Role of soil seed banks and newly dispersed seeds in population dynamics of the annual sunflower, Helianthus annuus


Helen M. Alexander, Haworth Hall, University of Kansas, 1200 Sunnyside Avenue, Lawrence, KS, USA 66045–7534 (fax 785 864 5321; e-mail


  • 1We explored the role of the seed bank in population dynamics of the summer annual Helianthus annuus. First, we determined seed survival under field conditions. Secondly, we conducted an experiment in which a dispersal treatment (plants allowed/not allowed to disperse seeds) was crossed with a soil disturbance treatment that was predicted to increase seed bank recruitment. Our goal was to evaluate the relative importance of the previous year's seed production vs. the remainder of the seed bank to numbers of plants.
  • 2Yearly seed survival was variable, ranging from 46 to 83% when seeds were buried in mesh packets and from 12 to 76% for seeds placed on the soil surface. Survival was higher for plots established in 1999 than in 2000. Survival was often higher in later years but was unaffected by the presence of litter.
  • 3By comparing seedling establishment between dispersal treatments, we inferred that approximately 10–23% of the seedlings were from the seed bank, independent of soil disturbance.
  • 4Although seed dispersal the previous year led to increased numbers of seedlings by at least 4.5 times, the number of adults only increased 2.5 times and head production only increased 1.5 times because of density-dependent processes.
  • 5Patches of seedlings emerging only from the seed bank (often at low densities) may have a disproportionate contribution to the next generation. Average head production/seedling was 3.6 for such seedlings vs. 0.8 for seedlings from both the previous year's seed set and the seed bank. Emergence of seeds from the seed bank in high-density seedling areas may, however, have little effect on patch reproduction as reproductive output was asymptotically related to the number of seedlings initially present. Studies of seed survival and seedling biology should therefore be done with consideration of the entire life cycle.


Dormant propagules or life stages are a widespread phenomenon in nature. For example, plants often have dormant seeds that persist in the soil (Leck et al. 1989; Rees 1997; Baskin & Baskin 1998), fungi can have dormant spores (Tommerup 1985; Gemma & Koske 1988), and diapause is a feature of fresh water crustacea (Hairston & Caceres 1996) and insects (Powell 2001). The existence of such dormant stages poses a challenge for population biologists, who often use models to predict the number of individuals in a subsequent year based on the number in the current year (i.e. Nt+1 = f (Nt)). If at least some propagules can remain dormant for more than 1 year, the numbers of individuals in a population may be uncoupled from the population size in the previous year (Crawley 1990).

There has been considerable theoretical work on the evolution of dormancy (reviewed by Olivieri 2001) and on the importance of seed banks in population dynamics (e.g. MacDonald & Watkinson 1981; Pacala 1986; Rees & Long 1992; Rees 1997; Maron & Gardner 2000). However, relatively few empirical studies explicitly focus on the role of seed banks at the population level, and specifically on the relative contribution to population size of recently produced seeds vs. those buried in the soil. These studies have used several approaches. First, Naylor (1972) used a creative mark-recapture technique to estimate the relative contribution of recently produced seeds vs. buried seeds to numbers of seedlings. In this method, he sowed seeds that were marked with a fluorescent paint and then compared the frequency of seeds marked with paint with the frequency of germinating seedlings with marks. Such an approach is, however, dependent both on the dispersal unit having structural characteristics that allow marking and the recovery of seedlings that are still attached to the remains of the marked seed. A second approach is to do detailed demographic studies of both buried seeds and seedlings (Sarukhan 1974) to infer the importance of the seed bank. Researchers have also noted population sizes in sites where seed production did not occur the previous season (due to extreme abiotic or biotic factors) and compared such sizes with sites where seed production did occur. Such ‘natural experiments’ provide information on natural germination from seed banks (Baskin & Baskin 1975, 1980), but the lack of experimental controls and replication make quantitative interpretation difficult. Thirdly, one can collect seed survival data from experimentally created seed banks (Kalisz 1991). Such survival estimates can be integrated with population models. By simulating conditions with and without a seed bank, and using sensitivity analyses, researchers have explored seed bank dynamics and the role of the seed bank in population persistence and dynamics (Kalisz & McPeek 1992, 1993; Jordan et al. 1995). Fourthly, field experiments can be performed with naturally established populations by creating exclosures to prevent seed dispersal (Peart 1989), eliminating seed production (Putwain et al. 1968) or eliminating seed bank contributions by soil sterilization (Bullock et al. 1994; Edwards & Crawley 1999a; Rogers & Hartnett 2001). These approaches allow explicit examination of the contribution to seedling recruitment of naturally established seed banks vs. seed inputs from the previous generations.

Our long-term goal is to use the last two approaches to examine the contribution of older seeds in the seed bank to the population ecology of Helianthus annuus L., the wild annual sunflower. Annual species persist in native vegetation, as well as in human-created disturbances, by successful dispersal through space (i.e. colonization of disturbances) and time (i.e. persisting as seeds at sites until a disturbance occurs). To explore the significance of dormant seeds in the population biology of H. annuus, we posed two questions. First, to collect data for future modelling exercises, we asked how long seeds persist in the soil and how burial conditions affect survival. We sowed seeds under different field conditions to evaluate seed persistence over several years. A study of seed survival was important because literature estimates of wild sunflower seed survival and germination are variable (Toole & Brown 1946; Burnside et al. 1981; Burnside et al. 1996; Teo-Sherrill 1996). We explored variation at a single site due to different seed burial methods and different years of plot establishment. Secondly, we explored the relative importance for seedling recruitment and subsequent reproduction of the previous year's seed production vs. older seeds in the seed bank. We conducted a field experiment to examine local plant recruitment and reproduction at sites where seed dispersal either had or had not occurred in the previous year. We crossed this seed dispersal manipulation with a soil disturbance treatment, as disturbance is often associated with germination of buried seeds (Chambers 1993; Chambers & MacMahon 1994; Baskin & Baskin 1998). We hypothesized that soil disturbance would increase seedling establishment and recruitment of seedlings from the seed bank.


study species

Wild sunflower, H. annuus, is a native North American annual that occurs in a range of habitats that receive frequent disturbance, including roadsides, open areas in prairies, crops, and crop margins (Heiser 1954). Individual plants can vary greatly in size and number of capitula (hereafter referred to as inflorescences or heads). Mature cypselas (hereafter referred to as seeds) of wild sunflower are large (7 mg; Alexander et al. 2001) and are eaten by a variety of animals including small mammals and quail (Robel et al. 1974; Teo-Sherrill 1996). Teo-Sherrill's (1996) work in Nebraska, USA, suggests a cyclical pattern to seed dormancy, with seeds not dormant from March to June (germinating if conditions are appropriate), but dormant in other months.

fate of seeds in the soil

We determined the fate of seeds over a 4-year period in two field situations, buried seed packets and on the soil surface. We predicted seed survival would be higher in the buried packets compared with the soil surface as Mohler's (1993) literature review reveals that for most species, seed survival increases with depth. This experiment was performed at the Nelson Environmental Study Area of the Kansas Field Station and Ecological Reserves (henceforth the KSR site), in Jefferson County, Kansas, USA. We set out replicate lots of seeds and destructively sampled one replicate each year to determine seed persistence. The experiment was also replicated in time, with initial establishment of plots in either November 1999 or December 2000 in two areas separated by 5 m. The seeds came from bulk collections made in Lawrence, Kansas, in either fall 1999 or 2000 and fates were followed for 4 years (2000–03) or 3 years (2001–03) for the 1999 and 2000 studies, respectively. The plots (15 plots per year, each 4.0 × 1.5 m, separated from each other by 4 m) were initially established by rototilling in an early successional field.

We buried fibreglass mesh packets (11.5 × 10 cm), each containing 140 seeds, 25 cm below the surface of the soil in the centre of each plot. Replicate packets were separated from each other by 50 cm. One hundred and forty seeds is a typical value of seed production per head (estimates of numbers of seeds per head range from 107 (Cummings & Alexander 2002) to 176 (this study)) and this is a number of seeds that could be deposited on a small patch of soil. We removed one randomly chosen packet from each plot in each subsequent late spring/early summer (after emergence of seedlings had stopped in the field) and intact seeds were removed and counted. To confirm that intact seeds were viable, we performed tetrazolium viability tests by cutting the recovered seeds laterally and placing them in a 0.2-mL tetrazolium solution in a cool, dark location for approximately 4 h. Red coloration of the embryo indicated the presence of respiring tissue, and thus a viable seed. As virtually 100% of the seeds visually judged to be viable also stained red, in 2002 we did tetrazolium staining for 70% of the samples and sorted the rest visually.

We also established replicate subplots on the surface of each plot. One hundred and forty seeds were placed in an open-topped fibreglass mesh basket that had been filled with field soil and buried so that the soil surface was even with the surrounding area. Baskets had surface dimensions of 20 × 20 cm, a depth of 13.75 cm, and had vertical sides rising 7 cm above the soil surface to retain seeds in the subplot. An additional, control, subplot per plot had no seeds added. Dried grass litter was added to a 2.5-cm depth on all but two subplots with seeds per plot. Litter treatments were randomly assigned to the subplot locations.

We counted the number of germinating seedlings in each subplot in the spring of each year (when nearly all seedlings emerge, Baskin & Baskin 1988). Seedlings were then removed by clipping stems at the soil surface to avoid causing soil disturbance. To determine the fate of non-germinating seeds, we removed two mesh baskets (one with and one without litter) from each plot for the first two summers of the study and one basket (with litter) per year thereafter. We sieved the soil, removed and counted intact seeds, and performed tetrazolium tests as previously described. As only a single seedling was found throughout the study in the control plots, these subplots were not sieved and not considered further.

We calculated yearly survival in several ways. For the first year (i.e. from establishment of plots in November or December until the following summer), it is straightforward to determine the number of seeds that survived out of the known total of 140. For second, third or fourth year survival, one can divide the number that survive for 2 years (or 3 or 4 years) by the estimated number of seeds that had survived the first year (or second or third year). However, as the numerator and denominator of these ratios are calculated by sieving seeds from different subplots, there is greater uncertainty in these estimates. Alternatively, if one assumes that there is constant yearly survival (s), one can estimate seed survival by solving for s in the equation: number sieved in nth year = 140 sn. Thus for 2 years s is estimated by the square root of the fraction that survives, and so on.

We calculated the number of eaten or decayed seeds by subtracting mean numbers of seeds (and seedlings for surface seeds) from the known 140 initial seeds per site. Note that the sum of the number of seeds that became seedlings, remained in the soil, and were eaten or decayed must equal 140 for each subplot. In the case of the buried seeds, we assume that either germination did not occur or that germinated seedlings at a 25-cm depth were unlikely to successfully emerge as seedlings. Thus statistical analyses on all three variables are, by definition, interdependent. We chose to perform analyses on seedling emergence and seed survival and present average predation levels for comparison.

contribution of previous year's seeds vs. the older seed bank to numbers of plants present

In August 2000, we initiated an experiment with 60 reproductive plants of H. annuus in a field site in Douglas County, Kansas, near Clinton Lake (subsequently referred to as the Clinton site). The site is a recent restoration area that was planted with wetland and prairie species in 2000; in previous years the field had been planted with corn and soybeans. In 2000 and 2001, the site was in early old-field succession and common ragweed (Ambrosia artemisiifolia) was dominant. Low-lying areas at the site flooded periodically, but not in the experimental area. Sunflower plants occurred as scattered, large individuals in the first year (2000), but as a higher density of smaller plants in 2001. The Clinton site was located 15 km south-west of the KSR site.

We chose the 60 individuals randomly along two transects, with the condition that each individual was initially between 25 and 50 cm tall and was at least 5 m away from neighbouring tagged plants. Plants were assigned randomly to one of four treatments (with every successive group of four plants along a transect including one plant of each treatment): (i) no seed dispersal and no soil disturbance; (ii) soil disturbance and no seed dispersal; (iii) seed dispersal and no soil disturbance; and (iv) seed dispersal and soil disturbance. For the no seed dispersal treatments, we removed heads at early stages of maturity. Otherwise, seed heads were tagged (to prevent double counting) and counted weekly throughout October to determine cumulative head production. We removed all individuals located up to 4 m from the experimental plants in late September 2000 (prior to seed dispersal) so that the soil at the base of ‘no seed dispersal’ plants was not likely to receive seeds from plants that year and all seeds deposited at the base of ‘seed dispersal’ plants came from the experimental plants. We based these procedures on the observations that sunflower plants were scattered at the site and that other experiments had shown that fewer than 4% of the seedlings produced from seeds from small patches of sunflowers were found more than 3 m away from the patches (D. Pilson and H. Alexander, unpublished data).

Prior to seed dispersal in 2000, we sampled 236 mL of soil from a surface area of 49 cm2 beneath each plant. This soil was subsequently sieved and examined for sunflower seeds. In November 2000 and February 2001, the soil in a 1-m radius around the remains of half of the original experimental plants was broken up to a depth of 15 cm (soil disturbance treatment). To estimate seed production in 2000 for plants with seed dispersal, we multiplied the cumulative number of heads by the average number of seeds per head at the site. We determined the latter by counting seeds per head for 45 heads haphazardly chosen in fall 2000 from non-experimental plants along the same transects as the experimental plants. Counts of non-damaged seeds were done when heads were mature but prior to dispersal. Although we did not observe extensive bird predation on the heads, as does sometimes occur in natural and crop populations (Berglund 1994), birds and other animals could have eaten seeds once dispersal began.

On 7 April and 8 May 2001, we counted seedlings (most at the cotyledon stage) within 1 m of the location of all the plants examined in 2000 except one that could not be found. Other studies have suggested that 87% of the seedlings produced from small patches of sunflowers are within 1 m of the source (D. Pilson and H. Alexander, unpublished data), and our sampling area should therefore have included most of the progeny of the plants examined in 2000. Total head production in 2001 was determined for all plants rooted within these 1 m radius circles by tagging and counting heads at each location every week between 8 August and 28 September. On the latter date, we also counted the number of reproductive plants in each circle. Heads initiated before October are likely to produce mature seeds before October frosts.

statistical analyses

Statistical analyses, were performed for both studies using SAS (version 8) (SAS Institute 1989). For the seed fate studies, anovas were used to explore the effect of establishment year and litter treatments on seed and seedling numbers, and on the differences in seed and seedling numbers between years over the course of the study. For the field experiment at the Clinton site, we used anovas to explore the effect of seed dispersal and disturbance treatments on numbers of plants, and correlations were performed between variables. In one-way anovas of the number of seeds or seedlings persisting each year, we used Tukey a posteriori tests at the 0.05 level to explore yearly differences. In the two-way anovas, we presented the Type III sums of squares (partial sums of squares not affected by ordering of effects in a model). Dependent variables were transformed (square root or log, as indicated in the Results) as necessary to meet assumptions of homogeneity of variance.


fate of seeds in the soil

Buried seeds

On average, 47% of the 140 original seeds were still viable after 4 years for plots initiated in 1999. More rapid loss of seeds occurred in the plots established in 2000, with only 16% still present after 3 years (Fig. 1). There were significant differences in the number of seeds found in different years (effect of year in plots established in 1999: F3,55 = 12.12, P < 0.0001; in 2000 established plots using log transformed data: F2,42 = 21.55, P < 0.0001); a posteriori tests revealed that attrition of seeds slowed in later years (Fig. 1). Estimates of yearly survival percentages varied from 71.3% to 83.1% for 1999 established plots and from 46.3% to 78.6% for 2000 established plots, and in all but one case differed significantly in the two sets of plots (Table 1).

Figure 1.

Mean number of viable seeds (± standard error) persisting in the soil for up to 4 years following burial of 140 seeds in mesh packets at a 25-cm depth in year 0 (a, 1999; b, 2000). Within each graph, bars with the same letter are not significantly different at the 0.05 level using a Tukey a posteriori test following a one-way anova examining the effect of year on numbers.

Table 1.  Estimates of average yearly probability of survival for wild sunflower seeds buried at 25-cm depth. Experimental plots were initiated in both 1999 and 2000. First year survival refers to survival from the initial establishment of the study to the following spring; second year survival refers to survival from 2000 to 2001 for plots established in 1999 and from 2001 to 2002 for the plots established in 2000, and so on. Annual survival ratios were calculated either from counts on separate replicates in the two years or assuming a constant ratio (see text). One-way anovas were performed to test for effect of establishment year; only significant differences in treatments are noted (*P < 0.05, **P < 0.01, ***P < 0.001, ****P < 0.0001)
Year of survivalPlotMean (SE)Effect of establishment year
First19990.815 (0.036)F1,28 = 23.0****
20000.488 (0.058)
Second19990.807 (0.063)F1,28 = 6.81*
20000.543 (0.079)
ThirdAll plots0.786 (0.047)Year effects were not significant
Fourth19990.713 (0.076)
Constant (2nd)19990.789 (0.038)F1,28 = 56.23****
20000.463 (0.020)
Constant (3rd)19990.831 (0.011)F1,27 = 140.87****
20000.526 (0.023)
Constant (4th)1999 plots0.819 (0.025)

Seeds deposited on the soil surface

We assumed that all seeds emerged as seedlings, persisted in the soil or were eaten or decayed, i.e. that secondary dispersal out of the baskets did not occur. Sunflower seedlings were only found in the baskets except in one case (out of 210 baskets), where several seedlings were found immediately adjacent. We deleted this data point, as it appeared that the seeds had been washed out of the basket.

Depending on treatment and year of establishment, 10.7% to 29.0% of the initial 140 seeds produced seedlings in the first year (Fig. 2). Only about half as many seedlings emerged in the presence of litter as in its absence (Fig. 2; F1,204 = 51.06, P < 0.0001); neither the establishment year nor its interaction with litter were significant (anova done on square-root transformed data). Seedlings were rare by the second year (Fig. 2), with a significantly higher number of seedlings in the litter treatments than in treatments without litter (F1,141 = 6.72, P < 0.05), and in the plots established in 1999 vs. 2000 (F1,141 = 24.51, P < 0.0001) (Fig. 2). There was no interaction between year of establishment and litter treatment. When plots with litter present were analysed (a posteriori test results in Fig. 2), there were significant differences in numbers of seedlings found in different years (plots established in 1999 using square root transformed data: F3,56 = 115.32, P < 0.0001; plots established in 2000 using log transformed data: F2,42 = 93.86, P < 0.0001).

Figure 2.

Mean number of viable seeds (± standard error) persisting at the soil surface for subsequent years following sowing of 140 seeds in year 0 (a, 1999; b, 2000). Mean number of seedlings (± standard error) emerging in years 1 and later (c, year 1 = 2000; d = 2001). For all plots, the dark bar indicates no litter and the light bar indicates litter treatments. Within each graph, bars with the same letter are not significantly different at the 0.05 level using a Tukey a posteriori test following an anova exploring the effect of year on numbers. Such a posteriori analyses were only performed on litter treatments so that the largest number of years could be analysed.

Yearly survival probabilities (or ratios) of seeds on the soil surface were not significantly affected by the litter treatment but were sometimes different between plots established in 1999 (27.1% to 75.7%) and 2000 (11.7% to 51.7%) (Table 2). Average yearly survival appeared higher for seeds that survived into the later years of the study compared with early years (after the first year, old-field succession occurred in the plots so that bare soil or litter treatments were no longer obvious). By the third year, 8% or less of the original 140 seeds were still present in the soil (Fig. 2). There were significant differences in numbers of seeds in different years (plots with litter established in 1999, using square root transformed data: F3,50 = 22.72, P < 0.0001; in 2000, using log transformed data: F2,42 = 17.31, P < 0.0001). A posteriori tests reveal that there was less attrition in the later years of the study (Fig. 2).

Table 2.  Estimates of average yearly probability of survival for wild sunflower seeds on the soil surface. Experimental plots were initiated in both 1999 and 2000. First year survival refers to survival from the initial establishment of the study to the following spring; second year survival refers to survival from 2000 to 2001 for plots established in 1999 and from 2001 to 2002 for the plots established in 2000, and so on. Annual survival ratios were calculated either from counts on separate replicates in the two years or assuming a constant ratio (see text). Two-way anovas were performed to test for effects of establishment year and litter effects; only significant differences in treatments are noted (*P < 0.05, **P < 0.01, ***P < 0.001, ****P < 0.0001)
Year of survivalPlotMean (SE)Effect of establishment year
First19990.271 (0.014)F1,54 = 55.97****
20000.117 (0.015)
SecondAll plots0.412 (0.046)Year effects were not significant
ThirdAll plots0.517 (0.085)Year effects were not significant
Fourth19990.757 (0.109)
Constant (2nd)19990.295 (0.019)F1,48 = 16.22***
20000.187 (0.018)
Constant (3rd)19990.404 (0.030)F1,28 = 28.11****
20000.157 (0.036)
Constant (4th)19990.450 (0.044)

Predation levels in the first year were higher with litter present. For example, 56.1% of the seeds in the litter treatment were lost to predation compared with 43.8% of seeds in the no litter treatments for the plots established in 1999. For plots initiated in 2000, predation levels were 77.7% vs. 59.6% for litter and no litter plots, respectively.

contribution of previous year's seeds vs. the older seed bank to numbers of plants present

Plants that were allowed to disperse seeds produced on average 130.4 ± 16.5 heads (± SE), with a range of 25–329. As average seed production per head was 176.1 ± 10.0, an average of 22, 964 seeds per plant were available for dispersal. At the time of the first census, seedlings were very small and emergence was incomplete, but by the second, the main emergence period was past and some early emerging seedlings may have died. Significantly more seedlings (4.5 times (April) to 9.8 times (May)) were found in locations where plants had dispersed seeds, as compared with sites where seed dispersal had been prevented (Table 3). Significant recruitment in the latter suggests the existence of a persistent seed bank. Disturbance had no effect on numbers of seedlings or other traits and is not discussed further.

Table 3.  Comparisons of mean 2001 population traits (SE) for circular plots of 1 m radius around wild sunflower plants that did and did not disperse seeds in 2000. All analyses are based on two-way analyses of variance with factors of presence/absence of seed dispersal in 2000 and soil disturbance/no disturbance; disturbance and its interaction term were never significant and so are not presented. Analyses denoted with an ‘a’ were log transformed and with a ‘b’ were square-root transformed to improve homogeneity of variances. Significant differences are noted: **P < 0.01, ***P < 0.001, ****P < 0.0001
 Seed dispersal allowed in 2000Seed dispersal not allowed in 2000F
Number of seedlingsa (April 2001)179.1 (36.9)40.5 (10.1)F1,56 = 23.45****
Number of seedlingsa (May 2001)202.0 (32.5)20.6 (3.5)F1,56 = 54.55****
Number of mature plants (September 2001)b 33.7 (2.7)13.7 (1.8)F1,55 = 38.15****
Survival percentage to September 2001 33.9 (7.1)68.1 (5.9)F1,53 = 14.13***
Number of heads (2001)150.3 (11.9)97.1 (13.4)F1,55 = 8.49**
Head production per plant (2001)  5.1 (0.5) 8.3 (0.9)F1,53 = 10.26**

There were 2.5× more mature plants and 1.5× more heads in September 2001 in locations where seed dispersal had occurred in 2000 compared with where it was prevented. This smaller difference between dispersal treatments than observed for number of seedlings implies higher survival percentages for seedlings emerging where there was no seed dispersal. Specifically, an average of 68.0% of May plants survived to September in locations without seed dispersal compared with 33.9% where there had been dispersal the previous year (Table 3). The number of heads per plant in 2001 was significantly higher in locations with no seed dispersal, calculated by dividing head production in 2001 by the number of mature plants present (Table 3).

Total number of mature plants in September 2001 initially increased, as expected, with the number of seedlings in May 2001 but the relationship then plateaued (Fig. 3a). We also found a curved relationship between total number of heads and the number of seedlings in May (Fig. 3b), and between total number of heads and total number of mature plants (Fig. 3c). To explore these curvilinear relationships statistically, we first examined whether quadratic functions would fit better than straight lines. Although parabolas fit significantly better than straight lines, we concluded that predictions of a downward turn at extremely large values were unrealistic for our data set. Thus, we chose to fit and use a hyperbolic equation, y = (bx)/(1 + ax). We found that the residual sum of squares was lower for the hyperbolic equation than the quadratic equation for all three cases, supporting a curvilinear relationship between the variables.

Figure 3.

Number of mature reproductive plants in fall 2001 (a) and number of heads (= heads) in fall 2001 (b) plotted against the number of seedlings in May 2001 for each 1 m radius circle at the Clinton site. For c, the number of heads in fall 2001 is plotted against the number of mature reproductive plants in fall 2001. For all plots, fitted lines are for a hyperbolic equation.

In analyses of only the seed dispersal plots, there was a significantly positive correlation between the number of heads in 2000 and the number of seedlings in April (r = 0.53, P = 0.0025) and in May 2001 (r = 0.60, P = 0.0005), and the number of reproductive plants in 2001 (r = 0.48, P = 0.008). The correlation between the number of heads in 2000 and in the same area in 2001 was r = 0.30 and was not significant at the 0.05 level (P = 0.10).

In the 55 pre-dispersal soil samples that were examined (five were lost), 14 seeds were found (46 samples with 0 seeds, and 5, 3 and 1 samples with, respectively, 1, 2 or 3 seeds present). This average of 0.25 seeds per 49 cm2 sample area extrapolates to a bank of 163.17 seeds in the soil of a 1-m radius circle.


fate of seeds in the soil

Seed survival and seed bank size can depend on seed burial and habitat conditions, seed age and density, seed predation, and on factors, such as herbivory, that affect seed inputs (reviewed in Baskin & Baskin 1998). We were interested in sunflower seed survival under natural conditions when seeds land on the soil surface, as well as the maximum survival that might occur with burial and low predation. For the latter, we felt that the 25-cm burial should represent a high level of seed persistence as predation should be further reduced by the packet. At this depth there was no light penetration, which enhances germination in H. annuus (Baskin & Baskin 1988), and Teo-Sherrill (1996) found no seedlings emerged from seeds of H. annuus buried 12.5–22.5 cm deep. Our seed survival results were often dependent on the year the experiment was established; in buried packets, plots established in 1999 still had, on average, 47% of the seeds remaining after 4 years, while the plots established in 2000 had only 16% of the seeds remaining after just 3 years. As expected, seed persistence on the soil surface was much lower than in the burial plots, given the high probability of germination, exposure to predation and exposure to more variable conditions (potential seed damage due to wetting and drying). Eight per cent of the seeds persisted for 4 years in the plots initiated in 1999 and fewer than 2% for 2 years in the plots set up in 2000.

Replication of the entire experiment in two different years provided surprising results. The observation that seed survival was often lower in the plots established in 2000 compared with those initiated in 1999 (whether considering calendar year or the age of the seeds) cannot be explained by yearly weather variation. The two sets of plots were adjacent to each other and only slightly different in topography (with the 2000 plots slightly lower), but the sites could still differ in abiotic or biotic factors that affect seed survivorship. Alternatively, the seeds used in 2000 may have had inherently lower survivorship than those in the 1999 cohort. Other studies have shown that seeds collected at different times or under different conditions can differ in characteristics that affect germination and dormancy (Cresswell & Grime 1981; Baskin & Baskin 1998; Qaderi et al. 2002). As we had no a priori reason to expect the observed differences in survival between the 1999 and 2000 plots, we lack the data on seed mass or other traits needed to test for seed quality differences.

The presence of litter on the soil surface reduced seed germination (or early survival of seedlings after emergence) in the first year, as has been found for other species (Jutila & Grace 2002) but, surprisingly, had no effect on numbers of viable seed. Litter may have made it more difficult for seeds to germinate while increasing seed predation, thus leading to no difference in numbers of seeds persisting in the soil. We do not understand why seedling numbers were higher in the presence of litter in the second year. However, this result may simply reflect differences in treatments in the number of remaining seeds available for germination; effects of litter on seedling emergence are often reported (e.g. Hamrick & Lee 1987).

Although seed survival data are often presented as log-linear seed-decay curves (see Lonsdale 1988; Rees & Long 1993; Rees 1997 for critiques), many population dynamic models require estimates of yearly seed survival (Kalisz & McPeek 1992, 1993; Jordan et al. 1995). Our study (Tables 1 and 2) highlights both the various ways that yearly seed survival can be estimated and the range of possible values. We agree with Teo-Sherrill (1996) that presenting a single average value across all experiments and years can be misleading. In addition to variation in survival due to year of establishment (plots initiated in 1999 vs. 2000) and environmental effects, an additional reason for heterogeneity in estimates is that seed sieving is destructive, and thus sampling errors are likely because some estimates use data collected in different subplots (Lonsdale 1988). Seed attrition was often reduced in the later years in the study (Figs 1 and 2). Such variation could indicate age-specific survival differences; for example Rees & Long's (1993) review illustrates that a Type III survivorship curve occurs for many composites. However, as noted by Rees (1997), the conditions experienced by the seeds are likely to change over the course of an experiment, making it difficult to test for age-specific survival. In our study, higher survival of seeds in the later years could result from any combination of differences due to age-specific seed mortality, density-dependent seed mortality (as fewer seeds are present in later years) or habitat-specific seed mortality (the habitat changed from tilled soil to a grass/forb mixture). Regardless of the exact cause, these data suggest that seeds that persist for more than 1 year after dispersal often have a greater probability of also persisting to future years.

It was unfortunately not feasible to carry out the seed survival study at the Clinton site, or the field experiment at KSR. We can note, however, that other studies of dormancy in wild sunflower seeds are roughly comparable with ours. For example, Teo-Sherill's (1996) report of 20% survival after 2.5 years for seeds of H. annuus buried 2.5–22.5 cm is similar to our 2000 results, although the two studies differ in both exact burial depth and in the presence of a mesh barrier in our studies. Similarly, our results are generally consistent with Toole & Brown's (1946) summary of Duvel's experiment, which reports that, depending on burial depth, 45–67% of seeds of H. annuus survived for 1 year. One report, the Burnside et al. (1981) buried seed experiment (at 23 cm depth), which revealed only 1–2% annual germination of wild sunflower, does differ greatly from our study. However, as Teo-Sherrill (1996) suggested, the low germination reported by Burnside et al. (1981) is misleading as tetrazolium tests of viability were not performed on exhumed seeds and seeds were tested in November (at a time when they would be intrinsically dormant). It is interesting that Burnside et al. (1996) found that 3% of the seeds from a 20-cm depth could germinate after 17 years of burial, suggesting that some seeds can persist for many years. However, the methodology used was the same as in the earlier study, complicating interpretations.

contribution of previous year's seed vs. the older seed bank to numbers of plants present

Seed fate studies, like those described above, show how long seeds persist but do not provide information on the ecological significance of seed bank survival for plant population biology. By comparing numbers of plants in manipulated sites where the previous year's seeds were, or were not, allowed to disperse, the role of the seed bank in the population may be inferred. From Table 3, we estimate that approximately 23% of the seedlings under the seed dispersal plants have their origins in the seed bank (an average of 41 seedlings are found in April even without seed dispersal and are likely to be among the 179 at sites with seed dispersal). Numbers of seedlings changed from April to May, with an unexpected decrease in the numbers found in plots without seed dispersal and increases in numbers in plots with seed dispersal, such that only a 10.4% seed bank contribution would be estimated from the May data. Our estimates of seed bank contributions are roughly consistent with the average number of seeds sieved from the soil at the site. For example, a mean of 41 seedlings from the seed bank would suggest 25% emergence for the estimated average of 163 seeds present in the 1-m radius area; this is of the same order of magnitude as emergence percentages found at the KSR site. However, the small number of seeds sampled in the soil and the fact that soil seed banks often have a patchy distribution (Benoit et al. 1992) makes it difficult to interpret estimates of average seeds per square cm of soil.

In contrast to our expectations, we found that soil disturbance had no significant effect on our results. Most likely, the relatively low vegetation cover at the site (bare soil was easily visible) reduced the impact of our manipulation. However, the presence or absence of seed dispersal in the previous year did alter the numbers of sunflowers at local sites. Specifically, prevention of seed dispersal, on average, reduced total numbers of seedlings, mature plants and subsequent heads at the local sites. The largest treatment effect was found for seedling counts, with a substantially reduced effect on number of mature plants and head production. The most likely explanation for this reduction in treatment effects with plant development is the presence of density-dependent processes and the considerable plasticity of H. annuus growth. As shown in Fig. 3, plots with high numbers of seedlings did not necessarily produce correspondingly more mature plants and heads, as compared with plots with fewer seedlings. Thus, at low-density seedling sites (usually without seed dispersal from the previous year), plants had, on average, a greater probability of summer survival and greater average head production on a per plant basis, relative to high-density seedling sites.

Results of the study at the Clinton site could have been quite different in a site with more ground cover. In that case, the presence or absence of a disturbance treatment is likely to have determined whether seed emerged from the seed bank. Unfortunately, we rarely know the vertical distribution of seeds in the soil, which has a large effect on how disturbance regimes are likely to affect seedling emergence from the seed bank (Mohler 1993). Further, plants at our sites in 2000 produced large numbers of heads; a similar study with smaller plants could have led to different results, as density dependence might be less pronounced at lower seed densities.

Our results are consistent with those from studies of annual plants that persist despite reproductive failure the previous year (Baskin & Baskin 1975, 1980). Recovery from the seed bank can be dramatic: a population of Sedum pulchellum that resulted from a seed bank had a density of 5.6 plants dm−2 (Baskin & Baskin 1980) and germination of buried seed produced a population of the annual Isanthus brachiatus comparable in size with those found in nearby habitats where seed production had occurred the previous year (Baskin & Baskin 1975). Rogers & Hartnett (2001) found seed rain to be of low importance in seedling recruitment in a tallgrass prairie. In contrast, seed dispersal was very important and seed bank contributions were negligible in two other grassland studies (Bullock et al. 1994; Edwards & Crawley 1999a). Peart (1989) estimated that, for the grass Vulpia bromoides, only 3% of the seedlings were a result of recruitment from the seed bank.

The experimental methods used to estimate seed bank vs. seed rain contributions under field conditions have potential drawbacks. For example, Peart (1989) estimated seed bank contributions by counting the number of grass seedlings appearing in seed exclosures that prevented seed fall. For some species, seedling establishment was higher in the exclosure than in the surrounding area, suggesting artifactual effects of exclosures may have inadvertently increased seedling emergence. In our study, it is possible that long-distance seed dispersal may have accounted for some of the seedlings emerging in sites where we clipped all the heads off the plants. In other studies seed bank contributions are determined by comparing numbers of seedlings in control sites with those in sites where field soil is either sterilized or replaced with other soil that lacks seeds (Bullock et al. 1994; Edwards & Crawley 1999a; Rogers & Hartnett 2001). In such studies, it is possible that the control and experimental treatments may have different seed germination conditions. However, there are many advantages to using natural seed banks and natural seed dispersal. It is difficult to create realistic experimental seed banks because the spatial patchiness and seed density are rarely known, and natural seed rain from plants is likely to lead to different seed dispersal patterns than treatments where seed is sown by researchers.

Data on seed bank size can be useful for interpretations of field experiments. In Edwards and Crawley's (1999a) study, for example, virtually all seedling recruitment was a result of recent seed dispersal. The fact that the community composition of the seed bank was unrelated to the established vegetation reinforces their conclusions. On a population level, Cabin & Marshall (2000) and Cabin et al. (2000) concluded that high seed production of the above-ground plants of a desert annual did not necessarily lead to a large number of seeds entering the seed bank, and that seed bank size was more related to the time between seed inputs into the seed bank and to seed survival rather than to the actual seed production at a particular time. In our study, we did take soil samples. However, in retrospect, they were inadequate to provide a detailed view of seed bank size, presumably because of the patchy nature of seeds in the soil. Given the difficulty of sieving large quantities of seeds, there are advantages to considering Naylor's (1972) mark-recapture technique for evaluating the contribution of recently produced seeds vs. buried seeds to numbers of seedlings.

In most studies of the role of the seed bank, collection of data is stopped after plants reach the seedling stage (e.g. Putwain et al. 1968; Peart 1989; Bullock et al. 1994; Edwards & Crawley 1999a; Rogers & Hartnett 2001). However, the contribution of seeds from the seed bank to numbers of seedlings and their contribution to the final reproductive output of the patch may not be the same. For example, because of density-dependent processes, our study showed a much larger effect of recent seed production on numbers of seedlings in the spring than on head production the following fall. Similar results have occurred in studies of herbivore and pathogen effects on plants. For instance, several studies have shown a strong effect of seedling pathogens or seed predators on numbers of seedling (Edwards & Crawley 1999b; Alexander & Mihail 2000; Cummings & Alexander 2002). However, in these same studies, the effect of the presence of the seed consumers on reproductive output of the patch is much less than expected based on their effects on seedling numbers. Instead, the low-density plant populations that occur in treatments with the seed consumer have reduced intraspecific competition and thus greater per-capita seed production than in treatments without the seed consumer.

In addition to field experiments examining the contribution of the seed bank to numbers of plants in a population (like at our Clinton site), a powerful approach is to create population models that explore the effect of the seed bank. Studies of the variability in seed survival under different conditions, like at the KSR site, are important for parameter estimation in such models. An excellent example is the work by Kalisz (1991) and Kalisz & McPeek (1992, 1993) who incorporated experimental data on seed bank persistence into demographic models of the herbaceous plant Collinsia verna. Simulated populations with a seed bank (compared with those without) grew faster and had an increased time to extinction coupled with a decreased likelihood of extinction. A community-level study (Stöcklin & Fischer 1999), where species with long-lived seeds had a lower probability of extinction in grassland remnants followed for 35 years, supported their findings. Kalisz & McPeek's (1993) results depended on the degree of predictability of environmental regimes. For sunflowers in the Great Plains, yearly variation in the abiotic environment and in population size is common. For example, numbers of sunflowers in western Nebraska vary greatly between years, presumably as a result of variation in environmental conditions (240 100 plants were found on a 28-km roadside survey in 2001; only 6720 were found on the same route in 2002, a year with an extreme drought; D. Pilson, unpublished data). Seed banks are likely to contribute to the persistence of the species in such variable environments.


There are two likely roles of seed banks for wild sunflowers, as well as for many plant species. First, in situations where reproduction occurs the previous year, seed banks increase the pool of available seeds for germination. Secondly, given appropriate germination conditions, seed banks may allow a species to persist at a site even when localized reproductive failure or past extinction of a population due to succession means that seeds have not been produced in recent years. The relative importance of these two roles and the relevant time frame depends on survival of seeds in different conditions. If seeds are buried deeply (due to animal digging or ploughing), and/or are unlikely to be eaten, our upper estimate of 83% annual survival suggests that 15% of the original seeds deposited at a site could still be present up to 10 years later. However, the majority of seeds are likely to be deposited on the soil surface, where seed survival is much lower than if buried (in our studies, often only 10–30%). Given such values, we expect that seed banks are primarily important for sunflower population dynamics in periods of 1–3 years after their establishment. At the Clinton site, we can infer that 10% to 23% of the seedlings establishing at this site came from the seed bank. However, we also estimated that the reproductive output per seedling was 3.6 (2.4–4.7) for seedlings at local sites emerging strictly from the seed bank as compared with 0.8 (0.74–0.84) for seedlings coming from a mixture of the previous year's seed set and seed banks (calculated by dividing head number by seedling numbers). Thus, seedlings existing in low-density sites where only the seed bank is responsible for germination can have a disproportionate contribution to seed production. Conversely, at high-density seedling sites, the addition of more seeds from the seed bank may have little effect on final seed output and population dynamics, as suggested by the fact that reproduction at a site was asymptotically related to numbers of seedlings. These results emphasize the need to integrate seed bank studies with research on the entire life cycle of the plant.


We greatly appreciate field and laboratory work by D. Brunin, J. Emry, S. Frain, M. Graham, G. Loving, J. Moody-Weis, J. Nash, G. Pederson and J. Zaudke, and comments on the manuscript by J. Antonovics, J. Emry, J. Moody-Weis and D. Pilson. J. Antonovics and N. Slade provided useful statistical advice. Graphs were done by M. Tourtellot. The Kansas Field Station and Ecological Reserves staff supported our work and we appreciate access to land by the City of Lawrence Parks and Recreation Department. Financial support was provided by USDA (9904008) and NSF (9876676) grants.