Convergence and divergence in plant community trajectories as a framework for monitoring wetland restoration progress


Correspondence author. E-mail:


1. Environmental policies that use ecological restoration to offset the destruction of natural ecosystems assume that restorations readily compensate for the losses because they progress reliably and predictably over time, following deterministic successional trajectories. However, succession and restoration are spatially and historically contingent processes, often characterized by divergent trajectories that deviate from expectations.

2. We develop a framework for monitoring restorations that integrates two ideas from succession theory: convergence vs. divergence in species composition among successional sites, and progression towards vs. deviation from an expected community state. We apply this framework to short- and long-term monitoring data from 11 restored wetlands in Illinois, USA, by comparing plant species composition among restored wetlands over time and between restored wetlands and two sets of reference wetlands (high integrity, ‘target’ wetlands and low integrity, degraded wetlands).

3. Over the first 4 years, restored wetlands that were initially similar in species composition diverged, progressing towards different high integrity target states. Planting a large number of native species in restorations increased their similarity to reference wetlands.

4. Over longer time scales (5–11 years post-restoration), however, restored wetlands deviated from the ideal trajectory and converged upon the species composition of degraded wetlands, mainly because of non-native species invasion.

5.Synthesis and applications. Framing restoration trajectories in terms of compositional convergence/divergence and progression towards/deviation from an acceptable range of reference sites is useful for monitoring restoration progress, identifying constraints to success and predicting restoration outcomes. Barriers to restoration, including non-native species and a lack of native propagules, can limit long-term progression towards target communities and constrain restoration to undesirable outcomes. Furthermore, convergence of restored wetlands on an undesirable community state limits the effectiveness of wetland mitigation policies.


Restoration goals are often embodied as a reference ecosystem, which guides restoration practice and provides a target species composition against which a restored site can be judged. Restoration progress towards the reference species composition can be measured as the increase in species similarity between a restored site and one or more reference sites (e.g. Mitchell et al. 2000; Holl 2002; Meyer, Whiles & Baer 2010). Specifically, progression can be quantified as a decrease in a dissimilarity metric, or ordination distance, between restoration and reference sites over time. This may be more meaningful than many commonly used univariate measures, which often misrepresent restoration progress (Matthews & Endress 2008; Matthews, Spyreas & Endress 2009c).

Environmental policies often dictate ecosystem restoration or creation to offset the destruction of natural ecosystems (e.g. compensatory wetland mitigation in the USA; Corps & EPA 2008). These policies assume that given enough time restorations compensate for losses, because once established, restorations will progress reliably. This notion of steady progression towards the species composition of a reference ecosystem is reminiscent of the classic deterministic model of vegetation succession, which predicted that following a disturbance, a site would pass through an orderly sequence of stages to arrive at a stable climax community (Clements 1936). Early models of succession also predicted that initially dissimilar successional communities would become more similar to each other over time, converging on the plant species composition of the singular climax type (Lepš & Rejmánek 1991; Peet 1992). Successional case studies failed to support these predictions suggesting instead that forces leading to divergence among sites and deviation from the expected climax are often greater than those leading to convergence upon a single climax (Matthews 1979; Christensen & Peet 1984; del Moral 2007).

Long-standing ideas of convergence and divergence among successional sites and progression towards a climax community have new practical relevance for ecological restoration. Restoration ecologists and practitioners have essentially replaced the notion of a stable climax with an idealized and desirable reference ecosystem. This leads to practical questions such as: In what situations might restorations progress towards the desired reference target? How can we assess whether restorations are moving towards targets? What can be done to actively encourage this progression?

To better address these questions, we developed a framework for monitoring restorations that integrates two key ideas from succession theory: convergence vs. divergence in species composition among successional sites, and progression towards vs. deviation from a desired reference state. Given a set of restored and reference communities, we can expect one of four possible successional patterns (Fig. 1). First, we might observe simultaneous convergence among restored sites and progression towards the species composition of the reference targets (Fig. 1a). This may occur in situations where abiotic conditions (e.g. light and nutrient availability) in restorations become more similar to conditions in mature reference ecosystems over time. Initial species composition may vary among sites, but restoration should be relatively easy because successional pathways eventually converge upon the same target (e.g. Mitsch et al. 1998). This scenario is most likely where initial abiotic variability among restored sites is minor, initial abiotic conditions are already similar to those of the target and propagules of target community species are readily available to colonize (Temperton & Hobbs 2004; White & Jentsch 2004). In previous studies where one or more of these conditions was not met, restorations often lingered in an intermediate, possibly transient, state between the initial and target states (e.g. Galatowitsch & van der Valk 1996; McLachlan & Knispel 2005; Fagan et al. 2008). In this case, decreasing site isolation from propagule sources, increasing planting effort or allowing more time might increase similarity to targets.

Figure 1.

 Given a set of restoration sites (grey circles) and reference targets (squares), four patterns are possible for temporal trajectories (arrows) in species composition: (a) convergence in species composition among restorations and progression towards the target composition, (b) divergence among restorations and progression towards a spectrum of potential targets, (c) convergence among restorations but deviation away from the intended reference targets, (d) divergence among restorations and deviation from intended targets.

Secondly, we might observe divergence among restorations and progression towards a spectrum of equally acceptable targets (Fig. 1b). Each restoration might diverge from all others or there may be a pattern of overall divergence but local convergence within subsets of restorations that trend towards a few alternative states (Woods 2007; Houseman et al. 2008). Vegetation is likely to diverge among sites if the same group of early colonizing, ubiquitous species dominate recently restored sites, but species establishing later respond to environmental differences among sites (Lepš & Rejmánek 1991). Alternatively, species composition may be initially similar if the same restoration techniques are applied at every site, for example where the same set of species are planted in all restorations regardless of whether those species are well-suited to a site’s specific environmental conditions. In either case, restorations will have similar initial species compositions but will diverge over time as plant communities respond to underlying environmental gradients among sites. Several studies suggest that this is a common successional pattern (Matthews 1979; Pineda et al. 1981; Christensen & Peet 1984; Seabloom & van der Valk 2003; del Moral 2007).

Thirdly, we might observe convergence among restorations, but deviation from desired reference communities (Fig. 1c). In this case, all restorations might converge upon a similar species composition but one that differs from the expected or desired composition (e.g. Wilkins, Keith & Adam 2003; Norman et al. 2006; Aronson & Galatowitsch 2008). Trajectories may be nonlinear, initially deviating from the expected target but eventually progressing towards it. Alternatively, the lack of progression may be permanent, revealing the importance of historical contingency (Matthews et al. 2009a). If the historical climate, disturbance regimes, land use patterns and species pools under which the reference communities assembled were markedly different from present-day conditions under which the restorations are developing then it may be impossible to recreate the set of conditions that would result in the targeted outcome. Instead, restorations would deviate towards historically novel communities that reflect modern conditions (Anderson, Schwegman & Anderson 2000; Ejrnæs et al. 2008).

Finally, we might observe a pattern of both divergence among restorations and deviation from the desired target composition (Fig. 1d). This pattern might be expected where species composition in restored communities is largely determined by stochastic events such as unpredictable hydrological regimes or long-lasting priority effects caused by chance colonization events (Egler 1952; Lockwood 1997; Chase 2007; Trowbridge 2007). In this case, the outcome of restoration would be largely unpredictable and potentially difficult to control.

Characterizing the successional dynamics for a set of restoration sites is essential for determining whether we can rely on passive restoration approaches to achieve restoration goals, whether more intensive intervention is required to achieve goals, or whether chosen restoration targets represent unrealistic expectations (Prach & Hobbs 2008; Suding & Hobbs 2009). We apply the framework depicted in Fig. 1 to vegetation succession in 11 restored wetlands and two sets of reference wetlands with the following objectives: (i) determine whether successional trajectories in restored wetlands were convergent or divergent, (ii) determine whether vegetation composition progressed towards the composition of natural reference wetlands of either high or low ecological integrity, (iii) relate the degree of progression towards reference wetlands to site isolation and planting effort, and (iv) determine whether trends observed over a short-term monitoring period (4 years) continued as the restorations aged.

Materials and methods

Study sites were wetlands restored by the Illinois Department of Transportation (IDOT) between 1993 and 2005 as mitigation for wetlands damaged during road construction. We have previously reported on trajectories of commonly used univariate measures of restoration progress, such as species richness, in these sites (Matthews, Spyreas & Endress 2009c). Here, we describe trajectories of multivariate species composition in 11 sites for which we have plant abundance data for 4 years. Restorations were located throughout Illinois, USA, from 38·8° to 42·5° latitude, and varied in size from 1·0 to 7·1 ha. All sites were initiated from bare ground, and restoration techniques varied among sites. Wetland hydrology was re-established through a combination of excavation with topsoil replacement, construction of berms or weirs to hold water and/or removal of artificial drainage features. Sites received between 0 and 48 native wetland plant species, introduced as seed and/or rootstock.

Species composition of natural, reference wetlands was obtained from the Illinois Critical Trends Assessment Program, which sampled herbaceous wetlands throughout Illinois (Molano-Flores 2002; see also Matthews, Spyreas & Endress 2009c). The ecological integrity of each wetland was graded on a scale of decreasing integrity from grade ‘A’ to ‘E’. Grades were assigned to sites based on their accumulated anthropogenic disturbance (e.g. historical farming, partial drainage, impoundment, herbicide application or intensive livestock grazing; White 1978). We specifically chose reference sites to represent both high and low ecological integrity wetlands. These included 23 grade A or B wetlands that represented the highest integrity wetlands in Illinois and served as ideal targets for wetland restoration. In addition, we randomly selected 23 wetlands from a set of 70 severely degraded, low integrity, grade D or E wetlands to represent undesirable outcomes for wetland restoration. Groundwater-fed fens and seeps, wetlands fringing lakes and wetlands south of 38·3° latitude were excluded because they were not representative of restorations in this study.

Herbaceous-layer vegetation in each restoration was quantitatively sampled for four consecutive years following site construction. Vegetation in each reference site was sampled once between 1998 and 2004. In each restored and reference site, vascular plants were sampled using at least twenty 0·25-m2 quadrats, evenly spaced along transects placed to span the hydrological gradient. All species observed in each quadrat were assigned cover class values (<1, 1–5, 6–25, 26–50, 51–75 or 96–100%). To equalize sampling effort across sites, we randomly selected 20 quadrats from each site in each year for analysis.

To determine if vegetation tended to converge (become more similar) or diverge (become more dissimilar) among restorations over time, we calculated Bray–Curtis dissimilarity, using species relative cover values, for pairs of restored wetlands, separately for each year. Bray–Curtis dissimilarity is one minus percentage similarity and varies from 0 (where two sites share no species in common) to 1 (where all species are shared and have equal abundance in the two sites; Lepš & Šmilauer 2003). Because the 55 site-to-site comparisons in each year were not independent, we used a randomization procedure analogous to a repeated measures one-way anova (Edgington 1995) to determine the effect of year on dissimilarity values. Dissimilarities were permuted 10 000 times among years, within site pairs, using the program Resampling Procedures v. 1·3 (D.C. Howell, University of Vermont, Burlington, VT, USA). The observed F-statistic for the effect of year on dissimilarity was then compared to the distribution of F-statistics generated under a null hypothesis of no effect of year to calculate a P-value.

To determine if restorations became more similar to high integrity references over time, we calculated Bray–Curtis dissimilarity between each restoration and every high integrity reference wetland, separately for each year the restoration was surveyed. Within each year, the 23 dissimilarity values were averaged for a total of 11 mean dissimilarities (one mean value for each restoration) in each of 4 years. We again used a randomization procedure to permute average dissimilarities among years (10 000 times), within each restoration, to determine the effect of year on average dissimilarity from the high integrity references. We repeated this procedure to determine whether the restorations became more similar to the low integrity references. Because reference sites were surveyed only once each, these analyses examined change in restorations relative to a stable reference baseline (the average reference composition). However, because we used several reference sites surveyed over multiple years, our analysis did not rely on an assumption that reference composition was static.

Vegetation composition in restored and reference wetlands was displayed using non-metric multidimensional scaling (NMDS) of Bray–Curtis dissimilarity matrices with the WinKyst 1·0 add-in for CANOCO 4·5 (Biometris, Wageninen, The Netherlands). The iterative NMDS algorithm was repeated 50 times, with a different starting configuration each time, to avoid convergence on a suboptimal solution. A three-dimensional solution was chosen based on a scree plot of the number of dimensions vs. stress (badness-of-fit). Species contributing strongly to dissimilarity between restorations and reference sites were identified using the procedure of Clarke (1993) for determining the relative contribution of each species to the average Bray–Curtis dissimilarity between two groups of sites.

We examined the influence of propagule availability on the dissimilarity between restored and reference wetlands using two variables related to propagule supply: number of species initially planted at restorations and isolation from nearby wetlands. For each restored wetland, we quantified site isolation as mean distance to the nearest three wetlands mapped by the National Wetlands Inventory (NWI, U.S. Fish and Wildlife Service) after verifying the existence of NWI wetlands using aerial photographs and field visits (see Matthews et al. 2009a). Mean dissimilarity, averaged across 4 years, between the restored and reference wetlands was regressed on number of planted species and log-transformed isolation using SYSTAT 11 (Systat Software, San Jose, CA, USA).

Eight of the 11 restored wetlands were resurveyed in 2006 as part of another study (Matthews et al. 2009a). At this time, restored wetlands ranged from 5 to 11 years since restoration. Vegetation was sampled, as above, in forty 0·25-m2 quadrats. We use data from 20 randomly selected quadrats in each of the eight sites to determine if the trends observed over the first 4 years continued as the restorations aged.


Short-term monitoring

Vegetation among restorations became less similar over 4 years (Fig. 2a). Although the trend was not statistically significant, it suggested divergence rather than convergence in species composition (repeated measures randomization test for decreasing similarity, = 2·32, = 0·07). Over the same time period, vegetation in restorations became more similar to vegetation in reference wetlands (Fig. 2a). However, restorations simultaneously became more similar to both high and low integrity reference wetlands (Fig. 2a; increasing similarity to high integrity references: = 9·10, = 0·0001; increasing similarity to low integrity references: = 3·12, = 0·02).

Figure 2.

 Mean (±SE) Bray–Curtis dissimilarity between the vegetation of restored and high integrity reference wetlands (open squares), mean dissimilarity between restored and low integrity reference wetlands (black diamonds) and mean dissimilarity among restored wetlands (grey circles). Mean values are displayed for 11 restored wetlands over the first 4 years since restoration (a), and for 8 restored wetlands that were resurveyed in 2006 when they ranged in age from 5 to 11 years (b).

Non-metric multidimensional scaling confirmed these patterns (Fig. 3a and b). Restored and natural wetlands were largely separated along the first NMDS axis (Fig. 3a), and restoration site scores across this axis increased with site age over 4 years (Spearman rank correlation: rS = 0·42, = 0·002). Therefore, the first axis probably reflected differences in successional age among sites. High and low integrity reference wetlands were well-separated along the second NMDS axis (Fig. 3a). Reference site position along this axis was correlated with an independent measure of ecological integrity, mean Coefficient of Conservatism (rS = −0·81, < 0·001), which is based on whether a site’s species are indicative of high integrity natural communities (Taft et al. 1997). Site position along the second axis was also closely correlated with cover by Phalaris arundinacea L., an invasive, non-native grass (rS = 0·75, < 0·001). Therefore, we interpreted the second NMDS axis as a gradient of ecological integrity among wetlands. Restoration site scores decreased along the second NMDS axis over 4 years for 10 of the 11 restorations. Thus, even though restorations were more similar, on average, to the low than the high integrity reference wetlands in all years (Figs 2 and 3a), trajectories across the second axis of the NMDS plot indicated that most restorations trended towards high rather than low integrity references over the first 4 years.

Figure 3.

 Non-metric multidimensional scaling (NMDS) of species composition in high integrity reference sites (open squares), low integrity reference sites (black diamonds), restored wetlands over the first 4 years since restoration (grey circles) and restored wetlands in 2006 (grey stars). Larger circles represent restored wetlands of increasing age, and years within a site are connected by lines to illustrate trajectories. Plots of NMDS axes 1 and 2 (a) and axes 1 and 3 (b) are shown (stress = 0·17).

Divergent trajectories among restorations were evident along the third NMDS axis (Fig. 3b). Initially, similar restored wetlands approached different reference community states over time (see Fig. 1b). This separation was not towards high vs. low integrity wetlands because reference sites were not differentiated along the third axis according to ecological integrity. Longitude and latitude were not significantly correlated with reference site position along the third NMDS axes (latitude: rS = 0·18, = 0·23; longitude: rS = 0·09, = 0·57), suggesting that the observed divergence was not driven by differences in regional species pools. Instead, site position along the third axis may have been related to an underlying hydrological gradient among sites, as site position along this axis was correlated with weighted wetland indicator status (rS = 0·57, < 0·001), an indicator of site hydrology based on plant species’ affinity for wetland vs. upland habitats (Atkinson et al. 1993).

Annual species (e.g. Echinochloa spp. P. Beauv., Persicaria lapathifolia [L.] S. F. Gray and Persicaria pensylvanica [L.] Small) and non-native perennials (e.g. P. arundinacea and Typha angustifolia L.) were characteristic of restorations, distinguishing them from high integrity reference wetlands (Table 1). Echinochloa spp., Persicaria spp. and T. angustifolia also distinguished restorations from low integrity reference wetlands (Table 2). Phalaris arundinacea was a characteristic species of low integrity references (Table 2 and Fig. 4), and its cover in restorations increased through time (Fig. 4).

Table 1.   Species distinguishing restored from high integrity reference wetlands, based on contribution to average Bray–Curtis dissimilarity between restored and reference wetlands (mean δ%)
YearRestoredMean δ%High integrity referenceMean δ%
  1. The 10 species with the highest mean δ% are shown for each of 4 years.

  2. Echinochloa spp. included Echinochloa muricata and E. crus-galli.

  3. *Annual species; †non-native species.

1Echinochloa spp.*6·7Carex stricta6·1
Phalaris arundinacea3·5Persicaria amphibium5·8
Persicaria pensylvanica*2·4Lemna minor3·1
Typha angustifolia2·4Leersia oryzoides3·1
Persicaria lapathifolia*2·2Bolboschoenus fluviatilis2·3
2Echinochloa spp.*4·4Carex stricta6·1
Phalaris arundinacea4·2Persicaria amphibium5·8
Leersia oryzoides4·0Lemna minor3·0
Typha angustifolia2·2Sagittaria latifolia2·1
Persicaria lapathifolia*2·1Bolboschoenus fluviatilis1·9
3Leersia oryzoides5·5Carex stricta6·2
Phalaris arundinacea5·2Persicaria amphibium5·9
Echinochloa spp.*3·3Lemna minor3·5
Eleocharis acicularis2·3Bolboschoenus fluviatilis2·3
  Sagittaria latifolia2·2
  Schoenoplectus acutus1·9
4Phalaris arundinacea6·6Carex stricta6·2
Leersia oryzoides5·2Persicaria amphibium6·0
Typha angustifolia2·9Lemna minor3·5
Eleocharis erythropoda2·7Bolboschoenus fluviatilis2·8
Iva annua*2·5Sagittaria latifolia2·2
Table 2.   Species distinguishing restored from low integrity reference wetlands, based on contribution to average Bray–Curtis dissimilarity between restored and reference wetlands (mean δ%)
YearRestoredMean δ%Low integrity referenceMean δ%
  1. The 10 species with the highest mean δ% are shown for each of 4 years.

  2. Echinochloa spp. included Echinochloa muricata and E. crus-galli.

  3. *Annual species; †non-native species.

1Echinochloa spp.*6·9Phalaris arundinacea27·4
Typha angustifolia2·6Amaranthus tuberculatus*4·0
Persicaria pensylvanica*2·4Bolboschoenus fluviatilis2·3
Persicaria lapathifolia*2·3Elymus virginicus1·8
Panicum dichotomiflorum*2·1  
Setaria faberi*†2·1  
2Echinochloa spp.*4·7Phalaris arundinacea27·6
Leersia oryzoides2·9Amaranthus tuberculatus*2·6
Typha angustifolia2·5Bolboschoenus fluviatilis1·9
Persicaria lapathifolia*2·2Elymus virginicus1·8
Setaria faberi*†1·8  
Eleocharis acicularis1·7  
3Leersia oryzoides5·0Phalaris arundinacea27·6
Echinochloa spp.*3·8Bolboschoenus fluviatilis2·3
Eleocharis acicularis2·4Amaranthus tuberculatus*2·3
Aster lanceolatus2·2Elymus virginicus1·8
Typha angustifolia2·2  
Bidens frondosa*1·7  
4Leersia oryzoides4·5Phalaris arundinacea27·5
Typha angustifolia3·3Bolboschoenus fluviatilis2·9
Echinochloa spp.*2·7Amaranthus tuberculatus*2·5
Iva annua*2·7Elymus virginicus1·9
Eleocharis erythropoda2·5  
Aster lanceolatus2·1  
Figure 4.

 Mean (±SE) Phalaris arundinacea relative percentage cover in restored wetlands of different ages (year 1 through year 4, = 11; year 5–11, = 8), low integrity reference wetlands (low reference, = 23) and high integrity reference wetlands (high reference, = 23).

Vegetation became more similar to both high and low integrity reference wetlands as the number of native species planted in restorations increased (high integrity references: β = −0·001, r= 0·67, F1,9 = 18·6, = 0·002; low integrity references: β = −0·004, r= 0·37, F1,9 = 5·3, = 0·05). However, degree of isolation from nearby natural wetlands was unrelated to average dissimilarity to references (high integrity references: β = 0·003, r= 0·08, F1,9 = 0·8, = 0·40; low integrity references: β = 0·012, r= 0·06, F1,9 = 0·6, = 0·47). Restoration site area also had no significant effect on average dissimilarity to references (high integrity references: β = −0·001, r= 0·03, F1,9 = 0·2, = 0·64; low integrity references: β = −0·013, r= 0·08, F1,9 = 0·8, = 0·40).

Longer term monitoring

Additional data, beyond the initial 4 years, for 8 of the 11 restorations indicated that the trend of increasing similarity to high integrity wetlands ended, whereas the trend of increasing similarity to low integrity wetlands continued unabated (Fig. 2b). Furthermore, site trajectories across the NMDS plot suggested that the restorations, which trended towards high integrity references over the first 4 years, strayed from this ideal trajectory by 2006 and began to approach the species composition of low integrity references (Fig. 3a). Specifically, site scores along the second NMDS axis increased for seven of eight restorations after the fourth year. Therefore, although we observed a pattern of divergence and progression towards targets over the first 4 years (Fig. 1b), we observed a pattern of weakening divergence and increasing deviation from targets over a longer timeframe (Fig. 1c and d).


Vegetation trajectories over 4 years

Species composition became more similar between restored and reference wetlands over 4 years, whereas over the same time period composition tended to diverge among restorations (see Fig. 1b). Successional sites with initially similar floras can diverge over time if early successional species have broader niches than late successional species (Christensen & Peet 1984). Early colonizing species may be similar across restorations, regardless of site environmental conditions, but over time the unique environmental conditions of sites become increasingly important. We have previously identified gradients in hydrology and fertility as important in differentiating vegetation among restored wetlands (Matthews et al. 2009a). Eventually species may be limited to those sites with environmental conditions where they are effective competitors (Matthews 1979; Pineda et al. 1981; Christensen & Peet 1984; Inouye & Tilman 1988). Other potential causes of divergence include stochastic variation in species composition or differences in initial restoration techniques that become amplified over time.

In situations where restorations diverge towards different community states, it may be difficult to define a precise target in advance, and restoration practitioners should allow for a range of acceptable outcomes (Temperton & Hobbs 2004). In contrast to the pattern observed over the first 4 years in this study, wetland mitigation programmes frequently employ ‘one size fits all’ restoration techniques that homogenize the wetlands of a region by replacing wetlands of diverse community and habitat types with a standard type, such as a pond with a fringe of Typha spp. L. cattails (Bedford 1999; Brooks et al. 2005). Divergence among restored wetlands towards acceptable, alternative targets should therefore be encouraged because it maintains habitat heterogeneity across a landscape.

For any restoration monitoring programme the observed degree of progression towards target communities will depend in part on the choice of reference sites (Matthews, Spyreas & Endress 2009c). The reference wetlands included in this study were varied, and restored wetlands became increasingly similar to the average reference composition. However, because we compared progress towards both high and low integrity wetlands, we were able to observe that even while becoming increasingly similar to target wetlands, the restorations grew increasingly similar to low integrity, undesirable wetlands. The species that best distinguished restored wetlands from both high and low integrity natural wetlands were early successional annuals (Tables 1 and 2). The progression towards reference wetlands appeared to be driven by the loss of these early successional species from restorations rather than by a gain of species typical of high integrity wetlands. The eventual decline of early successional and annual plants (Matthews & Endress 2010) generates an inevitable increase in similarity to established natural wetlands, be they of high or low ecological integrity. This should serve a note of caution for restoration ecologists and practitioners: evaluating restoration progress by measuring similarity against desirable reference communities alone can mask latent trajectories towards undesirable states.

The degree of progression towards reference vegetation varied among restorations, and the lack of progress in some sites may have been due to propagule limitation. Isolation from propagule sources can slow the rate of succession in wetlands (Bossuyt, Honnay & Hermy 2003), although we found no effect of distance from nearby wetlands in this study. Planting restorations with native species can facilitate progression towards restoration targets by overcoming seed limitation (Pywell et al. 2002; Galatowitsch 2006; Öster et al. 2009), enabling key ecological functions (Callaway, Sullivan & Zedler 2003; Smith et al. 2003) and inhibiting invasive species (Funk et al. 2008). However, other studies suggest that planted species sometimes hinder progress by preventing the establishment of desirable species (Holl 2002; Fagan et al. 2008). Here, planting a large number of native species had a positive effect on similarity to natural wetlands, probably by introducing species typical of mature wetlands that would not have otherwise colonized within 4 years.

Additional insights from long-term monitoring

Although restorations became increasingly similar to both high and low integrity references, trajectories displayed using NMDS indicated that restored wetlands more often progressed towards high rather than low integrity natural wetlands over the first 4 years, as indicated by decreasing site scores along the second axis (Fig. 3a). By 2006, however, trajectories had turned away from the high integrity targets and towards the low integrity sites. This shift was often associated with increasing dominance by P. arundinacea. Phalaris arundinacea may be the most prevalent and dominant non-native plant species in natural and restored wetlands in Illinois (Spyreas et al. 2004; Matthews et al. 2009b), and its dominance is associated with lower plant species richness and diversity (Spyreas et al. 2010). Dominance by P. arundinacea appears to be a stable condition maintained by a feedback mechanism, where dense P. arundinacea canopy growth monopolizes light, allowing greater production of leaves and leaf litter, which further reduces light and limits the establishment of other species (Zedler 2009). Invasion by highly dominant, non-native species such as P. arundinacea may lead to the eventual convergence of restorations on an undesirable, low-diversity, alternative state that is resistant to additional restoration efforts (see Fig. 1c). Thus, the initial divergence observed among restored wetlands may be transitory. Aronson & Galatowitsch (2008) also reported that species composition in restored wetlands converged over time as P. arundinacea invaded and expanded within sites. Convergence is likely where one or a few species dominate all sites (Inouye & Tilman 1988; Glenn-Lewin & van der Maarel 1992; but see Houseman et al. 2008), and convergence among restorations should be anticipated and avoided in community types and regions that are highly susceptible to invasive species.

In situations where restorations continually result in undesired community states, it may be possible to identify the particular biotic or abiotic constraints that prevent targeted outcomes and implement the management necessary for pushing sites across alternative state thresholds (Hobbs & Norton 1996; Suding, Gross & Houseman 2004; Cramer, Hobbs & Standish 2008; Martin & Kirkman 2009). However, where restoration of a particular target community composition is infeasible because constraints cannot be overcome, restoration practitioners may need to adjust their goals. For example, they might focus on the restoration of desired ecosystem functions that can be realistically achieved given current and projected future environmental conditions (Suding & Gross 2006; Choi et al. 2008).

Our previous studies have demonstrated that these restored wetlands often failed to comply with legal standards regarding plant species composition and non-native species presence (Matthews & Endress 2008). Although the restorations rapidly developed high species richness, univariate indicators of restoration progress based on species composition, such as cover by native plant species, often peaked and then declined (Matthews, Spyreas & Endress 2009c). Likewise, in this study vegetation appears to progress towards target species composition, only to later deviate towards a low integrity state. The eventual species composition in restored wetlands is determined by a combination of factors ranging from local hydrology and fertility to landscape elements that lie beyond the control of restoration practitioners, meaning that barriers to successful restoration occur at a variety of scales (Matthews et al. 2009a; Matthews & Endress 2010). As we demonstrate here, monitoring multivariate trajectories in species composition relative to high and low integrity reference sites can help identify these barriers to restoration progress.

The wetland permit programme established under the U.S. Clean Water Act allows wetland impacts to be offset through compensatory wetland mitigation (Corps & EPA 2008). This and other development policies that use ecosystem restoration or creation to offset the destruction of critical ecosystems often incorrectly assume that restoration can compensate for the losses through reliable, deterministic succession (Zedler 1996; Walker et al. 2009). The continuing failure of mitigation and compensation programmes to replace or re-create ecosystems of high ecological integrity will result in the replacement of regionally unique native ecosystems with homogeneous, low integrity restorations, leading to an overall net loss of regional biodiversity and ecosystem function even when the total area restored exceeds the area of natural ecosystem lost (Quigley & Harper 2006; Palmer & Filoso 2009).


Restored wetlands were constructed by IDOT and monitored by the Wetlands Group of the Illinois Natural History Survey (INHS). Reference sites were surveyed by Critical Trends Assessment Program (CTAP) personnel at INHS with funding from the Illinois Department of Natural Resources. Funding for 2006 site surveys was provided by IDOT and a National Great Rivers Research and Education Center grant to Anton Endress and J.W.M. Arun Soni, Patrick Baldwin and A. Endress assisted with field surveys.