Content uploaded by Sunil Kumar
Author content
All content in this area was uploaded by Sunil Kumar
Content may be subject to copyright.
Content uploaded by Curtis H Flather
Author content
All content in this area was uploaded by Curtis H Flather
Content may be subject to copyright.
RESEARCH ARTICLE
A habitat overlap analysis derived from Maxent for Tamarisk
and the South-western Willow Flycatcher
Patricia YORK (✉)
1
, Paul EVANGELISTA
1
, Sunil KUMAR
1
, James GRAHAM
1
, Curtis FLATHER
2
,
Thomas STOHLGREN
3
1 Natural Resource Ecology Lab, Colorado State University, 1499 Campus Delivery, Fort Collins, CO 80523, USA
2 Department of Agriculture, Forest Service, Rocky Mountain Research Station, Fort Collins, CO 80526, USA
3 Geological Survey, Fort Collins Science Center, Fort Collins, CO 80526, USA
© Higher Education Press and Springer-Verlag Berlin Heidelberg 2011
Abstract Biologic control of the introduced and inva-
sive, woody plant tamarisk (Tamarix spp, saltcedar) in
south-western states is controversial because it affects
habitat of the federally endangered South-western Willow
Flycatcher (Empidonax traillii extimus). These songbirds
sometimes nest in tamarisk where floodplain-level inva-
sion replaces native habitats. Biologic control, with the
saltcedar leaf beetle (Diorhabda elongate), began along the
Virgin River, Utah, in 2006, enhancing the need for
comprehensive understanding of the tamarisk-flycatcher
relationship. We used maximum entropy (Maxent) model-
ing to separately quantify the current extent of dense
tamarisk habitat ( > 50% cover) and the potential extent of
habitat available for E. traillii extimus within the studied
watersheds. We used transformations of 2008 Landsat
Thematic Mapper images and a digital elevation model as
environmental input variables. Maxent models performed
well for the flycatcher and tamarisk with Area Under the
ROC Curve (AUC) values of 0.960 and 0.982, respec-
tively. Classification of thresholds and comparison of the
two Maxent outputs indicated moderate spatial overlap
between predicted suitable habitat for E. traillii extimus
and predicted locations with dense tamarisk stands, where
flycatcher habitat will potentially change flycatcher
habitats. Dense tamarisk habitat comprised 500 km
2
within
the study area, of which 11.4% was also modeled as
potential habitat for E. traillii extimus. Potential habitat
modeled for the flycatcher constituted 190 km
2
, of which
30.7% also contained dense tamarisk habitat. Results
showed that both native vegetation and dense tamarisk
habitats exist in the study area and that most tamarisk
infestations do not contain characteristics that satisfy the
habitat requirements of E. traillii extimus. Based on this
study, effective biologic control of Tamarix spp. may, in
the short term, reduce suitable habitat available to E. traillii
extimus, but also has the potential in the long term to
increase suitable habitat if appropriate mixes of native
woody vegetation replace tamarisk in biocontrol areas.
Keywords Niche modeling, species interactions, Tamar-
isk, South-western Willow Flycatcher, habitat overlap
analysis
1 Introduction
Humans have dramatically altered the global distribution
of species over the past few centuries (Chapin III et al.,
2000). This movement of species coupled with the
disturbance of native habitats has facilitated the invasion
of exotic plants and animals around the world, threatening
the survival of many native species (Vitousek et al., 1997).
Although exotic and native species coexist in many
modern habitats, conservation efforts typically focus on
single-species management of either the introduced or the
threatened species. Chemical, mechanical, and biologic
control efforts geared at eradicating exotic species can
have negative effects on native populations, especially
when control efforts alter the critical habitat of sensitive
species of endangered, threatened, or endemic status (Innes
and Barker, 1999; Cory and Myers, 2000; Matarczyk et al.,
2002). New strategies are needed that simultaneously
consider conservation efforts for both introduced and
threatened species existing within the same landscape.
Here we focused on riparian corridors which often
represent only 1%–3% of the landscape in the arid south-
western United States, but are vital to biodiversity (Naiman
et al., 1993; Naiman and Décamps, 1997; Patten, 1998).
Riparian areas are highly susceptible to invasion by
Received February 14, 2011; accepted March 8, 2011
E-mail: pmyork0714@gmail.com
Front. Earth Sci. 2011, 5(2): 120–129
DOI 10.1007/s11707-011-0154-5
introduced species due to their relatively abunda nt
resources (Stohlgren et al., 1998). In the south-western
United States, alteration of the natural flow regime through
damming and withdrawals for agricultural, industrial, and
household consumption has dramatically altered riparian
habitats. Lowered water tables and reduced peak flows
associated with damming and diversions hinder successful
propagation of native cottonwood (Populus spp.) and
willow (Salix spp.) species, reducing the abundance of
mature native riparian forests (Lite and Stromberg, 2005).
Members and hybrids of the exotic genus Tamarix
(Tamarix parviflora, T. ramosissima, T. chinensis, etc.)
are adapted to the altered flow regime given their ability to
extract deep water through an extensive tap root (Everitt,
1980). Tamarisk (collectively, and also referred to as
saltceder) currently occurs within most large river systems
of the south-western United States and is estimated to have
replaced 470000–650000 ha of native riparian habitat
(Robinson 1965; Di Tomaso, 1998; Zavale ta, 2000).
Tamarisk’s success further suppresses the ability of
cottonwood and willows to reproduce (Lytle and Merritt,
2004; Stromberg et al., 2007; Merritt and Poff, 2010).
The altered ecosystems represent degraded habitat for
many native species, and restoration has become a high-
priority goal for many natural resource managers (Szaro
and Rinne, 1988). Millions of dollars have been spent by
United States government agencies at the local, state, and
federal level in efforts to remove tamarisk and restore
native habitats (S hafroth and Briggs, 2008). Control
techniques have included burning, herbicide treatments,
mechanical removal and most recently, biologic control by
the saltcedar leaf beetle (Diorhabda elongata) (Taylor and
McDaniel, 1998; Deloach et al., 2004; Bateman et al.,
2010). The saltcedar leaf beetle defoliates tamarisk plants
with repeated defoliation events leading to gradual die
back. Controversies surrounding biocontrol methods arose
due to the possible repercussions for suitable habitat of the
endangered South-western Willow Flycatcher (Empidonax
traillii extimus) (Dudley and Deloach, 2004; Sogge et al.,
2008).
The South-western Willow Flycatcher (hereafter, fly-
catcher) is one of four recognized subspecies of willow
flycatchers existing throughout the United States and
southern Canada. These birds winter in Central and South
America and occupy breeding territories in riparian areas
of North America for four to five months of the year. The
south-western subspecies occupies breeding sites in
Arizona, western New Mexico, and southern portions of
California, Nevada, Utah, and Colorado (Paxton et al.,
2007). Extensive surveys conducted since 1993 have
produced a current estimate of just over 1200 flycatcher
territories located at 284 breeding sites throughout the
bird’s range (Durst et al., 2007). A territory is defined as
the specific nesting location of one breeding pair, and a
breeding site is defined as a collective area containing one
or more flycatcher territories. Most breeding sites contain
five or less territories, and only a few sites contain more
than 50 territories (Durst et al., 2007).
Highly suitable flycatcher breeding habitat tends toward
heterogeneous mixes of riparian vegetation, both in terms
of plant age and species composition. Nesting sites
specifically tend toward riparian edge habitat (Paxton
et al., 2007). The United States Fish and Wildlife Service
listed the flycatcher as endangered in 1995 (United States
Fish and Wildlife Service, 1995). The alteration of riparian
habitat has been linked as the major factor in the
subspecies’ decline (Unitt, 1987). Durst et al. (2007)
estimated that throughout its range, approximately 27% of
flycatcher breeding sites were located in areas dominated
( > 50% cover) by tamarisk. Tamarisk often forms dense
homogenous thickets and may not provide the optimal
heterogeneous mix necessary for flycatcher
breeding.
Quantitative models are valuable tools when assessing
habitat suitability for species at landscape scales. Ground
surveys can be labor-intensive and expensive, and models
enable researchers to focus these efforts by determining
areas of investigative importance (Morisette et al., 2006).
Maxent (maximum entropy modeling; Phillips et al., 2006)
uses presence-only data to predict the distribution of a
species over the modeled landscape by correlating
occurrences (presence) with patterns in the included
environmental variables (Phillips et al., 2006). Comparison
of Maxent results for terrestrial and aquatic environments
indicted that Maxent performs as well with presence-only
data as other spatial distribution models, which often
require both presence and absence data (Elith et al., 2006;
Kumar et al., 2009; Elith and Graham, 2009). Maxent also
appears to fit general relationships even under situations
where sample sizes are small (Kumar and Stohlgren,
2009).
Both tamarisk (Evangelista et al., 2009) and the
flycatcher (Hatten and Paradzick, 2003; Paxton et al.,
2007; Hatten et al., 2010) have been modeled individually
at the landscape level. However, we could not find
examples of where they were comparatively modeled for
the same landscape. In this study we demonstrate how
single-species predictive modeling t echniques can be
combined to provide information for multi-species man-
agement purposes. Our goals were to quantify the amount
of habitat dominated by tamarisk and available to the
flycatcher within the study area, and to further the
unde rstanding of the factors contributi ng to the co-
occurrence of this endangered bird and invasive plant
species in the study area.
2 Methods
2.1 Study area
We conducted this study within the Great Basin Region.
The area includes the five United States Geological Survey
Patricia YORK et al. Tamarisk-Flycatcher Maxent habitat comparison 121
watershed cataloguing units flowing into the Overton
(northern) arm of Lake Mead (Fig. 1). We downloaded
Shapefiles of these watersheds from the United States
Geological Survey National Hydrology Data set Geodata-
base website
1)
. The total area studied equates to
34180 km
2
. The landscape is characterized by a mix of
Mojave Desert to the south and Great Basin Desert to the
north and represents the northern limits of the flycatcher’s
range. (Unitt 1987). Tamarisk is present in many riparian
corridors of these watersheds, and dense tamarisk habitat
( > 50% cover) is common along eastern drainages.
Biologic control of tamarisk with the saltcedar leaf beetle
(D. elongata) began along the Virgin River near St.
George, Utah, in 2006. The area impacted by the beetles
and intensity of defoliation events have increased in each
subsequent year (Hultine et al., 2009; Bateman et al.,
2010).
The study area encompasses 34180 km
2
of landscape.
The 11 flycatcher presence points represent nesting sites
and may contain more than one breeding pair. The 48
tamarisk points designate areas with dominant tamarisk
invasion ( > 50% cover).
2.2 Data sets
We used six Landsat Thematic Mapper (TM) scenes to
represent an eight month growing season (Path 40, Row
33; Path 39, Rows 33–35; and Path 38, Rows 34 and 35;
March through November, excluding July, 2008). We
downloaded the images at 30-m resolution from the United
States Geological Survey Earth Explorer website
2)
.We
collected images for each month dated within ten days of
each other when possible, although April, June, and
November data acquisition dates were further apart in
order to acquire cloud-free images. We created mosaics of
individual bands 1–5 and 7 and clipped each to the project
extent (Leica ERDAS Imagine 9.1, 2009). We used the
mosaics to create environmental model input variables for
both the tamarisk and flycatcher individual models.
We downloaded 175 tamarisk presence points from the
National Institute for Invasive Species Science website
collected between 2000 and 2004
3)
. We used these points
to estimate an initial tamarisk model. In November, 2009,
we traveled to the study area for initial model field
verification. We randomly selected 81 points within three
prediction categories (high, medium, and low) derived
from the initial tamarisk model. Of the 81 points, 48
contained tamarisk with a cover value greater than 50%.
We used these 48 points as input for all remaining tamarisk
models. We used the remaining 33 points, along with 44
additional randomly collected points, as absence points.
Our focus on points with tamarisk cover exceeding 50% is
important because it represents areas dominated by
tamarisk as opposed to the location of only a single or a
few plants. This designation is also consistent with
previous models predicting dense tamarisk habitat, along
with surveys regarding flycatcher habitat.
We received flycatcher breeding site presence points
from United States Geological Survey staff based out of
the Colorado Plateau Research Station. Eleven flycatcher
sites were located within the study area. Surveys conducted
at sites between 1997 and 2008 quantified total number of
territories per site per survey year, although not every site
was surveyed each year. Due to the dynamic nature of
flycatcher breeding, surveyed breeding sites did not
necessarily contain flycatcher territories each year. On
average, surveys documented 61 occupied territories in the
study area each year. More detailed survey results were not
provided due to the sensitivity of data associated with
endangered species. Of the 11 breeding sites, seven existed
in sites dominated by native vegetation and four existed in
sites dominated by tamarisk. Absence points were not
determined for the flycatcher. Unlike tamarisk “absence,”
flycatcher absence is more speculative because birds can
be present but go undetected. Therefore, the failure to
1) http://nhdgeo.usgs.gov/
Fig. 1 Study area map
2) http://edcsns17.cr.usgs.gov/EarthExplorer/
3) http:// www.niiss.org
122 Front. Earth Sci. 2011, 5(2): 120–129
detect a flycatcher nest does not necessarily indicate that
the area is unsuitable as habitat. Moreover, flycatchers
have high fidelity to sites where they successfully fledged
young (Paxton et al., 2007) – behavior that can result in
occupancy in subsequent breeding years even if the
territory has become marginal or unsuitable. Because of
detectability issues and site fidelity behavior, and a
relatively small number of observations, we expect the
estimation of a flycatcher model to be characterized by
higher uncertainty than the tamarisk model.
2.3 Tamarisk model variables
We based the Maxent model predicting tamarisk occur-
rence at densities greater than 50% on the published results
of Evangelista et al. (2009). The variables found in the
published results as most suitable to predict tamarisk
dominance within the landscape included tasselled cap
transformation, normalized difference vegetation index
(NDVI), and Landsat band 3 from October. Tasselled cap
transformations, originally developed to understand
changes in agricultural lands, generate three orthogonal
bands from the six-band Landsat composite (Huang et al.,
2002). The three generated bands represent measurements
of brightness (band 1, dominated by surface soils),
greenness (band 2, dominated by vegetation), and wetness
(band 3, includes interactions of soil, vegetation and
moisture patterns) (Kauth and Thomas, 1976). NDVI is
commonly used to detect characteristics of vegetation such
as canopy density. In a study of the Colorado River delta,
an arid landscape with similar riparian vegetation char-
acteristics to this study, NDVI performed best among
indices in identifying vegetation percent cover (Nagler
et al., 2001). NDVI is calculated with the third (red visible)
and fourth (near-infrared) Landsat TM bands and the
nonlinear equation:
NDVI ¼ðband 4 – band 3Þ=ðband 4 þ band 3Þ: (1)
Visible red-light reflectance in October (Landsat band 3)
has previously been used to detect tamarisk during
senescence when the plant turns a distinguishable bright
yellow-orange color (Everitt and Deloach, 1990). Due to
the lower elevation in this study area compared to that used
in Evangelista et al. (2009), and the presence of the
saltcedar leaf beetle which has been shown to extend the
growing season of tamarisk by three to four weeks (Dudley
and Deloach, 2004), we based our Landsat TM band 3
assessments on both October and November imagery.
The goal of the tamarisk modeling effort was to detect
areas already dominated by tamarisk, and not simply
suitable fo r tamarisk. Therefore, we did not include
variables such as distance to water, elevation, or slope in
the tamarisk models. These three variables describe
topographic features of the landscape, not the unique
spectral signatures that distinguish dense tamarisk infesta-
tion from other vegetation (see Evangelista et al., 2009).
Variables derived from remote sensing images enabled us
to use the unique spectral signatures of tamarisk to detect
actual areas of tamarisk dominance.
2.4 Flycatcher model variables
We based the model predicting habitat suitability for the
flycatcher on Paxton et al. (2007). Results from surveys
and previous models indicated that environmental vari-
ables thought to be important to flycatchers include
vegetation density, amount of edge habitat, and size of
patch (Sogge et al., 1997; Sogge and Marshall, 2000;
Paxton et al., 2007). Flycatchers typically arrive at nesting
areas from late April through May and return to wintering
grounds in September, and occasionally as late as October
(Finch and Stoleson, 2000). Where previous models used
variables from the months of June or July exclusively, we
added variables from five additional months (April, May,
August, September, and October) to span the entire
breeding season. We thought that environmental variables
from the months in which nesting sites were selected may
be important in predicting suitable habitat for the
flycatcher. In addition, knowing that flycatchers sometimes
produce more than one brood per year (Paxton et al.,
2007) , we thought environmenta l variables depicting
habitat quality toward the end of the breeding season
may be important as well.
We used the NDVI data sets to represent vegetation
density for each month of the growing season, and we
created a variable from a digital elevation model to depict
topographic slope. We reclassified each NDVI data set and
the slope data set using the ArcMap Spatial Analyst Iso
Cluster Unsupervised Classification tool and used the
created data sets to calculate circular neighborhood
statistics at four spatial extents (0.28, 1.13, 2.54, and
4.52 ha) (Paxton et al., 2007). We used FOCALSUM of the
reclassified NDVI and slope data sets to represent patch
size and floodplain size, respectively. We used
FOCALSTD of the reclassified NDVI and slope data sets
to represent habitat heterogeneity and edge proximity. We
used FOCALVARIETY of the reclassified NDVI data sets
to alternatively represent habitat het erogeneit y. Since
flycatchers are obligate riparian species, we included a
Euclidean “distance to water” variable, also derived from
the digital elevation model, to exclude densely-vegetated
upland areas that may otherwise appear suitable.
2.5 Data analyses
We separately modeled dense tamarisk habitat and habitat
suitability for the flycatcher with Maxent software
v.3.2
1)
. For each initial model, we used all candidate
1) http://www.cs.princeton.edu/~schapire/maxent/
Patricia YORK et al. Tamarisk-Flycatcher Maxent habitat comparison 123
environmental variables as inputs. After estimating the
initial model, we excluded all variables that contributed
less than 1.0% to tamarisk or flycatcher models (Evange-
lista et al., 2009). This step reduced the variables included
in the second model for tamarisk from 23 to 14 and for the
flycatcher from 69 to nine. From the second model outputs,
we tested the contributing variables for cross-correlations
with Predictive Analytics Software Statistics (SPSS for
Windows, Rel. 18.0.0. 2009. Chicago: SPSS Inc.). For
highly correlated variables (Pearson correlation coeffi-
cient > 0.80), we removed the variable with the lower
predictive power. After eliminating candidate predictors
due to cross-correlations, we estimated a final model. This
model specification process resulted in a tamarisk model
with ten environmental variables, and a flycatcher model
with five.
To measure the predictive performance of the tamarisk
model, we used features available through Schroder’s
ROC/AUC software
1)
. This software requires presence and
absence points to test threshold-dependent measures
including correct classification rate, sensitivity (true
positives), specificity (true negatives), and Cohen’s
maximized Kappa (the proportion of correctly classified
points), along with threshold-independent measures such
as Area Under the Receiver Operating Characteristic
(ROC) Curve (AUC). A ROC curve is created by plotting
sensitivity against ‘1-specificity’ for all possible thresh-
olds. The AUC is then calculated by measuring the
probability that a random presence point falls within the
predicted range of occurrences, and that a random absence
point falls out of the range. The AUC statistic ranges
between 0.5 (indicating that the analysis is no better than
random), and 1.0 (indicating perfect discrimination)
(Fielding and Bell, 1997; Pearce and Ferrier, 2000). We
report the “P Fair” criteria statistics where the difference
between sensitivity and specificity is minimized. We were
unable to use Schroder’s ROC/AUC software to test the
predictive performance of the flycatcher model due to the
lack of “true” absence points. For this analysis, we relied
on Maxent-generated AUC values alone.
2.6 Habitat overlap analysis
To compare the resulting continuous Tamarisk and
Flycatcher model outputs, we defined cut-off threshold
values and converted the continuous predictive values to
binary data sets. For the Tamarisk model, we used the
“ sensitivity-speci ficity difference minimizer” criteria
(0.255) (Jiménez-Valverde and Lobo, 2007), generated
from Schroder’s ROC/AUC analysis of prediction values
from the 48 training points used in the full model and a
random subset of 48 out of the 77 absence points. We
reclassified all prediction values less than 0.255 in the
continuous model output to the value of one, representing
the absence of dense tamarisk stands (defined as > 50%
cover), and all prediction values greater than 0.255 to the
value of two, representing presence of dense tamarisk
stands.
Because we lacked absence data for the flycatcher we
used a relatively robust approach to cut-off threshold
selection that tends toward high values of sensitivity and
specificity by averaging the prediction values for the
model-building presence points (Liu et al., 2005). This
approach is considered appropriate when the prevalence of
model-building data changes, as is the case with the
dynamic occupancy of flycatcher habitat (Liu et al., 2005).
We used this approach to determine a threshold value of
0.624 for the flycatcher model. We reclassified all
prediction values in the continuous model output below
0.624 to the value of three, representing habitat of low
suitability for the flycatcher, and all values above 0.624 to
the value of four, representing highly suitable habitat for
the flycatcher.
We chose reclassification values so that when multi-
plied, each product resulted in a unique value. We
multiplied the reclassified tamarisk output with the
reclas sified flycatcher output. This raster calculation
resulted in a habitat overlap analysis raster layer containing
four classes: habitat not dominated by tamarisk and low
suitability for flycatcher (overlap analysis value 3), habitat
not dominated by tamarisk and highly suitable for
flycatcher (overlap analysis value 4), habitat dominated
by tamarisk and low suitability for flycatcher (overlap
analysis value 6), and habitat both dominated by tamarisk
and highly suitable for flycatcher (overlap analysis value
8). We translated the number of 30m resolution pixels
occurring in each class to square kilometers of habitat and
calculated percentages of overlapping habitat.
3 Results
3.1 Tamarisk model
The Maxent model predicting the occurrence of dense
tamarisk stands performed quite well with a Maxent-
generated AUC score of 0.982. Ten variables contributed
to the final Tamarisk model (Table 1). The top two
contributing variables (June tasselled cap wetness and
band 3 from October) were also ranked in the top three
contributing variables in the study by Evangelista et al.
(2009).
The Tamarisk model also performed well according to
Schroder’s external model performance analysis of pooled
subset data. The AUC score was calculated as 0.952,
which significantly exceeds the AUC critical value (set at
0.70, p < 0.0001). Schroder’s external model analysis of
the “P-Fair” criteria calculated the correct classification
1) http://brandenburg.geoecology.uni-potsdam.de/users/schroeder/download.html
124 Front. Earth Sci. 2011, 5(2): 120–129
rate at 88.5%, sensitivity at 87.5%, specificity at 89.6%,
and the Cohen’s kappa statistic as 0.77. Since the correct
classification rate, sensitivity, and specificity statistics are
proportional indices, values closer to 100% indicate better
model performance (Pearce and Ferrier 2000). Fielding
and Bell (1997) suggested that a kappa score greater than
0.75 signifies excellent model performance.
3.2 Flycatcher model
The Maxent model predicting habitat suitability for the
flycatcher also performed well with a Maxent-generated
AUC score of 0.960. Five variables contributed to the final
Flycatcher model (Table 2). The top predicting variable
(May NDVI standard deviation neighborhood statistic,
radius 120 m) contributed 44.3% to model prediction.
Prediction of suitable habitat for the flycatcher increased
with greater standard deviation of NDVI, measured at the
4.5-ha-neighborhood scale (Fig. 2).The curve shows how
the logistic prediction changes as the standard deviation of
this variable increases and all other variables are kept at
their average sample value.
3.3 Habitat overlap analysis
The habitat overlap analysis compared the individual
model outputs to examine the relationship between
tamarisk and the flycatcher. Habitat overlap occurred in
four categories, according to raster calculations (Fig. 3).
The first category (habitat not dominated by tamarisk and
low flycatcher habitat suitability) included approximately
33000 km
2
, representing 97% of the modeled landscape.
The overlap analysis resulted in a calculation of approxi-
mately 500 km
2
of habitat dominated by tamarisk ( > 50%
cover) and approximately 190 km
2
of highly suitable
habitat for the flycatcher within the study area (Fig. 4). Of
the area modeled as suitable for the flycatcher, 30.7% was
also modeled as densely invaded by tamarisk. Of the area
Table 1 Contributions of variables for the Maxent model predicting
habitat dominated by tamarisk
Environmental variable Contribution/%
June tasselled cap wetness 28.3
October band 3 15.7
October tasselled cap wetness 13.2
September tasselled cap brightness 12
September NDVI 11.8
August NDVI 8.4
November band 3 5.6
October NDVI 2
August tasselled cap greenness 1.9
March NDVI 1.1
Table 2 Contributions of variables for Maxent model predicting highly
suitable habitat for the flycatcher
Environmental variable
Contribution/%
May NDVI standard deviation (r = 120 m)
a)
44.3
Distance to water
19.7
October NDVI cell variety (r =30m)
a)
15.8
Slope standard deviation (r =30m)
a)
10.8
Slope sum (r =60m)
a)
9.4
Note: a) Radius of the circular neighborhood statistic calculated for the variable
Fig. 2 Maxent prediction response curve
Fig. 3 Habitat overlap analysis map
Patricia YORK et al. Tamarisk-Flycatcher Maxent habitat comparison 125
modeled as densely invaded by tamarisk, only 11.4% was
also modeled as highly suitable habitat for the flycatcher.
4 Discussion
4.1 Data quality
The data were derived from multiple sources, representing
the best available data for modeling. We needed six
Landsat TM scenes to represent all parts of the study area,
meaning that we combined scenes from different days to
form one environmental variable. This can contribute to
varying spectral signatures, resulting in weaker models,
but this effect seemed minimal in this study. Atmospheric
“noise” can vary considerably, even within a few hours,
and differences in noise can produce differences in
adjoining Landsat TM images (Song et al., 2001; Song
and Woodcock, 2003). This issue also did not appear to
affect the outcome of the models, but it is difficult to
discern small differences that may not have occurred with
data from the exact same date and time.
We are confident that the 175 tamarisk points and field
verification produced reasonable results with minor
exceptions, where false-positive points arose under three
circumstances. Creosote bush (Larrea tridentata), agricul-
tural fields, and irrigated lawns were predicted as areas of
tamarisk habitat seven, four, and four times out of 81,
respectively. Without knowing any attributes of the
downloaded points, we had no indication if the points
had been taken in stands of tamarisk with greater than 50%
cover, or if there were flaws associated with data
acquisition. With the collection of both presence and
absence points during the verification trip, we were able to
re-run Maxent with data of known origin. The use of this
collected data greatly reduced the occurrence of false-
positives in the final tamarisk model (down from 46.3% to
26.2%).
4.2 Individual model performance
The final model predicting dense tamarisk habitat
performed quite well according to the various criteria
examined. The top contributing variables (June tasselled
cap wetness and band 3 from October) were consistent
with previous models that also used transformations of
remotely sensed images to detect tamarisk habitat (Everitt
and Deloach, 1990; Evangelista et al., 2009). As in these
previous modeling studies, we speculated that the spectral
signatures unique to tamarisk phenology provide us with
the ability to distinguish heavy infestations from other
vegetation. These results provided additional evidence that
modeling dense tamarisk habitat with remote sensing is
viable at the landscape scale.
The final model predicting habitat suitability for the
flycatcher also performed well according to model criteria.
An environmental variable representing the heterogeneous
character of a large (4.5 ha) habitat at the time of season
when flycatchers establish breeding territories contributed
to almost half of the Maxent model’s prediction. Hetero-
geneity within and beyond the breeding territory is thought
to be important to flycatcher breeding success (Durst et al.,
2007), and its importance as a predictor of suitable habitat
for the flycatcher was consistent with previous models
(Hatten and Paradzick, 2003; Paxton et al., 2007; Hatten
et al., 2010). However, these previous models investigated
this habitat characteristic in the months of June and July.
The results of this study suggest that environmental
variables from May describing habitat heterogeneity at
larger neighborhood spatial extents are important in
predicting suitable breeding habitat for the flycatcher.
4.3 Habitat overlap analysis
This analysis allowed us to examine the relationship
between tamarisk and the flycatcher within the study area.
The main reason for the flycatcher’s decline to endangered
status is thought to be the destruction of high quality,
native habitat, mostly due to the regulation of rivers in the
south-western United States and confounded by the
widespread invasion of tamarisk (Durst et al., 2007). The
models showed that 30.7% of highly suitable habitat for
the flycatcher is densely invaded by tamarisk, and only
11.4% of area densely invaded by tamarisk is also
considered suitable as breeding grounds for the flycatcher
within the study area. These results align well with survey
data that estimate approximately 27% of fl ycatcher nests
occur in dense tamarisk stands (Durst et al., 2007).
It is important to recognize the limitations of this
modeling approach when assessing management for the
flycatcher due to the sensitive nature of this endangered
bird and the limitations regarding habitat available for
breeding. The suggestion of introducing biocontrol agents
as a way to rid western rivers of tamarisk produced much
controversy because of these issues. Considering the
Fig. 4 Habitat overlap analysis
126 Front. Earth Sci. 2011, 5(2): 120–129
flycatcher is listed as endangered, any habitat that
facilitates successful breeding attempts is important in
terms of the species’ persistence. Although this modeling
method provides insight into the tamarisk-flyca tch er
relationship, field knowledge of species behavior should
always be assessed before determining management plans.
After verification of tamarisk and/or flycatcher presence
within the modeled l andscape, we r ecommend that
managers utilize different strategies according to the four
overlap categories presented in this model (Fig. 5).
Our study indicated that effective biocontrol coupled
with the reintroduction of native vegetation may have the
opportunity to increase the suitable habitat available to the
flycatcher by 30.7%. However, biocontrol efforts may
initially reduce the habitat available to the flycatcher within
the study area by the same percentage. Precautions must be
taken to ensure new native habitat is available for
individuals moving out of biocontrolled areas.
Overall, the habitat overlap analysis demonstrated how
comparison of single-species habitat models can help
determine multi-species management. The methods pre-
sented in this study offer a promising opportunity for
concurrent managem ent of invasive and end angered
species existing within the same landscape. When
attempting to manage separately for two or more
interacting species, this type of research is invaluable
(Zavaleta et al., 2001). Efforts to control exotic species can
have negative effects on native populations, especially
those of endangered, threatened, or endemic status
(Matarczyk et al., 2002). It is possible that areas of native
vegetation that once consisted of high-quality suitable
habitat for the South-western Willow Flycatcher within the
study area may now be dominated by tamarisk. Therefore,
restoration efforts, completed with concern for the
flycatcher, may have the potential to reestablish high
quality territory for this endangered songbird.
Acknowledgements We wish to thank the US Geological Survey for
funding and the Natural Resources Ecology Laboratory at Colorado State
University for logistical support. TJS contribution supported by US
Geological Survey Invasive Species Program, USGS Fort Collins Science
Program, and USDA CSREES/NRI 2008-35615-04666. Any use of trade,
product, or firm names is for descriptive purposes only and does not imply
endorsement by the US Government.
References
Bateman H L, Dudley T L, Bean D W, Ostoja S M, Hultine K R, Kuehn
M J (2010). A river system to watch: Documenting the effects of
saltcedar (Tamarix spp.) biocontrol in the Virgin River Valley. Ecol
Res, 28(4): 405–410
Chapin F S III, Zavaleta E S, Eviner V T, Naylor R L, Vitousek P M,
Reynolds H L, Hooper D U, Lavorel S, Sala O E, Hobbie S E, Mack
M C, Díaz S (2000). Consequences of changing biodiversity. Nature,
405(6783): 234–242
Cory J S, Myers J H (2000). Direct and indirect ecological effects of
biological control. Trends Ecol Evol, 15(4): 137–139
Deloach C J, Carruthers R, Dudley T, Eberts D, Kazmer D (2004). First
results for control of saltcedar (Tamarix spp.) in the open field in the
Western United States. In: Cullen, J M, Briese D T, Kriticos D J,
Lonsdale W M, Morin L, Scott J K, eds. Proceedings of the XI
International Symposium on Biological Control Weeds. Canberra,
Australia: CSIRO Entomology, 505–513
Di Tomaso J M (1998). Impact, biology, and ecology of saltcedar
(Tamarix spp.) in the southwestern United States. Weed Technol, 12:
326–336
Dudley T L, DeLoach C J (2004). Saltcedar (Tamarix spp.), endangered
species, and biological weed control – can they mix? Weed Technol,
18(sp1): 1542–1551
Durst S L, Sogge M K, Stump S D, Williams S O, Kus B E, Sferra S J
(2007). Southwestern Willow Flycatcher breeding site and territory
summary – 2006. USGS Open File Report2007–1391
Elith J, Graham C H (2009). Do they / How do they / Why do they differ?
– On finding reasons for differing pe rformanc es of spec ies
distribution models. Ecography, 32(1): 66–77
Elith J, Graham C H, Anderson R P, Dudík M, Ferrier S, Guisan A,
Hijmans R J, Huettmann F, Leathwick J R, Lehmann A (2006).
Novel methods improve prediction of species’ distributions from
occurrence data. Ecography, 29: 129–151
ESRI ArcGIS 9.2 (2006). Redlands, CA, USA, available online: www.
esri.com
Evangelista P H, Stohlgren T J, Morisette J T, Kumar S (2009). Mapping
Fig. 5 Recommended management strategies
Patricia YORK et al. Tamarisk-Flycatcher Maxent habitat comparison 127
invasive tamarisk (Tamarix): A comparison of single-scene and time-
series analyses of remotely sensed data. Remote Sens, 1(3): 519–533
Everitt B L (1980). Ecology of saltcedar – a plea for research.
Environmental Geology, 3(2): 77–84
Everitt J H, Deloach C J (1990). Remote sensing of Chinese tamarisk
(Tamarix chinensis) and associated vegetation. Weed Sci, 38: 273–
278
Fielding A H, Bell J F (1997). A review of methods for the assessment of
prediction errors in conservation presence/absence models. Environ
Conserv, 24(1): 38–49
Finch D M, Stoleson S H (2000). Status, ecology, and conservation of
the South-western Willow Flycatcher. General Technical Report
RMRS-GTR-60. Ogden, UT: U.S. Department of Agriculture, Forest
Service, Rocky Mountain Research Station
Hatten J R, Paradzick C E (2003). A multiscaled model of South-western
Willow Flycatcher breeding habitat. J Wildl Manage, 67(4): 774– 788
Hatten J R, Paxton E H, Sogge M K (2010). Modeling the dynamic
habitat and breeding population of Southwestern Willow Flycatcher.
Ecol Modell, 221(13–14): 1674–1686
Hultine K R, Belnap J, van Riper C III, Ehleringer J R, Dennison P E,
Lee M E, Nagler P L, Snyder K A, Uselman S M, West J B (2009).
Tamarisk biocontrol in the western United States: Ecological and
societal implications. Front Ecol Environ, 8(9): 467–474
Innes J, Barker G (1999). Ecological consequences of toxin use for
mammalian pest control in New Zealand: An overview. N Z J Ecol,
23: 111–127
Jiménez-Valverde A, Lobo J M (2007). Threshold criteria for conversion
of probability of species presence to either-or presence-absence. Acta
Oecol, 31(3): 361–369
Kauth R J, Thomas G S (1976). The tasseled cap – a graphic description
of the spectral-temporal development of agricultural crops as seen in
Landsat. In: Proceedings of the Symposium on Machine Processing
of Remotely Sensed Data; LARS, Purdue University: West Lafayette,
IN, USA, 41–51
Kumar S, Spaulding S A, Stohlgren T J, Hermann K A, Schmidt T S,
Bahls L L (2008). Potential habitat distribution for the freshwater
diatom Didymosphenia geminata in the continental US. Front Ecol
Environ, 7(8): 415–420
Kumar S, Stohlgren T J (2009). Maxent modeling for predicting suitable
habitat for threatened and endangered tree Canacomyrica monticola
in New Caledonia. Journal of Ecology and the Natural Environment,
1: 94–98
Leica ERDAS Imagine 9.1 (2009). Leica Geosystems Geospatial
Imaging, LLC: Atlanta G A , USA, 1991–2005; available online:
http://www.erdas.com
Lite S J, Stromberg J C (2005). Surface water and ground-water
thresholds for maintaining Populus-Salix forests, San Pedro River,
Arizona. Biol Conserv, 125(2): 153–167
Liu C, Berry P M, Dawson T P, Pearson R G (2005). Selecting
thresholds of occurrence in the prediction of species distributions.
Ecography, 28(3): 385–393
Lytle D A, Merritt D M (2004). Hydrologic regimes and riparian forests:
A structured population model for cottonwood. Ecology, 85(9):
2493–2503
Matarczyk J A, Willis A J, Vranjic J A, Ash J E (2002). Herbicides,
weeds and endangered species: Management of bitou bush
(Chrysanthemoides monilifera spp. rotundata) with glyphosate and
impacts on the endangered shrub, Pimelea spicata. Biol Conserv, 108
(2): 133–141
Merritt
D M, Poff N L R (2010). Shifting dominance of riparian Populus
and Tamarix along gradients of flow alteration in western North
American rivers. Ecol Appl, 20(1): 135–152
Morisette J T, Jarnevich C S, Ullah A, Cai W, Pedelty J A, Gentle J E,
Stohlgren T J, Schnase J L (2006). A tamarisk habitat suitability map
for the continental United States. Front Ecol Environ, 4(1): 11–17
Nagler P L, Glenn E P, Huete A R (2001). Assessment of spectral
vegetation indices for riparian vegetation in the Colorado River delta,
Mexico. J Arid Environ, 49(1): 91–110
Naiman R J, Décamps H (1997). The ecology of interfaces: Riparian
zones. Annu Rev Ecol Syst, 28(1): 621–658
Naiman R J, Décamps H, Pollock M (1993). The role of riparian
corridors in maintaining regional biodiversity. Ecol Appl, 3(2): 209–
212
Patten D T (1998). Riparian ecosystems of semi-arid North America:
Diversity and human impacts. Wetlands, 18(4): 498–512
Paxton E H, Sogge M K, Durst S L, Theimer T C, Hatten J R (2007). The
ecology of the Southwestern Willow Flycatcher in central Arizona – a
10-year synthesis report: U.S. Geological Survey Open-File Report
2007–1381
Pearce J, Ferrier S (2000). Evaluating the predictive performance of
habitat models developed using logistic regression. Ecol Modell, 133
(3): 225–245
Phillips S J, Anderson R P, Schapire R E (2006). Maximum entropy
modeling of species geographic distributions. Ecol Modell, 190(3–4):
231–259
Robinson T W (1965). Introduction, spread, and areal extent of saltcedar
(Tamarix) in the western states. US Geological Survey, Washington
D C, USA
Shafroth P B, Briggs M K (2008). Restoration ecology and invasive
riparian plants: An introduction to the special section on Tamarix spp.
in western North America. Restor Ecol, 16(1): 94–96
Sogge M K, Marshall R M (2000). A survey of current breeding habitats.
In: Finch D M, Stoleson S H, eds. Status, ecology, and conservation
of the South-western Willow Flycatcher. USDA Forest Service
General Technical Report RMRS-GTR-60. USDA Forest Service
Rocky Mountain Research Station, Ogden, UT, 43–56
Sogge M K, Marshall R M, Sferra S J, Tibbitts T J (1997). A South-
western Willow Flycatcher natural history summary and survey
protocol. USGS Biological Resources Division, Colorado Plateau
Research Station, Northern Arizona University
Sogge M K, Sferra S J, Paxton E H (2008). Tamarix as habitat for birds:
Implications for riparian restoration in the southwestern United
States. Restor Ecol, 16(1): 146–154
Song C, Woodcock C E, Seto K C, Lenney M P, Macomber S A (2001).
Classification and change detection using Landsat TM data: When
and how to correct atmospheric effects? Remote Sens Environ, 75(2):
230–244
Song C H, Woodcock C E (2003). Monitoring forest succession with
multitemporal Landsat images: Factors of uncertainty. IEEE Trans
Geosci Rem Sens, 41(11): 2557–2567
Stohlgren T J, Bull K A, Otsuki Y, Villa C, Lee M (1998). Riparian
zones as havens for exotic plant species in the central grasslands.
128 Front. Earth Sci. 2011, 5(2): 120–129
Plant Ecol, 138(1): 113–125
Stromberg J C, Beauchamp V B, Dixon M D, Lite S J, Paradzick C
(2007a). Importance of low-flow and high-flow characteristics to
restoration of riparian vegetation along rivers in arid south-western
United States. Freshw Biol, 52(4): 651–679
Stromberg J C, Lite S J, Marler R, Paradzick C, Shafroth P B, Shorrock
D, White J M, White M S (2007). Altered stream-flow regimes and
invasive plant species: the Tamarix case. Glob Ecol Biogeogr, 16(3):
381–393
Szaro R C, Rinne J N (1988). Ecosystem approach to management of
southwestern riparian communities. Transactions of the 53rd North
American Wildlife and Natural Resources Conference, 53: 502–511
Taylor J P, McDaniel K C (1998). Restoration of saltcedar (Tamarix sp.)-
infested floodplains on the Bosque del Apache National Wildlife
Refuge. Weed Technol, 12: 345–352
Unitt P (1987). Empidonax traillii extimus: an endangered subspecies.
West Birds, 18: 137–162
United States Fish and Wildlife Service (1995). Final rule determining
endangered status for the Southwestern Willow Flycatcher. Fed
Regist, 60: 10694–10715
Vitousek P M, DAntonio C M, Loope L L, Rejmanek M, Westbrooks R.
(1997). Introduced species: A significant component of human-
caused global change. N Z J Ecol, 21: 1–16
Zavaleta E S (2000). The economic value of controlling an invasive
shrub. AMBIO: A Journal of the Human Environment, 29: 462–467
Zavaleta E S, Hobbs R J, Mooney H A (2001). Viewing invasive species
removal in a whole-ecosystem context. Trends Ecol Evol, 16(8):
454–459
Patricia YORK et al. Tamarisk-Flycatcher Maxent habitat comparison 129
