Eric W. Seabloom, National Center for Ecological Analysis and Synthesis, 735 State Street, Suite 300, Santa Barbara, CA 93101, USA (fax 805 892 2510; e-mail firstname.lastname@example.org).
1The spatial structure of plant communities can have strong impacts on ecosystem functions and on associated animal communities. None the less, spatial structure is rarely used as a measure of restoration success.
2The restoration of hundreds of wetlands in the prairie pothole region in the mid-western USA provided an excellent opportunity to determine whether the re-establishment of abiotic conditions is sufficient to restore structure, composition and spatial patterning of the vegetation.
3We mapped the topography and vegetative distributions in 17 restored and nine natural wetlands. We used these data to compare the composition and spatial structure of the vegetation in both wetlands types.
4The composition of the plant communities differed between restored and natural wetlands; the restored wetlands lacked the well-developed sedge-meadow community found in most natural wetlands. However, the spatial heterogeneity was similar, although the zonation patterns were less well-developed in the restored wetlands.
5Although the overall structure was similar, species distributions differed among wetland types, such that species were found more than 10 cm higher in restored wetlands than in natural wetlands.
6Synthesis and applications. This study illustrates that restored plant community composition and spatial structure may converge on their targets at different rates. Evaluations of restoration success should consider spatial structure of communities along with compositional and functional metrics.
Despite the importance of restoration (Dobson, Bradshaw & Baker 1997), it is still unclear whether our current ecological understanding is sufficient to allow us to restore natural communities in areas that have been converted to human-dominated land uses (Zedler 2000). In part, the difficulty lies in determining what constitutes successful restoration. The criteria for determining the successful restoration of a plant community include re-establishment of vegetative cover plant productivity (Mitsch & Wilson 1996; Mitsch et al. 1998) to successful introduction of specific suites of species (Galatowitsch & van der Valk 1996a,b). However, most criteria are based on site-level measures, and there have been relatively few attempts to incorporate spatial structure as a measure of restoration success.
None the less, the spatial distribution of vegetation can have strong effects on ecosystem processes. Most of the studies of the relationship between restoration success and spatial structure have focused on the effects of landscape-scale configuration of sites. This large-scale configuration can affect the ecosystem functions of restored plant communities (Johnston, Detenbeck & Niemi 1990; Zedler 2000) and associated animal populations (Schultz & Crone 2001). Many animals use multiple habitats, and the degree to which these habitats are interspersed with one another can have strong impacts on populations dynamics and long-term viability (Dunning, Danielson & Pulliam 1992).
The development of zones or bands of vegetation along environmental gradients (coenoclines) is often the dominant factor controlling the spatial distribution of species (Whittaker 1960, 1967; Whittaker & Niering 1975; Spence 1982). While the importance of environmental gradients has been evident to community ecologists since the inception of the field (von Humboldt & Bonpland 1985, originally written in 1807; Shreve 1922), little is known about the actual processes that result in the formation of these gradients in natural systems, not to mention the less well understood restored systems.
The rate at which environmental gradients begin to dominate the spatial structure of plant communities can vary widely. In some cases, abiotic conditions can account for much of the variability in species composition, and coenoclines rapidly re-assemble following environmental fluctuations. For example, Seabloom, Moloney & van der Valk (2001) found that that in an experimental wetland complex, the effects of water depth overwhelmed the effects of historical recruitment events in 2–3 years. In this case, the length of the water level fluctuation was brief and there were ample propagules available in the soil seed bank. On the other hand, the findings of Werner & Platt (1976) suggest that in grassland communities this process occurs over the course of decades. Numerous other studies suggest that the ultimate composition of communities is more strongly determined by historical recruitment events and pre-emption than by environmental conditions (Grace 1987; Robinson & Dickerson 1987; van der Valk & Welling 1988; Drake 1991).
Determining the rate of coenocline formation is critical to planning successful restorations. If environmental gradients have strong and rapid impacts on species distribution, then the restoration of spatial patterns in restored communities will be a natural outcome of restoring the appropriate abiotic environment. On the other hand, if coenoclines develop slowly, the initial patterns created by historical recruitment events may be evident for long periods of time. In this case, it may be necessary to include spatial structure in restoration plans and to distribute plant species in spatial patterns that mimic those found in natural wetlands, a much more complicated process.
An unprecedented restoration effort in the prairie pothole region in the mid-western USA provided an excellent opportunity to determine whether the simple re-establishment of abiotic conditions is sufficient to restore structure, composition and spatial patterning of the vegetation. In this region, the hydrology of thousands of wetlands was restored decades after the areas were originally drained. In most cases, there was no attempt to introduce wetlands species into these potholes, and restoration entailed simply reflooding a particular basin by building a levee or blocking a drainage tile. For this reason, successful re-establishment of vegetation relied on natural colonization.
Another major advantage of the prairie wetland system is that plant distributions in freshwater wetlands are ultimately determined by a single environmental gradient (i.e. water depth) (Spence 1982), making it relatively simple to track the formation of coenoclines. In contrast, the distribution of species in many other systems may be determined by complex interactions between multiple gradients (e.g. water depth and salinity; Snow & Vince 1984; Vince & Snow 1984; Bertness & Ellison 1987).
In addition, the prairie wetland system is well suited to observational studies, because the restored and natural wetlands are discrete units of similar size and shapes that are well interspersed throughout the landscape. In contrast, many studies of restorations are anecdotal due to the lack of adequate replication.
In this study, we determined the degree to which the spatial patterning within restored prairie wetlands has become re-established in the absence of artificial seeding. To do this, we mapped the vegetation and topography of two types of prairie pothole wetlands: restored wetlands that were 5–7 years old and natural wetlands that were centuries old. We use these data to (i) determine whether the level of spatial heterogeneity in plant composition is similar in restored and natural wetlands; (ii) determine whether restored wetland plant communities have developed the distinctive zonation patterns found in natural wetlands; and (iii) test whether the species that have become established in restored wetlands are found over a similar elevational range in restored and natural wetlands.
In the process of performing these tests on restored prairie wetlands, we demonstrated the use of several statistical analyses that can be used to compare spatial structure in other types of restored plant communities. In our first analyses, we used a multivariate permutation test, the Mantel test, to compare the level of spatial heterogeneity independent of a specific environmental gradient. Plant–plant or plant–animal interactions can lead to the development of distinctive spatial patterns in plant communities that do not have strong abiotic gradients. We then used modifications of this test to incorporate explicitly the effects of environmental gradients on species distribution. While we specifically examined the effects of hydrological changes associated with a gradient in elevation, the approach we used is generally applicable to any system with strong environmental gradients such as salinity or fertility.
Materials and methods
Prairie pothole wetlands form in saucer-like depressions typical of the recently (12 000–14 500 years before present) glaciated regions of the Great Plains region in central North America (Kemmis 1991). One of the most striking features of these wetlands is the distinct vegetative zonation patterns that form perpendicular to the elevational gradient in each wetland (Stewart & Kantrud 1971). These concentric rings result from differential effects of flooding on species-specific rates of seed germination, seedling survival and adult mortality (van der Valk & Welling 1988).
We mapped the vegetation in nine natural and 17 restored prairie pothole wetlands during the summers of 1993 and 1994. These wetlands were small (mean area = 1 ha) with closed drainages, and there was no detectable difference in size or shape between the natural and restored wetlands (Seabloom & van der Valk, in press). The restored wetlands were between 5 and 7 years old at the time of sampling.
To make comparisons of species’ distributions among wetlands, we use the full-pool elevation as a consistent reference elevation. This was taken as the elevation of maximum flooding during the spring of 1993, a year of record flooding levels in the mid-western USA. In natural wetlands, this elevation was measured directly by placing survey stakes at the upper limit of standing water at peak flooding. In restored wetlands, this upper bound was often determined by the presence of water control structures, such as spillways or standpipes. In these cases, we used the elevation of the water control structure as the most accurate estimate of the full pool elevation. In all future discussions, we assign this high-water mark an elevation of 0·0 cm.
In each wetland, we placed stakes in a rectangular grid orientated to a random bearing. We adjusted the distance between these grid nodes on the basis of the size of the wetlands, so that there were approximately five sample points in each wetland zone (sensuStewart & Kantrud 1971). The distance between grid nodes ranged from 15 to 30 m and the number of nodes per wetland ranged from 11 to 42.
We surveyed the elevation (±1·0 cm) of each node in the grid using an optical transit (theodolite). Distances were determined using the stadia method (tacheometry). After completing the elevational survey, we recorded the percentage cover of all emergent plant species within a 1·0-m2 quadrat centred on each node. These cover estimates were based on six classes: < 1%, 1–4%, 5–24%, 25–49%, 50–74% and 75–100%. In the analyses, samples were assigned the cover from the mid-point of the range of the appropriate class. All plant identifications followed the nomenclature of Gleason & Cronquist (1991).
We used Mantel tests to investigate the relative importance of three sources of variability in community composition: among wetland type (restored vs. natural), among individual wetlands within a type, and among elevations within a wetland. For example, we wished to determine whether samples taken from very different elevations in a wetland differed in their plant composition. The Mantel test is a permutation test that calculates the correlation among the corresponding elements of two distance matrices (Livshits, Sokal & Kobyliansky 1991; Sokal & Rohlf 1995).
We chose to use the Mantel test because it is a multivariate technique that allows us to compare the change in overall community composition. In contrast, univariate tests require the use of metrics such as species richness or individual ordination axes. In addition, the test makes few distributional assumptions about the data, because significance is assessed using permutation methods.
We begin by describing the construction of each of the distance matrices. First, we constructed the community distance matrix, D. If we have n samples, we can construct an n × n matrix that contains the compositional distance between all pairs of the original n samples. This matrix is symmetrical and so can be displayed as a half matrix composed of n × n elements. Each element of D is calculated as:
where aik is the percentage cover of the kth species in the ith quadrat (Pielou 1984). Dij scales from 1 (when there are no shared species) to 0 (when cover values are exactly equal for all species). We tested alternate distance measures, such as correlation and covariance, and found no qualitative difference in the results.
We compared D to three n × n distance matrices that represented the effects of wetland type (natural or restored), individual wetlands within a type, and elevation within a wetland. We constructed a matrix of dummy variables, T, such that samples from the same type of wetland (e.g. Restored) had a distance of 0 and samples from wetlands of different types were assigned a distance of 1. Specifically:
Similarly, we constructed the individual wetland matrix, W, such that samples from the same wetland had a distance of 0 and samples from different wetlands were assigned a distance of 1. Specifically:
The elements of the elevational distance matrix, E, were calculated as the absolute value of the difference in their associated elevations, such that:
where ei is the elevation of the ith quadrat. In addition, we tested for an interaction between elevation and wetland type by conducting a separate Mantel test of the effects of wetland type for samples collected in eight elevational classes (−100 to −75 cm, −75 to −50 cm, −50 to −25 cm, −25 to 0 cm, 0–25 cm, 25–50 cm, 50–75 cm, 75–100 cm). In this way, we were able to determine if the differences among natural and restored wetlands were stronger at certain elevations.
The test statistic for the Mantel test (Z) is simply the product–moment correlation, r, between all corresponding elements of the two distance matrices (e.g. community composition and elevation). As is the case when calculating the correlation between two variables measured on independent random samples, Z ranges from −1 to 1. In the context of the Mantel test, significantly positive values of Z indicate that there is a positive correlation between the elements of the two distance matrices. For example, if large compositional distances correspond to large elevational distances, Z will be larger than average. The significance of the statistic Z is established by repeatedly permutating the rows and columns of one of the distance matrices and recalculating the test statistic. The true value of Z is compared with this distribution to determine if it is exceptionally large. We used Matlab to perform the Mantel tests (Anonymous 1992), and all significance tests were based on 999 permutations of the distance matrix.
While the Mantel tests can detect an overall relationship between community composition and elevation, we also wished to determine the range of elevations over which community composition remained similar. Are samples that are 10 cm apart much more similar than those that are 20 or 30 cm apart? This is a direct measure of the strength of the zonation patterns. If the zonation patterns are strong, community composition will change rapidly with elevation. We constructed a Mantel correlogram (Oden & Sokal 1986; Koenig 1999; Koenig & Knops 2000) using the community distance matrix, D, and elevation matrix, E. In this analysis, we divided the elevational distances into 11 distance categories by rounding the elements of the elevational distance matrix down to the nearest multiple of 10 cm (e.g. 0–9·9, 10–19·9, etc.). We simultaneously calculated Z within each distance category and plotted Z as a function of elevational distance. All distances of greater than 100 cm were combined into a single category, due to low replication at these larger distance categories. We calculated the Mantel correlogram separately for the natural and restored wetlands to compare the rate at which community composition changes along the elevational gradient in both wetland types.
The Mantel tests of the whole data set indicated that wetland species composition had a similar level of correlation with all three explanatory variables: wetland type, individual wetland variability and elevation (Table 1). The difference between the two wetland types reflected the suites of species that were missing from the restored wetlands. In Table 2, we show the proportional cover of the five most dominant species in restored and natural wetlands in 25-cm elevational increments. One of the key differences is the lack of sedges (Carex sp.) in the restored wetlands. In natural wetlands, there was at least one sedge species among the dominant species at all elevations less than 25 cm. In contrast, there were no sedge species among the dominants in the restored wetlands.
Table 1. Mantel test of multivariate plant community distance at three scales: elevation within wetland, among wetlands and among wetland types (i.e. natural and restored). Significance tests are based on 999 permutations of the community distance matrix. Mean distance among and within classes are shown for the matrices composed of nominal distances
Mean community distance
Table 2. Mean proportional cover of the five most abundant species in six elevational classes in restored and natural wetlands. An elevation of 0 cm is the point of maximum flooding in the wetlands (i.e. full-pool elevation). The lower bound of the elevational classes is listed in the table (e.g. the category > −100 contains samples collected between −100 and −50 cm in elevation
Largely S. canadensis.
Includes S. glauca, S. verticillata, and S. viridis.
We repeated the same tests separately for each wetland type, and found that variability among wetlands and elevation accounted for similar amounts of variability in natural and restored wetlands. In addition, the mean variability of samples taken within a wetland was similar across types, indicating similar levels of spatial heterogeneity (Table 1). We found some evidence that there was an interaction between wetland type and elevation, such that community composition in restored and natural wetlands was more similar at elevations greater than 50 cm (Table 3).
Table 3. Mantel test of multivariate plant community distance among natural and restored wetlands at eight elevational classes. The lower bound of the elevational classes is listed in the table (e.g. the category > −100 contains samples collected between −100 and −75 cm in elevation). Mean distance among and within wetland types are shown. Significance tests are based on 999 permutations of the community distance matrix
Mean community distance
Compositional variability increased with increasing elevation in natural wetland (Fig. 1). In contrast, compositional variability in restored wetlands showed a unimodal response to elevation. Community composition was similar among samples taken either deep in the restored wetland basins or from the surrounding mesic prairie. However, community variability in restored wetlands was similar to that of natural wetlands at moderate elevations.
development of zonation patterns
The Mantel correlogram was very similar in both wetland types; community composition was positively correlated among samples taken from similar elevations and negatively correlated among widely separated samples (Fig. 2). This general result confirmed that coenoclines were easily detectable in natural and restored wetlands. However, the structure of the coenocline was not as well developed in the restored wetlands. Community composition in natural wetlands was significantly positively correlated among samples that were separated by less than 30 cm elevation. In contrast, samples less than 60 cm apart were significantly positively correlated in the restored wetlands.
It should be noted that the magnitude of the correlations in the Mantel correlogram was fairly low because the effects of elevation were compared among samples collected at similar elevation in different wetlands. We calculated the distance in this way so that the strength of the elevational effect was comparable to that of wetland type and individual wetland variability. The absolute correlations were much higher when the analysis was restricted to a single wetland, but the replication was low within each distance class.
distribution of species
In addition to comparing the spatial pattern in restored and natural wetlands, we compared the distribution of individual species in both wetland types. For each of the species that were present in both restored and natural wetlands (n = 100), we calculated the mean and standard deviation of the elevation at which each species was found. We used the standard deviation as a measure of the elevational range of each species. The mean elevation of each species was 10·9 cm higher in restored wetlands (P < 0·001) based on a paired t-test. There was no detectable difference in the standard deviation of the elevations at which each species was found (P = 0·895).
It is clear from previous work that 2–3-year-old restored prairie wetlands do not resemble natural wetlands in terms of species richness and composition (Galatowitsch & van der Valk 1996a,b). We found that there were still large difference in species composition among restored and natural wetlands after 5–7 years. As noted by Galatowitsch & van der Valk (1996b), we found that these differences increased with elevation.
Despite these floristic differences, our analyses of spatial structure indicated that the spatial variability in restored wetlands was similar to that in natural wetlands, such that community composition had similar levels of variability within and among natural and restored wetlands. While overall variability was similar, the zonation patterns in the restored wetlands were less distinctive in restored wetlands. Species composition remained similar over larger segments of the elevational gradient in restored wetlands (< 60 cm) than in natural wetlands (< 30 cm). Bird species often have strong preferences for certain vegetative zones, and it is possible that the more homogeneous plant communities in restored wetlands may lead to lower avian diversity (Weller & Spatcher 1965; Delphey 1991).
Our results are similar to the pattern found by Werner & Platt (1976) in their study of coenocline formation in old-field communities. They compared the distribution of six species of perennial forbs (Asteraceae: Solidago sp.) in a prairie (the site of three of the natural wetlands in this study) and a 23-year-old abandoned field. They found a less distinct coenocline (i.e. there was more overlap among the species) in the old field relative to the prairie. Although the study was not replicated, this finding suggested that these differences were due to a successional process in which early distributional patterns reflect the pre-emption by early colonizers rather than environmental gradients. Over time, however, competition led to increased niche separation and more distinctive zonation patterns.
Restored wetlands also differed from natural wetlands in the degree of community variability at a given elevation in our study. At moderate elevations, near the full pool level, the community variability of restored wetlands was similar to that in natural wetlands. However, the restored wetland plant communities were more homogeneous than the natural wetland communities at low and high elevations. The homogeneity in restored wetlands reflects the dominance of two invasive species: Bromus inermis at high elevations and Typha× glauca at low elevations. These species often form nearly monospecific stands in the centres and at the perimeters of restored wetlands.
Only four species were found between −100 and −50 cm in restored wetlands, and Typha was by far the most dominant. In contrast, Typha was only the fifth most abundant species in natural wetlands, and its mean cover was half of that in restored wetlands. The invasive nature of Typha is not confined to the prairie wetland systems (Reinartz & Warne 1993), and has been referred to as the ‘cattailization of America’ (Odum 1988).
Species distributions were different in the two wetland types: species were found higher in the restored wetland basins than in the natural wetlands. One possible reason for species being found at higher elevations in restored wetlands is that they are not colonizing the lower bounds of their potential distributions. It is interesting to note that Werner & Platt (1976) also found species’ distributions to be clustered at the upper end of the moisture gradient. In contrast to their work, we were not able to detect a change in niche width between the natural and restored wetlands.
In addition to the differences in the location of species that were found in both natural and restored wetlands, we found that there were critical components of the wetland plant community that were missing in the restored wetlands (i.e. sedges). The low success of sedges in restored wetlands has been noted elsewhere (Galatowitsch & van der Valk 1996a,b) and has been cited as a potential cause for the low populations of certain components of the avian community, such as swamp sparrows Melospiza georgiana, common yellowthroats Geothlypis trichas and marsh wrens Cistothorus palustris (Delphey 1991).
Our results indicate that the spatial structure of restored wetlands communities may converge with that found in natural communities. If this is generally true, it may be unnecessary to manipulate within-wetland spatial patterns during restoration to match that found in natural wetlands, at least at moderate elevations. In contrast, more active management of the restored wetlands may be necessary to increase spatial heterogeneity at low and high elevations in those restored wetlands where a few invasive species dominate.
It is possible that careful management of water levels may eliminate the differences in species’ elevational distributions. Flooding inhibits the germination of many wetland plants, and these species rely on periodic drawdowns for recruitment (van der Valk & Davis 1978; Seabloom, van der Valk & Moloney 1998). It is possible that the artificial water control structures in restored wetlands cause water levels to be more constant (Shaffer, Kentula & Gwin 1999) and prevent colonization of the lower parts of the wetland basin. Periodic planned drawdowns may be a necessary component of long-term management of restored wetlands.
A more intractable problem is correcting the current low species diversity in restored wetlands (Galatowitsch & van der Valk 1996a,b). Even if the necessary abiotic conditions are re-established, natural colonization can be very slow due to depauperate seed banks and isolation of restored wetlands from sources of native seeds (Godwin 1923; MacArthur & Wilson 1967; Wienhold & van der Valk 1988; Reinartz & Warne 1993; Bakker et al. 1996). Historically, up to 60% of the prairie pothole region was covered by wetlands, while extant wetlands are typically isolated in a matrix of agricultural lands (Galatowitsch 1993). Furthermore, the presence of high densities of exotic species in natural and restored wetlands is likely to limit the spread of native species once the dispersal barrier is overcome due to pre-emption (Green & Galatowitsch 2001, 2002). It may be possible to reduce the prevalence of some exotic species by reducing nitrate loads to restored wetlands and curtailing the use of exotic species for soil conservation projects (Green & Galatowitsch 2002).
Special thanks are due to Kirk Moloney, Susan Galatowitsch, and Elizabeth Borer for assistance in project design and editorial input. In addition, we wish to thank James Krumm, Julia Tauber and Deven Nice for their assistance in the collection of field data. Funding was provided by the US Environmental Protection Agency, Iowa Lakeside Laboratory, as well as the Department of Botany, the Ecology and Evolutionary Biology Interdepartmental Program, and the Geographic Information Systems Support and Research Facility at Iowa State University. The writing and analysis for this research was while E. W. Seabloom was a Postdoctoral Associate at the National Center for Ecological Analysis and Synthesis, a Center funded by NSF (grant no. DEB-0072909), the University of California, and the Santa Barbara campus.