Severe ecosystem disturbances including species invasion, anthropogenic land use change, and extinction of carnivores have resulted in dramatic changes in ecosystem composition and losses of global biodiversity (Rambo and Faeth, 1999; Myers et al., 2000; Raghubanshi and Tripathi, 2009; Strong and Frank, 2010). Especially when disturbances directly impact the abundance and distribution of vegetation, their effects can propagate across ecosystems, reducing foodweb stability and resilience (Nakamura et al., 2000; Raghubanshi and Tripathi, 2009; Strong and Frank, 2010; Ripple and Betschta, 2012). Understanding both the scale and cascading impacts of disturbance across ecosystems, and exploring their mitigation is imperative to prevent the loss of unique species (Raghubanshi and Tripathi, 2009). Species loss, even of uncommon and weakly interacting species, can degrade an ecosystem's resiliency to disturbance and invasion, increase variability of ecosystem processes, and result in the loss of additional unique species (O'Gorman et al., 2010).

In the past few decades, white-tailed deer (Odocoileus virginianus) populations have expanded across the eastern United States, due to local extinction of natural predators, landscape and ecological changes, reduction in hunting pressure, and the ability of deer to adapt to human-modified ecosystems (Côté et al., 2004; Latham et al., 2005; Creacy, 2006; Rawinski, 2008; Strong and Frank, 2010). At high densities, deer alter community dynamics and ecosystem processes through trampling and selective browsing, especially of reproductive plant parts (Augustine and Frelich, 1998; Latham et al., 2005; Geddes and Mopper, 2006; Averill et al., 2017). Strong browsing pressure limits the reproductive potential of preferred plants (Nakahama et al., 2020) and can reduce the density and diversity of herbaceous plants over time, causing local extinctions (Augustine and Frelich, 1998; Strong and Frank, 2010). As a result, white-tailed deer are considered a keystone species, because their browsing has such strong direct and indirect impacts on the ecosystems they live in (Rooney, 2001; Rooney and Waller, 2003).

The effects of deer browsing on plant communities may be magnified in low productivity areas (Olff and Ritchie, 1998; Proulx and Mazumder, 1998; Bakker et al., 2006) where low soil nutrient availability increases the energetic costs of replacing lost plant tissue (Janzen, 1974). This may be especially true for serpentine barrens, which are globally rare, heavily fragmented habitats that occur on resource-poor soils derived from ultramafic rock (Latham and McGeehin, 2012). Due to the increased cost of regrowth in resource-poor environments, serpentine plant species have reduced resilience to herbivory (Coley et al., 1985; Lau et al., 2008; Strauss and Boyd, 2011).

In the eastern United States, serpentine barrens are biodiversity hotspots for rare endemic serpentine plant species, as well as many locally or regionally endangered species (Rajakaruna et al., 2009; Flinn et al., 2017). In recent decades, these ecosystems have been overwhelmed by increasing browsing pressure from white tailed deer (Prince et al., 2004; Floyd, 2006). Several endangered flowering forbs, including downy lobelia (Lobelia puberula) and white heath aster (Symphyotrichum ericoides) have been locally extirpated from eastern serpentine barrens in the past decades (Latham and McGeehin, 2012). While it is unclear that deer overbrowsing is directly responsible for these declines, it is clear that rare and endangered species highly preferred by deer are vulnerable. Latham and McGeehin (2012) suggest that any attempt to restore rare eastern serpentine grasslands without limiting the deer population is a “lesson in futility” and “the only practical method” to protect vulnerable endemic species is the complete exclusion of deer.

Herbivore exclusion, particularly deer exclusion, has been a common technique to restore rare plant communities worldwide (Rambo and Faeth, 1999; Ruhren and Handel, 2003; Stephan et al., 2017). This method of restoration has been widely successful at increasing herbaceous plant diversity for ecosystems that were overbrowsed by deer (Ruhren and Handel, 2003; Stephan et al., 2017; Nakahama et al., 2020). However, herbivore exclusion as a method of restoration may not always improve plant diversity. The efficacy of restoration by herbivore exclusion is dependent on the length and severity of herbivory disturbance (Nakahama et al., 2020). Areas that have experienced chronic browsing disturbance for decades may have depleted seedbanks and dwindling soil nutrients, and may not recover from disturbance, even when herbivores are excluded. Additionally, certain species, particularly those with energetically costly adaptations, are sensitive to disturbance and may not recover when herbivores are excluded (Tamura, 2010). Alternatively, herbivore exclusion may decrease biodiversity (Mortenson et al., 2017). If deer preferentially feed on highly competitive or exotic species, browsing may provide less competitive species an advantage. In this scenario, herbivore removal could destabilize plant communities by allowing dominance of one or a few competitive plant species (Mortenson et al., 2017). This may be particularly true in low productivity environments, characterized by plants with limited competitive abilities.

In order to assess serpentine ecosystem recovery from deer browsing, we deployed deer exclusion plots to quantify differences in flowering plant community composition and key response traits (total plant number per plot, total flower number per plot, average blossom area per plot, average vegetative area per plot, and average height per plant) between areas of deer presence and absence. Response traits describe how plants respond to changes in their environments (Violle et al., 2007). Plant community response to deer exclusion was assessed over the course of one growing season. In a recovery scenario, we expected increases in diversity and response trait values where deer were excluded. In a depleted seedbank scenario, we expected no difference between treatments. In a scenario where deer played a role in increasing plant diversity through indirect effects on plant competition, we expected lower plant diversity in excluded areas, and the proliferation and dominance of one or a few plant species. This research was conducted in an eastern serpentine savannah, a globally rare and highly fragmented habitat for which restoration from deer overbrowsing has not been attempted.

The total number of flowering plants on the landscape fluctuated over the course of the growing season, peaking during the late summer (Figure 1). NMDS was used to assess how plant community functional responses differed between control and exclusion treatments. Both the magnitude and variability of all plant response traits were higher in the exclusion treatment compared to the control treatment (dimensions = 2, stress = 0.02; Figure 2). Permanova results showed that plant functional responses were significantly different in exclusion plots relative to control plots [F_{(1, 8)} = 5.6, p = 0.02], and pairwise analysis showed that all plant response variables differed significantly between the two treatments (Supplementary Table 1).

Flowering plant count in exclusion plots was consistently greater than in control plots (X1,942 = 12.79, p < 0.001). Deer presence had a significant negative impact on all measured plant response variables: species and floral counts, vegetative area per plot, number of flowers per plant, and average plant height per species (Table 1, Supplementary Table 2). Based on NMDS analysis, flower species that responded most positively to deer exclusion were lyrate rockcress (Arabis lyrata) and serpentine aster (S. depauperatum) (dimensions = 2, stress <0.001; Figure 3, Supplementary Table 3). Shannon's Diversity did not vary between control plots and exclusion plots [F_{(1, 40)} = 0.38, p = 0.82].

This study examined serpentine plant community recovery after one year of deer exclusion. If deer exclusion promoted community recovery, we predicted excluded plots would exhibit increased diversity and response trait values in excluded plots. If deer browsing caused an increase in diversity by limiting the abundance of competitive species, we expected to find lower diversity in excluded areas, the proliferation of one or a few plant species and their trait values, and a decrease in trait values for all other species. If however, decades of deer browsing resulted in irreversible degradation of the seedbank, we expected no changes between treatments. Rather than finding strong support for one scenario, after only one year of data collection, we found evidence to support all three.

No differences in overall herbaceous plant Shannon's Diversity Index were observed between control and exclusion treatments. These results could be explained by a number of factors, including exclusion of annual seedlings by established perennials, changes in soil composition due to deer trampling, or a depleted seedbank at Soldiers Delight. Tamura (2010) reports that if deer overbrowsing remains unchecked for over 10 years, recovery may be a long and difficult process. Deer have existed at high densities (approximately 40/km² since the 1990s) around Soldiers Delight for several decades (Porter, 1991; Walters et al., 2016). Serpentine barrens are already vulnerable to herbivory pressures, due to their low plant productivity, low plant species diversity, and nutrient poor soils (Safford and Mallek, 2011; Strauss and Boyd, 2011; Flinn et al., 2017). As a result, chronic browsing pressure could have driven preferentially consumed species to extinction (Anderson et al., 2001; Latham and McGeehin, 2012). These results are consistent with studies of Midwest prairies, that have attributed low floral diversity to decades of intense deer browsing (Anderson et al., 2001). However, this research was only conducted for one year, and a continuation of this study might reveal different results.

Although plant diversity was unaffected by deer exclusion, flowering plant functional responses and abundances dramatically increased in exclusion plots. Throughout the growing season (Figure 2) and across the plant community (Figure 3), deer exclusion consistently increased values for all plant response traits measured (Table 1), including plant and flower count, vegetative area, flowers per plant, and plant height for herbaceous species. In other words, deer exclusion resulted in plant communities with traits indicative of greater competitive (increased height and percent cover) and reproductive ability (increased species and floral counts). These results are consistent with numerous studies that found deer exclusion dramatically increases the abundance and response trait values of flowering plants in prairie (Anderson et al., 2017) and temperate forest ecosystems (Ruhren and Handel, 2003; Wang and Mopper, 2008; Sakata and Yamasaki, 2015; Nakahama et al., 2020). Stephan et al. (2017) suggests that in addition to increases in height, size and abundance, plants can allocate more resources to defense against other herbivores in the absence of significant biomass loss from deer.

Although deer exclusion caused higher response values at the community level, not all plant species responded in the same way. Species most impacted by deer herbivory included A. lyrata and S. depauperatum. Exclusion had a strong positive impact on the height of S. depauperatum plants, and both the height and plant area of A. lyrata plants. At our sites, both species were perennials with long flowering times of up to 2 months. Plant area increased most for S. rugosa, and flower count increased most for O. biennis (Figure 3). The only species that did not follow the community trend of increasing abundance or trait values in the absence of deer was slender knotweed (Polygonum tenue). In control plots this species grew taller and was more abundant. The unexpected results from this species may reflect overall plant scarcity, or competitive exclusion of P. tenue when deer were absent. During the late summer when P. tenue was flowering, Liatris spicata was hyper-abundant in exclusion plots, at a density as high as 41 flowers per square meter. In control plots, L. spicata was never more abundant than 11 flowers per square meter. The dominance of L. spicata in exclusion plots may result in limited nutrients or space for P. tenue to thrive.

Serpentine barrens are considered to be zones of high conservation concern (Latham and McGeehin, 2012; Flinn et al., 2017). Of the flowering plants observed in this study, L. spicata, C. arvense and S. depauperatum are ranked as critically imperiled and highly rare in the state of Maryland (Maryland Natural Heritage Program, 2021). S. depauperatum is additionally classified as critically endangered and near-endemic to eastern serpentine barrens (Rajakaruna et al., 2009; Latham and McGeehin, 2012; Maryland Natural Heritage Program, 2021). If left unprotected, eastern serpentine grasslands may lose these unique rare species, and be permanently degraded by continued intense browsing pressure (Latham and McGeehin, 2012; Flinn et al., 2017). Although plant diversity did not increase in exclusion plots, and P. tenue performed worse in exclusion plots than control plots, we observed dramatic recovery in the abundance and response traits of rare imperiled species in a single growing season. If fences were maintained and vegetation was monitored over additional years, we expect the herbaceous plant community would continue to recover biomass, increase reproductive potential, and possibly increase species diversity. However, due to the long legacy of deer overbrowsing, and possible seed bank depletion, we recommend that effective restoration include reintroductions of extirpated species, and additional planting of competitively excluded species, in combination with deer exclusion (Ruhren and Handel, 2003; Nakahama et al., 2020). Constructing fences to exclude deer may be a simple yet effective strategy to protect endemic species and restore ecosystem functionality to historically degraded eastern serpentine habitats.

The raw data supporting the conclusions of this article will be made available by the authors, without undue reservation.

AR contributed to the conceptualization, methodology, investigation, formal analysis, visualization, and writing of this manuscript. CH contributed to the conceptualization, methodology, reviewing and editing, and funding acquisition for this manuscript. Both authors contributed to the article and approved the submitted version.

The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

All claims expressed in this article are solely those of the authors and do not necessarily represent those of their affiliated organizations, or those of the publisher, the editors and the reviewers. Any product that may be evaluated in this article, or claim that may be made by its manufacturer, is not guaranteed or endorsed by the publisher.