Ecology, 94(2), 2013, pp. 424–434� 2013 by the Ecological Society of America

Understory plant communities and the functional distinctionbetween savanna trees, forest trees, and pines

JOSEPH W. VELDMAN,1,3 W. BRETT MATTINGLY,1,4 AND LARS A. BRUDVIG2

1Department of Zoology, University of Wisconsin, Madison, Wisconsin 53706 USA2Department of Plant Biology, Michigan State University, East Lansing, Michigan 48824 USA

Abstract. Although savanna trees and forest trees are thought to represent distinctfunctional groups with different effects on ecosystem processes, few empirical studies haveexamined these effects. In particular, it remains unclear if savanna and forest trees differ intheir ability to coexist with understory plants, which comprise the majority of plant diversity inmost savannas. We used structural equation modeling (SEM) and data from 157 sites acrossthree locations in the southeastern United States to understand the effects of broadleafsavanna trees, broadleaf forest trees, and pine trees on savanna understory plant communities.After accounting for underlying gradients in fire frequency and soil moisture, abundances (i.e.,basal area and stem density) of forest trees and pines, but not savanna trees, were negativelycorrelated with the cover and density (i.e., local-scale species richness) of C4 graminoid species,a defining savanna understory functional group that is linked to ecosystem flammability. Inanalyses of the full understory community, abundances of trees from all functional groupswere negatively correlated with species density and cover. For both the C4 and fullcommunities, fire frequency promoted understory plants directly, and indirectly by limitingforest tree abundance. There was little indirect influence of fire on the understory mediatedthrough savanna trees and pines, which are more fire tolerant than forest trees. We concludethat tree functional identity is an important factor that influences overstory tree relationshipswith savanna understory plant communities. In particular, distinct relationships between treesand C4 graminoids have implications for grass–tree coexistence and vegetation–fire feedbacksthat maintain savanna environments and their associated understory plant diversity.

Key words: fire suppression; flammability; functional group; longleaf pine; Pinus palustris; plantdiversity; prescribed fire; Quercus spp.; southeastern United States; species coexistence; woodland.

INTRODUCTION

In savannas, the relationship between trees and

understory vegetation is often framed as an antagonistic

interaction mediated by fire (e.g., van Langevelde et al.

2003). Indeed, tree canopy cover is often negatively

correlated with grass abundance (Scholes 2003), and

surface fires fueled by the herbaceous understory can kill

trees (Prior et al. 2010). Recently, the conceptual

framework of two contrasting life forms (i.e., grasses

and trees) has been replaced by models that recognize

that tree species adapted to savanna environments are

functionally distinct from species adapted to forests

(Hoffmann et al. 2005b, Ratnam et al. 2011). In

particular, savanna trees differ from forest trees in ways

related to fire tolerance (e.g., bark thickness; Hoffmann

et al. 2003), light capture (e.g., specific leaf area; Prior et

al. 2003), fire facilitation (e.g., leaf litter flammability;

Kane et al. 2008), and architecture (e.g., crown area;

Rossatto et al. 2009). Evaluating the functional differ-

ences between savanna and forest trees has improved

our understanding of how tree demographics influence

savanna–forest boundaries (Hoffmann et al. 2009), and

how savanna trees might contribute to ecosystem

flammability via positive feedbacks with fire (Beckage

et al. 2009). Yet, it remains unclear if savanna trees and

forest trees differ in their effects on savanna understory

plant communities, which are often species-diverse

assemblages of C4 graminoids, herbs, and shrubs of

high conservation value (Bond and Parr 2010).

Functional differences between savanna trees and

forest trees likely influence ecosystem processes in

complex ways that, in turn, influence understory plant

communities. For example, savanna trees might facili-

tate understory vegetation by producing highly flamma-

ble leaf litter that helps to carry surface fires, whereas

forest trees produce litter that impedes fire (Fonda 2001,

Kane et al. 2008, Beckage et al. 2009). The crown

architecture of savanna trees allows relatively large

amounts of light to reach plants in the understory,

whereas forest trees intercept more light and thereby

limit understory productivity (Hoffmann et al. 2005a,

Rossatto et al. 2009). Because direct sunlight helps to

dry fuels (and shade can permit fuels to stay too damp to

Manuscript received 18 June 2012; accepted 29 August 2012.Corresponding Editor: B. D. Inouye.

3 E-mail: jveldman@wisc.edu4 Present address: Department of Biology, Eastern Con-

necticut State University, Willimantic, Connecticut 06226USA.

424

burn, but see Hoffmann et al. 2012b), light capture and

litter characteristics interact to influence the accumula-

tion of leaf litter, a factor that can limit understory

vegetation (Hiers et al. 2007). Whether these differences

are relevant for understory plant communities has yet to

be clarified, because most studies that investigate factors

influencing savanna understory plants consider all trees

collectively in canopy cover measurements (e.g., Peter-

son and Reich 2008). In this study, we distinguish

between tree functional groups, and ask: Do savanna

trees, forest trees, and pines differ in their effects on

savanna understory plant cover and diversity?

To answer this question, we used structural equation

modeling (SEM) to evaluate the influences of broadleaf

savanna trees, broadleaf forest trees, and pine trees on

understory plant communities at 157 sites from three

locations in the longleaf pine (Pinus palustris) ecosystem

of the southeastern United States. Like many of the

world’s savannas (e.g., Bond and Parr 2010), fire

suppression in longleaf pine savannas results in

increased abundances of forest trees, the accumulation

of litter, and a decline in understory plant species

(Hartnett and Krofta 1989, Hiers et al. 2007). Most

studies that assess the relationship between tree cover

and understory plants in this ecosystem either do not

distinguish between savanna and forest trees (e.g.,

Brudvig and Damschen 2011), exclude pines from the

analyses (e.g., Hiers et al. 2007), or are focused on pines

alone (e.g., Platt et al. 2006). For this study, we

distinguish broadleaf savanna trees from forest tree

species and consider pines as a distinct functional

group.

SEM is an ideal approach for understanding complex

multivariate problems (Grace 2006), and allows us to

explicitly test hypothesized relationships between tree

functional groups and understory plant communities,

while also incorporating the direct and indirect effects of

underlying environmental gradients (i.e., fire frequency

and soil moisture; Fig. 1). Increased abundance of forest

trees is typically concurrent with fire suppression in

savannas (Roitman et al. 2008, Geiger et al. 2011),

necessitating an approach that disentangles the effects of

trees vs. fire history on understory plant communities

(Peterson et al. 2007). Additionally, prescribed fire and

soil moisture both influence understory species diversity

(Brockway and Lewis 1997, Kirkman et al. 2001) and

tree species distributions (Cavender-Bares et al. 2004) in

the study region; any approach to understanding the

influence of trees on understory vegetation must account

for these gradients.

In this study, we combine a large-scale vegetation data

set with SEM to determine how abundances of savanna

trees, forest trees, and pines relate to understory plant

communities (Fig. 1a). We also examine the relation-

ships between trees and C4 graminoids (Fig. 1b),

motivated by the importance of C4 species as a defining

functional group in savannas and their role in vegeta-

tion–fire feedbacks (Bond and Parr 2010). In sum, this

study provides an assessment of whether the distinction

between tree functional groups is relevant to our

understanding of the controls over savanna understory

plant communities in ways that influence plant diversity

and grass–tree coexistence.

METHODS

We conducted this study in savannas and woodlands

(hereafter ‘‘savannas,’’ due to the prevalence of fire and

coexistence of trees and C4 graminoids; Ratnam et al.

2011) within the historical range of the longleaf pine

ecosystem in the southeastern United States. Longleaf

pine savannas are characterized by an overstory of

scattered pine trees and highly diverse understory plant

communities (Walker and Peet 1983, Peet 2006). Pine

savannas once covered much of the coastal plain in the

southeastern United States (Frost 2006), but currently

occupy ,3% of this original area due to human land use

and fire suppression (Peet 2006). Coexistence of

overstory trees and understory plants is maintained by

frequent, low-intensity surface fires that kill or top-kill

woody plants and consume leaf litter (Drewa et al. 2002,

Thaxton and Platt 2006, Hiers et al. 2007). Although

high abundances of broadleaf trees are associated with

fire suppression (e.g., Gilliam and Platt 1999), certain

fire-tolerant oaks (Quercus spp.) are a natural compo-

nent of the otherwise pine-dominated tree community

(Greenberg and Simons 1999; see Plate 1).

Site selection

We selected 157 study sites at three locations that

support longleaf pine savannas: Fort Bragg, North

Carolina (73 000 ha, elevation, 43–176 m; mean annual

precipitation [MAP], 1270 mm; mean annual tempera-

ture [MAT], 168C; n¼ 61 sites); Savannah River Site (a

National Environmental Research Park), South Caro-

lina (SRS, 80 000 ha, elevation, 20–130 m; MAP, 1225

mm; MAT, 188C; n ¼ 47 sites), and Fort Stewart,

Georgia (114 000 ha, elevation, 2–56 m; MAP, 1220

mm; MAT, 198C; n ¼ 49 sites). These locations span ;

450 km and 38 latitude and are representative of the

hydrological and topographic conditions typical of the

eastern range of the longleaf pine ecosystem (Frost

2006). To inform the selection of study sites, we created

a geographic information system that included historic

maps and aerial photographs from the time of public

acquisition (1919 for Fort Bragg, 1951 for SRS, 1947 for

Fort Stewart), contemporary aerial photographs, and

annual prescribed fire records from 1991 to 2009. We

limited our study to sites that were savannas at the time

of public acquisition (i.e., we excluded formerly

cultivated land) because a history of agriculture results

in reductions in plant diversity (Flinn and Vellend 2005,

Brudvig and Damschen 2011) that could limit our ability

to detect relationships between understory plants and

trees. We ensured that all potential sites were at least

250 m apart and covered �1 ha of relatively uniform

habitat (i.e., sites did not cross topographical, hydro-

February 2013 425UNDERSTORY PLANTS AND SAVANNA TREES

FIG. 1. Path diagrams of structural equation models relating tree functional groups to either (a) all understory plants or (b) C4

graminoids only. Black denotes hypotheses that are the focus of this study (i.e., relationships between trees of different functionalgroups and understory plants). Gray denotes underlying environmental gradients (i.e., fire and soil moisture) that can influenceboth trees and understory plants. Measured variables are represented by rectangles, composite variables by hexagons, and latentvariables by circles. Arrows indicate direction of influence, with solid arrows significant at P , 0.05 and arrow thicknessproportional to the strength of the correlations. Dashed lines indicate hypothesized paths that were modeled but were notsignificant (NS). Dotted lines are paths of fixed (equal) regression weights used to define composite and latent variables.Standardized regression weights are listed for significant (P , 0.05) paths. BA is basal area.

JOSEPH W. VELDMAN ET AL.426 Ecology, Vol. 94, No. 2

logical, or land use boundaries). After identifying

potential sites, we used fire records to select sites that

were stratified across the range of fire frequencies at each

location (Table 1). Although wildfires do occur, the

overwhelming majority of fires are prescribed burns

implemented by managers at each study location.

Vegetation sampling

Within each 1-ha site, we randomly positioned a 20 3

50 m sampling plot. Between August and November

2009, we counted, identified, and measured the diameter

of all trees � 2.5 cm diameter at breast height (1.37 m)

within the plot. To assess species density (i.e., local-scale

species richness), we recorded all plant species present

within each of eight 1-m2 subplots nested within the

larger 203 50 m plot (Brudvig and Damschen 2011) and

calculated the mean number of species per 1-m2 subplot.

To determine understory cover, we estimated the percent

cover for each species in each 1-m2 subplot and used the

sum of all species cover values to determine total

understory cover. We chose to sample at the 1-m2 scale

because species density is a common measurement in

herbaceous communities (Grace 1999), including long-

leaf pine savannas (e.g., Myers and Harms 2009). In

addition to the full understory community, we calculat-

ed the density and cover of C4 graminoids species for

separate analyses. We measured soil moisture holding

capacity (a major determinant of understory diversity

and site productivity in the longleaf pine ecosystem;

Kirkman et al. 2001) by extracting six soil cores (2.5 cm

diameter and 15 cm in depth) at 10-m intervals along the

centerline of the plot; soil cores were homogenized and

processed following Brudvig and Damschen (2011) to

determine moisture content at field capacity by mass.

Tree classification

We used literature from the southeastern United

States related to tree habitat preferences along fire

gradients (Harnett and Krofta 1989, Cavender-Bares et

al. 2004) to classify all trees as broadleaf savanna trees,

broadleaf forest trees, or needleleaf trees (all pines),

hereafter referred to as ‘‘savanna trees,’’ ‘‘forest trees,’’

and ‘‘pines,’’ respectively (Appendix A). Habitat associ-

ations are used to distinguish savanna trees and forest

trees in Neotropical savannas (e.g., Hoffmann et al.

2003, Veldman and Putz 2011), and in combination with

published data on bark thickness (Jackson et al. 1999),

our savanna tree classification is consistent with

definitions of savanna trees used internationally (Rat-

nam et al. 2011). Most of the pines in our study fit the

general model of savanna trees (i.e., thick bark and

highly flammable litter) but they differ from broadleaf

savanna trees in a number of ways that warrant

treatment as a functionally distinct group. Among these

many functional differences (e.g., Cook et al. 2001) are

thicker bark than broadleaf savanna trees (Jackson et al.

1999) and vastly different leaf morphologies. All tree

species not defined as savanna trees or pines were

considered forest trees, based on their relatively thin

bark (Jackson et al. 1999) and typical habitat associa-

tions with infrequently burned mesic hardwood forests

(Harnett and Krofta 1989, Jacqmain et al. 1999).

TABLE 1. Summary of measured variables included in the structural equation models from 157 sites at three locations in thesoutheastern USA.

Variable Mean Range Transformation Description

Understory species density (species/m2) 6.6 1.5–17.9 log10 mean number of plant species per 1 m2

subplot; 8 subplots/siteUnderstory plant cover (total %) 71 7–225 log10 mean total percent cover for all plant species

in 8 1-m2 subplots/siteC4 species density (species/m2) 1.1 0–3.1 log10 mean number of C4 graminoid species per 1

m2 subplot; 8 subplots/siteC4 cover (total %) 11 0–63 log10 mean total percent cover for all C4

graminoid species in 8 1-m2 subplots/siteSavanna BA (m2/ha) 0.7 0–9.4 log10 basal area of savanna trees � 2.5 cm dbh in

one 20 3 50 m plot/siteForest BA (m2/ha) 1.0 0–21.3 log10 basal area of forest trees � 2.5 cm dbh in

one 20 3 50 m plot/sitePine BA (m2/ha) 15.1 1.3–35.7 none basal area of pine trees � 2.5 cm dbh in one

20 3 50 m plot/siteSavanna stems (stems/ha) 109 0–2030 log10 density of savanna trees � 2.5 cm dbh in

one 20 3 50 m plot/siteForest stems (stems/ha) 97 0–1520 log10 density of forest trees � 2.5 cm dbh in one

20 3 50 m plot/sitePine stems (stems/ha) 342 30–2350 log10 density of pine trees � 2.5 cm dbh in one 20

3 50 m plot/siteFire frequency (number of fires) 5 0–17 none number of prescribed fires between 1991 and

2009Soil moisture holding capacity (% by mass) 41 30–57 none moisture content at field capacity

determined from a composite of six 20-cmsoil cores per site

Note: Locations were: Fort Bragg, North Carolina; Savanna River Site, South Carolina; and, Fort Stewart, Georgia.

February 2013 427UNDERSTORY PLANTS AND SAVANNA TREES

Structural equation modeling

We developed two structural equation models (SEmodels) to test the direct effects of savanna trees, forest

trees, and pines on (1) all understory plants and (2) C4

graminoids only, while accounting for the direct and

indirect effects of soil moisture availability and firefrequency. We began by specifying our hypotheses in

multivariate conceptual models that formed the basis forthe SEM (Fig. 1). Each model included 10 observed

variables (Table 1), three composite overstory variables(savanna trees, forest trees, and pines), and one latent

understory variable (Fig. 1). As such, we modeled theunderstory community as a latent variable that was

represented by an equally weighted combination oflocal-scale species richness and total plant abundance.

We modeled overstory trees in each functional group ascomposite variables defined by stem density and basal

area. Latent variables are unmeasured variables (oftentheoretical constructs) for which we have no directmeasure but that we infer from other (often measured)

variables. Composite variables are unmeasured variablescompletely specified by causal (often measured) vari-

ables (see Grace et al. 2010 for an explanation of latentand composite variables). Prior to fitting the models, we

generated bivariate plots and univariate density plots inR 2.9.2 (R Development Core Team 2009) to assess the

linearity of relationships and skewness of variables(Appendix B). Because SEM models linear relationships

between variables, and because extreme skew can affectcovariance (Grace 2006), we applied a log10 transfor-

mation to all of the understory variables and most of thetree variables (Table 1; Appendix B); no transforma-

tions were required for fire frequency, soil moistureholding capacity, or pine basal area. We fit the SE

models using the maximum likelihood function in IBMSPPS Amos 20.0 (Amos Development Corporation,Meadville, Pennsylvania, USA).

We combined data from all three study locations to

achieve analyses that both encompass a wide range ofconditions that support longleaf pine savannas, and thatspan regional gradients in edaphic factors that influence

tree–understory interactions. A statistical reason foranalyzing all locations collectively, SEM requires large

data sets, and models repeated at the location levelwould risk spurious results due to small sample sizes

(Grace 2006). Although statistical analyses were per-formed on all sites combined (n ¼ 157), we coded

bivariate graphics to facilitate visual interpretation ofhow sites from each location contribute to results.

Additional statistical analyses

A challenge to presenting SEM results is that, whilepotentially informative, multivariate relationships can-

not be easily plotted in two dimensions. Additionally,relationships identified in SE models can be obscured ormisrepresented by plotting simple bivariate relation-

ships, if strong underlying gradients exist among sites, asin our study. In an attempt to provide visually

interpretable bivariate plots between the tree and

understory variables in our models, while controlling

for underlying gradients, we plotted residuals of

understory variables against tree abundances by func-

tional group. To calculate the residuals, we used the lm

function in R 2.9.2 (R Development Core Team 2009) to

model the linear main effects of fire frequency and soil

moisture holding capacity on understory species density,

total cover, C4 species density, and C4 cover. Similar to

SEM, this analysis accounts for underlying gradients in

fire and soil moisture. Unlike SEM, this approach

ignores covariance in abundances between tree func-

tional groups and does not allow for composite or latent

variables. As such, these residual bivariate plots should

be interpreted as supplementary to the SEM, with the

understanding that they do not incorporate much of the

complexity of the SE models.

RESULTS

Our study sites encompassed a broad range of

conditions spanning fire-maintained to fire-suppressed

longleaf pine savannas (Table 1). Our sites varied in

total tree stem density (80–2950 stems/ha) and basal

area (4–43 m2/ha), with wide ranges in savanna tree,

forest tree, and pine tree abundances (Table 1) that

allow us to test hypotheses about the effects of tree

functional groups on understory plant communities. We

encountered a total of 35 tree species, including four

savanna species, 27 forest species, and four species of

pine (Appendix A). Not surprisingly, pines accounted

for the majority of all stems (63% relative abundance),

followed by savanna trees (20%), and forest trees (17%).

Pines also accounted for the majority of total tree basal

area (90% relative dominance) compared to savanna and

forest trees (4.3% and 5.7%, respectively). See Appendix

A for lists of tree species, frequencies, and abundances.

Full understory community

The SE model relating tree functional groups to

understory plant species density and cover accounted for

46% of the variation in the understory plant community

(R2¼ 0.46, Fig. 1a). Savanna trees, forest trees, and pine

trees each had direct negative effects on understory

plants (Fig. 1a) driven by negative relationships between

tree abundances and understory species density (Fig.

2a, b). Relationships between trees and understory cover

were less consistent, with stem density unrelated to

understory cover for each tree functional group (Fig.

2b). Fire frequency had a direct positive effect on

understory plants, as well as an indirect positive effect

by limiting forest trees (Fig. 1a). Soil moisture had a

direct positive influence on understory richness and pine

basal area, and a negative relationship with savanna

stems (Fig. 1a). The model was relatively effective at

explaining variation in forest tree stem density (R2 ¼0.26), forest tree basal area (R2 ¼ 0.17), and savanna

tree stem densities (R2 ¼ 0.11), but not pine basal area

JOSEPH W. VELDMAN ET AL.428 Ecology, Vol. 94, No. 2

FIG. 2. Full understory plant community species density (species/m2) and cover (total %) in relation to (a) tree basal area (m2/ha) and (b) tree stem density (stems/ha) of savanna trees, forest trees, and pines. To account for effects of fire and soil moisture,understory variables are displayed as residuals from generalized linear models of the main effects of fire frequency and soilmoisture. Symbols correspond to the three study locations. Linear regression lines are shown for significant relationships (P , 0.05)between tree variables and understory residuals. Note the different x-axis scale for pine basal area compared to savanna and foresttrees.

February 2013 429UNDERSTORY PLANTS AND SAVANNA TREES

(R2 ¼ 0.07), pine density (R2 ¼ 0.01), or savanna tree

basal area (R2 ¼ 0.02).

C4 graminoids

The SE model relating tree functional groups to C4

species density and cover accounted for 56% of the

variation in C4 graminoids (R2¼ 0.56; Fig. 1b). Whereas

forest trees and pine trees were negatively correlated

with the C4 understory community (Fig. 1b), savanna

tree abundance was unrelated to C4 graminoids. Overall

negative relationships of forest trees and pines with C4

graminoids appear to be primarily due to negative

effects of forest tree and pine basal area on C4 species

density and cover (Fig. 3a). In supporting analyses (Fig.

3a, b), there was no relationship between savanna stem

density and C4 species density and cover, and no

relationship between savanna tree basal area and C4

cover, but there was a weak negative relationship

between savanna basal area and C4 species density. Fire

frequency had a direct positive effect on C4 graminoids,

as well as an indirect positive effect by limiting forest

trees (Fig. 1b). Unlike the full community analysis, soil

moisture was not correlated with C4 graminoids (Fig.

1b). Because of the identical structure of the full

community and C4 graminoid models, relationships

between fire frequency, soil moisture, and tree abun-

dances are identical in both models (Fig. 1a, b).

DISCUSSION

Our results show that savanna trees, in contrast to

forest trees and pines, do not limit a defining group of

savanna understory plants: C4 graminoids. Tree func-

tional identity was also important in mediating the

indirect effects of fire on the full understory, although all

functional groups were negatively correlated with

understory diversity and plant cover. By linking tree

functional identity to understory plant community

diversity and C4 graminoid abundance, our results

support a growing body of research on the distinct roles

of tree functional groups in savanna vegetation dynam-

ics (Hoffmann et al. 2012a).

The full understory results are consistent with

numerous studies showing that increasing tree abun-

dances are concurrent with declines in savanna under-

story plant communities (e.g., Ratajczak et al. 2012). By

incorporating fire frequency in our models, our results

extend this literature to show that trees exert direct

negative control over understory communities and are

not simply correlated with declines in understory

communities driven by fire suppression. Further, our

results suggest that aboveground differences in tree

functional groups (i.e., those associated with light

interception and litter characteristics) do not result in

qualitatively different effects on full understory plant

community diversity and cover (although the negative

overstory effect on understory plants was weakest for

savanna trees; Fig. 1a). It is plausible that the negative

effects of trees could be due to similar function in

belowground processes among tree groups (e.g., root

competition for limiting resources; Harrington et al.

2003); little is known about the belowground functional

differences (or equivalencies) among tree functional

groups, and more work is needed to assess this

hypothesis.

In contrast to effects on understory communities as a

whole, savanna trees appear better able to coexist with

C4 graminoids compared to forest trees and pines. Our

results do not support the hypothesis that savanna trees

facilitate (i.e., are positively correlated with) C4 species,

but rather that C4 graminoid diversity and abundance

are controlled by factors other than savanna tree

abundance (Fig. 1b). Among these factors are fire,

which strongly promotes C4 species density and cover,

and pine and forest tree abundances, which are

negatively correlated with C4 graminoids. Although

distinct functional characteristics might explain coexis-

tence of savanna trees and C4 species, high rates of self-

thinning (Sea and Hanan 2012) and inferior competitive

abilities relative to shade-tolerant forest trees (e.g.,

Geiger et al. 2011) could also contribute to coexistence.

Indeed, maximum stem densities for savanna trees,

forest trees, and pines were comparable (Table 1), yet

maximum savanna tree basal area (9.4 m2/ha) was just

half that of forest trees (21.3 m2/ha) and only one-

quarter of maximum pine basal area (35.7 m2/ha);

savanna trees may not reach sufficient basal area to limit

C4 cover. Although the mechanisms need further

exploration, our results do support the differentiation

between savanna and forest trees (Ratnam et al. 2011)

by providing evidence that savanna trees and forest trees

differ in their relationships with C4 graminoids, a

distinction with important implications for savanna–

forest dynamics (Hoffmann et al. 2009) and grass–fire

feedbacks (Hoffmann et al. 2012b).

Pine abundance in this study was negatively correlat-

ed with both C4 graminoids and full understory

community species density and cover. Pines are consid-

ered a keystone functional group in the longleaf pine

ecosystem because they produce fire-promoting leaf

litter (Fonda 2001) and are associated with high-

diversity plant communities (Walker and Peet 1983).

Pines likely limit succession of savannas to forests via

vegetation–fire feedbacks (e.g., Beckage et al. 2009); in

this regard, pines may be viewed as facilitators of

understory plant communities, at least relative to

infrequently burned mesic forests (e.g., Nowacki and

Abrams 2008). Nonetheless, across the range of sites we

sampled there was a negative influence of pines on

understory plants. This result is consistent with studies

that document increased species richness following

removal of longleaf pine overstory trees (Platt et al.

2006). Surprisingly, pine tree abundances were unaffect-

ed by fire frequency in our study. This may be because

prescribed fires are implemented within prescription

conditions (i.e., high fuel moisture, low ambient

temperatures) that are unlikely to kill fire-tolerant pines

JOSEPH W. VELDMAN ET AL.430 Ecology, Vol. 94, No. 2

FIG. 3. C4 graminoid species density (species/m2) and cover (total %) in relation to (a) tree basal area (m2/ha) and (b) tree stemdensity (stems/ha) of savanna trees, forest trees, and pines. To account for effects of fire and soil moisture, understory variables aredisplayed as residuals from generalized linear models of the main effects of fire frequency and soil moisture. Symbols correspond tothe three study locations. Linear regression lines are shown for significant relationships (P , 0.05) between tree variables and C4

residuals. Note the different x-axis scale for pine basal area compared to savanna and forest trees.

February 2013 431UNDERSTORY PLANTS AND SAVANNA TREES

(e.g., Beckage et al. 2005), or because pine abundance

reflects historical timber management more so than

contemporary fire frequency.

Among trees in this study that fit the general model of

savanna species (Ratnam et al. 2011), broadleaf savanna

trees and pines appear to represent two functional

groups that differ in their effects on C4 graminoids and

relationship with fire (Fig. 1b), including different effects

on fire behavior (Wenk et al. 2011). Pines are dominant,

fire-promoting, and highly fire tolerant compared to

broadleaf savanna trees, which are relatively less

abundant, less flammable, and less fire tolerant (but

far more fire tolerant than forest trees; Fig. 1a, b;

Appendix B). With two-functional groups of fire-

adapted trees (one dominant and fire promoting), the

longleaf pine ecosystem may have more in common with

highly flammable eucalypt-dominated savannas of

Australia (Lawes et al. 2011, Bond et al. 2012) than

with savannas of Africa and South America that lack

highly flammable tree species (Lehmann et al. 2011).

Studies seeking to understand global distributions of

savannas or tree coexistence with savanna understory

plants should consider not only the distinction between

savanna trees and forest trees (Ratnam et al. 2011,

Hoffmann et al. 2012a), but also functional variation

within fire-adapted trees as a group (Lawes et al. 2011,

Wenk et al. 2011). We should expect these different

functional groups to have different effects on savanna–

forest dynamics (Beckage et al. 2009, Lehmann et al.

2011), and thus different consequences for understory

plant communities.

Our C4 model (Fig. 1b), in combination with previous

work on grass and leaf litter flammability, suggests that

savanna trees, forest trees, and pines may influence fire–

vegetation relationships via different mechanisms.

Broadleaf savanna trees are able to coexist with C4

graminoids and thereby permit the accumulation of

graminoid fuels that are the primary fine fuel source in

most savannas (Hoffmann et al. 2012b). Unlike broad-

leaf savanna trees, pines limit C4 graminoids but

nonetheless contribute to ecosystem flammability by

producing highly flammable leaf litter (Fonda 2001). In

contrast, forest trees are incompatible with savanna

environments because they are susceptible to fire and

suppress understory plant communities, including light-

demanding, fire-promoting C4 graminoids (Hoffmann et

al. 2012a). By studying understory plant communities in

relation to overstory trees, we can achieve a more

comprehensive understanding of the factors that main-

tain savanna environments and the plant diversity they

support.

CONCLUSIONS

This study links overstory tree functional identity to

savanna understory plant communities, and shows that

the distinction between savanna trees, forest trees, and

pines is important for understanding tree coexistence

with C4 graminoids. The functional distinction among

trees appears less important for the full understory

community, given that abundances of all tree functional

groups were negatively correlated with understory

species density and cover. By illustrating relationships

between tree functional groups and savanna understory

PLATE 1. Example of a longleaf pine savanna at Ft. Stewart, Georgia, USA, with fire-tolerant broadleaf savanna trees and anunderstory dominated by C4 grasses. Photo credit: J. W. Veldman.

JOSEPH W. VELDMAN ET AL.432 Ecology, Vol. 94, No. 2

plant communities, we provide the basis for future

inquiry into likely mechanisms (e.g., light interception,

litter flammability, belowground competition) by which

trees interact with understory plant communities. As a

practical implication of this study, we suggest that

ecosystem managers wishing to promote C4 grasses and

sedges via management of overstory trees should

explicitly consider tree species functional groups and,

in our study system, manage abundances of forest trees

(through fire) and pines (through cutting), but not

remove broadleaf savanna tree species.

ACKNOWLEDGMENTS

We thank E. Damschen, J. Orrock, and J. Walker for helpingto develop the network of research sites associated with thisstudy. Thanks to L. Bizarri, C. Christopher, C. Collins, A.Powell, and R. Ranalli for help with vegetation sampling. Forlogistical support we thank: the Fish and Wildlife Branch andForestry Branch of Fort Stewart; the USDA Forest Service–Savannah River; and the Endangered Species Branch, ForestryBranch, and the Cultural Resources Program of Fort Bragg.This project was funded by the Strategic EnvironmentalResearch and Development Program (Project RC-1695) andby the Department of Agriculture, Forest Service, SavannahRiver, Interagency Agreement (DE-AI09-00SR22188) with theDepartment of Energy, Aiken, South Carolina. E. Grman, N.Swenson, and R. Globus Veldman provided helpful suggestionson this manuscript.

LITERATURE CITED

Beckage, B., W. J. Platt, and L. J. Gross. 2009. Vegetation, fire,and feedbacks: a disturbance-mediated model of savannas.American Naturalist 174:805–818.

Beckage, B., W. J. Platt, and B. Panko. 2005. A climate-basedapproach to the restoration of fire-dependent ecosystems.Restoration Ecology 13:429–431.

Bond, W. J., G. D. Cook, and R. J. Williams. 2012. Which treesdominate in savannas? The escape hypothesis and eucalyptsin northern Australia. Austral Ecology 37:678–685.

Bond, W. J., and C. L. Parr. 2010. Beyond the forest edge:ecology, diversity and conservation of the grassy biomes.Biological Conservation 143:2395–2404.

Brockway, D. G., and C. E. Lewis. 1997. Long-term effects ofdormant-season prescribed fire on plant community diversi-ty, structure and productivity in a longleaf pine wiregrassecosystem. Forest Ecology and Management 96:167–183.

Brudvig, L. A., and E. I. Damschen. 2011. Land-use history,historical connectivity, and land management interact todetermine longleaf pine woodland understory richness andcomposition. Ecography 34:257–266.

Cavender-Bares, J., K. Kitajima, and F. A. Bazzaz. 2004.Multiple trait associations in relation to habitat differentia-tion among 17 Floridian oak species. Ecological Monographs74:635–662.

Cook, E. R., J. S. Glitzenstein, P. J. Krusic, and P. A.Harcombe. 2001. Identifying functional groups of trees inwest Gulf Coast forests (USA): a tree-ring approach.Ecological Applications 11:883–903.

Drewa, P. B., W. J. Platt, and F. B. Moser. 2002. Fire effects onresprouting of shrubs in headwaters of southeastern longleafpine savannas. Ecology 83:755–767.

Flinn, K. M., and M. Vellend. 2005. Recovery of forest plantcommunities in post-agricultural landscapes. Frontiers inEcology and the Environment 3:243–250.

Fonda, R. W. 2001. Burning characteristics of needles fromeight pine species. Forest Science 47:390–396.

Frost, C. C. 2006. History and future of the longleaf pineecosystem. Pages 9–42 in S. Jose, E. J. Jokela, and D. L.

Miller, editors. The longleaf pine ecosystem: ecology,silviculture, and restoration. Springer, New York, NewYork, USA.

Geiger, E. L., S. G. Gotsch, G. Damasco, M. Haridasan, A. C.Franco, and W. A. Hoffmann. 2011. Distinct roles ofsavanna and forest tree species in regeneration under firesuppression in a Brazilian savanna. Journal of VegetationScience 22:312–321.

Gilliam, F. S., and W. J. Platt. 1999. Effects of long-term fireexclusion on tree species composition and stand structure inan old-growth Pinus palustris (longleaf pine) forest. PlantEcology 140:15–26.

Grace, J. B. 1999. The factors controlling species density inherbaceous plant communities: an assessment. Perspectives inPlant Ecology, Evolution and Systematics 2:1–28.

Grace, J. B. 2006. Structural equation modeling and naturalsystems. Cambridge University Press, Cambridge, UK.

Grace, J. B., T. M. Anderson, H. Olff, and S. M. Scheiner.2010. On the specification of structural equation models forecological systems. Ecological Monographs 80:67–87.

Greenberg, C. H., and R. W. Simons. 1999. Age, composition,and stand structure of old-growth oak sites in the Floridahigh pine landscape: implications for ecosystem managementand restoration. Natural Areas Journal 19:30–40.

Harrington, T. B., C. M. Dagley, and M. B. Edwards. 2003.Above- and belowground competition from longleaf pineplantations limits performance of reintroduced herbaceousspecies. Forest Science 49:681–695.

Hartnett, D. C., and D. M. Krofta. 1989. Fifty-five years ofpost-fire succession in a southern mixed hardwood forest.Bulletin of the Torrey Botanical Club 116:107–113.

Hiers, J. K., J. J. O’Brien, R. E. Will, and R. J. Mitchell. 2007.Forest floor depth mediates understory vigor in xeric Pinuspalustris ecosystems. Ecological Applications 17:806–814.

Hoffmann, W. A., R. Adasme, M. Haridasan, M. T. Carvalho,E. L. Geiger, M. A. B. Pereira, S. G. Gotsch, and A. C.Franco. 2009. Tree topkill, not mortality, governs thedynamics of savanna–forest boundaries under frequent firein central Brazil. Ecology 90:1326–1337.

Hoffmann, W. A., A. C. Franco, M. Z. Moreira, and H.Haridasan. 2005a. Specific leaf area explains differences inleaf traits between congeneric savanna and forest trees.Functional Ecology 19:932–940.

Hoffmann, W. A., E. L. Geiger, S. G. Gotsch, D. R. Rossatto,L. C. R. Silva, O. L. Lau, M. Haridasan, and A. C. Franco.2012a. Ecological thresholds at the savanna–forest boundary:how plant traits, resources and fire govern the distribution oftropical biomes. Ecology Letters 15:759–768.

Hoffmann, W. A., S. Y. Jaconis, K. L. Mckinley, E. L. Geiger,S. G. Gotsch, and A. C. Franco. 2012b. Fuels ormicroclimate? Understanding the drivers of fire feedbacksat savanna–forest boundaries. Austral Ecology 37:634–643.

Hoffmann, W. A., B. Orthen, and P. K. V. Nascimento. 2003.Comparative fire ecology of tropical savanna and forest trees.Functional Ecology 17:720–726.

Hoffmann, W. A., E. R. Silva, G. C. Machado, S. J. Bucci,F. G. Scholz, G. Goldstein, and F. C. Meinzer. 2005b.Seasonal leaf dynamics across a tree density gradient in aBrazilian savanna. Oecologia 145:307–316.

Jackson, J. F., D. C. Adams, and U. B. Jackson. 1999.Allometry of constitutive defense: a model and a comparativetest with tree bark and fire regime. American Naturalist153:614–632.

Jacqmain, E. I., R. H. Jones, and R. J. Mitchell. 1999.Influences of frequent cool-season burning across a soilmoisture gradient on oak community structure in longleafpine ecosystems. American Midland Naturalist 141:85–100.

Kane, J. M., J. M. Varner, and J. K. Hiers. 2008. The burningcharacteristics in southeastern oaks: discriminating firefacilitators from fire impeders. Forest Ecology and Manage-ment 256:2039–2045.

February 2013 433UNDERSTORY PLANTS AND SAVANNA TREES

Kirkman, L. K., R. J. Mitchell, R. C. Helton, and M. B. Drew.2001. Productivity and species richness across an environ-mental gradient in a fire-dependent ecosystem. AmericanJournal of Botany 88:2119–2128.

Lawes, M. J., B. P. Murphy, J. J. Midgley, and J. Russell-Smith. 2011. Are the eucalypt and non-eucalypt componentsof Australian tropical savannas independent? Oecologia166:229–239.

Lehmann, C. E. R., S. A. Archibald, W. A. Hoffmann, andW. J. Bond. 2011. Deciphering the distribution of thesavanna biome. New Phytologist 191:197–209.

Myers, J. A., and K. E. Harms. 2009. Local immigration,competition from dominant guilds, and the ecologicalassembly of high-diversity pine savannas. Ecology 90:2745–2754.

Nowacki, G. J., and M. D. Abrams. 2008. The demise of fireand ‘‘mesophication’’ of the eastern United States. BioSci-ence 58:123–138.

Peet, R. K. 2006. Ecological classification of longleaf pinewoodlands. Pages 51–93 in S. Jose, E. J. Jokela, and D. L.Miller, editors. The longleaf pine ecosystem: ecology,silviculture, and restoration. Springer, New York, NewYork, USA.

Peterson, D. W., and P. B. Reich. 2008. Fire frequency and treecanopy structure influence plant species diversity in a forest–grassland ecotone. Plant Ecology 194:5–16.

Peterson, D. W., P. B. Reich, and K. J. Wrage. 2007. Plantfunctional group responses to fire frequency and tree canopycover gradients in oak savannas and woodlands. Journal ofVegetation Science 18:3–12.

Platt, W. J., S. M. Carr, M. Reilly, and J. Fahr. 2006. Pinesavanna overstorey influences on ground-cover biodiversity.Applied Vegetation Science 9:37–50.

Prior, L. D., D. Eamus, and D. M. J. S. Bowman. 2003. Leafattributes in the seasonally dry tropics: a comparison of fourhabitats in northern Australia. Functional Ecology 17:504–515.

Prior, L. D., R. J. Williams, and D. M. J. S. Bowman. 2010.Experimental evidence that fire causes a tree recruitmentbottleneck in an Australian tropical savanna. Journal ofTropical Ecology 26:595–603.

R Development Core Team. 2009. R version 2.9.2. RFoundation for Statistical Computing, Vienna, Austria.

Ratajczak, Z., J. B. Nippert, and S. L. Collins. 2012. Woodyencroachment decreases diversity across North Americangrasslands and savannas. Ecology 93:697–703.

Ratnam, J., W. J. Bond, R. J. Fensham, W. A. Hoffmann, S.Archibald, C. E. R. Lehmann, M. T. Anderson, S. I. Higgins,and M. Sankaran. 2011. When is a ‘forest’ a savanna, andwhy does it matter? Global Ecology and Biogeography20:653–660.

Roitman, I., J. M. Felfili, and A. V. Rezende. 2008. Treedynamics of a fire-protected cerrado sensu stricto surroundedby forest plantations, over a 13-year period (1991–2004) inBahia, Brazil. Plant Ecology 197:255–267.

Rossatto, D. R., W. A. Hoffmann, and A. C. Franco. 2009.Differences in growth patterns between co-occurring forestand savanna trees affect the forest–savanna boundary.Functional Ecology 23:689–698.

Scholes, R. J. 2003. Convex relationships in ecosystemscontaining mixtures of trees and grass. Environmental andResource Economics 26:559–574.

Sea, W. B., and N. P. Hanan. 2012. Self-thinning and treecompetition in savannas. Biotropica 44:189–196.

Thaxton, J. M., and W. J. Platt. 2006. Small-scale fuel variationalters fire intensity and shrub abundance in a pine savanna.Ecology 87:1331–1337.

van Langevelde, F., C. A. D. M. van de Vijver, L. K. J. van deKoppel, N. de Ridder, J. van Andel, A. K. Skidmore, J. W.Hearne, L. Stroosnijder, W. J. Bond, H. H. T. Prins, and M.Rietkerk. 2003. Effects of fire and herbivory on the stabilityof savanna ecosystems. Ecology 84:337–350.

Veldman, J. W., and F. E. Putz. 2011. Grass-dominatedvegetation, not species-diverse natural savanna, replacesdegraded tropical forests on the southern edge of theAmazon Basin. Biological Conservation 144:1419–1429.

Walker, J. L., and R. K. Peet. 1983. Composition and speciesdiversity of pine–wiregrass savannas of the Green Swamp,North Carolina. Vegetatio 55:163–179.

Wenk, S. E., G. G. Wang, and J. L. Walker. 2011. Within-standvariation in understorey vegetation affects fire behaviour inlongleaf pine xeric sandhills. International Journal ofWildland Fire 20:866–875.

SUPPLEMENTAL MATERIAL

Appendix A

Classification, frequencies, and abundances of tree species (Ecological Archives E094-036-A1).

Appendix B

Bivariate plots of all (untransformed) measured variables included in the SEM (Ecological Archives E094-036-A2).

JOSEPH W. VELDMAN ET AL.434 Ecology, Vol. 94, No. 2