1 23
Biological Invasions
 
ISSN 1387-3547
 
Biol Invasions
DOI 10.1007/s10530-016-1136-z
Local and regional disturbances associated
with the invasion of Chesapeake Bay
marshes by the common reed Phragmites
australis
M. Benjamin Sciance, Christopher
J. Patrick, Donald E. Weller, Meghan
N. Williams, Melissa K. McCormick &
Eric L. G. Hazelton
1 23
Your article is protected by copyright and
all rights are held exclusively by Springer
International Publishing Switzerland (outside
the USA). This e-offprint is for personal
use only and shall not be self-archived
in electronic repositories. If you wish to
self-archive your article, please use the
accepted manuscript version for posting on
your own website. You may further deposit
the accepted manuscript version in any
repository, provided it is only made publicly
available 12 months after official publication
or later and provided acknowledgement is
given to the original source of publication
and a link is inserted to the published article
on Springer's website. The link must be
accompanied by the following text: "The final
publication is available at link.springer.com”.
PHRAGMITES INVASION
Local and regional disturbances associated with the invasion
of Chesapeake Bay marshes by the common reed Phragmites
australis
M. Benjamin Sciance . Christopher J. Patrick . Donald E. Weller . Meghan N. Williams .
Melissa K. McCormick . Eric L. G. Hazelton
Received: 15 October 2015 /Accepted: 25 March 2016
 Springer International Publishing Switzerland (outside the USA) 2016
Abstract The invasion of wetlands by Phragmites
australis is a conservation concern across North
America. We used the invasion of Chesapeake Bay
wetlands by P. australis as a model system to examine
the effects of regional and local stressors on plant
invasions. We summarized digital maps of the distri-
butions of P. australis and of potential stressors
(especially human land use and shoreline armoring) at
two spatial scales: for 72 subestuaries of the bay and
their local watersheds and for thousands of 500 m
shoreline segments. We developed statistical models
that use the stressor variables to predict P. australis
prevalence (% of shoreline occupied) in subestuaries
and its presence or absence in 500 m segments of
shoreline. The prevalence of agriculture was the
strongest and most consistent predictor of P. australis
presence and abundance in Chesapeake Bay, because
P. australis can exploit the resulting elevated nutrient
levels to enhance its establishment, growth, and seed
production. Phragmites australis was also positively
associated with riprapped shoreline, probably because
it creates disturbances that provide colonization
opportunities. The P. australis invasion was less
severe in areas with greater forested land cover and
natural shorelines. Surprisingly, invasion was low in
highly developed watersheds and highest along shore-
lines with intermediate levels of residential land use,
possibly indicating that highly disturbed systems are
Guest editors: Laura A. Meyerson and Kristin Saltonstall/
Phragmites invasion.
M. B. Sciance  C. J. Patrick  D. E. Weller (&) 
M. N. Williams  M. K. McCormick  E. L. G. Hazelton
Smithsonian Environmental Research Center, 647
Contees Wharf Rd., Edgewater, MD 21037, USA
e-mail: wellerd@si.edu
M. B. Sciance
e-mail: sciancemb@gmail.com
C. J. Patrick
e-mail: Christopher.Patrick@tamucc.edu
M. N. Williams
e-mail: williamsme@si.edu
M. K. McCormick
e-mail: mccormickm@si.edu
E. L. G. Hazelton
e-mail: eric@hazelton-ecological.com
M. B. Sciance
Department of Geography and Geology, University of
North Carolina Wilmington, 601 S. College Rd.,
Wilmington, NC 28403, USA
E. L. G. Hazelton
Ecology Center and Department of Watershed Sciences,
Utah State University, 5205 Old Main Hill-NR 314,
Logan, UT 84322, USA
Present Address:
C. J. Patrick
Department of Life Sciences, Texas A&M University Corpus
Christi, 6300 Ocean Dr., Corpus Christi, TX 78412, USA
123
Biol Invasions
DOI 10.1007/s10530-016-1136-z
Author's personal copy
uninhabitable even to invasive species. Management
strategies that reduce nutrient pollution, preserve
natural shorelines, and limit nearshore disturbance of
soils and vegetation may enhance the resilience of
shorelines to invasion.
Keywords Phragmites australis  Land use 
Subestuary  Shoreline armoring  Invasion 
Disturbance
Introduction
Disturbance has long been considered to support the
invasion of biological communities by exotic species
(Colautti et al. 2006), but disturbance is also a natural
part of most environments. In a recent meta-analysis,
Moles et al. (2012) reported that disturbance alone is
often a poor predictor of community invasibility; but
measures of change in the natural disturbance regime,
particularly anthropogenic disturbance, provide better
predictors. Other factors such as propagule availability
and land use surrounding a site also affect invasibility,
and these factors may obscure the effects of changes in
the disturbance regime (Moles et al. 2012). Pysˇek et al.
(2010) report that national wealth and human popu-
lation density (which they considered to be proxies for
anthropogenic disturbance, eutrophication, and
propagule pressure) were strongly correlated with
invasion by alien species.
Understanding and managing the effects of anthro-
pogenic disturbance on invasions requires identifying
and integrating proximal factors acting near invaded
sites with broader factors acting across the surrounding
landscape. Field studies have investigated the effects of
disturbances at specific sites, but there has been less
attention to comparing the spatial pattern of invasion to
the spatial distribution of disturbance across large areas.
Broader-scale investigations are needed because the
factors driving regional distributions may differ from
drivers of distribution within individual ecosystems.
Broader-scale analyses have been completed for other
ecological responses in estuaries through spatial analysis
of digital maps across large regions (Comeleo et al.
1996; Paul et al. 2002; Li et al. 2007; Rodriguez et al.
2007; Patrick et al. 2014, 2016).
Plant invasions of coastal estuarine wetlands
provide an ideal system for integrating the effects of
disturbance at different scales on invasion success. At
the broad scale of entire watersheds and their receiving
waters, human activities on the land release nutrients
that drive eutrophication in coastal ecosystems (Nixon
1995; Smith 2006). Nitrogen and phosphorus are
introduced from many activities, including crop and
livestock agriculture, urban and suburban land use,
fossil-fuel combustion, and point sources such as
sewage treatment plants (Anderson et al. 2002;
Shuster et al. 2005; Conley et al. 2009). Anthro-
pogenic eutrophication is one of the biggest manage-
ment concerns for aquatic systems (Smith and
Schindler 2009) where excess nutrients are often
exploited by undesirable or harmful flora and fauna
(e.g., algae blooms). Eutrophication is a central
problem in Chesapeake Bay (Boesch et al. 2001;
Kemp et al. 2005) that has led to federal mandates to
reduce nutrient inputs from the land and to multistate
efforts to meet that mandate (USEPA 2010).
Disturbances along the shoreline may also con-
tribute to invasion of coastal estuarine systems.
Shoreline disturbance includes the removal of vege-
tation (Silliman and Bertness 2004) and construction
of erosion control structures such as riprap, bulkheads,
jetties, and groins (Long et al. 2011; Bertness et al.
2002). Removal of vegetation (and its associated
nutrient demand) can significantly increase nutrient
delivery to the adjacent water (Sweeney and Newbold
2014; Weller and Baker 2014). The construction of
shoreline armoring creates disturbed soils that provide
openings for invasive plants to establish and prolifer-
ate (Strayer et al. 2012). Coastal estuarine wetlands
simultaneously integrate the impacts of these multiple
stressors operating at different scales.
Coastal marshes in the Chesapeake and throughout
North America are being invaded by an aggressive
subspecies of common reed (P. australis ssp. aus-
tralis). The native strain (P. australis ssp. americanus)
was previously a small part of marsh assemblages, but
recently the invasive subspecies has been aggressively
extending its distribution in wetlands across North
America (Chambers et al. 1999; Saltonstall 2002). In
the Chesapeake Bay, the invasive strain was noted in
small numbers in several watersheds in the 1970s, but
its distribution in the Bay has expanded, reaching
10–30 % of the shoreline in different sections of the
upper Bay by 2005 (Chambers et al. 1999, 2008;
Meadows and Saltonstall 2007).
The invasion of marshes by P. australis ssp. australis
has a number of ecological and environmental
M. B. Sciance et al.
123
Author's personal copy
consequences. Invaded wetlands have experienced
declines in native marsh grass biodiversity (Burdick
and Konisky 2003; Chambers et al. 1999); epifaunal
abundance (Robertson and Weis 2005); and fish, birds,
and swimming crustaceans (Osgood et al. 2003; Dibble
et al. 2013). The widespread invasion and the negative
impacts it has on native species have broughtP. australis
ssp. australis to the forefront of wetland studies
(Mozdzer et al. 2013; Hazelton et al. 2014), and made
it one of the most studied plant systems in North
America (Meyerson et al. 2009, 2016).
P. australis stands have larger nitrogen pools than
other marsh communities (Meyerson et al. 1999, 2000;
Windham and Meyerson 2003); and anthropogenic
nitrogen can promote the expansion of the invasive P.
australis strain (King et al. 2007; Chambers et al. 2008;
Kettenring et al. 2011). Compared to the native
subspecies, the invasive one demands four times more
nitrogen to support its biomass than the native sub-
species (Mozdzer and Zieman 2010), but the invasive
strain can exploit high nitrogen levels to achieve higher
growth rates and biomass than the native strain and other
species of marsh plants (Mozdzer et al. 2013). Proximity
to agriculture (and the associated nitrogen inputs from
fertilizer and animal waste) is positively correlated with
P. australis abundance (Chambers et al. 2008; Mazur
et al. 2014). Other anthropogenic disturbance besides
nitrogen inputs may also promote the invasive strain,
including shoreline alteration, sea level rise, increased
anthropogenic carbon, and altered salinity (e.g., Meyer-
son et al. 2010;Mozdzer andMegonigal 2012; Guo et al.
2013). Few studies have looked at the role of shoreline
disturbance in contributing to P. australis invasion
(Bertness et al. 2002; Silliman and Bertness 2004; Long
et al. 2011).
We used the invasion of Chesapeake Bay marshes
by P. australis as a model system for examining the
roles of regional and proximal anthropogenic distur-
bance in biological invasion. P. australis has been
acknowledged as a good model organism for advanc-
ing understanding of plant invasions (Meyerson et al.
2016). We used available digital spatial data to
quantify P. australis presence and prevalence and to
develop potential predictor variables representing
anthropogenic disturbance. At the regional scale, we
related P. australis prevalence across many embay-
ments (subestuaries) to variables quantifying human
activities in the local watershed and along the entire
shoreline of each subestuary. At the more proximal
scale, we related P. australis presence in 500 m
segments of shoreline to variables representing human
activities in each segment. We compared and inte-
grated different statistical modeling approaches within
and between the two spatial scales. We addressed two
basic questions: which anthropogenic factors affect P.
australis invasion success at the two spatial scales
(subestuary and shoreline), and do factors correlated
with invasion success differ between the two scales?
At both the shoreline segment and subestuary
scales, we expected that the amounts of agricultural
and developed land would be positively correlated
with P. australis prevalence because of nutrient
enrichment from those land uses. We also expected
to find that armored shorelines would be positively
related to P. australis presence because the associated
disturbance of soils and native marsh communities
provides an opening for invasion. We expected these
effects to act synergistically with the nutrient-enrich-
ing effects of agricultural and developed lands.
Methods
Study area
The Chesapeake Bay (Fig. 1) is the largest estuary in
North America, and its 166,000 km2 watershed
includes parts of six states and Washington, DC
(Weller and Baker 2014; CBP 2015). Nearly 18
million people live in the watershed (Kemp et al.
2005), many in large metropolitan areas, including
Baltimore MD,Washington DC, and Hampton Roads-
Norfolk VA. In rural parts of the watershed, forests
and other natural lands are interspersed with livestock
and row crop agriculture.
The Chesapeake Bay’s complex shoreline encloses
over 100 tributary embayments (here called subestuar-
ies), each of which has its own local watershed (Li et al.
2007). The local watersheds differ widely in their
proportions of forest, agricultural, developed, and other
land cover types; and the bay and its subestuaries are
commonly divided into four major salinity zones:
polyhaline, mesohaline, oligohaline, and tidal fresh
(Fig. 1, Li et al. 2007; Patrick et al. 2014). Thus, the
subestuaries provide a population of replicate study units
well suited for exploring the interacting effects of land
use and salinity on estuaries. That replication has been
exploited in recent statistical analyses relating watershed
Local and regional disturbances associated with the invasion of Chesapeake Bay marshes
123
Author's personal copy
and subestuary characteristics to ecological responses in
the subestuaries (King et al. 2004, 2007; Li et al. 2007;
Patrick et al. 2014, 2016; Patrick and Weller 2015).
Phragmites australis and shoreline data
The data on P. australis distribution and shoreline
characteristics for this study came from a shoreline
inventory (VIMS-CCRM 2009) that mapped
17,047 km of shoreline, roughly 80 % of the Chesa-
peake Bay shore. The shoreline was mapped once, but
it took 11 years to cover the study area. Eighty percent
of the data were collected between 1998 and 2002, and
20 % between 2003 and 2008. The data set contains
three spatial layers. For 99,763 individual segments
comprising the surveyed shoreline, a land use layer
and bank condition layer reports adjacent land use
(agriculture, forest, marsh, and residential land), bank
height (0–5, 5–10, 10–30, and[30 m), and vegetative
cover of the bank. All banks[5 m high were steep, so,
we simplified the bank height information from five to
two categories, low banks (0–5 m) and high banks
([5 m). For a separate set of 74,010 individual
segments also comprising the surveyed shoreline, a
shoreline armoring layer maps linear armoring struc-
tures, such as riprap, bulkhead, groin field, seawall,
and breakwater. The third spatial layer maps 47,762
point structures along the shoreline, such as docks,
boat ramps, or marinas.
Importantly, the land use and bank condition layer
reports the presence or absence of P. australis in each
mapped segment. That reporting does not distinguish
between the native and invasive non-native P. aus-
tralis subspecies; however, in 2003–2004, Meadows
and Saltonstall (2007) found the invasive strain in five
times more patches than the native strain on the Bay’s
eastern shore. A 2006 survey of P. australis stands on
both shores of the Bay found the native strain in only
7 % of the stands (Tulbure et al. 2012). Because most
of the P. australis stands are the invasive strain,
observed spatial relationships of P. australis preva-
lence and abundance with potential controlling factors
(below) are strongly dominated by the responses of the
invasive strain.
Analyses
We summarized data and developed statistical models
at two scales of analysis: The coarser-scale analysis
related subestuary-wide P. australis prevalence to
land use in the local watershed and characteristics of
the entire shoreline of each subestuary. The finer-scale
analysis related P. australis presence or absence in
individual 500 m segments of shoreline to land use
and other characteristics at each 500 m segment.
Spatial analyses were performed with ArcGIS 10
geographic information software (ESRI 2011), and
statistical analyses were implemented within the R
statistical software system (R Core Team 2014).
Phragmites australis prevalence in subestuaries
From a larger group of previously defined subestuaries
(Li et al. 2007; Patrick et al. 2014), we selected the
subset of 72 Chesapeake Bay subestuaries that had been
Fig. 1 The Chesapeake Bay and the local watersheds of 72
subestuaries containing Phragmites australis mapped in the
shoreline survey (VIMS-CCRM2009). The local watersheds are
shaded by the salinity zone of the subestuary: tidal fresh (TF),
oligohaline (OH), mesohaline (MH), and polyhaline (PH). The
inset shows the location of the Chesapeake Bay watershed and
six states in the mid-Atlantic region of the U.S
M. B. Sciance et al.
123
Author's personal copy
surveyed in the shoreline survey (VIMS-CCRM 2009)
and contained P. australis. We omitted subestuaries
without P. australis because we could not determine
whether it was absent because of environmental factors
or because of a lack of invasive propagules. Including
systems lacking propagules would confound analyses
relating prevalence to environmental factors. Subestu-
aries were assigned to salinity zones (tidal fresh,
oligohaline, mesohaline, or polyhaline) based on the
Chesapeake Bay Program Segmentation salinity seg-
mentation scheme (http://www.chesapeakebay.net/
segmentscheme.htm, see Patrick et al. 2014).
Land cover proportions in the local watershed of
each subestuary were estimated from the circa 2000
National Land Cover Dataset (called NLCD 2001,
Homer et al. 2004) by intersecting the polygons
defining the local watersheds of the subestuaries with
the land cover map within the GIS and then calculating
the proportion of each land cover class within each
watershed. We aggregated some of the NLCD cate-
gories into five broader land cover types: cropland
(NLCD code 82, cropland), developed (all developed
categories), forest (all types of forest and woody
wetland), and wetland (emergent herbaceous wet-
lands, NLCD code 95).
We summarized the subestuary polygons and their
intersection with a digital bathymetry map (NOAA
1998) to calculate the metrics describing subestuaries,
including the shoreline perimeter, mouth width,
subestuary volume, and the fractal dimension of the
shoreline. The fractal dimension is a measure of
shoreline complexity calculated as two times the natural
logarithm of subestuary perimeter divided by the natural
logarithm of subestuary area (Li et al. 2007). Average
tidal range was estimated by intersecting the subestuary
boundaries with a digital map of coastal vulnerability
(Thieler and Hammar-Klose 1999) based on a coastal
hazards database (Gornitz and White 1992) updated
with more recent data (Thieler and Hammar-Klose
1999). That map included tidal range data interpolated
among 657 tide stations by Hubertz et al. (1996).
Shoreline characteristics were summarized for each
subestuary and recorded as percentage of total
subestuary shoreline length occupied by each shore-
line land use and shoreline armoring type (such as
bulkhead or riprap); or as number of structures per km
of shoreline for point structures like docks, boat-
houses, or wharves. Phragmites australis prevalence
was quantified as the percentage of total subestuary
shoreline length occupied byP. australis by adding the
lengths of individual segments containing P. australis
and dividing by the total subestuary shoreline length.
We used univariate linear regression (lm function,
R Core Team 2014) to test for hypothesized relation-
ships (see Introduction) between P. australis preva-
lence (% of subestuary shoreline with P. australis) and
31 potential predictors from watershed, estuary, and
shoreline variables. When considering many univari-
ate regressions, it is necessary to control for experi-
ment wise error across the entire set of regressions The
Dunn–Sˇida´k correction (Ellison and Gotelli 2004)
accomplishes this by estimating a more stringent
(lower) P level for statistical significance of an
individual regression to preserve the desired overall
significance (P B 0.05) for the set of regressions.
With 31 regressions, the corrected P level for signif-
icance of an individual regression is P B 0.0016.
We also applied multiple linear regression to all of
the independent variables to identify a multivariate
linear model for predicting P. australis in a subestu-
ary. All possible multiple regression models with 2–5
predictors were estimated with best subsets analysis
(Hosmer et al. 1989) and ranked with the Bayesian
information criterion (BIC) using the R Leaps package
(Tumley 2009).
As an alternative multivariate model to relate P.
australis prevalence in subestuaries to watershed,
estuary, and shoreline characteristics; we also imple-
mented regression tree analysis with the randomForest
package in R (Breiman and Cutler 2014). Regression
tree analysis requires fewer assumptions than linear
regression analysis and can work well with non-normal
variables, non-linear responses, and non-continuous
variables (De’Ath and Fabricius 2000). Random forest
(RF) analysis is an implementation of tree analysis that
works by creating multiple decision trees, each with a
different bootstrapped sample of the original data. The
final prediction is obtained by aggregating over the
ensemble of trees (Biau 2012). We assessed the results
of the RF model by examining the rankings of variable
importance to model construction and examining partial
dependence plots of the most important variables
(Cutler et al. 2007). Partial dependence plots visualize
the effects of individual predictors on response vari-
ables. A partial dependence plot shows the average
response of the predicted variable to the predictor,
across all other combinations of predictor variables used
in the RF analysis (Cutler et al. 2007).
Local and regional disturbances associated with the invasion of Chesapeake Bay marshes
123
Author's personal copy
Phragmites australis presence or absence
in individual shoreline segments
To quantify Phragmites presence in individual shoreline
segments, we processed the shoreline data differently
from the summary of Phragmites prevalence in subestu-
aries (above). The survey mapped land use, bank
condition and P. australis presence for 99,763 segments
comprising 17,047 km of shoreline (VIMS-CCRM
2009). After eliminating some segments where shoreline
features were not recorded or where extensive marshes
were not carefully surveyed for P. australis presence
(Marcia Berman, Virginia Institute of Marine Science,
personal communication), there were 13,872 km of
shoreline left for further analysis. The lengths of
segments reporting P. australis presence varied from
less than 1 meter to 15.4 km. The variation in size
presents a problem for statistical analysis because the
factors affecting very small segments may be different
from the factors affecting larger segments. Therefore, we
resampled the land use and bank condition map to
produce a new map divided into shoreline segments of
consistent 500 m length. To accomplish this, we merged
adjacent segments with identical attributes and then
divided the shoreline layer at 500 m intervals. There
were 26,590 segments available for analysis after
residual segments less than 500 m were eliminated.
We also resampled the map layer reporting shoreline
armoring to match the 500 m segments created above.
We summarized the shoreline attributes for each
500 m segment. From the resampled land use and
bank condition layer, we recorded the presence or
absence of P. australis in each 500 m segment and the
percentages of the segment’s 500 m length with three
different land uses (agricultural land, residential land,
and forest). From the resampled shoreline armoring
layer, we recorded the percentages of three different
shoreline armoring conditions (natural, riprap, or
bulkhead). Minor land uses and armoring types were
omitted from the analysis. We also calculated the
average bottom slope within 90 m of each 500 m
shoreline segment using a slope grid derived from a
digital bathymetric model (NOAA 1998).
We used univariate logistic regression to test our
hypotheses (see ‘‘Introduction’’ section) about the
effects of land use and shoreline armoring in shoreline
segments on P. australis presence in those segments.
We implemented logistic regression with the R glm
function (R Core Team 2014), and we related
P. australis presence or absence to seven candidate
predictor variables: three land uses (% forest, %
agricultural land, and % residential land), three shore-
line armoring conditions (% unarmored, % bulkhead
and % riprap), and average slope at the shoreline. The
number of shoreline segments was quite large (26,590),
so it was possible to omit some of the segments from
model fitting and use the reserved data for model
validation. We randomly selected half of the 500 m
aggregated shoreline segments for fitting the logistic
models. As above, we used the Dunn–Sˇida´k correction
(Ellison and Gotelli 2004) to control for experiment
wise error across the set of univariate predictors when
testing whether models were statistically significant.
With seven logistic regressions, setting the corrected
P level for significance of an individual regression to
P B 0.0073 yields an experiment wise significance
level of P\ 0.05. The optimal classification threshold
for each significant logistic model was calculated using
receiver operator characteristic (ROC) analysis, which
evaluates the percentages of correctly predicted pres-
ences and absences across a range of probability
thresholds to identify the threshold that maximizes
correct predictions (Hosmer and Lemeshow 2000).
We evaluated model performance with Cohen’s
kappa (j). Values of j can range from-1 to 1. Positive
values indicate that the classification is more successful
than would be expected from chance alone, whereas
negative values indicate worse results than expected
from chance alone. A value of ?1 indicates perfect
agreement between the modeled and measured classi-
fications (Cohen 1960; Manel et al. 2001). We selected
a best model from the seven univariate logistic models
based on the statistical significance levels (P), the j
values, and AIC scores (Burnham and Anderson 2002).
Then, we applied that model to the other half of the data
that had been reserved for validation and assessed
model performance for the validation data with j.
We implemented multiple logistic regression (R glm
function; R Core Team 2014) to assess whether some of
the seven predictor variables could be combined in a
multivariate model to improve success in predicting P.
australis presence or absence. We programmed a simple
permutation procedure to generate potential models and
then performed model comparison (see ‘‘Methods’’
section above) to identify which combination of land
use and shoreline structure variables constituted the best
model. As above, that model was further tested by
applying it to the validation data and calculating j.
M. B. Sciance et al.
123
Author's personal copy
Results
Phragmites australis distribution
within Chesapeake Bay
Of the 13,872 km of shoreline analyzed, 6.6 % was
occupied by P. australis (Fig. 1). Phragmites aus-
tralis prevalence in the Chesapeake Bay decreased
from the north to south. Subestuaries with the
greatest percentage of shoreline occupied by P.
australis were in the northeastern portion of the bay
(Fig. 2). Phragmites australis was more prevalent in
subestuaries within the oligohaline and mesohaline
salinity zones than in the tidal fresh and polyhaline
subestuaries (Fig. 3).
Fig. 2 Phragmites australis distribution in the Chesapeake Bay
as mapped by the shoreline survey (VIMS-CCRM 2009).
a Individual shoreline segments containing P. australis, b local
watersheds of 72 Chesapeake Bay subestuaries (Fig. 1) shaded
by the percent of shoreline occupied by P. australis. Some of the
shoreline occupied by P. australis (e.g., arrows in a) is not
within one of the 72 study subestuaries in b. Some subestuaries
are small but have relatively large local watersheds (arrows in
b). The shading of local watersheds in b represents the
prevalence of P. australis along the shoreline of the corre-
sponding subestuary near the bay, not throughout the local
watershed
Fig. 3 Phragmites australis prevalence within in the surveyed
subestuaries (Fig. 2b) summarized by salinity zone (Fig. 1)
Local and regional disturbances associated with the invasion of Chesapeake Bay marshes
123
Author's personal copy
Predicting Phragmites australis prevalence
in subestuaries
Three of the 31 variables representing shoreline,
estuary, or watershed characteristics had statistically
significant univariate linear relationships with P.
australis prevalence in subestuaries after applying
the Dunn–Sˇida´k correction (P\ 0.0016) to control
experiment wise error at P\ 0.05 (Table 1). The
percentage of agricultural land use on the shoreline
and the percentage of agricultural cover in the local
watershed had the strongest positive relationships with
P. australis prevalence (both R2 = 0.20, P\ 0.001),
while the percent of forested cover within the local
watershed was negatively related to P. australis
prevalence (P\ 0.001, R2 = 0.19).
Among multiple regression models with five or
fewer independent variables, there were four models
that were equivalently good at predicting P. australis
prevalence in subestuaries (Table 2). All four explained
more of the variability in prevalence among subestu-
aries (R2[ 0.5, P\ 0.0001, Table 2) than did the best
Table 1 Univariate
regressions of the P.
australis prevalence in
subestuaries versus 16
shoreline variables from the
shoreline survey (VIMS-
CCRM 2009) and versus 15
other variables describing
the subestuary-watershed
system
Variables are sorted by
decreasing R2 within
groups. The three
relationships with
P\ 0.001 are significant
below the Dunn–Sˇida´k
threshold (P\ 0.0016) that
controls experiment wise
error to P\ 0.05
Predictor Intercept Slope R2 P
Shoreline characteristics
% Agriculture 7.01 0.60 0.20 \ 0.001
% Beach 15.91 -0.49 0.04 0.08
% High banks 15.47 -0.18 0.04 0.09
Boat ramp density 9.39 8.63 0.03 0.13
% Forested shoreline 16.87 -0.15 0.03 0.15
Length of surveyed shoreline 15.10 -0.03 0.03 0.16
% Vegetated shore 17.87 -0.13 0.03 0.17
% Riprap 9.34 0.33 0.02 0.22
% Developed shoreline 16.12 -0.11 0.02 0.24
% Bulkhead shoreline 13.68 -0.15 0.01 0.40
Average slope at shore 15.69 -1.19 0.01 0.41
% Other shoreline hardening 13.44 -0.44 0.01 0.51
% Low banks 9.03 0.05 0.00 0.58
Dock and boathouse density 13.00 -0.14 0.00 0.71
Marina and wharf density 12.79 -3.62 0.00 0.72
% Residential shoreline 13.01 -0.02 0.00 0.81
Subestuary or watershed characteristics
% Cropland 3.90 47.99 0.20 \ 0.001
% Forest 26.72 -35.36 0.19 \ 0.001
Tidal range 28.28 -35.13 0.10 0.01
Watershed perimeter 17.28 -0.05 0.06 0.04
% Grassland 8.23 57.62 0.05 0.05
Fractal dimension 95.41 -59.34 0.04 0.09
Watershed area (km2) 14.04 -0.01 0.03 0.17
Subestuary shoreline perimeter (km) 14.09 -0.01 0.02 0.19
Watershed/subestuary area ratio 14.38 -0.17 0.02 0.23
Subestuary basin volume (km3) 13.79 -39.51 0.02 0.24
Subestuary area (km2) 13.83 -0.09 0.02 0.29
% Wetland 11.07 37.43 0.01 0.37
Subestuary mouth width 14.03 -0.70 0.01 0.43
% Developed land 13.43 -6.54 0.01 0.48
Salinity (ppt) 14.11 -0.18 0.00 0.60
M. B. Sciance et al.
123
Author's personal copy
univariate models (R2 = 0.20, Table 1). Across the
fourmultiple regressionmodels,P. australis prevalence
was positively related to three variables (cropland and
wetland land cover in the watershed and boat ramp
density) and negatively related to three others (propor-
tion of shore with high banks, the average slope at
shore, and average tidal range).
The random forest model explained 39 percent of the
variation in P. australis prevalence among subestuaries
(P\ 0.0001). The six most important variables in the
model were the percentage of cropland in the water-
shed, the percentage of agriculture land use at the
shoreline, tidal range, the percentage of forest in the
watershed, the slope at shoreline, and the percentage of
shoreline with beach (see variable importance plot,
Fig. 4). Phragmites australis prevalence increased
sharply at watershed cropland levels above 25 %, and
increased gradually with the percentage of shoreline
with agricultural land use (see partial dependence plots,
Fig. 5). Phragmites australis prevalence decreased
gradually with increasing forest cover in the watershed,
but had sharp negative threshold responses to increas-
ing tidal range, slope at the shoreline, and the
percentage of shoreline with beach (Fig. 5).
Predicting Phragmites australis presence
or absence in shoreline segments
After applying the Dunn–Sˇida´k correction to control
experiment wise significance to P\ 0.05, five of the
seven characteristics of shoreline segments yielded
significant univariate logistic models for predicting P.
australis presence or absence in 500 m shoreline
segments. The probability of P. australis presence was
positively related to the percentage of shoreline with
agricultural land use, the percentage with residential
land use, and the percentage with riprap; and nega-
tively related to the percentages of forested and
unarmored shoreline (Table 3). The models with the
percentage of riprap and the percentage of agricultural
shoreline had similar success classifying the training
data (j = 0.16 Table 3), but the AIC score of the
model based on the percentage of agricultural shore-
line was far lower (DAIC = 139), so we selected it as
the best univariate model of the set (Table 3).
Table 2 Variables included and R2 values of the four best
multiple linear regression models for predicting Phragmites
australis prevalence in subestuaries from combinations of the
predictors in Table 1
Predictor Model number
1 2 3 4
% Cropland in the watershed X X X X
% Wetland in the watershed X X
Tidal range X X X X
Boat ramp density X X X X
% High banks X X
Average slope at the shoreline X X
R2 0.51 0.51 0.54 0.55
All four models are statistically significant at P\ 0.001
Fig. 4 Variable importance plot of the top ten predictor
variables from the random forest analysis relating Phragmites
australis prevalence in subestuaries to predictor variables at the
shoreline and throughout the watershed (Table 1). The
independent variables are ranked from top to bottom in order
of importance, as measured by a the average increase in node
purity (IncNodePurity) and by b the mean squared error
(%IncMSE)
Local and regional disturbances associated with the invasion of Chesapeake Bay marshes
123
Author's personal copy
The best multiple logistic regression model for
predicting P. australis presence or absence in shore-
line segments contained the percentages of shoreline
with forest and bulkhead as negative predictors and the
percentages of shoreline with agricultural land, resi-
dential land, and riprap as positive predictors
(Table 3). When compared to the best univariate
model based on shoreline agriculture only, the mul-
tiple logistic model had a slightly lower classification
success (j = 0.17 vs. 0.18) but a lower AIC (DAIC
score = 180), so the metrics fail to conclusively rate
one model as better than the other in the training data.
However, model performance diverged when the
models were used to predict P. australis presence in
the validation data. While classification success on the
validation data was lower than success on the training
data for both models, the drop was far smaller
(Dj = 0.02) for the multivariate model than for the
univariate agriculture model (Dj = 0.07). Both mod-
els were better at predicting P. australis absence
(C83 % success) than at predicting presence (B35 %
success, Table 4).
Fig. 5 Partial dependence plots from the random forest analysis relating Phragmites australis prevalence in subestuaries to 31 possible
predictor variables. Only the six most important variables are shown
Table 3 Univariate logistic regression models relating the probability of Phragmites australis presence in shoreline segments to
seven characteristics of those shoreline segments
Predictor Intercept Slope AIC P j
% Agriculture -1.64 1.53 11,577 \0.001 0.18
% Forest -1.28 -0.70 11,715 \0.001 0.07
% Residential shoreline -1.64 0.50 11,784 \0.001 0.10
% Riprap -1.61 1.17 11,739 \0.001 0.16
% Bulkhead -1.51 0.25 11,846 0.42
% Unhardened shore -1.09 -0.55 11,779 \0.001 0.12
Average slope at shore -1.56 0.02 11,842 0.01
j values, which quantify success in classifying the training data, were calculated for significant models. The five relationships with
P\ 0.001 are significant below the Dunn–Sˇida´k threshold (P\ 0.0016) that controls experiment wise error to P\ 0.05
M. B. Sciance et al.
123
Author's personal copy
Discussion
We took an integrative, spatial modeling approach to
analyzing the regional and local factors affecting an
ongoing invasion. Our study of the Phragmites
australis invasion of Chesapeake Bay exploits the
value of P. australis as a model organism for advanc-
ing more general understanding of plant invasions
(Meyerson et al. 2016). Our analysis is unprecedented
in the spatial extent of the analysis, the breadth and
types of predictors considered, and the simultaneous
investigation of drivers that affect entire watershed-
subestuary systems as well drivers that affect individ-
ual shoreline segments. The analysis revealed strong
anthropogenic impacts on the P. australis invasion of
estuarine wetlands and distinguished factors affecting
invasion at both scales from specific factors affecting
only one.
Agriculture and shoreline armoring were signifi-
cant predictors of P. australis occurrence in subestu-
aries and in individual shoreline segments (Table 5).
Conversely, variables associated with reduced anthro-
pogenic impact, specifically forested and unmodified
shoreline, were negatively correlated with P. australis
occurrence in the Chesapeake Bay (Table 5). Some
variables were significant predictors only at the
shoreline segment scale, including: % residential
shoreline, % bulkhead, and % unarmored shoreline
(Table 5). These local-scales associations likely
reflect more direct mechanistic connections of shore-
line activities to P. australis invasion (see below).
Agriculture was the strongest and most consistent
correlate of P. australis prevalence, and the
percentage of cropland in the watershed was a
significant positive predictor of P. australis preva-
lence in a subestuary in every test where it was used
(Table 5). Phragmites australis commonly grows at
the edges of agricultural fields throughout the world
(Haslam 2010). Agricultural activities increase nitro-
gen levels in receiving waters (Jordan et al. 1997a, b;
Liu et al. 2000; Savage et al. 2010). Total nitrogen
concentrations were measured in 20 of our 72
subestuaries (Fig. 1) that had low percentages of
developed land in their local watersheds, and estuarine
nitrogen concentration increased strongly with the
percentage of cropland in the local watershed
(R2 = 0.63, P\ 0.001, Thomas E. Jordan et al.
unpublished data).
Elevated nitrogen levels promote P. australis in
several ways. Elevated nitrogen increases P. australis
sexual reproduction and expansion in Chesapeake Bay
(Kettenring et al. 2011); increases P. australis density,
height, and above-ground shoot biomass (Bastlova
et al. 2004; Engloner 2009); and allows seedlings to
rapidly escape from a vulnerable life stage (Saltonstall
and Stevenson 2007; Kettenring et al. 2015; Hazelton
et al. 2014). P. australis stands have larger nitrogen
pools than other marsh communities (Meyerson et al.
1999, 2000; Windham and Meyerson 2003). Foliar
nitrogen was higher in P. australis from Chesapeake
Bay subestuaries with more agriculture, even if total
human land use was relatively low, suggesting that
even a relatively small amount of agriculture may
raise nitrogen levels enough to benefit P. australis
(King et al. 2007). Watersheds dominated by forests
export less nitrogen than watersheds with agriculture
Table 4 Classification success for two models predicting
Phragmites australis presence in shoreline segments: the best
univariate logistic model (based on shoreline agriculture only)
and the multivariate logistic model (based on five variables, the
percentages of the shoreline with agriculture, forest, or
residential land and the percentages of the shoreline with
bulkhead and riprap)
Data j Total proportion
correct
Proportion of predicted
presence that is correct
Proportion of predicted
absence that is correct
Univariate model
Training 0.18 0.77 0.35 0.84
Validation 0.11 0.75 0.28 0.83
Multiple logistic model
Training 0.17 0.74 0.31 0.85
Validation 0.15 0.68 0.28 0.85
Half of the data were used to fit the models (training data), and the other half was reserved as validation data to provide independent
model testing
Local and regional disturbances associated with the invasion of Chesapeake Bay marshes
123
Author's personal copy
(Dillon and Kirchner 1975; Jordan et al. 1997a, b; Liu
et al. 2000), and this is consistent with negative
relationship of watershed forest with P. australis
prevalence (Table 5).
Agriculture on the shoreline increased the proba-
bility of P. australis invasion in individual shoreline
segments (Tables 3, 5). Previous studies also reported
a correlation between P. australis invasion and
agriculture at the shoreline (King et al. 2007; Cham-
bers et al. 2008). Shoreline agriculture can deliver
nitrogen directly to adjacent waters. However, natural
riparian zones not only release less nitrogen directly to
adjacent waters, they can absorb nitrogen moving
downhill and reduce nitrogen inputs to the water from
disturbed uphill systems (Sweeney and Newbold
2014; Weller and Baker 2014). Natural riparian
vegetation may also be more difficult to invade than
other shoreline land uses with reduced vegetative
cover and disturbed soils (Bertness et al. 2002;
DeSimone et al. 2010; Finnegan et al. 2012).
Surprisingly, P. australis prevalence in subestuaries
was not significantly related to developed land in the
watershed or to shoreline development (Tables 1, 2;
Fig. 4). We expected these factors to increase P.
australis prevalence because developed lands can
release nitrogen and can contain disturbed sites that
are open for invasion (see ‘‘Introduction’’ section). An
earlier study did report a positive association between
watershed development and P. australis abundance
(King et al. 2007); however, that study only considered
existing wetlands, while this study accounted for the
entire shoreline, including segments lacking wetlands.
Other studies have reported that residential develop-
ment in any part of a subestuary is associated with
higher foliar nitrogen concentrations and abundance of
P. australis for the entire subestuary (King et al. 2007;
McCormick et al. 2010a). Our study did find a
significant relationship between P. australis presence
in 500 m shoreline segments and adjacent shoreline
development (Table 3). Among subestuaries, therewas
a tendency for higherP. australis prevalence to occur in
subestuaries with intermediate levels of residential
shoreline but low levels of developed land in their local
watersheds (Fig. 6). Subestuaries with highly devel-
oped watersheds and mostly armored shoreline may be
so impacted that little tidal wetland remains.
Table 5 Summary of 14 statistically significant predictors of Phragmites australis across five statistical models predicting P.
autrtalis prevalence in subestuaries or P. australis presence in 500 m shoreline segments
Predictor Prevalence in subestuaries Presence in shoreline segments
Univariate regression Multiple regression Random forest Univariate logistic Multiple logistic
Shoreline characteristics
% Agriculture ? ? ? ?
% Forest – –
% Residential ? ?
% Riprap ? ? ?
% Bulkhead –
% Unhardened –
Boat ramp density ? NA NA
Average slope – –
% High banks – NA NA
% Beach – NA NA
Watershed or estuary characteristics
% Cropland ? ? ? NA NA
% Forest – – NA NA
% Wetland ? NA NA
Tidal range – – NA NA
NA indicates that a variable was not included in a particular analysis, while ‘‘?’’ or ‘‘–’’ mark variables that had significant positive
or negative relationships with P. australis
M. B. Sciance et al.
123
Author's personal copy
The presence of riprap was also positively corre-
lated with P. australis prevalence in subestuaries and
with its presence in individual shoreline segments.
These positive associations likely arise because the
construction of riprap creates disturbed sites for
colonization, and wave action can continue to disturb
vegetation and substrate at the end of the protected
shoreline. Phragmites australis stands often grow at
the edges of structures such as riprap and bulkhead
(Melissa McCormick, unpublished data), and distur-
bances as small as 30 cm in diameter can promote P.
australis invasion (Kettenring et al. 2015).
Tidal range and average slope at the shoreline were
negatively related to P. australis prevalence in
subestuaries (Table 5), but these relationships were
driven by three subestuaries (Chester River, Swan
Creek, and Tavern Creek) that are very heavily
invaded (60–80 % of shoreline occupied) and also
have low values of tidal range and average slope at the
shoreline. It is possible that high tidal amplitude could
push propagules above the flood zone during lunar and
storm tides, contributing to establishment. However,
we refrain from interpreting these relationships
because further study is needed to determine if they
are robust or are artifacts of this particular data set.
Our study has identified several important corre-
lates of P. australis occurrence in estuarine wetlands.
By considering a large population of subestuary-
watershed systems, testing for both regional and local
associations, and applying different statistical models;
we demonstrated the correlates of P. australis invasion
more clearly than previous studies (King et al. 2007;
Chambers et al. 2008). Our analyses complement site-
specific field studies and support the importance of
factors identified in those studies (reviewed in Hazel-
ton et al. 2014). Thus, our study enhances understand-
ing of how nitrogen and disturbance drive the P.
australis invasion of Chesapeake Bay and is also
relevant to understanding the broader importance of
those factors in plant invasions in general. Despite
these contributions, the predictive power of our
models was somewhat low. We explained at most
55 % of the variation in P. australis prevalence among
subestuaries (Tables 1, 2), and the best models for P.
australis presence in 500 m shoreline segments were
far better at predicting absence than at predicting
presence (Table 4). Our models for the 500 m shore-
line segments may have been limited by quantifying P.
australis as a binary response (present or absent in
each segment), but we were reluctant to interpret the
P. australis response more finely because the data
were collected over a ten-year interval. More impor-
tantly, poor predictive ability reflects a key gap in the
information available for our models. Plant invasions
typically depend on three factors: nutrient amendment,
altered disturbance regime, and propagule pressure
(Colautti et al. 2006). The independent variables that
we considered capture the first two factors well, but
supply no information on propagule availability or on
spatial variation in availability across the Chesapeake
Bay. We do know that P. australis seeds increased as
the invasion in Chesapeake Bay progressed because
Fig. 6 Relationships of Phragmites australis prevalence in
subestuaries to two measures of residential development. a The
percentage of developed land in the local watershed, b the
percentage of subestuary shoreline with residential
development. Subestuaries with higher P. australis prevalence
([26 % of shoreline occupied, highlighted as filled symbols)
occur at low levels of watershed development (\20 % a) and
intermediate levels residential shoreline (19–41 % b)
Local and regional disturbances associated with the invasion of Chesapeake Bay marshes
123
Author's personal copy
greater genetic diversity promotes seed production in
P. australis (McCormick et al. 2010b). Further genetic
studies may provide the missing information on the
spatial distribution of propagules that is needed to
build more powerful predictive models.
Conclusion
Our results add to the growing body of evidence that P.
australis ssp. australis growth and proliferation are
aided and accelerated by system-wide anthropogenic
disturbances and by localized disturbances directly on
the shoreline. Our findings strengthen the established
relationship between P. australis and agriculture by
demonstrating that agriculture is a robust positive
predictor at regional and local scales. Riprapped
shorelines are also associated with P. australis inva-
sion, likely because of the disturbances associated with
installation and erosion at the edges. We suggest that
systems with low watershed development, some agri-
cultural land, and intermediate levels of development
and disturbance at the shoreline are most at risk of
invasion because there is enough anthropogenic
eutrophication and disturbance to help P. australis
ssp. australis invade, but not so much shoreline
alteration that available habitat is lacking. Phragmites
australis ssp. australis is well established in North
America, but large swathes of habitat are still unin-
vaded. For example, the polyhaline and tidal fresh parts
of Chesapeake Bay have low levels of invasion, and
there are some systemswithin themiddle salinity zones
that are relatively free ofP. australis.While eradication
in heavily invaded areas is probably not possible (see
review in Hazelton et al. 2014), management efforts
focused on preventing the continued spread of this
invasive plant into areas with low levels of invasion
might be successful.Management strategies that reduce
nutrient pollution, preserve natural shorelines, and limit
nearshore disturbance of soils and vegetation may
enhance the efforts to increase the resilience of
shorelines to new invasion. These efforts are not likely
to succeed unless they consider both regional and local
factors and can address disturbance, nitrogen, and
propagules together (Hazelton et al. 2014; Kettenring
et al. 2011; 2015; McCormick et al. 2016).
Acknowledgments We thank the Virginia Institute of Marine
Sciences, the Maryland Department of Natural Resources, and
the Chesapeake Bay Program for providing data used in our
analysis. This work was supported by award number
NA09NOS4780214 from the National Oceanic and
Atmospheric Administration (NOAA) Center for Sponsored
Coastal Ocean Research (CSCOR).
References
Anderson DM, Glibert PM, Burkholder JM (2002) Harmful
algal blooms and eutrophication: nutrient sources, com-
position, and consequences. Estuaries 25:704–726
Bastlova D, Cizkova H, Bastl M, Kvet J (2004) Growth of
Lythrum salicaria and Phragmites australis plants origi-
nating from a wide geographical area: response to nutrient
and water supply. Glob Ecol Biogeogr 13:259–271
BertnessMD, Ewanchuk PJ, Silliman BR (2002) Anthropogenic
modification of New England salt marsh landscapes. Proc
Natl Acad Sci 99:1395–1398
Biau G (2012) Analysis of a random forests model. J Mach
Learn Res 13:1063–1095
Boesch DF, Brinsfield RB,Magnien RE (2001) Chesapeake Bay
eutrophication: scientific understanding, ecosystem
restoration, and challenges for agriculture. J Environ Qual
30:303–320
Breiman L, Cutler A (2014) Package ‘randomForest’ http://
cran.r-project.org/web/packages/randomForest/random
Forest.pdf
Burdick DM, Konisky RA (2003) Determinants of expansion for
Phragmites australis, common reed, in natural and
impacted coastal marshes. Estuaries 26:407–416
Burnham KP, Anderson DR (2002) Model selection and mul-
timodel inference, 2nd edn. Springer, New York
CBP (Chesapeake Bay Program) (2015) The Chesapeake Bay
watershed. http://www.chesapeakebay.net/discover/
baywatershed. Accessed 3 Feb 2016
Chambers RM,Meyerson LA, Saltonstal K (1999) Expansion of
Phragmites australis into tidal wetlands of North America.
Aquat Bot 64:261–273
Chambers RM, Havens KJ, Killeen S, Berman M (2008)
Common reed Phragmites australis occurrence and adja-
cent land use along estuarine shoreline in Chesapeake Bay.
Wetlands 28:1097–1103
Cohen J (1960) A coefficient of agreement for nominal scales.
Educ Psychol Meas 20:37–46
Colautti RI, Grigorovich IA, MacIsaac HJ (2006) Propagule
pressure: a null model for biological invasions. Biol
Invasions 8:1023–1037
Comeleo RL, Paul JF, August PV, Copeland J, Baker C, Hale
SS, Latimer RW (1996) Relationships between watershed
stressors and sediment contamination in Chesapeake Bay
estuaries. Land Ecol 11:307–319
Conley DJ, Paerl HW, Howarth RW, Boesch DF, Seitzinger SP,
Havens KE, Lancelot C, Likens GE (2009) Controlling
eutrophication: nitrogen and phosphorus. Science
323:1014–1015
Cutler DR, Edwards TC Jr, Beard KH, Cutler A, Hess KT,
Gibson J, Lawler JJ (2007) Random forests for classifica-
tion in ecology. Ecology 88:2783–2792
M. B. Sciance et al.
123
Author's personal copy
De’Ath G, Fabricius KE (2000) Classification and regression
trees: a powerful yet simple technique for ecological data
analysis. Ecology 8:3178–3192
DeSimone J, Macrae ML, Bourbonniere RA (2010) Spatial
variability in surface N2O fluxes across a riparian zone and
relationships with soil environmental conditions and
nutrient supply. Agric Ecosyst Environ 138:1–9
Dibble KL, Penelope SP, Meyerson LA (2013) Impacts of plant
invasions can be reversed through restoration: a regional
meta-analysis of faunal communities. Biol Invasions
15:1725–1737
Dillon PJ, Kirchner WB (1975) The effects of geology and land
use on the export of phosphorus from watersheds. Water
Res 9:135–148
Ellison GN, Gotelli NJ (2004) A primer of ecological statistics.
Sinauer, Sunderland
Engloner AI (2009) Structure, growth dynamics and biomass of
reed (Phragmites australis)—a review. Flora Morphol
Distrib Funct Ecol Plants 204(5):331–346
ESRI (Environmental Systems Research Institute) (2011) Arc-
GIS desktop release 10. Environmental Systems Research
Institute, Redlands
Finnegan J, Regan JT, de Eyto E, Ryder E, Tiernan D, Healy
MG (2012) Nutrient dynamics in a peatland forest riparian
buffer zone and implications for the establishment of
planted saplings. Ecol Eng 47:155–164
Gornitz V, White TW (1992) A coastal hazards database for the
USWest Coast: ORNL CDIAC-81 NDP-043C. Oak Ridge
National Laboratory, Oak Ridge, TN
GuoW-Y, Lambertini C, Li X-Z, Meyerson LA (2013) Invasion
of Old World Phragmites australis in the New World:
precipitation and temperature patterns combined with
human influences redesign the invasive niche. Glob
Change Biol 19:3406–3422
Haslam SM (2010) A book of reed:(Phragmites australis (Cav.)
Trin. ex Steudel, Phragmites communis Trin.). Forrest
Text, Cardigan
Hazelton EL, Mozdzer TJ, Burdick DM, Kettenring KM,
Whigham DF (2014) Phragmites australis management in
the United States: 40 years of methods and outcomes. AoB
Plants 6:plu001
Homer C, Huang C, Yang L, Wylie B, Coan M (2004) Devel-
opment of a 2001 national land-cover database for the
United States. Photogramm Eng Remote Sens 70:829–840
Hosmer DW, Lemeshow S (2000) Applied logistic regression.
Wiley, New York
Hosmer DW, Jovanovic B, Lemeshow S (1989) Best sub-sets
logistic regression. Biometrics 45:1265–1270
Hubertz JM, Thompson EF, Wang HV (1996) Wave informa-
tion studies of U.S. coastlines: annotated bibliography on
coastal and ocean data assimilation. WIS Report 36, U.S.
Army Engineer Waterways Experiment Station, Vicks-
burg, MS
Jordan TE, Correll DL,Weller DE (1997a) Effects of agriculture
on discharges of nutrients from coastal plain watersheds of
Chesapeake Bay. J Environ Qual 26:836–848
Jordan TE, Correll DL, Weller DE (1997b) Nonpoint source
discharges of nutrients from Piedmont watersheds of
Chesapeake Bay. J Am Water Res Assoc 33:631–645
KempW, BoyntonW, Adolf J, Boesch D, Boicourt W, Brush G,
Stevenson J (2005) Eutrophication of Chesapeake Bay:
historical trends and ecological interactions. Mar Ecol Prog
Ser 303:1–29
Kettenring KM, McCormick MK, Baron HM, Whigham DF
(2011) Mechanisms of Phragmites australis invasion:
feedbacks among genetic diversity, nutrients, and sexual
reproduction. J Appl Ecol 48:1305–1313
Kettenring KM, Whigham DF, Hazelton ELG, Gallagher SK,
Baron HM (2015) Biotic resistance, disturbance, and mode
of colonization impact the invasion of a widespread,
introduced, wetland grass. Ecol Appl 25:466–480
King RS, Beaman JR, Whigham DF, Hines AH, Baker ME,
Weller DE (2004) Watershed land use is strongly linked to
PCBs in white perch in Chesapeake Bay subestuaries.
Environ Sci Technol 38:6546–6552
King RS, Deluca WV, Whigham DF, Marra PP (2007)
Threshold effects of coastal urbanization on Phragmites
australis (common reed) abundance and foliar nitrogen in
Chesapeake Bay. Estuaries Coasts 30:469–481
Li X, Weller DE, Gallegos CL, Jordan TE, Kim HC (2007)
Effects of watershed and estuarine characteristics on the
abundance of submerged aquatic vegetation in Chesapeake
Bay subestuaries. Estuaries Coasts 30:840–854
Liu ZJ,Weller DE, Correll DL, Jordan TE (2000) Effects of land
cover and geology on stream chemistry in watersheds of
Chesapeake Bay. J Am Water Res Assn 36:1349–1365
Long WC, Grow JN, Majoris JE, Hines AH (2011) Effects of
anthropogenic shoreline hardening and invasion by
Phragmites australis on habitat quality for juvenile blue
crabs (Callinectes sapidus). J Exp Mar Biol Ecol
409:215–222
Manel S, Williams HC, Ormerod SJ (2001) Evaluating pres-
ence–absence models in ecology: the need to account for
prevalence. J Appl Ecol 38:921–931
Mazur MLC, Kowalski KP, Galbraith D (2014) Assessment of
suitable habitat for Phragmites australis (common reed) in
the Great Lakes coastal zone. Aquat Invasions 9:1–19
McCormick MK, Kettenring KM, Baron HM, Whigham DF
(2010a) Extent and reproductive mechanisms of Phrag-
mites australis spread in brackish wetlands in Chesapeake
Bay, Maryland (USA). Wetlands 30:67–74
McCormick MK, Kettenring KM, Baron HM, Whigham DF
(2010b) Spread of invasive Phragmites australis in estu-
aries with differing degrees of development: genetic pat-
terns, Allee effects and interpretation. J Ecol
98:1369–1378
McCormick MK, Brooks HEA, Whigham DF (2016)
Microsatellite analysis to estimate realized dispersal dis-
tance in Phragmites australis. Biol Invasions. doi:10.1007/
s10530-016-1126-1
Meadows RE, Saltonstall K (2007) Distribution of native and
introduced Phragmites australis in freshwater and oligo-
haline tidal marshes of the Delmarva peninsula and
southern New Jersey. J Torrey Bot Soc 134:99–107
Meyerson LA, Chambers RM, Vogt KA (1999) The effects of
Phragmites removal on nutrient pools in a freshwater tidal
marsh ecosystem. Biol Invasions 1:129–136
Meyerson LA, Cronin JT, Pysek P (2016) Phragmites australis
as a model organism for studying plant invasions. Biol
Invasions. doi:10.1007/s10530-016-1132-3
Meyerson LA, Saltonstall K, Windham L, Kiviat E, Findlay S
(2000) A comparison of Phragmites australis in freshwater
Local and regional disturbances associated with the invasion of Chesapeake Bay marshes
123
Author's personal copy
and brackish marsh environments in North America. Wetl
Ecol Manag 8:89–103
Meyerson LA, Saltonstall K, Chambers RM, Silliman BR,
Bertness MD, Strong D (2009) Phragmites australis in
eastern North America: a historical and ecological per-
spective. In: Silliman BR, BertnessMD, Grosholz ED (eds)
Human impacts on salt marshes: a global perspective.
University of California Press, Berkeley, pp 57–82
Meyerson LA, Lambert AM, Saltonstall K (2010) A tale of three
lineages: expansion of common reed (Phragmites aus-
tralis) in the US Southwest and Gulf Coast. Invasive Plant
Sci Manag 8:515–520
Moles AT, Flores-Moreno H, Bonser SP, Warton DI, Helm A
et al (2012) Invasions: the trail behind, the path ahead, and
a test of a disturbing idea. J Ecol 100:116–127
Mozdzer TJ, Megonigal JP (2012) Jack-and-Master trait
responses to elevated CO2 and N: a comparison of native
and introduced Phragmites australis. Public Libr Sci One
7:e42794
Mozdzer TJ, Zieman JC (2010) Ecophysiological differences
between genetic lineages facilitate the invasion of non-
native Phragmites australis in North American Atlantic
coast wetlands. J Ecol 98:451–458
Mozdzer TJ, Brisson J, Hazelton ELG (2013) Physiological
ecology and functional traits of North American native and
Eurasian introduced Phragmites australis lineages. AoB
Plants 5:plt048
Nixon S (1995) Coastal marine eutrophication: a definition,
social causes, and future concerns. Ophelia 41:199–219
NOAA (National Oceanographic and Atmospheric Adminis-
tration) (1998) Chesapeake Bay, VA/MD (M130) Bathy-
metric ditigal elevation model (30 meter resolution)
derived from source hydrogrpahic survey soundings
collected by NOAA. https://data.noaa.gov/dataset/
chesapeake-bay-va-md-m130-bathymetric-digital-elevation
model-30-meter-resolution-derived-from-s. Accessed 3
Feb 2016
Osgood DT, Yozzo DJ, Chambers RM, Jacobson D, Hoffman T,
Wnek J (2003) Tidal hydrology and habitat utilization by
resident nekton in Pragmites and non-Phragmitesmarshes.
Estuaries 26:522–533
Patrick CJ, Weller DE (2015) Interannual variation in sub-
merged aquatic vegetation and its relationship to water
quality in subestuaries of Chesapeake Bay. Mar Ecol Prog
Ser 537:121–135
Patrick CJ, Weller DE, Li X, Ryder M (2014) Effects of
shoreline alteration and other stressors on submerged
aquatic vegetation in subestuaries of Chesapeake Bay and
mid-Atlantic coastal bays. Estuaries Coasts 37:1516–1531
Patrick CJ, Weller DE, Ryder M (2016) The relationship
between shoreline armoring and adjacent submerged
aquatic vegetation in Chesapeake Bay and nearby Atlantic
Coastal Bays. Estuaries Coasts 39:158–170
Paul JF, Comeleo RL, Copeland J (2002) Landscape metrics and
estuarine sediment contamination in the mid-Atlantic and
southern New England regions. J Environ Qual
31:836–845
Pysek P, Jarosık V, Hulme PE, Kuhn I, Wild J et al (2010)
Disentangling the role of environmental and human pres-
sures on biological invasions across Europe. Proc Natl
Acad Sci 107:12157–12162
R Core Team (2014) R: a language and environment for sta-
tistical computing R Foundation for Statistical Computing,
Vienna, Austria. http://www.R-project.org/. Accessed 16
Feb 2016
Robertson TL, Weis JS (2005) A comparison of epifaunal
communities associated with the stems of salt marsh
grasses Phragmites australis and Spartina alterniflora.
Wetlands 25:1–7
Rodriguez W, August PV, Wang YQ, Paul JF, Gold A, Rubin-
stein N (2007) Empirical relationships between land use/-
cover and estuarine condition in the Northeastern United
States. Land Ecol 22:403–417
Saltonstall K (2002) Cryptic invasion by a non-native genotype
of the common reed, Phragmites australis, into North
America. Proc Natl Acad Sci 99:2445–2449
Saltonstall K, Stevenson JC (2007) The effect of nutrients on
seedling growth of native and introduced Phragmites
australis. Aquat Bot 86:331–336
Savage C, Leavitt PR, Elmgren R (2010) Effects of land use,
urbanization, and climate variability on coastal eutrophi-
cation in the Baltic Sea. Limnol Oceanogr 55:1033–1046
Shuster WD, Bonta J, Thurston H, Warnemuende E, Smith DR
(2005) Impacts of impervious surface on watershed
hydrology: a review. Urban Water J 2:263–275
Silliman BR, Bertness MD (2004) Shoreline development
drives invasion of Phragmites australis and the loss of
plant diversity on New England salt marshes. Conserv Biol
18:1424–1434
Smith VH (2006) Responses of estuarine and coastal marine
phytoplankton to nitrogen and phosphorus enrichment.
Limnol Oceanogr 51:377–384
Smith VH, Schindler DW (2009) Eutrophication science:Where
do we go from here? Trends Ecol Evol 24:201–207
Strayer DL, Findlay SEG, Miller D, Malcom HM, Fischer DT,
Coote T (2012) Biodiversity in Hudson River shore zones:
influence of shoreline type and physical structure. Aquat
Sci 74:597–610
Sweeney BW, Newbold JD (2014) Streamside forest buffer
width needed to protect stream water quality, habitat, and
organisms: a literature review. J Am Water Res Assoc
50:560–584
Thieler ER, Hammar-Klose ES (1999) National assessment of
coastal vulnerability to sea-level rise: preliminary results
for the U.S. Atlantic coast. U.S. Geological Survey, Open-
File Report 99-593
Tulbure MG, Ghioca-Robrecht DM, Johnston CA, Whigham
DF (2012) Inventory and ventilation efficiency of nonna-
tive and native Phragmites australis (common reed) in
tidal wetlands of the Chesapeake Bay. Estuaries Coasts
35:1353–1359
Tumley T (2009) Package ‘leaps’ Version 2.9. http://cran.r-
project.org/web/packages/leaps/leaps.pdf. Accessed 3 Feb
2016
USEPA (U. S. Environmental Protection Agency) (2010) Che-
sapeake Bay total maximum daily load for nitrogen,
phosphorus and sediment. http://www.epa.gov/sites/
production/files/2014-12/documents/cbay_final_tmdl_
exec_sum_section_1_through_3_final_0.pdf. Accessed 3
Feb 2016
VIMS-CCRM (Virginia Institute of Marine Science Center for
Coastal Resources Management) (2009) GIS data and
M. B. Sciance et al.
123
Author's personal copy
maps: shoreline inventories. http://ccrm.vims.edu/gis_
data_maps/shoreline_inventories/index.html. Accessed 3
Feb 2016
Weller DE, Baker ME (2014) Cropland riparian buffers
throughout Chesapeake Bay watershed: spatial patterns
and effects on nitrate loads delivered to streams. J Am
Water Res Assoc 50:696–715
Windham L, Meyerson LA (2003) Effects of common reed
(Phragmites australis) expansions on nitrogen dynamics of
tidalmarshes of the northeasternU. S. Estuaries 26:452–464
Local and regional disturbances associated with the invasion of Chesapeake Bay marshes
123
Author's personal copy