence on overall seed bank dynamics or on Paulownia seed densities, they did have strong influences on seed bank dynamics of both Rubus and Phytolacca (Table 1). Not surprisingly, Rubus seeds were added in significantly greater numbers to Rubus plots than to non-Rubus plots (F1,49 = 8.83, P < 0.01), but despite these differences Rubus seedlings did not germinate any more frequently within Rubus patches (F1,49 = 0.03, P > 0.1). Nearly three times as many Rubus seeds were found in the surviving seed bank as initial germination suggested were present in the input. Because these increases that were presumably due to release of seeds from deep dormancy were proportional to the original numbers of seeds, the seed bank in Rubus patches continued to be larger than in non-Rubus patches (Table 1). Losses from the seed bank due to germination were in the same proportion in the two patch types (F1,58 = 0.04, P > 0.1), so we may assume that this relationship will continue to hold, although our ability to estimate the exact number of seeds may be limited.

Table 1.  Mean seedling and species number contrasted by patch type. Mean number of seedlings m−2 and standard deviation (in parentheses) reported. Cell entries in bold indicate statistically significant differences between patch types in seedling density at that stage as determined by two-way ANOVA. See Analysis section for description of methods
SpeciesPatch typeInputGermination in situSurviving
Mean seedlingsRubus293.16 (189.15)96.00(106.45)248.77 (163.58)
Non-Rubus393.14 (371.20)90.86 (83.20)194.28 (112.19)
Paulownia tomentosaRubus133.68 (184.96)0.00 (0.00)50.58 (57.85)
Non-Rubus221.71 (291.07)1.71 (9.07)54.86 (67.80)
Phytolacca americanaRubus58.84 (54.87)65.03 (79.78)38.71 (48.14)
Non-Rubus109.14 (135.21)45.71 (60.92)62.86 (53.50)
Rubits allegheniensisRubus46.45 (53.92)11.35 (29.85)128.0 (137.64)
Non-Rubus15.43 (21.98)6.86 (14.72)37.71 (56.65)
Mean nondominantsRubus54.19 (50.88)19.61 (34.98)31.48 (28.76)
Non-Rubus46.85 (38.93)36.57 (41.27)38.85 (38.06)
Mean number of speciesRubus4.87 (1.58)1.58 (1.31)4.16 (1.31)
Non-Rubus4.28 (1.86)2.21 (1.31)4.16 (1.39)
Total number of speciesRubus221021

Phytolacca showed opposite patterns (Table 1). Fewer Phytolacca seeds were initially added to sample units in Rubus patches than to non-Rubus patches, and of these seeds fewer of them survived, both numerically (F1,49 = 4.41, P < 0.05) and proportionally (F1,58 = 2.76, P = 0.10), within Rubus patches than outside (Table 1). We are far more confident of the seed numbers estimated for Phytolacca at each interval, as apart from a small amount of initial physiological dormancy, this species is not known to produce seeds that are technically dormant between years (Baskin & Baskin 1998). Numerically, the conflicting seed bank dynamics of these two dominant species are likely to have cancelled each other out, leading to a lack of patch-type effect on seed survival when all species of seeds were considered together (F1,49 = 0.86, P > 0.1, Table 1).

Whereas the average species composition of the seed bank in Rubus patches was virtually unchanged over time, both in terms of the relative abundance and expected numbers of species (Fig. 4b), non-Rubus seed banks showed slight changes in species composition (Fig. 4a). Relative abundance curves showed a greater equitability of species abundances in non-Rubus patches and a somewhat greater number of species expected inside these patches (not significant, Fig. 4a, inset). Although the ES values for non-Rubus patches appeared to be slightly different, the NESS analysis showed no detectable changes in species distributions in either patch type (non-Rubus patches: 81.1 ± 53.3%; Rubus patches: 81.5 ± 52.4%; both confidence intervals overlap 100%).

Figure 4.

Relative abundance and rarefaction curves for patch types: (a) non-Rubus plots, and (b) Rubus plots. See Fig. 3 legend for explanations.

Transect effects

Transects did not influence seed input density in a statistically significant way (F4,4 = 1.64, P < 0.1), but they marginally influenced both total numbers and proportions of seeds germinated in situ (F4,4 = 4.61, P = 0.08); seed germination increased from transect 1 to 5, an effect echoed by the three dominant species analysed separately. There was no association of transects with the number or proportion of total seeds surviving in the seed bank.

Relative abundance changes were minor on all transects with a slight increase in equability on transects 3, 4, and 5, but slight decreases on transects 1 and 2 (data not shown). ES analysis revealed no significant changes in the expected number of species for any transect. The NESS analysis, however, showed that transect 2 had a significant change in species composition (Table 2), resulting from a shift in the identity of the dominant species from Rubus to Phytolacca on that transect.

Table 2.  Normalized Expected Species Shared (NESS): normalized percentage (± 95% confidence interval) of species expected to be shared between random samples of size 50 from the initial seed bank and the surviving seed bank for each transect
TransectNESS50 (%)
179.3 (± 29.9)
266.8 (± 23.1)
386.0 (± 41.3)
475.0 (± 25.6)
583.5 (± 35.0)


A large proportion of dispersed seeds remained in the seed bank in this system for at least a year. Losses to germination and seed mortality were, on average, relatively low. Thus, a substantial part of the seed bank is likely to consist of seeds dispersed in previous growing seasons. How long seeds are able to survive in the seed bank, once incorporated, will determine the age structure of the seed bank in the future and its potential contribution to changes in community structure, population dynamics and genetics.

In this early successional forest system, post-dispersal seed loss, although low, altered the initial spatial patterns of seed bank input. Species-specific patterns of seed mortality and in situ germination reduced the dominance by the common species, slightly increasing overall seed bank diversity. At the same time, these loss-related processes began to reduce site-to-site variation in overall seed bank density. Seed input, post-dispersal seed mortality and germination all played a role in determining the resulting seed bank density and composition.

How might we integrate this information about the balance of gains and losses to the seed bank with what we know about seed bank size and the dominant species’ germination behaviour and requirements? Paulownia has a large seed bank and an individual can produce millions of seeds each year. These seeds require high light conditions and a brief cold stratification period for germination (Carpenter & Smith 1981), and are only thought to remain viable for 2 years at most (Young & Young 1992). In this study, only two seeds germinated in situ, and a very low percentage, fewer than 30% of the dispersed seeds, survived (Table 1). The large Paulownia seed bank is characterized by very high turnover, with little between-year build-up. Although Paulownia has a large seed bank, seed banking is probably not a major component of its reproductive strategy, although the window of opportunity for successful recruitment is apparently open for more than 1 year after dispersal.

Phytolacca seed bank dynamics are strikingly different. In this herbaceous perennial species, germination requires a short cold stratification period and is stronger under light than dark conditions. If presented with appropriate environmental conditions, germination can occur throughout the year (Baskin & Baskin 1998). Despite liberal germination requirements, a substantial Phytolacca seed bank can build up; in this particular case, 46% of dispersed seeds survived in the soil for at least a year. Under changed environmental conditions, a great many of these seeds may respond by germinating, although this may vary with seed age (Hyatt 1999). Phytolacca clearly employs a seed bank strategy which enables seeds to rapidly respond to favourable environmental changes.

Data presented here suggest that Rubus employs a different type of seed banking strategy. Rubus seeds require a long cold stratification period for germination, are not dependent on light conditions, and can retain viability under laboratory conditions for at least 25 years (Brinkman 1974; Jobidon 1993). Further, because seed count estimates increased over time (Table 1), clearly some dispersed seeds are incapable of germinating the first year after dispersal. It is likely that some of these seeds are incapable of germinating for several years after dispersal. If these seeds escape predation and pathogenesis, their entry into a between-year seed bank is guaranteed. It is not clear whether this seed banking behaviour in Rubus evolved as a bet-hedging strategy against somewhat predictably fluctuating environmental conditions in temperate deciduous forests, but it clearly contrasts with the strategy adopted by Phytolacca, which maintains a bank in which seed are nearly always germinable. More long-term data sets concerning the interaction between seed ageing and germination requirements are needed to explore the function of this seed banking strategy further. In addition, further examination of the evolutionary history and physiological mechanisms of the functional seed bank strategies of early successional species would be profitable.

The case of Rubus illustrates the difficulties of making accurate estimates of the relative probabilities of alternative seed fates. Unfortunately, the profound dormancy of Rubus seeds compromised our ability to discern the precise numerical balance of gains to and losses from the seed bank and changes in species diversity. Our estimates of seed retention in this species were therefore highly conservative, as would be those for any other species in this study with innately dormant seeds.

Vegetation patch effects

Through their initial effects on seed bank dynamics, Rubus patches may have observable effects on succession that become apparent after disturbance events, possibly decades later. Phytolacca seed survival was negatively affected in patches occupied by Rubus through both inhibited seed input and increased seed mortality. Conspecific Rubus seeds, on the other hand, were maintained in high densities. Rubus also reduced the germination of rarer species, although to a lesser degree (Hyatt 1996). By accumulating high densities of conspecific seeds and excluding the seeds of other species, seed bank dynamics in Rubus patches may facilitate future site pre-emption after disturbance (Shmida & Ellner 1984; Bergelson 1990). The effects of initial colonizers on subsequent successional processes have already been explored through their facilitative effects on seedlings (Li & Wilson 1998; Werner & Harbeck 1982; Callaway 1995) and their effects on dispersal agent behaviour (Holl 1998; McDonnell 1986; Debussche & Isenmann 1994). In large forest disturbances, seed availability of early colonizers like Rubus is suspected to influence successional trajectories and patch dynamics in a significant way (Peterson & Carson 1996).

In this system, the influence of above-ground vegetation on seed bank dynamics was not equally strong for all species. For example, surviving seed densities of Paulownia were mainly determined by the spatial distribution of seed input and were hardly affected by above-ground vegetation. Other species may have been similarly unaffected by patterns of above-ground vegetation, but most were at such low densities at all stages that any effect would not be detectable. While patterns of germination and seed banking behaviour may be explained by patterns of dormancy and viability in some species (Baskin & Baskin 1998), this is not true for all species. We wish to stress that for two of the dominant species in this system, Phytolacca and Rubus, both seed input and seed–environment interactions played strong roles in determining seed bank densities.


The approach described here for studying propagule dynamics of whole plant communities is novel and needs to be applied to other temperate-zone deciduous forests as well as to other communities and taxa. Many species of plants and animals that inhabit temporally variable environments can persist in unfavourable environments in a dormant stage (e.g. eggs and spores), and the fates of these ‘sleeping’ propagules may have marked effects on populations and communities. For example, dynamic effects are created in copepod communities by pond-specific differences in emergence patterns (DeStasio 1989; Hairston et al. 1995; (Càceres 1997; Hairston et al. 1997) and it is likely that spatial patterns of propagule survivorship influence the population dynamics of other diapausing invertebrates. Breaking down seed and egg banks into their budgetary components can reveal the extent to which between-year survival occurs for different species in these communities and will enable us to pursue tests of theories concerning its consequences.

We note that other workers have taken similar approaches to understanding the seed bank dynamics of single species (Garcia-Fayos et al. 1995; Cavers et al. 1995), but dynamic studies following cohorts through time can be far more broadly applied. While seed loss through mortality can be estimated by the approach taken here, the causes of mortality are likely to include factors we have yet to understand, such as fungal and bacterial pathogenesis (Augspurger 1988), soil–water–seed interactions (Chambers & MacMahon 1994), and burial depth (Blackshaw 1992). These causal factors should also be the focus of future research.


We wish to acknowledge the generosity of the Brandywine Conservancy for allowing us to conduct research on their property. Their ongoing cooperation with scientific researchers has been immeasurably valuable. The staff of the glasshouse at the University of Pennsylvania, especially T. Byford, helped census and maintain the seed bank assays. M. Leck provided suggestions as to the directions analyses should take, and P. Petraitis introduced us to rarefaction analysis. J.C. Cahill and T. Howard and two anonymous referees helped to make improvements on an earlier draft of the manuscript.

Received 27 May 1999revisionaccepted 4 April 2000