Spatial Dependency of Vegetation?
Environment Linkages in an
Anthropogenically Influenced
Wetland Ecosystem
Ryan S. King,1,2* Curtis J. Richardson,1 Dean L. Urban,1 and
Edwin A. Romanowicz1
1Nicholas School of the Environment and Earth Sciences, Duke University, Box 90328, Durham, North Carolina 27708, USA;
2Smithsonian Environmental Research Center, Box 28, Edgewater, Maryland 21037, USA
ABSTRACT
Management and restoration of vegetation patterns
in ecosystems depends on an understanding of al-
logenic environmental factors that organize species
assemblages and autogenic processes linked to as-
semblages. However, our ability to make strong
inferences about vegetation?environment linkages
in field studies is often limited due to correlations
among environmental variables, spatial autocorre-
lation, and scale dependency of observations. This is
particularly true in large, heterogeneous ecosys-
tems such as the Everglades. Here, an extensive
canal-and-levee system has modified historical fire
regimes and hydropatterns while contributing large
inputs of surface-water phosphorus (P), nitrogen
(N) and cations such as sodium (Na). Some of these
anthropogenic influences have been implicated as
factors leading to the shift of sawgrass (Cladium
jamaicense Crantz) and slough communities to an
assemblage of weedy species such as cattail (Typha
domingensis Pers.). To untangle the independent ef-
fect of multiple variables, we used a spatially ex-
plicit, multivariate approach to identify linkages
among spatial patterns, environmental factors, and
vegetation composition along a 10-km gradient of
anthropogenic influence in the Everglades, an area
immediately downstream from canal inflow struc-
tures. Clusters of plots were stratified among three
zones (Impacted, Transition, and Reference), a design
that allowed us to contrast vegetation?environ-
ment linkages and spatial patterns at multiple scales
and degrees of ecosystem alteration. Along the
10-km gradient, partial Mantel tests showed that
nutrients (phosphorus, nitrogen, and potassium)
and hydropattern (frequency of dryness) were in-
dependently linked to patterns in fine-scale vegeta-
tion composition, but phosphorus was the only en-
vironmental variable linked to patterns of coarse-
scale composition. Regardless of scale, the effect of
distance from canal inflows accounted for variation
in vegetation that could not be explained by other
variables. A significant residual effect of spatial
proximity among sampling locations also was de-
tected and was highly suggestive of dispersal or
other spatial determinants of vegetation pattern.
However, this pure spatial effect was significantly
stronger in the Transition and Impacted zones than
in the Reference zone?fine-scale environmental
variables explained all of the spatial structure in
vegetation in the Reference zone. A further exam-
ination of spatial patterns in vegetation by using
Mantel correlograms revealed significant heteroge-
neity at fine, local scales in the Reference zone, but
this pattern progressively degraded toward homo-
geneity among closely neighboring locations in the
Impacted zone. However, the fine-scale vegetation
pattern in the Reference zone was hierarchically
nested at a broader scale and yielded a similar
coarse pattern across the landscape, whereas the
coarse pattern in the Transition and Impacted zones
Received 10 July 2002; accepted 2 January 2003; published online 12
January 2004.
*Corresponding author; e-mail: KingRy@si.edu
Ecosystems (2004) 7: 75?97
DOI: 10.1007/s10021-003-0210-4 ECOSYSTEMS
? 2004 Springer-Verlag
75
was relatively heterogeneous and fragmented. Col-
lectively, these results indicate that allogenic spatial
and environmental factors related to the canal sys-
tem have disrupted the coupling between pattern
and process by altering fine-scale vegetation?envi-
ronment linkages and spatial patterns characteristic
of the natural Everglades ecosystem.
Key words: scale; pattern; hierarchy theory; mac-
rophytes; partial Mantel test; ordination; spatial au-
tocorrelation; hydropattern; nutrients; Everglades.
INTRODUCTION
Environmental factors are traditionally considered
one of the predominant determinants of vegetation
pattern [for example, see Whittaker (1956) and
Bray and Curtis (1957)], a view that has resulted in
numerous observational studies of community?en-
vironment relationships in plant ecology. Unfortu-
nately, studies of this type often produce equivocal
results due to the confounded nature of ecological
data [for example, see Legendre (1993) and Thom-
son and others (1997)]. Correlations among envi-
ronmental variables, spatial autocorrelation, and
scale dependency of observations all degrade the
interpretative value of results produced from many
analytical techniques in use today [see Legendre
and Legendre (1998)]. This is especially true for
large-scale observational studies in which supple-
mentary experimental manipulations are either im-
practical or impossible to produce at the appropriate
scale (Legendre and Fortin 1989; Wiens 1989), a
fact that is troubling given our current awareness of
the importance of landscape-scale processes in ecol-
ogy (Levin 1992). Thus, conducting field studies
that afford strong inference about vegetation?envi-
ronment linkages can be a daunting task, particu-
larly in large, heterogeneous ecosystems (Beyers
1998; Urban 2000).
The Everglades is an example of a large, hetero-
geneous ecosystem in which vegetation?environ-
ment linkages have received much attention. Veg-
etation in this wetland ecosystem has been affected
by a variety of anthropogenic influences in the past
several decades, which has led to a surge of recent
studies designed to infer the causes of observed
changes. Reputedly sustained by fire and hydropat-
tern (Loveless 1959; Craighead 1971) while limited
by phosphorus (Steward and Ornes 1975a, 1975b;
Davis 1991; Noe and others 2001), the historic het-
erogeneous mosaic of vegetation communities de-
scribed by Davis (1943) has been altered by disrup-
tion of natural environmental variation and
degradation of water quality (SFWMD 1992; Davis
and Ogden 1994). However, intercorrelations and
spatial autocorrelation of multiple factors have
made it difficult to isolate the linkages between
specific environmental variables and observed
changes in vegetation patterns.
Although modifications to hydrology, fire fre-
quency and intensity, and other environmental fac-
tors are suggested to play a role in the alteration of
plant distributions in the Everglades, phosphorus
(P)-enriched runoff from the Everglades Agricul-
tural Area (EAA) has been identified as the primary
stressor (SFWMD 1992). The extensive canal-and-
levee system that compartmentalizes the remnant
Everglades also serves as a conduit for P from the
EAA, and water-control structures along the canals
function as point sources of P to downstream por-
tions of the wetland ecosystem. In areas near water-
control structures, P has been found to be at least
partially responsible for the transformation of Cla-
dium jamaicense Crantz (sawgrass) stands and open-
water sloughs to dense stands of invasive Typha
domingensis Pers. (cattail) (Davis 1991; Urban and
others 1993; Newman and others 1998). Typha dis-
tribution and growth is positively correlated with
both soil and water total P and is limited in areas
with low P (Craft and Richardson 1997; Doren and
others 1997; Miao and Sklar 1998; Miao and others
2000). Mesocosm studies also have demonstrated
that Typha is more competitive than Cladium under
high P conditions [for example, see Newman and
others (1996)]. However, fertilizer experiments
have been unable to show that adding P alone
necessarily results in competitive exclusion of Cla-
dium (Craft and others 1995; Chiang and others
2000). These experimental findings suggest that
other factors, such as hydropattern [for example,
see Toth (1987), Urban and others (1993), and
Newman and others (1996)], fire [for example, see
Gunderson and Snyder (1994), Urban and others
(1993), and Newman and others (1998)], or cations
such as sodium (Craft and Richardson 1997) may
be important synergists in Typha expansion.
Because the autecology of both Typha and Cla-
dium has received most of the attention from re-
searchers, few studies have examined patterns of
entire macrophyte communities in response to
these human influences. The general observation
has been that high-P areas are dominated by mo-
notypic stands of dense Typha [for example, see
Jensen and others (1995) and Rutchey and Vilchek
(1999)], resulting in a relatively homogeneous
landscape pattern when compared to the reference
Cladium?slough mosaic (Obeysekera and Rutchey
1997; Wu and others 1997). However, this obser-
vation has been largely based on photointerpreta-
76 R. S. King and others
tion of satellite imagery, which is limited in spatial
resolution and is not appropriate for assessing fine-
scale pattern in species composition (Obeysekera
and Rutchey 1997; Richardson and others 1997).
Recent field studies have indicated that many other
macrophyte species coexist with Typha in high-P
areas, and that diversity is actually greater near
canals than in interior-wetland locations (Doren
and others 1997; Vaithiyanathan and Richardson
1999). However, little is known about the spatial
patterns of these communities, the environmental
factors that are responsible for generating these pat-
terns, or the influence of such spatial patterns on
autogenic environmental variation [for example,
biogeochemical properties of peat soils (Noe and
others 2001)].
Despite the large body of research that continues
to expand our understanding of the Everglades eco-
system, no field study has attempted to untangle
the independent role of multiple abiotic factors in
community-level changes in vegetation composi-
tion. Thus, we evaluated linkages among spatial
factors, environmental conditions, and vegetation
composition along a gradient of anthropogenic in-
fluence in the northern Everglades. We used a mul-
tiscale sampling approach (a) to examine vegeta-
tion?environment linkages along the full gradient
of anthropogenic influence, (b) to contrast linkages
and spatial patterns in composition as a function of
relatively discrete levels of impact to the ecosystem,
and (c) to contrast linkages and spatial patterns
across fine and coarse spatial scales. We contend
that an understanding of the characteristic scales
and patterns of vegetation?environment linkages is
critical for the Everglades and other ecosystem res-
torations if these efforts are to facilitate natural
processes effectively at multiple scales (Holling and
others 1994; Redfield 2000).
METHODS
Study Area and Sampling Design
We sampled in Water Conservation Area 2A (WCA-
2A) in the northern Everglades (Figure 1). WCA-2A
is a 43,280-ha diked wetland landscape, with wa-
ter-control structures governing the inflow and
outflow of surface water. Inflow primarily occurs
along the northern levee through three water-con-
trol structures (S10-A, C, and D) on the Hillsboro
Canal, a conduit for outflow from Lake Okeechobee
and P-enriched runoff from the EAA (Figure 1).
Inflow from the Hillsboro Canal has induced a steep
longitudinal eutrophication gradient in WCA-2A,
Figure 1. Map of south Flor-
ida showing the location of
Water Conservation Area 2A
(WCA-2A); Impacted, Transi-
tion, and Reference zones;
locations of S-10 water-con-
trol structures on the Hills-
boro Canal; centroids of sam-
pling clusters; and plot-
cluster sampling design.
EAA, Everglades Agricultural
Area.
Space, Environment, and Wetland Vegetation 77
due primarily to large inputs of P (SFWMD 1992).
Aerial photographs (US Geological Survey unpub-
lished data) and descriptive studies [for example,
see Davis (1943)] prior to impoundment have dem-
onstrated that vegetation pattern and composition
across this region of the WCA-2A landscape was
once very similar. Today, however, three relatively
distinct vegetation zones exist along this gradient
(Figure 1): (a) an impacted zone (hereafter, Im-
pacted) approximately 0?3 km downstream of the
canal inflow structures, where surface water and
soil are heavily enriched with P, and vegetation
characteristic of the pristine Everglades reportedly
has been all but replaced by dense stands of Typha
and other invasive species; (b) a transition zone
(hereafter, Transition) that ranges from 3?7 km
from the canal, where P concentrations diminish
but remain elevated, and vegetation is a mix of
Typha, other invasive species, Cladium, and infre-
quent open-water slough habitats; and (c) a rela-
tively unaffected reference zone (hereafter, Refer-
ence) beyond 7 km from the canal that exhibits
water and soil chemistry representative of the his-
torical Everglades, with vegetation structured as a
mosaic of Cladium stands interlaced with open-wa-
ter sloughs. Because boundaries among these three
zones have been well established in recent studies
[for example, see Doren and others (1997), Obey-
sekera and Rutchey (1997), Van der Valk and Ros-
burg (1997), Wu and others (1997), Richardson
and others (1999), Vaithiyanathan and Richardson
(1999), SFWMD (2000), and King and Richardson
(2002)], we felt these zonal categorizations were
valid and would provide us with an opportunity to
make contrasts among three relatively discrete, yet
contiguous, levels of ecosystem alteration.
Previous to this study, three 10-km-long
transects were established, each aligned with one of
the S-10 inflow structures and parallel to the veg-
etation gradient (Figure 1) (Richardson and others
1999). Six long-term sampling stations were
marked along each transect, starting 1.0?1.5 km
from the canal and spaced at 1.5-km intervals. We
selected 14 of these stations as centroids for our
sampling, with all six selected from the central
transect (C transect), and random draw of four of
the six from each of the A and D transects (Figure
1). In aggregate, five centroids were considered Im-
pacted, five Transition, and four Reference based on
the a priori zone classifications (Figure 1).
We used a stratified-cluster sampling design as
described by Urban (2000). In this study area, we
hypothesized that variation in vegetation pattern
was likely to occur on fine, local scales (for exam-
ple, tens of meters) as well as coarse or landscape
scales (thousands of meters). Conventional sam-
pling designs such as random or stratified random
would not have captured local patterns in variation
if conducted with coarse or landscape-scale separa-
tion distances among plots; similarly, at fine scales,
these designs would require thousands of plots and
thus would not be practical at a large spatial extent.
Thus, this cluster approach (see Figure 1) provided
a highly efficient method for considering multiple
scales in spatial, environmental, and compositional
variables (Fortin and others 1989; Urban 2000).
A single plot at each of the 14 stations served as a
plot-cluster centroid. Eight additional plots were
marked in a constellation, with four plots placed at
50-m distances from centroids and four others at
200-m distances, each in the four cardinal direc-
tions (Figure 1). Separation distances among plots
within clusters ranged from 50 to 400 m, with a
total of nine plots per cluster and 126 plots across
the landscape. Distances among plot clusters ranged
from approximately 1000 to 10,000 m. Random
allocation of plots within clusters was not practical,
as this would have caused excessive damage to
vegetation in the areas adjacent to plots because of
the airboat used for transportation.
We chose a plot size of 10 m2, large enough to
integrate across microhabitats and thus reduce
noise, but not so large that they averaged across
distinct patches of vegetation (Fortin and others
1989). Plots were semicircular to facilitate sampling
from the perimeter and to minimize disturbance.
Environmental and Vegetation Variables
One assumption inherent to our study was that the
environmental variables that we measured were
indeed important to variation in vegetation compo-
sition and expressed in an ecologically meaningful
manner. We selected a suite of variables that we
hypothesized were causing changes to vegetation
patterns or were linked to patterns autogenically.
We considered four types of variables: (a) spatial,
(b) soil chemical, (c) hydrological, and (d) physical
disturbance (that is, fire).
Spatial Variables. Relative proximity to inflow
structures on the Hillsboro Canal was expected to
be the underlying, indirect determinant of changes
in vegetation in patterns in the study area. Distance
(in meters) from these structures (hereafter, Canal)
was calculated using Universal Trans-Mercator
(UTM) coordinates of inflow structures and individ-
ual plots. Coordinates were estimated using a global
positioning system (GPS) at inflow structures and
plot-cluster centroids. For greater precision, coordi-
nates of the eight additional plots per cluster were
determined using direct measurements (in meters)
78 R. S. King and others
from a meter tape and bearings from plot-cluster
centroids.
These coordinates were also used to calculate
separation distances (in meters) among plots. Sep-
aration distances (hereafter, Space) were used for
estimating the effect of spatial proximity among
plots on vegetation?environment correlations. This
was important because vegetation samples that are
close together tend to be more similar than ones far
apart, regardless of environmental determinants. In
other words, we used Space to remove the con-
founding effect of spatial autocorrelation on the
statistical significance of vegetation?environment
linkages, as well as to detect spatial structure in
vegetation that could not be accounted for by en-
vironmental variables (Legendre and Fortin 1989;
Legendre 1993). Greater explanation on the use of
the variable Space is presented in Data Analysis:
Vegetation?Environment Linkages.
Soil Chemical Variables. We limited our chemical
variables to those obtained from soils because soils
are an integrator of long-term water-chemistry
conditions and the variables of interest are highly
correlated to both water chemistry and loading
rates across the study area [for example, see De-
Busk and others (1994), Craft and Richardson
(1997), Richardson and others (1999), and King
and Richardson (2002)]. Moreover, soil chemistry
was shown by Pan and others (2000) to explain
more variation in community attributes than did
water chemistry in the Everglades, particularly for
P, which was a nutrient of great interest for the
present study. Preliminary analysis and results re-
ported by King (2001), who used the same plots
examined in this study, showed that soil chemistry
was more strongly related to vegetation composi-
tion than was water chemistry data collected from
1995 to 1998 at these same locations. Specifically,
soil total P (TP) explained all of the variation ex-
plained by either surface-water TP or soluble reac-
tive P (SRP), while soil TP still remained signifi-
cantly related to vegetation after removing the
combined variance explained by surface-water TP
and SRP (P 
 0.0001, partial Mantel test?see Data
Analysis: Vegetation?Environment Linkages). Finally,
variation in many soil-chemical variables [for ex-
ample, soil total carbon (C)] is a product of varia-
tion in vegetation composition; thus, soil chemistry
enabled us to more directly evaluate the linkage
between composition and potential autogenic
processes.
Soil samples were collected for chemical analyses
from every plot during 20?29 October 1998. Each
sample was analyzed for total C, P, nitrogen (N),
calcium (Ca), potassium (K), magnesium (Mg), and
sodium (Na). Field and laboratory methods are de-
scribed by King (2001). Soil chemical concentra-
tions were expressed per unit dry mass rather than
volume because bulk density has been shown to be
similar along the vegetation gradient (Reddy and
others 1991). All soil variables were retained for
analysis because no pair was deemed to be collinear
(Kleinbaum and others 1988).
Hydrological Variables. The hydropattern in
WCA-2A is regulated, and no longer occurs as nat-
ural response to precipitation and surface-water
flow as it did before water control structures were
built. Within WCA-2A, the hydropattern varies
considerably (Romanowicz and Richardson 1997,
2004). Variations in hydropattern result from the
relative rate of water flow through the water con-
trol structures, the topography of the peat surface,
and equipotential surface of the water. We esti-
mated hydropattern at each plot by linking field
measurements to a spatially explicit hydrological
model developed by Romanowicz and Richardson
(1997, 2004) for the same study area.
The model was developed using historical sur-
face-water stage and discharge data (SFWMD 1992)
and a network of 11 continuous-recording stage
recorders equipped with 10-psi pressure transduc-
ers (error, 
 0.1% full-scale reading). These sta-
tions were monitored for between 2 and 3 years to
relate the equipotential surface of surface water to
concurrent surface-water stage data and discharges
through the canal inflow structures into WCA-2A
(r2  0.99). This model was then used to predict
water elevation at numerous locations in WCA-2A,
including the 14 plot-cluster stations, using histor-
ical (1981?98) stage and discharge data (SFWMD
1995). Predicted water depths at plot-cluster cen-
troids were validated using multiple field measure-
ments in 1998?99 (r2  0.98).
To estimate hydropattern at each of our plots,
water depths (in centimeters) were measured at
three random locations within each plot and aver-
aged during 20?29 October 1998. Field water-
depth measurements taken from each plot were
then linked to the water-depth estimates produced
by the model at each of the respective plot-cluster
centroid benchmarks. In other words, we used the
validated water-depth estimates generated by the
model at each of the 14 plot-cluster centroids as
benchmarks, and the relative difference in instan-
taneous field measurements of water depth be-
tween the centroids and surrounding plots, to cor-
rect for differences in local elevation among plots in
each respective cluster. The model was then used to
generate temporal estimates of water depth for ev-
ery plot for the period of 1981?98.
Space, Environment, and Wetland Vegetation 79
Using the temporal water-depth estimates, we
considered metrics of (a) mean water depth, (b)
frequency of exceeding or falling below certain
depths (for example, percent of days at  150 cm),
and (c) stability of water depth (for example, met-
rics of variance or range in water depth). We eval-
uated all metrics by using both short-term (1 year)
and long-term (1981?98) data. Short-term hydrol-
ogy was expected to have a greater influence on
subtle compositional patterns (Gunderson 1989),
whereas long-term hydrology was expected to have
a greater influence on coarse-scale species distribu-
tions (Urban and others 1993; Busch and others
1998; Shay and others 1999). Preliminary results
indicated that short-term mean water depth (here-
after, Depth), long-term frequency of depth less
than 10 cm (index of severe dryness calculated as
the percent of days during 1981?98 in which water
depth was less than 10 cm; hereafter Freq
10cm),
and long-term interquartile range of depth [a ro-
bust index of stability of water depth calculated as
the range between the 25th and 75th percentiles of
the distribution of daily water depths during 1981?
98; hereafter IQR(Depth)] were most strongly re-
lated to vegetation. These three variables were not
collinear, and each accounted for unique variation
in vegetation composition along the anthropogenic
influence gradient (P  0.05, partial Mantel test?
see Data Analysis). Thus, all three were retained for
further analysis.
Fire Variables. Fire disturbance was expected to
play a role in vegetation pattern because Cladium is
considered fire tolerant (Steward and Ornes
1975b), and intense fires are believed to be partially
responsible for maintaining the sawgrass?slough
mosaic (Craighead 1971; Gunderson and Snyder
1994). A composite map of recent, large fires con-
structed using aerial photographs indicated that
more than 50% of WCA-2A had burned at some
point during 1981?98 (Florida Game and Freshwa-
ter Fish Commission unpublished data). We calcu-
lated an index of fire frequency for the period of
1981?98 by using these data. Coordinates of fire
boundaries were related to plot coordinates and
used to determine the presence of fires at each plot,
following methods used by Newman and others
(1998). Frequencies of large fires ranged from zero
to two for individual plots during this period. Be-
cause the time since the last fire was potentially as
important as the frequency of fires (Allen and
Wyleto 1983; Gunderson and Snyder 1994), fire
frequency was weighted as 1/log10(t  1), where t
 time since fire (years), and summed for all fires
during the period 1981?98. Preliminary analysis
indicated that vegetation composition had a higher
correlation to this fire index (hereafter, Fire) than
did simple fire frequency; thus, the index was re-
tained for analysis.
Vegetation Composition. We estimated vegetation
species composition and cover within each plot by
using Braun?Blanquet cover classes (Phillips 1959).
This cover-estimation technique was used to mini-
mize disturbance to vegetation within the plots and
because it was ideal for our distance-based statistical
analyses [for example, see Leduc and others
(1992)]. Classes ranged from 0 to 6, and approxi-
mated a log-linear relationship with increasing per-
cent cover. Two observers estimated cover by wad-
ing around the perimeter of the plot. All
macrophytes were identified to the lowest practical
level, usually species.
We also estimated cover classes of calcareous and
noncalcareous periphyton mats, as calcareous mats
have been indicated to be sensitive to P (Flora and
others 1988; Vymazal and others 1994; McCormick
and others 1998) and are an important feature of
pristine slough habitats in the Everglades (Turner
and others 1999a; Noe and others 2001). Periphy-
ton mats either disappear or shift to noncalcareous
forms where P is elevated (McCormick and others
1998). Coverage of open water (no vegetation in
water column) was included as a cover type be-
cause reportedly it has been reduced in impacted
areas due to invasive vegetation [for example, see
Obeysekera and Rutchey (1997)]. All vegetation
sampling was conducted during 20?29 October
1998 (end of wet season), a period when periphy-
ton mats were at their peak biomass (McCormick
and others 1998) and water levels were sufficient to
allow airboat travel to most areas of WCA-2A.
Data Analysis
Environmental Variation Among Zones. We used
mixed-model nested analysis of variance (ANOVA)
to compare means of soil chemical (C, Ca, K, Mg, N,
Na, and P), hydrological (Depth, Freq
10cm, and
IQR(Depth)), and fire (Fire) variables among the
three vegetation impact zones to evaluate whether
they differed environmentally at this relatively
broad scale. Plots were replicates nested within
clusters (Cluster, random effect), whereas clusters
were replicates nested within zones (Zone, fixed
effect). Means were contrasted using Tukey?s HSD
test for Zone effects deemed significant by ANOVAs;
means for significant Cluster effects were not com-
pared because this effect was considered random
(Bennington and Thayne 1994). ANOVAs were
conducted using the Variance Components module
of Statistica 5.5 (Statsoft, Tulsa, OK, USA).
80 R. S. King and others
Macrophyte Species Distributions Among Zones. To
assess affinities of macrophyte species to impact
zones and test the hypothesis that the transition
zone was truly an area of ecological transition, we
performed Indicator Species Analysis (INSPAN)
(Dufre?ne and Legendre 1997) on vegetation cover
data. INSPAN is a nonparametric technique used to
identify species with a high fidelity for a particular
group or class, as defined by the user. Indicator
Values (IVs) were the percent of perfect indication
among the three impact zones and were calculated
using both the relative frequency and cover of each
species [see Dufre?ne and Legendre (1997) for ana-
lytical details]. Here, we hypothesized that the
Transition zone would have fewer indicator species
than either the Impacted or Reference zones be-
cause it should host patchy distributions of species
more frequent to the other zones. Significance
(Bonferroni-corrected P 
 0.05) of IVs was esti-
mated using 10,000 random permutations of the
vegetation data (Manly 1997). INSPAN was per-
formed using PC-ORD 4.08 (MjM Software,
Gleneden Beach, OR, USA).
Vegetation?Environment Linkages. We used two
complementary distance-based procedures to esti-
mate relationships between vegetation composition
and environmental variables. First, we ordinated
plots and species based on species composition by
using nonmetric multidimensional scaling (nMDS)
(Minchin 1987). Ordination provided a visual as-
sessment of gradients in species composition among
and within impact zones and was conducted to aid
in the interpretation of partial Mantel tests, our
second, more tactical approach to estimating spatial
and environmental linkages to composition. We
used Bray?Curtis dissimilarity (BCD) as the distance
metric, a coefficient shown to be one of the most
robust and ecologically interpretable (Faith and
others 1987). Once plots were ordinated, species
centroids were mapped into ordination space by
using weighted averaging (Legendre and Legendre
1998). A two-dimensional solution was the most
appropriate for all ordinations, as stress values (an
indicator of agreement between BCDs and the con-
figuration of the ordination) were relatively low
and exhibited small decreases when additional axes
were included in the ordination.
We performed ordinations for all plots (n  126)
as well as three subsets of plots corresponding to
each zone (n  45 for Impacted and Transition; and
n  36 for Reference). We used subsets of data to
examine potential interactions in vegetation?envi-
ronment relationships among impact zones that
would potentially be obscured when using the full
data set. An additional coarse-scale ordination was
produced by using average composition of each of
the 14 plot clusters in contrast to vegetation pat-
terns produced at the fine scale (Allen and Wyleto
1983; Turner and others 1999b).
To relate environmental variables to gradients in
composition in nMDS ordinations, we used rota-
tional vector fitting (Faith and Norris 1989). Vector
fitting was used to find the direction of the maxi-
mum correlation of an environmental variable in
an ordination of the vegetation data. Vector fitting
was performed on all ordinations. Environmental
values from each plot were used in fine-scale vector
fitting, whereas average values from within each
plot cluster were used in the coarse-scale analysis.
Significance (P 
 0.05) of environmental vectors
was estimated using 10,000 random permutations
of the data. Ordination and vector fitting were per-
formed using DECODA 2.05 (University of Mel-
bourne, Parkville, Victoria, Australia).
Partial Mantel tests were used to measure the
partial correlation (Mantel r) between spatial, en-
vironmental, and vegetation distance matrices
(Mantel 1967; Smouse and others 1986). Funda-
mentally, the analysis examines whether plots that
are similar environmentally also are similar compo-
sitionally (Urban and others 2002). Mantel r coef-
ficients are typically relatively small in magnitude
(usually  0.5 for all but the strongest relation-
ships) because the analysis considers the full rather
than reduced dimensionality in multivariate data
[for example, see Legendre and Fortin (1989), Le-
duc and others (1992), and Foster and others
(1999)]. Because it uses distance matrices, this ap-
proach enables the user to extract variation caused
by spatial autocorrelation (Legendre 1993) as well
as other environmental variables to yield pure-par-
tial correlations?relationships that represent vari-
ation that cannot be explained by all other variables
included in the analysis.
We used spatial (Space and Canal) and environ-
mental (C, N, P, Ca, K, Mg, Na, Depth, IQR(Depth),
Freq
10cm, and Fire) variables as individual pre-
dictors in the Mantel analysis. Individual variables
were converted to distance matrices using euclid-
ean distance (Legendre and Legendre 1998). Vege-
tation species composition was expressed using
BCD, as in the nMDS analysis. We first examined
simple relationships between Space and individual
environmental variables, Canal and environmental
variables, and each spatial or environmental vari-
able and vegetation. We then examined the effect
of spatial autocorrelation on vegetation?environ-
ment linkages by factoring out the effect of Space.
Finally, we estimated pure-partial vegetation?envi-
ronment linkages. Here, the strength of a relation-
Space, Environment, and Wetland Vegetation 81
ship between a predictor variable and vegetation
was assessed after variation explained by all other
variables had been removed (except for Canal, be-
cause it was assumed to be causing variation in
several environmental variables). We also exam-
ined the pure-partial Space?vegetation relation-
ship, which would indicate residual spatial pattern
in vegetation that can be explained only by spatial
processes such as dispersal or other unmeasured
factors with spatial structure.
Partial Mantel tests were conducted by using data
from all plots, three subsets of data corresponding
to the impact zones, and coarse-scale averages of
composition and the environment within the 14
plot clusters. Significance (Bonferroni-corrected P

 0.05) of Mantel r coefficients was assessed using
10,000 permutations. As a visual framework for
these results, linkages among spatial, environmen-
tal, and vegetation variables were synthesized using
path diagrams, a schematic that depicts significant
paths of relationships among variables (Leduc and
others 1992; Zmyslony and Gagnon 2000).
Bootstrapped confidence limits (95% CLs) were
estimated for partial Mantel r coefficients to allow
for comparison of the strength of linkages among
zones. Bootstrapping was conducted by resampling
distance matrices at a level of 90%, with 1000
resamples (Manly 1997; King and Richardson 2002).
Spatial Pattern of Vegetation Among Zones. To as-
sess spatial patterns in vegetation at multiple scales,
we used an extension of the Mantel test called a
Mantel correlogram (Oden and Sokal 1986). Cor-
relograms produce an index of spatial autocorrela-
tion for classes of separation distances?samples
that are compositionally more similar than average-
yield positive autocorrelation, while those that are
less similar are negatively autocorrelated (Legendre
and Fortin 1989).
We contrasted pattern in vegetation among the
three zones at both fine and coarse scales. Fine-
scale pattern was examined within clusters by using
separation distances of 50?400 m, with 50 m inter-
vals (for example, 251- to 300-m separation dis-
tances  300-m distance class). Coarse-scale pat-
tern was assessed by comparing autocorrelation
among clusters (within-cluster versus among-clus-
ter variation). Because we were interested in how
the average pattern within each cluster differed
with respect to the average pattern among remain-
ing clusters, these correlograms were limited to
only one 400-m interval. Significant differences in
Mantel r values among distance classes were con-
trasted using 95% CLs estimated by bootstrapping.
Distance classes were considered significantly dif-
ferent if CLs did not overlap (Manly 1997). Mantel
tests and bootstrapping were performed using S-
Plus 5.0 for Unix.
RESULTS
Environmental Variation Among Impact
Zones
Three of the 11 environmental variables analyzed
by using ANOVA differed among landscape impact
zones (Table 1). All three zones differed for P and
IQR(Depth), while Na differed between Impacted
and Reference zones. Values of Ca, Depth, and
Freq
10cm tended to be greatest in the Reference
zone but were not statistically different than other
zones because of high heterogeneity among plots
within clusters in that zone (reflecting the hetero-
geneity of the Cladium?slough mosaic). All 11 vari-
ables yielded a significant Cluster effect, indicating
coarse-scale spatial differences in the environment
among clusters within one, two, or all of the three
zones.
Macrophyte Species Distributions
Forty macrophyte species and three other cover
types (open water, noncalcareous periphyton, and
calcareous periphyton) were identified (Table 2).
Eight species were significant indicators of the Im-
pacted zone, whereas six were indicators of the
Reference zone. Despite having the most species, no
species or cover types were indicators of the Tran-
sition zone, reaffirming it to be an ecological area of
transition between the Impacted and Reference
zones because it hosted patchy distributions of spe-
cies common to both other zones (Table 2). Typha
domingensis and invasive vines, Mikania scandens and
Sarcostemma clausum, were the best indicators of the
Impacted zone, although small floating species,
Lemna spp. and Salvinia minima, and the understory
herb Rumex verticillatus also had strong affinities for
sparse-canopied areas there. The invasive willow
Salix caroliniana was the only woody shrub with a
significant indicator value; it was common in the
Impacted zone. Cladium jamaicense was a significant
indicator of the Reference zone despite being fairly
common in the Transition zone as well. Four
slough-community species also showed high fidel-
ity to the Reference zone: Nymphaea odorata (water
lily), Eleocharis elongata (spikerush), and Utricularia
purpurea and U. fibrosa (bladderworts). Calcareous
periphyton mat also was a significant indicator and
was almost exclusively found in the Reference
zone.
82 R. S. King and others
Vegetation?Environment Linkages
Ordination. Ordination of fine-scale composition
resulted in two gradients: a coarse-scale gradient
significantly associated with Canal, P, IQR(Depth),
and Depth; and a fine-scale gradient related to
Freq
10cm, Fire, K, N, and Na (Figure 2). Canal
was most strongly linked to variation in composi-
tion but was closely followed by P and IQR(Depth)
(Figure 2a).
Composition differed markedly among impact
zones, as plots were sorted accordingly along nMDS
axis 1 (Figure 2b). Reference-zone vegetation, al-
though tightly banded at one end of nMDS axis 1,
showed much fine-scale variation as evidenced by
great dispersion ( diversity) along nMDS axis 2.
However, the Transition zone showed the greatest
overall variation in composition, with an extensive
distribution of plots along nMDS axes 1 and 2, and
several plots intermingling among Impacted and
Reference zones.
Species centroids were unevenly dispersed in or-
dination space. Remarkably few species were pro-
jected in the region corresponding to the Transition
zone; most centroids were clearly associated with
the Impacted or Reference zones (Figure 2a and b).
In the Reference zone, slough-species centroids
closely corresponded to vectors of Depth, N, and K,
whereas Cladium (CLADJAMA) was at the opposite
end of this within-zone gradient and corresponded
to Freq
10cm and Fire. Centroids were most
tightly aggregated in the Impacted zone, suggesting
less distinction of discrete communities.
Coarse-scale ordination of vegetation composi-
tion revealed that Canal, P, IQR(Depth), and Depth
were the primary correlates of coarse-scale pattern
(Figure 2c). Virtually all variation in composition
occurred along nMDS Axis 1, and this gradient
mirrored that of Axis 1 in the fine-scale ordination
(Figure 2a and b). Ordination of average composi-
tion in clusters essentially eliminated fine-scale
variation recovered by nMDS axis 2, subsequently
eliminating fine-scale vegetation?environment re-
lationships. Clusters were also completely separated
into distinct strata corresponding to the three zones
of impact, which provided further support for the
dual gradient/zone concept in the study area (Fig-
ure 2c).
Ordinating subsets of data for each zone largely
corroborated trends found in the full data set but
revealed a few other potential relationships (Figure
Table 1. Results from Mixed-model Nested Analysis of Variance (ANOVA) on Selected Environmental
Characteristics of Impacted, Transition, and Reference Zones
Variablea Code Units F(2,11)
b P
Landscape Zone
Impacted
(n  45)c
Transition
(n  45)
Reference
(n  36)
Distance from canal Canal m ?d ? 2495 (869) 5541 (914) 9050 (924)
Total carbon (soil) C g/kg 0.15 NS 435.0 (20.3) 435.0 (27.1) 428.0 (47.7)
Total calcium (soil) Ca g/kg 0.46 NS 37.1 (16.5) 42.8 (20.8) 47.0 (34.5)
Total potassium (soil) K mg/kg 0.06 NS 581.7 (29.5) 557.4 (58.4) 527.0 (28.3)
Total magnesium (soil) Mg mg/kg 0.11 NS 3707 (111) 3859 (140) 3625 (146)
Total sodium (soil) Na mg/kg 4.31 0.032 3058 (160)A 2900 (173)AB 2162 (187)B
Total nitrogen (soil) N g/kg 0.02 NS 29.3 (2.2) 29.0 (3.7) 29.2 (4.4)
Total phosphorus (soil) P mg/kg 102.30 
 0.001 1434 (174)A 1198 (184)B 578 (152)C
Water depth (1 year) Depth cm 3.74 NS 35.7 (8.3) 41.8 (9.6) 46.4 (10.4)
Interquartile range,
water depthe IQR(Depth) cm 20.90 
 0.001 28.2 (0.1)A 29.7 (0.2)B 33.6 (0.1)C
Frequency, water depth
less than 10 cme Freq
10cm % 2.79 NS 3.1 (0.4) 3.1 (0.3) 6.0 (0.8)
Fire indexe Fire Sumf 0.46 NS 0.2 (0.4) 0.4 (0.5) 0.3 (0.5)
aCanal, distance from Hillsboro Canal; Depth, short-term mean water depth; Fire, fire index; Freq
10cm, water depth less than 10 cm, and IQR(Depth), interquartile range
of depth. C, carbon; Ca, calcium; K, potassium; Mg, magnesium; N, nitrogen; Na, sodium; and P, phosphorus.
bF ratios and associated P values correspond to the Zone main effect (fixed effect). Cluster was a random effect nested within Zone and was used as the error term. Cluster
(random effect, F11,112) was significant (P 
 0.05) for all variables.
cMean (
 1 SD) values are based on measurements collected at individual plots within landscape zones. LSD tests were used to compare means among levels of the Zone effect
when deemed significant from ANOVAs. Means with the same superscript letters do not differ (P  0.05).
dANOVA not conducted on Canal since it was not independent of the landscape zones.
e1981?98.
fSum of total number of fires/plot during 1981?98, weighted as 1/log10(t  1) for each fire, where t  time (years) since fire.
Space, Environment, and Wetland Vegetation 83
3). In the Impacted zone, cations (Ca and Mg) and
Fire were associated with the few plots in which
Cladium occurred (Figure 3a). However, of these
three environmental variables, only Mg was signif-
icant after Bonferroni correction. IQR(Depth), N,
and C were inversely related to this Cladium gradi-
ent, with the floating plants Salvinia minima
(SALVMINI) and Lemna spp. (LEMNA) along with
Table 2. List of Macrophyte Species and Cover Types, Including Corresponding Codes (See Figures 2 and
3) and Indicator Values (IVs) for Each Landscape Zone
Species/Cover Type Code
Indicator Value (IV)
PImpacted Transition Reference
Acrostichum danaeifolium Langsd. and Fitch. ACRODANA 0 6 0 NSa
Alternanthera philoxeroides (Mart.) Griseb. ALTEPHIL 7 0 0 NS
Amaranthus australis (Gray) Sauer. AMARAUST 1 1 0 NS
Aster sp. ASTER 2 0 0 NS
Bacopa sp. BACOPA 2 0 0 NS
Calcareous periphyton mat CALMAT 0 2 37 
 0.0001
Cephalanthus occidentalis L. CEPHOCCI 5 3 1 NS
Ceratophyllum demersum L. CERADEME 0 4 0 NS
Chara sp. CHARA 0 5 19 0.0055*
Cladium jamaicense Crantz. CLADJAMA 6 28 46 
 0.0001
Crinum americanum L. CRINAMER 0 1 8 0.0317*
Cyperus odoratus L. CYPEODOR 2 0 0 NS
Eichornia crassipes (Mart.) Solms. EICHCRAS 9 0 0 0.0353*
Eleocharis cellulosa Torr. ELEOCELL 0 1 13 0.0107*
Eleocharis elongata Chapm. ELEOELON 0 0 29 
 0.0001
Hydrocotyle umbellata Lamark. HYDRUMBE 11 1 0 0.0190*
Ipomoea sagittata Poir. IPOMSAGI 1 1 2 NS
Lemna sp. LEMNA 30 4 0 
 0.0001
Limnobium spongia (Bosc.) Steud. LIMNSPON 0 4 0 NS
Ludwigia leptocarpa (Nutt.) LUDWLEPT 4 0 0 NS
Ludwigia repens Forst. LUDWREPE 3 5 0 NS
Noncalcareous periphyton mat MAT 3 20 0 0.0024*
Mikania scandens (L.) Willd. MIKASCAN 61 8 0 
 0.0001
Nymphaea odorata Aiton. NYMPODOR 0 9 39 
 0.0001
Open water?no cover OPEN 19 34 34 NS
Panicum repens L. PANIREPE 0 0 3 NS
Peltandra virginica (L.) Schott and Endl. PELTVIRG 0 0 3 NS
Pistia stratiotes L. PISTSTRA 7 2 0 NS
Poaceae sp. POACEAE 13 0 0 0.0047*
Polygonum densiflorum Meisn. POLYDENS 6 1 0 NS
Polygonum punctatum Ell. POLYPUNC 25 16 0 0.0116*
Pontederia cordata L. PONTCORD 8 6 0 NS
Rumex cf. verticillatus L. RUMEVERT 36 0 0 
 0.0001
Sagittaria lancifolia L. SAGILANC 31 14 1 0.0011
Salix caroliniana Michx. SALICARO 20 1 1 
 0.0001
Salvinia minima Baker. SALVMINI 25 0 0 
 0.0001
Sarcostemma clausum (Jacq.) Schult. SARCCLAU 39 1 0 
 0.0001
Scirpus validus Vahl. SCIRVALI 3 1 0 NS
Typha domingensis Pers. TYPHDOMI 58 31 2 
 0.0001
Utricularia fibrosa Walt. UTRIFIBR 0 10 37 
 0.0001
Utricularia foliosa L. UTRIFOLI 0 6 14 0.0312*
Utricularia purpurea Walt. UTRIPURP 0 0 48 
 0.0001
Wolfiella sp. WOLFIELL 4 0 0 NS
aIndicates IVs that are not significant. IVs are percent of perfect indication, with significant (Bonferroni-corrected P 
 0.05) scores shown in bold.
*Not significant after Bonferroni correction.
84 R. S. King and others
the willow Salix caroliniana (SALICARO) reaching
greatest abundance here. Typha (TYPHDOMI) was
widespread throughout; thus, its centroid was lo-
cated at the midpoint of the ordination. Neither P
nor Canal was linked to composition in the Im-
pacted zone.
In the Transition zone, N, K, Na, Depth, and
Canal were the variables most closely linked to
composition (Figure 3b). Noncalcareous periphyton
mat (MAT) and Nymphaea odorata (NYMPODOR)
were the two most notable centroids associated
with these variables; their cover was chiefly located
in slough habitats or other deep, open-canopied
areas. Cladium abundance was inversely related to
these variables.
Ordination of Reference-zone vegetation repro-
duced the Cladium?slough gradient revealed in the
full landscape ordination (Figures 2 and 3c). Slough
species were associated with deeper water, but also
N. Cladium stands were more likely to experience
periods of dryness and also had greater soil C and P
relative to sloughs.
Partial Mantel Tests. Partial Mantel tests on fine-
scale data from the full vegetation gradient indi-
cated that most environmental variables were spa-
tially autocorrelated (column Space in Table 3 and
Figure 4). However, only four variables, Na, P,
IQR(Depth), and Depth, were directly related to
Canal (column Canal in Table 3), corroborating
results from ANOVA among the landscape zones
(Table 1). Considering simple relationships, vegeta-
tion composition was significantly related to all but
two variables (Na and Fire; column Veg in Table 3).
However, after extracting spatial autocorrelation
(Space), Ca and Mg were no longer significant (col-
umn Veg/Space in Table 3). The variables that cor-
responded to nMDS axis 1 (Figure 2)?Canal, P,
and IQR(Depth)?were most closely tied to compo-
sition regardless of spatial dependencies. N, K, and
C exhibited the strongest link to composition of the
remaining fine-scale variables, although Depth and
Freq
10cm remained significantly correlated to
vegetation as well.
Pure-partial tests revealed that C, Depth, and
IQR(Depth) could not account for unique variation
in the vegetation composition (column Veg/* in
Table 3 and Figure 4). However, P remained highly
significant. The relationship between Freq
10cm
and vegetation actually improved slightly after vari-
ation from all other variables had been removed. N
and K also remained highly significant as pure par-
tials. Confidence limits (95% CLs) generated by
bootstrapping indicated that Canal was the stron-
gest factor (partial Mantel r  0.26, lower 2.5% 
0.24, and upper 2.5%  0.28)?a relationship that
Figure 2. Nonmetric multidimensional scaling (nMDS)
ordination of individual (a) plots and (b) species/cover-
type centroids using fine-scale vegetation species compo-
sition, and (c) clusters using coarse-scale species compo-
sition (calculated as averages of the 14 plot clusters).
Symbols indicate membership among the three impact
zones. Environmental vectors show the direction and
magnitude of significant correlations (r values are adja-
cent to vectors) within the ordination space. See Tables 1
and 2 for codes for environmental and species variables,
respectively. *P 
 0.05, **P 
 0.001, and ***P 
 0.0001.
Canal, distance from Hillsboro Canal; Depth, short-term
mean water depth; Fire, fire index; Freq
10cm, water
depth less than 10 cm; and IQR(Depth), interquartile
range of depth. K, potassium; N, nitrogen; Na, sodium;
and P, phosphorus.
Space, Environment, and Wetland Vegetation 85
was indicative of variation in composition caused
by the canal-and-levee system but unexplained by
any of the other measured spatial or environmental
variables. Finally, a highly significant spatial resid-
ual suggested that spatial factors contributed to
variation in vegetation pattern that could not be
explained by the environment or Canal.
Mantel results from the coarse-scale analysis gen-
erally supported those of the coarse-scale nMDS
ordination, as Canal, P, and IQR(Depth) were
highly associated with vegetation (Table 3 and Fig-
ure 4). However, IQR(Depth) was not significant as
a pure partial. Canal and P were the only variables
that significantly accounted for coarse-scale varia-
tion that could not be accounted for by other vari-
ables. Although spatially autocorrelated, coarse-
scale composition did not exhibit a significant
spatial residual, indicating that coarse spatial pat-
terns were mostly attributed to Canal and the
environment.
Mantel analysis of data from just the Impacted
zone revealed that fewer environmental variables
were dependent upon Space or Canal than in the
full data set, reflecting greater homogeneity in the
environment (Table 4 and Figure 5a). Vegetation
also had weaker linkages to the environment in this
zone. N was the only environmental variable that
was significant as a pure partial. Space explained
the most variation in composition in the Impacted
zone.
Environmental variables measured in Transition
plots showed greater spatial dependency than in the
Impacted zone (Table 4 and Figure 5b). Neverthe-
less, numerous environmental variables were
linked to vegetation even after correcting for spatial
autocorrelation and mutual correlations among
variables. N and K had the strongest relationships to
vegetation, although Depth, Freq
10cm, and C
also were significant. Space was the most influential
factor, with lower confidence limits of its pure-
partial Mantel coefficient exceeding the upper lim-
its of any other variable (partial Mantel r  0.48,
lower 2.5%  0.45, and upper 2.5%  0.50).
Figure 3. Nonmetric multidimensional scaling (nMDS)
ordinations of plots using fine-scale vegetation composi-
tion (Bray?Curtis dissimilarity) within the (a) Impacted,
(b) Transition, and (c) Reference zones. For clarity, only
common species or cover-type centroids were projected.
Environmental vectors show the direction and magni-
tude of significant correlations (r values are adjacent to
vector labels) within the ordination space. See Tables 1
and 2 for codes for environmental and species variables,
respectively. *P 
 0.05, **P 
 0.001, and ***P 
 0.0001.
Canal, distance from Hillsboro Canal; Depth, short-term
mean water depth; Fire, fire index; Freq
10cm, water
depth less than 10 cm; and IQR(Depth), interquartile
range of depth. C, carbon; Ca, calcium; K, potassium; Mg,
magnesium; N, nitrogen; Na, sodium; and P, phosphorus.
86 R. S. King and others
In the Reference zone, most environmental vari-
ables were spatially autocorrelated (Table 4 and
Figure 5c). However, Depth, Freq
10cm, C, Ca, N,
and P were significantly linked to vegetation re-
gardless of spatial dependencies. Freq
10cm, C, K,
N, and P all remained linked to vegetation after
variance explained by other variables had been re-
moved. Vegetation was not spatially autocorrelated,
nor was there a significant spatial residual, which
indicated that the spatial distribution of macrophyte
species was similar across the Reference landscape.
Based on 95% CLs of partial Mantel coefficients,
Space was significantly more important to vegeta-
tion patterns in Transition and Impacted zones than
in the Reference zone.
Contrasting Vegetation Pattern Among
Impact Zones
Mantel correlograms showed that fine-scale pattern
(within clusters) differed significantly among zones
Table 3. Results from Simple (r) and Partial () Mantel Tests Among Spatial, Environmental, and
Vegetation (Veg) Distance Matrices at Fine and Coarse Scales Along the Full Vegetation Gradient
Variable (X)a
Space (S)b Canal (C)c Veg (Y)d Veg/Spacee Veg/* (Pure Partial)f
rSX P rCX P rXY P XY S P XY * P
Fine scale
(n  126 plots)
Space ? ? 0.624 
 0.0001 0.392 
 0.0001 ? ? 0.178 
 0.0001
Canalh 0.624 
 0.0001 ? ? 0.502 
 0.0001 0.349 
 0.0001 0.262 
 0.0001
C NSa NS 0.104 
 0.0001 0.108 
 0.0001 NS
Ca 0.115 
 0.0001 NS 0.071 0.0004 NS NS
K NS NS 0.088 
 0.0001 0.127 
 0.0001 0.187 
 0.0001
Mg 0.315 
 0.0001 NS 0.050 0.0021 NS NS
Na 0.060 0.0030 0.064 0.0002 NS NS NS
N 0.067 
 0.0001 NS 0.230 
 0.0001 0.222 
 0.0001 0.210 
 0.0001
P 0.444 
 0.0001 0.714 
 0.0001 0.397 
 0.0001 0.266 
 0.0001 0.178 
 0.0001
Depth 0.215 
 0.0001 0.160 
 0.0001 0.140 
 0.0001 0.062 
 0.0001 NS
Freq-10cm NS NS 0.080 
 0.0001 0.071 
 0.0001 0.121 
 0.0001
IQR(Depth) 0.593 
 0.0001 0.786 
 0.0001 0.376 
 0.0001 0.193 
 0.0001 NS
Fire 0.166 
 0.0001 NS NS NS NS
Coarse scale
(n  14 clusters)
Space ? ? 0.551 
 0.0001 0.527 
 0.0001 ? ? NS
Canal 0.551 
 0.0001 ? ? 0.763 
 0.0001 0.668 
 0.0001 0.406 
 0.0001
C NS NS NS NS NS
Ca 0.406 
 0.0001 NS NS NS NS
K NS NS NS NS NS
Mg 0.509 
 0.0001 NS NS NS NS
Na NS NS NS NS NS
N NS NS NS NS NS
P 0.503 
 0.0001 0.859 
 0.0001 0.658 
 0.0001 0.536 
 0.0001 0.465 
 0.0001
Depth 0.420 
 0.0001 0.297 0.0031 NS NS NS
Freq-10cm NS NS NS NS NS
IQR(Depth) 0.526 
 0.0001 0.775 
 0.0001 0.543 
 0.0001 0.370 0.0003 NS
Fire NS NS NS NS NS
aCanal, distance from Hillsboro Canal; Depth, short-term mean water depth; Fire, fire index; Freq
10cm, water depth less than 10 cm; and IQR(Depth), interquartile range
of depth; and Space, separation distance. C, carbon; Ca, calcium; K, potassium; Mg, magnesium; N, nitrogen; Na, sodium; and P, phosphorus.
bSimple Mantel test between Space and individual variables.
cSimple Mantel test between Canal and individual variables.
dSimple Mantel test between vegetation and individual variables.
ePartial Mantel test between vegetation and individual variables after accounting for variation explained by spatial autocorrelation.
fPartial Mantel test between vegetation and individual-variables after accounting for variation explained by remaining variables. Variation explained by Canal was not
removed in pure-partial tests on vegetation since it was assumed to have caused changes to several environmental variables. However, variation explained by all other variables
was removed from the estimate of the pure-partial effect of Space and Canal on vegetation.
gIndicates Mantel coefficients that are not significant (Bonferroni-corrected P  0.05).
hThe effect of distance from canal inflow structures.
Space, Environment, and Wetland Vegetation 87
of differing impact (Figure 6a). Neither Impacted
nor Transition zones displayed much local variation
in composition, which manifested itself as relatively
similar levels of autocorrelation among distance
classes. However, all distance classes for these two
zones exhibited positive autocorrelation, indicating
that within-cluster composition, regardless of sepa-
ration distance, was more similar than average
composition in other areas of their respective zones.
A subtle spike of positive autocorrelation occurred
at the 250-m distance class in both Impacted and
Transition zones.
In contrast, the Reference zone showed much
fine-scale variation in composition, with the corre-
logram effectively capturing pattern of the natural
vegetation mosaic. Vegetation was positively auto-
correlated at the smallest separation distances (50
and 100 m), whereas no autocorrelation was ob-
served at intermediate distances (Figure 6a). Plots
separated by 400 m were negatively autocorrelated.
Spatial pattern in the Reference zone differed sig-
nificantly from that of the Transition and Impacted
zones at 150-, 300-, and 400-m distance classes.
Similar to the other two zones, the Reference zone
exhibited a distinct and significant spike of positive
autocorrelation at the 250-m distance class.
Coarse vegetation pattern differed among impact
zones. At the cluster scale, the Transition zone had
the highest degree of autocorrelation but was not
significantly different from the Impacted zone (Fig-
ure 6b). Positive autocorrelation also was detected
at this coarse scale in the Reference zone but to a
significantly lesser degree than in the other two
zones.
DISCUSSION
Vegetation?Environment Linkages:
Allogenic or Autogenic?
Results from partial Mantel tests suggested that sev-
eral of the spatial and environmental variables were
independently related to vegetation composition.
Although our approach effectively removed varia-
tion accounted for by space or other variables and
thus provided strong evidence for variables directly
linked to vegetation patterns, it did not conclusively
establish the nature of these relationships. That is,
were environmental factors causing variation in
vegetation composition (allogenic), or were existing
patterns in composition causing observed environ-
mental variation (autogenic)?
The variables most closely related to the primary
vegetation gradient?P, Canal, and IQR(Depth)?
were allogenic factors. In the case of P, it is well
established that Everglades vegetation is P limited,
with most species adapted to highly oligotrophic
conditions [reviewed by Noe and others (2001)].
Several species, particularly those comprising the
slough community, have been shown to be sensi-
tive to elevated P in both observational (Vaithiy-
anathan and Richardson 1999) and experimental
(Walker and others 1989; Craft and others 1995;
Chiang and others 2000) settings. Conversely,
many species found to be most abundant in the
Impacted and Transition zones are opportunistic,
?weedy? species that are highly competitive in
Figure 4. Path diagram depicting the linkages among
spatial factors, environmental variables, and vegetation
composition at (a) fine and (b) coarse scales along the
vegetation gradient, as estimated using Mantel tests. Sig-
nificant pure-partial linkages between variables and veg-
etation are shown by solid arrows, while linkages that
were significant after accounting for spatial autocorrela-
tion, but not as pure partials, are shown as dotted arrows.
Arrow thickness is proportional to magnitude of relation-
ship (see Table 3). Canal, distance from Hillsboro Canal;
Depth, short-term mean water depth; Fire, fire index;
Freq
10cm, water depth less than 10 cm; IQR(Depth),
interquartile range of depth; and Space, separation dis-
tance. C, carbon; Ca, calcium; K, potassium; Mg, magne-
sium; N, nitrogen; Na, sodium; and P, phosphorus.
88 R. S. King and others
high-P environments (Davis 1994). Typha in partic-
ular is clearly most competitive under elevated nu-
trient conditions (Urban and others 1993; Newman
and others 1996; Miao and Sklar 1998; Miao and
others 2000) and has been positively associated
with P gradients in many locations in the Ever-
glades [for example, see Doren and others (1997)].
Thus, that P was significantly and independently
linked to variation in vegetation pattern is not sur-
prising and corroborates results from other studies
Table 4. Results from Simple (r) and Partial () Mantel Tests Among Fine-scale Spatial, Environmental,
and Vegetation (Veg) Distance Matrices within Each of the Three Impact Zones (See Table 3 for Details)
Variable
Space (S) Canal (C) Veg (Y) Veg/Space Veg/* (Pure Partial)
rSX P rCX P rXY P XY S P XY * P
Impacted (n  45)
Space ? ? 0.171 0.0025 0.286 
 0.0001 ? ? 0.186 
 0.0001
Canal 0.171 
 0.0001 ? ? NS NS NS
C NS NS NS NS NS
Ca 0.591 
 0.0001 NS 0.185 
 0.0001 NS NS
K NS NS NS NS NS
Mg 0.455 
 0.0001 NS NS NS NS
Na NS 0.110 0.0026 NS NS NS
N NS NS 0.085 0.0007 NS 0.116 0.0005
P NS NS NS NS NS
Depth NS NS NS NS NS
Freq-10cm NS NS NS NS NS
IQR(Depth) 0.837 
 0.0001 0.286 
 0.0001 0.222 
 0.0001 NS NS
Fire 0.470 
 0.0001 NS NS NS NS
Transition (n  45)
Space ? ? 0.216 
 0.0001 0.376 
 0.0001 ? ? 0.480 
 0.0001
Canal 0.216 
 0.0001 ? ? 0.178 
 0.0001 0.107 0.0034 NS
C 0.117 0.0004 NS 0.220 
 0.0001 0.191 0.0107 0.139 
 0.0001
Ca 0.177 
 0.0001 NS NS NS NS
K NS 0.310 
 0.0001 0.298 
 0.0001 0.321 
 0.0001 0.261 
 0.0001
Mg 0.342 
 0.0001 NS 0.178 
 0.0001 NS NS
Na 0.166 
 0.0001 0.165 
 0.0001 0.210 
 0.0001 0.162 0.0002 NS
N NS 0.198 
 0.0001 0.458 
 0.0001 0.465 
 0.0001 0.370 
 0.0001
P NS NS NS NS NS
Depth 0.114 0.0013 0.208 
 0.0001 0.137 
 0.0001 0.103 0.0030 0.117 0.0003
Freq-10cm NS NS NS NS 0.114 0.0012
IQR(Depth) 0.516 
 0.0001 0.543 
 0.0001 0.194 
 0.0001 NS NS
Fire 0.501 
 0.0001 0.274 
 0.0001 NS NS NS
Reference (n  36)
Space ? ? 0.318 
 0.0001 NS ? ? NS
Canal 0.318 
 0.0001 ? ? NS NS NS
C NS NS 0.149 0.0004 0.147 0.0006 0.120 0.0020
Ca NS NS 0.121 0.0018 0.118 0.0020 NS
K NS NS NS NS 0.212 
 0.0001
Mg 0.484 
 0.0001 NS NS NS NS
Na 0.310 
 0.0001 NS NS NS NS
N 0.208 
 0.0001 NS 0.246 
 0.0001 0.239 
 0.0001 0.166 
 0.0001
P NS 0.184 0.0006 0.231 
 0.0001 0.227 
 0.0001 0.102 0.0028
Depth 0.399 
 0.0001 NS 0.336 
 0.0001 0.342 
 0.0001 NS
Freq-10cm 0.424 
 0.0001 0.184 0.0004 0.228 
 0.0001 0.225 
 0.0001 0.103 0.0022
IQR(Depth) 0.440 
 0.0001 0.667 
 0.0001 NS NS NS
Fire 0.471 
 0.0001 0.367 
 0.0001 NS NS NS
Canal, distance from Hillsboro Canal; Depth, short-term mean water depth; Fire, fire index; Freq
10cm, water depth less than 10 cm and IQR(Depth), interquartile range
of depth; and Space, separation distance. C, carbon; Ca, calcium; K, potassium; Mg, magnesium; N, nitrogen; Na, sodium; and P, phosphorus.
Space, Environment, and Wetland Vegetation 89
that suggest P enrichment is a major source of per-
turbation to Everglades plant communities.
Stability in water depth may have acted in con-
cert with P to promote the establishment of invasive
species such as Typha. Frequency of severe dryness
(Freq
10cm) and variability of water depth
[IQR(Depth)] were significantly related to vegeta-
tion composition along this gradient, although
IQR(Depth) was not significant as a pure-partial
coefficient. Several authors have suggested that hy-
dropattern plays a role in the observed expansion of
Typha in the Everglades, as Typha is highly compet-
itive in deeper, more stable water conditions but
intolerant of drought (Toth 1988; Urban and others
1993; Newman and others 1996). Cladium, on the
Figure 5. Path diagrams depicting the linkages among
spatial factors, environmental variables, and fine-scale
vegetation composition within (a) Impacted, (b) Transi-
tion, and (c) Reference zones, as estimated using Mantel
tests. Significant pure-partial linkages between variables
and vegetation are shown by solid arrows, while linkages
that were significant after accounting for spatial autocor-
relation, but not as pure partials, are shown as dotted
arrows. Arrow thickness is proportional to magnitude of
relationship (see Table 4). Canal, distance from Hillsboro
Canal; Depth, short-term mean water depth; Fire, fire
index; Freq
10cm, water depth less than 10 cm;
IQR(Depth), interquartile range of depth; and Space, sep-
aration distance. C, carbon; Ca, calcium; K, potassium;
Mg, magnesium; N, nitrogen; Na, sodium; and P, phos-
phorus.
Figure 6. Mantel correlogram of vegetation composition
showing spatial autocorrelation among impact zones us-
ing (a) fine-scale (within-cluster) separation distances
(50 m), and (b) coarse-scale (among-cluster) separation
distance (400-m distance class only). Significantly auto-
correlated distance classes are indicated by the filled sym-
bols, and unfilled symbols indicate no autocorrelation. Error
bars indicate bootstrapped 95% confidence limits.
90 R. S. King and others
other hand, is well adapted for dynamic hydrolog-
ical conditions but exhibits a diminished capacity to
resist invasion by Typha under deep, stable water
conditions (Toth 1987; Davis 1994; Newman and
others 1996). Our data showed that mean water
depths from the previous year actually were greater
with increasing distance from the canal, whereas
variation in water depth was most stable near the
canal inflow structures where Typha was most pro-
lific. Severe dryness tended to be experienced more
frequently in the Reference zone, and plots in this
area also exhibited the greatest fluctuation in water
levels. Concomitantly, Cladium dominated these
drought-susceptible and hydrologically dynamic
plots, as evidenced in the nMDS ordinations. The
significant relationship between vegetation and hy-
drological variability supports the hypothesis that
modifications to the hydropattern related to the
canal-and-levee system have at least partially con-
tributed to observed vegetation patterns. It further
supports the hypothesis that hydrology may act
synergistically with P to promote Typha and other
invaders, since experimental fertilizer studies have
been unable to demonstrate that P enrichment
alone results in competitive exclusion of Cladium
(Craft and others 1995; Chiang and others 2000).
Perhaps the most interesting linkage revealed
along the full vegetation gradient was the residual
variation in composition that could only be ex-
plained by distance from the canal inflow structures
(Canal). Canal had a direct effect on P and
IQR(Depth), a strong indication that variation in
these factors was due to water entering the wetland
through canal inflow structures (SFWMD 1992; Ro-
manowicz and Richardson 1997, 2004). However,
even after variation from these and all other re-
maining variables was removed, Canal remained a
significant correlate of composition?its partial cor-
relation was greater than any other variable. Possi-
bly, Canal captured the synergistic effect of P and
hydropattern that was not accounted for when
those respective factors were considered individu-
ally. In addition, this observation may have resulted
from one or a combination of several other factors:
(a) environmental variables not measured but di-
rectly influenced by the canal were acting to struc-
ture vegetation (for example, micronutrients)
(G. M. Ferrell unpublished data); (b) proximity to
the canal affected seed and/or floating plant dis-
persal (Vaithiyanathan and Richardson 1999); (c) a
?front? of succeeding vegetation assemblages that
began near the canal has progressed into the inte-
rior of the study area but asynchronously with en-
vironmental changes (that is, level of P enrichment)
(Wu and others 1997; C. J. Richardson unpublished
data); and (d) other contagious spatial effects di-
rectly related to proximity to canal (for example,
herbivory or disease (Richardson and others 1999).
Indeed, other potential factors beyond these surely
exist and warrant examination in future research.
At a minimum, it seems reasonable to conclude that
influences directly attributable to canal-and-levee
systems represent a serious threat to the long-term
integrity of the Everglades and other large, wetland
ecosystems (for example, delta marsh) (Shay and
others 1999).
Whereas proximity to the canal inflow structures,
elevated P, and modified hydropatterns were allo-
genic determinants of vegetation patterns along the
anthropogenic influence gradient, the nature of the
linkages between significant fine-scale environ-
mental variables and vegetation was much less
clear. Of particular interest was the significant link-
age between N and vegetation composition. Several
studies have shown that N additions do not stimu-
late growth in Everglades plant communities
(Steward and Ornes 1975a; Walker and others
1988; Craft and others 1995; Chiang and others
2000) and that all but the most heavily P-enriched
locations are P limited (Richardson and others
1999). Nevertheless, N accounted for variation in
vegetation that no other variable could, along the
full vegetation gradient and within all three impact
zones. Within the Transition and Reference zones,
highest N (and K) concentrations were associated
with periphyton mats and slough-community spe-
cies, particularly the water lily Nymphaea odorata.
Decomposing Nymphaea tissue has been shown to
have greater N concentrations than more recalci-
trant species like Cladium (Steward and Ornes
1975b; J. Vymazal unpublished data). Periphyton
mats are largely composed of blue-green algae,
many of which are heterocystic N fixers (Swift and
Nicholas 1987; Browder and others 1994), and also
tend to have relatively high tissue N concentrations
(Vymazal and Richardson 1995). Indeed, Gleason
and Stone (1994) describe two principal sediment
types in the Everglades that closely correspond to
our observations: Everglades peat, generated by Cla-
dium, and Loxahatchee peat, a product of Nymphaea
slough communities. Thus, N may have been an
excellent indicator of fine-scale composition be-
cause the vegetation itself may have been causing
variation in N concentrations?an excellent illustra-
tion of the effect of pattern on process (Watt 1947).
Nitrogen may have played a more direct role in
generating fine-scale compositional patterns in the
Impacted zone, however. It was the only environ-
mental variable that was independently linked to
vegetation composition in this zone. High soil P
Space, Environment, and Wetland Vegetation 91
concentrations adjacent to canal inflows have been
suggested to result in some instances of N limitation
[for example, see Richardson and others (1999)].
This fact, coupled with the autogenic soil legacies
described by Gleason and Stone (1994), suggests
that remnant patches of soil containing elevated N
may influence vegetation patterns in locations
where P is no longer limiting. Salix caroliniana (wil-
low), an indicator of the Impacted zone, was par-
ticularly abundant in areas of elevated N. Under-
story species (Rumex verticillatus, Lemna spp., and
Salvinia minima) were closely associated with Salix
stands, an indication that these species may have
benefited from the reduction in canopy cover im-
posed by Typha (Grimshaw and others 1997). The
pre-impact physical template may have influenced
observed fine-scale pattern in the Impacted zone
and subsequently may have contributed to the sig-
nificant relationship between N and vegetation
(Peterson 2002). A closer examination of spatial
patterns of N and vegetation is needed to better
evaluate its role in structuring fine-scale commu-
nity composition in P-enriched areas of the Ever-
glades.
Considering cation?vegetation relationships, our
results support the conclusions of Craft and Rich-
ardson (1997), who examined the potential rela-
tionship between Typha expansion and elevated Na,
Ca, and Mg in soils near inflow structures. Using
simple correlations, they showed that Typha was
most strongly related to soil P and concluded that
cations were not important to its distribution even
though Typha also was significantly correlated Na.
In our study, none of these cations was able to
account for unexplained variation in vegetation.
Although Na and other cations cannot be com-
pletely ruled out as partial determinants, our results
suggest that they may not be significant agents of
pattern formation.
Fire has been shown to be an important distur-
bance in the maintaining of the mosaic of Ever-
glades plant communities (Craighead 1971; Gun-
derson and Snyder 1994). A wide variety of
responses of vegetation to fire have been reported,
primarily focusing on Typha and Cladium. These
responses have depended largely on intensity of
fires and the presence of nutrient enrichment [for
example, see Urban and others (1993), Richardson
and others (1997), and Newman and others
(1998)]. In our study, Fire was weakly but signifi-
cantly correlated to composition in the ordination
of the full vegetation gradient. Fire also covaried
with Freq
10cm, as plots that were dry most fre-
quently tended to burn most frequently. Addition-
ally, Cladium was the dominant species in plots that
were most susceptible to both drying and burning.
However, Fire was not linked to vegetation compo-
sition in the partial Mantel analysis. Thus, our re-
sults are equivocal. In general, these findings imply
that Fire covaries with other, stronger patterns of
environmental variation and thus is not able to
explain a unique component of variation in the
vegetation. It is important to note that our fire
index only considered large fires and only those
that had occurred since 1981. Fires also were spa-
tially contagious; hence, autocorrelation inhibited
our ability to detect fire effects. Due to the tremen-
dous number of potential interactions with other
variables and the high variability of fire frequency,
intensity, and spatial extent, it is highly unlikely
that simple patterns will emerge [for example, see
Newman and others (1998)]. For long-term man-
agement and restoration of Everglades vegetation
to succeed, a better understanding of the influence
of fire on vegetation patterns is needed.
Implications of Spatial Pattern and Scale
One of the greatest limitations of conclusions drawn
from observational studies in ecology has been the
phenomenon known as spatial autocorrelation
(Legendre 1993). Our Mantel test results revealed
some implications of spatial structuring in the en-
vironment and the observed effect it can have on
descriptive vegetation?environment relationships.
Had Space not been considered, we may have er-
roneously concluded that virtually every variable
measured was directly related to vegetation compo-
sition. On the contrary, many of these relationships
were artifacts of spatial autocorrelation in both the
environment and vegetation. Simply put, samples
that were close together tended to be more similar
environmentally and ecologically than ones far
apart, regardless of whether the environment was
mechanistically causing the observed vegetation
patterns or vice-versa (Legendre 1993). Although
the finding of nonsignificance after accounting for
Space does not rule out the possibility that these
variables were causally linked to vegetation, it does
suggest that many observed ?significant? relation-
ships reported in the Everglades and elsewhere may
have been due to spurious correlations that arose
from spatially structured data (Thomson and others
1997). Factoring out variation explained by other
variables provides further evidence that a signifi-
cant variable indeed may be a determinant of ob-
served pattern because its independent contribu-
tion can be more accurately assessed (Leduc and
others 1992). Thus, our Mantel approach did not
establish causation (Beyers 1998) but, we believe,
represented an improvement in analytical ap-
92 R. S. King and others
proaches used to infer direct linkages between spe-
cies and their environment, particularly in large,
spatially heterogeneous landscapes.
An added benefit afforded by the partial Mantel
approach was the ability to assess spatial variation
in vegetation composition not accounted for by the
environment?a spatially autocorrelated residual.
We were able to show that spatial proximity among
sampling locations (Space) accounted for significant
variation in vegetation pattern across the landscape,
particularly in the Impacted and Transition zones.
In both of these zones, Space accounted for the
most variation in composition, highly suggestive of
dispersal- or disturbance-influenced pattern [for
example, see Pickett and White (1985) and Pacala
and Levin (1997)]. Moreover, other than a weak
linkage to N, Space was the only variable to explain
variation in vegetation in the Impacted zone. On
the contrary, all of the spatial structure in the Ref-
erence zone was explained by environmental vari-
ation. Thus, these results suggest that, with increas-
ing impact to the ecosystem, spatial factors
overwhelmed and decoupled the linkages between
fine-scale vegetation pattern and environmental
factors observed in the Reference zone.
We further examined spatial pattern in vegeta-
tion between scales by using Mantel correlograms,
which indicated that the spatial structure in the
Transition and Impacted zones was occurring pri-
marily at coarse rather than fine scales. The Tran-
sition zone exhibited the greatest degree of coarse-
scale heterogeneity, as plots within clusters tended
to be much more similar than plots among clusters.
Coarse-scale heterogeneity diminished slightly in
the Impacted zone, but not to a statistically signifi-
cant degree (95% CLs). That spatial pattern in the
Impacted zone was so prevalent was surprising,
especially considering the widespread opinion that
this area is almost exclusively composed of mono-
typic stands of dense Typha [for example, see Wu
and others (1997) and Rutchey and Vilchek
(1999)]. However, these earlier studies relied on
remotely sensed data that may not have been suf-
ficient in resolution to detect all but the most dom-
inant species. It is likely that the resolution neces-
sary to quantify the patterns we observed would be
cost prohibitive and unrealistic to obtain by remote
sensing (Obeysekera and Rutchey 1997). Thus, our
results reinforce the importance of field studies in
landscape ecology and suggest that synergy be-
tween field and remote-sensing approaches is
needed to better understand the role of pattern and
process across multiple scales (Levin 1992).
Historical data indicate that vegetation in the Im-
pacted and Transition zones was very similar to that
of the Reference zone prior to the construction of
the canal-and-levee system (Davis 1943; US Geo-
logical Survey unpublished data). Using the Refer-
ence zone as a benchmark, an examination of pat-
terns between scales and among impact zones
revealed an intriguing paradox?one that sug-
gested different patterns through time and space,
depending on scale. In Figure 7, we illustrate a
conceptual model based on our results showing
generalized trajectories of Everglades vegetation
pattern at fine and coarse scales, and as intensity of
human influence increases. Our results suggest that
the fine-scale mosaic of Cladium stands and slough
communities was degraded by locally invasive veg-
etation through time in the Transition zone, con-
verging toward greater similarity among closely
neighboring locations. The once-heterogeneous
fine-scale pattern in the Impacted zone was de-
graded even further, with local pattern converging
toward greater homogeneity.
Moving across scales, however, the Reference
mosaic was homogeneous at a coarse scale (Figure
7). Here, fine-scale heterogeneity was repeated
across the landscape, and coarse scales integrated
this nested pattern much like Watt?s (1947) unit
pattern?fine-scale elements undergo shifts through
time and space, but the aggregate coarse pattern
essentially appears constant. In the presence of hu-
man influence, however, this was disrupted, and
fine-scale components gave rise to different pat-
terns at broader scales. In the Transition zone,
coarse pattern diverged from the Reference land-
scape, moving in many different trajectories based
on a variety of possible factors such as dispersal and
interspecific competition (Smith and Huston 1989;
Figure 7. Conceptual diagram showing generalized tra-
jectories of Everglades vegetation pattern among impact
zones at fine and coarse scales.
Space, Environment, and Wetland Vegetation 93
White 1994). Possibly, as different plants invaded
and competed here, early colonists amplified coarse
pattern through seed dispersal and vegetative prop-
agation, reinforcing their presence through positive
feedback (Pacala and Levin 1997; He and Mladenoff
1999). Typha, Mikania scandens, and Sarcostemma
clausum, three of the dominant species near the
canal inflow structures, effectively propagate vege-
tatively; thus, they may have contributed to coarse-
scale variation by differentially proliferating in areas
of the Transition zone. This high degree of fragmen-
tation at a coarse scale was a divergence from fine-
scale pattern in the Transition zone.
Finally, in the Impacted zone, coarse pattern con-
verged toward greater homogeneity from that ob-
served in the Transition zone but still remained
fragmented relative to the Reference zone (Figures
2c and 7). Similar to the Transition zone, the Im-
pacted zone was not dominated exclusively by
Typha but also by a host of other invasive species.
Thus, residual coarse pattern from its previous tran-
sitional phase, coupled with contagious distur-
bances and dispersal processes, may have been re-
sponsible for significantly greater heterogeneity at a
coarse scale than that observed in the Reference
landscape.
Using a hierarchical approach (Urban and others
1987), we took a ?snapshot? of vegetation compo-
sition at multiple scales and described the potential
trajectories of vegetation pattern as a function of
anthropogenic influence. Our results indicate that
human effects on the Everglades have resulted in
more than just a replacement of Cladium with
Typha. Collectively, these results indicate that allo-
genic spatial and environmental factors related to
the canal system have disrupted the coupling be-
tween pattern and process by altering fine-scale
vegetation?environment linkages and spatial pat-
terns characteristic of the natural Everglades eco-
system. These alterations certainly affect the spatial
ecology of higher organisms, such as invertebrates,
fish, birds, reptiles, and other wildlife, in a variety of
ways that may depend largely on landscape con-
nectivity and critical scales in their individual life
histories [for example, see MacArthur and Wilson
(1967) and Levin (1976)]. The implications of this
are great for restoration and management of vege-
tation, and suggest that serious attention needs to
be given toward mimicking the characteristic spatial
and temporal scales of pattern in the environment
(DeAngelis 1994; Holling and others 1994), as these
patterns ultimately drive self-organization in land-
scapes (Phillips 1999). The field of landscape ecol-
ogy has already begun to address many of these
scaling issues for terrestrial wildlife (for example,
spotted owl); such management approaches should
be extended to aquatic systems as well. One simple
management application of our findings could be
the use of correlograms of vegetation and environ-
mental spatial patterns from reference areas as
models to guide restoration efforts in disturbed ar-
eas of the Everglades ecosystem. Principles of hier-
archy theory [for example, see Allen and Star
(1982) and O?Neill and others (1986)] may provide
a framework for such efforts. In conclusion, our
study indicates a need for more research on the
spatial and temporal scales that are responsible for
vegetation patterns, and the role these patterns play
in the demographic processes of the flora and fauna
across the Everglades and other anthropogenically
influenced ecosystems.
ACKNOWLEDGEMENTS
We thank P. Heine, J. Rice, and W. Willis for help
with soil chemical analyses, J. Johnson and L.
Karppi for contributing long hours of assistance in
the field, and P. Vaithiyanathan for logistical sup-
port and study design suggestions. The article was
improved by comments from M. Baker, B. Bedford,
G. Bruland, D. DeAngelis, M. Hanchey, J. Vymazal,
D. Whigham, and three anonymous reviewers.
Funding was provided by a grant from the EAA
Environmental Protection District to the Duke Uni-
versity Wetland Center.
REFERENCES
Allen TFH, Starr TB. 1982. Hierarchy: perspectives for ecological
complexity. Chicago: University of Chicago Press.
Allen TFH, Wyleto EP. 1983. A hierarchical model for the com-
plexity of plant communities. J Theor Biol 101:529?40.
Bennington CC, Thayne WV. 1994. Use and misuse of mixed-
model analysis of variance in ecological studies. Ecology 75:
717?22.
Beyers DW. 1998. Causal inference in environmental impact
studies. J North Am Benthol Soc 17:367?73.
Bray JR, Curtis JT. 1957. An ordination of the upland forest
communities of southern Wisconsin. Ecol Monogr 27:325?49.
Browder JA, Gleason PJ, Swift DR. 1994. Periphyton in the
Everglades: spatial variation, environmental correlates, and
ecological implications. In: Davis SM, Ogden JC, Eds. Ever-
glades: the ecosystem and its restoration. Boca Raton (FL): St.
Lucie. p 379?418.
Busch DE, Loftus WF, Bass OL. 1998. Long-term hydrological
effects on marsh plant community structure in the southern
Everglades. Wetlands 18:230?41.
Chiang C, Craft CB, Rogers DW, Richardson CJ. 2000. Effects of
4 years of nitrogen and phosphorus additions on Everglades
plant communities. Aquat Bot 68:61?78.
Craft CB, Richardson CJ. 1997. Relationships between soil nu-
trients and plant species composition in Everglades peatlands.
J Environ Qual 26:224?32.
Craft CB, Vymazal J, Richardson CJ. 1995. Response of Ever-
94 R. S. King and others
glades plant communities to nitrogen and phosphorus addi-
tions. Wetlands 15:258?71.
Craighead FC. 1971. The trees of south Florida. Coral Gables
(FL): University of Miami Press.
Davis JH. 1943. The natural features of southern Florida. Fla
Geol Soc Geol Bull 25.
Davis SM. 1991. Growth, decomposition, and nutrient retention
of Cladium jamaicense Crantz and Typha domingensis Pers. in the
Florida Everglades. Aquat Bot 40:203?24.
Davis SM. 1994. Phosphorus inputs and vegetation sensitivity in
the Everglades. In: Davis SM, Ogden JC, Eds. Everglades: the
ecosystem and its restoration. Boca Raton (FL): St. Lucie. p
357?78.
Davis SM, Ogden JC. 1994. Toward ecosystem restoration. In:
Davis SM, Ogden JC, Eds. Everglades: the ecosystem and its
restoration. Boca Raton (FL): St. Lucie. p 769?96.
DeAngelis DL. 1994. Synthesis: spatial and temporal character-
istics of the environment. In: Davis SM, Ogden JC, Eds. Ever-
glades: the ecosystem and its restoration. Boca Raton (FL): St.
Lucie. p 307?22.
DeBusk WF, Reddy KR, Koch MS, Wang Y. 1994. Spatial distri-
bution of soil nutrients in a northern Everglades marsh: Water
Conservation Area 2A. Soil Sci Soc Am J 58:543?52.
Doren RF, Armentano TV, Whiteaker LD, Jones RD. 1997.
Marsh vegetation patterns and soil phosphorus gradients in
the Everglades ecosystem. Aquat Bot 56:145?63.
Dufre?ne M, Legendre P. 1997. Species assemblages and indicator
species: the need for a flexible asymmetrical approach. Ecol
Monogr 67:345?66.
Faith DP, Minchin PR, Belbin L. 1987. Compositional dissimilar-
ity as a robust measure of ecological distance. Vegetatio 69:
57?68.
Faith DP, Norris RH. 1989. Correlation of environmental vari-
ables with patterns of distribution and abundance of common
and rare freshwater macroinvertebrates. Biol Conserv 50:77?
98.
Flora MD, Walker DR, Scheidt DJ, Rice RG, Landers DH. 1988.
The response of the Everglades marsh to increased nitrogen
and phosphorus loading: Part I. Nutrient dosing, water chem-
istry, and periphyton productivity. Homestead (FL): Ever-
glades National Park.
Fortin M-J, Drapeau P, Legendre P. 1989. Spatial autocorrelation
and sampling design in plant ecology. Vegetatio 83:209?22.
Foster DR, Fluet M, Boose ER. 1999. Human or natural distur-
bance: landscape-scale dynamics of the tropical forests of
Puerto Rico. Ecol Appl 9:555?72.
Gleason PJ, Stone P. 1994. Age, origin, and landscape evolution
of the Everglades peatland. In: Davis SM, Ogden JC, Eds.
Everglades: the ecosystem and its restoration. Boca Raton
(FL): St. Lucie. p 149?98.
Grimshaw HJ, Wetzel RG, Brandenburg M, Segerblom K, Wen-
kert LJ, Marsh GA, Charnetzky W, Haky JE, Carraher C. 1997.
Shading of periphyton communities by wetland emergent
macrophytes: decoupling of algal photosynthesis from micro-
bial nutrient retention. Arch Hydrobiol 139:17?27.
Gunderson LH. 1989. Historical hydropatterns in wetland com-
munities of Everglades National Park. In: Sharitz RR, Gibbons
JW, Eds. Freshwater wetlands and wildlife. Oak Ridge (TN):
US Department of Energy. p 1099?111.
Gunderson LH, Snyder JR. 1994. Fire patterns in the southern
Everglades. In: Davis SM, Ogden JC, Eds. Everglades: the
ecosystem and its restoration. Boca Raton (FL): St. Lucie. p
291?306.
He HS, Mladenoff DJ. 1999. The effects of seed dispersal on the
simulation of long-term forest landscape change. Ecosystems
2:308?19.
Holling CS, Gunderson LH, Walters CJ. 1994. The structure and
dynamics of the Everglades ecosystem: guidelines for ecosys-
tem restoration. In: Davis SM, Ogden JC, Eds. Everglades: the
ecosystem and its restoration. Boca Raton (FL): St. Lucie. p
741?756.
Jensen JR, Rutchey K, Koch MS, Narumalani S. 1995. Inland
wetland change detection in the Everglades Water Conserva-
tion Area 2A using a time series of normalized remotely sensed
data. Photogramm Eng Rem Sens 61:199?209.
King RS. 2001. Dimensions of invertebrate assemblage organi-
zation across a phosphorus-limited Everglades landscape [dis-
sertation]. Durham (NC): Duke University
King RS, Richardson CJ. 2002. Evaluating subsampling ap-
proaches and macroinvertebrate taxonomic resolution for
wetland bioassessment. J North Am Benthol Soc 21:150?71.
Kleinbaum DG, Kupper LL, Muller KE. 1988. Applied regression
analysis and other multivariable methods. Belmont (CA):
Duxbury.
Leduc A, Drapeau P, Bergeron Y, Legendre P. 1992. Study of
spatial components of forest cover using partial Mantel tests
and path analysis. J Veg Sci 3:69?78.
Legendre P. 1993. Spatial autocorrelation: trouble or a new
paradigm? Ecology 74:1659?73.
Legendre P, Fortin M-J. 1989. Spatial pattern and ecological
analysis. Vegetatio 80:107?38.
Legendre P, Legendre L. 1998. Numerical ecology. 2nd ed. Am-
sterdam: Elsevier.
Levin SA. 1976. Population dynamic models in heterogeneous
environments. Ann Rev Ecol Syst 7:287?311.
Levin SA. 1992. The problem of pattern and scale in ecology.
Ecology 73:1943?67.
Loveless CM. 1959. A study of the vegetation in the Florida
Everglades. Ecology 40:1?9.
MacArthur RH, Wilson EO. 1967. The theory of island biogeog-
raphy. Princeton: Princeton University Press.
Manly BFJ. 1997. Randomization, bootstrap, and Monte Carlo
methods in biology. 2nd ed. London: Chapman and Hall.
Mantel N. 1967. The detection of disease clustering and a gen-
eralized regression approach. Cancer Res 27:209?20.
McCormick PV, Shuford RB, Backus JG, Kennedy WC. 1998.
Spatial and seasonal patterns of periphyton biomass and pro-
ductivity in the northern Everglades, Florida, U.S.A. Hydrobi-
ology 362:185?208.
Miao S, Newman S, Sklar FH. 2000. Effects of habitat nutrients
and seed sources on growth and expansion of Typha domingen-
sis. Aquat Bot 68:297?311.
Miao S, Sklar FH. 1998. Biomass and nutrient allocation of
sawgrass and cattail along a nutrient gradient in the Florida
Everglades. Wetlands Ecol Manag 5:245?63.
Minchin PR. 1987. An evaluation of the relative robustness of
techniques for ecological ordination. Vegetatio 69:89?107.
Newman S, Grace JB, Koebel JW. 1996. Effects of nutrients and
hydroperiod on Typha, Cladium, and Eleocharis: implications for
Everglades restoration. Ecol Appl 6:774?83.
Newman S, Schuette J, Grace JB, Rutchey K, Fontaine T, Reddy
Space, Environment, and Wetland Vegetation 95
KR, Peitrucha M. 1998. Factors influencing cattail abundance
in the northern Everglades. Aquat Bot 60:265?80.
Noe GB, Childers DL, Jones RD. 2001. Phosphorus biogeochem-
istry and the impacts of phosphorus enrichment: why is the
Everglades so unique? Ecosystems 4:603?24.
Obeysekera J, Rutchey K. 1997. Selection of scale for Everglades
landscape models. Landscape Ecol 12:7?18.
Oden NL, Sokal RR. 1986. Directional autocorrelation: an exten-
sion of spatial correlograms to two dimensions. System Zool
35:608?17.
O?Neill RV, DeAngelis DL, Waide JB, Allen TFH. 1986. A hier-
archical concept of the ecosystem. Princeton: Princeton Uni-
versity Press.
Pacala SW, Levin SA. 1997. Biologically generated spatial pattern
and the coexistence of competing species. In: Tilman D, Ka-
reiva P, Eds. Spatial ecology: the role of space in population
dynamics and interspecific interactions. Princeton: Princeton
University Press. p 204?32.
Pan Y, Stevenson RJ, Vaithiyanathan P, Slate J, Richardson CJ.
2000. Changes in algal assemblages along observed and ex-
perimental phosphorus gradients in a subtropical wetland,
U.S.A. Freshwater Biol 44:339?53.
Peterson GD. 2002. Contagious disturbance, ecological memory,
and the emergence of landscape pattern. Ecosystems 5:329?
38.
Phillips EA. 1959. Methods of vegetation study. New York: Holt.
Phillips JD. 1999. Divergence, convergence, and self-organiza-
tion in landscapes. Ann Assoc Am Geogr 89:466?88.
Pickett STA, White PS. 1985. The ecology of natural disturbance
and patch dynamics. Orlando (FL): Academic.
Reddy KR, DeBusk WF, Yang Y, DeLaune R, Koch M. 1991.
Physico-chemical properties of soils in Water Conservation
Area 2A of the Everglades. West Palm Beach (FL): Report to
the South Florida Water Management District.
Redfield GW. 2000. Ecological research for aquatic science and
environmental restoration in south Florida. Ecol Appl 10:990?
1005.
Richardson CJ, Ferrell GM, Vaithiyanathan P. 1999. Nutrient
effects on stand structure, resorption efficiency, and secondary
compounds in Everglades sawgrass. Ecology 80:2182?192.
Richardson CJ, Vaithiyanathan P, Romanowicz EA, Craft CB.
1997. Macrophyte community responses in the Everglades
with an emphasis on cattail (Typha domingensis) and sawgrass
(Cladium jamaicense) interactions along a gradient of long-term
nutrient additions, altered hydroperiod, and fire. In: Richard-
son CJ, Eds. Effects of phosphorus and hydroperiod alterations
on ecosystem structure and function in the Everglades. Report
to the Everglades Agricultural Area Environmental Protection
District. Duke Wetland Center Publication 97-05. Durham
(NC): Duke University. p 14.1?56.
Romanowicz EA, Richardson CJ. 1997. Hydrologic investigation
of Water Conservation Area 2A. In: Richardson CJ, Eds. Ef-
fects of phosphorus and hydroperiod alterations on ecosystem
structure and function in the Everglades. Report to the Ever-
glades Agricultural Area Environmental Protection District.
Duke Wetland Center Publication 97-05. Durham (NC): Duke
University. p 12.1?29.
Romanowicz EA, Richardson CJ. 2004. Hydrology gradients in
the Everglades. In: Richardson CJ, editor. The Everglades ex-
periments: lessons for ecosystem restoration. New York:
Springer-Verlag. [Forthcoming.]
Rutchey K, Vilchek L. 1999. Air photointerpretation and satellite
imagery analysis techniques for mapping cattail coverage in a
northern Everglades impoundment. Photogramm Eng Rem
Sens 65:185?91.
SFWMD (South Florida Water Management District). 1992. Sur-
face water improvement plan for the Everglades. West Palm
Beach (FL): Supporting Information Document, South Florida
Water Management District.
SFWMD (South Florida Water Management District). 1995. Us-
er?s guide to REMO, remote access to the South Florida Water
Management District?s water quality and hydrometeorological
databases. West Palm Beach (FL): South Florida Water Man-
agement District.
SFWMD (South Florida Water Management District). 2000.
2000 Everglades consolidated report. West Palm Beach (FL):
Supporting Information Document, South Florida Water
Management District.
Shay JM, De Geus PMJ, Kapinga MRM. 1999. Changes in shore-
line vegetation over a 50-year period in the Delta Marsh,
Manitoba in response to water levels. Wetlands 19:413?25.
Smith T, Huston M. 1989. A theory of the spatial and temporal
dynamics of plant communities. Vegetatio 83:49?69.
Smouse PE, Long JC, Sokal RR. 1986. Multiple regression and
correlation extensions of the Mantel test of matrix correspon-
dence. System Zool 35:627?32.
Steward KK, Ornes WH. 1975a. Assessing a marsh environment
for wastewater renovation. Water Pollut Control Fed J 47:
1880?91.
Steward KK, Ornes WH. 1975b. The autecology of sawgrass in
the Florida Everglades. Ecology 56:162?71.
Swift DR, Nicholas RB. 1987. Periphyton and water quality
relationships in the Everglades Water Conservation Areas,
1978?1982. West Palm Beach (FL): South Florida Water Man-
agement District.
Thomson JD, Weiblen G, Thomson BA, Alfaro S, Legendre P.
1997. Untangling multiple factors in spatial distributions: lilies,
gophers, and rocks. Ecology 77:1698?715.
Toth LA. 1987. Effects of hydrologic regimes on lifetime produc-
tion and nutrient dynamics of sawgrass. West Palm Beach
(FL): South Florida Water Management District, Technical
Publication 87?6.
Toth LA. 1988. Effects of hydrologic regimes on lifetime produc-
tion and nutrient dynamics of cattail. West Palm Beach (FL):
South Florida Water Management District, Technical Publica-
tion 88?6.
Turner AM, Trexler JC, Jordan CF, Slack SJ, Geddes P, Chick JH,
Loftus WF. 1999a. Targeting ecosystem features for conserva-
tion: standing crops in the Florida Everglades. Conserv Biol
13:898?911.
Turner SJ, Hewitt JE, Wilkinson MR, Morrisey DJ, Thrush SF,
Cummings VJ, Funnell G. 1999b. Seagrass patches and land-
scapes: the influence of wind?wave dynamics and hierarchical
arrangements of spatial structure on macrofaunal seagrass
communities. Estuaries 22:1016?32.
Urban DL. 2000. Using model analysis to design monitoring
programs for landscape management and impact assessment.
Ecol Appl 10:1820?1832.
Urban DL, O?Neill RV, Shugart HH. 1987. Landscape ecology: a
hierarchical perspective can help scientists understand spatial
patterns. Bioscience 37:119?27.
Urban D, Goslee S, Pierce K, Lookingbill T. 2002. Extending
community ecology to landscapes. Ecoscience 12:200?12.
Urban NH, Davis SM, Aumen NG. 1993. Fluctuations in sawgrass
96 R. S. King and others
and cattail densities in Everglades Water Conservation Area
2A under varying nutrient, hydrologic and fire regimes. Aquat
Bot 46:203?23.
Vaithiyanathan P, Richardson CJ. 1999. Macrophyte species
changes in the Everglades: examination along a eutrophica-
tion gradient. J Environ Qual 28:1347?58.
Van der Valk AG, Rosburg TR. 1997. Seed bank composition
along a phosphorus gradient in the northern Florida Ever-
glades. Wetlands 17:228?36.
Vymazal J, Craft CB, Richardson CJ. 1994. Periphyton response
to nitrogen and phosphorus additions in Florida Everglades.
Algol Stud 73:75?97.
Vymazal J, Richardson CJ. 1995. Species composition, biomass,
and nutrient content of periphyton in the Florida Everglades.
J Phycol 31:343?54.
Walker DR, Flora MD, Rice RG, Scheidt DJ. 1989. Response of
the Everglades marsh to increased nitrogen and phosphorus
loading: II. Macrophyte community structure and chemical
composition. Homestead (FL): Everglades National Park.
Watt AS. 1947. Pattern and process in the plant community. J
Ecol 35:1?22.
White PS. 1994. Synthesis: vegetation pattern and process in the
Everglades ecosystem. In: Davis SM, Ogden JC, Eds. Ever-
glades: the ecosystem and its restoration. Boca Raton (FL): St.
Lucie. p 445?60.
Whittaker RH. 1956. Vegetation of the Great Smoky Mountains.
Ecol Monogr 26:1?80.
Wiens JA. 1989. Spatial scaling in ecology. Funct Ecol 3:385?97.
Wu Y, Sklar FH, Rutchey K. 1997. Analysis and simulations of
fragmentation patterns in the Everglades. Ecol Appl 7:268?76.
Zmyslony J, Gagnon D. 2000. Path analysis of spatial predictors
of front-yard landscape in an anthropogenic environment.
Landscape Ecol 15:357?71.
Space, Environment, and Wetland Vegetation 97