1 Upper Mississippi River and Great Lakes Region Joint Venture Technical Report No. 2014-1 Modelling Great Lakes Piping Plover Habitat Selection during the Breeding Period From Local to Landscape Scales Benjamin M. Kahler, Upper Mississippi River and Great Lakes Region Joint Venture, U.S. Fish and Wildlife Service, 2651 Coo- lidge Road, East Lansing, MI 48823 Vincent S. Cavalieri, Ecological Services, U.S. Fish & Wildlife Service, 2651 Coolidge Road, East Lansing, MI 48823 INTRODUCTION Effective planning, evaluation and delivery of conservation actions rely on an under- standing of species’ ecology. In the absence of perfect knowledge, researchers and managers should work collectively and use the best available information to develop habitat management prescriptions in an adaptive management framework. Important decisions should be continual- ly improved through evaluation addressing key knowledge gaps, particularly those factors per- ceived to most limit population growth for a target species. Our understanding of species ecol- ogy and life-history requirements drives the formulation of conceptual and inferential models for bird habitat conservation (National Ecological Assessment Team 2006). Meaningful man- agement objectives can be set by understanding the biological processes involved in controlling the distribution of a species (Young and Hutto 2002). Furthermore, new information and a more refined understanding of species ecology can increase the efficacy of habitat conservation planning and delivery efforts (USFWS 2008). Monitoring is essential to assess wildlife population status and trends of a species. Mon- itoring data can also be used to inform understanding of species ecology including its distribu- tion in time and space, and to assess species-habitat associations at multiple spatial scales. In- creasingly the prediction of a species distribution from survey data is an important component of conservation planning (Guisan and Zimmerman 2000, Scott et al. 2002, Johnson and Gilling- ham 2005). For example, distribution modeling is used to predict species occurrence in previously unsampled sites by identifying biologically important variables (Young and Hutto 2002). Model- ing the distribution of a species, also called an ecological or environmental niche, uses associa- tions between species occurrence records and environmental variables to identify conditions in
30
Embed
Modelling Great Lakes Piping Plover habitat selection during the ...
This document is posted to help you gain knowledge. Please leave a comment to let me know what you think about it! Share it to your friends and learn new things together.
Transcript
1
Upper Mississippi River and Great Lakes Region Joint Venture
Technical Report No. 2014-1
Modelling Great Lakes Piping Plover Habitat Selection during the Breeding Period From Local
to Landscape Scales
Benjamin M. Kahler,
Upper Mississippi River and Great Lakes Region Joint Venture, U.S. Fish and Wildlife Service, 2651 Coo-
lidge Road, East Lansing, MI 48823
Vincent S. Cavalieri,
Ecological Services, U.S. Fish & Wildlife Service, 2651 Coolidge Road, East Lansing, MI 48823
INTRODUCTION
Effective planning, evaluation and delivery of conservation actions rely on an under-
standing of species’ ecology. In the absence of perfect knowledge, researchers and managers
should work collectively and use the best available information to develop habitat management
prescriptions in an adaptive management framework. Important decisions should be continual-
ly improved through evaluation addressing key knowledge gaps, particularly those factors per-
ceived to most limit population growth for a target species. Our understanding of species ecol-
ogy and life-history requirements drives the formulation of conceptual and inferential models
for bird habitat conservation (National Ecological Assessment Team 2006). Meaningful man-
agement objectives can be set by understanding the biological processes involved in controlling
the distribution of a species (Young and Hutto 2002). Furthermore, new information and a
more refined understanding of species ecology can increase the efficacy of habitat conservation
planning and delivery efforts (USFWS 2008).
Monitoring is essential to assess wildlife population status and trends of a species. Mon-
itoring data can also be used to inform understanding of species ecology including its distribu-
tion in time and space, and to assess species-habitat associations at multiple spatial scales. In-
creasingly the prediction of a species distribution from survey data is an important component
of conservation planning (Guisan and Zimmerman 2000, Scott et al. 2002, Johnson and Gilling-
ham 2005).
For example, distribution modeling is used to predict species occurrence in previously
unsampled sites by identifying biologically important variables (Young and Hutto 2002). Model-
ing the distribution of a species, also called an ecological or environmental niche, uses associa-
tions between species occurrence records and environmental variables to identify conditions in
2
which populations can be maintained (Austin 2002, Pearson 2007). Because there is no one
best method for modeling distribution, many studies have used a comparative approach (e.g.,
Moisen and Frescino 2002, Thuiller et al. 2003, Muñoz and Felicísimo 2004).
Ecological niche modeling methods can include those that use combinations of recorded
species presence, absence, and/or pseudo-absence (Austin 2002). Detailed descriptions of
these commonly used presence only, presence/pseudo-absence, and presence/absence meth-
ods have been previously described (Austin 2002, Guisan and Thuiller 2005, and Elith et al.
2006). Applying different modeling algorithms to the same survey data may lead to very differ-
ent predictions of a species’ distribution, mainly due to slight differences in the statistical struc-
ture and assumptions of each algorithm (Austin 2002). Brotons et al. (2004) found pres-
ence/absence models were more accurate than presence-only models, concluding that absence
data provides information that is useful and reliable in model calibration.
Many techniques have been developed to model the distribution of a species for con-
servation planning (Guisan and Thuiller 2005). Advances in quantitative methods, including ge-
ographic information systems (GIS) and remote sensing techniques, have enabled researchers
and managers to incorporate “landscape-level” measures into ecological research (Bissonette
1997, Klopatek and Gardner 1999, Turner et al. 2001) and to model species-habitat associations
at multiple spatial scales to predict species distributions (Scott et al. 2002). Predictive models
illustrate and infer the suitability of potential habitats when linked to a GIS. Many studies have
used GIS to analyze wildlife-habitat associations at multiple spatial scales (e.g., Naugle et al.
1999, Scott et al. 2002, and Elith et al. 2006).
The Piping Plover (Charadrius melodus) is a small species of shorebird endemic to
North America (Eliot-smith and Haig 2004, Wemmer et al. 2001). There are three recognized
breeding populations: the Northern Great Plains, Atlantic Coast and the Great Lakes (Eliot-
Smith and Haig 2004, USFWS 2003). While all three are imperiled, and were federally listed un-
der provisions of the Endangered Species Act
(ESA) in 1986, the Great Lakes population is by
far the smallest and most at risk and is listed as
endangered; the other two are listed as threat-
ened (Wemmer et al. 2001, USFWS 1985,
USFWS 2003). The Great Lakes Piping Plover
population was estimated to be between 492
and 682 breeding pairs in the late 1800s-early
1900s but only 17 pairs were recorded in the
entire Great Lakes basin at the time of listing
(Russell 1983, USFWS 1985).
V. Cavalieri
3
Habitat loss and degradation from development, as well as increased disturbance due to
an increase in recreational use of beaches, are believed to be among the primary causes leading
to the decline of the Great Lakes Piping Plover population, although other factors have also
contributed (Russell 1983, USFWS 1986, Wemmer et al. 2001, USFWS 2003, Haffner et al.
2009). Since listing, development of Great Lakes shoreline for residential, commercial and rec-
reational properties has continued to accelerate in some areas, resulting in reduced Piping
Plover nesting habitat at these sites (Wemmer et al. 2001). Conversely, fluctuating Great Lakes
water levels can significantly change coastal shorebird habitat (Potter et al. 2007), and the area
available to breeding Piping Plovers has likely expanded to relatively low Great Lakes water lev-
els since 1997. Protecting and monitoring as many Piping Plover nests as possible is very im-
portant for the recovery of this critically endangered species. Additionally, a better understand-
ing of Piping Plover habitat selection and use will provide managers with the tools they need to
be able to make important management decisions that will aid in the recovery of this very rare
shorebird.
Piping Plover research in the Great Lakes basin has enabled managers to determine
basic habitat characteristics for nesting locations. Typical nesting habitat consists of wide sand
or cobble beaches with sparse vegetation on Great Lakes shorelines (Powell and Cuthbert 1992,
Pike 1985, Wemmer et al. 2001, USFWS 2003). This understanding allowed for the designation
of Great Lakes Piping Plover critical habitat. Critical habitat is defined by the ESA as areas on
which have the physical or biological features (i.e., primary constituent elements) that are es-
sential to the conservation of the species and where there may be special management consid-
erations for protection (USFWS 2003). Only locations that have the primary constituent ele-
ments are considered critical habitat. The primary constituent elements for Great Lakes Piping
Plovers as defined in the critical habitat rule are “island and mainland shorelines that support
open, sparsely vegetated, sandy habitats, such as sand spits or sand beaches, that are associat-
ed with wide, unforested systems of dunes and inter-dune wetlands” (USFWS 2003). Additional
primary constituent elements include a total beach area of at least 2 ha, at least 50 m of beach
where beach width is greater than 7 m and a distance to tree line from the normal high water-
mark of more than 50 m. The location must also have low levels of disturbance from humans
and pets (USFWS 2003). While these primary constituent elements and associated critical habi-
V. Cavalieri
4
tat have allowed for the protection of many Great Lakes Piping Plover habitats, the recent ad-
vancements in Geographic Information Systems and remote sensing techniques allow for a
broader examination of Piping Plover habitat across different spatial scales. We developed and
assessed spatially explicit models of Piping Plover presence and habitat use in Lakes Michigan,
Huron, and Superior. This allowed us expand efforts to predictively map Piping Plover habitat in
the Great Lakes region.
In addition to its federal status, the Piping Plover is a Joint Venture (JV) focal shorebird
species representing beach foraging habitats in the Upper Mississippi River and Great Lakes JV
region (Potter et al. 2007). Biological models for Piping Plovers were not developed in the 2007
JV Shorebird Habitat Conservation Strategy (Potter et al. 2007) due to a lack of adequate spatial
data and the dynamic nature of Piping Plover breeding and migration habitats. Nevertheless,
population estimates and goals for Piping Plover were translated into protection and restora-
tion objectives for beach habitat in the JV region (UMRGLR JV 2007).
The goal of this study was to use three analytical approaches to determine the land-
scape habitat features associated with the distribution of Piping Plovers in the U.S. Great Lakes
region. Our study is the first effort to evaluate landscape suitability for the Great Lakes Piping
Plover across the majority of its current geographic breeding extent in the U.S. Great Lakes re-
gion and our results will further JV planning efforts by updating understanding of quantifiable
habitat associations and assist in evaluating regional habitat requirements and objectives. In
addition, understanding the distribution of landscape suitability for Great Lakes Piping Plover
will enhance targeting of monitoring activities, understand what it may require to make areas
more suitable for Piping Plover through habitat restoration efforts, and provide a means to as-
sess areas designated as critical habitat for the Piping Plover. To achieve this goal, we asked two
questions.
1. What is the distribution of landscape suitability for Piping Plover across its current
breeding extent in Michigan, USA?
2. What are landscape metrics associated with Piping Plover nest locations in this re-
gion?
To address these questions we compare three analytical approaches to estimate landscape
suitability for Piping Plover across its geographic breeding extent in Michigan, USA (Figure 1).
This approach provides a suite of model predictions within a matrix of user familiarity and com-
prehension and statistical complexity and appropriateness. All models presented have utility in
understanding habitat selection by Piping Plover. By presenting and evaluating the performance
of different models we empower the user to choose which model(s), if any, have potential for
assessment in the field and/or the enhancement of population monitoring and habitat conser-
vation and restoration activities.
5
Figure 1. Geographic extent of nest records and landscape modelling for
Piping Plovers in Michigan, USA.
METHODS
Nest locations
Monitoring efforts for Great Lakes Piping Plover have lacked systematic, unbiased col-
lection of data out of logistical and financial necessity. Biologists attempt to locate every Piping
Plover nest in the Great Lakes each breeding season (LeDee et al. 2010). Monitoring efforts are
largely targeted at locations where Piping Plovers have been previously recorded, particularly in
recent years (LeDee et al. 2010). Other locations having appropriate habitat conditions, but no
recent records, are checked more opportunistically. Each nest location is recorded with a GPS
unit and added to a nest location database. Piping Plovers exhibit strong breeding philopatry
(site fidelity; Haig and Oring 1988), nesting in close proximity to breeding territories from previ-
ous years for the duration of their adult life (believed to be ~5 years; Wilcox 1959, Elliot-Smith
and Haig 2004). In an effort to reduce pseudo-replication among nest locations in the analysis,
we used Piping Plover nest locations from recent years (2000—2010), separated by 5 years be-
tween time periods (further description below).
Predictor variable estimation
We hypothesized local and landscape variables (Table 1) that would predict presence of
Piping Plover nests during the breeding season based on current understanding of life-history
requirements, species-habitat associations, and expert opinion. These variables were surro-
6
gates for local and landscape processes that influence habitat selection by Piping Plover either
directly or indirectly. We gathered spatial data from several sources (Table 1) and generated 30
m resolution raster coverages for each variable across the study area using ModelBuilder work-
flows in ArcGIS 10.0 (ESRI, Redlands, California, USA).
Analysis
We modeled the probability of Piping Plover nest occurrence using three modeling algo-
rithms: Generalized Additive Models (GAM; Hastie and Tibshirani, 1986, 1990, Yee and Mitchell
1991), Boosted Regression Trees (BRT; Elith et al. 2008), and Maximum Entropy models
(MAXENT; Phillips et al. 2006, Phillips and Dudík 2008). The first two algorithms are more wide-
ly understood and best suited for modeling species-habitat associations where systematic sam-
pling has led to recorded species presence and absence; two aspects absent in monitoring Pip-
ing Plover in the Great Lakes region. The third algorithm is lesser known and was developed to
study species-habitat associations and predict geographic distribution of species in the absence
of systematic sampling and known absence (e.g., museum specimens).
Each method required us to generate pseudo-absence or background locations to pair with
nest locations in analyses. We used generalized random-tessellation stratified (GRTS) design to
select pseudo-absence and background locations (Kincaid et al. 2008), and we used National
Land Cover Database (NLCD) from years most closely matching Piping Plover nest site data used
for model development.
We created separate sampling timeframes for the years 2001 and 2006 based on NLCD spa-
tial data. For GAM and BRT analyses, we paired nest locations from the year 2000 (n=34) with
spatial data from 2001 (Homer et al. 2007); paired nest locations from the years 2005 (n=57)
and 2010 (n=51) with NLCD spatial data from 2006 (Fry et al. 2011) (Table 2). We created coast-
line extents by buffering coastline derived from NLCD spatial data by 2.4 km for each time peri-
od. For GAM and BRT analyses we drew pseudo-absence locations only from areas within
coastal extents classified as barren cover. We drew unstratified, equal probability samples for
pseudo-absence locations at a 1:3 ratio (nests: pseudo-absence) for 2001 (n=102) and 2006
(n=324) coastal extents in R 2.15.1 (R Development Core Team 2012).
We combined nest records and pseudo-absence locations for GAM and BRT analyses and
divided them into separate training (70%) and testing (30%) sets using equal probability GRTS
sampling, stratifying the sample in proportion to nests in the year 2000 and nests in the years
2005 and 2010. For MAXENT analyses we paired nest locations (n=53) and spatial data from the
year 2006 (Table 2). We drew unstratified, equal probability samples for background locations
(n=5000) within the 2006 coastal zone extent in R. We extracted values of predictor variables
(i.e., local and landscape habitat characteristics) to nest and pseudo-absence/background loca-
tions.
7
Table 1. Predictor variables used in analyses of Piping Plover nest occurrence in its recorded geographic breeding extent in Michigan, USA.
Scale
Type Abbreviation Description Local Landscape Unit Source
Non-anthropogenic BA100M Barren cover within 0.1 km radius X % National Land Cover Database
BA1KM Barren cover within 1 km radius X % "
WDY100M Woody cover within 0.1 km radius X % "
WDY1KM Woody cover within 1 km radius X % "
WDY10KM Woody cover within 10 km radius X % "
D2WDY Distance to woody cover X m "
D2RIV06 Distance to stream/river X m National Hydrography Dataset
Anthropogenic DVT100M Developed cover within 0.1 km radius X % National Land Cover Database
DVT1KM Developed cover within 1 km radius X % "
DVT10KM Developed cover within 10 km radius X % "
RD100M Road density within 0.1 k m radius X internal TIGER/Line roads
RD1KM Road density within 1 km radius X internal "
RD10KM Road density within 10 km radius X internal "
D2RD Distance to road X m "
D2DVT Distance to developed cover X m National Land Cover Database
Topographic PERSLOPE Slope X % x10 Digital Elevation Model
D2COAST Distance to coastline X m National Land Cover Database
Categorical SANDY Geomorphological shoreline classification X binary U.S. Army Corps of Engineers
MAXENTg 0.214 0.805 0.794 (0.070) 0.808 (0.035) 0.509 (0.074) 0.820h (0.044) a Boosted Regression Tree (BRT), Generalized Additive Model (GAM), Maximum Entropy (MAXENT) b Proportion of all cases correctly predicted (OPS = Overall prediction success) c Proportion of true positives correctly predicted. d Proportion of true negatives correctly predicted. e Proportion of specific agreement. Model performance based on values 0.0-0.4 = poor, 0.4-0.75 = good,
> 0.75 = excellent (after Landis and Koch 1977). f Area under the Receiver Operating Characteristic Curve; Model discriminatory ability based on values
g Evaluation indices for MAXENT were calculated from predicted values at the same nest and pseudo-
absence locations used to evaluate GAM and BRT models. h AUC value on original testing data = 0.989 (0.003), indicating excellent model discriminatory ability.
17
Figure 5. Distribution of predicted probabilities from three models (Generalized Additive Model, GAM;
Boosted Regression Tree, BRT; Maximum Entropy, MAXENT) of probability of Piping Plover nest occur-
rence in its recorded geographic breeding extent in Michigan, USA. For models with good discriminatory
ability, distribution of sites with recorded presence (black) will occur disproportionately at higher pre-
dicted probabilities and sites with recorded absence (grey) will occur disproportionately at lower pre-
dicted probabilities. Note: cross-hatched bars (number of pseudo-absence or background plots with low
predicted probability) are truncated.
Figure 6. Threshold-dependent evaluation indices (sensitivity, specificity, and Kappa) as a function of
threshold from three models (Generalized Additive, GAM; Boosted Regression Tree, BRT; Maximum
Entropy, MAXENT) of probability of Piping Plover nest occurrence in its recorded geographic breeding
extent in Michigan, USA. Model quality is better where Kappa reaches a maximum value and stays at a
higher value for a greater range of threshold values and where lines representing sensitivity and speci-
ficity cross at a higher value.
18
Figure 7. Receiver Operative Characteris-
tic (ROC) plots from models predicting the
probability of Piping Plover nest occur-
rence in its recorded geographic breeding
extent Michigan, USA. ROC plots and as-
sociated Area Under the Curve (AUC) val-
ues are reported for three algorithms
(Generalized Additive, GAM; Boosted Re-
gression Tree, BRT; Maximum Entropy,
MAXENT). Model discriminatory ability
based on AUC value 0.5-0.7 = low, 0.7-0.9
= moderate, > 0.9 = excellent.
V. Cavalieri
19
Figure 8. Probability of Piping Plover nest occurrence in its recorded geographic breeding extent in Michigan, USA, obtained with Generalized
Additive Model (GAM), Boosted Regression Tree (BRT), and Maximum Entropy (MAXENT) models. The mosaic model displays predicted probabil-
ities based on averaged model predictions.
20
Figure 8 (continued).
21
DISCUSSION
Model results point to a variety of factors, including anthropogenic, non-anthropogenic
and topographic variables that influence the selection of Great Lakes Piping Plover habitat. Like
previous work, these model results point to wide beaches on undeveloped Great Lakes shore-
line areas as being the best locations for Piping Plover nesting habitat. What this work does
however, is for the first time predicts the suitability of habitat across much of the state of Mich-
igan based on best fit models. Our analyses of landscapes associated with Great Lakes Piping
Plover nest sites provide a broader understanding of breeding habitat characteristics at multi-
ple scales. These results will help inform annual monitoring program efforts such as determin-
ing new survey locations and associated resource allocations. Additionally, models can contrib-
ute to important habitat management decisions, such as which locations might be most suita-
ble for habitat restoration efforts.
Although our understanding of potential factors limiting population growth throughout
the annual cycle is incomplete, effective habitat conservation and population monitoring for
the Great Lakes Piping Plover during the breeding period remains a critical management foci
(USFWS 2003, Haffner et al. 2009). As part of this effort, each season biologists attempt to lo-
cate every Great Lakes Piping Plover pair so each individual nest can be monitored and protect-
ed. Additionally, population demography has been studied through a long term banding effort,
with > 1100 color-banded Piping Plovers in the Great Lakes since 1993 (LeDee et al. 2010). Even
with high rates of annual banding and detection probability, each season adult Piping are ob-
served that are unmarked. Between 1993 and 2009 an average of 6 breeding adults per year in
the Great Lakes population was un-banded suggesting there may be areas in the Great Lakes
harboring Piping Plovers that remain unmonitored (LeDee et al. 2010). Model results can be
used to identify and prioritize additional suitable habitats to be surveyed for breeding pairs.
Once found new nesting areas can be assessed for protection and monitored to increase nest
success. Furthermore, population estimates (a measure of goal achievement) may be refined
for this Great Lakes cohort.
The Great Lakes Piping Plover population appears to be increasing slowly in recent
years, yet the only population viability analysis (PVA) completed on the Great Lakes segment of
the population suggests it will likely remain vulnerable to extirpation for decades as a result of
environmental and demographic factors (Wemmer et al. 2001). This PVA recommends many
conservation actions such as protecting as many physically suitable breeding sites as possible
and restoring marginally suitable areas to increase habitat quality for breeding Great Lakes Pip-
ing Plover (Wemmer et al. 2001). The PVA indicated that there is likely a shortage of habitat in
the Great Lakes where human disturbance is low enough for the successful breeding of Piping
Plovers and that an increase in permanent protected habitat is likely required for the long-term
recovery of the Great Lakes Piping Plover population (Wemmer et al. 2001). The amount by po-
tential nesting habitat for Piping Plover can now be estimated by combining predictions from
22
models presented here with information about area use during the breeding period (Haffner et
al. 2009).
This project has shown that there are many areas on the Michigan Great Lakes shoreline
where habitat for Plovers is available but perhaps marginal (Figure 9). Some of these areas may
be suitable locations for Great Lakes Piping Plover habitat restoration projects. The models also
suggest that habitat may not be the limiting factor for Piping Plover at this time. However our
methods may not have been able to assess human and domestic animal disturbance at sites or
other factors such as encroachment of vegetation on beaches. It is possible that variables cho-
sen as a proxy of disturbance, such as distance to roads, do not adequately measure true levels
of disturbance at some locations. It may be that many sites that appear suitable according to
these models may have high levels of disturbance making them unattractive locations for
breeding Piping Plovers or that plant succession has led to habitat conditions unsuitable for
Plovers. Future work should focus on including these factors into a similar analysis. Suggested
future actions include efforts to ground-truth model results using localized maps created with a
model layer (Figure 9). For example, biologists familiar with Piping Plover habitat requirements
should visit high likelihood of occurrence sites identified by the model but currently without
breeding plovers to evaluate model usefulness and accuracy. Biologists should also take meas-
urements of habitat variables known to be required by Piping Plovers or to validate model re-
sults or discover potential weaknesses in the model. Additionally, localized maps developed for
ground-truthing efforts should also be used to target locations for additional Piping Plover sur-
veys and ground-truthing results may also help select locations where habitat management can
be used to improve habitat conditions for Great Lakes Piping Plovers.
V. Cavalieri
23
Figure 9. Example of high resolution spatial prediction of Piping Plover nest occurrence two ar-
eas in the study area (bottom two images) based on average landscape suitability among three
algorithms (i.e., mosaic model). The two areas shown are intended to illustrate areas with rela-
tively high (bottom left) and relative low (bottom right) overall probability of nest occurrence.
Gray shading within model predictions (2.4 km coastline buffer) indicates minimal landscape
suitability (probability of Piping Plover nest occurrence = 0).
Model limitations:
An important distinction exists between geographic space and environmental space in
the context of distribution modeling. Ecological niche modeling relates spatially referenced oc-
currence records with environmental conditions (environmental space) to project modeled dis-
tributions onto geographic space (Figure 10). Hutchinson (1957) defined a fundamental niche as
the full range of abiotic conditions within which a species can persist. When plotted in geo-
graphic space, the fundamental niche is referred to as the potential distribution. The actual dis-
24
tribution of a species is the area within its fundamental niche that it truly occupies. When the
actual distribution is plotted in environmental space it is known as a species’ occupied niche
(Pearson 2007). The niche occupied by a species is the actual area where a species occurs and is
viable. However, most methods used to model ecological niches estimate a species’ realized
niche which is not the same as its occupied niche.
Figure 10. Illustration of the relationship between a hypothetical species’ distribution in
geographical space and environmental space. Adapted from Pearson (2007).
Implementing ecological niche modeling methods without accounting for population
demographic parameters results in a species’ realized niche (Dias 1996, Pearson 2007). The re-
alized niche includes the niche occupied by a species and other areas where it cannot persist
(i.e., non-habitat or habitat sinks). A species will occupy areas where it cannot persist locally,
also known as habitat sinks (Pulliam 1988, Pulliam and Danielson 1991). However, habitats may
fluctuate between being habitat sinks (λ < 1) and sources (λ > 1) depending on annual resource
fluctuations (e.g., water and food), predation, and conspecific competition (Pulliam and Dan-
ielson 1991, Dias 1996). Including all spatially referenced occurrence locations for a species in
geographic space in its occupied niche is of limited value in conservation planning without es-
timates of population demography (e.g., survival probability). Therefore, if ecological niche
modeling with species presence/absence data is applied without accounting for source/sink dy-
namics, results will include predictions of areas likely to serve as either population sources or
sinks.
25
ACKNOWLEDGEMENTS
We would like to thank Francesca Cuthbert, Jennifer Stucker and Sarah Saunders for da-
ta access, advice on study design and for reviewing the manuscript. Darin Simpkins, Christie
Deloria, Ted Koehler and Jack Dingledine were instrumental in developing the idea for this pro-
ject and for advice on study design and analysis, as well as manuscript review. We would also
like to thank Greg Soulliere and Rachel Pierce for additional manuscript reviews. Special thanks
to the dozens of plover monitors, grad students and volunteers who collected the Piping Plover
nesting data used in this analysis. Funding for this project came from the Coastal Program of
the U.S. Fish and Wildlife Service.
LITERATURE CITED
Bissonette, J.A., ed. 1997. Wildlife and landscape ecology: effects of pattern and scale. Springer-Verlag
New York, Inc., New York, USA.
Brotons, L., W. Thuiller, M.B. Araujo, and A.H. Hirzel. 2004. Presence-absence versus presence-only
modelling methods for predicting bird habitat suitability. Ecography 27:437-448.
Burnham, K.P., and D.R. Anderson. 2002. Model selection and multimodel inference, Second edition.
Springer-Verlag New York, Inc., New York, USA.
Cohen, J. 1960. A coefficient of agreement for nominal scales. Educational and Psychological Measure-
ment 20:37-46.
Cramer. J.S. 2003. Logit models: from economics and other fields. Cambridge University Press, UK.
Elith, J., C.H. Grahm, R.P. Anderson, M. Dudík, S. Ferrier, A. Guisan, R.J. Hijmans, F. Heuttmann, J.R.
Leathwick, A. Lehmann, J. Li, L.G. Lohmann, B.A. Loiselle, G. Manion, C. Moritz, M. Nakamura, Y.
Nakazawa, J.McC. Overton, A.T. Peterson, S.J. Phillips, K. Richardson, R. Scachetti-Pereira, R.E.
Schapire, J. Soberón, S. Williams, M.S. Wisz, and N.E. Zimmerman. 2006. Novel methods improve
predictions of species’ distributions from occurrence data. Ecography 29:129-151.
Eilth, J., S.J. Phillips, T. Hastie, M. Dudik, Y.E. Chee, and C.J. Yates. 2011. A statistical explanation of
MaxEnt for ecologists. Diversity and Distributions 17:43-57.
Elith, J., J.R. Leathwick, and T. Hastie. 2008. A working guide to boosted regression trees. Journal of Ani-
mal Ecology 77:802-813.
Elliot-Smith, E., and S.M. Haig. 2004. Piping plover (Charadrius melodus), in The Birds of North America
Online (A. Poole, Ed.). Ithaca: Cornell Lab of Ornithology.
Fielding, A.H., and J.F. Bell. 1997. A review of methods for the assessment of prediction errors in conser-
bNatural logarithm of the odds of detection as a function of area and water permanance categories.cK, number of parameters in model; AIC, Akaike's Information Criterion; ∆AIC, difference in AIC relative to top ranked model; w i , relative Akaike
weight
aModel derived from variables considered to be anthropogenic (A), non-anthropogenic or "natural" (N), topographic (T), or including