Geographic variation in dynamics of an annual plant with a seed bank


  • Helen M. Alexander,

    Corresponding author
    1. Department of Ecology and Evolutionary Biology, University of Kansas, Lawrence, KS 66045-7534, USA
    Search for more papers by this author
  • Diana Pilson,

    1. School of Biological Sciences, University of Nebraska, Manter Hall, Lincoln, NE 68588-0118, USA
    Search for more papers by this author
  • Jennifer Moody-Weis,

    1. William Jewell College, 500 College Hill, Campus Box 1059, William Jewell College, Liberty, MO 64068, USA
    Search for more papers by this author
  • Norman A. Slade

    1. Department of Ecology and Evolutionary Biology, University of Kansas, Lawrence, KS 66045-7534, USA
    2. Natural History Museum/Biodiversity Research Center, 1345 Jayhawk Blvd., University of Kansas, Lawrence, KS 66045-7561, USA
    Search for more papers by this author

*Correspondence author. E-mail:


1. Most population dynamics studies are geographically restricted, yet species ranges are large. We performed multiyear roadside surveys of the sunflower, Helianthus annuus, at two locations that differ in precipitation (eastern Kansas, KS; western Nebraska, NE). Our goals were to (i) document if there was geographic variation in dynamics and evaluate the role of habitat variables and the landscape matrix; (ii) determine the likely amount of occupiable habitat and (iii) explore the role of seed banks in dynamics.

2. Geographical variation: Occupancy and mean numbers of plants per occupied roadside unit were generally higher in NE than KS. Sunflower abundance was linearly related to spring precipitation in NE but not in KS. Soil disturbance was associated with increased occupancy and apparent colonization, and reduced apparent extinction. Variation in the landscape adjacent to roads had a larger effect on occupancy in KS than in NE. In KS, smaller populations were more prone to apparent extinction; NE results were more variable. Note that we refer to ‘apparent’ colonization or extinction because seed banks may persist even when above-ground plants are absent.

3. Occupiable habitat: 25% of the roadside was never occupied by sunflowers in KS, despite surveying for 6 years. An asymptotic limit to occupancy in NE was not apparent, but fewer years were surveyed.

4. Seed banks: Seed banks appear prevalent. The strongest evidence comes from a year following a spring drought in NE, when 100s to 1000s of plants were found in units that lacked plants the year before.

5.Synthesis. We found both geographical similarities (role of soil disturbance, seed banks) and differences (response to rainfall variation, importance of landscape matrix) in sunflower dynamics. Our work suggests that for appropriate species (including many weedy and invasive plants), replicate roadside surveys are an efficient way to evaluate geographic variation in dynamics, the landscape matrix and habitat characteristics across a broad geographic area. Such data help bridge the gap between broad-scale distributional studies and small experimental plot studies, and provide insights on the population dynamics that underlie species ranges.


Dynamics of plant populations at a regional scale can vary depending on the proportion of habitat that is suitable and occupied by a species, which in turn affects the likelihood of dispersal among sites (Bullock et al. 2002; Freckleton & Watkinson 2002; Ouborg & Eriksson 2004). At one extreme, habitat patches may be isolated. Thus, once a population goes extinct, there is little chance of recolonization. These ‘remnant populations’ or ‘regional assemblages’ may persist, however, for long periods because of seed banks (Eriksson 1996; Freckleton & Watkinson 2002). The other extreme comprises ‘patchy populations’ (Harrison & Taylor 1997), where suitable habitat is abundant, dispersal occurs frequently among sites, and local population dynamics largely determine regional dynamics (‘spatially extended populations’, Freckleton & Watkinson 2002). Classic metapopulations are an intermediate type and describe situations where populations occupy habitat patches within a matrix of unsuitable habitat, and regional dynamics result from a balance of colonization and extinction.

Regardless of where a species exists on this continuum, research on regional dynamics is difficult. The conceptual and logistical challenges include: (i) that although geographical variation in dynamics is likely, it is difficult to ‘replicate’ studies across a species range; (ii) the problem of defining ‘occupiable’ habitat and (iii) the fact that viable seeds in the soil (=seed banks) complicate interpretation of colonization and extinction (Freckleton & Watkinson 2002; Gaston 2003; Ouborg & Eriksson 2004). The first issue is self-evident: we expect colonization and extinction dynamics of plants to vary geographically given differences in the abiotic and biotic environment and landscape matrix. However, despite extensive studies of plant distributions, little data exist on geographic variation in dynamics. For example, of the 26 studies that Freckleton & Watkinson (2002; Table 1) categorize to have a ‘regional’ spatial scale, only four include yearly data on dynamics for multiple regions across the range of a species. In contrast, most researchers either focus on dynamics of multiple populations within a single region, or they emphasize broad geographical surveys, yet lack yearly data on population colonization and extinction. Studies that have been done show, for example, that probability of extinction of a dune plant varied depending on beach characteristics (Thrall, Burdon & Bock 2001), while dynamics of the annual Vulpia ciliata did not differ greatly across a range of habitats (Watkinson, Freckleton & Forrester 2000). A Brazilian annual exhibited metapopulation dynamics with geographic variation in the density of suitable habitat patches (Husband & Barrett 1998).

Table 1.   Total number of plants on the survey route, proportion of 80-m units occupied and mean number of plants per occupied unit (SE) for 6 years in KS and 4 years in NE (data from both sides of the road were combined for these summaries)
YearTotal number of plantsProportion occupiedMean plants per occupied unit (SE)
 199975420.4925.7 (2.55)
 200016 5320.4462.9 (8.27)
 200116 6470.5847.9 (5.93)
 200224690.4010.4 (1.23)
 200398740.4933.7 (3.33)
 20049490.217.6 (1.52)
 2001143 8810.77403.0 (30.1)
 200214270.2810.8 (2.6)
 200398 1240.81261.0 (30.3)
 200437 7600.76106.5 (12.4)

As an illustration of the second problem, researchers were unable to define suitable habitat patches in over half of the studies reviewed by Freckleton & Watkinson (2002). Empiricists have addressed this issue with experimental additions of plants to assess ‘occupiability’ (Ehrlén & Eriksson 2000; Mildén et al. 2006) and descriptions of suitable habitat (analysed with data on environmental variables (Bourg, McShea & Gill 2005) or patterns of species co-occurrence (Münzbergová & Herben 2004)). Such methods, however, may not be feasible at large spatial scales. When habitat patches cannot be easily defined, one can take a pragmatic approach of superimposing a grid over a region and recording occupancy and abundance for arbitrarily defined ‘populations’ within each grid square (Antonovics et al. 1994; Thomas & Kunin 1999). Unidimensional grids along roadsides have been effectively used to study range limits (Prince, Carter & Dancy 1985), invasion and spread (Wilcox 1989; Crawley & Brown 1995, 2004), scaling-down of colonization and extinction rates (Moody-Weis et al. 2008), and host–pathogen interactions (Antonovics et al. 1994; Antonovics 2004). Grid-based approaches result in complete coverage of a defined area; hence one does not have to make decisions about where to look for a particular species and risk missing plants in places judged to be unsuitable habitat. However, grid cells may differ in terms of suitability for a species, and it is difficult to know whether cells that lack plants are ‘occupiable’ or whether they would not be colonized even if seeds dispersed to such sites (‘unoccupiable’).

Finally, seed banks pose a problem since a ‘colonization’ event could be due to seed dispersal or to emergence from buried seeds. Similarly, the disappearance of plants at a site is not a true extinction if viable seeds remain in the soil. We thus refer to these events as ‘apparent’ colonizations and extinctions. Seed banks are notoriously variable in space and hard to quantify (Baskin & Baskin 1998). Although small-scale experimental studies can determine the relative importance of seed rain vs. seed bank (Alexander & Schrag 2003), such studies are difficult to scale-up to a regional level. Despite these challenges, seed banks are important to regional dynamics because they imply a source of colonists that is not directly dependent on the current proportion of sites that are deemed occupied based on above-ground plants (Gotelli 1991).

We studied regional dynamics of Helianthus annuus, an annual sunflower. We used identical methods to survey roadside populations in two locations (eastern Kansas (KS), western Nebraska (NE); 600 km apart) within the species’ broad distribution in North America. We addressed three issues:

  • 1Geographical variation: Dynamics could differ across a species’ geographical range if the habitat, landscape matrix or environmental conditions change the processes of apparent colonization and apparent extinction. KS and NE differ greatly in precipitation and the landscape types found adjacent to roads. Are there differences in dynamics between KS and NE? Do habitat variables (e.g. soil disturbance, precipitation) or the landscape matrix predict occupancy or apparent colonization and apparent extinction dynamics?
  • 2Occupiable habitat: Roadside plants have been cited as examples of spatially extended populations (Freckleton & Watkinson 2002); if so, the proportion of occupiable habitat should be high. Over a multi-year survey, are all roadside units occupied by sunflowers at some time or is there evidence of roadside units that are not occupiable?
  • 3Seed banks: Small spatial scale studies reveal that seed banks affect local sunflower population dynamics (Alexander & Schrag 2003; Moody-Weis & Alexander 2007). Are patterns of regional dynamics consistent with a major role of seed banks?

Materials and methods

Study organism

Helianthus annuus L. (Asteraceae, ‘sunflower’) is a native, self-incompatible weedy annual. In KS and NE, seeds germinate in April and early May. By early fall, plants are typically 0.5–2.0 m in height and produce showy yellow inflorescences. Annual seed survival in the soil varies, suggesting seed banks could be relatively short lived (1–3 years) or persist for long periods (≥10 years) depending on seed burial and environmental conditions (Alexander & Schrag 2003). In a field experiment, 10–23% of emerging seedlings came from soil seeds (Alexander & Schrag 2003). The timing and extent of soil disturbance had a large effect on seed bank formation and seedling emergence (Moody-Weis & Alexander 2007).

Roadside survey

Within North America, H. annuus occurs at high densities in the Great Plains (Moody-Weis 2006). This area, extending roughly from 95° to 105° W within the USA, has an east-west precipitation gradient with a major influence on flora and fauna (Hayden 1998). We conducted surveys along rural roads at two sites along this gradient: in eastern KS (1999–2004, 1000 mm annual precipitation) and western NE (2001–2004, 420 mm annual precipitation). We followed survey methods successfully used by Antonovics and coworkers (Antonovics et al. 1994; Antonovics 2004; and see Moody-Weis et al. 2008 for details of our methods). In KS, a 23.8-km continuous route was selected that went through Douglas and Jefferson Counties. The KS landscape is a mosaic of grasslands (primarily fields of Bromus inermis; occasional crop fields) and deciduous woodlands. In NE, an 18.6-km route was chosen in Keith County; the NE landscape includes cropland (typically winter wheat, corn and fallow in rotation) and rangeland. We knew that sunflowers were common in these regions, but exact survey routes were chosen without prior knowledge of sunflower locations. Along each route, and on each side of the road, a continuous string of 80-m roadside units was delineated. The KS route had 298 units, the NE route 233. We used a combination of automobile and bicycle odometers, GPS data, photos and landmarks to locate the ends of each 80 m unit. The width of each unit was 3–5 m and was defined by the width of the road verge.

Surveys were conducted over a 3–5 day period at peak flowering (late August (NE), early September (KS)) when plants were large, producing multiple yellow inflorescences, and thus easily detected. By walking, biking and driving, teams of 2–4 people counted the number of plants in the 80-m units on each side of the road (arbitrarily defined as roadside 1 and 2 at each site). We resurveyed the same routes and units each year. In 2001, the landscape matrix (i.e. the landscape type of the area immediately beyond the verge for each unit) was categorized as either woods or grassland (including a small amount of cropland) in KS and as crop or rangeland in NE. This categorization did not change across the years of our study. It was unusual for an 80-m unit to be bordered by a mixture of landscapes; when this did occur, the unit was recorded as the landscape that occurred over the majority of the length. In 2001–2004, we recorded the level of soil disturbance in each 80-m unit in spring, using a 3-level categorical scale. In 1999 and 2000 in KS, soil disturbance was instead recorded late in the fall using a 0–1 (1999) or a 0–2 (2000) scale. For all scales, the lowest category represents the absence of obvious soil disturbance, while the highest category reflects units with most soil overturned. Habitat disturbance categories did not include mowing, which was rare; low disturbance levels were likely because of animal burrowing while high disturbance levels were human-mediated (hand or power tools for digging or agricultural machinery).

All roadside verges were open (not wooded) and thus potentially favourable habitat. Roadsides were typically unmanaged. In KS, we only observed herbicide use in a localized area in 1 year. KS mowing was sporadic and did not typically occur during the early fall census period. In NE, we never observed herbicide use; mowing was rare and typically was done after surveys were complete. When large sunflower plants are mown, they quickly regrow and flower, so counts still could be made. Sunflowers were uncommon outside of the roadside verge since the surrounding landscape was managed for hay, crops, grazing or was wooded. When sunflower patches were outside of the roadside area, they were usually 10s to 100s of metres away. We do not recall any cases where sunflowers were directly adjacent to the roadside verge, but were not in the verge area.

We used daily precipitation data collected at the University of Kansas Field Station and Ecological Reserves (on the KS route) and from an automated weather station operated by the High Plains Regional Climate Center (6.4 km from the NE route). Since germination and early establishment of seedlings depends on spring conditions, we examined total precipitation from 1 April to 15 May, the time of peak seedling emergence (Alexander & Schrag 2003; Moody-Weis & Alexander 2007; D. Pilson, unpubl. data.).



A population was arbitrarily defined as plants within an 80-m roadside unit on a particular side of the road. Occupancy was defined as the proportion of all units with sunflowers, the probability of apparent colonization was defined as the proportion of units with plants in year t that lacked plants in year t − 1, and the probability of local apparent extinction (‘apparent extinction’ hereafter) was defined as the proportion of units without plants in year t that had plants in year t − 1. We based our definitions on the presence or absence of above-ground plants; by using the terms ‘apparent’ colonization and ‘apparent’ extinction, we are recognizing that seed banks may occur in units that have no sign of above-ground plants.

Predictors of occupancy

We were curious whether the landscape type adjacent to the roadside, roadside characteristics (recent soil disturbance) and seed numbers in previous years were predictive of occupancy. Since we lacked data on seeds, we used the number of plants per unit in previous years as a proxy. For KS 2001–2004 and NE 2003–2004, we used logistic regression with occupancy as the response variable and year, landscape type, disturbance class the previous spring or fall, and the number of plants present in both year t − 1 and year t − 2 as predictor variables. Since roadside units were adjacent to each other, variables were likely to be autocorrelated. Therefore, we examined partial autocorrelation coefficients for the residuals from these analyses. Typically, the first-order coefficient was much larger than any other coefficients. We calculated an effective sample size (n′) adjusted for the first-order coefficient, r, where n′ = n[(1 − r)/(1 + r)] (Cressie 1991). Z-values were used to test whether the regression coefficients were significantly different from zero (Z-test statistic = estimated coefficient/SE of coefficient); the Z-values presented were adjusted for the effective sample size (multiplying the original Z-value by the squareroot of the ratio of n′/n).

Each yearly survey provides data on what units are occupied or not occupied. With multiple surveys, one may find that all units are occupied in at least 1 year, or alternatively, that the same units may remain empty year after year. The latter situation could be consistent with either unsuitable habitat or habitat isolated from seed sources. For each site and roadside, we thus calculated the cumulative number of unoccupied units observed over successive years of the survey and determined whether data could be fit with a nonlinear regression equation (y = a + becx) using the ‘nlin’ procedure in sas 9.1 (SAS Institute. 2002). In this equation, y = the cumulative number of unoccupied units, x = number of survey years, and a, b and c are parameters that define the curve. An estimate of c < 0 suggests an asymptote for the number of never-occupied units. We used these equations to predict the number of ‘never occupied’ units when x = 10 years.

Predictors of apparent colonizations and apparent extinctions

Apparent colonization should be more likely if soil disturbance was recent; soil disturbance brings buried seeds to the surface and creates an ideal environment for seeding emergence and growth. Similarly, apparent extinction should be less likely with recent disturbance since an annual plant could continually establish at such a location. We also predicted that the landscape type adjacent to a unit could alter apparent colonization and apparent extinction probabilities. The irregular spacing of apparent colonization and apparent extinction events along the roadsides meant that we could not adjust for autocorrelation using simple partial autocorrelation coefficients. We therefore used partial Mantel tests (Fortin & Gurevitch 1993) in the program r (Venables et al. 2002) to adjust for the physical distance separating units in the analyses, and thus examine associations with soil disturbance or adjacent landscape type. Physical distance was defined as the number of roadside units that separated two populations. Analyses were done for each year, site and roadside. The significance of the coefficients was assessed using a permutation test, and adjustments for multiple tests were done with the Dunn–Sidak method (Gotelli & Ellison 2004). However, since Gotelli & Ellison (2004) provide several arguments why the Dunn–Sidak method and other adjustments for multiple tests are excessively conservative and should not be used, we provide both the original and adjusted significance values.

Apparent colonization events include both true colonizations due to seed dispersal and colonizations due to seed banks. Although we cannot distinguish between these two kinds of events, we would expect that true colonizations should occur close to potential sources of seeds. Thus, for each apparent colonization event, we examined the distance (in 80-m units) to the nearest possible source of colonists in the previous year, considering populations from both sides of the road. An apparent colonization in an adjacent or opposite unit to a previously occupied unit was defined to have a distance of 1 unit. True colonization events should typically generate small population sizes because of small numbers of colonizing propagules. This relationship should be most obvious in apparent colonizations that were spatially isolated. We thus determined the number of plants per unit for units that were newly colonized and isolated versus those that were persistent and isolated. An isolated unit was defined as one where the nearest possible source population from the previous year was more than one unit away.

True extinctions (i.e. absence of all plants and seeds from a particular location that had plants the year before) should be more likely to occur in units with small population sizes. To examine if this occurred with the apparent extinction events we could document, we compared the number of plants per unit for units that persisted to the next year and for units that lost all above-ground plants. Such comparisons were done for each site and year.

Analyses were done in Minitab 14 (Minitab, Inc. 2005) unless otherwise indicated.


Geographic variation

Habitat, landscape and climate

KS and NE roadsides were qualitatively similar in disturbance levels, but differed in landscape. On average, 17.5% (KS) and 18.2% (NE) of the roadside units were in the highest disturbance classes (most likely associated with sunflower germination). In KS, the majority of the landscape adjacent to the roadside was in grassland, with 15.7% in woods. Most of the adjacent roadside landscape in NE was in crops, with 17.8% in rangeland. The typical length of a KS roadside bordered by woods was small (mean = 3.25 units, median = 1.5 units, with a minimum of 1 unit and a maximum of 30 units). In NE, rangeland bordered the road in a few large sections (mean = 15.3 units, median = 14.5 units, with a minimum of 7 units and a maximum of 27 units). Over the years studied, the average KS precipitation from 1 April to 15 May was 174.7 mm (SE 36.2, range 42.9–312.8) while average NE precipitation was 7.62 mm (SE 1.73, range 3.1–11.0).

Occupancy, apparent colonization and apparent extinction

The proportion of occupied units and the mean number of plants per occupied unit was higher in NE than KS with the exception of 2002 in NE, a year with unusually low spring precipitation (Table 1, Fig. 1). Despite data from only 4 years, precipitation in early spring was significantly positively correlated with the total number of plants observed on the NE route; no pattern was evident for KS (Fig. 2). In KS, roadside units adjacent to grasslands had higher occupancy than roadsides adjacent to woods; differences were not seen between NE landscape types (Table 2, Fig. 3a,c). Previous soil disturbance was often positively associated with higher occupancy (Table 2, Fig. 3b,d).

Figure 1.

 Number of plants per 80-m unit along the survey routes, where 0 refers to the start of the route and 298 (KS) and 233 (NE) refer to the end, measured in 80-m units. Data for one of the roadsides for (a) KS 2001, (b) KS 2002, (c) KS 2003, (d) NE 2001, (e) NE 2002 and (f) NE 2003 are presented as examples. Note variation in the y-axis scale.

Figure 2.

 Total number of plants summed over both roadsides in the survey route versus the total precipitation (mm) during the period of seedling germination (1 April to 15 May, mm). Open circles: NE (F1,2 = 32.62, P = 0.03, R2 = 94.2%); closed circles: KS (F1,4 = 0.51, P = 0.51, R2 = 11.2%). NE data are also presented in the insert.

Table 2.   Logistic regression analyses on occupancy in KS and NE. Occupancy was the dependent variable, with year, landscape, disturbance and number of plants in the previous year and 2 years earlier as the independent variables. Analyses were done only for KS 2001–2004 (defined years 1–4) and NE 2003–2004 (defined years 1–2) since we lacked data from the previous 2 years for KS 1999–2000 or NE 2000–2001. Interactions with year were included to evaluate the degree to which effects were year-dependent. Analyses were done for each side of the road to adjust for autocorrelation (roadside number noted in parentheses). Z statistics are presented, with values adjusted for the first order partial autocorrelation coefficient (see text for details). For each variable, effects are interpreted relative to the lowest value (i.e. for disturbance, the significance of ‘1’ refers to differences between the low (0) and medium (1) level of disturbance, and ‘2’ refers to the difference between the low (0) and high (2) level of disturbance). Significant values are noted in bold, with symbols indicating levels of significance (#0.05 < < 0.1; *< 0.05; **< 0.01; ***P < 0.001). The letter ‘a’ refers to variables that could not be included in particular analyses because the statistical model did not converge
PredictorSite (roadside)
  1. †In KS, years 1–4 refer to 2001–2004; in NE, years 1–2 refer to 2003–2004.

 N(t − 1)1.85#1.250.651.01
 N(t − 2)2.25*2.59**1.582.18*
Year × Landscape
 2 × 21.44#0.520.880.29
 3 × 20.71−0.48
 4 × 2−0.69−0.05
Year × Disturbance
 2 × 11.101.37a1.22
 2 × 21.92#2.92**a2.00*
 3 × 11.70#1.64
 3 × 21.572.31*
 4 × 11.192.03*
 4 × 21.71#3.65***
Year × N(t − 1)
Year × N(t − 2)
Figure 3.

 Proportion of units occupied for KS and NE routes depending on the landscape adjacent to the roadside unit and the disturbance level. SEs are indicated, followed methods in Agresti & Coull (1998). Landscape comparisons are shown for KS (a) where adjacent landscape was woods (black) or grassland (white) and NE (c) where adjacent landscape was cropland (black) or rangeland (white). Disturbance comparisons are shown for KS (b) and NE (d). Disturbance categories are arranged by increasing levels from left to right, with lowest white and highest black. Data for both roadsides per route are combined for presentation.

In KS, higher probabilities of apparent colonization were consistently found in units adjacent to grassland compared with those bordered by woods (Fig. 4a); differences were not statistically significant after considering the spatial autocorrelation of landscape types (Table 3). Apparent colonization was positively associated with greater disturbance in KS (Fig. 4c, Table 3). Apparent extinctions in KS were more likely to occur in units bordered by woods and in the least disturbed units (Table 3, Fig. 4). Few landscape or disturbance patterns were apparent in NE (Table 3). KS units that lost above-ground plants had a smaller mean and median initial population size than units in which populations persisted (see Table S1 in Supporting Information: test of means using paired t-test, t4 = −2.99, = 0.04; test of medians: sign test, = 0.06). In two of the three NE years, apparent extinctions commonly occurred in populations with 10s to 100s of plants.

Figure 4.

 Proportion colonized and extinct for KS depending on landscape adjacent to the roadside unit and disturbance. All data refer to apparent colonizations and apparent extinctions (see text), since seed banks may occur. SE are indicated, following methods in Agresti & Coull (1998). Proportion colonized (a) and proportion extinct (b) depending on whether adjacent landscape was woods (black) or grassland (white). Proportion colonized (c) and proportion extinct (d) depending on disturbance level. Disturbance categories are arranged by increasing levels from left to right, with lowest white and highest black. Data for both roadsides per route were combined for presentation.

Table 3.   Landscape and disturbance effects on apparent colonization and apparent extinction in KS and NE. Partial Mantel tests were done for each side of the road to adjust for autocorrelation (roadside number noted in parentheses). Correlation coefficients are presented; significant values are noted in bold, with symbols indicating levels of significance (#0.05 < < 0.1; *< 0.05; **< 0.01; ***P < 0.001). The values before the slash (/) refer to values without adjustment for multiple tests; the values after the slash are adjusted for multiple tests using the Dunn–Sidak method, with NS = not significant (Gotelli & Ellison 2004). See text for details. Dashes indicate analyses that could not be performed because of the very small number of extinctions that occurred in a particular year [two extinctions for 2003(1) and three extinctions for 2003(2) in NE]

Occupiable habitat

Approximately one-fourth of the 298 KS units were never occupied in any of the 6 years (Fig. 5a). The never-occupied units for one side of the road had lower disturbance over the survey period (based on a combined disturbance index over the 6 years) compared with units that did have plants (Z = 2.50, P < 0.05); no significant pattern was seen for the other side. Never-occupied units were significantly more likely to be bordered by woods, with average woodland percentages of 44.2% compared with 15.7% woodland over the entire survey route (Z = 3.62, P < 0.001 and Z = 3.88, P < 0.001, respectively, for each roadside).

Figure 5.

 Proportion of the total roadside units that were never occupied versus the cumulative number of years surveyed. Black and white dots refer to data from the two roadsides; there were 298 and 233 total units per roadside, respectively, in the KS and NE routes. (a) KS, (b) grassland-adjacent KS units (total units = 256 and 249 for each roadside, respectively) and (c) NE. Nonlinear regressions using number of unoccupied units provided a good fit for KS data (roadside 1: y = 71.96 + 153.6e−0.63x, F2,3 = 178.5, = 0.0008; roadside 2: y = 69.54 + 155.4e−0.65x, F2,3 = 546.31, = 0.0001). In KS, the projected number of never-occupied units after 10 years of surveys [based on nonlinear regressions (a)] would be 72 (24.2%) and 70 (23.5%) for roadsides 1 and 2, respectively. Nonlinear regressions for KS grassland (b) were roadside 1: y = 42.79 + 145.4e−0.65x, F2,3 = 233.9, P =0.0005; roadside 2: y = 33.72 + 137.8e−0.65x, F2,3 = 582.7, P =0.0001. The projected number of never-occupied units after 10 years was 43 (16.8%) and 34 (13.3%), for each roadside, respectively.

Given the association between low occupancy and adjacent woodlands (Table 2, Fig. 3a), we excluded the woodland units and examined the remaining 256 units with an adjacent grassland landscape type (Fig. 5b). The percentage of never-occupied units with adjacent grassland after 6 years averaged 16.7%. Most (83%) of the 84 never-occupied grassland units had plants occurring in an adjacent unit in at least one of the years of the survey. For comparison, we examined a random sample of 84 units that were occupied in at least 1 year. A significantly higher percentage (96.4%) of units in this random sample had plants occurring in adjacent units at some point in the survey period (χ2 = 7.9, d.f. = 1, P = 0.005). Thus, never-occupied units showed a tendency towards being spatially isolated.

In NE, the average percentage of units never-occupied was lower than in KS after only 4 years of surveys (14.0%) (Fig. 5c). There was no significant difference between occupied and never-occupied units in the adjacent landscape. Never-occupied units were less likely to be disturbed (= 3.10, < 0.01 and = 2.11, < 0.05 for the two roadsides). There was no suggestion of an asymptote (Fig. 5c) so nonlinear regression was not performed.

Sunflower dynamics and the role of seed banks

Several lines of evidence are consistent with an important role of seed banks in sunflower dynamics. First, stretches of roadside with 100s to 1000s of plants in 1 year (t) could have virtually no plants the year before (t − 1), but large numbers 2 years earlier (t − 2) (Fig. 1d–f). Numbers of plants in year t − 2 were often significant predictors of occupancy in year t (Table 2). Second, although most apparent colonization occurred near units that were occupied the previous year, for two NE years (2002–2003, 2003–2004) and one KS year (2003–2004), >25% of the apparent colonizations were at least 160 m from the nearest possible source unit the previous year (see Fig. S1). Given typical seed dispersal distances (<10m, D. Pilson & H. M. Alexander, unpubl. data.), seed bank colonization is more plausible than long-distance movement. Third, true colonization events resulting from dispersal should be expected to have small numbers of plants, while seed bank colonizations could potentially be of any size. However, there was no difference in mean or median population size for isolated newly colonized units versus isolated units that had persistent populations for the five KS years (Table S1; test of means using paired t-test: t4 = −0.81, = 0.46, test of medians: sign test, = 0.5). Statistical analyses could not be done with only 3 NE years. Note, however, that newly occupied units in NE in 2003 could have 100s to 1000s of plants (Fig. 1e,f). Finally, with the exception of the high apparent colonization probabilities in NE in 2003 (following the low occupancy drought year of 2002), we saw no pattern relating the probability of apparent colonization to occupancy in the previous year, as expected if seed bank colonization is common (see Fig. S2).


Geographic variation

Even though the KS and NE study sites are separated by 600 km and differ in average annual precipitation by more than twofold, sunflower roadside dynamics share basic similarities. At both sites, for example, soil disturbance is associated with sunflower ecology and seed banks appear prevalent. Sites differed, however, in the landscape matrix and whether it was associated with sunflower dynamics. In NE, for instance, all landscapes were open (crop, range) and had little apparent effect on roadside dynamics. KS and NE sites also differed in the relationship between precipitation and plant numbers. In the low-rainfall year in NE, precipitation was apparently inadequate for germination or seedling establishment. When rainfall is sufficient, it is likely that large numbers of sunflowers result because of the large seed banks that accumulated in favourable years. Further, the relatively sparse and open roadside vegetation in NE is associated with high seedling survival and thus large numbers of adult plants in high seedling years (D. Pilson, unpubl. data.). In KS, it is likely that all years had adequate rainfall for germination and establishment. KS roadside vegetation also has a dense structure; thus high rainfall likely leads to a more competitive environment for sunflowers and reduces the ability of sunflower numbers to respond to changes in precipitation. The higher precipitation in eastern KS also allows trees to grow. The KS study area thus consists of a mosaic of open grasslands and woods, with roadside sunflower dynamics depending on the identity of the adjacent landscape. Occupancy and local densities in KS were also typically lower than in NE, and populations were less continuous along the verge. The physical array of KS roadside populations is consistent with the classic metapopulation concept because of the juxtaposition of suitable and less suitable habitat patches. However, in contrast to metapopulation theory, seed dispersal between suitable patches does not appear to dominate dynamics. Instead, local processes like emergence from the soil seed bank appear likely. We found, for example, that an isolated apparent colonization event had, on average, the same number of plants as a population that persisted, and that the numbers of plants present at a unit 2 years earlier was often predictive of the occupancy of the unit.

We thus suggest that roadside populations in NE (and to a lesser extent in KS) exist as spatially extended populations (sensuFreckleton & Watkinson 2002), where plants exist in a largely continuously suitable habitat and local processes predominate and can be scaled up to predict regional abundance and dynamics. Although we have not explored such scaling-up, it is noteworthy that Moody-Weis et al. (2008) found sunflower apparent colonization and apparent extinction could be ‘scaled down’ from a 640-m to a 80-m spatial scale using a subset of these data. We hypothesize that regional dynamics of H. annuus in more heavily forested regions of eastern North America would be closer to non-equilibrium dynamics given the isolated nature of H. annuus populations (Moody-Weis 2006). Dynamics west of our NE site, where rainfall declines further, may be even more strongly driven by seed banks and critical periods of precipitation, similar to many desert annuals (Venable 2007).

Occupiable habitat

Defining suitable habitat is a challenge for studies of plant regional dynamics. In NE, occupancy was high, and virtually all areas of the roadside verge had plants in at least 1 year. However, inter-year variability and the shorter number of survey years made it harder to explore if any units appeared ‘unoccupiable’. In KS, a quarter of the units were never occupied, and the data suggested that increased years of sampling would be unlikely to alter the results. Such units were, on average, bordered by woods and had low disturbance, providing information on what is needed for suitable habitat. Although not surprising for an annual, these results, and knowledge of which units were never occupied, would allow us to ‘target’ particular locations for future seed or transplant studies to quantify ‘occupiability’ (Ehrlén & Eriksson 2000; Mildén et al. 2006). It is also noteworthy that never-occupied grassland-adjacent units tended to be more isolated from seed sources. Such a result could suggest dispersal limitation, but such patterns could also result from clustering of suitable units across the landscape.

Sunflower dynamics and the role of seed banks

Directly quantifying seed banks on a geographic spatial scale is logistically challenging, if not impossible. The few metapopulation studies that focus on the role of seed banks consider systems (such as ant mounds) where patches are less than a metre wide (Dostál 2005). Our roadside surveys, however, allowed us to make inferences about the importance of seed banks for sunflower populations on a scale of kilometres. The NE data provide the most striking evidence. For example, a 4.5 km distance of roadside (units 21–76, Fig. 2d–f) had 100% occupancy in 2001, with a total of 58 365 plants. This same stretch had 4% occupancy in 2002 (2 units occupied, each with a single plant). However, in 2003, the stretch had 100% occupancy with a total of 39 191 plants, with individual units having 100s to 1000s of plants. Clearly the 2003 plants came from seeds produced in 2001 or earlier. Given the high occupancy and densities of plants in NE, we expect that seed banks are prevalent along roadsides in most years.

In KS, we lack a year of extreme low occupancy in the middle of the time series, making it harder to quantify the importance of seed banks. We expect that in KS (and in NE in other years), apparent colonizations occur through both seed dispersal and germination of buried seeds. Most apparent colonizations occur near potential sources of seeds in the previous year, as consistent with the importance of seed dispersal. However, in three site–year combinations, 25% or more of the apparent colonizations occurred 160 m or more from a source population, while seed dispersal gradients for individual plants rarely exceeded 10 m (D. Pilson & H. M. Alexander, unpubl. data.). Similarly, the lack of difference in mean and median population size for newly colonized versus persistent populations is consistent with seed bank emergence but not with seed dispersal. Finally, apparent colonization was decoupled from the prevalence of the plant in the previous time period, as expected with a major role of seed banks (Gotelli 1991).

Theoretical work suggests that seed banks are important in population and metapopulation persistence, especially in habitats with considerable temporal and spatial environmental heterogeneity (Kalisz & McPeek 1993; Perry & Gonzalez-Andujar 1993; Claessen, Gilligan & van den Bosch 2005). We suggest a combination of empirical and statistical approaches to refine our knowledge of the relative contributions of seed dispersal versus seed banks to colonizations. For the former, one could experimentally manipulate soil in isolated areas where seed dispersal is not likely and observe if plants emerge. For the latter, statistical models together with long-term survey data could explore if a ‘colonization’ event is better predicted by the temporal history at a particular site (i.e. numbers of plants in years − 2, − 3, etc.) or by the physical distance between a site and nearby populations the previous year (− 1) (J. Antonovics, pers. comm.).

The disappearance of above-ground plants, termed apparent extinctions in this study, is consistent with the role of stochastic processes in population dynamics, i.e. smaller populations were more prone to extinction. Similar patterns have been seen for other plant species (Antonovics et al. 1994). Higher probabilities of apparent extinction, however, were also observed in areas with low disturbance. We note, for example, that although it is easy to establish experimental populations (Cummings & Alexander 2002), these populations quickly disappear without additional disturbance. Thus, unlike classic metapopulation predictions, the loss of H. annuus and other disturbance specialists (e.g. Crawley & Brown 2004) is often caused by deterministic processes such as secondary succession (Johnson 2000; Bossuyt & Honnay 2006).

Our multi-year data set at two locations provides rich information on dynamics. However, any study of colonization and extinction must be reported with caveats. For example, long-distance seed dispersal could occur through movement of road maintenance vehicles. We, however, do not expect a major role for machinery; these rural roadsides were not heavily maintained. We also cannot eliminate the possibility of off-road populations, or populations along nearby roads, contributing to roadside dynamics. Our general knowledge of the landscape suggests that such effects are minimal, but we recognize this is a potential problem for interpretation. In other studies, off-road populations did not appear to influence roadside dynamics (Crawley & Brown 2004). Documenting extinction is always a challenge: Kéry et al. (2006) emphasized that <100% detectability of plants in surveys can lead to biased estimates of extinction parameters. Although we lack the data to calculate detection probabilities, we deliberately counted plants when they were large and flowering, and covered nearly all the routes on foot. Finally, we note that our data, and thus our inferences, are restricted to roadsides, which are a subset of all sunflower habitats. Shuster et al. (2005) found, however, that a random and a roadside sampling scheme were both effective in distributional studies of the exotic Alliaria petiolata.


The extensive network of roads (>6.3 million km in the US alone) can create conservation challenges: roadsides limit wildlife dispersal and are a conduit for the spread of invasive exotic plant species (Forman et al. 2003). However, given that such networks exist, ecologists should take advantage of the fact that roads provide ready-made transects through diverse geographical regions. Roadside surveys have already been used to document the spread of exotics (Wilcox 1989; Shuster et al. 2005); expansion to multiple regions across a species range would allow analysis of geographic variation in dynamics (i.e. Gaston 2003) and provide crucial monitoring data for control programs.

Roadsides are also important habitat for many wild relatives of crop plants, and thus are likely locations for crop–wild hybridization. In the case of sunflower, it is known that crop genes persist in wild populations (Whitton et al. 1997; Linder et al. 1998) and insertion of crop transgenes can increase seed production in wild H. annuus (Snow et al. 2003). USA companies are currently not using transgenes, but this could change over time. However, as a general question, concerns about transgene ‘escape’ often revolve around the question ‘if a transgene that increased seed production became fixed in a species, would we expect increased regional plant abundance and distribution?’ (Ellstrand 2003; Pilson & Prendeville 2004). Spread of a wild plant with a trangene would depend on key variables examined in our study, including the likelihood of seed bank formation, the relationship between geographic variation in precipitation and plant numbers, and whether suitable, but unoccupied habitat exists. For example, although NE occupancy often exceeded 75%, we might expect even higher occupancy could occur given that no asymptotic limits were apparent. However, F1 crop–wild hybrid seed typically have less dormancy than wild seed (Mercer, Shaw & Wyse 2006), so early generation hybrid seed would be less likely to accumulate in a seed bank.

Conclusions and future directions

Habitat structure and population dynamics of wild sunflowers varied at our two geographically distinct study sites. Populations in KS existed in a matrix of suitable and unsuitable habitat patches while the NE site often had continuous sunflower populations. Between years, variation in population size was largely driven by spring precipitation in NE, but not in KS. Landscape characteristics were an important determinant of within-year variation in KS occupancy, but such patterns were not evident in NE. At both sites, soil disturbance and the seed bank had important effects on apparent colonization. The patterns we observe at this large spatial scale are not easily open to experimentation. However, extensive spatial and temporal data such as ours could be integrated with spatially explicit models to explore the importance of species’ traits (e.g. seed dormancy) or hybridization on regional distributions (e.g. Antonovics 1999; Claessen, Gilligan & van den Bosch 2005). In such an approach, empirical data to parameterize and validate models should ideally come from multiple locations in a species range. By incorporating information about geographic variability in population dynamics, as well as habitat and landscape variables in such models, we can better evaluate the generality of theoretical predictions, and more broadly explore the population dynamics of species ranges.


We particularly appreciate insights of J. Antonovics, C. Cummings, A. Snow and M. Tourtellot in the development of this study. C. Cummings was instrumental in establishing the KS survey route, and many students, too numerous to name, assisted in data collection in KS and NE. The staff of the University of Kansas Field Station and Ecological Reserves provided KS weather data. We appreciate comments on earlier versions of this manuscript by J. Antonovics, F. Ballantyne, C. Collins, K. Mercer, and members of H.A’s laboratory group. Funding was provided by USDA grants 9904008 and 9601405 and USDA-CSREES 2006-39454-17438, by a University of Kansas General Research Fund grant (2301446), and by the University of Nebraska Layman Fund.