Close
About
FAQ
Home
Collections
Login
USC Login
Register
0
Selected
Invert selection
Deselect all
Deselect all
Click here to refresh results
Click here to refresh results
USC
/
Digital Library
/
University of Southern California Dissertations and Theses
/
Assessing the connectivity for the jaguar (Panthera onca) in the United States-Mexico border ecoregions using species distribution modeling and factorial least cost path analysis
(USC Thesis Other)
Assessing the connectivity for the jaguar (Panthera onca) in the United States-Mexico border ecoregions using species distribution modeling and factorial least cost path analysis
PDF
Download
Share
Open document
Flip pages
Contact Us
Contact Us
Copy asset link
Request this asset
Transcript (if available)
Content
Assessing the Connectivity for the jaguar (Panthera onca) in the United States-Mexico Border
Ecoregions Using Species Distribution Modeling and Factorial Least Cost Path Analysis
by
Cirenia A. Torres
A Thesis Presented to the
FACULTY OF THE USC GRADUATE SCHOOL
UNIVERSITY OF SOUTHERN CALIFORNIA
In Partial Fulfillment of the
Requirements for the Degree
MASTER OF SCIENCE
(GEOGRAPHIC INFORMATION SCIENCE AND TECHNOLOGY)
August 2021
Copyright © 2021 Cirenia A. Torres
ii
Epigraph
Gentle eyes
that see so much,
paws that have
the quiet touch.
Purrs to signal
"all is well"
and show more love
than words can tell.
Graceful movements
touched with pride,
a calming presence
by our side.
A friendship
that will last and grow -
small wonder
why we love them so.
- Author Unknown
iii
Dedication
To Panthera onca (Linnaeus, 1758)
Eric Kilby https://www.flickr.com/photos/ekilby/6703324115/
iv
Acknowledgements
Special thanks to Drs. Cushman and Wan whose help was integral to completing my thesis. Dr.
Wan provided training using his python and R scripts for multi-scale covariate processing and
producing SDMs. My sincere thanks to Dr. Cushman whose research motivated and inspired me
to push myself to learn beyond what I learned in my SSI classes. He was very encouraging
throughout this whole process and without his help in answering modeling questions I would not
have been able to complete this thesis. Dr. Lopes-Gonzales recommend an ecology book that
was very helpful for estimating mammalian home ranges which further helped me with the
jaguar modeling. Lastly, special thanks to Drs. Wilson and Fleming for their meticulous
feedback and helping me reach the summer graduation deadline.
Without Google Earth Engine it would not have been possible to preprocess my vegetation and
water index covariate datasets using Landsat data. This is a valuable and free resource which I
had not previously used in my SSI classes. It challenged me to learn JavaScript, other
programming languages, and understand how to process Landsat data. My sincere thanks and
appreciation to Google Earth Engine and the wonderful team who made this free for anyone to
use.
Special thanks to my two cat children (Oniyuri and Yoruichi) and Larry Tran who have been
patient with me throughout this process.
Before starting the SSI GIST program, I had very limited knowledge of GIS. I am happy that I
learned the theory and applications for spatial thinking. All classes were fun and challenging.
v
Table of Contents
Epigraph .................................................................................................................................... ii
Dedication ................................................................................................................................. iii
Acknowledgements ................................................................................................................... iv
List of Tables ........................................................................................................................... vii
List of Figures ......................................................................................................................... viii
Abbreviations ............................................................................................................................ ix
Abstract .................................................................................................................................... xi
Chapter 1 : Introduction ..............................................................................................................1
1.1. Landscape Connectivity .................................................................................................. 1
1.1.1. Habitat Loss and Fragmentation ............................................................................. 1
1.1.2. Connectivity .......................................................................................................... 2
1.2. Co-distributed Species in the US-Mexico Border Regions .............................................. 4
1.2.1. Border Ecoregions and Biodiversity Hotspots ........................................................ 4
1.2.2. Species at Risk Due to Border Dispersal Barriers ................................................... 4
1.3. Research Objectives and Thesis Organization ................................................................. 6
Chapter 2 : Related Work ............................................................................................................8
2.1. Multispecies Connectivity Modeling............................................................................... 8
2.2. Species Distribution Modeling with Presence-only Data ............................................... 10
2.2.1. Maximum Entropy Method (MaxEnt) for modeling species geographic
distributions............................................................................................................ 10
2.2.2. Generalized Linear Models .................................................................................. 13
2.2.3. Generalized Boosted Models ................................................................................ 15
2.2.4. Random Forests Models ....................................................................................... 15
2.3. Species Connectivity Modeling .................................................................................... 17
vi
2.3.1. Resistant kernel modeling .................................................................................... 17
2.3.2. Factorial least cost path modeling ........................................................................ 18
2.3.3. Graph theory ........................................................................................................ 20
2.4. Focal Species Selection ................................................................................................ 22
2.4.1. Jaguar (Panthera onca) Background ..................................................................... 22
2.4.2. Jaguar Modeling Applications .............................................................................. 25
Chapter 3 : Methods .................................................................................................................. 33
3.1. Research Design ........................................................................................................... 33
3.2. Focal Species Datasets .................................................................................................. 34
3.3. Focal Species Covariate Datasets .................................................................................. 35
3.4. Focal Species Habitat and Corridor Modeling ............................................................... 41
3.4.1. Random Forests ................................................................................................... 41
3.4.2. Calculating resistant kernels and factorial least-cost paths .................................... 42
3.4.3. Using the graph network algorithm ...................................................................... 43
3.5. Model Selection, Validation, and Evaluation ................................................................ 43
Chapter 4 : Results .................................................................................................................... 45
4.1. Jaguar Habitat Map ....................................................................................................... 45
4.2. Jaguar Resistance Surface ............................................................................................. 48
4.3. Jaguar Corridor Modeling ............................................................................................. 50
Chapter 5 : Conclusion .............................................................................................................. 54
References ................................................................................................................................ 57
vii
List of Tables
Table 1 List of the eight family types and their associated default link functions to be used with
the glm() function in R ...................................................................................................... 14
Table 2 Jaguar habitat modeling studies ............................................................................. 28-29
Table 3 Data layers used by different studies for creating jaguar permeability matrices ............ 31
Table 4 Cost values (cost to jaguar movement) obtained from 15 jaguar experts assigned to
various landscape layers, from Rabinowitz and Zeller (2010) ............................................ 32
viii
List of Figures
Figure 1 Map of study area used for this thesis project showing US-Mexico ecoregions .............5
Figure 2 Graph showing the distribution produced using the cloglog and logistic transforms .... 13
Figure 3 Example factorial least-cost path analysis .................................................................. 19
Figure 4 Map showing approximate current and historic geographic range of the jaguar and the
range limits based on found fossils from late Pleistocene and mid Pleistocene ................. 23
Figure 5 Average human population density for the period 2000 – 2020 ................................... 37
Figure 6 16 km univariate scaling for the compound topographic index .................................... 39
Figure 7 16 km univariate scaling for annual mean temperature ............................................... 39
Figure 8 8 km univariate scaling for slope position ................................................................... 40
Figure 9 8 km univariate scaling for roughness ......................................................................... 40
Figure 10 Jaguar potential habitat map in the US/Mexico border ecoregions from the RF model
.......................................................................................................................................... 46
Figure 11 The scaled variable importance (a), bootstrap error convergence (b), and the ROC
curve (c) graphs for the random forest jaguar potential habitat map ................................... 47
Figure 12 Partial dependency plots (a) – (c), and (d) Presence/Absence proximity matrix ......... 48
Figure 13 Jaguar resistance surface created using a negative exponential function and rescaled
from 1 to 100 ..................................................................................................................... 49
Figure 14 Map of jaguar resistance surface with major roads and the US-Mexico border wall
overlaid ............................................................................................................................. 50
Figure 15 Map of predicted corridors for the jaguar with the US-Mexico border wall and areas in
Figures 16 and 17 overlaid on top of the corridor density map ........................................... 52
Figure 16 The predicted corridors for the jaguar with the previous (old) US-Mexico border wall
overlaid on top ................................................................................................................... 53
Figure 17 The predicted corridors for the jaguar with the new US-Mexico border wall overlaid
on top................................................................................................................................. 53
ix
Abbreviations
AUC Area Under the (ROC) Curve
CBP U.S. Customs and Border Protection
CITES Convention on International Trade in Endangered Species of Wild Fauna and
Flora
CTI Compound Topographic Index
DEM Digital Elevation Model
DHS Department of Homeland Security
EM Ensemble Model
ENFA Ecological Niche Factor Analysis
EVI Enhanced Vegetation Index
FLCP Factorial Least Cost Path
FPR False Positive Rate
GARP Genetic Algorithm for Ruleset Prediction
GBIF Global Biodiversity Information Facility
GBM Generalized Boosted Model
GEE Google Earth Engine
GIS Geographic Information System
GLM Generalized Linear Model
GPS Global Positioning System
GPW Gridded Population of the World
IIC Integral Index of Connectivity
IPP Inhomogeneous Poisson Process
x
IUCN International Union for Conservation of Nature
JCMA Jaguar Management and Conservation Area
JCU Jaguar Conservation Unit
JOD Jaguar Observations Database
LCP Least Cost Path
MaxEnt Maximum Entropy Method
MD Mahalanobis Distance
NDVI Normalized Difference Vegetation Index
OLS Ordinary Least Squares
OOB Out of Bag
PC Probability of Connectivity
RF Random Forest
ROC Receiver Operating Characteristic
SDM Species Distribution Model
SSI Spatial Sciences Institute
SVM Support Vector Machines
TPR True Positive Rate
TRI Terrain Ruggedness Index
UNICOR UNIversal CORridor Network Simulator
US United States of America
USGS U.S. Geological Survey
xi
Abstract
Connectivity is important for biodiversity conservation because it can offset the impacts of
habitat loss and fragmentation, allowing migration, dispersal, and adequate gene flow. Barriers
that cut across a species range such as the United States-Mexico border wall can block dispersal
and negatively impact gene flow between populations. It is therefore important to understand
how to establish or re-establish wildlife corridors in order to help species survive. The focal
species selected for this thesis project was the jaguar (Panthera onca). The study area comprised
several ecoregions that covered portions of the United States of America (US) and Mexico. The
jaguar’s suitable habitat was identified using a Random Forest model to predict potential
habitats. The factorial least-cost path analysis was used to identify the jaguar’s potential
corridors. Results predict there is good habitat for jaguars in the Sonoran-Sinaloan subtropical
dry forest, Sinaloan dry forests, Sierra Madre Occidental, California montane chaparral and
woodlands, Arizona Mountains forest, Sierra Madre Oriental pine-oak forests, Veracruz moist
forests, Sierra de la Laguna pine-oak forests, Sierra de la Laguna dry forests, Tamaulipan
matorral, and small portions of the Sonoran desert ecoregion. The jaguar's potential corridor
modeling suggests that there were previously two high density corridors between the US and
Mexico allowing jaguar connectivity. However, if the partially constructed border barriers are
completed those jaguar corridors will be lost. Work on nine co-distributed mammals (orders:
Carnivora and Artiodactyla): jaguar (Panthera onca), mountain lion (Puma concolor), ocelot
(Leopardus pardalis), bobcat (Lynx rufus), black bear (Ursus americanus), gray fox (Urocyon
cinereoargenteus), Mexican gray wolf (Canus lupus baileyi) Sonoran pronghorn (Antilocapra
americana sonoriensis), and Bighorn sheep (Ovis canadensis) in the US-Mexico border
ecoregions will continue after the completion of this work.
1
Chapter 1 : Introduction
Establishing connectivity between species habitats is important because it can offset the impacts
of habitat loss and fragmentation. As anthropogenic pressures continue to push species into
isolated habitat patches it is important to understand how to establish wildlife corridors that can
help species migrate, disperse, and maintain adequate gene flow. This chapter introduces the
concept of landscape connectivity, highlights some of the species at risk from loss of
connectivity in the US-Mexico border ecoregions, describes the research objectives, and thesis
organization.
1.1. Landscape Connectivity
1.1.1. Habitat Loss and Fragmentation
Current rates of species extinction are about 1,000 times greater than background rates of
extinction and are likely underestimated (Pimm et al. 2014). The major threats that can drive
worldwide species extinction are habitat degradation (including habitat loss and fragmentation),
overexploitation, invasive species, climate change, pollution, and disease (Groom, Meffe, and
Carroll 2006). Human caused habitat loss and degradation of natural habitat have large negative
effects on biodiversity (Fahrig 2003) and are the main immediate threat to global biodiversity
(Groom, Meffe, and Carroll 2006). Habitat destruction usually leads to habitat fragmentation
which can significantly reduce biodiversity and damage ecosystems (Haddad et al. 2015).
Habitat fragmentation which reduces and isolates natural habitat, despite having differing
effects on species, is “one of the greatest threats to regional and global biodiversity” (Groom,
Meffe, and Carroll 2006, 250). Only species that have small home ranges and whose life history
requirements can be met within a fragment may survive in fragmented landscapes (Groom,
Meffe, and Carroll 2006). The negative effects of habitat fragmentation include edge effects that
2
create barriers and filters that prevent some species from moving, lack of access to important
habitat due to large distance between important habitat patches, and other effects that can
negatively impact animal dispersal and seasonal migration. Additionally, habitat fragmentation
can negatively impact gene flow which can contribute to a greater extinction risk (Riley et al.
2006; Ernest et al. 2014; Wan, Cushman, and Ganey 2018). If species populations become too
genetically isolated in fragmented landscapes, they can suffer from inbreeding and low genetic
variation which in turn can lead to local extinction (Rudnick et al. 2012). Habitat loss and
fragmentation threaten the ability of organisms to stay connected to one another in a landscape
by disrupting the ability of individual organisms and their genes to move across landscapes.
1.1.2. Connectivity
Establishing connectivity between natural areas is one way to help counter the effects of
fragmentation (Groom, Meffe, and Carroll 2006; Haddad et al. 2015). Landscape connectivity is
the amount of movement made possible by the landscape composition (Rudnick et al. 2012). The
movement which is made possible by landscape connectivity can be the movement of genes,
propagules, individuals, and populations of species. Landscape connectivity can facilitate two
important movement processes of species: migration (i.e. seasonal movements) and dispersal
(i.e. to occupy previously unoccupied territory by that species). Connectivity can be further
classified into two types: structural and functional. Structural connectivity relates to physical
landscape characteristics and functional connectivity “describes how well genes, propagules,
individuals, or populations move through the landscape” (Rudnick et al. 2012, 2). We can either
identify and protect currently existing landscape connectivity or reestablish connectivity in
fragmented landscapes to help species gain resilience to future environmental conditions that
may put them at a greater extinction risk. Maintaining or reestablishing landscape connectivity
3
across ecoregions or continents for long periods of time is important for the long-term survival of
species because it allows them to shift their ranges in response to climate change and other long-
term ecological changes; if change happens too fast a species may not be able to adapt (Rudnick
et al. 2012).
There are several modeling techniques that can be used to identify and quantify landscape
connectivity. Cushman et al. (2013) describes several methods which can be used to estimate
resistance including telemetry, landscape genetics, habitat quality, mark-recapture, and
combinations of methods. Potential corridors can be identified using least-cost modeling or
factorial least-cost paths (Cushman et al. 2013). For example, Elliot et al. (2014) used Global
Positioning System (GPS) tracking data from dispersing African lions and resistance surfaces for
calculating the factorial least-cost path network. Circuit theory, centrality analyses, resistant
kernels, and network-based models are some of the other types of ways we can analyze
connectivity (Cushman et al. 2013). Measuring functional connectivity can be achieved more
cost effectively and with greater sample sizes using genetic approaches compared to tracking
individual animals but the concern with genetic studies is that “current genetic patterns may not
reflect the impact of current landscape features” (Rudnick et al. 2012, 6) and “genetic
connectivity may be masked in some instances by local adaptation” (Rudnick et al. 2012, 6).
Comparing resistance surfaces created with movement data to those created through landscape
genetic analyses can be useful to evaluate the robustness of a connectivity model (Cushman et al.
2013). How well a species can travel across a landscape can depend on many factors and is
typically predicted using landscape resistance surfaces created by assigning values to cells in a
raster. Simple estimates of landscape resistance might not be able to adequately predict
resistance to movement of a specialist species or of other species with “species-specific
4
movements” with specific habitat needs (Rudnick et al. 2012, 6). Connectivity modeling needs to
consider both functional and structural connectivity components that can ensure “long-term
habitat shifts” (Rudnick et al. 2012, 6).
1.2. Co-distributed Species in the US-Mexico Border Regions
1.2.1. Border Ecoregions and Biodiversity Hotspots
An ecoregion is defined as a geographic area that has a distinct collection of “natural
communities that share a large majority of their species, ecological dynamics, and similar
environmental conditions and whose ecological interactions are critical for their long-term
persistence” (Dinerstein et al. 1995, 4). The US-Mexico border crosses eight ecoregions
described by Dinerstein et al. (2017): (1) the California Coastal Sage and Chaparral; (2) the
Chihuahuan Desert; (3) the Sierra Madre Occidental Pine-Oak Forests; (4) the Sonoran Desert;
(5) the Tamaulipan Mezquital; (6) the Western Gulf Coastal Grasslands; (7) the California
Montane Chaparral and Woodlands; and (8) the Sierra Madre Oriental Pine-Oak Forests (Figure
1). Within these ecoregions there are five biodiversity conservation hotspots along the border,
the Californias, Sonora Desert, Sky Islands, Big Bend, and Lower Rio Grande Valley (Defenders
of Wildlife 2018). These biodiversity hotspots are areas of high biological diversity which are
threatened with habitat degradation.
1.2.2. Species at Risk Due to Border Dispersal Barriers
Studies have found that the US-Mexico border wall threatens biodiversity. Lasky, Jetz,
and Keitt (2011) performed the first large scale evaluation of the border ecoregions to assess the
potential threats presented by the border dispersal barrier to species that are non-volant terrestrial
vertebrates. They identified the California, Madrean archipelago, and Gulf coast border regions
as having high species richness with high number of species at risk from existing border barriers
5
Figure 1 – Map of study area used for this thesis project showing US-Mexico ecoregions.
6
and from construction of potential new border barriers. They identified several amphibians,
reptiles, and mammals for further study. Peters et al. (2018) found that the border bisects the
geographic ranges of 1,077 native terrestrial and freshwater animals and 429 native plants. Of
these, 62 species are listed as Critically Endangered, Endangered, or Vulnerable by the
International Union for Conservation of Nature (IUCN). Endangered species including
Peninsular bighorn sheep (Ovis canadensis nelsoni), Mexican gray wolf (Canis lupus baileyi)
and Sonoran pronghorn (Antilocapra americana sonoriensis). The study also mentions “if cut off
by a border wall, 17% of the 346 species we analyzed, including jaguar (Panthera onca) and
ocelot (Leopardus pardalis), would have residual US populations covering 20,000 square
kilometers or less” (Peters et al. 2018, 741). Besides border fences and walls, other
anthropogenic threats exist along the border such as artificial night lighting and noise pollution
that can further threaten endangered Texas ocelots by discouraging dispersal of Mexican ocelots
into Texas (Grigione and Mrykalo 2004). GPS tracked bobcats have died due to forced dispersal
during previous border wall construction programs (Gaskill 2011).
1.3. Research Objectives and Thesis Organization
How do old and new border barriers affect connectivity for sensitive borderland
mammals? Broadly speaking, this thesis has the objective to contribute to a better understanding
of co-distributed border ecoregion species connectivity. The aim of this study is to identify areas
on the US-Mexico border that would disrupt cross-border connectivity for one border species
that may be sensitive to new border barriers. To accomplish this goal, core habitat areas and
corridors for the jaguar in the border ecoregions and surrounding ecoregions were identified
(Figure 1). It is important to find out how new border barriers would affect connectivity for
mammals since existing and new border barriers can disrupt migration, dispersal, and
7
adequate gene flow; specifically, in this study for the jaguar (Panthera onca). The findings can
help with management decisions for biodiversity conservation efforts in these border ecoregions.
To the best of my knowledge this is the first study to use the Factorial Least Cost Path (FLCP)
modeling approach for the jaguar (Panthera onca) in this study area. The specific objectives for
completing this study are as follows:
1. Identify appropriate border ecoregion focal species that could be impacted by
border barriers and that would likely serve as umbrella species.
2. Determine the most important environmental and anthropogenic predictor
variables that influence suitable habitat for focal mammals by doing a thorough
literature review.
3. Create habitat suitability maps for the jaguar (Panthera onca) by building species
distribution models.
4. Identify core habitat patches for the jaguar (Panthera onca) by using the resistant
kernel method.
5. Identify potential corridors through the factorial least cost path method.
The remainder of this thesis is organized as follows. Chapter 2 reviews the related
research which others have conducted using the methods outlined above for delineating habitat
patches and modeling potential corridors. Chapter 3 describes the data and methods used in this
study. Chapter 4 describes the results of the study, and Chapter 5 discusses the significance of
these results and concludes by noting the main findings related to the focal species.
8
Chapter 2 : Related Work
This chapter describes work on multispecies connectivity modeling and the techniques
commonly used. The four sections that follow describe multispecies connectivity modeling,
species distribution modeling with presence only data, potential corridor modeling methods, and
focal species selection.
2.1. Multispecies Connectivity Modeling
Multispecies connectivity modeling approaches use different methods to understand how
to best connect habitat patches in fragmented landscapes. Some studies combine methods that
predict suitable habitat and identify potential corridors. There are different algorithms for
predicting suitable habitat, such as the widely used maximum entropy method (MaxEnt), and
potential corridors, such as least cost path methods.
DeMatteo et al. (2017), for example, used MaxEnt to evaluate habitat use, habitat
suitability, and potential species richness for jaguars (Panthera onca), pumas (Puma concolor),
ocelots (Leopardus pardalis), oncillas (Leopardus tigrinus), and bush dogs (Speothos
venaticus) across northern-central Misiones, Argentina. They next determined the optimal
location for primary/secondary corridors that would link the northern-central zones of the Green
Corridor in Misiones and identified areas within these corridors needing priority management.
This study performed a secondary analysis that compared the multispecies corridor results with
the jaguar’s unique requirements and effectively demonstrated that their “multispecies approach
balanced the preferences of all five species and effectively captured areas required by this highly
restricted and endangered carnivore” (DeMatteo et al. 2017, 1). They prefer multispecies
approaches over a single species approach when creating corridors. They used data collected
9
with dogs that could detect scat and a DNA extraction protocol to identify species from the scat.
The optimal locations for primary/secondary corridors were identified by overlaying the five
species-specific least cost path and least cost corridor models. They identified a primary corridor
with a width of 7 km and a secondary corridor with a width of 14 km. They choose the five
aforementioned species because they were identified from scat analysis to be co-distributed in
the study area.
Ersoy, Jorgensen, and Warren (2019) demonstrate an approach that identifies
multispecies networks and compared it to a previous study in Sheffield, UK. They modeled least
cost corridors for four bird, three mammal, and three reptile species. They identified a mix of
landcovers important for supporting a maximum number of species and proposed ways that
current Sheffield green networks can be improved.
Khosravi, Hemami, and Cushman (2018) identified core habitat areas and connectivity
corridors for six species found in the same geographic area in the central Iranian plateau. They
used an ensemble model (EM) of habitat suitability using MaxEnt, GLM, GBM, and the
biomod2 package in R to predict potential habitats of Persian leopard (Panthera pardus
saxicolor), Asiatic cheetah (Acinonyx jubatus venaticus), caracal (Caracal caracal), wild cat
(Felis silvestris), sand cat (Felis margarita), and grey wolf (Canis lupus). They then use resistant
kernels and factorial least-cost path modelling to predict important core habitats and potential
corridors between habitat patches using the UNIversal CORridor network simulator (UNICOR)
(Landguth et al. 2012). This program uses Dijkstra’s algorithm (Dijkstra, 1959) to solve the
single shortest path problem from each species occurrence point to every other occurrence point
(Landguth et al. 2012). The contribution of each core habitat to the connectivity network was
assessed using graph network algorithms in Conefor 2.2 (Saura and Tornė 2009). The next
10
section describes species distribution modeling with presence-only data. These three
aforementioned examples covered in this section show how species corridor modeling can be
combined with species distribution modeling.
2.2. Species Distribution Modeling with Presence-only Data
Species distribution modeling (SDM) has many names including climatic envelope-
modeling, habitat modeling, environmental niche modeling, and ecological niche modeling
(Hijamans and Elith 2017). SDMs can use species occurrences in the form of presence and/or
absence georeferenced locations along with predictor variables (environmental, topographic,
climatic, anthropogenic, etc.) with the goal of predicting the species distribution, abundance,
presence, or occurrence. The outcome one ends up predicting depends on the modeling method.
There are numerous statistical methods that can be used to produce predictions and primarily
depend on the type of data that is available. If you have a well-defined sampling method with
presence and absence data, you should use statistical methods that were developed to handle
those data. If you have information only about a species presence, you should use statistical
methods developed to handle presence-only data; however, it is possible to use a modeling
method that requires absence data so long as you can substitute absences with background data
(Hijamans and Elith 2017). The following review of SDM methods is restricted to presence-only
data because these were the data available for this study.
2.2.1. Maximum Entropy Method (MaxEnt) for modeling species geographic distributions
MaxEnt is a maximum-entropy technique that uses a machine learning algorithm called
sequential-update which is analogous to the AdaBoost algorithm to predict a species’ potential
distribution (π) (Phillips, Dudik, and Schapire 2004). The potential distribution π can exhibit
sampling bias. This can happen for several reasons such as when some locations are sampled
11
more because of easier access to them. Sampling bias will cause the distribution to be weighted
more towards the areas and environmental conditions where species localities were sampled
more. For this reason, π and are best interpreted as a relative index of environmental
suitability (Phillips, Anderson, and Schapire 2006). The unknown distribution π is estimated by
of maximum entropy which depends on the condition that is equal to the empirical
distribution for all of the features fj (i.e. [fj] = [fj]) (Phillips, Dudik, and Schapire 2004). In
reality, is close but not exactly equal to and the regularization equation estimates how close
it is to the empirical value (Phillips, Dudik, and Schapire 2004). The maximum entropy
(MaxEnt) distribution equation in which entropy is maximized is equivalent to the maximum
likelihood Gibbs distribution equation, which is the distribution that minimizes relative entropy
(Kullback-Leiber divergence) (Phillips, Dudik, and Schapire 2004; Phillips, Anderson, and
Schapire 2006).
The MaxEnt program, used for maximum entropy modelling of species geographic
distributions (Phillips, Dudík, Schapire, nd, Ver. 3.4.1), uses the sequential-update algorithm
“that modifies one weight λj at a time” (Phillips, Dudik, and Schapire 2004, 667) and converges
to the optimal maximum entropy distribution. MaxEnt predicts environmental suitability (or
habitat suitability) as a function of the environmental variables by finding the probability
distribution of maximum entropy, which is the distribution that is as close to uniform as possible.
It uses presence-only data (positive examples), such as from museum records and herbarium
collections and does not require information about where species are absent (negative examples)
(Phillips, Dudik, and Schapire 2004; Phillips, Anderson, and Schapire 2006). The MaxEnt
models show probability distributions over pixels but pixels without species records cannot be
12
considered as places where a species is absent because it is only working with data about
presence localities (Phillips, Anderson, and Schapire 2006).
MaxEnt has always been freely available but its source code was not publicly available
because it was owned by AT&T. In December 2016, MaxEnt became open-source, and a new
version (ver. 3.4.0) was released with an update to include a default log-log (cloglog) transform
which can be interpreted as an estimate of the probability of occurrence (Phillips et al. 2017).
This new cloglog transform uses a maximum likelihood exponential model obtained from an
inhomogeneous Poisson process (IPP). Although it is an unlikely assumption, if sampling effort
is unbiased then the new raw MaxEnt output can be interpreted as the relative abundance for the
species and the cloglog transformation converts it to the probability of species presence (Phillips
et al. 2017) with the following equation:
Probability of presence = 1−exp(−exp(H)pλ (z)) (1)
The previous versions of MaxEnt used the logistic transform as the default output with the
following equation to give an estimate of the probability of presence (Phillips and Dudik 2008):
Q (y = 1|z) = e
H
q λ
(x(z))
/ 1+ e
H
q λ
(x(z))
(2)
The logistic transform is still available in the new version. Figure 2 shows a graph comparing
these two equations.
The species datasets used in this study are from presence-only mammal species
occurrence locations recorded from human observations, camera traps, telemetry, and scientific
literature without knowledge of species absence locations. This method was proven to perform
better than other methods that do not require absence data, such as the Genetic Algorithm for
Ruleset Prediction (GARP), which was tested with presence-only species occurrence records for
birds and mammals (Phillips, Dudik, and Schapire 2004; Phillips, Anderson, and Schapire 2006).
13
Figure 2 – Graph showing the distribution produced using the cloglog and logistic transforms.
Source: Phillips et al. (2017).
2.2.2. Generalized Linear Models
In statistics, a generalized linear model (GLM) describes any model with an expected
value (μ) of the response (dependent) variable (Y) with linear explanatory (independent)
variables x1, x2, … xp (Upton and Cook 2014).
The following model described in Upton and Cook (2014), is a link function where the
parameters β1, β2,…, βp are not known:
g(μ) = β0 + β1x1 + … + βpxp (3)
Nelder and Wedderburn (1972) developed this class of GLMs which includes the mathematical
likelihood procedure to fit models based on normal, binomial, Poisson, or gamma distributions.
14
A GLM is a generalization of ordinary least squares regression (OLS) (Hijamans and Elith
2017). Species presence and absence (or background) data can be analyzed using logistic
regression methods commonly used with GLMs (Hijamans and Elith 2017).
Species distribution modeling involves model fitting of species distributions to
environmental variables. The environmental variables are the explanatory (predictor) variables,
and the response variables are the species distributions. In R (R Core Team 2018), GLMs are
fitted using a maximum likelihood procedure and related to response variables through link
functions that allow the variance of each measurement to be a function of the predicted value
(Hijamans and Elith 2017). These models can be fitted using the glm() function in R (Kabacoff
2017). Table 1 shows the different family types that can be used as part of the glm() function in
R to fit GLMs.
Table 1 – List of the eight family types and their associated default link functions to be used
with the glm() function in R. The glm function has the form: glm(formula,
family=familytype(link=linkfunction), data=). Source: Kabacoff (2017).
Family Default Link Function
binomial (link = “logit”)
gaussian (link = “identity”)
Gamma (link = “inverse”)
inverse.gaussian (link = “1/mu^2”)
poisson (link = “log”)
quasi (link = “identity”, variance = “constant”)
quasibinomial (link = “logit”)
quasipoisson (link = “log”)
15
2.2.3. Generalized Boosted Models
The Generalized Boosted Models (GBM), gbm package in R is based on Friedman’s
(2001, 2002) gradient boosting functions such as the gradient boosting machine and Freund and
Shapire’s (1997) AdaBoost algorithm (Ridgeway 2019). The gbm package uses the AdaBoost
exponential loss function and Friedman’s gradient descent algorithm (Ridgeway 2019). The gbm
package supports the Gaussin, AdaBoost, Bernoulli, Laplace, Quantile regression, Cox
Proportional Hazard, Poisson, and Pairwise distributions. Friedman (2001, 2002) describes an
estimation function which estimates f(x) and includes a loss function Ψ(y, F(x)) written in the
regression function below as L(y, F(x)):
F* = arg min E y,xL(y, F(x)) = arg min Ex[Ey(L(y, F(x))) | x] (4)
F F
Sometimes this is written as the regression function that estimates f(x) (Ridgeway 2019):
f
̂ (x) = arg min Ey|x [Ψ(y, f(x))|x] (5)
f(x)
Freedman’s gradient boosting machine contains Friedman’s gradient boost algorithm.
The AdaBoost algorithm is a machine learning algorithm, specifically described by
Freund and Shapire (1997) as an adaptive boosting algorithm with the goal of identifying a
hypothesis that has a low error relative to the distribution over the training examples. The first
hypothesis that is identified is used for obtaining the next weight vector and the process repeats
to generate more weight vectors in an iterative process that adjusts adaptively to errors in weak
hypotheses (Freund and Shapire 1997).
2.2.4. Random Forests Models
Random forests (RF) is an ensemble machine learning algorithm developed by Breiman
(2001). The general process by which RF works is through growing an ensemble of trees,
allowing each tree to vote for the most popular class, and finally the forest chooses the
16
classification with the most votes from all the trees in the forest. To grow the ensembles, random
vectors are produced independent of past random vectors with the same distribution. The
procedures implemented by Breiman’s (2001) RF algorithm are called random forests. Breiman
(2001) provides a definition for an RF classifier as follows:
“A random forest is a classifier consisting of a collection of tree-structured
classifiers {h(x, k ), k = 1,...} where the {k } are independent identically
distributed random vectors and each tree casts a unit vote for the most popular
class at input x” (Breiman 2001, 6).
Random Forests has a well-established history in many disciplines but is less commonly
implemented in ecological studies. There are numerous benefits to implementing an RF method
in ecology including a high classification accuracy, the ability to determine variable importance,
the ability to model complex interactions among predictor variables, it can handle thousands of
input variables, it is robust against overfitting, it can perform regression and classification even if
there are missing data, and it performs well compared to other classifiers. Cutler et al. (2007)
used data from three different species to compare the accuracies of an RF model to those of
classification trees, a logistic regression, and linear discriminant analysis, and found RF
performed the best. Mi et al. (2017) found RF performed better than MaxEnt for predicting rare
species distributions with a limited number of samples over a large area and missing data for
several Asian crane species. Torres et al. (2012) evaluated 11 SDMs and although all generally
had high AUC values (≥ 0.88), the RF model had the highest AUC (0.96) when testing with 30%
of the occurrence locations. Although RF is not as commonly used in ecology compared to other
fields it provides many advantages and studies have proven it performs better than many other
commonly used algorithms.
17
The next section describes resistant kernel modeling, factorial least cost path modeling,
and graph theory. These techniques can be combined with species distribution modeling with
presence-only data as well.
2.3. Species Connectivity Modeling
The resistant kernel, factorial least cost path, and graph theory modeling techniques
discussed in this section can be used to model species connectivity for either single or multiple
species.
2.3.1. Resistant kernel modeling
The resistant kernel modeling (or resistant kernel estimator) approach calculates the
expected density of dispersing individual animals in each pixel on a landscape (Cushman, Lewis,
and Landguth 2014). The resistant-kernel estimator method is a hybrid between the kernel
estimator method and least-cost paths which use resistance surfaces (Compton et al. 2007). To
calculate the resistant kernel estimator a least cost kernel for each cell is first calculated. Each
cell represents a source for dispersers, for example a vernal pool, and then all kernels in each cell
are added together (Compton et al. 2007). Worton (1989) described the fixed kernel and adaptive
kernel methods for estimating the utilization distribution in species home range analyses. Patch
isolation can be influenced by distance and the type of land cover matrix (Ricketts 2001).
Ricketts (2001) described a maximum likelihood method to estimate relative resistances of two
types of landcover for butterfly movement. Resistant surfaces are typically created by assigning
resistance values to land cover types, while the least-cost path (LCP) method finds the shortest
(less) costly distance between two origins (Compton et al. 2007). A variation to the LCP method
involves using a multidirectional approach called a least-cost kernel surface because it measures
the functional distance from one cell (source of dispersers) to all other cells in a landscape,
18
giving a probability of dispersal (Compton et al. 2007). A cost is assigned to each cover type
which represents a cost an animal incurs when moving across that surface. The lower the cost the
easier it will be for an animal to move across that landscape surface.
The UNIversal CORridor network simulator (UNICOR; Landguth et al. 2012) program is
used for identifying species connectivity and corridors. The UNICOR program uses a modified
version of Dijkstra’s algorithm which computes the single shortest path from all locations on a
landscape that are specified (Landguth et al. 2012). Dijkstra’s algorithm solves two problems, it
constructs a tree, which is a graph, that has only one path between every two nodes and finds a
path of shortest length between two nodes (Dijkstra 1959). UNICOR version 2.0 includes the
resistant kernel technique used by Compton et al. (2007) which predicts habitat connectivity and
corridor paths using a resistance surface as an input. There are three advantages to using resistant
kernels: 1) They predict and map expected movement rates for every pixel in the study area; 2)
the scale dependency of a species dispersal ability can be used to understand the effect of
landscape change and fragmentation; and 3) they simulate and map different geographic extents
using a combination of species (Landguth et al. 2016). The resistant kernel method used in
UNICOR uses the modified version of Dijkstra’s algorithm to compute the least-cost dispersal
around every specified source cell to create expected density surfaces for dispersing individuals
at any location on a landscape (Landguth et al. 2016). This is accomplished by creating surfaces
of cost to movement for every specified source which is then transformed to indicate a scale
from zero to one (Landguth et al. 2016).
2.3.2. Factorial least cost path modeling
A major limitation to traditional least-cost paths and corridor analysis is that there are
only two locations considered (i.e. the source and the destination). Factorial least cost path
19
analysis is a type of least cost path method that relies on a spatially synoptic view to understand
connectivity (Cushman, Lewis, and Landguth 2014). By adopting a synoptic view one can
understand connectivity from multiple locations to all other locations simultaneously. In this
way, factorial least cost path analysis helps us to better understand connectivity by calculating
least cost paths for “thousands or millions of combinations of locations” across the study area
(Cushman, Lewis, and Landguth 2014, 845). Figure 3 shows a factorial least cost path analysis
where densities of all paths are shown from blue to red with red representing the lowest cost
paths (Rudnick et al. 2012).
Figure 3 – Example factorial least-cost path analysis. Source: Rudnick et al. (2012)
Both the resistant kernel and factorial least cost path modeling methods can be modeled
with the UNICOR software described above. This program can be used for factorial least cost
path modeling to calculate the least cost paths for all source pair locations and to create several
least cost paths (Cushman, Lewis, and Landguth 2014). The corridor strength is indicated by the
20
higher grid cell values in the raster that is created. Each cell value is calculated by summing the
number of cost paths that cross the cell (Cushman, Lewis, and Landguth 2014).
2.3.3. Graph theory
Graph theory is a branch of geometry that originated in 1735 when the Swiss
mathematician Leonard Euler solved the Königsberg bridge problem that involved finding a path
over seven bridges that traversed a river without crossing any bridge twice (Hosch 2010). Graph
theory is now considered a branch of mathematics that is concerned primarily with the statistical
description of static networks (Proulx, Promislow, and Phillips 2005). The modern version of the
theorem proved by Euler can be stated as follows: “If there is a path along edges of a multigraph
that traverses each edge once and only once, then there exist at most two vertices of odd degree;
furthermore, if the path begins and ends at the same vertex, then no vertices will have odd
degree” (Hosch 2010, 108). In graph theory the term graph refers to a set of vertices which are
called points or nodes and edges which are the lines connecting vertices (Hosch 2010). A
multigraph is one that has any two nodes connected by more than one line and the graph is
complete when each one of its nodes is connected to every other node by a line (Hosch 2010).
There are also paths in graph theory which can take on different routes in a graph. There are
different types of paths (e.g. Eulerian circuits) that can be defined in graphs. Graph theory has
applications in many fields ranging from sociology to evolution. It has been widely used in
biological networks (Proulx, Promislow, and Phillips 2005). It is useful for problems related to
finding optimal paths in a graph, given different criteria, and efficient algorithms (Hosch 2010).
Graph theory has been applied to connectivity analyses in conservation biology. In this
case a graph represents a landscape made up of nodes, which could be habitat patches, and lines
connecting pairs of nodes, which could represent dispersal (Urban and Keitt 2001).
21
Pascual-Hortal and Saura (2008) used a graph-based approach to determine functionally
connected forest areas within a species distribution and identify the forest habitat areas that are
more important for maintaining connectivity for capercaillie. With their results they provided
recommendations that could help with the conservation of capercaillie.
Urban and Keitt (2001) used the minimum spanning tree, a graph construct, to understand
the relative importance of habitat patches for the Mexican Spotted Owl. They used data from
Keitt et al. 1995, USDI Fish and Wildlife Service 1995, to demonstrate the application of
minimum spanning trees and found that using this method, large core owl populations and their
dispersal routes between core habitats were well maintained.
Another study used a graphed-based approach to understand landscape connectivity
indices for prioritizing habitat patches and corridors (Pascual-Hortal and Saura 2006). This study
compared 10 graph-based connectivity indices which included a new index, the integral index of
connectivity (IIC), with seven different habitat changes, such as habitat patch loss and corridor
loss, to see how well each one could identify important landscape elements. They found
limitations to existing indices and that the new index was more appropriate because it performed
consistently with different habitat changes (Pascual-Hortal and Saura 2006). In a later study,
Saura and Pascual-Hortal (2007) looked at the response from nine connectivity indices again
including the IIC. This time the IIC came in second place and a new index, the probability of
connectivity (PC), placed first, performing consistently across the 13 landscape properties
measured.
Conefor Sensinode 2.2 (CS22) (Saura and Torne 2009) is a free software package that is
used for quantifying the importance of habitat patches. This program is based on graph theory
and works well with geographic information systems (GIS). It can process thousands of nodes
22
and can be used with any standard computer. The time required to complete the analysis will
depend on the computer and number of nodes that need to be processed (Saura and Tornė 2009).
The next section describes the species which was selected from the numerous species
found to be at risk of border barriers from prior studies.
2.4. Focal Species Selection
Focal species were selected based on available data from various sources and databases
including species occurrence data from literature, museums, public organizations, human
observations, camera traps, and genetic samples, that indicated their presence occurred across the
US-Mexico border and from coast to coast. Menke (2008) created a mountain lion least cost path
model (LCP) for locating potential corridors in New Mexico for this felid. The final model found
overlap between the mountain lion LCP model of potential corridors and habitat for gray wolf,
jaguar, swift fox, and kit fox. This thesis project focuses on the jaguar (Panthera onca) and
covers a much larger geographic extent than Menke (2008) did. The background for this highly
endangered species and prior modeling are described in the two subsections below that conclude
this chapter.
2.4.1. Jaguar (Panthera onca) Background
The jaguar (Panthera onca) is a keystone species whose historic range once stretched
from southwestern United States to southern Argentina, but their range is much smaller today
(Seymour 1989) (Figure 4). They are the largest felid predator and the only remaining
representative of the genus Panthera in the American hemisphere. Presently the jaguar
subspecies classification is unclear. Eight subspecies have been recognized by Pocock (1939)
and Seymour (1989) but morphological and genetic analysis do not indicate that there are
23
separate subspecies (Larson 1997, Eizirik et al. 2001, Ruiz-Garcia et al. 2006) and Larson (1997)
recommends captive jaguars should be managed as a single species.
Figure 4 – Map showing approximate current and historic geographic range of the jaguar
and the range limits based on found fossils from late Pleistocene and mid Pleistocene. Source:
Seymour (1989).
The IUCN Red List classifies jaguars as Near Threatened and they have a decreasing
global population trend (Quigley et al. 2017). Major threats to jaguars vary by geographic region
and include habitat loss and fragmentation, retaliatory killings due to livestock depredation,
illegal body part trade, trophy hunting, and human competition for wild meat (Quigley et al.
2017). Threats to jaguar survival have resulted in severely fragmented populations, a loss of
habitat connectivity at local and regional scales, a 49% loss of their historic geographic range,
24
and a 21% loss of their important prey species (white-lipped peccary) (Quigley et al. 2017).
Morcatty et al. (2020) found the illegal body part trade for several cat species including the
jaguar are connected to Chinese-led development in Central and South America. They found
jaguar seizers of body parts to have increased between 2012 – 2018 and found jaguar canines
were the most common body part. The main threats to jaguars in northern Mexico include
“illegal predator control, illegal hunting, depletion of prey species, and habitat degradation and
fragmentation” (Rosas-Rosas and Valdez 2010, 366). In the southwestern United States, the
major threats appear to have been livestock settlers. According to Brown (1983), kill data from
jaguars indicates that they were eliminated by livestock operators and predator control agents
which is concurrent with human settlement and development of the livestock industry. Reports
and photographs of jaguars killed in Arizona and New Mexico have been documented as early as
1986 (Brown and López-González, 2000). There are records of jaguars occurring in the
southwest US including California, Texas, Arizona, and New Mexico (Brown 1983).
Jaguars have been reported to prey on more than 85 different species with their favorite
prey reported as peccaries, capybaras, pacas, agoutis, armadillos, caimans, and turtles (Seymour
1989). In two specific case studies in northeastern Sonora, Mexico, jaguar prey species have
been identified but it is not possible to assume that such findings would apply to the entire state
of Sonora. In a study area (about 400 km
2
) in northeastern Sonora, Mexico, where cattle ranching
dominates, jaguars were found to prey mainly on cattle but they also preyed on white-tailed deer,
lagomorphs, collard peccary, and coati (Rosas-Rosas, Bender, and Valdez 2008). This study
lasted six years and comprised just 27 scat samples that were analyzed by microscope (Rosas-
Rosas, Bender, and Valdez 2008). This same study attributed confirmed killings of calfs from
ranch records and field surveys to three individual jaguars over a period of six years, and yet
25
there was still a very high calf survival rate averaging 93.3%, since most calves were sold
(5728/6136) (Rosas-Rosas, Bender, and Valdez 2008). Cassaigne et al. (2016) investigated the
jaguar and puma diet in a study area of about 700 km
2
which partially overlapped the study area
of Rosas-Rosas, Bender, and Valdez (2008) using molecular analysis of opportunistically
collected scat samples (2012-2013) and kill sites (2011-2013) from two collard jaguars and
seven collard pumas. The study found that a variety of small prey weighing less than 15 kg made
up the majority of jaguar kill sites (52%) (Cassaigne et al. 2016). Despite a low jaguar density in
the study area and just five jaguar scats and kill data for two collard jaguars, they identified a
variety of prey species which included birds, deer, calves, coati, and skunks (Cassaigne et al.
2016).
Important native prey for jaguars along the border have been suggested to be Cous white-
tailed deer, javelina (a.k.a. collared peccary), coati, opossum, and other medium sized mammals
(Brown and López-González 2001). Other examples of borderland jaguar prey have been
documented as horse, cattle, elk carrion, white-tailed deer, white-nosed coati, javelina, desert
tortoise, frogs, and skunk (Brown and López-González 2001). Potential native prey species for
jaguars in the southwestern United States (Arizona and New Mexico) include white-tailed deer,
collared peccary, mule deer, coatis, skunks, raccoons, and jackrabbits (Hatten, Averill-Murray,
and Van Pelt 2005). Potential domestic prey includes animals such as livestock and horse
(Hatten, Averill-Murray, and van Pelt 2005).
2.4.2. Jaguar Modeling Applications
Tôrres et al. (2012) evaluated 11 modeling methods including MaxEnt to predict species
distributions and test whether species distribution modeling in general could provide estimates of
jaguar population densities in the Neotropics. They used jaguar occurrences from scientific
26
books, research papers, online databases, and field records and complied 1,409 spatially unique
jaguar records. The cell size was 0.0417 degrees which is about 4 km
2
because SDM accuracy is
limited by “the quality of distribution data and the available climatic and topographic data sets”
(Tôrres et al. 2012, 617). The predictor variables used to evaluate all models were precipitation
of the coldest quarter, precipitation of the warmest quarter, precipitation seasonality (coefficient
of variation), annual precipitation, mean temperature of the wettest quarter, mean temperature of
the driest quarter, maximum temperature of the warmest period, minimum temperature of the
coldest period, temperature seasonality (coefficient of variation), annual mean temperature,
altitude and slope. The number of iterations used was 1,000 and they used pseudo-absences. All
the models they evaluated had high AUC values (≥ 0.88), but the Random Forest (RF) model had
the highest value (0.96) when testing a subset of 30% of the occurrence locations (Tôrres et al.
2012).
Rosas-Rosas, Bender, and Valdez (2010) found that jaguar cattle kill sites in northeastern
Sonora (study site ~ 400 km
2
) were positively associated with oak, semitropical thornscrub, and
xeric thronscrub and negatively associated with upland mesquite. They also found a positive
association of cattle kills sites with proximity to water and roads (Table 2). Other potentially
important variables found from the type of vegetation recorded at associated jaguar kill sites in
northeastern Sonora, Mexico (400 km
2
)
include semitropical thornscrub, oak patches, and
tropical deciduous forest (Rosas-Rosas and Valdez 2010). Natural prey of jaguar also prefers
these types of habitats including white-tailed deer, coatimundi, and collard peccaries (Rosas-
Rosas and Valdez 2010).
Table 2 lists important predictor variables found, study information, habitat modeling
algorithms used, the input predictor variables, and source for the jaguar modeling studies. Many
27
of these studies use a variety of different SDM algorithms individually or in combination which
include MaxEnt, GARP, ENFA, MD, SVM, and EM’s.
Important vegetation for jaguars in the borderlands is described in other non-modeling
literature from direct observation of the biotic community from jaguars reportedly killed or
photographed and include Sinaloan thornscrub (mostly found in Sinaloa), Madrean evergreen
woodland (including woodlands of oak and pine), chaparral, and shrub-invaded semidesert
grasslands (Brown and López-González 2001). In Arizona and New Mexico, it seems that
montane conifer forest and pinon-juniper woodland might also be important for jaguars (Brown
and López-González 2001). Other habitat that has been associated with individually studied
jaguars with camera traps in the southwestern US includes Sonoran lowland desert, Sonoran
desert scrub, mesquite grassland, Madrean oak woodland, and pine-oak woodland (McCain and
Childs 2008).
There have been potential corridor models created for jaguars of varying scales, including
continental, countrywide, and finer scales. Rabinowitz and Zeller (2010) created a model of
potential corridors for the jaguar at a continental scale (or range-wide scale) using GIS and
expert opinion to create a cost surface and identify least cost corridors connecting the 90 known
jaguar populations from northern Mexico to northern Argentina. They identified 182 potential
corridors ranging between 3 to 1,607 km in length. They used six datasets for creating a
28
Table 2 – Jaguar habitat modeling studies
Important
variables found
Study information Programs used Input predictor
variables
Source
Oak, semitropical
thornscrub, and xeric
thronscrub, proximity to
water, proximity to roads
(positive) and upland
mesquite (negative)
Location: northeastern
Sonora, Mexico
Extent: 400 km
2
Years: 1999-2004
Occurrences: 45
confirmed jaguar cattle
kill-sites
Accuracy: unknown
MaxEnt
Best model found:
vegetation type, distance
to permanent water type,
and distance to roads.
Elevation, vegetation
cover type (primary oak
forest, semitropical
thronscrub, xeric
thornscrub, disturbed
semitropical thornscrub,
tropical deciduous forest,
mesquite, and disturbed
oak forest), distance to
roads, distance to
permanent water sources
(perennial streams and
rivers, springs, ponds,
and permanent water
developments), elevation,
slope, and aspect.
Resolution: unknown
Rosas-Rosas, Bender,
and Valdez (2010).
Elevation between 1,220
and 1,829 m with scrub
grasslands, elevation and
biomes, distance to water
within 10 km or
perennial or intermittent
water sources (caution:
Euclidean distances
measured in mountainous
terrain), and terrain
ruggedness (intermediate
to extreme ruggedness)
Location: Arizona
Extent: 295,234 km
2
Years: 1901-2001
Occurrences: 57 jaguar
sightings alive or dead
Accuracy: < 8 km (<1.7
km to 8 km)
ArcGIS – habitat
suitability map
Coarse-scaled: vegetation
biomes (ecosystems) and
series (defined by
dominant or
characteristic species),
elevation and terrain
ruggedness, proximity to
perennial or intermittent
water sources (streams,
rivers, lakes, or springs),
and human density.
Resolution: 30 m DEM
resampled to 1 km
2
cells.
Hatten, Averill-
Murray, and van Pelt
(2005).
Important variables not
possible to obtain from
discriminant analysis.
Important variables
obtained from
histograms: precipitation,
elevation, slope, and
temperature, shrubland,
grassland, and forest
(females: shrubland,
deciduous broadleaf
forest, and grassland; but
also, needleleaf forest
and mixed forest).
Location: Arizona, New
Mexico, Texas
panhandle, Sonora, and
Chihuahua.
Extent: approximately
1.1 million km
2
Years: ~1900 -2003
Occurrences: 142 (100
male, 42 female) jaguar
occurrence records from
museums, photos,
verified kills,
universities, conservation
organizations, interviews
with residents in Mexico.
Accuracy: 25km
2
Genetic Algorithm for
Rule Set Production
(GARP)
3 models:
Males + Females
Males
Females
20 environmental layers
for climate and
landscape, including
temperature, wetness,
vapor pressure, frost
days, snow
accumulation, radiation,
soil type, elevation,
aspect, slope, compound
topographic index, water
flow, and runoff
Resolution: resampled to
25km
2
pixels
Boydston and Lopez-
Gonzalez (2005).
From ENFA and
MaxEnt:
Prefers: Tropical rain
forest, prey, and
regularly flooded
vegetation,
Avoids: Higher
Elevations, arid
vegetation, and grassland
Location: Mexico
Extent: 1,972,550 km
2
Years: Calibration
dataset: 1990 – 2008
Evaluation dataset:
2000 – 2008
Occurrences: Two data
sets: 1) Calibration:
197 (1 data point per
locality, 1 record of
presence for each 5 km
2
)
occurrences from
literature, CONABIO,
Ensemble Model (EM)
Ecological Niche Factor
Analysis (ENFA)
Mahalanobis distance
(MD)
MaxEnt
6 environmental and
anthropogenic factors
assumed to be important
Dry forest, tropical rain
forest, other forest, arid
vegetation, grassland,
regularly flooded
vegetation, agriculture,
anthropogenic
perturbation: roads
(distance) and human
population density,
Rodríguez-Soto et al.
(2011).
29
GBIF, MaNIS, and
Jaguar GIS. 2)
Evaluation: 104 from
VHF-locations for 5
collard adult female
jaguars in 5 different
regions, camera-traps
from 8 additional
regions.
Accuracy: unknown
elevation, slope, prey-
species richness.
Resolution: All layers
resampled to 1-km
2
cell
size.
From MaxEnt: positive
relation to risk of attack
by jaguar: tree cover
percentage, percentage of
animals in free grazing
areas, and altitude
Negative correlation to
risk of attack by jaguar:
arid vegetation
Location: Mexico
Extent: 1,953,162 km
2
Years: 1990 – 2010
Occurrences: 222 from
felid attacks on livestock
(jaguar – 152, puma –
70)
Accuracy: unknown
Jaguar: MaxEnt, GARP-
with best subsets, and
Support Vector Machines
(SVM)
Puma: MaxEnt,
Environmental distance,
ENFA, GARP- with best
subsets, and GARP –
single run
Landscape, livestock
management, and
anthropogenic:
Topographic: altitude
and slope
Vegetative associations
of the National Forest
Inventory: forest
(conifers, oaks and
riverside vegetation), dry
forest, rainforests, arid
vegetation, underwater
vegetation, and
agriculture
Percent tree cover
Livestock density,
percentage of free
grazing
Human population
density, and
distance to paved roads.
1 datum per pixel
Resolution: 1 -km
2
Zarco-González et al.
(2013).
permeability matrix which was used to create a cost surface to be used in a least cost path
analysis (Table 3).
Other studies have focused on countrywide scale corridor studies such as in Mexico
(Rodrí guez-Soto, Monroy-Vilchis, and Zarco-Gonzá lez 2013), or a biome-scale in Brazil
(Morato et al. 2014), and for smaller study areas within countries such as Nicaragua (Zeller et al.
2011) and Argentina (DeMatteo et al. 2017). Rodrí guez-Soto, Monroy-Vilchis, and Zarco-
Gonzá lez (2013) used a previously created ensemble model (Rodrí guez-Soto et al. 2011, Figure
2) of the potential distribution of the jaguar in Mexico and identified jaguar management and
conservation areas (JCMA) which they used for identifying potential jaguar corridors between
them in their 2013 study (Table 3). The variables used for the ensemble model in Rodrí guez-Soto
et al. (2011) are listed in Table 2. This ensemble model of potential jaguar distribution in Mexico
30
was used as a cost raster or permeability map to create the potential jaguar corridors in Mexico
(Rodrí guez-Soto, Monroy-Vilchis, and Zarco-Gonzá lez 2013).
Rodrí guez-Soto, Monroy-Vilchis, and Zarco-Gonzá lez (2013) used the inverse of the
habitat suitability map previously created from an ensemble model (Rodríguez-Soto et al. 2011)
as the permeability (cost raster) map. They identified JCMAs as habitat patches using the
Corridor Designer in ArcGIS. The permeability map and habitat patches were used as inputs to
create the potential jaguar corridors for Mexico by calculating the cost-distance of each pixel
using Corridor Designer in ArcGIS. In contrast, Rabinowitz and Zeller (2010) assigned cost
values to landscape layer pixels obtained from 15 jaguar experts. The cost values ranged from 0,
which indicates no cost to jaguar movement to a value of 10, which indicates a high cost for
jaguar movement as shown in Table 4. A final cost surface or permeability matrix was created by
using the Raster Calculator in ArcGIS and reclassifying the output raster. They created
movement cost grids from each of the 90 Jaguar Conservation Units (JCU) and delineated least
cost corridors using the Corridor function from the Spatial Analysist toolbox in ArcGIS
(Rabinowitz and Zeller 2010).
31
Table 3 – Data layers used by different studies for creating jaguar permeability matrices
Study / Notes Input layers Dataset name and
scale
Year of data Data Source
Rabinowitz and
Zeller (2010) datasets
for creating the jaguar
permeability matrix /
Extent: Continental
(northern Mexico to
northern Argentina)
Elevation Global 30 arc-second
elevation data set
1 km resolution
1996 Center for earth resources
observation and science
(EROS)
Landcover type Global land cover 2000
1 km resolution
1999–2000 Global land cover 2000
Percent tree and shrub
cover
Continuous vegetation
fields
500 m resolution
2000 Global land cover facility
Population settlements Vector map level 0
population settlements
1:1,000,000 scale
1960s–1990s National imagery and
mapping agency (NIMA)
Human population density Gridded population of the
world v3
2.5 min resolution
2000 Center for international
earth science information
network
(CIESIN)
Roads Vector map level 0 roads
1:1,000,000 scale
1960s–1990s National imagery and
mapping agency (NIMA)
Rodríguez-Soto,
Monroy-Vilchis, and
Zarco-Gonz á lez
(2013) variables
Extent: country of
Mexico
* variables for layers
and datasets are from
Table 1 in Rodríguez-
Soto, Monroy-Vilchis,
and Zarco-Gonzá lez
(2013), but see
variables listed in
Table 2 for Rodríguez-
Soto et al. (2011).
*Vegetation cover National Forest Inventory
1:250 000
2001 SEMARNAT et al.
*Human disturbance:
Agriculture, Road network,
and Human population
density
Agriculture, Road
network, and Human
population density
2001, 2008, and 2005 SEMARNAT et al.,
CONABIO, and FAO
*Protected Natural Areas
(PNA)
Protected Natural Areas 2007 Consejo Nacional de
Areas Protegidas
(CONAP)
*Elevation Digital elevation model 2007 U.S. Geological Survey
(USGS)
32
Table 4 – Cost values (cost to jaguar movement) obtained from 15 jaguar experts assigned to
various landscape layers, from Rabinowitz and Zeller (2010).
Landscape Layer Cost Value Landscape Layer Cost Value
Land cover type Tree and Shrub Cover (%)
Tree Cover, broadleaved,
evergreen
0 0 – 10 9
Tree Cover, broadleaved,
deciduous
0 10- 20 7
Tree Cover, needle- leaved,
evergreen
1 20 - 40 5
Tree Cover, mixed leaf Type 0 40 - 60 2
Tree Cover, regularly flooded,
fresh water
2 60 - 80 0
Tree Cover, regularly flooded,
saline water
2 80 - 100 0
Mosaic: Tree cover/other
natural vegetation
1 Human Population Density
(people/Km2)
Shrub Cover, evergreen 2 0-20 1
Shrub Cover, deciduous 3 20-40 5
Herbaceous Cover 5 40-80 7
Sparse herbaceous or sparse
shrub cover
6 80-160 9
Regularly flooded shrub and/or
herbaceous cover
5 160-320 10
Cultivated and managed areas 8 >320 N/A
Mosaic: Cropland/Tree Cover/
Other natural Vegetation
5 Elevation (m)
Mosaic: Cropland/Shrub or
grass cover
7 0 – 1000 0
Bare areas 8 1000–2000 2
Water Bodies 6 2000–3000 7
Snow and Ice N/A 3000–5000 10
Artificial surfaces and
associated areas
10 >5000 N/A
Distance from Roads (Km) Distance from Settlements (Km)
0 to 2 7 0–2 8
2 to 4 4 2–4 5
4 to 8 2 4–8 4
80 to 160 1 8-16 1
> 16 0 > 16 0
33
Chapter 3 : Methods
3.1. Research Design
The overarching research question that this study hopes to answer is: How will new border
barriers affect connectivity for jaguars? By answering this question, it will be possible to identify
dispersal corridors which may be disrupted by new border barriers. Additionally, these results
can be compared to the connectivity for the eight co-distributed charismatic mammals (orders:
Carnivora and Artiodactyla): mountain lion (Puma concolor), ocelot (Leopardus pardalis),
bobcat (Lynx rufus), black bear (Ursus americanus), gray fox (Urocyon cinereoargenteus),
Mexican gray wolf (Canus lupus baileyi) Sonoran pronghorn (Antilocapra americana
sonoriensis), and Bighorn sheep (Ovis canadensis.) All data were processed using ArcGIS Pro
(Version 2.3.2, Esri Redlands, CA) and Google Earth Engine 2019 (GEE, Google, Mountain
View, CA). All modeling and analysis tasks were completed using two software programs: R
(Version 3.6.0 (2019-04-26) -- "Planting of a Tree"), and UNICOR (Version 2.0, 2016).
The general methodology which was employed was inspired by and constitutes a
variation to that used by Khosravi, Hemami, and Cushman (2018). This earlier study focused on
a small geographic region unlike this thesis project which focused on a greater geographic extent
and therefore required a revised methodology. Different species data points cover different areas
along the US-Mexico border ecoregions which can help meet the intended goal of understanding
how new border barriers would affect connectivity for the jaguar and several border mammals
that may be impacted by new border barriers. The realization that the occurrence points for
several individual species span the US-Mexico border helped define a study area. For example,
jaguar records are found in the above mentioned US and Mexican states, but they are also found
in Texas and northeastern Mexico. The desire to understand how the connectivity of border
34
mammals would be affected by border barriers motivated this study which followed the US-
Mexico border from coast to coast.
Important predictor variables were identified from scientific literature for the jaguar and
several focal mammals from relevant study sites, in, or as close as possible to the borderlands of
US and Mexico. This approach provided a greater probability of selecting appropriate predictor
variables for use in the chosen study area. However, literature was also reviewed for focal
species from other geographic areas, such as Argentina, Brazil, and other central and south
American countries. It is important to select appropriate covariates to build habitat suitability
maps for the chosen species. Building habitat suitability maps helps to identify core habitat
patches for specific species by using the resistant kernel method. Potential corridors can be
identified using the factorial least cost path method, and the importance of patches and corridors
can be assessed using graph network algorithms.
3.2. Focal Species Datasets
Several species of concern whose dispersal, migration, and gene flow might be impacted,
have been identified and are listed in Chapter 1 and described in Chapter 2. The focal species are
jaguar (Panthera onca), mountain lion (Puma concolor), ocelot (Leopardus pardalis), bobcat
(Lynx rufus), black bear (Ursus americanus), gray fox (Urocyon cinereoargenteus), Mexican
gray wolf (Canus lupus baileyi) Sonoran pronghorn (Antilocapra americana sonoriensis), and
Bighorn sheep (Ovis canadensis). These species represent the mammalian orders, Carnivora and
Artiodactyla.
The occurrence localities that were used for the jaguar and other focal species were
obtained from scientific literature, museums, universities, institutions, online databases, and a
conservation organization. This includes occurrence data with GPS coordinates from
35
photographs, camera trap image/video data, scat, hair, tracks, reports, preserved specimens,
species mortalities, genetic sampling location data, human observations, hunter harvested
individuals, road kills, and GPS telemetry. Three main online databases were used to obtain most
of the occurrence data: (1) the Global Biodiversity Information Facility (GBIF); (2) VertNet; and
(3) the Jaguar Observations Database (JOD). Occurrence records were cleaned using ArcGIS Pro
by removing occurrence points found in the ocean, records with high uncertainties, fossils, zoo
records, duplicate records, and by examining metadata files. Final cleaned species records were
saved in separate AcrGIS Pro geodatabases including for the jaguar which was used in this thesis
project.
3.3. Focal Species Covariate Datasets
There are many studies that advocate for multi-scale optimization when modeling species
habitats because species may respond differently to predictor variables at different spatial scales.
Multi-scale habitat modeling refers to the scale(s) that is(are) important to an individual
organism because of how the individual interacts with the environment. Wiens (1989) mentions
that when studies asking the same questions are conducted at different scales their findings are
not always consistent. Different species may respond differently at different scales which can
lead to issues when designing nature reserves. As Wiens (1989, 385) mentions, “the very
foundation of geography is scaling”. The term “scale” is context specific but regarding spatial
analysis and modeling, O’Sullivan and Perry (2013) define it as a term that describes spatial
grain, spatial extent, temporal grain, and temporal extent. Spatial grain refers to the resolution
used to collect the data which the data (e.g. pixel, or cell size), temporal grain refers to the
frequency used to collect the data, spatial extent refers to the total area that the dataset covers,
and temporal extent is the date range over which the data were collected (O’Sullivan and Perry
36
2013). When considering spatial scale in ecological studies “expanding the extent of a study
usually also entails enlarging the grain. The enhanced ability to detect broad-scale patterns
carries the cost of a loss of resolution of fine-scale details” (Wiens 1989, 387). When spatial
scale is increased temporal scale should also be increased since processes will operate at slower
rates. The acquisition of occurrence data for a broad temporal range supports the large spatial
extent chosen for this study. Finally, Wiens (1989) recommends that ecologists should adopt a
multiscale approach to species studies.
Several studies have found that multi-scale species models outperform single-scale
species models in terms of predictive power (e.g., McGarigal et al. 2016; McGarigal, Zeller, and
Cushman 2016; Timm et al. 2016; Wan et al. 2017) and provide greater predictive capacity
(Timm et al. 2016). For models to be robust they should include a full set of “covariates relevant
to habitat selection by the species as is possible” and at spatial scales important to the focal
species (Timm et al. 2016, 1210). This can be accomplished by varying the bandwidth for each
covariate and each location in ArcGIS Pro. Unfortunately scale optimization procedures are very
rarely used in habitat modeling studies (McGarigal et al. 2016). Timm et al. (2016, 1210)
recommends using multi-scale models to reduce “investigator-driven bias”. This author
determined the covariates from literature reviews and discussions with species experts. This
thesis project also used literature reviews to determine important variables and understand the
jaguar’s ecology. Feedback for variable selection was received from Dr. Cushman. This
approach provided a full list of the best possible variables to use and the inclusion of species-
important scale optimization provided a more robust modeling framework.
All covariate datasets required varying levels of preprocessing with ArcGIS Pro and GEE
prior to implementing a multi-scale optimization method. Figure 5 shows the average human
37
population density covariate layer for the period 2000-2020. It was created using ArcGIS Pro
Cell Statistics and 5 input rasters for the years 2000, 2005, 2010, 2015, and 2020. A total of 52
covariate layers were ultimately processed.
Figure 5 – Average human population density for the period 2000 - 2020. Average number of
persons per 30m pixel, 2000 – 2020. Source: GPW ver. 4 rev. 11.
Eight anthropogenic, two climate, 15 ecoregion, 10 land cover, six topographic, seven
vegetation, and four water covariate layers were further processed at five different scales, to
generate a total of 260 covariates. Dr. Wan’s python script was used for multi-scale optimization.
0 – 0.000001
38
Results for some of the 260 covariates processed are shown in Figures 6 to 9. The univariate
scaling of the compound topographic index (CTI) and mean annual temperature layers at 16 km
are shown in Figures 6 and 7. Results for the univariate scaling of slope position and roughness
layers at 8 km are shown in Figures 8 and 9. Roughness was rescaled from 1 to 10 using a log
transform function for proper visualization of landscape features with low (1) to high (10)
surface roughness (Figure 9).
The slope position, roughness, and compound topographic index were processed using
ArcGIS Pro and the ArcGIS Geomorphometry and Gradient Metrics toolbox (Evan et al. 2014).
Slope position calculates the scalable slope position by subtracting a focal mean raster from the
original elevation raster based on Berry’s (2002) methodology for calculating a surface area
ratio. Slope position values range from low (negative) to high (positive). Evan et al. (2014)
describe roughness as a representation of a continuous raster within a specified window and this
metric is based on the research conducted by Riley, DeGloria and Elliot (1999) and Blaszczynski
(1997). Roughness (or ruggedness) is an estimate of terrain heterogeneity which is an important
variable for predicting species potential habitats and densities (Riley, DeGloria, and Elliot 1999).
It is related to the Terrain Ruggedness Index (TRI) which describes level ground to extremely
rugged terrain. The CTI is a steady state wetness index and a function of the slope and upstream
contributing area per unit width orthogonal to the flow direction. CTI is described by the
following equation:
CTI = ln (As / (tan (β)) (6)
where As is equal to the area value calculated as flow accumulation plus 1 multiplied by the pixel
area in square meters. β is the slope expressed in radians (Evan et al. 2014).
39
F
Figure 6 –16 km univariate scaling for the compound topographic index.
Figure 7 – 16 km univariate scaling for annual mean temperature.
°C
-2
10
12
15
16
18
21
23
24
40
Figure 8 – 8 km univariate scaling for slope position.
Figure 9 – 8 km univariate scaling for roughness.
-1057
-27
-4
15
803
Slope Position
41
3.4. Focal Species Habitat and Corridor Modeling
The focal species suitable habitat models and corridor modeling followed the
methodology of Khosravi, Hemami, and Cushman (2018) with some modifications. A Random
Forest model was built and a resistance surface was created for input into the UNICOR program.
Scale optimization was performed for each covariate layer determined to be important to the
jaguar once the final set was determined.
3.4.1. Random Forests
After reviewing several commonly used SDM algorithms as described in Chapter 2 the
Random Forests (RF) algorithm was chosen. Random forests is a machine learning algorithm
that generates multiple tree predictors where each tree depends on the values of a random vector
which is sampled independently and with the same distribution for all trees of a forest (Breiman
2001). It can perform both classification and regression accurately even if there are missing data.
The improvements to classification documented in numerous applications can be attributed to the
growth of an ensemble of trees and “letting them vote for the most popular class” (Breiman
2001, 5). Random Forests models will not be overfit, and they will work with large datasets of
higher dimensionality. The growth of the trees occurs by allowing random vectors to be
generated leading the way for the ensemble tree growth. The random vector is generated
independently but with the same distribution as the past random vectors. The error rates
accompanying the random selection of features to split up each node are comparable to the error
rates of the Adaboost algorithm, but these methods are more robust when dealing with noise
(Breiman 2001).
The Random Forest R package “randomForest” is an ensemble machine learning
algorithm for classification and regression which implements the Breiman (2001) random forests
42
algorithm and it is based on Breiman’s and Cutler’s Fortran code (Breiman et al. 2018). Random
Forests was used because it is a strong classifier that uses a robust algorithm and prior studies
have used this approach. For example, Mi et al. (2017) found that it performed better than
MaxEnt for predicting rare species distributions with a limited number of samples over a large
area and missing data for several Asian crane species. Torres et al. (2012) evaluated 11 SDMs
and although all generally had high AUC values (≥ 0.88), the RF model had the highest AUC
(0.96) when testing with 30% of the occurrence locations. Random Forest has an established
history and good predictive performance ability. One great feature of MaxEnt is the capability to
perform a jackknife analysis to identify the most important predictor variables; however,
Random forest can use the extractor function to measure variable importance. Dr. Wan’s R script
with a slight modification by the author was used to create the Random Forest potential habitat
model. The author added some lines of code to make use of multiple computer cores to speed up
the prediction map output.
3.4.2. Calculating resistant kernels and factorial least-cost paths
The resistant kernels and factorial least-cost paths were generated using the UNICOR
program described in Chapter 2. Once the SDM was created for the jaguar, the habitat suitability
model was converted to a resistance map following the methodology of Khosravi, Hemami, and
Cushman (2018). The predicted habitat patches were then determined and a sensitivity analysis
was performed to determine the robustness of predicted patches that reflect the dispersal abilities
of the jaguar.
43
3.4.3. Using the graph network algorithm
The graph network algorithm described in Chapter 2 was used next to determine the
contribution of core areas to the network following the methodology of Khosravi, Hemami, and
Cushman (2018).
3.5. Model Selection, Validation, and Evaluation
Both the receiver operating characteristic (ROC) and area under the ROC curve (AUC)
have been widely used in SDMs to evaluate model accuracy. In general, the ROC is a curve that
plots the true positive rate (TPR) on the y-axis against the false positive rate (FPR) on the x-axis.
The ROC is used for visualizing the performance of a binary classifier widely used in machine
learning. The TPR is also known as the sensitivity (or the probability of detection) and the FPR
is also known as “1-specificity” (or probability of false detection). The AUC summarizes the
performance of the classifier. AUC values closer to 1 indicate better model performance
(Phillips, Dudík, and Schapire, nd). Generally, a model is considered good if it has an AUC
value greater than 0.75 (Elith et al. 2006). The ROC and AUC was used to determine the success
of the models to predict occurrence patterns.
Cohen's (1960) Kappa is a coefficient of interjudge agreement for nominal scales and has
the following equation:
k = Po – Pc / 1 - Po (7)
where Po is equal to the proportion of units in which the judges are in agreement, Pc is equal to
the proportion of units in which the agreement is expected by chance, therefore k is the
proportion of agreement after chance agreement has been removed. It can measure the agreement
between predicted presences and absences (or pseudo-absences) with the actual presences and
absences (or pseudo-absences) corrected for agreement that might occur only by chance. It has a
44
range between -1 to +1 and if the value is less than 0 this indicates that the agreement is less than
expected by chance (or performance is no better than by random chance). Since SDMs were
created using RF with presence-only data, a methodology like that used by Evans and Cushman
(2009) and Mi et al. (2017) was used to validate SDMs in this thesis project.
45
Chapter 4 : Results
This chapter describes the results for the RF habitat model, resistance surface, and the
UNICOR corridor model for the jaguar.
4.1. Jaguar Habitat Map
The jaguar habitat map was modeled using 51 covariates at the optimal scale (i.e the
lowest Out of Bag (OOB) error rates from univariate scaling). The multicollinearity test found
three of the 51 covariates to be correlated and these were removed from further analysis. Those
variables were the Enhanced Vegetation Index (EVI) at 16 km for all years averaged, the
Normalized Difference Vegetation Index (NDVI) at 16 km for all years averaged, and major
roads and links at 16 km. The data with removed variables was found to be balanced and the
final most parsimonious model determined with the model improvement ratio which was not
fitting noise selected a model with 36 variables. The selected model had an Area under the ROC
curve value of 0.852, a cross-validation Kappa value of 0.7073, a cross-validation OOB Error of
0.147561, and a cross-validation error variance of 1.642579e-05. Figure 10 shows the final
jaguar potential habitat map. The color gradient displays areas of high probability of occurrence
(red) to low (blue) for the jaguar.
Results predict there is good habitat for jaguars in the Sonoran-Sinaloan subtropical dry
forest, Sinaloan dry forests, Sierra Madre Occidental, California montane chaparral and
woodlands, Arizona Mountains forests, Sierra Madre Oriental pine-oak forests, Veracruz moist
forests, Sierra de la Laguna pine-oak forests, Sierra de la Laguna dry forests, Tamaulipan
matorral, and small portions of the Sonoran desert ecoregions.
Figure 11 shows the results for the scaled variable importance graph (a), bootstrap error
convergence (b), and the ROC curve (c) for the 36 selected covariates. The scaled variable
46
Figure 10 – Jaguar potential habitat map in the US/Mexico border ecoregions from the RF
model. The color gradient shows the probability of species occurrence.
importance graph (b) lists the variables scaled from less (0) to more (0.05) important for the RF
model. The bootstrap error convergence graph (b) shows the convergence of bootstrap error
estimates with the error on the y-axis and the number of trees on the x-axis. Lastly, the ROC
curve (c) graphically displays the hit rate (y-axis) and the false alarm rate (x-axis). These results
provide confidence in the RF model.
The partial dependency plots for Pav_16km, u17_16km, and MaRd_16km and a presence
/ absence proximity matrix are shown in Figure 12. The partial dependency plots (a) – (c)
indicate the probability for either Pav_16 km, u17_16 km, or MaRd_16 km. The presence /
47
Figure 11 – The scaled variable importance (a), bootstrap error convergence (b), and the ROC
curve (c) graphs for the random forest jaguar potential habitat map.
absence proximity matrix (d) is a two-dimensional graph which plots the absences versus the
presences used in the RF model.
a
b
c
48
Figure 12 – Partial dependency plots (a) – (c), and (d) Presence/Absence proximity matrix.
4.2. Jaguar Resistance Surface
Figure 13 shows the jaguar resistance surface created using a negative exponential
function rescaled from 1 to 100. The color gradient represents areas of high resistance (blue) to
low areas (red) for jaguars to traverse the landscape. Areas of low resistance are assumed to
comprise landscape features which would be easier or less costly for jaguars to traverse the
landscape. Areas of high resistance represent a greater cost to traversal. The less costly areas in
this jaguar resistance surface are assumed to be more favorable for use by jaguars.
a b
c d
49
Figure 13 - Jaguar resistance surface created using a negative exponential function and rescaled
from 1 to 100.
Figure 14 shows the resistance surface with the major interstate freeways, national and
divided roads, and the border wall for reference. From this map we can see that even though
there is habitat which is assumed to have a lower resistance for jaguars there may still be
complete barriers or partially complete barriers to jaguar movement. For example, a complete
barrier could be assumed to be an extremely tall border wall since this would completely block
50
Figure 14 – Map of jaguar resistance surface with major roads and the US-Mexico border wall
overlaid.
jaguar movement. Complete to partial barriers can be assumed for different types of roads such
as interstate highways, primary national roads, or other important typically divided roads.
4.3. Jaguar Corridor Modeling
The jaguar's potential corridor modeling suggests that there were previously two high
density corridors between the US and Mexico allowing jaguar connectivity. However, if the
51
partially constructed border barriers are completed those jaguar corridors will be lost.
Additionally, only one jaguar corridor was completely unobstructed, one partially unobstructed,
and two already blocked by previous border wall construction. The predicted corridors for the
jaguar are shown in Figure 15. Figures 16 and 17 show where the portions of the US-Mexico
border wall intersect the jaguar predicted corridors. Figure 16 shows the intersections and
unobstructed areas of the previous border wall and the predicted jaguar corridors. Figure 17
shows the new border wall with a status of “partially constructed”, previous border wall, and the
intersections with the predicted jaguar corridors. These three maps show the color gradients from
high (red) to low (blue) corridor density. These results suggest that if the partially constructed
border wall goes to completion, it will block these remaining unobstructed corridors found
between the US and Mexico.
52
Figure 15 – Map of predicted corridors for the jaguar with the US-Mexico border wall and areas
in Figures 16 and 17 overlaid on top of the corridor density map. The US and Mexico coastlines
are shown in green.
53
Figure 16 – The predicted corridors for the jaguar with the previous (old) US-Mexico border
wall overlaid on top.
Figure 17 – The predicted corridors for the jaguar with the new US-Mexico border wall overlaid
on top.
54
Chapter 5 : Conclusion
The potential habitat for the jaguar was mapped using the best available data for two
countries, which included occurrence records and numerous covariate datasets. Covariate
datasets and occurrence records were carefully selected from the best available sources to
produce habitat maps, resistance surfaces, and potential corridors. The multiscale optimization
and RF model provided the best possible jaguar model using this available data. The FLCP
analysis incorporated the robust Dijkstra’s (1959) algorithm to predict the best possible potential
corridors for the jaguar.
The choice of covariate dataset sources was based on well documented covariates and
metadata files were carefully inspected. However, country specific covariate datasets were a
major challenge to use because some cannot be combined and others are not of comparable
quality. This did limit the desire to include more covariates. However, future studies can
certainly add more and compare model results. Open-Street-Map provides free global data and
has excellent descriptions of numerous features including roads. However, there may still be
differences in data collection when considering the US and Mexico. Sourcing covariate datasets
that encompass two or more countries can be a challenging task, but future work can be done by
choosing the best available datasets with the best possible descriptions and metadata files as is
reasonably possible.
Several species occurrence records were carefully sourced and stored in geodatabases for
future modeling. These other species records were found along other sections of the US-Mexico
border and will likely produce corridors for one or more of these focal species along other
sections of the US-Mexico border. Modeling for several species will provide more information
about potential corridors which may be hindered by border walls and/-or roads throughout this
55
study area. The effects of using different species datasets can be compared in future research. For
example, if more datasets become available geodatabases can be updated and it would be
interesting to see if the outcomes change.
Future research can be conducted to assess the core habitats for the jaguar as well as to
include different modeling scenarios for this species. For example, scenarios with different
dispersal barriers can be considered and modelled in future studies. In addition, different habitat
models for the jaguar can be compared in future studies. For example, it would be interesting to
compare a MaxEnt output as well as other model outputs to the results provided in this thesis
project. Since covariate datasets were processed and completed to accommodate nine species,
they could be used in future studies to model all nine species.
The potential corridor results for the jaguar in Figure 15 show their locations
concentrated in parts of the western and eastern portions of this study area. With only one
previously completely unobstructed corridor and one partially obstructed corridor available it
becomes even more important to advocate for change. Jaguar corridors are found in only a small
area relative to the entire US-Mexico border and require specialized habitats.
The Department of Homeland Security (DHS) was established June 2002 under the
presidency of George W. Bush as a response to the September 11, 2001, terrorist attacks
(https://dhs.gov). The DHS has an important and honorable duty to protect and defend the
American people through established protocols that allow efficient information sharing between
numerous federal departments and agencies. The U.S. Customs and Border Protection (CBP) is
one of the many departments collaborating with the DHS. The CBP has worthy core values with
the intention to make the US safer. However, keeping Americans safer could extend beyond
identifying potential threats caused by humans to include strategies to enrich and improve the
56
quality of human life by ensuring safe and resilient ecoregions within the US and across borders.
There are numerous threatened species in many countries which affect the structure of ecological
interactions needed for species long term persistence. Mexico has the third highest total number
of threatened species and the US ranks six (IUCN 2021). The DHS and CBP have the potential
to afford species in peril greater opportunities to overcome numerous threats by facilitating
potentially important northern range expansions through potential corridors for endangered
species such as the jaguar which is listed as endangered in both the US and Mexico. By doing so
the connectivity of peripheral jaguar populations may improve and protect against future
environmental conditions that may put them at a greater extinction risk. It is important to protect
core areas, peripheral areas, and cross border connectivity to reduce a species probability of
extinction, in this case the jaguar. The DHS and CBP could re-evaluate border barriers, including
recently built ones because they can cut off corridors for jaguars and potentially other species.
For endangered species it is even more critical to allow dispersal corridors given continuously
shrinking habitats.
57
References
Blaszczynski, J.S. 1997 “Landform characterization with Geographic Information Systems.”
Photogrammetric Engineering and Remote Sensing, 63(2), 183-191.
Boydston, E. E., and C. A. López González. 2005. “Sexual Differentiation in the Distribution
Potential of Northern Jaguars (Panthera onca).” In G. J. Gottfried, B. S. Gebow, L. G. Eskew,
and C. B. Edminster (Eds.), Connecting mountain islands and desert seas: biodiversity and
management of the Madrean Archipelago II. (pp. 51-56). Proc. Fort Collins, CO: U.S.
Department of Agriculture, Forest Service, Rocky Mountain Research Station.
Breiman, L. 2001. “Random Forests.” Machine Learning, 45(1), 5-32.
https://doi.org/10.1023/A:1010933404324.
Breiman, L., A. Cutler, A. Liaw, and M. Wiener. 2018. “Breiman and Cutler's Random Forests
for Classification and Regression.” Retrieved from https://cran.r-
project.org/web/packages/randomForest/randomForest.pdf.
Berry, J. K. 2002. “Use surface area for realistic calculations.” GeoWorld 15(9), 20-21.
Brown D. E. 1983. “On the Status of the Jaguar in the Southwest.” The Southwestern Naturalist,
28(4), 459-460. doi:10.2307/3670828.
Brown, D. E., and C. A. López-González. 2000. “Notes on the Occurrences of Jaguars in
Arizona and New Mexico.” The Southwestern Naturalist, 45(4), 537-542.
https://www.jstor.org/stable/3672607.
Brown, D. E., and C. A. López-González. 2001. “Borderland Jaguars: Tigres de la Frontera.”
Salt Lake City, UT: The University of Utah Press.
Cassaigne, I., R. A. Medellín, R. W. Thompson, M. Culver, A. Ochoa, K. Vargas, … A. Torres-
Gómez. 2016. “Diet of pumas (Puma concolor) in Sonora, Mexico, as determined by GPS kill
sites and molecular identified scat, with comments on jaguar (Panthera onca) diet.” The
Southwestern Naturalist, 61(2) 125-132. https://doi.org/10.1894/0038-4909-61.2.125.
Cohen, J. 1960. “A Coefficient of Agreement for Nominal Scales.” Educational and
Psychological Measurement, 20, 37-46. http://dx.doi.org/10.1177/001316446002000104.
Compton, B. W., K. McGarigal, S. A. Cushman, and L. R. Gamble. 2007. “A Resistant-Kernel
Model of Connectivity for Amphibians That Breed in Vernal Pools.” Conservation
Biology, 21(3), 788-799. https://doi.org/10.1111/j.1523-1739.2007.00674.x.
Cushman, S. A., B. McRae, F. Adriaensen, P. Beier, M. Shirley, and K. Zeller. 2013. “Biological
corridors and connectivity.” Key Topics in Conservation Biology, 2, 384-404.
https://doi.org/10.1002/9781118520178.ch21.
58
Cushman, S. A., J. S. Lewis, and E. L. Landguth. 2014. “Why Did the Bear Cross the Road?
Comparing the Performance of Multiple Resistance Surfaces and Connectivity Modeling
Methods.” Diversity, 6(4), 844-854. https://doi.org/10.3390/d6040844.
Cutler, D. R., Edwards, T. C., Beard, K. H., Cutler, A., Hess, K. T., Gibson, J., and Lawler, J. J.
2007. Random Forests for classification in ecology.” Ecology, 88(11), 2783–2792.
https://doi.org/10.1890/07-0539.1.
Defenders of Wildlife. 2018. “In the Shadow of the Wall: Executive Summary.” Retrieved from
https://defenders.org/newsroom/shadow-of-wall.
DeMatteo, K. E., M. A. Rinas, J. P. Zurano, N. Selleski, R. G. Schneider, and C. F. Argüelles.
2017. “Using niche-modelling and species-specific cost analyses to determine a multispecies
corridor in a fragmented landscape.” PLoS ONE, 12(8), e0183648.
https://doi.org/10.1371/journal.pone.0183648.
Dinerstein, E., D. M. Olson, D. J. Graham, A. L. Webster, S. A. Primm, M. P. Bookbinder, and
G. Ledec. 1995. “A Conservation Assessment of the Terrestrial Ecoregions of Latin America
and the Caribbean.” Washington (DC): World Bank.
Dinerstein, E., D. Olson, A. Joshi, C. Vynne, N. D. Burgess, E. Wikramanayake, … M. Saleem.
2017. “An ecoregion-based approach to protecting half the terrestrial realm.” Bioscience, 67,
534-545.
Dijkstra, E. W. 1959. “A Note on Two Problems in Connexion With Graphs.” Numerical
Mathematics, 1(1), 269-271. https://doi.org/10.1007/BF01386390.
Eizirik, E., J. H. Kim, M. Menotti-Raymond, P. G. Crawshaw, S. J. O’ Brien, and W. E. Johnson.
2001. “Phylogeography, Population History and Conservation Genetics of jaguars (Panthera
onca, Mammalia, Felidae).” Molecular Ecology, 10(1), 65-79. https://doi.org/10.1046/j.1365-
294X.2001.01144.x.
Elith J., C. H. Graham, P. R. Anderson, M. Dudik, S. Ferrier, A. Guisan, … N. E. Zimmermann
2006. “Novel methods improve prediction of species’ distributions from occurrence
data.” Ecography (Copenhagen), 29(2), 129-151. doi:10.1111/j.2006.0906-7590.04596.x.
Elliot, N. B., S. A. Cushman, D. W. Macdonald, and A. J. Loveridge. 2014. “The devil is in the
dispersers: predictions of landscape connectivity change with demography.” Journal of
Applied Ecology, 51(5), 1169–1178. https://doi.org/10.1111/1365-2664.12282.
Ernest, H. B., T. W. Vickers, S. A. Morrison, M. R. Buchalski, and W. M. Boyce. 2014.
“Fractured Genetic Connectivity Threatens a Southern California Puma (Puma Concolor)
Population.” PLoS ONE, 9(10), e107985. https://doi.org/10.1371/journal.pone.0107985.
Ersoy, E., A. Jorgensen, and P. H. Warren. 2019. “Identifying multispecies connectivity
corridors and the spatial pattern of the landscape.” Urban Forestry & Urban Greening, 40,
308–322. https://doi.org/10.1016/j.ufug.2018.08.001
59
Evans, J. S., and S. A. Cushman. 2009. “Gradient modeling of conifer species using random
forests.” Landscape Ecology, 24(5), 673-683. https://doi.org/10.1007/s10980-009-9341-0.
Evans J. S., J. Oakleaf, S. A. Cushman, and D. Theobald. 2014. “An ArcGIS Toolbox for Surface
Gradient and Geomorphometric Modeling (Version 2.0-0).” Retrieved
from http://evansmurphy.wix.com/evansspatial.
Fahrig, L. 2003. “Effects of Habitat Fragmentation on Biodiversity.” Annual Review of Ecology,
Evolution, and Systematics, 34(1), 487–515.
https://doi.org/10.1146/annurev.ecolsys.34.011802.132419.
Freund, Y., and R. E. Schapire. 1997. “A Decision-Theoretic Generalization of On-Line
Learning and an Application to Boosting.” Journal of Computer and System Sciences, 55(1),
119–139. https://doi.org/10.1006/jcss.1997.1504.
Friedman, J. H. 2001. “Greedy Function Approximation: A Gradient Boosting Machine.” The
Annals of Statistics, 29(5), 1189-1232. https://doi.org/10.1214/aos/1013203451.
Friedman, J. H. 2002. “Stochastic Gradient Boosting.” Computational Statistics and Data
Analysis, 38(4), 367-378. https://doi.org./10.1016/S0167-9473(01)00065-2.
Gaskill, M. 2011. “United States Border Fence Threatens Wildlife.” Retrieved from
https://doi.org/10.1038/news.2011.452.
Groom, M. J., G. K. Meffe, and C. R. Carroll. 2006. “Principles of Conservation Biology.”
Sunderland, MA: Sinauer Associates.
Haddad, N. M., L. A. Brudvig, J. Clobert, K. F. Davies, A. Gonzalez, R. D. Holt, … J. R.
Townshend. 2015. “Habitat fragmentation and its lasting impact on Earth’s ecosystems.”
Science Advances, 1(2), e1500052. https://doi.org/10.1126/sciadv.1500052.
Hatten, J. R., A. Averill-Murray, and W. E. van Pelt. 2005. “A spatial model of potential jaguar
habitat in Arizona.” Journal of Wildlife Management, 69(3): 1024–1033.
https://doi.org/10.2193/0022-541X(2005)069[1024:ASMOPJ]2.0.CO;2.
Hijmans, R. J., and J. Elith. 2017. “Species Distribution Modeling with R. (Packages. January 8,
2017).” Retrieved from https://cran.r-project.org/web/packages/dismo/vignettes/sdm.pdf.
Hosch, W. L., ed. 2010. “The Britannica Guide to Geometry.” Chicago, IL: Britannica
Educational Publishing.
IUCN. 2021. “The IUCN Red List of Threatened Species (Version 2021).” Retrieved from
https://www.iucnredlist.org/statistics.
Kabacoff, R. I. 2017. “Generalized Linear Models. (Quick-R: Data Types).” Retrieved from
https://www.statmethods.net/advstats/glm.html.
60
Keitt, T. H., A. Franklin, and D. L. Urban. 1995. “Landscape analysis and metapopulation
structure. (Chapter 3, Recovery Plan for the Mexican Spotted Owl; Volume II, Technical and
supporting information).” Albuquerque, NM: U.S. Department of the Interior Fish and
Wildlife Service, Southwestern Region.
Khosravi, R., M.-R. Hemami, and S. A. Cushman. 2018. “Multispecies assessment of core areas
and connectivity of desert carnivores in central Iran.” Diversity and Distributions, 24(2), 193–
207. https://doi.org/10.1111/ddi.12672.
Landguth, E. L., B. K. Hand, J. Glassy, S. A. Cushman, and M. A. Sawaya. 2012. “UNICOR: A
Species Connectivity and Corridor Network Simulator.” Ecography, 35(1), 9–14.
https://doi.org/10.1111/j.1600-0587.2011.07149.x.
Landguth, E. L., B. K. Hand, J. M. Glassy, and S. A. Cushman. 2016. “UNICOR User Manual
(Version 2.0).” Retrieved from https://github.com/ComputationalEcologyLab.
Larson, S. E. 1997. “Taxonomic re-evaluation of the Jaguar.” Zoo Biology, 16(2), 107-120.
Lasky, J. R., W. Jetz, and T. H. Keitt. 2011. “Conservation Biogeography of the US-Mexico
Border: a Transcontinental Risk Assessment of Barriers to Animal Dispersal.” Diversity and
Distributions 17(4), 673–687. https://doi.org/10.1111/j.1472-4642.2011.00765.x.
McCain, E. B., and J. L. Childs. 2008. “Evidence of Resident jaguars (Panthera onca) in the
southwestern United States and the Implications for Conservation.” Journal of Mammalogy
89(1), 1–10. https://doi.org/10.1644/07-MAMM-F-268.1.
McGarigal, K., K. A. Zeller, and S. A. Cushman. 2016. “Multi-scale habitat selection modeling:
Introduction to the special issue.” Landscape Ecology, 31(6), 1157-1160.
https://doi.org/10.1007/s10980-016-0388-4.
McGarigal, K., H. Y. Wan, K. A. Zeller, B. C. Timm, and S. A. Cushman. 2016. “Multi-scale
Habitat Selection Modeling: A Review and Outlook.” Landscape Ecology, 31(6), 1161-1175.
https://doi.org/10.1007/s10980-016-0374-x.
Menke, K. 2008. “Locating Potential Cougar (Puma concolor) Corridors in New Mexico Using
a Least-Cost Path Corridor GIS Analysis.” Albuquerque, MN: Bird’s Eye View.
Mi, C., F. Huettmann, Y. Guo, X. Han, and L. Wen. 2017. “Why choose Random Forest to
predict rare species distribution with few samples in large undersampled areas? Three Asian
crane species models provide supporting evidence. PeerJ, 5, e2849.
https://doi.org/10.7717/peerj.2849.
Morato, R. G., K. M. Ferraz, R. C. de Paula, and C. B. Campos. 2014. Identification of Priority
Conservation Areas and Potential Corridors for Jaguars in the Caatinga Biome, Brazil. PLoS
ONE, 9(4), e92950. https://doi.org/10.1371/journal.pone.0092950.
61
Morcatty, T. Q., J. C. Bausch Macedo, K. A. I. Nekaris, Q. Ni, C. C. Durigan, M. S. Svensson,
and V. Nijman. 2020. Illegal trade in wild cats and its link to Chinese‐led development in
Central and South America. Conservation Biology, 34(6), 1525–1535.
https://doi.org/10.1111/cobi.13498.
Nelder, J. A., and R. W. M. Wedderburn. 1972. “Generalized Linear Models.” Journal of the
Royal Statistical Society. Series A (General), 135(3), 370-384.
https://www.jstor.org/stable/2344614.
O'Sullivan, D., and G. L. W. Perry. 2013. “Spatial simulation: Exploring pattern and process.”
New York, NY: John Wiley & Sons.
Pascual-Hortal, L., and S. Saura. 2006. “Comparison and Development of New Graph-Based
Landscape Connectivity Indices: Towards the Priorization of Habitat Patches and Corridors
for Conservation.” Landscape Ecology, 21(7), 959-967. https://doi.org/10.1007/s10980-006-
0013-z.
Pascual-Hortal, L., and S. Saura. 2008. “Integrating Landscape Connectivity in Broad-Scale
Forest Planning through a New Graph-Based Habitat Availability Methodology: Application
to Capercaillie (Tetrao Urogallus) in Catalonia (NE Spain).” European Journal of Forest
Research, 127(1), 23–31. https://doi.org/10.1007/s10342-006-0165-z.
Peters, R., W. J. Ripple, C. Wolf, M. Moskwik, G. Carreón-Arroyo, G. Ceballos, … J. R. Miller.
2018. “Nature Divided, Scientists United: US–Mexico Border Wall Threatens Biodiversity
and Binational Conservation.” BioScience, 68(10), 740–743.
https://doi.org/10.1093/biosci/biy063.
Phillips, S. J., R. P. Anderson, M. Dudík, R. E. Schapire, and M. E. Blair. 2017. “Opening the
Black Box: an Open-Source Release of MaxEnt.” Ecography, 40(7), 887–893.
https://doi.org/10.1111/ecog.03049.
Phillips, S. J., R. P. Anderson, and R. E. Schapire. 2006. “Maximum Entropy Modeling of
Species Geographic Distributions.” Ecological Modeling, 190(3-4), 231-259.
https://doi.org/10.1016/j.ecolmodel.2005.03.026.
Phillips, S. J., and M. Dudík. 2008. “Modeling of Species Distributions with MaxEnt: New
Extensions and a Comprehensive Evaluation.” Ecography, 31(2), 161-175.
https://doi.org/10.1111/j.0906-7590.2008.5203.x.
Phillips, S. J., M. Dudík, and R. E. Schapire. [nd] “MaxEnt software for modeling species niches
and distributions (Version 3.4.1).” Retrieved from
http://biodiversityinformatics.amnh.org/open_source/maxent/.
Phillips, S. J., M. Dudík, and R. E. Schapire. 2004. “A Maximum Entropy Approach to Species
Distribution Modeling.” In Proceedings of the Twenty-First International Conference on
Machine Learning, Banff, Alberta, Canada (pp. 1-8). New York, NY: ACM.
https://doi.org/10.1145/1015330.1015412.
62
Pimm S. L., C. N. Jenkins, R. Abell, T. M. Brooks, J. L. Gittleman, L. N. Joppa, …
J. O. Sexton. 2014. “The biodiversity of species and their rates of extinction, distribution, and
protection.” Science, 344(6187), 1246752. https://doi.org/10.1126/science.1246752.
Pocock, R. I. 1939. “The Races of Jaguar (Panthera onca).” Novitates Zoologicae, 41(4), 406-
422.
Proulx, S. R., D. E. L. Promislow, and P. C. Phillips. 2005. “Network Thinking in Ecology and
Evolution.” Trends in Ecology and Evolution, 20(6), 345-353.
https://doi.org/10.1016/j.tree.2005.04.004.
Quigley, H., R. Foster, L. Petracca, E. Payan, R. Salom, and B. Harmsen. 2017. “Panthera onca
(Errata version published in 2018) The IUCN Red List of Threatened Species 2017).”
Retrieved from http://dx.doi.org/10.2305/IUCN.UK.2017-3.RLTS.T15953A50658693.en.
R Core Team. 2018. “R: A language and environment for statistical computing.” Vienna,
Austria: R Foundation for Statistical Computing.
Rabinowitz, A., and K. A. Zeller. 2010. “A range-wide model of landscape connectivity and
conservation for the jaguar, Panthera onca.” Biological Conservation, 143(4), 939–945.
https://doi.org/10.1016/j.biocon.2010.01.002.
Ricketts, T. H. 2001. “The Matrix Matters: Effective Isolation in Fragmented Landscapes.” The
American Naturalist, 158(1), 87-99. https://doi.org/10.2307/3078900.
Ridgeway, G. 2019. “Generalized Boosted Models: A Guide to the gbm Package.” Retrieved
from https://cran.r-project.org/web/packages/gbm/vignettes/gbm.pdf.
Riley, S. P. D., J. P. Pollinger, R. M. Sauvajot, E. C. York, C. Bromley, T. K. Fuller, and R. K.
Wayne. 2006. “FAST-TRACK: A Southern California Freeway Is a Physical and Social
Barrier to Gene Flow in Carnivores.” Molecular Ecology, 15(7), 1733–1741.
https://doi.org/10.1111/j.1365-294x.2006.02907.x.
Riley, S. J., S. D. DeGloria and R. Elliot. 1999. “A terrain ruggedness index that quantifies
topographic heterogeneity.” Intermountain Journal of Sciences, 5, 23-27.
Rodríguez-Soto, C., O. Monroy-Vilchis, L. Maiorano, L. Boitani, J. C. Faller, M. Á. Briones, …
A. Falcucci. 2011. Predicting potential distribution of the jaguar (Panthera onca) in Mexico:
identification of priority areas for conservation. Diversity and Distributions, 17(2), 350–361.
https://doi.org/10.1111/j.1472-4642.2010.00740.x.
Rodríguez-Soto, C., O. Monroy-Vilchis, and M. M. Zarco-González. 2013. Corridors for jaguar
(Panthera onca) in Mexico: Conservation strategies. Journal for Nature Conservation, 21(6),
438–443. https://doi.org/10.1016/j.jnc.2013.07.002.
Rosas-Rosas, O., L. Bender, and R. Valdez. 2008. “Jaguar and Puma Predation on Cattle Calves
in Northeastern Sonora, Mexico.” Rangeland Ecology and Management, 61(5), 554-560.
63
Rosas-Rosas, O. and R. Valdez. 2010. The Role of Landowners in Jaguar Conservation in
Sonora, Mexico. Conservation Biology, 24(2), 366–371. https://doi.org/10.1111/j.1523-
1739.2009.01441.x.
Rudnick, D. A., S. J. Ryan, P. Beier, S. A. Cushman, F. Dieffenbach, C. W. Epps, … S. C.
Trombulak. 2012. “The Role of Landscape Connectivity in Planning and Implementing
Conservation and Restoration Priorities.” Ecological Society of America Issues in Ecology,
16, 1-20.
Ruiz-Garcia, M., E. Payán, A. Murillo, and D. Alvarez. 2006. “DNA Microsatellite
Characterization of the jaguar (Panthera onca) in Colombia.” Genes and Genetic Systems,
81(2), 115-127. https://doi.org/10.1266/ggs.81.115.
Saura, S., and L. Pascual-Hortal. 2007. “A New Habitat Availability Index to Integrate
Connectivity in Landscape Conservation Planning: Comparison with Existing Indices and
Application to a Case Study.” Landscape and Urban Planning, 83(2-3), 91–103.
https://doi.org/10.1016/j.landurbplan.2007.03.005.
Saura, S., and J. Torne´. 2009. “Conefor Sensinode 2.2: A Software Package for Quantifying the
Importance of Habitat Patches for Landscape Connectivity.” Environmental Modeling and
Software, 24(1), 135-139. https://doi.org/10.1016/j.envsoft.2008.05.005.
SEMARNAT (Secretary of Natural Resources of Mexico). 2002. “Normas Oficiales Mexicanas
(NOM-059-Ecol.) que determina las especies de flora y fauna silvestres terrestres y acuaticas
en peligro de extincion, menazadas, raras y las sujetas a proteccion especial, que establece
especificaciones para su proteccion.” Mexico City, Mexico: SEMARNAT.
Seymour, K. L. 1989. “Panthera onca.” Mammalian Species, 340, 1-9.
https://doi.org/10.2307/3504096.
Tôrres, N. M., P. De Marco, T. Santos, L. Silveira, A. T. de Almeida Jácomo, and J. A. Diniz-
Filho. 2012. Can species distribution modelling provide estimates of population densities? A
case study with jaguars in the Neotropics. Diversity and Distributions, 18(6), 615–627.
https://doi.org/10.1111/j.1472-4642.2012.00892.x.
Timm, B. C., K. McGarigal, S. A. Cushman, and J. L. Ganey. 2016. “Multi-scale Mexican
spotted owl (Strix occidentalis lucida) nest/roost habitat selection in Arizona and a
comparison with single-scale modeling results.” Landscape Ecology, 31(6), 1209-1225.
https://doi.org/10.1007/s10980-016-0371-0.
Upton, G., and I. Cook. 2014. “A Dictionary of Statistics (3 ed.).” Oxford, UK: Oxford
University Press. https://doi.org/10.1093/acref/9780199679188.001.0001.
Urban, D., and T. Keitt. 2001. “Landscape connectivity: a graph-theoretic perspective.” Ecology,
82(5), 1205-1218. https://doi.org/10.2307/2679983.
64
Wan, H. Y., S. A. Cushman, and J. L. Ganey. 2018. “Habitat Fragmentation Reduces Genetic
Diversity and Connectivity of the Mexican Spotted Owl: A Simulation Study Using Empirical
Resistance Models.” Genes, 9(8), 403. https://doi.org/10.3390/genes9080403.
Wan, H. Y., K. McGarigal, J. L. Ganey, V. Lauret, B. C. Timm, and S. A. Cushman. 2017.
“Meta-replication reveals nonstationarity in multi-scale habitat selection of Mexican Spotted
Owl.” The Condor: Ornithological Applications, 119(4), 641–658.
https://doi.org/10.1650/CONDOR-17-32.1.
Wiens, J. A. 1989. “Spatial Scaling in Ecology.” Functional Ecology, 3(4), 385-397.
https://www.jstor.org/stable/2389612.
Worton, B. J. 1989. Kernel Methods for Estimating the Utilization Distribution in Home-Range
Studies. Ecology, 70(1), 164–168. https://doi.org/10.2307/1938423.
Zarco-González, M. M., O. Monroy-Vilchis and J. Alaníz. 2013. Spatial model of livestock
predation by jaguar and puma in Mexico: Conservation planning. Biological Conservation,
159, 80–87. https://doi.org/10.1016/j.biocon.2012.11.007.
Zeller, K. A., S. Nijhawan, R. Salom-Pérez, S. H. Potosme, and J. E. Hines. 2011. “Integrating
occupancy modeling and interview data for corridor identification: A case study for jaguars in
Nicaragua.” Biological Conservation, 144(2), 892-901.
https://doi.org/10.1016/j.biocon.2010.12.003.
Abstract (if available)
Abstract
Connectivity is important for biodiversity conservation because it can offset the impacts of habitat loss and fragmentation, allowing migration, dispersal, and adequate gene flow. Barriers that cut across a species range such as the United States-Mexico border wall can block dispersal and negatively impact gene flow between populations. It is therefore important to understand how to establish or re-establish wildlife corridors in order to help species survive. The focal species selected for this thesis project was the jaguar (Panthera onca). The study area comprised several ecoregions that covered portions of the United States of America (US) and Mexico. The jaguar’s suitable habitat was identified using a Random Forest model to predict potential habitats. The factorial least-cost path analysis was used to identify the jaguar’s potential corridors. Results predict there is good habitat for jaguars in the Sonoran-Sinaloan subtropical dry forest, Sinaloan dry forests, Sierra Madre Occidental, California montane chaparral and woodlands, Arizona Mountains forest, Sierra Madre Oriental pine-oak forests, Veracruz moist forests, Sierra de la Laguna pine-oak forests, Sierra de la Laguna dry forests, Tamaulipan matorral, and small portions of the Sonoran desert ecoregion. The jaguar's potential corridor modeling suggests that there were previously two high density corridors between the US and Mexico allowing jaguar connectivity. However, if the partially constructed border barriers are completed those jaguar corridors will be lost. Work on nine co-distributed mammals (orders: Carnivora and Artiodactyla): jaguar (Panthera onca), mountain lion (Puma concolor), ocelot (Leopardus pardalis), bobcat (Lynx rufus), black bear (Ursus americanus), gray fox (Urocyon cinereoargenteus), Mexican gray wolf (Canus lupus baileyi) Sonoran pronghorn (Antilocapra americana sonoriensis), and Bighorn sheep (Ovis canadensis) in the US-Mexico border ecoregions will continue after the completion of this work.
Linked assets
University of Southern California Dissertations and Theses
Conceptually similar
PDF
Use of least-cost path analysis to identify potential movement corridors for jaguars across the US-Mexico border
PDF
A threat-based least-cost path decision support model for national security resource allocation along the US-Mexico border
PDF
A Maxent-based model for identifying local-scale tree species richness patch boundaries in the Lake Tahoe Basin of California and Nevada
PDF
Assessing the transferability of a species distribution model for predicting the distribution of invasive cogongrass in Alabama
PDF
Light rail expansion in Houston using viable path corridors and least cost path: alternatives for the failed University and Uptown lines
Asset Metadata
Creator
Torres, Cirenia Argelia
(author)
Core Title
Assessing the connectivity for the jaguar (Panthera onca) in the United States-Mexico border ecoregions using species distribution modeling and factorial least cost path analysis
School
College of Letters, Arts and Sciences
Degree
Master of Science
Degree Program
Geographic Information Science and Technology
Degree Conferral Date
2021-08
Publication Date
07/18/2021
Defense Date
06/14/2021
Publisher
University of Southern California
(original),
University of Southern California. Libraries
(digital)
Tag
border,connectivity,dispersal barriers,ecoregions,factorial least cost path,Jaguar,landscape connectivity,OAI-PMH Harvest,Panthera onca,random forest,random forests,species distribution modeling,UNICOR,United States-Mexico border,UNIversal CORridor Network Simulator
Format
application/pdf
(imt)
Language
English
Contributor
Electronically uploaded by the author
(provenance)
Advisor
Wilson, John P. (
committee chair
), Cushman, Samuel A. (
committee member
), Fleming, Steven D. (
committee member
)
Creator Email
catorres@usc.edu,cireniaatorres@gmail.com
Permanent Link (DOI)
https://doi.org/10.25549/usctheses-oUC15609009
Unique identifier
UC15609009
Legacy Identifier
etd-TorresCire-9781
Document Type
Thesis
Format
application/pdf (imt)
Rights
Torres, Cirenia Argelia
Type
texts
Source
University of Southern California
(contributing entity),
University of Southern California Dissertations and Theses
(collection)
Access Conditions
The author retains rights to his/her dissertation, thesis or other graduate work according to U.S. copyright law. Electronic access is being provided by the USC Libraries in agreement with the author, as the original true and official version of the work, but does not grant the reader permission to use the work if the desired use is covered by copyright. It is the author, as rights holder, who must provide use permission if such use is covered by copyright. The original signature page accompanying the original submission of the work to the USC Libraries is retained by the USC Libraries and a copy of it may be obtained by authorized requesters contacting the repository e-mail address given.
Repository Name
University of Southern California Digital Library
Repository Location
USC Digital Library, University of Southern California, University Park Campus MC 2810, 3434 South Grand Avenue, 2nd Floor, Los Angeles, California 90089-2810, USA
Repository Email
cisadmin@lib.usc.edu
Tags
border
connectivity
dispersal barriers
ecoregions
factorial least cost path
Jaguar
landscape connectivity
Panthera onca
random forest
random forests
species distribution modeling
UNICOR
United States-Mexico border
UNIversal CORridor Network Simulator