Restoration success varies based on time since restoration in a disturbance-dependent ephemeral wetland ecosystem

Introduction

Ecological restoration of habitats in disturbance-dependent ecosystems remains a challenge (Suding 2011). Restoration programs often focus on an idealized definition of high-quality habitat and assume postrestoration habitat conditions are either static over time or steadily improving toward some preferred state (Matthews et al. 2009). However, disturbance-dependent ecosystems are highly dynamic and, due to natural successional processes, over time restored sites may decline in quality and show increased variability postdisturbance (Hobbs & Harris 2001; Suding et al. 2004; Atkinson et al. 2022). Many ecosystems experience natural degradation–improvement–succession cycles initiated by disturbance events such as fires, beavers creating and then abandoning ponds, hurricanes forming forest gaps, and periodic grazing by migratory ungulates (Suding et al. 2004). Restoration that aims to mimic disturbance through multiple cycles of restoration over time often fails to revert habitat to idealized standards; as a result, successful restoration of disturbance-dependent ecosystems over the long term can be elusive or transient (Young et al. 2015; Abella et al. 2020; Collins et al. 2021).

One of the most challenging disturbance-dependent ecosystems to restore is temporary or ephemeral wetlands (Middleton 1999; Calhoun et al. 2017). Many ephemeral wetlands require regular disturbances to retard successional forces, such as woody encroachment (Conway et al. 2010; Calhoun et al. 2017). The restoration of such ecosystems requires extensive knowledge about how local communities respond to disturbance, and whether restoration actions that reestablish specific targeted species or overall biodiversity are effective in the long term (Moreno-Mateos et al. 2012). After the initial restoration occurs, subsequent restoration that is either too soon or too late in the successional cycle may impede recovery of target species. Striking the right balance in timing is critical for both successful recovery and cost-effective management, but where that balance lies is unknown for many targets of restoration (Brudvig 2011). Implementing restoration at the right time is critical given that many ephemeral wetland ecosystems that depend on disturbance also harbor rare or threatened species (Gorman et al. 2013; Klaus & Noss 2016; Baecher et al. 2018) or high levels of biodiversity (Seabloom & van der Valk 2003; Mitchell 2016).

We conducted a decade-long restoration experiment to assess the temporal variability of disturbance-dependent ecosystems in response to restoration, with a focus on ephemeral wetlands that support the only known populations of St. Francis' satyr (Neonympha mitchellii francisci), a U.S. federally endangered butterfly. The degradation of St. Francis' satyr habitat, combined with increased human activity in the region, have necessitated management intervention to ensure butterfly persistence (Cayton et al. 2015); however, it is unclear how often that intervention should occur. Because habitat conditions are naturally short-lived across the landscape, multiple cycles of restoration are likely crucial to recovery success (Fig. 1). In the case of St. Francis' satyrs, temporal shifts in host plant population size and butterfly population numbers may mirror each other, creating the need to better assess and monitor target species as conditions change postrestoration. Many at-risk butterfly species are particularly dependent on habitat maintained by disturbance (Haddad 2018; Schultz et al. 2019), which, even though it initially reduces populations, is required for butterfly persistence (Weiss 1999; Johst et al. 2006; Warchola et al. 2018).

Following several years of declining St. Francis' satyr population size, we began efforts across multiple streams to restore wetland habitat and determine how butterflies and their host plant populations respond to restoration over a multiyear time scale. Although the region includes disturbance-dependent ephemeral wetlands that support several species of conservation concern (Erwin et al. 2016), to date few studies have looked at how invertebrate communities might respond over time to restoration activities in these wetlands (Schowalter 2012). Previous work on St. Francis' satyr has shown that the creation of restored habitat patches has tripled the amount of butterfly host plants (Aschehoug et al. 2015), and that understanding butterfly response to restored habitat is critical for creating and maintaining sustainable populations (Cayton et al. 2015; Henry et al. 2019). With this experiment, we demonstrate how restoration impacts are not consistent across time, and that planning for multiple restoration cycles is critical for managing disturbance-dependent ephemeral wetlands for long-term success.

We asked the following questions: (1) does restoration have a significant effect on the size of butterfly host plant populations; (2) how does the effect of restoration on plant population size change over time since restoration and number of restoration cycles; and (3) how does butterfly population size change as a function of plant population size?

Methods

Results

Both sites had a significant interaction between the effects of treatment and time on sedge cover, but patterns of response to treatments differed. At site A, average percent cover of all Carex species across time since restoration was significantly higher in restored plots than in control plots (F_[1,31] = 22.428, p < 0.001) (Fig. 2A), with percent cover changing nonlinearly with time since restoration (F_[4,139] = 28.126, p < 0.001) and a significant interaction between treatment and time (F_[4,139] = 2.547, p = 0.04). At site D, average percent cover of all Carex species did not differ significantly between restored and control plots (Fig. 2B); however, percent cover changed nonlinearly with time since restoration (F_[4,138] = 4.247, p = 0.002), and there was a significant interaction between treatment and time (F_[4,138] = 7.666, p < 0.001).

The temporal dynamics of sedge cover differed across sites. Total Carex at site A averaged 11% in control plots (n = 225, range = 0–60, SD = 11.1) and 21% in restored plots (n = 543, range = 0–80, SD = 17.4) across all years. The highest average percentage of total Carex at site A occurred 2 years after restoration in control plots (n = 54, mean = 19, SD = 12.9) and 2 years after restoration in restored plots (n = 116, mean = 27, SD = 3.7) (Fig. 2A). Total Carex at site D averaged 12% in control plots (n = 375, range = 0–90, SD = 11.8) and 19% in restored plots (n = 1,110, range = 0–95, SD = 17.3) across all years. The highest average percentage of total Carex at site D occurred 6 years after restoration in control plots (n = 26, mean = 21, SD = 20.6) and 2 years after restoration in restored plots (n = 216, mean = 28, SD = 19.6) (Fig. 2B). These overall patterns shifted for the two key Carex species, with highest percentage of C. mitchelliana in restored plots occurring 4 and 9 years after restoration in sites A and D, respectively, and highest percentage of C. atlantica in restored plots occurring 2 years after restoration for both sites A and D (Supplement S1; Fig. S2).

The top model (ΔAIC ≤2) selected for predicting total amount of Carex species included re-restored status, and time since restoration as both a linear and quadratic variable (Table S1). This pattern held true regardless of site. When the main and random effects from the top model were plotted, peak Carex amounts varied according to restoration cycle: the top model predicted that restored plots would reach a maximum of 17% sedge cover approximately 2.3 years postrestoration (Fig. 3A) while re-restored plots would to reach a maximum of 12% sedge cover approximately 2.4 years postrestoration (Fig. 3B).

St. Francis' satyr PY index across time since restoration showed differing patterns between restored and control plots at the site-level: control and restored plots were significantly different at site A (F_[1,4] = 17.692, p < 0.001) (Fig. 2C), although not at site D across all years (Fig. 2D). PY index at site A averaged 1.3 in control plots (n = 14, range = 0–4.2, SD = 1.4) and 5.7 in restored plots (n = 47, range = 0–32.6, SD = 7.6). PY index at site A peaked 4 years after restoration in control plots (n = 2, mean = 3.9, SD = 0.4) and 4 years after restoration in restored plots (n = 5, mean = 10.8, SD = 6.0) (Fig. 2C). PY index at site D averaged 0.1 in control plots (n = 23, range = 0–0.5, SD = 0.1) and 0.5 in restored plots (n = 92, range = 0–7.9, SD = 1.4). PY index at site D peaked 6 years after restoration in control plots (n = 2, mean = 0.3, SD = 0.1) and 4 years after restoration in restored plots (n = 10, mean = 1.1, SD = 2.2) (Fig. 2D).

The top models (ΔAIC ≤2) selected for predicting PY index of butterflies all included flight period, total sedge, and year (Table S2). Butterfly counts maintained a positive correlation with total sedge for both periods but were expected to be higher in the second flight period (Fig. 4A) than in the first flight period (Fig. 4B). Site and the interaction between flight and year had weak or equivocal influence on butterfly counts, and no factors strongly predicted excess zero counts.

Discussion

Our use of a carefully designed experiment that explicitly tests the difference between treatment and control over a decade of management actions provides direct evidence that the impacts of restoration in the ephemeral wetland ecosystem are significant and change over time. This evidence supports our conceptual model of temporal shifts in species responses, and allows us to determine the extent of system stability and evaluate how management practices can better adapt to changing conditions over longer time frames. Restored plots contained higher amounts of sedge than control plots in all years postrestoration at site A and initial years postrestoration at site D, and butterfly populations in restored plots remained larger than in control plots for almost all years at both sites. However, peak levels of sedge and butterflies seen in early years postrestoration decline in later years for both sites, with both sedge amount and butterfly populations decreasing or approximating levels seen in control plots. This shift provides a template for planning management response for disturbance-dependent ecosystems in general and highlights how long-term monitoring is necessary to accurately assess the impacts of restoration (Schultz & Crone 1998; Brudvig 2011; Wilsey 2021).

The status of a plot as being restored for the first or second time is also significant, which suggests that not only time since restoration matters for predicting sedge amount, but also the number of cycles of restoration that have occurred. Predictions of sedge amount split by restoration status show that plots on their second restoration cycle are likely to have peak sedge amount at the same time as plots on their first cycle, but that peak is likely to be lower. This may be due to ecological processes, such as changes in soil composition or hydrology over time that impact sedges (Allessio Leck & Schütz 2005; De Steven et al. 2010), or possibly differences in methodology, in that restoration in the second cycle is less intensive in removing hardwoods (particularly woody undergrowth) in an effort to protect already established sedges. Regardless of the underlying mechanisms, this result has long-term implications for management: if subsequent restoration actions are increasingly less effective than the initial action, then restoration as a management strategy for improving habitat quality may have a time limit on how long it can be useful. Additional strategies, such as a greater focus on preserving abiotic factors in the soil (Morris et al. 2013) or community involvement that prevents wetland degradation (Scholte et al. 2016; Sinthumule 2021) may need to be integrated into planning to ensure sustainable habitat improvement.

Predicted butterfly population sizes reflect this temporal trend of sedge growth. The most important factor in explaining butterfly numbers after flight period is amount of sedge cover, followed by year. Butterfly population size increases directly in response to increased amount of overall sedge, with higher numbers seen during the second flight period when St. Francis' satyrs are known to be more abundant (Kuefler et al. 2008). Effective management for St. Francis' satyrs will likely need to focus on maintaining habitat with abundant sedges through regular hardwood removal at least every 4 years to ensure that butterfly populations do not drop below sustainable levels. The variability seen in butterfly population size over time underscores the dynamic nature of the disturbance-dependent ecosystem and highlights the regularity of population fluctuations that necessitate continual management intervention.

Examples of at-risk butterflies that inhabit disturbance-dependent ecosystems are numerous; however, studies on the effects of habitat restoration on these butterfly populations rarely use both experimental framework and a long-term approach (e.g. Weiss 1999; Warchola et al. 2018; Henry et al. 2020). Our use of a long-term experimental design provides a novel result for understanding how butterfly populations may fluctuate over time, and does so in a more robust way than the tradition of comparing restoration results to either dynamic reference sites or historical knowledge (Balaguer et al. 2014; Higgs et al. 2014; Atkinson et al. 2022). Determining the effectiveness of restoration requires several years of monitoring, especially with at-risk butterflies, where a short time span of only one flight period or one season is unlikely to accurately assess critical shifts (Himes Boor et al. 2018). Continuous monitoring over several seasonal cycles can best determine focal species response to shifts in host plant populations and population trends.

The inconsistent effect of time on the focal species and its host plants highlights the complexities of managing habitats where variability is integral to species' persistence. Key to successful management is the ability to regularly monitor both habitat conditions and target species population size, and adapt management goals to current conditions across long time frames and according to widely accepted restoration standards (Palmer et al. 2005). This applies across numerous types of disturbance-dependent ecosystems that benefit from management intervention, such as fire-adapted ecosystems (Warchola et al. 2018) or storm-influenced habitats (Henry et al. 2020). Ephemeral wetlands, where complex hydrology and high biodiversity can greatly impact conservation priorities, represent the most iconic of this type of habitat under threat (Calhoun et al. 2017). The long-term (decade or longer) restoration of wetlands can lead to mixed results, in some cases creating wetlands comparable to natural reference sites (Hopple & Craft 2013), and in other cases showing degradation over time that creates suboptimal habitat conditions (Matthews & Spyreas 2010). Our results show that patterns of overall host plant amount may mask variation in individual species trends, with adaptive management plans needing to account for differences in how each target species may respond over time. Because of this, long-term monitoring is needed to assess the extent to which restoration actually mimics beneficial disturbance events and when peak habitat quality occurs.

The importance of long-term monitoring was emphasized by site differences in both sedge growth and butterfly response. While site A showed a clear pattern of significantly more sedge at restored sites than control sites across all years, site D was less consistent in sedge population trends. The sedge response at site D may be due to several factors: more pronounced effect of spillover of sedge seeds between restored and control plots (Craig et al. 2011; Blitzer et al. 2012); differences in local microclimate or abiotic factors that dampen the effects of restoration at a site level (Raney et al. 2014); or differing initial seed banks that lead to altered conditions between sites (Donath et al. 2003; Allessio Leck & Schütz 2005). These site differences extended to butterfly populations, with site D supporting a smaller population and showing less consistent population trends, particularly once it has been several years since restoration. Regardless of underlying mechanisms, these differences underscore the necessity of conducting continual, long-term monitoring in order to discern local patterns of restoration response across any type of disturbance-dependent ecosystem.

Here we have focused on the temporal variability of restoration in disturbance-dependent ecosystems; however, one limitation to our study is the exclusion of spatial variability. It is unlikely in many cases that all habitat can be restored at once, as a strong source population is often needed to successfully repopulate restored habitat (Sundermann et al. 2011; Aavik & Helm 2018; Lanta et al. 2020). This creates a trade-off between spatial and temporal restoration, where only a portion of degraded habitat can be restored at any given time in order to leave some target populations intact while restoration activities take effect. In many cases, both spatial and temporal restoration need to be weighed in order to optimize restoration success (Wilson et al. 2011). Future research should examine whether restoration of a fraction of plots within the landscape at staggered time intervals can increase habitat quality while simultaneously maintaining focal species populations for recolonization.

Successful restoration is rarely achieved with a “one and done” approach, especially in ecosystems where temporal variation in habitat quality due to disturbance is an inherent characteristic. Long-term monitoring and active, adaptive management are integral to ensure restoration success across dynamic landscapes and should be better integrated into planning (Valkó et al. 2021; Wilsey 2021). With a clear timeline for determining how restoration activities are effective, it is easier to fully understand which actions influence population size, and therefore assess other critical conservation measures such as demographic parameters and species behavior that better inform management plans.