Interactive effects of flooding and deer (Odocoileus virginianus) browsing on floodplain forest recruitment

Forest Ecology and Management 303 (2013) 11–19

Contents lists available at SciVerse ScienceDirect

Forest Ecology and Management journal homepage: www.elsevier.com/locate/foreco

Interactive effects of flooding and deer (Odocoileus virginianus) browsing on floodplain forest recruitment Nathan R. De Jager a,⇑, Benjamin J. Cogger b, Meredith A. Thomsen b a b

USGS Upper Midwest Environmental Sciences Center, La Crosse, WI 54603, United States Department of Biology and River Studies Center, University of Wisconsin-La Crosse, La Crosse, WI 54601, United States

a r t i c l e

i n f o

Article history: Received 6 December 2012 Received in revised form 12 February 2013 Accepted 15 February 2013 Available online 30 April 2013 Keywords: Tree mortality Succession Upper Mississippi River Flooding Herbivory Reed canarygrass

a b s t r a c t Floodplain forests have historically been resilient to the effects of flooding because the tree species that inhabit these ecosystems regenerate and grow quickly following disturbances. However, the intensity and selectivity of ungulate herbivory in floodplains has the potential to modify the community-level effects of flooding by delaying forest recruitment and leaving sites vulnerable to invasive species. We established a series of exclosures along an elevation gradient in an actively recruiting floodplain forest along the Upper Mississippi River prior to three large-magnitude flood events. Pre-flood browsing by Odocoileus virginianus (white-tailed deer) ranged from 20% to 85% of all available stems, and reduced subsequent annual tree height growth from 60 cm/yr to approximately 35 cm/yr, regardless of elevation. Tree mortality, in contrast, was positively correlated with both pre-flood browsing rates and the duration of the growing season that the ground elevation of plots was flooded. Mortality rates ranged from approximately 40% in plots that experienced low levels of deer browsing (<30% of stems) and short flood durations (<40 days) to as high as 98% in plots that experienced high levels of deer browsing (>80% of stems) and long flood durations (>50 days). Longer flood durations led to larger shifts in tree community composition, away from heavily browsed and less flood tolerant Acer saccharinum L. (silver maple) and Populus deltoides (cottonwood) and toward species that were more flood tolerant and not preferred by deer. Phalaris arundinacea (reed canarygrass) colonized some portions of all plots, except for those situated at high elevations and protected by exclosures. Hence, herbivory can interact with the local flooding regime of rivers to delay recruitment of some tree species, resulting in shifts in successional trajectories, and leaving young forests vulnerable to invasion by exotic herbaceous species. Published by Elsevier B.V.

1. Introduction Flooding is the main driver of the structure and function of floodplain ecosystems (Junk et al., 1989). Although large magnitude floods can initially cause high rates of tree mortality, floodplain forests have historically been resilient to flooding as the tree species that inhabit these ecosystems have a high capacity to regenerate on frequently disturbed sites (Whitlow and Harris, 1979; Yin, 1998; Yin et al., 2009). However, flooding also provides a constant supply of invasive plant propagules from upstream sources, increasing the likelihood of invasion in disturbed floodplain sites (Rood et al., 2010; Eschtruth and Battles, 2011). Additionally, herbivore overabundance is now common worldwide (Côté et al., 2004; Mysterud, 2006) and selective foraging on tree species and avoidance of herbaceous species may further promote invasion (Kellogg and Bridgham, 2004). Understanding how ⇑ Corresponding author. Tel.: +1 608 781 6232; fax: +1 608 781 6066. E-mail address: [email protected] (N.R. De Jager). 0378-1127/$ - see front matter Published by Elsevier B.V. http://dx.doi.org/10.1016/j.foreco.2013.02.028

flooding interacts with herbivory to affect the resilience of floodplain forests in the presence of invasive species is important given the potential for more frequent large magnitude floods (Watson et al., 1998; Osborn and Hulme, 2002; Robson, 2002; Knox, 2009), the broad distribution of invasive wetland species (Zedler and Kercher, 2004), and overabundant herbivore populations (Côté et al., 2004). Effects of flooding in bottomland hardwood ecosystems are mainly determined by variation in hydroperiod (Sharitz and Mitsch, 1993; Hodges, 1997). Longer periods of soil saturation and anoxic conditions decrease plant photosynthetic rates and respiratory efficiency (Sena Gomes and Kozlowski, 1980; Pezeshki, 2001), resulting in overall higher rates of mortality at the lowest floodplain elevations. Some tree species can tolerate extended periods of anoxia by inducing the formation of morphological structures that aid in the transport of oxygen between roots and shoots (e.g. aerenchyma, lenticels, and adventitious roots (Whitlow and Harris, 1979)). Such adaptations allow flood-tolerant species to dominate in low-elevation sites (De Jager et al., 2012).

12

N.R. De Jager et al. / Forest Ecology and Management 303 (2013) 11–19

Herbivory may alter the ability of individual trees to tolerate the effects of flooding. Browsing by large ungulates often reduces height growth via the removal of apical meristems (Danell et al., 1994) which reduces auxin levels in axillary buds, increasing cytokinin levels and thereby promoting axillary bud development (Senn and Haukioja, 1994; Tanaka et al., 2006). The result is a highly pruned, shrub-like architecture (De Jager and Pastor, 2008, 2010). Browsing-induced reductions to the height growth of young trees could contribute to mortality in floodplains because lower stature prolongs the duration of terminal bud submergence during flood events, which is known to increase the likelihood of tree mortality (Hosner, 1958, 1960). Browsing could also increase mortality rates of trees not completely submerged by altering the production and balance of auxin and cytokinin, which are thought to be responsible for the formation of aerenchyma, lenticels, and adventitious roots in response to prolonged conditions of soil anoxia (Whitlow and Harris, 1979). By selectively foraging on some tree species and avoiding others, deer could cause shifts in tree species composition. Or, by focusing their feeding activity on woody species and avoiding herbaceous plants, deer could induce community type conversion in floodplains, especially when these ecosystems face pressure from highly invasive herbaceous species (e.g. Kellogg and Bridgham, 2004). In the winter of 2009, we established a series of exclosures to compare alternative methods to limit browsing by Odocoileus virginianus (white-tailed deer) along an elevation gradient in an actively recruiting forest in the floodplain of the Upper Mississippi River (UMR). The site had been successfully treated to control Phalaris arundinacea (reed canarygrass), but a previous study had demonstrated that deer browsing was limiting tree seedling height growth, slowing forest establishment (Thomsen et al., 2012). By chance, three large magnitude flood events occurred during the following two growing seasons, giving us the additional opportunity to examine a series of questions related to tree and reed canarygrass recruitment patterns in response to flood duration (elevation) and herbivory. The specific questions we addressed included: (1) What are patterns of plant forage selection by deer in bottomland hardwoods? (2) How do different tree species respond to lost tissue and variable flood durations? (3) Do browsing and flooding cause shifts in tree species compositions? (4) Does browsing and flooding facilitate invasion by reed canarygrass? (5) What are the implications for management actions aimed at promoting floodplain forest regeneration?

an estimated 11.1 deer/km2 in fall 2010 and 8.4 deer/km2 in winter 2011 (Wisconsin Department of Natural Resources, 2011). Densities exceed the estimated sustainable density of 2–4 deer/km2 for Wisconsin, suggested by Alverson et al. (1988). In late November 2009 we constructed three types of 20  20 m exclosures to reduce deer browsing of young trees: (1) mesh fences, (2) electric fences and (3) chemical fences. Mesh fences consisted of 2.4 m tall polypropylene mesh (Kencove Farm Fence Inc., Blairsville, PA) strung on 2.4 m tall metal fence posts spaced approximately 5 m apart. Mesh fences were further supported by a 3.26 mm nylon cable wire affixed to post tops. To prevent sub-fence entry, a 1.94 mm high tensile wire was woven through the base of the mesh fence and secured with ground staples. Electric fences consisted of 2.4 m tall solar-powered six-strand electrical fencing (1.2 cm electric ribbon; StafixÒ, Mineral Wells, TX; Magnum 12 V fence charger; ParmakÒ, Kansas City, MO) also strung on 2.4 m tall metal fence posts. Finally, chemical fences were constructed from 1.78 cm woven plastic fabric ribbon strung along plastic fence posts approximately 1 m tall, spaced every 5 m. Plotsaver™ (Messina Wildlife Management, Washington, NJ), a strong-smelling chemical deer repellent, was sprayed on the ribbon once a month for the duration of the study. Each exclosure type was replicated five times and applied to randomly assigned treatment plots. Five additional plots of equal size with no barriers served as controls, for a total of 20 plots. All experimental plots were located in the area in which Phalaris had previously been controlled and tree seedlings were abundant. Plot elevations were determined by recording the water depth at the four corners of each plot during a single day of stable inundation during spring 2010. The river stage height of the UMR at the longitudinal position of the study site was used to estimate water surface elevation (M.A.S.L). The difference between the measured water depths at the four corners of each plot and the elevation of the water surface of the UMR was then used to estimate the elevation of each plot. Plot elevations were later confirmed with LiDAR (Light Detection and Ranging) imagery. The mean of the four corner elevations was used as a single measure of plot elevation and revealed a gradient of ground elevations within each exclosure treatment (Fig. 1a). We transformed elevation from M.A.S.L to the number of days that the ground elevation was submerged during the growing seasons of 2010 and 2011 (April 1-Sept 30) by comparing plot elevation with river stage hydrographs at the longitudinal river position of the study site Fig. 1b). 2.2. Data collection

2. Methods 2.1. Study site Deer exclosures were constructed in a Mississippi River floodplain site owned by the U.S. Army Corps of Engineers (USACE) south of La Crosse, WI (43°44.30 N, 91°12.60 W). The low head dams constructed along the UMR do not store flood waters and thus the river still experiences flood pulses; the dams do however prevent low-flow conditions during the summer growing season. In 1998, straight-line winds blew down approximately 4.2 ha of forest within the site; it was salvaged logged and subsequently invaded by P. arundinacea L. (reed canary grass). By the start of the study described here, a previous restoration experiment (Thomsen et al., 2012) had reduced Phalaris cover and resulted in a mean tree seedling density of 30 seedlings per m2 of 2–3 year old Acer saccharinum L. (silver maple), Fraxinus pennsylvaticus Marshall (green ash), Salix exigua Nutt. and S. nigra Marshall (sandbar and black willow) and Populus deltoides W. Bartram ex Marshall (eastern cottonwood). White-tailed deer densities near the site during the study period were comparable to Wisconsin state averages with

Track surveys were conducted after snowfall events during the winter of 2009 (December 2009 – February 2010). The number of tracks was counted in a 1 m2 area at three points along three evenly spaced transects per plot. The mean number of tracks m2 in each plot was divided by the number of days since the last snowfall to estimate the number of tracks m2 day1. Track densities were qualitatively inspected during 2010 because there was little evidence of deer presence at the site. Vegetation sampling took place within five 1.5  1.5 m subplots within each of the 20 plots. Four subplots were located approximately 3 m from the corners of each plot and one subplot was centered within each plot. Sampling occurred during the spring (April) and late summer (August) of 2010 and 2011. Tree density was estimated within each subplot and used to estimate the percent mortality per plot over the course of the entire study period. Tree heights were measured for individuals within subplots and used to estimate mean annual height growth (cm yr1) per plot as well as the percentage of trees per plot escaping both the height reach of deer and maximum flooding depth at the site (>200 cm). Because our measure of flood duration extended past

13

N.R. De Jager et al. / Forest Ecology and Management 303 (2013) 11–19



 ri ð1  pi Þ Ei ¼ ln pi ð1  r i Þ

A Elevation (MASL)

193.5

where ri is the proportion of stems browsed from species i and pi is the proportion of stems of species i available. Index values less than zero indicate avoidance, whereas values greater than zero suggest a preference for a given species. X2 values were calculated to evaluate the statistical significance of Ei based on one degree of freedom (Jenkins, 1979):

193.4 193.3 193.2

X2 ¼ 

1 xi

193.1

B 100

Days Flooded

ð1Þ

80 60 40 20 0

Control

Chemical

Electric

Mesh

Fig. 1. (A) Mean (±1 SE) ground elevations (meters above sea level, MASL) for the four corners of the 20 forest plots under study, organized by exclosure treatment. (B) Estimated duration of ground-level inundation during the growing season (April 1 – September 30) for forest plots (days flooded). The 30-year mean growing season flood durations are indicated by black bars, while the 2-year mean growing season flood durations for 2010–2011 (±1 SE) are indicated by the black bars plus the white bars.

August 2011, we resampled plots in April 2012 to verify that no additional changes in tree height growth or mortality had occurred after August 2011. All measurements were made according to species. Percent cover of reed canarygrass was estimated visually within each subplot and then aggregated at the plot level using subplot averages. These estimates were made during the two summer sampling episodes of 2010 and 2011, and one additional sampling effort was conducted during early November of 2012. Browsing intensity was estimated as the percentage of plant stems browsed (% consumption) in the subplots. During summer, percent consumption was estimated as the percentage of all stems bearing leaves that were browsed. Percent consumption was used to represent deer browsing rather than the number of stems browsed because our study included both single-stemmed seedlings and multi-stemmed saplings. When this is the case, percent consumption better represents the intensity of browsing as it relates to likely effects on plant growth and survival because it accounts for the total number of stems available. 2.3. Data analysis Differences in herbivory among the exclosure treatments were examined using one-way ANOVAs of plot-level means for winter track densities as well as winter and summer percent consumption; Tukey–Kramer post hoc analyses were used to evaluate pairwise differences. Whether herbivory occurred at random with respect to tree species (i.e., in proportion to species abundances) was tested using an electivity index:

1 þ mx

 i

E2i þ



1 yi

1 þ ny



ð2Þ

i

where xi is the number of stems browsed of species i, yi is the number of stems of species i available, m is the total number of stems browsed across all species and n is the total number of stems available across all species. We used measurements of the percentage of plant stems browsed by deer as a continuous measure of herbivory, rather than retaining the exclosure treatments as categorical variables for examining effects of herbivory on tree growth, mortality, species composition, and reed canarygrass cover. The effects of herbivory and flooding on tree height growth, mortality, species composition and reed canarygrass cover were examined at the plot level using multiple linear regressions to avoid pseudoreplication. Best models were determined starting with the full model (all factors) and then backward stepwise regression was used with AIC for small sample sizes as the model selection criteria (Burnham and Anderson, 2001). Model selection was carried out using the ‘step’ command in R 2.7.1 (R development Core Team, 2008), and the final model had the lowest AIC value. We evaluated model selection uncertainty by calculating AIC weights (wr), which sum to one and can be interpreted as the probability that the selected model is in fact the best model given the set of candidate models and the data at hand (Burnham and Anderson, 2001). Changes in tree species composition from the start of the study to the end of the study were examined at the plot level using nonmetric multi-dimensional scaling (NMDS) of Bray-Curtis similarity estimates (Legendre and Legendre, 1998) with PRIMER 6 (PRIMER-E, Plymouth, UK, Clarke and Gorley, 2006). Vectors were applied to an ordination biplot to identify changes in the location of each study plot from the start of the study to the end of the study along each ordination axis. Shifts in the position of each plot from the start to the end of the study were then interpreted with respect to correlation coefficients between the proportional abundance of each species and each ordination axis. 3. Results 3.1. Herbivory Track surveys in winter 2009 indicated that the effectiveness of the exclosures differed (F3,16 = 14.65, P < 0.001). Mesh fences completely excluded deer and electric fences appeared to provide moderate protection for seedlings, but no difference was found between track densities inside chemical fences and unfenced controls (Table 1). Mesh fences were the only type to significantly decrease the amount of plant tissue removed by deer during the first winter (F3,16 = 10.58, P < 0.001, Table 1). The percentage of stems browsed by deer during winter 2009 in the mesh exclosures averaged 20% (browsing that occurred prior to fence construction) as compared to 40–85% in the other treatments (Table 1). Across all treatments, deer highly preferred silver maple stems (Ei = 0.49), which constituted 76% of all browsed stems (Table 2). Silver maple was also the most abundant forage species at the start of the experiment, making up 66% of the stems available. Although the other

14

N.R. De Jager et al. / Forest Ecology and Management 303 (2013) 11–19

Table 1 Mean (±1 SE) winter track densities and the percentage of plant stems browsed by white-tailed deer measured within three exclosure treatments (fence types) and an unfenced control treatment (five replicates within each treatment) in an early successional floodplain forest site along the Upper Mississippi River during the winter of 2009. Browsing in the mesh exclosures occurred prior to fence construction in November 2009 because deer tracks were never observed in those plots. Significant differences among treatments are identified by different letters. Treatment

Mean track (m2 wk1)

Mean% of stems browsed

Control Chemical Electric Mesh

2.8 ± 0.13a 2.2 ± 0.33ab 1.5 ± 0.28bc 0 ± 0c

57.3 ± 5.9a 73.0 ± 6.5a 58.6 ± 8.9a 23.6 ± 2.9b

Table 2 Estimates of browse availability, consumption, and an index of selective foraging (electivity) for browse data collected during the winter of 2009.

*

Species

# Stems available

% Stems browsed

% Of all stems browsed

Electivity

Green ash Cottonwood Silver maple Willow

258 26 738

32.2 53.4 58.4

14.6 2.4 76.1

0.55* 0.06 0.49*

96

39.5

6.7

0.26*

Indicates significant avoidance or preference for a given species (P = 0.05).

three species were browsed by deer, green ash and willow were avoided (Ei = 0.55 and 0.26, respectively) and cottonwood was browsed in proportion to its abundance (Ei = 0.06). Following the winter of 2009, a precipitous drop in deer activity was noted at the study site, corresponding with three consecutive large magnitude flood events (Fig. 2). 3.2. Flooding The three flood events that occurred during 2010 and 2011 surpassed the 90th percentile of daily river stage readings taken during the past 50 growing seasons (April 1-Sept. 30), and the latter two events surpassed the 95th percentile of daily stages (Fig. 2). The peak river stage observed during 2011 was surpassed during at least one day during 5 of the past 50 growing seasons, indicating that flooding of this magnitude is a decadal phenomenon. Water

depths during these events ranged from less than 0.5 m in higher elevation plots during the first flood pulse to as deep as 1.75 m in the lower elevation plots during the third flood pulse. These events doubled the duration that the ground elevations of the study plots were inundated during the growing season, from a 30-year mean that ranged from 13 to 42 days to a 2-year mean (2010 and 2011) that ranged from 27 to 85 days (Fig. 1b), and very little browsing occurred at the site during the two years of high water, regardless of exclosure treatments (Fig. 2). The percentage of stems browsed by deer during winter 2009 was therefore used as the measure of percent consumption in subsequent analyses, reflecting the intensity and selectivity of deer browsing that occurred prior to the flood events and in response to the exclosure treatments. 3.3. Effects of herbivory and flooding on tree growth and mortality The rate of annual height growth of trees (all species grouped) from the winter of 2009 to the end of our experiment in late summer 2011 was negatively associated with the percentage of stems browsed by deer during the winter of 2009, regardless of flood duration (Tables 3 and 4, Fig. 3a). Trees from heavily browsed plots grew approximately 35 cm yr1 as compared with a height growth of 60 cm yr1 for trees in lightly browsed plots. Best models based on AIC for cottonwood and silver maple, species that deer displayed neutral and positive electivities for, respectively, indicated similar negative associations between the percentage of plant stems browsed by deer and plant height growth (Tables 3 and 4). There was some evidence (wr = 0.44) indicating that flood duration may have had additional effects on the height growth of cottonwood. In contrast, there was very little evidence to suggest that the height growth rates of green ash and willow, two species not preferred by deer, were related to percent consumption or flood duration (Tables 3 and 4). One consequence of browsing-induced reductions to height growth was that the percentage of live trees reaching 200 cm (and thus escaping both the height reach of deer and terminal bud submergence during the flood events) at the conclusion of the study was negatively associated with the percentage of stems browsed by deer, again regardless of flood duration (Fig. 3b). Willow comprised the majority of trees escaping browse and flood height (n = 41, 65%). The small number of trees of each species reaching 200 cm (green ash = 8, cottonwood = 3, silver maple = 12) precluded the statistical investigation of species-level effects on escape likelihood.

195.0

80

95% 194.0

60

90% 40

193.5 193.0

% Consumption

Elevation (MASL)

194.5

20 192.5 192.0 2006

0 2007

2008

2009

2010

2011

Time Fig. 2. Daily water surface elevation (meters above sea level, MASL) at the longitudinal river position of the study site from 2006 through the end of the study in 2011. Dots indicate the percentage of stems browsed by deer averaged across all plots not protected by mesh fences (% consumption) during the winter of 2009 (sampled in spring 2010) and winter 2010 (sampled in spring 2011), as well as summer of 2010 and 2011. Solid gray lines denote the lowest and highest plot elevations at the study site. Dashed lines represent the 90th and 95th percentiles of all daily river stage readings taken during the past 50 growing season.

15

N.R. De Jager et al. / Forest Ecology and Management 303 (2013) 11–19

Table 3 AIC weights for candidate models for predicting annual tree height growth (cm yr1) and percentage mortality for all species pooled (All), green ash, cottonwood, silver maple, and willow species (Best models are in bold). Candidate models included a null model (y = intercept-only), linear models consisting of the percentage of stems browsed by deer (c), the mean growing season duration of ground-level inundation (f), c + f, and an interaction term c⁄f, in addition to the individual main effects. AIC weights can be interpreted as the probability that the selected model is indeed the best model would emerge as the best model given the set of candidate models and the data at hand. Response

Species

Model type y = intercept-only

y=c

y=f

y=c+f

y=cf

Ht. growth

All Green ash Cottonwood Silver maple Willow

<0.01 0.63 <0.01 <0.01 0.53

0.93 0.28 0.52 0.97 0.10

<0.01 0.04 <0.01 <0.01 0.36

0.05 0.04 0.44 0.03 0.01

0.02 0.00 0.03 <0.01 <0.01

Mortality

All Green ash Cottonwood Silver maple Willow

<0.01 0.60 <0.01 <0.01 0.82

<0.01 0.34 0.98 <0.01 0.02

<0.01 0.03 <0.01 <0.01 0.17

0.93 0.02 0.02 0.90 <0.01

0.07 0.01 <0.01 0.10 <0.01

Table 4 Best models based on AIC for annual tree height growth (cm yr1) and percentage mortality for all species pooled (All), green ash, cottonwood, silver maple, and willow species, where c is the percentage of stems browsed by deer, and f is the mean growing season duration of ground-level inundation. The full set of candidate models and AIC weights (wr) are provided in Table 3. Response

Species

Model

wr

R2

Ht. growth

All Green ash Cottonwood Silver maple Willow

y = 65.5–0.35c y = 38.7 y = 53.8–0.2c y = 56.45–0.226c y = 63.95

0.93 0.63 0.52 0.97 0.53

0.39

0.002

0.4 0.34

0.06 0.04

All Green ash Cottonwood Silver maple Willow

y = 17.63 + 0.76f + 0.21c y = 30.37 y = 51.6 + 0.47c y = 18.72 + 0.92f + 0.21c y = 37.57

0.93 0.60 0.98 0.90 0.82

0.57

<0.001

0.38 0.60

0.04 <0.001

Mortality

The best model based on AIC for predicting percent mortality (across all species) consisted of the effects of flood duration and percent consumption, accounting for 57% of the variance in mortality rates (Tables 3 and 4, Fig. 4). Mortality rates at the highest elevations ranged from approximately 40% in plots that experienced low levels of deer browsing (<30% of stems) to as high as 75% in plots that experienced high levels of deer browsing (>80% of stems). Mortality rates at the lowest elevation sites ranged from approximately 75% in plots that experienced low levels of deer browsing to as high as 98% in plots that experienced high levels of deer browsing. The model in Fig. 4 predicts just 17% mortality in the absence of deer browsing at the highest elevation sites and 100% mortality at the lowest elevation sites when deer remove 100% of available plant stems. At the end of the study, total tree densities in plots that experienced long flood durations ranged from approximately 0.06 trees m2 in heavily browsed plots to 5.4 trees m2 inside the lowest elevation mesh exclosure. In comparison, tree densities at higher elevations ranged from approximately 7.3 trees m2 in heavily browsed plots to as high as 13 trees m2 in the highest elevation mesh exclosure, further suggesting that the effect of pre-flood browsing on tree mortality was consistent across the elevation gradient. The best species-specific model for percent mortality of silver maple was similar to the model applied to all species (Tables 3 and 4), suggesting that this species was primarily responsible for the patterns in flood- and browsing-induced mortality across all species. Mortality rates for green ash and willow averaged 30.3% and 37.5% respectively across all plots and there was very little evidence to suggest association with flooding or browsing (Tables 3 and 4). Mortality of cottonwood increased with percent consumption from a low of 51% to a high of 100%, regardless of flood duration (Tables 3 and 4).

P

3.4. Effects of herbivory and flooding on community composition The best model based on AIC for predicting changes in tree species composition consisted of the effects of flood duration (Fig. 5a). Bray-Curtis similarity of the study plots at the beginning versus end of the experiment ranged from near 100% for high elevation plots to near 0% for low elevation plots, indicating that resistance to changes in species composition decreased with increasing flood duration (Fig. 5a). However, changes in species composition in the lower elevation plots occurred along NMDS axis 1 of the biplot shown in Fig. 5b, which was negatively associated with the relative abundance of silver maple (r = 0.99) and positively associated with the relative abundance of green ash (r = 0.81) and willow (r = 0.75). Hence, changes in species composition consisted of shifts away from dominance by silver maple, the species most preferred and affected by deer, and toward dominance by green ash and willow, the two species not preferred or affected by deer. Percent cover of reed canarygrass was uniformly low across our plots during the abbreviated growing seasons of 2010 and 2011, occupying less than 20% cover in all plots. However, we observed a marked increase in reed canarygrass cover during a follow-up survey in early November 2012, following a growing season more typical of the 30-year mean (Fig. 1b). The best model for predicting percent cover of reed canarygrass in November 2012 consisted of the effects of flood duration, percent consumption, and the interaction between these two factors (Fig. 6). Percent cover of reed canarygrass was uniformly high (50–60%) in plots that were flooded for long durations (regardless of consumption) and in plots that received high pre-flood browsing rates (regardless of flood duration), but much reduced in plots that experienced short flood durations and low browsing rates (<20% cover).

N.R. De Jager et al. / Forest Ecology and Management 303 (2013) 11–19

Annual Height Growth (cm yr-1)

16

70

A

60 50 40 30

y = 65.5 - 0.35x

20

R2 = 0.39, P = 0.002, wr = 0.93 10 0

% of Trees > 200 cm

40

20

40

60

B

80

100

y = 35.6 - 0.43x R2 = 0.57, P < 0.001, wr = 0.97

30 20

Fig. 4. The effects of flood duration (f, mean days inundated from April 1 – September 30 in 2010 and 2011) and deer browsing (c, percent of stems browsed by deer during the winter of 2009) on tree mortality calculated from the start of the study in winter 2009 to the end of the summer of 2011. Open circles indicate plots protected by mesh fences; solid circles represent plots of the other exclosure types and unprotected control plots.

10 0

0

20

40

60

80

100

% Consumption Fig. 3. (A) The mean annual height growth of trees in each study plot (winter 2009 – summer 2011) versus the percentage of stems browsed by deer during the winter of 2009 (% consumption). (B) Percentage of trees escaping both browse and flood height (>200 cm) at the completion of the study in 2011 across all species versus percent consumption. Open circles indicate plots protected by mesh fences; solid circles represent the other exclosure types and unprotected control plots.

4. Discussions 4.1. Flooding and herbivory increase tree mortality and change species composition Three consecutive large magnitude floods occurred during the course of our study (2010 and 2011), impacting the survival of tree seedlings. However, herbivory that took place prior to the floods critically influenced plant mortality in response to flooding. In higher elevation sites, variation in pre-flood browsing from 20% to 85% of available stems corresponded to tree mortality rates ranging from approximately 40% to 60%. The same range of browsing intensities in lower elevation sites increased mortality from approximately 80% to near 100%. Hence, although flood duration had a relatively larger effect on tree mortality than herbivory, and although the magnitude of the effect of herbivory on tree mortality did not vary with flood duration, the added effect of herbivory was sufficient to induce a near complete loss of tree cover in plots at the lowest elevations. Effects of browsing were most pronounced on the species that deer showed a preference for, silver maple and cottonwood, and the lower elevation plots experienced large shifts away from these species and toward less palatable green ash and willow. Our results indicate that high deer densities can reduce the resistance of early successional forests to effects of large magnitude floods by increasing tree mortality rates above those expected to occur in response to flood events, especially

for highly palatable species. Other studies have also found that browsing by white-tailed deer can limit the success of floodplain forest restoration (Sweeney et al., 2002; Ruzicka et al., 2010; Thomsen et al., 2012), further suggesting the potential for improved management practices if levels of herbivory can be reduced. Each of the species present at our study site can withstand long periods of soil saturation, but complete submergence is known to cause high rates of mortality (Whitlow and Harris, 1979). Although it was not possible to attain accurate estimates of the duration of terminal bud submergence for each tree in our study, approximate estimates based on mean tree height at the start of each growing season were on the order of 15–35 days. Complete submergence for such durations has been shown to cause 100% mortality of silver maple and cottonwood seedlings, while green ash and willow seedlings were able to survive such flood durations (Hosner, 1958, 1960). Hence, the species that deer tended to avoid in our study are also likely to be the most tolerant of terminal bud submergence. Selective foraging and flooding may act synergistically to promote relatively predictable shifts in tree species composition away from highly palatable and less flood tolerant species. This finding could result from direct effects of browsing on tree height growth, which would have prolonged the duration of terminal bud submergence for species already relatively intolerant of complete submergence.

4.2. Flooding and herbivory facilitate invasion Large floods are important components of bottomland ecosystems because they reset the sequence of forest succession and promote vegetation turnover (Yin et al., 2009). Floodplain trees grow rapidly (Barancˇeková et al., 2007), making bottomland forests generally resilient to the effects of disturbance (i.e. they can recover). However, competition from reed canarygrass decreases growth and survival of tree seedlings of species typical of earlysuccessional bottomland forests (Reinhardt Adams et al., 2011;

17

A

100

NMDS AXIS 2

Similarity

80 60 40 20 y=126.1-1.14x R2=0.61, P<0.001, wr = 0.96 0 30 40 50 60 70

80

90

Days Flooded

ASH (-0.58) WIL (+0.60), CW (+0.37), SM (+0.01)

N.R. De Jager et al. / Forest Ecology and Management 303 (2013) 11–19

1.5

B

Stress: 0.04

1.0 0.5 0.0 -0.5 -1.0 -1.5 -1.5

-1.0

-0.5

0.0

0.5

1.0

1.5

2.0

SM (-0.99) ASH (+0.81), CW (+0.26), WIL (+0.75)

NMDS AXIS 1 Fig. 5. (A) Biplot of the Bray–Curtis similarity of tree assemblages for each plot at the start of the study in winter 2009 as compared to the end of the study in the summer of 2011. Open symbols represent mesh exclosures and solid circles represent other exclosure types and unfenced control plots. (B) NMDS ordination plot showing tree assemblages at the start of the study (open triangles) and at the end of the study (closed circles). Note the shift in tree assembly along the NMDS axis 1, which had a negative correlation coefficient when compared with the relative abundance of silver maple (SM, 0.99) and a positive correlation coefficient when compared with the relative abundance of green ash (ASH, 0.81), willow (WIL, 0.75), and cottonwood (CW, 0.26), signifying shifts away from silver maple and toward other species for some study plots.

Fig. 6. The effects of flood duration (f, mean days inundated from April 1 to September 30 in 2010 and 2011) and deer browsing (c, percent of stems browsed by deer during the winter of 2009) on percent cover of reed canarygrass at the end of the summer of 2012. Open circles represent plots protected by mesh fences; solid circles represent plots of the other exclosure types and unprotected control plots.

management intervention. In contrast, the highest elevation plots that were protected from herbivory now have high densities of tall trees which shaded the soil surface (personal observations). Reed canarygrass germination is decreased by shading (Lindig-Cisneros and Zedler, 2002), leading us and other authors to hypothesize that early-successional bottomland hardwoods will resist reed canarygrass invasion once saplings grow tall enough (Hovick and Reinartz 2007, Reinhardt Adams et al., 2011). Similar positive feedbacks among shading, litter accumulation and competition are believed to limit re-invasion by reed canarygrass in restored sedge meadows dominated by Carex stricta (tussock sedge) (Zedler, 2009). The outcome in the higher elevation exclosures is similar to that reported from upland studies, which show that protection from deer promotes woody plant dominance, preventing herbaceous plant expansion (Knight et al., 2009; Rooney, 2009). In protected, low elevation plots and unfenced, higher elevation plots we observed smaller reductions in tree density as compared with lower elevation unfenced plots, but similar increases in reed canarygrass cover. In these areas, we predict that long-term site trajectory will be determined by the outcome of competition between the remaining saplings and reed canarygrass. Future monitoring of these sites will shed light on the nature of transitions between the alternative stable states (reed canarygrass monoculture vs. restored forest) highlighted by Zedler (2009), and the role flooding and herbivory play in tipping the balance toward one or the other.

4.3. Management implications Thomsen et al., 2012), raising concerns that the invader threatens the long-term resilience of bottomland hardwood ecosystems (Romano, 2010). Our study began with low reed canarygrass foliar cover in all study plots, as a result of prior restoration efforts (Thomsen et al., 2012). However, the near complete loss of tree cover from plots that experienced both high browsing rates and long flood durations resulted in a large increase in reed canarygrass cover the following growing season (i.e., re-invasion). The previous lack of tree seedling establishment in plots dominated by reed canarygrass at our study site (Thomsen et al., 2012) strongly suggests that these areas will remain dominated by the grass in the absence of

This study examined patterns of tree growth and mortality in response to an herbivore population estimated to be at least two times greater than suggested sustainable densities (Alverson et al., 1988; Wisconsin Department of Natural Resources, 2011) and flooding durations that lasted twice as long as the 30-year mean. Our results suggest that where such high rates of browsing and long flooding durations combine to result in high levels of tree mortality, they can shift species composition and promote invasive plant establishment, potentially reducing long-term forest ecosystem resilience and the success of management actions. Reed canarygrass recolonized areas from which it had previously been

18

N.R. De Jager et al. / Forest Ecology and Management 303 (2013) 11–19

eliminated (Thomsen et al., 2012), reinforcing the need for crepeated control efforts and long-term monitoring in successful restorations (Lavergne and Molofsky, 2006; Zedler, 2009; Annen, 2011), especially in the face of chronic high deer densities (Waller and Alverson, 1997) and predicted increases in the frequency of high magnitude floods (Watson et al., 1998; Osborn and Hulme, 2002; Robson, 2002; Knox, 2009). There is widespread interest in managing deer herd size to decrease the negative impacts of deer browsing (McShea, 2012). It has been shown, however, that legacies of herbivory can persist after herd sizes are reduced, preventing the restoration of historical forest composition, structure and function (Royo et al., 2010; Tanentzap et al., 2011). The collective results from this study and Thomsen et al. (2012) suggest that in our study system a short period of protection from browsing may provide young trees with a window of opportunity to escape herbivory, complete submergence during flood events, and competition with invasive species for light. Tanentzap et al. (2012) provide a list of suggestions to improve the likelihood of success of management interventions to decrease browsing pressure in forests, all of which are met by the natural characteristics of bottomland hardwood ecosystems. Recovery from browsing is predicted to take the longest on nutrient-poor soils (Tanentzap et al., 2012), and floodplain ecosystems are characterized by high nutrient availability, high productivity, and high growth rates (Malanson, 1993). Propagule availability for palatable species can be limited in sites with a history of browsing (Tanentzap et al., 2012); canopy trees surrounding the study plots were observed to provide high rates of seed rain in our site (Thomsen et al., 2012), and seasonal flooding serves to deliver propagules in riparian ecosystems (Eschtruth and Battles, 2011). Topographic features can provide refugia from herbivores (e.g. boulder-top communities in eastern hardwood forests, Rooney, 1997), and in geomorphically complex floodplains, isolated islands exhibit lower browsing rates than mainland sites (Cogger et al., in Press). Finally, disturbances that expose mineral soil may promote the establishment of palatable species after browsing intensity is decreased (Tanentzap et al., 2012), and in riparian ecosystems that is accomplished through episodic flooding (e.g. Yin, 1998). In summary, bottomland ecosystems are characterized by environmental conditions that are predicted to promote forest recovery after herbivore abundances are decreased, maximizing the costeffectiveness of management interventions. While effects of herbivory may be manageable on a site-by site basis, the effects of flooding have to be addressed at the regional scale. Although recurrent flooding is generally considered an important component of large floodplain rivers (Junk et al., 1989), there is growing evidence that the frequency of large magnitude floods may increase over the next century in response to climate change (Watson et al., 1998: Osborn and Hulme, 2002: Robson, 2002; Knox, 2009). As a response, it could be useful for floodplain managers to consider the role of regional hydrological manipulations, especially the ways in which such actions might interact with herbivory. For example, in the UMR, summertime draw-downs are used to simulate historic low-flow conditions. Such actions might provide an opportunity for the establishment of tree seedlings at lower elevations than is usually possible, particularly if new seedlings can be protected from herbivory and if the window between large floods is long enough to allow newly generated trees to escape effects of future floods. Yin et al. (2009) recently modeled responses of trees to floods and showed that repeated draw-downs could improve the long-term regeneration of some floodplain tree species. Information about landscape distributions of browsing pressure and annual flood durations could be used to identify places on the landscape most susceptible to flood and browsing-induced forest turnover and compositional change.

Acknowledgements This work was partially supported by the National Great Rivers Research and Education Center (NGRREC), the University of Wisconsin-La Crosse (UW-L), the UW-L River Studies Center and the Long Term Resources Monitoring Program, a component of the Upper Mississippi River Restoration-Environmental Management Program. The authors thank the Environmental Stewardship staff at the La Crescent Field Office of the United States Army Corps of Engineers, St. Paul District for help in the field and site treatment expenses, and numerous individuals at UW-L for their help in the field. Use of trade, product, or firm names does not imply endorsement by the US Government.

References Alverson, W.S., Waller, D.M., Solheim, S.L., 1988. Forests too deer: edge effects in northern wisconsin. Conservation Biology 4, 348–358. Annen, C.A., 2011. Manipulating internal system feedbacks to accelerate reed canarygrass (Phalaris arundinacea) control: from theory to practice. Ecological Restoration 29, 222–224. Barancˇeková, M., Krojerová-Prokešová, J., Homolka, M., 2007. Impact of deer browsing on natural and artificial regeneration in floodplain forest. Folia Zoologica 54, 354–364. Burnham, K.P., Anderson, D.R., 2001. Kullback–Leibler information as a basis for strong inference in ecological studies. Wildlife Research 28, 111–119. Clarke, K.R., Gorley, R.N., 2006. PRIMER v6. User Manual and Tutorial. PRIMER-E, Plymouth, UK. Cogger, B.J., De Jager, N.R., Thomsen, M.T., in Press. Winter browse selection by white-tailed deer and implications for bottomland forest restoration in the Upper Mississippi River valley, USA. Natural Areas Journal. Côté, S.D., Rooney, T.P., Tremblay, J., Dussault, C., Waller, D.M., 2004. Ecological impact of deer overabundance. Annual Review of Ecology Evolution and Systematics 35, 113–147. Danell, K., Bergström, R., Edenius, L., 1994. Effects of large mammalian browsers on architecture, biomass and nutrients of woody plants. Journal of Mammalogy 75, 833–844. De Jager, N.R., Pastor, J., 2008. Effects of moose Alces alces population density and site productivity on the canopy geometries of birch Betula pubescens and B. pendula and Scots pine Pinus sylvestris. Wildlife Biology 14, 251–262. De Jager, N.R., Pastor, J., 2010. Effects of simulated moose Alces alces browsing on the morphology of rowan Sorbus aucuparia. Wildlife Biology 16, 307–310. De Jager, N.R., Thomsen, M.T., Yin, Y., 2012. Threshold effects of flood duration on the vegetation and soils of the Upper Mississippi River floodplain, USA. Forest Ecology and Managment 270, 135–146. Eschtruth, A.K., Battles, J.J., 2011. The importance of quantifying propagule pressure to understand invasion: an examination of riparian forest invisibility. Ecology 92, 1314–1322. Hodges, J.D., 1997. Development and ecology of bottomland hardwood sites. Forest Ecology and Management 90, 117–125. Hosner, J.F., 1958. The effects of complete inundation upon seedlings of six bottomland tree species. Ecology 39, 371–373. Hosner, J.F., 1960. Relative tolerance to complete inundation of fourteen bottomland tree species. Forest Science 6, 246–251. Hovick, S.M., Reinartz, J.A., 2007. Restoring forest in wetlands dominated by reed canarygrass: the effects of pre-planting treatments on early survival of planted stock. Wetlands 27, 24–39. Jenkins, S.H., 1979. Seasonal and year-to-year differences in food selection by beavers. Oecologia 44, 112–116. Junk, W.J., Bayley, P.B., Sparks, R.E., 1989. The flood pulse concept in river-floodplain systems. In: Dodge, D.P. (Ed.), Proceeding of the International Large River Symposium, vol. 106. Canadian Special Publications in Fisheries and Aquatic Sciences, pp. 110–127. Kellogg, C.H., Bridgham, S.D., 2004. Disturbance, herbivory, and propagule dispersal control dominance of an invasive grass. Biological Invasions 6, 319–329. Knight, T.M., Dunn, J.L., Smith, L.A., Davis, J., Kalisz, S., 2009. Deer facilitate invasive plant success in a Pennsylvania forest understory. Natural Areas Journal 29, 110–116. Knox, J.C., 2009. Evidence of Mississippi River flood history preserved in floodplain sedimentary deposits. In: Finding the Balance Between Floods. Flood Protection, and River Navigation. Center for Environmental Sciences, St. Louis, St. Louis, USA, pp. 28–33, (Accessed March 2011). Lavergne, S., Molofsky, J., 2006. Control strategies for the invasive reed canarygrass (Phalaris arundinacea L.) in North American wetlands: the need for an integrated management plan. Natural Areas Journal 26, 208–214. Legendre, P., Legendre, L., 1998. Numerical Ecology. Elsevier Scientific, Oxford, UK. Lindig-Cisneros, R., Zedler, J.B., 2002. Relationships between canopy complexity and germination microsites for Phalaris arundinacea L. Oecologia 133, 159–167. Malanson, G.P., 1993. Riparian Landscapes. Cambridge University Press, Cambridge, UK.

N.R. De Jager et al. / Forest Ecology and Management 303 (2013) 11–19 McShea, W., 2012. Ecology and management of white-tailed deer in a changing world. Annals of the New York Academy of Sciences 1249, 45–56. Mysterud, A., 2006. The concept of overgrazing and its role in management of large herbivores. Wildlife Biology 12, 129–141. Osborn, T.J., Hulme, M., 2002. Evidence for trends in heavy rainfall events over the UK. Philosophical Transactions of the Royal Society of London Series A 360, 1313–1325. Pezeshki, S.R., 2001. Wetland plant response to soil flooding. Environmental and Experimental Botany 46, 299–312. R Development Core Team., 2008. R: a language and environment for statistical computing. Reinhardt Adams, C., Kauth, P.J., Sorenson, J.W., 2011. Assessing competition between reed canary grass (Phalaris arundinacea) and swamp white oak (Quercus bicolor). Ecological Restoration 29, 332–338. Robson, A.J., 2002. Evidence for trends in UK flooding. Philosophical Transactions of the Royal Society of London Series A 360, 1327–1343. Romano, S.P., 2010. Our current understanding of the Upper Mississippi River System floodplain forest. Hydrobiologia 640, 115–124. Rood, S.B., Braatne, J.H., Goater, L.A., 2010. Favorable fragmentation: river reservoirs can impede downstream expansion of riparian weeds. Ecological Applications 20, 1664–1677. Rooney, T.P., 1997. Escaping herbivory: refuge effects on the morphology and shoot demography of the clonal forest herb, Maianthemum canadense. Journal of the Torrey Botanical Society 124, 280–285. Rooney, T.P., 2009. High white-tailed deer densities benefit graminoids and contribute to biotic homogenization of forest ground layer vegetation. Plant Ecology 202, 103–111. Royo, A.A., Stout, S.L., de Calesta, D.S., Pierson, T.G., 2010. Restoring forest herb communities through landscape-level deer herd reductions: is recovery limited by legacy effects? Biol. Conservation 143, 2425–2434. Ruzicka, K.J., Groninger, J.W., Zaczek, J.J., 2010. Deer browsing, forest edge effects, and vegetation dynamics following bottomland forest restoration. Restoration Ecology 18, 702–710. Sena Gomes, A.R., Kozlowski, T.T., 1980. Growth responses and adaptations of Fraxinus pennsylvanica seedlings to flooding. Plant Physiology 66, 267–271. Senn, J., Haukioja, E., 1994. Reactions of the mountain birch to bud removal: effects of severity and timing and implications for herbivores. Ecology 75, 494–501. Sharitz, R.R., Mitsch, W.J., 1993. Southern floodplain forests. In: Martin, W.H., Boyce, S.G., Echternacht, A.C. (Eds.), Biodiversity of the southeastern United States/ lowland terrestrial communities. Wiley, London, UK, pp. 311–372.

19

Sweeney, B.W., Czapka, S.J., Yerkes, T., 2002. Riparian forest restoration: Increasing success by reducing plant competition and herbivory. Restoration Ecology 10, 392–400. Tanaka, M., Takei, K., Kojima, M., Sakakibara, H., Mori, H., 2006. Auxin controls local cytokinin biosynthesis in the nodal stem in apical dominance. The Plant Journal 45, 1028–1036. Tanentzap, A.J., Bazely, D.R., Koh, S., Timciska, M., Haggith, E.G., Carleton, T.J., Coomes, D.A., 2011. Seeing the forest for the deer: do reductions in deer disturbance lead to forest recovery. Biological Conservation 144, 376– 382. Tanentzap, A.J., Kirby, K.J., Goldberg, E., 2012. Slow responses of ecosystems to reductions in deer (Cervidae) populations and strategies for achieving recovery. Forest Ecology and Management 264, 159–166. Thomsen, M., Brownell, K., Urich, R., Groshek, M., Kirsch, E., 2012. Control of reed canarygrass promotes wetland herb and tree seedling establishment in Upper Mississippi River floodplain. Wetlands 32, 543–555. Waller, D.M., Alverson, W.S., 1997. The white-tailed deer: a keystone herbivore. Wildlife Society Bulletin 25, 217–226. Watson, R.T., Zinyowera, M.C., Moss, R.H., 1998. The regional impacts of climate change: an assessment of vulnerability. Cambridge University Press, Cambridge, UK. Whitlow, T.H., Harris, R.W., 1979. Flood tolerance in plants: a-state-of-the art review, Technical Report E-79-2. US Army Engineer Waterways Experiment Station Environmental Laboratory, Vicksburg, Mississippi, USA. Wisconsin Department of Natural Resources., 2011. . Yin, Y., 1998. Flooding and forest succession in a modified stretch along the Upper Mississippi River. Regulated Rivers: Research and Management 14, 217– 225. Yin, Y., Wu, Y., Bartell, S.M., Cosgriff, R., 2009. Patterns of forest succession and impacts of flood in the Upper Mississippi river floodplain ecosystem. Ecological Complexity 6, 463–472. Zedler, J.B., 2009. Feedbacks that might sustain natural, invaded, and restored states in herbaceous wetlands. In: Hobbs, R.J., Suding, K.N. (Eds.), New models for ecosystem dynamics and restoration (the science and practice of ecological restoration series). Island Press, Washington, DC, USA, pp. 236– 258. Zedler, J.B., Kercher, S., 2004. Causes and consequences of invasive plants in wetlands: opportunities, opportunists, and Outcomes. Critical Reviews in Plant Sciences 23, 431–452.