Introduction

Understanding the ecological requirements of an endangered plant species is essential to its preservation and successful recovery (Menges and Dolan 1998; Pfab and Witkowski 2000; Brys et al. 2004; Colling and Matthies 2006). In the absence of other information, habitat management and the selection of reintroduction sites are usually based on the environmental conditions of the sites where the remaining populations occur (Scott et al. 2002). However, a species that has become rare may be restricted to sub-optimal sites where its numbers will at best remain low and at worst decline. In this situation, experiments are necessary to determine optimal habitat characteristics (Maschinski et al. 2004; Guerrant and Kaye 2007), yet are rarely done (Wright et al. 2006).

A rare species may be restricted to sub-optimal sites because the best sites have been converted to other uses, but also because the environmental change has lowered site quality: sites that were once quite favorable for the species may no longer be. Environmental changes that have negatively affected rare species include decreases in fire frequency (Hawkes and Menges 1995; Bowles et al. 1998; Kaye et al. 2001), changes in grazing and mowing (Hutchings 1987; Schrautzer et al. 2011), woody plant encroachment (Waite and Farrell 1998), succession (Menges 1990), and invasive species (Seabloom et al. 2006). Third, some sites may always have been “sinks” in which the species only persisted because of migration from favorable areas (Van Hecke 1991). A population declining due to environmental changes or because its “source” population has been lost may persist for a long time with recruitment below the replacement rate (Eriksson 1996; Hanski and Ovaskainen 2002), due to long-lived individuals or a seed bank.

Streptanthus bracteatus, a rare annual of central Texas, may occur in sub-optimal sites for all three reasons. This study examined the potential effects of environmental change on site quality, specifically the effects of an increase in canopy and understory cover. Although existing populations occur in dense woodland, S. bracteatus grows very well in full sun (N. Fowler, personal observation). Our initial hypothesis that higher light levels would improve site quality was also supported by anecdotal reports that sites where it once grew abundantly were more open than sites it presently occupies and by observations of wild S. bracteatus plants apparently thriving in full sun (Bracted Twistflower Working Group, personal communication). More open canopies and understories in the past may have been caused by crown or by surface fires, both of which are now suppressed.

Endangered species can sometimes be “canaries in the coal mine” that provide clues to vanished habitats. As a result of fire suppression and other alterations of the disturbance regime, the vegetation now present in a region may no longer include all the types of habitat once present (Hobbs and Huenneke 1992). Even if species have not been lost, vegetation properties such as canopy openness may have changed throughout the region. If so, some of the species that are now rare may be rare because they require habitats that are now rare or even entirely lost. Therefore, the requirements of some rare species may provide insights into these rare or lost habitats. We suggest that the habitat requirements of S. bracteatus provide indirect evidence about past woodland composition and structure, and the past role of fire. This is the opposite of the usual approach in which the community is already well-known (e.g., to be fire-dependent) and the effects of an environmental factor such as fire are studied in the context of rare-species conservation only to fine-tune its use (e.g., Bradstock et al. 1996; Kaye et al. 2001; Liu et al. 2005).

Methods

Study species

Streptanthus bracteatus (bracted twistflower; Brassicaceae) is a rare endemic of the eastern and southern Edwards Plateau of central Texas (Poole et al. 2007). Most of the known populations are in Travis County (Pepper 2010). It is a winter annual with a persistent seed bank (Zippin 1997). All known Travis County populations are in woodlands (Zippin 1997). Previously known threats include the development of private land, recreational activities on public land, and herbivory by white-tailed deer (Zippin 1997; Pepper 2010). S. bracteatus has Candidate status under the Endangered Species Act [United States Fish and Wildlife Service (USFWS) 2011].

Phenology differs substantially from year to year. Seeds germinate in the autumn (October–December) when cool weather coincides with moist soil. In years with very little autumn rainfall, very few seeds germinate. Plants overwinter as basal rosettes. Reproduction is initiated in April when plants form flowering stalks. Flowers are produced in April or May, and seed dispersal occurs in May–July. Seedpods (siliques) 3–12-cm long split to release the seeds. There is no obvious mechanism of seed dispersal other than gravity.

Garden experiment

This experiment measured the effects of three light environments on S. bracteatus plants grown in pots in a garden. Seeds were germinated in small pots in a greenhouse in the winter of 2007 and later transplanted into 20-cm diameter pots. In June 2007, 39 plants were moved from the greenhouse into an unshaded outdoor area in Austin. On July 18, 2007, three experimental treatments were imposed: shadecloth, shade wall, or full sun. Full sun plants were unshaded from sunrise to sunset. In the shadecloth treatment, representing the light conditions S. bracteatus experiences in woodlands, each plant was individually surrounded (top and four sides) by shadecloth. The shade wall treatment represented the light environment a plant would experience at the edge of a woodland. Each shade wall plant had a vertical 0.5 m2 of opaque foamboard next to its pot on its west side. Pots were randomly intermingled. Each pot was at least 2 m away from any other pot, which prevented shading from other plants and their shade walls and shadecloth. Plants were watered, fertilized, and treated for powdery mildew and insects as needed throughout the experiment. See online supplementary material for additional information about experimental methods.

Light availability was measured with a LI-190SA quantum sensor (Li-Cor, Inc., Lincoln, Nebraska, USA). The average of three readings was recorded at plant height on a clear day in August at 1500 h for each treatment, while shade wall plants were shaded. Photosynthetically active radiation (PAR) ±1 standard error for the three treatments was: no shade, 1324.33 ± 11.06; shadecloth, 550.67 ± 9.44; and shade wall, 160.67 ± 5.66 μmol−2 s−1. Though 50 % shadecloth was used, only 42 % of full sunlight actually reached the plants inside. On clear days, shade wall plants received 5 h of full sunlight, mostly in the morning, 2 h midday in which only part of the plant was shaded, and 1.67 h of full shade in the afternoon.

Nine weeks after imposing the treatments, the 29 surviving plants were harvested, dried, and weighed. All seedpods were collected just before they dehisced and the number of seeds in each counted. Treatment means were compared with analyses of variance (ANOVAs). Survival rates and rates of reproduction were compared with Fisher’s exact test.

Field experiment

This experiment measured the effects of cover (canopy plus understory) on S. bracteatus. It was conducted in woodland dominated by Quercus buckleyi (Texas red oak) and Juniperus ashei (Ashe juniper) in Austin. There were ten pairs of plots (Fig. 1). Paired plots had similar slope, aspect, and elevation. One plot of each pair was a control, located under continuous canopy, and the other had the understory thinned. In addition, the thinned plots of plot-pairs 1–6 were in canopy gaps and the thinned plots of plot-pairs 7–10 were in canopy openings created by oak wilt (C. fagacearum) that had killed Q. buckleyi. By putting as many thinned plots as possible in this oak wilt area, we were able to achieve a broader range of light environments than we could with canopy gaps alone. The two plots of each plot-pair were 4–10 m apart, i.e., as close together as possible without affecting each other’s light environment.

Fig. 1
figure 1

Schematic diagram of the design of the field experiment following herbivore-caused mortality of all but 38 of the batch 1 transplants and their replacement with 62 batch 2 transplants. Please note that this is not a map, does not represent physical locations or distances at any scale, and does not show randomization. For example, control plots were not consistently north of thinned plots

Full size image

In November and December 2008, total cover was measured in all plots with a densitometer ~1 m above the ground. Woody plants were then pruned in the thinned plots until total cover was reduced by 50 % measured by a densiometer. Each plot was deer-fenced in December 2008.

The experiment was designed for 100 transplants, five per plot. Because only 38 of the 100 transplants in first batch of transplants survived (see below), we planted a second batch of 62 transplants. The transplants of batch 2 were planted wherever the transplants of batch 1 had not survived, replacing the dead transplants one-for-one (Fig. 1). All transplants were germinated and grown in individual pots. The first batch of transplants was germinated and grown indoors in the fall of 2008 and moved out-of-doors in November or December of 2008 to harden off. All seeds of the second batch of transplants were planted outside on February 1, 2009, where they remained until transplanted into the field site. Plants were treated as needed for powdery mildew and insects. The number of leaves was counted just before transplanting. No transplants had initiated reproduction before transplanting. Between January 19 and 21, 2009, we transplanted 100 plants, five per plot. Ten replacement transplants were planted on 18 February. Together these transplants are the batch 1 transplants. Herbivory was intense: leaves were repeatedly removed, and sometimes the entire plant was removed. All but 37 of the original plants and all but one replacement plant died, often after being repeatedly damaged, leaving 38 surviving transplants in batch 1 (diamonds, Fig. 1).

As soon as herbivory was observed, soon after transplanting the first batch of plants, anti-herbivore measures in addition to the deer fencing were taken. Neither bird netting over the deer exclosures nor poultry wire around the bases of the deer exclosures were effective, but cayenne was temporarily effective in reducing leaf damage; from this, we infer that climbing mammals, probably tree or ground squirrels, were responsible. Herbivore damage ceased when each plant was protected by an individual cage of poultry wire (2.54-cm hexagonal mesh) ~30-cm diameter and ~60-cm high.

The second batch of transplants was transplanted March 23–25, 2009 into the 62 empty points where plants had died. Each of the batch 2 transplants was individually caged immediately after transplanting, and all survived.

Transplants were watered daily for at least a week after transplanting to prevent transplant shock. We continued to provide lesser amounts of supplementary water because the soil was exceptionally dry. Rainfall during the 6 months from January to June 2009 was 66 % of the January–June average (NOAA NCDC), exacerbating a drought (July–December 2008 rainfall was 36 % of average). We estimate that the extra water did not quite bring the total up to an average year. For example, May + June average Austin rainfall is ~19 cm, which was more than the total of rainfall during that 2-month period in 2009 (7 cm) + added water (~10 cm).

Each week, we measured maximum rosette diameter (leaf tip to leaf tip) and counted the number of leaves. We recorded the date of first reproduction as the day on which bolting (production of a flowering stalk) was first visible. The length of each flowering stalk and seedpod (silique) was recorded weekly. For analysis, we summed the length of all flowering stalks on a plant on each date, and analyzed the maximum sum, regardless of which date it occurred on. Likewise, we summed the length of all seedpods on a plant on each date, but only analyzed the maximum sum. Most of the seedpods were left on the plants in the hopes of establishing a new population, but one pod was collected from each of 14 different plants of batch 1 and its seeds counted to estimate seeds per cm of pod length.

To measure cover during the experiment, hemispherical photographs were taken at each of the 100 planting points 3–7 March, and again on 22 June, with a Sigma 4.5 mm F2.8 EX DC Circular Fisheye lens after leveling the camera. Each photograph was taken directly above a transplant, ≤0.5 m above the ground. An estimate of cover (%) at this height was calculated from each photograph using Gap Light Analyzer© Version 2.0. Average plot cover for each date was calculated by averaging the estimated cover at all five planting positions in a plot.

Hemispherical photographs were also taken above plants in three natural S. bracteatus populations, Cat Mountain, Mount Bonnell, and Ullrich Water Treatment Plant, on May 27, 2009 and cover estimated from each.

Cover was compared among treatments, plot-pairs, and plots for each date separately with ANOVAs. Plant responses were analyzed for the two batches of transplants separately so that the ANOVAs and analyses of covariance (ANCOVAs) met the assumption of normal, homoscedastic residuals. All dead plants were excluded from the analyses of plant responses other than survival. Non-reproductive plants were excluded from the analyses of summed flowering stalk length and summed seedpod length. Plant responses to treatments were compared with ANOVAs or, if the model included a plant-size covariate, ANCOVAs. Plot-pair (the block term) and plot were random effects and treatment was a fixed effect, so treatment was tested over plot in these analyses. (Note that plot and treatment × plot-pair form a single term in these analyses.) Plant responses to cover itself were analyzed with models containing average plot cover in March or June (a continuous variable), plot (a random effect), and in some cases a plant-size covariate; cover was tested over plot. We report analyses of the effects of March cover on batch 1 plants and June cover on batch 2 plants, because these best represent the light environment experienced by each batch during the experiment. A plant-size covariate was included in an analysis if a preliminary analysis showed that it had a significant effect on the response variable. No more than one plant-size covariate was included in any model. See online materials for additional information about field experiment methods and analyses.

Results

Garden experiment

Differences in survival rates among the three light treatments were not significant (Fisher’s exact test, df = 2, χ2 = 0.55, P = 0.81). Differences in biomass also were not significant, mostly due to small sample sizes (ANOVA on total above-ground dry biomass, F 2,28 = 2.48, P = 0.10; Fig. 1). There was, however, a strong trend for plants grown in full sun all day or half of each day to be larger, by 69 and 97 % respectively, than plants grown under shadecloth.

Seven of the 29 surviving plants initiated reproduction: 46 % (5/11) of shade wall plants, 25 % (2/8) of full sun plants, and 0 % (0/10) of shadecloth plants. These differences were significant (Fisher’s exact test, χ2 = 5.91, P = 0.0492). The difference between full sun and shade wall plants, however, was not significant (Fisher’s exact test, χ2 = 0.83, P = 0.63). The difference in average seed set between these two treatments was also not significant.

Field experiment

Cover

Control plots had more cover than treated plots (Fig. 2; F 1,9 = 4.31, P = 0.068 and F 1,9 = 5.12, P = 0.050 in March and June, respectively). Average cover on the two dates was positively correlated (r s = 0.66, N = 20 plots). However, canopy gaps and understory pruning had little effect on cover in plot-pairs 1–6 once the canopy began to leaf out (Fig. 3). Only in three pairs of plots (pairs 7, 8, and 10), all located where oak wilt had killed the overstory Q. buckleyi trees (Fig. 1), did the control and thinned treatments continue to differ substantially in cover. This was reflected in large treatment × plot-pair interactions, i.e., plot-to-plot variation (F 9,80 = 14.42, P < 0.0001, and F 9,80 = 13.13, P < 0.0001, in March and June, respectively). Because cover was measured at the height of the S. bracteatus plants, ≤0.5 m above the soil surface, it included shading by both canopy and understory plants and, in some instances, by the slope of the hillside itself.

Fig. 2
figure 2

Average above-ground dry biomass of garden experiment plants at harvest. Vegetative + reproductive biomass = total biomass. Thin vertical lines ±1 SE. Shadecloth plants did not reproduce

Full size image
Fig. 3
figure 3

March and June total cover (canopy + understory) in each plot in the field experiment, calculated from hemispherical photographs taken at plant height (≤0.5 m). Vertical lines ±1 SE

Full size image

Survival

Almost all the deaths of the transplants of batch 1 were due to herbivory, probably by squirrels (see “Methods” section). Herbivory rates varied greatly among pairs of plots but not between paired plots (Fig. 1), so there was no evidence that treatment affected herbivory. Each of the plants in the second batch was individually caged at transplanting, none was eaten by squirrels, and all survived.

Reproduction: general

All the surviving plants of batch 1 initiated reproduction. There were on average 3.65 seeds/cm of seedpod length (N = 14 pods, SE = 0.29). Only 18 of the 62 plants in batch 2 had produced flowering stalks by the end of the experiment. Summed seedpod length was positively correlated with flowering stalk height (r s,batch1 = 0.56, P < 0.05; r s,batch2 = 0.67, NS), number of leaves (r s,batch1 = 0.11, NS, r s,batch2 = 0.68, NS), and rosette diameter (r s,batch1 = 0.39, P < 0.05, r s,batch2 = 0.27, NS).

Growth and reproduction: treatment effects

Treatment did not have a significant effect on any of the five tested responses of batch 1 transplants (Table 1; Fig. 4). Batch 2 plants were larger in thinned plots than in control plots: they had significantly more leaves and significantly greater total flowering stalk length (Table 2; Fig. 5).

Table 1 Results of tests of treatment effects on batch 1 transplants
Full size table
Fig. 4
figure 4

Means and standard errors of batch 1 transplants, adjusted for any covariates used in the model. Date of first reproduction is the date when a flowering stalk was first visible. Means ± 1 SE. P values of treatments from ANCOVA F tests (type 1 SS, denominator = MSplot)

Full size image
Table 2 Results of tests of treatment effects on batch 2 transplants
Full size table
Fig. 5
figure 5

Means and standard errors of batch 2 transplants. P values of treatments from ANCOVA F tests (type 1 SS, denominator = MSplot)

Full size image

Growth and reproduction: cover effects

More cover significantly reduced plant size: the greater the cover, the lesser the summed pod length of batch 1 plants (Table 3; Fig. 6) and the fewer the leaves on the batch 2 transplants (Table 4; Fig. 7). There was a non-significant trend for batch 2 plants that initiated reproduction before the end of the experiment to be in plots with lower cover than plants that did not initiate reproduction (only plot-pairs with batch 2 plants in both plots included in an analysis comparing June cover between treatments, N = 54; Wilcoxon χ2 = 3.1124, P = 0.0777).

Table 3 Effects of March cover on batch 1 transplants
Full size table
Fig. 6
figure 6

Relationships between March total cover (canopy + understory), as measured by hemispherical photographs taken at plant height, and performance of batch 1 plants in each plot. Fitted lines, shown only if P < 0.30, are from regressions of response variables against plot means, weighted by the number of plants in each plot

Full size image
Table 4 Effects of June cover on batch 2 transplants
Full size table
Fig. 7
figure 7

Relationships between June total cover (canopy + understory), as measured by hemispherical photographs taken at plant height, and performance of batch 2 plants in each plot. Fitted lines, shown only if P < 0.30, are from regressions of response variables against plot means, weighted by the number of plants in each plot

Full size image

Cover at three existing populations

Average cover over S. bracteatus plants at three existing Travis County populations was 63–68 % (Table 5). If all plants are pooled, the average was 66 %. By comparison, the average cover in the experimental plots in June was 70 % in the control plots, 72 % in the thinned plots in gaps, but only 50 % in the thinned plots in the oak wilt area.

Table 5 Cover at the height of a S. bracteatus plant (≤0.5 m above the ground) at three existing populations
Full size table

Discussion

Our initial hypothesis was supported: S. bracteatus thrives under less cover than the remaining Travis County populations experience. In pots with ample water, plants in full sun most or all the day grew as well or better than plants in continuous shade. In the more realistic field experiment, transplants in plots with lower cover were larger and had greater fecundity, even when differences in cover were relatively small. These positive responses to light are consistent with anecdotal reports (see “Introduction” section) and with recent studies of southern populations of S. bracteatus (Leonard 2010).

Our results strongly imply that the habitat occupied by the remaining Travis County populations is suboptimal. The results of the garden experiment suggest that optimum cover is quite low, perhaps close to zero, when water is not limiting. The results of the field experiment (Figs. 6, 7) suggest that optimum cover was as low as 37–48 % (lowest cover in March and June, respectively), and perhaps lower. The optimal cover for S. bracteatus at any given time likely is affected by soil water content and temperature. During hotter, drier seasons or years the effect of shading to reduce transpiration rate may be relatively more important and its competitive effects relatively less important, which would tend to increase the optimum level of cover for S. bracteatus. We cannot eliminate the effects of competition for below-ground resources. It is likely that shading by neighboring plants, in the field experiment and in the wild, is associated with greater competition for water and nutrients as well as for light, and this too affects optimum cover.

Environmental change is the most likely reason why the sites where S. bracteatus now grows have more cover than is optimal for this species. We suggest that S. bracteatus once occupied more open, patchy woodlands than it does now. Crown and surface fires, now suppressed, are the most likely pre-settlement creators of such woodlands. S. bracteatus may have even occupied the ecological niche of “fire-follower,” i.e., a species adapted to germinate and grow in the conditions that follow a fire. Its seedbank (Zippin 1997) could have allowed it to persist between fires. Some California species of Streptanthus are fire-followers (Moreno and Oechel 1991; Hickman 1993) and others live in the high-light environment of serpentine outcrops (Kruckeberg 1986; Mayer et al. 1994; Dolan 1995; Rodríguez-Rojo et al. 2001). The pre-settlement woodland fire regime in central Texas is not known; it is not even established that these woodlands experienced fires. The frequency, season, intensity, and spatial scale of past woodland fires are completely unknown, although observations of wildfires in this region suggest that woodland fires would have been quite patchy and produced a wide range of cover values (K. Doyle, personal communication). Most of the woody species now found in woodlands in the region regrow after fire (Reemts and Hansen 2008), but woodland fires, like oak wilt and our pruning, would have simultaneously reduced both cover and below-ground competition for a time. Whether competing perennials were completely killed or merely lost their leaves, “fire-following” annuals would have experienced less competition for both light and water from them after a fire.

A role for fire in the oak forests of the eastern United States, which are more mesic than those of central Texas, is gaining acceptance, although the ubiquity, frequency, and importance of fire are debated (Clark and Royall 1996; Clark 1997; Foster et al. 2002; Nowacki and Abrams 2008). Because failure of oak regeneration is a widespread concern (Loftis and McGee 1993), there are many studies of the effects of fire on oak regeneration. The magnitude of the positive effects they have found ranges from absent to slight and transient to substantial (e.g., Dey and Hartman 2005; Hutchinson et al. 2005b; Albrecht and McCarthy 2006; Green et al. 2010; Fan et al. 2012). Fire in eastern forests can create more open understories (Dey and Hartman 2005) and increase abundances of some herbaceous species (Elliott et al. 1999; Bourg et al. 2005; Hutchinson et al. 2005a). Prescribed burning and overstory thinning in one mixed-oak Ohio forest enhanced the performance of long-lived herbaceous perennials (Huang et al. 2007; Boerner and Huang 2008). Our approach of determining the ecological requirements of a rare species to infer past community structure and dynamics could perhaps help clarify the role of fire in these forests.

The requirements of rare species may provide insights into past community structure and dynamics in many ecosystems—the “canary in the coal mine” role described in the Introduction. For example, the role of fire in the oak-dominated woodlands of many Mediterranean-climate regions (Huntsinger and Bartolome 1992; Kutiel 1994; Acácio et al. 2009) might be clarified by studying their herbaceous species. Excellent examples of using rare species to guide conservation management (grazing, coppicing) come from English forests (Rackham 1990).

Management recommendations

In general, managers should be alert to the possibility that existing habitat may be a poor guide to a species’ optimal habitat requirements. When the mismatch between current and optimal habitat is not obvious, the ecological requirements of the species may be misunderstood, reducing the effectiveness of conservation actions.

Specifically, (1) sites selected for establishing new S. bracteatus populations should not be in dense J. ashei stands or other sites with dense cover. (2) One-time understory thinning to 50 % cover is not very effective in reducing cover. (3) If cover from all species 0.5 m above the soil surface is >50 %, more aggressive, ongoing thinning of woody plants (by removal or pruning) should be implemented in part of the site on an experimental basis. However, understory thinning should not be undertaken without providing protection from deer. The understory plants may be providing S. bracteatus plants with some physical protection from deer, as they do for oaks in this region (Zippin 1997; Russell and Fowler 2004). Understory thinning should also not be undertaken if it could lead to an increase in foot traffic or mountain bike traffic, both of which are significant threats to some populations (Pepper 2010). (4) Prescribed burns should be considered as an alternative tool for cover reduction. (5) Management for S. bracteatus may be more compatible with management for the endangered black-capped vireo (Vireo atricapilla), which requires open shrubland (Wilkins et al. 2006), than management for the endangered golden-cheeked warbler (Dendroica chrysoparia; USFWS 1990). Most of woodland conservation management in Travis County is for this warbler and neither mechanical thinning nor prescribed fires are allowed (City of Austin 2007). S. bracteatus conservation is, therefore, unlikely to be fully compatible with golden-cheeked warbler conservation as currently practiced. An investigation of the suitability of vireo habitat for S. bracteatus would be useful.

Average seed set/plant is a critical variable for any annual species. Summed seedpod length is a suitable surrogate for it when monitoring S. bracteatus populations; summed flowering stalk height is the next best choice. Furthermore, individual fecundity integrates the effects of all of the environmental factors that affect a plant between germination and death. It is, therefore, an excellent way to determine the suitability of habitat and the success of habitat management. Given the wide swings in population size from year to year (Zippin 1997), seedset/plant will give reliable information about habitat suitability long before a consistent trend in population size can be detected.