University of Southern California Dissertations and Theses

Using Maxent modeling to predict habitat of mountain pine beetle in response to climate change

(USC Thesis Other)

Using Maxent modeling to predict habitat of mountain pine beetle in response to climate change

PDF

Download a page range

Download transcript

Copy asset link

Request this asset

Transcript (if available)

Content

USING MAXENT MODELING TO PREDICT HABITAT OF MOUNTAIN PINE
BEETLE IN RESPONSE TO CLIMATE CHANGE

by

Caitlan R. Dowling

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)

DECEMBER 2015

Copyright 2015 Caitlan R. Dowling

ii

iii
ACKNOWLEDGMENTS
So many people have encouraged me throughout this process, from close friend to
strangers on the street. I would like to specifically thank the Spatial Sciences Institute
faculty and staff, including Kate and Ken, and Richard and the GIST Tech Support. My
committee Dr. Jennifer Swift and Dr. Darren Ruddell, and my advisor Dr. Travis
Longcore. I am lucky to have such a positive advisor- thank you for your direction.
Thank you to my parents, who have supported me throughout this process- as they have
all my endeavors- along with the rest of my family. And to my partner, Mark Henspeter,
thank you for always reminding me I can do it.

iv
TABLE OF CONTENTS
ACKNOWLEDGMENTS .......................................................................................................................................... iii
TABLE OF CONTENTS ........................................................................................................................................... iv
LIST OF TABLES ...................................................................................................................................................... vi
LIST OF FIGURES ................................................................................................................................................... vii
LIST OF ABBREVIATIONS ................................................................................................................................... ix
ABSTRACT ................................................................................................................................................................... x
CHAPTER ONE: INTRODUCTION ...................................................................................................................... 1
1.1 Description of Species ................................................................................................................................ 1
1.2 Description of Mountain Pine Beetle Habitat .................................................................................. 3
1.3 Study Area ...................................................................................................................................................... 7
1.4 Study ................................................................................................................................................................. 9
CHAPTER TWO: RELATED WORK .................................................................................................................. 10
2.2 Vegetation ..................................................................................................................................................... 12
2.4 Past Maxent study of the Mountain Pine Beetle ........................................................................... 14
2.4.1 Biological Data .................................................................................................................................... 14
2.4.2 Environmental Variables ................................................................................................................ 14
2.4.3 Analysis ................................................................................................................................................. 16
2.5 Maxent with CMIP5 .................................................................................................................................. 18
CHAPTER THREE: METHODS ........................................................................................................................... 19
3.2 Research Design ......................................................................................................................................... 19
3.3 Research Data ............................................................................................................................................. 20
3.3.1 Biological Data .................................................................................................................................... 21
3.3.2 Environmental Variables ............................................................................................................... 23
3.4 Procedures ................................................................................................................................................... 25
CHAPTER FOUR: RESULTS ................................................................................................................................ 27
4.1 Current Species Distribution Map ....................................................................................................... 27
4.2 Future Species Distribution Map ......................................................................................................... 29
4.3 Model Performance ................................................................................................................................... 32
4.3.1 Current .................................................................................................................................................. 32
4.3.2 Future ..................................................................................................................................................... 32
4.4 Variable Importance ................................................................................................................................. 33

v
4.4.1 Current .................................................................................................................................................. 33
4.4.2 Future ..................................................................................................................................................... 34
CHAPTER FIVE: DISCUSSION ............................................................................................................................ 36
5.1 Model Strengths and Weaknesses ...................................................................................................... 36
5.2 Geographic Results ................................................................................................................................... 37
5.3 Current to 2050 and 2070 Changes ................................................................................................... 39
5.4 Variable Results ......................................................................................................................................... 42
5.5 Future Research ......................................................................................................................................... 42
REFERENCES ........................................................................................................................................................... 45
APPENDICES ............................................................................................................................................................ 49
Appendix A – Jackknife Tests ....................................................................................................................... 49
Current ............................................................................................................................................................. 49
2050, RCP 4.5 ................................................................................................................................................. 50
2050, RCP 8.5 ................................................................................................................................................. 52
2070 RCP 4.5 .................................................................................................................................................. 53
2070 RCP 8.5 .................................................................................................................................................. 55
Appendix B – Stretch Maps with 2.5 Standard Deviation ................................................................ 57
Current ............................................................................................................................................................. 57
2050 RCP 4.5 .................................................................................................................................................. 57
2050 RCP 8.5 .................................................................................................................................................. 58

vi
LIST OF TABLES
Table 1 Safranyik Climate Factors effecting the MPB ............................................................ 11
Table 2 WorldClim Bioclimatic Variables .................................................................................. 15
Table 3 Percent Habitat Change ..................................................................................................... 16
Table 4 Square Kilometers ............................................................................................................... 17
Table 5 Square Kilometers deemed suitable ............................................................................ 30
Table 6 Percent decrease from current suitable area, 490,075 km
2
............................... 31
Table 7 Future AUC Results ............................................................................................................. 33
Table 8 Most Influential Variables ................................................................................................ 34

vii
LIST OF FIGURES
Figure 1 Left, Adult Mountain Pine Beetle; Right, Mountain Pine Beetle Larva
(Colorado State University 2011) ........................................................................................... 2
Figure 2 Forest Service IDS Mountain Pine Beetle Survey Data from 1997 to 2014 ... 3
Figure 3 Sawdust at the base of a tree is residue from the MPB burrowing into the
tree (British Columbia 2013) ................................................................................................... 4
Figure 4 Lodgepole Pine produce resin in an attempt to keep the beetle out. While
this can sometimes be successful, ‘pitching out’ the beetle, often the beetle
overcomes the resin and leaves behind pitch tubes (NPS 2011). .............................. 4
Figure 5 Study Area ............................................................................................................................... 8
Figure 6 2014 MPB Survey Data (Forest Service 2014) ....................................................... 21
Figure 7 The raster grid cells are 30 arc-seconds: 0.008333333333333 degrees
or approximately 1 kilometer squared. ............................................................................. 22
Figure 8 ArcMap model used to process GeoTIFF images for Maxent ............................ 25
Figure 9 Current MPB Suitability (Threshold 0.4666) .......................................................... 27
Figure 10 Current Scaled MPB Suitability .................................................................................. 28
Figure 11 Current MPB Suitability compared to MPB Damage from 1997 to 2014 .. 29
Figure 12 Binary and Scaled Results for 2050 ......................................................................... 30
Figure 13 Binary and Scaled Results for 2070 ......................................................................... 31

viii
Figure 14 Area Under ROC (Receiver operating characteristic) Curve (AUC),
Current Climate ........................................................................................................................... 32
Figure 15 These Jackknife results for 2050/RCP 4.5 serve as representative
Jackknife results for the future models. Results for each individual model
are available in Appendix B. ................................................................................................... 35
Figure 16 By 2070, both RCP 4.5 and RCP 8.5 show high suitability for the MPB
near the Great Lakes. Maps showing all results with a 2.5 Standard Deviation
are available in Appendix B. ................................................................................................... 38
Figure 17 The change in suitability for 2050 ............................................................................ 40
Figure 18 The change in suitability for 2070 ............................................................................ 41

ix
LIST OF ABBREVIATIONS
AUC Area Under the ROC Curve
GIST Geographic Information Science and Technology
IDS Insect and Disease Detection Survey
MPB Mountain Pine Beetle
RCP Representative Concentration Pathways
ROC Receiver Operating Characteristic
SSI Spatial Sciences Institute
USC University of Southern California

x
ABSTRACT
The Mountain Pine Beetle (Dendroctonus ponderosae) is a unique indicator species in
the face of climate change. Since the beginning of this century, it has expanded from its
historic territory in the Rocky Mountains at an unprecedented rate. As climate variables
continue to change, it is uncertain how the MPB will spread throughout the continental
United States. Existing habitat models have studied the current MPB territory, but have
not yet been expanded to look at how a changing climate might influence the habitable
range for the MPB. In response to recent climate shifts, host tree species have become
increasingly susceptible to MPB attack. As their historical habitat is consumed the MPB
may also be expanding into new host species. This study applied Maximum Entropy
modeling (Maxent) processes to look at habitat suitability for the Mountain Pine Beetle
under future climate scenarios. Results for two different emissions scenarios for 2050 and
2070 both showed a change in the MPB’s range across the United States. Habitable areas
became more concentrated to cooler areas, typically at higher elevations. These models
show that as climate change progresses, the Mountain Pine Beetle will be a dynamic
variable in forest management across the country as it alters not only its distribution, but
also impacted species. Maxent modeling techniques allow a look into the future under
varying scenarios to effectively predict the impacts of climate change on the Mountain
Pine Beetle and its presence in our forest system.

1
CHAPTER ONE: INTRODUCTION
The mountain pine beetle (Dendroctonus ponderosae) (MPB) is a small dark beetle that
burrows into mature pine trees in Western North America. In the twentieth century, MPB
habitat spanned from the Black Hills of South Dakota to the West Coast (Wood 1982).
With climate change shifting potential beetle habitat, this study looks at prospective new
habitats for the pine beetle within the continental United States of America. In
determining what areas could be suitable beetle habitat, forest caretakers can plan for the
impeding beetle arrival and develop an appropriate management strategy. GIS can be
used to refine predictions of the areas of U.S. forests at risk from pine beetle infestation
under climate change scenarios.

1.1 Description of Species
The MPB is one of many bark beetle species that have been part of a healthy
forest cycle for thousands of years. In addition to the fossil record of bark beetles
preserved in tree resin (Nikiforuk 2011, 44 - 45) studies of tree rings in Alaska and
Canada show that beetle outbreaks have been a part of a natural cycle to thin forests
(Berg et al. 2006, 22). It is hard to view the life cycle of a forest within the life span of a
human, which makes the MPB seem as a destructive agent. Andrew Delmar Hopkins
(1857 – 1948), considered the “father of North American entomology” (Nikiforuk 2011,
56) studied the MPB in the late 19
th
century and deemed it “the enemy of pine forests.”
Stephen Lane Wood (1924 – 2009) of Brigham Young University was considered the
premier expert on bark beetles. He authored and co-authored over 100 publications and
categorized over 1000 of the world’s beetles (Cognato and Knizek 2010). In his The

2
Bark and Ambrosia Beetles of North and Central America (Coleoptera: Scolytidae): A
Taxonomic Monograph, he calls the MPB “the most destructive species of
Dendroctonus.”
The MPB’s life cycle typically takes one full year to complete. However, in
colder climates this cycle can take up to two years to complete, while warmer areas may
see two to three cycles per year (Logan and Bentz 1999, 925). Adult MPBs take flight in
early summer, traveling anywhere from the next tree to over 200 miles to find new trees
to infest (Nikiforuk 2011, 73). They then burrow into the tree, eating the phloem of the
tree. During this process they also deposit funguses such as the Blue Stain Fungus
(grosmannia clavigera) which they carry in their mycangia, or pouches within their
cheeks (Halter 2011, 58). After laying their eggs, which develop into larva, the pulp
created by the fungus then feeds the growing larva into adulthood, in which the cycle
starts again. Adult MPB are 1/8 to 1/3 of an inch, while larva are 1/8 to1/4 of an inch,
depicted in Figure 1.

Figure 1 Left, Adult Mountain Pine Beetle; Right, Mountain Pine Beetle Larva
(Colorado State University 2011)

3
1.2 Description of Mountain Pine Beetle Habitat
The MPB is naturally found in the Rocky Mountain region, from the Black Hills
of South Dakota to the west coast (Wood 1982). Figure 2 depicts the span of the MPB
range from 1997 to 2014. The pine (Pinaceae) family represents a range of trees
susceptible to the MPB. The MPB predominantly attacks the Lodgepole Pine (Pinus
contorta), typically trees over 10 inches in diameter, 85 to 100 years old. During large
outbreaks, the MPB will attack trees as small as 4 inches in diameter (Logan and Powell
2011). However, they can infest up to 22 different species of pine (Safranyik et al. 2010,
416).

Figure 2 Forest Service IDS Mountain Pine Beetle Survey Data from 1997 to 2014

4
Several telltale signs can detect the MPB’s presence in a tree. As MPBs burrow
into a tree, they leave ample sawdust at the base of the tree (Figure 3). Trees with natural
defenses against the MPB, such as the Lodgepole Pine, produce resin to block additional
beetles from entering the tree and attempts to suffocate the ones that have already entered
the tree (Figure 34).

Figure 3 Sawdust at the base of a tree is residue from the MPB burrowing into the
tree (British Columbia 2013)

Figure 4 Lodgepole Pine produce resin in an attempt to keep the beetle out. While
this can sometimes be successful, ‘pitching out’ the beetle, often the beetle
overcomes the resin and leaves behind pitch tubes (NPS 2011).

5
After a successful attack, trees are broken down into three classifications; green
attack (the tree is infested but still photosynthesizing, keeping the needles green), red
attack (one year after a successful attack, 90% of all killed trees will have red needles),
and gray attack (three years after a successful attack, a tree has lost all of its needles)
(Wulder 2005, 18-41). Remote sensing can identify areas of red attack trees. While
Landsat TM/ETM+ data can only identify large areas of red attack trees, multispectral
IKONOS data can detect small patches of red attack trees due to the pixel size being
approximately the size of a pine crown (White et al. 2005, 7).
Being able to identify small outcroppings of infestation is one of the most
successful forms of preventing further outbreaks, as once an infestation has reached a
certain size little can be done to stop it (White et al. 2005, 4). However, remote sensing
can only be used to identify infested trees a full year after an attack has started, as there is
currently no method to remotely identify green attack trees (Wulder and Dymond 2003,
2). This makes accurate prediction models of what areas would be suitable habitat for the
MPB important in order for forest managers to be on high alert for the first signs of a
MPB attack. Additionally, areas predicted to have a high possibility of a MPB attack can
create a response plan to implement at the first sign of the MPB.
Pine forests naturally maintain a symbiotic relationship with both the MPB and
wildfires. The MPB would enter a forest, leaving conditions ripe for forest fires, which
would clear the underbrush and open the pinecones of old growth Lodgepole Pines
(Nikiforuk 2011, 58). America’s stance on wildfire suppression was to extinguish all
fires no matter the size or the cause until the 1960s, when a more holistic approach was
taken. It was realized that fire is a necessary part of the forest lifecycle to retain a

6
balance between old growth and new growth vegetation. Significant damage had already
been done, and many of the pine forests of North America are full of dense old growth
trees. This has upset the natural relationship between the MPB, trees and wildfires,
which historically had worked together to renew forests.
The presence of densely packed, old growth pine leaves forests susceptible to
both wildfire and epidemic levels of beetle infestations. The MPB can travel great
distances to infest a forest, and can spread over thousands of acres in a few short years
(Nikiforuk 2011, 73). While an overly dense forest can in itself create a higher fire
hazard, the presence of the MPB can drastically change the condition of the forest as a
fuel source (Jenkins 2013, 2). In some cases, if a bark beetle is carrying the blue stain
fungus, the dead forest dramatically drops in susceptibility to fire, as seen after the spruce
beetle (Dendroctonus rufipennis) infestation on the Kenai Peninsula, Alaska (Berg et al.
2006). This is due to the fungus speeding up the rate of decomposition in the tree,
leaving the dead trees soft rather than dry. In other cases, the MPB can leave forests as
ready fuel for a fire. This is due to the MPB leaving behind dry, brittle trees when the
fungus is not present. The relationship grows more complex as firefighters rely on
predicting the movement and severity of a wildfire based on the known vegetation
present in a forest. Each level of attacked trees, even green attack trees, presents a
different level of fire susceptibility (Jenkins 2013, 6). This makes the identification of
potential new MPB outbreak areas highly important for effective fire management,
including the safety of surrounding communities and the fire fighters themselves.
There is debate over whether the relationship between the blue stain fungus and
the MPB is traditional symbiotic or parasitic on behalf of the fungus. MPBs can

7
successfully kill a tree without the presence of the blue stain fungus, and trees can survive
with the fungus (Six 2011, 6-9). However, when trees are already stressed by higher
temperatures and lower precipitation their defense system is compromised (Chapman et
al. 2012, 2176). This can result in the presence of the fungus increasing their mortality
rate (Halter 2011, 59). MPB larva cannot survive below -40 degrees F, which limits its
available habitat (Carroll et al. 2006, 2). In the past 30 years, increasingly fewer areas in
North America reach this temperature during winter. Longer, warmer summers have
extended the season which adult MPB can emerge from their trees and take flight,
attacking new trees. Additionally, a combination of higher temperatures and lower
precipitation have left trees stressed, increasing their susceptibility to a beetle attack
(Carroll et al. 2006, 2).

1.3 Study Area
This study looks at the contiguous United States as potential habitat for the MPB.
The data sets used are nation wide data sets, although MPB presence has historically
appeared in the western United States. The study boundaries (Figure 5) are set by the
extent of the rasters used as the environmental variables and future climate variables.
These rasters are clipped to the same coordinates NASA NEX uses for their downscaled
model of the continental United States: -125.02083333, 49.9375, -66.47916667, 24.0625.
Moving beyond the immediate boarder of the United States to a slightly larger
rectangular prevents error along the borders of the result where the raster results may be
skewed.

8

Figure 5 Study Area

This study will look beyond the western United States, as the MPB has recently
been identified in Michigan, where it has attacked the Jack Pine (Pinus banksiana)
(Klutsch and Erbilgin 2012). Additionally, studies in Canada have modeled how the
MPB may move across the boreal forest, which contains significant amounts of Jack Pine
(Safranyik et al. 2010.) Study participant Allan Carroll has stated he believes the MPB
would continue to move across the continent (National Geographic 2015.) The Jack Pine
does not have natural defenses against the MPB such as the Lodgepole Pine. As suitable
habitat shifts, the vegetation the MPB occupies may shift as well. This study will expand
the range typically used to examine MPB habitat to determine if new areas are likely to
become suitable habitat under climate change scenarios. The western states where the
MPB has historically inhabited are also included however, as an increase in temperatures
may change their habitat boundaries there as well.

9
1.4 Study
The MPB has been the subject of many scientific studies. The literature review
section covers the various work that have defined what variables the MPB does or does
not thrive under, including climate and vegetation. How Maxent can be used as a Species
Distribution Model is then explained, along with why the MPB is a good candidate for
Maxent. The methods section explains what data will be used for the Maxent Models.
The results show the output of the Maxent runs, while the discussion explains the
modeling process, the results and their application. This study hopes to answer what areas
may be suitable habitat for the MPB in the future, and the variables that determine this.
By using Maxent to create future habitat models, GIS modeling allows for the prediction
of suitable habitat into 2050 and 2070.

10
CHAPTER TWO: RELATED WORK
Starting in the early 2000s, the dramatic outbreak of MPB across the country spurred an
increase in research efforts focusing on the species. These studies are generally composed
of two categories; variables affecting the MPB, and how those factors will affect the
MPB in the future.

2.1 Variables affecting the Mountain Pine Beetle
The Logan et al model looks into how univoltine seasonality (one brood a season)
is related to epidemics (Logan and Bentz 1999, 925). The Regniere and Bentz model
studies at what temperatures MPB larva survive verses succumb to cold temperatures
(Regniere and Bentz 2007, 559-72). The Safranyik et al model focuses on a collection of
climate variables for predicting MPB spread through Lodgepole Pine (Carroll et al. 2006,
1). Dr. Les Safranyik and Allan Carroll have been major players in the Canadian Forest
Service’s research into the MPB. Together they have worked to modify the model, and
apply it to predicting range expansion into the Canadian boreal forest (Safranyik et al.
2010, 415-442).
The Safranyik et al model focuses on identifying areas that are climatically
suitable for the MPB, considering temperature, length of growing season and
precipitation. For this, they developed a series of true/false statements based off the
original model from the 1970s, shown in Table 1.

11
Table 1 Safranyik Climate Factors effecting the MPB
Criteria Description Rationale
P1 > 350 degree-days above 5.5 degree
Celsius from August 1
st
to end of
growing season (Boughner 1964) and
>833 degree-days from August 1
st
to
July 31
st

A univoltine lifecycle synchronized
with critical seasonal events is
essential for MPB survival (Logan
and Powell 2001). 305 degree-days is
the minimum heat requirement from
peak flight to 50% egg hatch, and 833
degree-days is the minimum required
for a population to be univoltine
(adapted from Reid 1962)
P2 Minimum winter temperatures >-40
degrees Celsius
Under-bark temperature at or below -
40 degree causes 100% morality
within a population (Safranyik and
Linton 1998)
P3 Mean Maximum August
temperatures >/= 18.3 degrees
Celsius
The lower threshold for MPB flight is
~18.3 degrees Celsius (McCambridge
1971). It is assumed that when the
frequency of maximum daily
temperatures >/= 18.3 degrees Celsius
is <=5% during August, the peak of
MPB emergence and flight will be
protracted and mass attack success
reduced.
P4 Sum of precipitation from April to
June < long-term average
Significant increases in MPB
population have been correlated with
periods of two or more consecutive
years of below-average precipitation
over large areas of western Canada
)Thomson and Shrimpton 1984)
Y1 Variability of growing season
precipitation
Since P4 is defined in terms of a
deviation from average, the
coefficient of variation of
precipitation was included. Its
numerical values were converted to a
relative sale from 0 to 1.
Y2 Index of water deficit (the index of
water deficit replaces the water
deficit approximation (National Atlas
of Canada 1970) in the original
model of Safranyik et al. (1975).
Water deficit affects the resistance of
lodgepole pine to MPB, as well as
subsequent development and survival
of larvae and associated blue-stain
fungi. Water deficit is the yearly sum
of (rainfall-evapotranspiration) in
months with mean air temperature >0)
Source: Adapted from Safranyik et al. 2010, 439

Although in close proximity to Canada, the contiguous United States possess
different climate regions and vegetation types than that of Canada. However, these

12
studies developed the variables which subsequent studies have deemed particularly
important when studying the MPB. Two dominant factors from the Safrayik et al. model:
winter temperature above -40 degrees Fahrenheit and the mean maximum temperature for
August over 65 degrees Fahrenheit (Safranyik et al. 2010, 439) are important benchmarks
for the beetle. MPB larva cannot survive below -40 degrees F, and optimum emergence
occurs at a 65 degree mean August temperature. Precipitation is also an important
variable, as precipitation during the growing season greatly affects the ability for trees to
defend themselves during a MPB attack.

2.2 Vegetation
Throughout the contiguous United States the MPB has predominantly infested
Lodgepole Pine (Pinus contorta), yet can also inhabit 21 other species of Pine. The Jack
Pine (Pinus banksiana), found east of the Rocky Mountains, is predicted to be the second
most suitable tree for the MPB due to its biological similarities and ecological niche
(Safranyik et al. 2010, 419). Unlike the Lodgepole, the Jack Pine does not produce the
same protective resin to resist MPB infestation by “pitching out” the beetle. Research in
Canada has shown that naïve hosts (forests that have not gone through a beetle attack in
the past) are more susceptible to attacks (Cudmore et al. 2010). As the MPB has spread
into the western edge of the Jack Pine range of Alberta, Canada, research into how native
pathogens would affect the Jack Pine’s defense against the MPB have begun (Klutsch
and Erbilgin 2012). Hydration requirements are also an important factor, but so far have
only been used when evaluating Lodgepole Pine; other species are still untested

13
(Safranyik et al. 2010, 421). As such, it is unknown how the Jack Pine’s defenses to the
MPB may be affected by drought.

2.3 Presence/Absence or Presence-Only Data
One of the most powerful new tools in predicting MPB impacts is habitat
modeling through GIS tools such as Maxent. Maxent is a type of Species Distribution
Model (SDM) that uses presence only data to predict the habitat of a given species.
Typically, SDMs will look at both presence and absence data to give a more complete
picture of the species habitat. However, historic data is often given only as presence data;
a particular species is known to exist at this location, but cannot be confirmed to be
absent at a nearby location. Because of this, Maxent has become increasingly popular as
large presence only datasets become more widely available and concern over climate
change grows (Phillips and Dudik 2007).
Maxent can be used to determine the density of a species within its’ habitat, or to
predict what area may be suitable for a species outside of its current habitat. To create a
model, Maxent randomly generates background points to compare against observed
presence data; all locations are equally likely to be sampled. The range of your
background points should be based on your ecological interests – the current habitat
extent or a wider range (Merow et al. 2013). In addition to presence data for your species
of interest, Maxent models also require importing raster layers that describe the
environmental conditions intended to measure against the study species. The WorldClim
BioClim 19 climatic variables are frequently used as the environmental variables, as they
contain variables such as temperature and precipitation.

14
2.4 Past Maxent study of the Mountain Pine Beetle
In 2011, Maxent was used to evaluate three different bark beetles’ habitat,
including the MPB habitat under current and future climate scenarios in eight Rocky
Mountain states in the western United Sates (Evangelista et al. 2011). These states are
currently habited by the MPB. The parameters used are detailed below, followed by an
analysis.

2.4.1 Biological Data
The United States Forest Service annually conducts aerial surveys and publishes
Insect and Disease Detection Survey (IDS) data for their survey areas, including the
acreage damaged by the MPB that year. This dataset is “the most accurate representation
of Mountain Pine beetle damage” (USDA Forest Service). After a successful attack, trees
are broken down into three classifications; green attack, red attack, and gray attack
(Wulder 2005, 18-41). Trees are not easily identifiable from the air until the red attack
phase, a year after the attack has begun. Evangelista et al. relied on a range of data from
this Forest Service data set, dating from 1991 to 2008, the most current species presence
data at the time.

2.4.2 Environmental Variables
The WorldClim’s 19 bioclimatic variables were used to conduct this analysis are
detailed in Table 2. This dataset is available as individual rasters spanning the inhabited
continents, presented as latitude/longitude coordinates in WGS84 with approximately
1km
2
cell size.

15

Table 2 WorldClim Bioclimatic Variables
BIO1 = Annual Mean Temperature
BIO2 = Mean Diurnal Range (Mean of monthly (max temp - min temp))
BIO3 = Isothermality (BIO2/BIO7) (* 100)
BIO4 = Temperature Seasonality (standard deviation *100)
BIO5 = Max Temperature of Warmest Month
BIO6 = Min Temperature of Coldest Month
BIO7 = Temperature Annual Range (BIO5-BIO6)
BIO8 = Mean Temperature of Wettest Quarter
BIO9 = Mean Temperature of Driest Quarter
BIO10 = Mean Temperature of Warmest Quarter
BIO11 = Mean Temperature of Coldest Quarter
BIO12 = Annual Precipitation
BIO13 = Precipitation of Wettest Month
BIO14 = Precipitation of Driest Month
BIO15 = Precipitation Seasonality (Coefficient of Variation)
BIO16 = Precipitation of Wettest Quarter
BIO17 = Precipitation of Driest Quarter
BIO18 = Precipitation of Warmest Quarter
BIO19 = Precipitation of Coldest Quarter
Source: WorldClim - Global Climate Data

For the future climate models, three were used under two different carbon
emission scenarios for both 2020 and 2050. These models were the Canadian Centre for
Climate Modeling and Analysis (CCMA), the Hadley Centre Coupled Model x3
(HadCM3), and the Commonwealth Scientific and Industrial Research Organization
(CSIRO). The emission scenarios represented a conservative (a2a) and liberal (b2a)
estimation of carbon levels. This resulted in six different possibilities for both 2020 and
2050. Additionally, they averaged the three models for each emissions scenario for both
2020 and 2050.

16
2.4.3 Analysis
The Maxent model was considered to work reasonably well for current climate
scenarios with an average AUC of 0.898 (+- 0.003). Precipitation during the warmest
quarter (Bio18) was shown to be the best predicting variable, followed by the mean
temperature of the warmest quarter (Bio10). The Maxent internal jackknife test, (another
method of testing the importance of each variable) showed that Bio18 and Bio12
(precipitation in the warmest quarter) as the best predictors for range.
All three climate models, CCMA, HadCM3, and CSIRO, predicted a decrease in
habitat suitability for the MPB. Taking the Maxent results, they then compared them to
the LANDFIRE Vegetation Type maps (Table 3). This method was chosen as vegetation
is not homogenous in growth, and the MPB have begun infesting urban forests beyond
what is represented by the vegetation level. As such, the results are shown both before
and after being overlaid with the vegetation layer.

Table 3 Percent Habitat Change

Year

Emission
Scenario
Total Habitat
Decrease
% Current
Habitat
Suitable
% New Habitat % Previous
Habitat
Decrease
2020 a2a 16% 74% 9% 26%
2020 b2a 27% 64% 9% 36%
2050 a2a 28% 58% 11% 42%
2050 b2a 28% 57% 15% 43%
Source: Evangelista et al. 2011

17
Between 1990 and 2001 the MPB infested less than 400,000 ha in the United States, or
4,000 Square Kilometers. In 2008, 25,000 Square Kilometers were infested, a 625%
increase. Modeled results for 2020 and 2050 range greatly (
Table 4).

Table 4 Square Kilometers
Year Model Square
Kilometers
Model Square
Kilometers
2020 CCCMA b2a 208,400 HadCM3 b2a 158,600
2050 CCCMA b2a 226,700 HadCM3 b2a 132,700
Source: Evangelista et al. 2011

The results showed an overall shift in regions hospitable to the MPB. However,
the results are to be viewed as possible hypothesis for beetle habitat (Evangelista et al.
2011, 314). Uncertainties in the study did arise, such as contrasting predictions between
models for the Western Pine Beetle (Dendroctonus brevicomis). Additionally, the results
for the MPB indicate that it currently inhabits only a small portion of its potential range.
“…results for current climate conditions suggest that an area of 244,800 km^2 is suitable
for the Mountain pine beetle. In 2008, mountain pine beetle infestation reached
approximately 25,000 km
2
(USDA, 2009) or one-tenth of the predicted suitable habitat”
(Evangelista et al. 2011).
Evangelista et al. concluded that additional studies at different scales for bark
beetle infestation would be necessary. Additionally, the climate models that they used
were from the CMIP3 (Coupled Model Intercomparison Project), developed in 2001,
were almost out dated at the time, and they believed the new models should be tested
once available.

18

2.5 Maxent with CMIP5
In 1995, the Working Group on Coupled Modelling under the World Climate
Research Programme started studying and standardizing atmosphere-ocean general
circulation models as the Coupled Model Intercomparison Project (CMIP). New models
are continuously developed and fine-tuned, with the most recent set released in 2013.
These models, the CMIP5 General Circulation models, were released alongside research
evaluating the models to be used for future climate scenario studies.
As the transition from CMIP3 to CMIP5 models progresses, research using the
CMIP5 climate models is starting to be published. A Maxent study, Forecasting
Distributional Responses of Limber Pine to Climate Change at Management-Relevant
Scales in Rocky Mountain National Park (Monahan et al. 2013) was published December
31, 2013 using the CMIP5 models. In addition to the 19 BioClimatic variables for the
current climate conditions, this study used a downscaled version of 31 different climate
models under two different emission scenarios. When looking at a regional issue, it has
been determined that a downscaled version of global climate models provide the
necessary resolution needed for the smaller scale and eliminates error that would be
present using a global model for a regional study (Thrasher et al. 2013) This study looked
at a broad range of variables, yet most notably found an increase in elevation for limber
pine under both emission scenarios.

19

CHAPTER THREE: METHODS
In the face of a changing climate around the world, many unique biological responses are
beginning to appear. As climate change progresses, the areas habitable for the Mountain
Pine Beetle are predicted to change. As new areas become hospitable, the MPB is
predicted to expand beyond its historical habitat due to amenable climate variables such
as warmer winters and weaker target species, in addition to a depletion of their traditional
host forests (Evangalista et al. 2011). Evaluating the continental 48 states of America as a
whole will provide a broader look at areas hospitable to the MPB. Previous studies have
looked at the MPBs historical habitat within the western US and Canada, however have
not yet focused on the eastern United States or California. Specifically, the area around
the states of Michigan, New York and California will be of interest as possible new
territory for the MPB.

3.2 Research Design
In this study, Maxent (Maximum Entropy Modeling) will be used to create a
Species Distribution Model for the MPB across the United States. Maxent Software
Version number 3.3.3K is available for free online (Phillips et al. 2006). The creators of
the Maxent software provide a tutorial guiding the user through the features and
capabilities of Maxent. Their tutorial has downloadable data for use in following the
lessons. Additionally, several university professors have tutorials posted online that are
useful in gaining a broader understanding of Maxent. These tutorials also explain how to
use ArcGIS to prepare data for Maxent and to view and interpret results.

20
Maxent results are displayed as an html file, with additional capability to edit
results. Several charts are produced, including the Area Under the Receiver Operating
Characteristic (ROC) Curve, or the ACU. This shows how well the model preforms, with
a value of .5 indicating the results could be random and confidence increasing the nearer
to 1.0. Two different results test the contribution of the environmental variables. The
Analysis of Variable Contributions shows the percent predictive contribution each
variable contributes to the model, while the Jackknife tests identify the most important
variables by running a test for each variable in isolation and comparing it to all of the
variables.
Maxent produces a raster that automatically displays habitat suitability as a 0 – 1
range, with a habitat suitability threshold defined by the user. A 90% sensitivity criterion
will be used to distinguish suitable from unsuitable habitat. These graphic images will
then be transferred to ArcGIS, using the ASCII to Raster tool. The results can then be
compared to past results to further analyze MPB habitat suitability.

3.3 Research Data
Multiple types of data are needed for this Maxent study. The data includes
biological data for the species of interest, in this case the presence data for the MPB.
Current environmental data is needed to train Maxent on what conditions the MPB lives
in. Lastly, future climatic data is needed to then project a prediction of where the MPB
may reside in the future.

21
3.3.1 Biological Data
The United States Forest Service’s IDS data for 2014 is currently available for
download. This includes damage identified in the year 2014 only; however it is not
necessarily unique to 2014. Some of this area may overlap with damage identified in
previous years. This data is downloadable through a database created for each region,
with the damage shapefiles related to a descriptive table. To receive only the MPB data,
the shapefile attribute table is joined with the connected table, and Select by Attribute is
used to select only areas damaged by the Mountain Pine Beetle. No unknown values
where included. The selected attributes where then exported into a MPB shapefile for
each region, and merged to show MPB presence polygon data for the continental United
States. Shown in Figure 6, this data shows the 1,781,025.4 acres of damage inflicted by
the MPB across 12 states in 2015, or 7,207.5 km².

Figure 6 2014 MPB Survey Data (Forest Service 2014)

22
In contrast to the MPB data obtained, Maxent models rely on point data to
identify species occurrence. While appropriate for modeling, most species’ habitat is not
confined to one XY location. MPBs can cover significant territory, and are capable of
traveling many miles to infest a forest (Nikiforuk 2011). As such, this polygon data
collected by the USDA Forest Service must be transformed into point data that accurately
represents the breadth of infestation without overestimating spread.
After querying the dataset to extract MPB data, the polygons where ready for the
next step. A bioclimate raster layer was turned into point data, using the “Raster to Point”
tool, creating a point within each raster cell. The raster used was geographically projected
in WGS 84, but the cell size had been standardized to the smallest cell. This prevents a
bias toward the northern region of the country.

Figure 7 The raster grid cells are 30 arc-seconds: 0.008333333333333 degrees or
approximately 1 kilometer squared.

23
The bioclimatic grid points were then clipped by the MPB presence data, with the results
shown in Figure 7. This method provided a thorough sampling of points while reducing
spatial autocorrelation effects.
These points will become the presence data suitable for use in Maxent. From here,
the Add XY Coordinates tool is used to assign coordinate values for each point. The data
is then saved as a database file, and is editable as an excel sheet to include the header for
the Species, Longitude, and Latitude columns. The completed table is saved as a CSV
file. In this study there are 11,316 points, comparable to the 10,775 points used in
Evangelista’s study.

3.3.2 Environmental Variables
In addition to presence data, Maxent models also require importing raster layers
that describe the environmental conditions of specific interest to measure the study
species against. In this case, the WorldClim’s 19 bioclimatic variables were used (see
Chapter 2 Table 2). These bioclimatic variables are shown to provide more meaning then
monthly data alone, as insects are easily affected by changes in temperature (Kumar et al.
2013). Additionally, these were the environmental layers used in the previous MPB
Maxent study (Evangelista et al. 2011, 309). However, the data set described as “current”
actually includes data from approximately 1950 to 2000. This is 14 years behind the most
current MPB data. The implications of this will be covered in the results.
The raster grid from these environmental layers is required for formatting
species presence into the appropriate form for Maxent, as discussed above. Additionally,
limited editing is required to prepare the rasters themselves for Maxent, as the layers

24
must have the same cell size, extent and projection system (Young et al. 2011). This
requires co-registering each layer with the same extent, and every cell size the same. This
is done using the Extract by Raster tool within the Spatial Analyst toolbox. Each layer is
processed to have the same extent and cell size as bio_1, the smallest cell set (Young et
al. 2011). The rasters are then clipped to cover only the continental United States, using
the coordinates NASA uses for their continental United States models as discussed in
Section 1.3. These coordinates were also used as the extent for the future climate models.
At this point, the rasters were converted to the necessary ASCII format. This was done
with the Raster to ASCII tool within Conversions Tools. The resulting .ASC files are all
sent to one new directory for use in Maxent.
Future climate models are available from WorldClim as well, already bias
corrected and spatially downscaled. The 19 bioclimatic variables can be downloaded for
each model and Representative Concentration Pathways (RCP.) RCP are different from
the CMIP3 emission scenarios. Instead, they model how much greenhouse gas will be
emitted, and at what year emissions will peak. The CMIP5 provides four different
scenarios, with RCP 8.5 and RCP 4.5 comparable to A2 and B2 scenarios of CMIP3
(Maloney et al. 2013). RCP 4.5 predicts emissions will peak in 2040 and then stabilize,
while with RCP 8.5 they will continue to grow past 2100. The Community Climate
System Model, version 4 (CCSM4) was developed by the National Center for
Atmospheric Research, United States. This model was chosen as its focus is the United
States. The model’s BioClimatic variables are downloaded as GeoTiff files. These files
are then individually opened in Arc and re-projected to standardize the datum to WGS
1984. The file is then clipped to the same extent as the current bioclim variables, and

25
saved as an ASCII using the Raster to ASCII conversion tool. It is important that the
future variables have the same format and name as the current variables. This process
was expedited after the first emission scenario was complete through the use of a model,
shown in Figure 7.

Figure 8 ArcMap model used to process GeoTIFF images for Maxent

3.4 Procedures
The Mountain Pine Beetle presence data entered is in the “samples” file, and the
location of the BioClim layers folder is linked to under “Environmental Layers”. Future
projection data is entered under “Projection layers directory/file”. An output directory is
designated where results will originate. A first run with only the current data was
conducted. Each climate model and emissions scenario for 2050 and 2070 are then tested
independently. 30% of the points were withheld for model verification with ten
replications averaged (Evangelista et al. 2011).

26
The results of the Maxent run are summarized within an HTML file in the output.
This includes a graphic representation of the results. To analyze the graphic results, the
averaged ASCII file was displayed as a raster in ArcGIS using Conversion tools, To
Raster, ASCII to Raster. This has to be processed as a “Float”, as the results are a range
between 0 and 1. The preset of ‘Integer” will display as empty. This produces a gray
scale map of the image, that can be modified to a color ramp. To display a binary
suitable/unsuitable habitat, you must reclassify the raster using the Spatial Analysis
toolbox, Reclass, Reclassify. Classify the raster to have only two breaks entered
manually. The threshold for suitability will be a 90% sensitivity. This information is
available from the Maxent in the results CSV file under “10 percentile training presence
logistic threshold.” The lowest number to this threshold is classified as zero, while above
is classified as one. The results is a two coded map showing suitable versus unsuitable
habitat. The area of suitable habitat can be calculated by looking at the raster count. The
suitable area is calculated for the current climate along with each projection.

27
CHAPTER FOUR: RESULTS
The results of the Maxent models are represented both graphically and in numeric values.
Numerically, there is a decrease in suitable habitat for the MPB with an increase in time
and greenhouse gas. The maps can be seen below. A 90% sensitivity was set within the
Maxent model for determining suitability, and that numeric threshold is also noted in the
figure captions. Additionally, the models performance is evaluated, along with which
variables most effect the results.
4.1 Current Species Distribution Map
When looking at current climate variables, Maxent produced 490,075 km
2
as
suitable for the MPB, shown in Figure 9. This is almost 98% more land than the acreage
found to be inhabited by the Forest Service’s presence data.

Figure 9 Current MPB Suitability (Threshold 0.4666)

28
The scaled maps are run with a standard deviation of 10, with the high suitability set to
the same threshold as the binary maps. This result was shown to most closely match the
binary results for all of the results. At this standard deviation, the MPB suitability is
confined to the historic western states, as seen in Figure 10.

Figure 10 Current Scaled MPB Suitability

The MPN does not occupy the entire “suitable” range at all times. This is consistent with
the MPB life cycle, as it continuously moves through an area as it depletes its host. While
climatically it is still suitable, it no longer has a supply of host vegetation. These results
are also mirrored in the Evangelista et al. study. However, one element to consider is the
temporal gap between the MPB data set and the current Bioclimatic variables. The MPB
data is from 2014, while the current bioclimatic data ends in 2000. When looking at MPB

29
presence data ranging from 1997 to 2014, 93,942 km
2
have been infested by the MPB.
This data set overlaid with the area currently deemed suitable is depicted in Figure 12.
While still 80% more land is deemed suitable then inhabited in the 17 year time span,
there is significant overlap between these two layers.

Figure 11 Current MPB Suitability compared to MPB Damage from 1997 to 2014

4.2 Future Species Distribution Map
When looking into the future, results were developed for both 2050 and 2070
using the two different RCPs, shown in Table 5.

30
Table 5 Square Kilometers deemed suitable

CCSM4
2050 2070
RCP 4.5 RCP 8.5 RCP 4.5 RCP 8.5
220,177 km
2
137,220 km
2
180,273 km
2
69,161 km
2

Both RCPs show a decrease in suitable habitat from now to 2050 and again in 2070.
However, RCP 8.5 shows a great decrease then RCP 4.5. This decrease is visibly tracked
in the maps below, Figure 12 and Figure 13.

2050/RCP 4.5 Binary Results
(Threshold .4701)
2050/RCP 4.5 Scaled Results
2050/RCP 8.5 Binary Results
(Threshold .4674 )
2050/RCP 8.5 Scaled Results
Figure 12 Binary and Scaled Results for 2050

31

2070/RCP 4.5 Binary Results
(Threshold .4722)
2070/RCP 4.5 Scaled Results
2070/RCP 8.5 Binary Results
(Threshold .4685)
2070/RCP 8.5 Scaled Results
Figure 13 Binary and Scaled Results for 2070

When compared to the area Maxent determined to currently be suitable for the MPB, this
is a significant decrease. The percent decrease is shown in Table 6.

Table 6 Percent decrease from current suitable area, 490,075 km
2

2050 2070
CCSM4 RCP 4.5 RCP 8.5 RCP 4.5 RCP 8.5
Percent Decrease 55% 72% 63% 86%

32
4.3 Model Performance
Model Performance can be measured by looking at the AUC in the Maxent
results. A result of .5 is believed to be the result of a random sampling, while a score of 1
is considered perfect. All of the results were found to be significant.

4.3.1 Current
The AUC for the current climate is 0.772 with a standard deviation of 0.003,
proving the model is significant. The graphic depiction of this is shown in Figure 14.

Figure 14 Area Under ROC (Receiver operating characteristic) Curve (AUC),
Current Climate

4.3.2 Future
For the four future models, the AUC results were very similar. This similarity is to
be expected with similar data. The results, graphically similar in appearance to Figure 14,
are shown in Table 7.

33
Table 7 Future AUC Results
Model AUC Standard Deviation
2050/RCP 4.5 0.772 0.002
2050/RCP 8.5 0.773 .002
2070/RCP 4.5 0.774 .003
2070/RCP 8.5 0.774 .002

4.4 Variable Importance
Maxent produces several outputs that address each variables importance in the
results. First, the variables are ranked by variable contribution. Second, the variable
importance are represented through the jackknife test; each model is run eliminating one
variable at a time, and then by running each variable independently. This shows if there
are variables less important than others in the final result.

4.4.1 Current
When analyzing the variable contribution of the 19 Bio Climatic variables for the
current climate model, Bio 8, mean temperature for the wettest quarter, is ranked as
having the highest contribution of 39.5%. This is followed by Bio 10, mean temperature
of the warmest quarter, with 33.8%. Previous studies also showed Bio 10 as the second
leading predictor (Evangelista et al. 309, 2011.)
The results of the jackknife test show that Bio 10 had the highest gain when run
independently for all three jackknife tests- training gain, test gain, and AUC on test data.

34
4.4.2 Future
Results for all four future models closely resembled the variable importance for
the current model. Bio 8 is always the highest contribution, followed by Bio 10. The
percent contribution for each is depicted in Table 8.

Table 8 Most Influential Variables
Model Most Influential Variable; % Second Most Influential; %
2050/RCP 4.5 Bio 8; 39.5%. Bio 10; 33.2%.
2050/RCP 8.5 Bio 8; 41.8% Bio 10; 30.1%
2070/RCP 4.5 Bio 8; 39.7% Bio 10; 34.4%
2070/RCP 8.5 Bio 8; 40.5% Bio 10; 32.4%.

Bio 8, mean temperature for the wettest quarter, is ranked as having the highest
contribution This is followed by Bio 10, mean temperature of the warmest quarter
The results of the jackknife test show that Bio 10 also had the highest gain when
run independently for all three jackknife tests- training gain, test gain, and AUC on test
data. This held true for all four future models. When looking at the Jackknife test for the
AUC test data, Bio 10 results in .76, close to the .78 total for all 19 variables combined.
Bio 8 was ranked as having the second most useful information, while Bio 4 has the most
information not present in other variables.

35

Figure 15 These Jackknife results for 2050/RCP 4.5 serve as representative
Jackknife results for the future models. Results for each individual model are
available in Appendix B.

36
CHAPTER FIVE: DISCUSSION
The Maxent results show that a changing climate will have a profound effect on the
MPB. However, spatially the amount of suitable land for the MPB is predicted to
decrease rather than increase. While this goes against early worries that the beetle
epidemic would continue to increase with climate change, it does follow the trend
documented by the US Forest Service IDS data (Forest Service 2014). The climatic
variables that were of most importance to the Maxent models are variables that are
commonly looked at in past MPB research as well. As these variables have a profound
effect on the MPB, they will likely also affect the vegetation the MPB infests, leaving
many questions as to how exactly climate change will affect the MPB ecosystem.

5.1 Model Strengths and Weaknesses
Maxent was an appropriate tool to evaluate the MPB habitat over a large area.
The ability to only use presence data is important, as absence data would be difficult, if
not impossible to produce due to the large study area and the difficulty in detecting green
attack trees. All of the models proved to be statistically significant with an AUC of .77,
rounded. While there is debate over the reliability of the AUC, the resulting variable
importance ranking falls in line with the variables known to be most important to the
MPB. Additionally, a clear pattern between RCP’s is evident, further confirming the
models were significant.
There were difficulties in working with data sets that cover the span of the United
States. Preparing the data was difficult. The MPB data did not initially contain only the

37
location of the MPB, but of all pests the Forest Service surveyed. This made the data set
even larger then it needed to be, and it was not immediately apparent how to separate the
MPB data from the rest of the data. Once the table connection was identified, this was no
longer a problem. On a simpler level, the climate layers took time to load with pyramid
structures and significant storage space to process, requiring enough space to store both
the raw data and the processed data. For instance, the future climate variables were
downloaded, then processed in groups of three while the 19 GeoTiffs were accessible
from the temporary download folder. Additionally, three GB of available space was
necessary for a Maxent run to be completed.

5.2 Geographic Results
When looking at the binary suitable/unsuitable results, no major geographic shifts
were evident. However, when the default 2.5 standard deviation is used for the scaled
results, the maps showed high probability in the Great Lakes region. A subtle “medium”
suitability is seen in the results for 2050. By 2070, RCP 4.5 has a patch of high suitability
on the Canadian side of the Great Lakes across from Michigan. When looking at RCP 8.5
for 2070, the patch of high suitability has decreased from where it was with RCP 4.5 and
increased closer to Minnesota, as seen in Figure 16. These results border the Great Lakes
and may be in error, due to the lakes mimicking variables favorable to the MPB.
However, it is worth further research to determine if these anomalies are caused by the
lakes or if they are the direct results of favorable conditions for the MPB.

38

Figure 16 By 2070, both RCP 4.5 and RCP 8.5 show high suitability for the MPB
near the Great Lakes. Maps showing all results with a 2.5 Standard Deviation are
available in Appendix B.

While the results all showed the MPB remaining in the western United States,
there were major shifts in the amount of area deemed suitable. Under the most extreme
scenario, RCP 8.5, if emission continue to increase, only 69,161 km
2
would be suitable

39
for the MPB by 2070, an 86% decrease for the amount of land suitable currently. It would
be, however, still significantly greater by 89% then the 7,207 km
2
currently inhabited by
the MPB in 2014.
While no huge shifts where documented, the results do indicate a shift in
elevation for the MPB into higher elevations. This is still a notable change in territory for
the MPB, as higher elevations have different vegetation that often have not had to defend
themselves from a MPB attack before. These shifts were also echoed in the Maxent study
of Limber Pine in Rocky Mountain National Park (Monahan et al. 2013.) Further
research into how these results would affect each other would provide a clearer picture as
to what the higher elevations of the Rocky Mountains will look like throughout the next
century.

5.3 Current to 2050 and 2070 Changes
The movement from what land is currently suitable for the MPB to what land may
be suitable in the future can be depicted by comparing the rasters. The Raster Calculator
tool can be used to subtract each future projection from the current suitability. These
results are seen in Figure 17 and 18.

40

Suitable only in 2014 Suitable only in 2050/RCP 4.5 Suitable in 2014 and 2050/RCP 4.5
322,350 km² 52,452 km² 167,725 km²

Suitable only in 2014 Suitable only in 2050/RCP 8.5 Suitable in 2014 and 2050/RCP 8.5
388,987 km² 36,132 km² 101,088 km²
Figure 17 The change in suitability for 2050

41

Suitable only in 2014 Suitable only in 2070/RCP 4.5 Suitable in 2014 and 2070/RCP 4.5
358,397 km² 48,595 km² 131,678 km²

Suitable only in 2014 Suitable only in 2070/RCP 8.5 Suitable in 2014 and 200/RCP 8.5
443,228 km² 22,314 km² 46,847 km²
Figure 18 The change in suitability for 2070

42
5.4 Variable Results
The climatic variables important to this study are echoed throughout MPB
research. The Safrayik model indicates mean maximum temperature for August over 65
degrees Fahrenheit (Safranyik et al. 2010, 439) are important benchmarks for the beetle.
The longer, warmer a summer, the more time a beetle has to propagate. This is reflected
in the variables Maxent found most important in this study – Bio8, Mean Temperature of
the Wettest Quarter, and Bio 10, Mean Temperature of the Warmest Quarter.
Evangelista et al. found Bio18, Precipitation of the Warmest Quarter, as the most
important, followed by Bio 10 as well. Bio 8 and 18 are similar, both having to do with
precipitation during the growing season. This is an important variable, as precipitation
during the growing season greatly affects the ability for trees to defend themselves during
a MPB attack. While these variables have the most effect on the MPB, they likely would
affect the surrounding vegetation as well, potentially leading to even more favorable
conditions for the MPB.

5.5 Future Research
This study is the beginning of applying the CMIP5 climate models to the MPB.
As Evangelista et al. stated, “We view our models as hypotheses: possible future
scenarios of ecological change.” There are many opportunities for future studies to build
upon this process, with a variety of ecological and spatial variables possible. Future
studies should apply additional climate models to a similar process. Which climate
models are favored will continue to evolve as additional research is published using the
CMIP5 models. Additionally, this study did not include vegetation as an environmental

43
variable within Maxent as vegetation is also expected to change with climate change.
However, a larger study may apply a two-step process that models climate change’s
effect on the vegetation, and then applies those results to the MPB.
This study did not apply a sampling bias as the MPB has been recorded outside of
the sampling area (such as in Michigan) and in areas not sampled by the Forest Service,
such as metropolitan areas. Results could be compared to a study where a sampling bias
is applied under the Maxent Bias field. Similarly, this study only incorporated MPB
presence data from 2014 for the models. While this narrows in on where the MPB
currently is, the results could be compared to the data set containing survey data from
1997 to 2014. This larger data set could also be paired down further to only include data
before the start of the outbreak in 2000, which would also align with the “Current”
Bioclimatic variables that end in 2000. This option would allow for testing of the model
to see if it could accurately predict current MPB habitat.
Further, there is room for scaling in both directions. Research into individual
states and ecosystems is likely more useful for forest managers to assess the probability
of a MPB attack within their management area. However, a study into the span of North
America including both Canada and the United States may be able to give a more
complete picture to how the MPB may move in elevation and direction. A larger presence
data set that covered the United States and Canada would be necessary. Differences in
survey data would need to be accounted for, and a longer temporal range for the presence
data would likely give a more complete picture of the MPB presence by eliminating error
caused by a gap in survey method.

44
This study aimed to look at potential habitat for the MPB under climate change
scenarios. As the results show, possible implications of climate change may have
unexpected results on the expansion or contraction of species such as the MPB. The
application of this modeling can give forest managers a look at what the landscape may
look like in the future and plan accordingly. There are many possible ways Maxent and
GIS can continue to be used to model how the MPB’s habitat may change under
developing climate change scenarios.

45
REFERENCES
Berg, Edward E., J. David Henry, Christopher L. Fastie, Andrew D. De Volder, and
Steven M. Matsuoka. 2006. Spruce beetle outbreaks on the Kenai Peninsula,
Alaska, and Kluane National Park and Reserve, Yukon Territory: Relationship to
summer temperatures and regional differences in disturbance regimes. Forest
Ecology and Management, no. 277:219-232.

Carroll, Allan L., Terry L. Shore, and Les Safranyik. 2006. Chapter 6 Direct Control:
Theory and Practice. The Mountain Pine Beetle- A Synthesis of Biology,
Management, and Impacts in Lodgepole Pine.

Carroll, Allan L., J. Régnière, J.A. Logan, S.W. Taylor, B.J. Bentz, and J.A. Powell.
2006. Impacts of Climate Change on Range Expansion by the Mountain Pine
Beetle. Natural Resources Canada, Canadian Forest Service, Pacific Forestry
Centre.

Chapman, Teresa B., Thomas T. Veblen, and Tania Schoennagel. 2012. Spatiotemporal
patterns of mountain pine beetle activity in the southern Rocky Mountains.
Ecology 93, no. 10: 2175-85.

Cognato AI. and Knizek M. 2010. A Festschrift for Stephen L. Wood. In: Cognato AI,
Knizek M (Eds) Sixty years of discovering scolytine and platypodine diversity: A
tribute to Stephen L. Wood. ZooKeys 56: 1-5.

Cudmore, T. J., Björklund, N., Carroll, A. L. and Staffan Lindgren, B. (2010), Climate
change and range expansion of an aggressive bark beetle: evidence of higher
beetle reproduction in naïve host tree populations. Journal of Applied Ecology
47: 1036–1043. doi: 10.1111/j.1365-2664.2010.01848.x

Elith, Jane, Steven J. Phillips, Trevor Hastie, Miroslav Dudik, Yung En Chee, and Colin
J. Yates. 2011. A statistical explanation of MaxEnt for ecologists. Diversity and
Distributions 17: 43-57.

Evangelista, Paul H., Sunil Kumar, Thomas J. Stolhgren, and Nicholas E. Young. 2011.
Assessing forest vulnerability and the potential distribution of pine beetles under
current and future climate scenarios in the Interior West of the US. Forest
Ecology and Management 206: 307-316.

USDA Forest Service. Explore Forest Insect and Disease Conditions in the United States
using Forest Health Protection Mapping and Reporting Tools. 2014.
http://foresthealth.fs.usda.gov/portal

Halter, Reese. 2011. The Insatiable Bark Beetle. Toronto, Canada: Rocky Mountain
Books.

46
Jenkins, Michael J., Justin B. Runyon, Christopher J. Fettig, Wesley G. Page, and
Barbara J. Bentz. 2013. Interactions among the mountain pine beetle, fires, and
fuels. Forest Science 60.

Klutsch, Jennifer, and Nadir Erbilgin. 2012 Interaction of an Invasive Bark Beetle with a
Native Forest Pathogen: Potential Effect of Dwarf Mistletoe on Range Expansion
of Mountain Pine Beetle in Jack Pine Forests. University of Alberta, Department
of Renewable Resources.

Kumar, S., Stohlgren, T. J. 2009. Maxent modeling for predicting suitable habitat for
threatened and endangered tree Canacomyrica monticola in New Caledonia.
Journal of Ecology and Natural Environment 1(4): 94 – 98.

Leatherman, D.A., I. Aguayo, and T.M. Mehall. 2011. Mountain Pine Beetle Fact Sheet
#5.528, Colorado State University.

Logan, Jesse A., and Barbara J. Bentz. 1999. Model analysis of mountain pine beetle
(Coleoptera: Scolytidae) seasonality. Environmental Entomology 28, no. 6: 924-
34.

Logan, J. and Powell, J. 2011. Ghost forests, global warming, and the mountain pine
beetle (Coleoptera: Scolytidae). American Entomologist, 47(3): 160-173.

Maloney, Eric D., Suzana J. Camargo, Edmund Chang, Brian Colle, Rong Fu, Kerrie L.
Geil, Qi Hu, Xianan Jiang, Nathaniel Johnson, Kristopher B. Karnauskas, James
Kinter, Benjamin Kirtman, Sanjiv Kumar, Baird Langenbrunner, Kelly
Lombardo, Lindsey N. Long, Annarita Mariotti, Joyce E. Meyerson, Kingtse C.
Mo, J. David Neelin, Zaitao Pan, Richard Seager, Yolande Serra, Anji Seth, Justin
Sheffield, Julienne Stroeve, Jeanne Thibeault, Shang-Ping Xie, Chunzai Wang,
Bruce Wyman, and Ming Zhao. 2013. North American Climate in CMIP5
Experiments: Part III: Assessment of Twenty-First-Century Projects. J. Climate,
27, 2230–2270.

Merow, C., Smith, M., & Silander, J. 2013. A practical guide to MaxEnt for modeling
species’ distributions: What it does, and why inputs and settings matter.
Ecography 1058-1069.

Monahan WB, Cook T, Melton F, Connor J, Bobowski B. 2013. Forecasting
Distributional Responses of Limber Pine to Climate Change at Management-
Relevant Scales in Rocky Mountain National Park. PLoS ONE 8(12): e83163.

Nikiforuk, Andrew. 2011. Empire of the Beetle: How Human Folly and a Tiny Bug Are
Killing North America’s Great Forests. Vancouver, BC Canada: Greystone Books
and David Suzuki Foundation.

Phillips, Steven. A Brief Tutorial on Maxent. AT&T Research.

47

Phillips, Steven J. and Miroslav Dudik. 2007. Modeling of species distributions with
Maxent: new extensions and a comprehensive evaluation. Exography 31: 161-
175.

Photos, Mountain Pine Beetle. British Columbia Government.

Photos, Mountain Pine Beetle. National Park Service.

Regniere, Jacques, and Barbara Bentz. 2007. Modeling cold tolerance in the mountain
pine beetle, Dendroctonus pnderosea. Journal of Insect Physiology no. 53:559-72.

Rosner, H. (2015, April 1). The Bug That’s Eating the Woods. National Geographic.

Safranyik, L., A.L. Carroll, J. Re´gnie` re, D.W. Langor, W.G. Riel, T.L. Shore, B. Peter,
B.J. Cooke, V.G. Nealis, S.W. Taylor. 2010. Potential for range expansion of
mountain pine beetle into the boreal forest of North America. Entomological
Society of Canada 142: 415–442.

Six, D.L. and M. J. Wingfield. 2011. The role of phytopathogenicity in bark beetle-
fungus symbioses: A challenge to the classic paradigm. Annual Review of
Entomology 56: 255-272.

Thrasher, B., J. Xiong, W. Wang, F. Melton, A. Michaelis and R. Nemani. 2013.
Downscaled Climate Projections Suitable for Resource Management. Eos
Transactions, American Geophysical Union 94(37), 321.

White, Joanne C., Michael A. Wulder, Darin Brooks, Richard Reich, and Roger D.
Wheate. 2005. Detection of red attack stage mountain pine beetle infestation with
high spatial resolution satellite imagery. Canadian Forest Service, Mountain Pine
Beetle Initiative Working Paper 2005-24.

Wood, Stephen L. 1982. The bark and ambrosia beetles of North and Central America
(Coleoptera: Scolytidae), a taxonomic monograph. Memoirs of the Great Basin
Naturalist.

WorldClim - Global Climate Data. http://www.worldclim.org/

Wulder, Michael, and Caren Dymond. 2003. Remote sensing technologies for mountain
pine beetle surveys. Mountain Pine Beetle Symposium: Challenges and Solutions.
Canada.

Wulder, Michael A., Caren C. Dymond, and Joanne White. 2005. Remote sensing in the
survey of mountain pine beetle impacts: review and recommendations. Natural
Resources Canada, Canadian Forest Service, Pacific Forestry Centre. BC-X-401.

48
Young, Nick, Lane, Carter, Evangelista, Paul. 2011. A MaxEnt Model v3.3.3e Tutorial
(ArcGISv10). Natural Resource Ecology Laboratory at Colorado State University
and the National Institute of Invasive Species Science.

49
APPENDICES
Appendix A – Jackknife Tests
Current

50

2050, RCP 4.5

51

52
2050, RCP 8.5

53

2070 RCP 4.5

54

55
2070 RCP 8.5

56

57
Appendix B – Stretch Maps with 2.5 Standard Deviation
Current

2050 RCP 4.5

58
2050 RCP 8.5

Abstract (if available)

Abstract The Mountain Pine Beetle (Dendroctonus ponderosae) is a unique indicator species in the face of climate change. Since the beginning of this century, it has expanded from its historic territory in the Rocky Mountains at an unprecedented rate. As climate variables continue to change, it is uncertain how the MPB will spread throughout the continental United States. Existing habitat models have studied the current MPB territory, but have not yet been expanded to look at how a changing climate might influence the habitable range for the MPB. In response to recent climate shifts, host tree species have become increasingly susceptible to MPB attack. As their historical habitat is consumed the MPB may also be expanding into new host species. This study applied Maximum Entropy modeling (Maxent) processes to look at habitat suitability for the Mountain Pine Beetle under future climate scenarios. Results for two different emissions scenarios for 2050 and 2070 both showed a change in the MPB’s range across the United States. Habitable areas became more concentrated to cooler areas, typically at higher elevations. These models show that as climate change progresses, the Mountain Pine Beetle will be a dynamic variable in forest management across the country as it alters not only its distribution, but also impacted species. Maxent modeling techniques allow a look into the future under varying scenarios to effectively predict the impacts of climate change on the Mountain Pine Beetle and its presence in our forest system.

Linked assets

University of Southern California Dissertations and Theses

Conceptually similar

PDF

Modeling burn probability: a Maxent approach to estimating California's wildfire potential

PDF

Building better species distribution models with machine learning: assessing the role of covariate scale and tuning in Maxent models

PDF

Habitat suitability modeling of Mexican spotted owl (Strix occidentalis lucida) in Gila National Forest, New Mexico

PDF

Predicting Hydromantes shastae occurrences in Shasta County, California

PDF

Predicting archaeological site locations in northeastern California’s High Desert using the Maxent model

PDF

Using volunteered geographic information to model blue whale foraging habitat, Southern California Bight

PDF

Detection and accuracy assessment of mountain pine beetle infestations using Landsat 8 OLI and WorldView02 satellite imagery: Lake Tahoe Basin-Nevada and California

PDF

Identification and analysis of future land-use conflict in Mecklenburg County, North Carolina

PDF

A Maxent-based model for identifying local-scale tree species richness patch boundaries in the Lake Tahoe Basin of California and Nevada

PDF

Spatial distribution of the greater sage-grouse in the Powder River Basin in northeastern Wyoming

PDF

A comparison of GLM, GAM, and GWR modeling of fish distribution and abundance in Lake Ontario

PDF

The wildland-urban interface in Lassen County, California: a change analysis 2000–2015

PDF

Using Maxent to model the distribution of prehistoric agricultural features in a portion of the Hōkūli‘a subdivision in Kona, Hawai‘i

PDF

Utilizing GIS and remote sensing to determine sheep grazing patterns for best practices in land management protocols

PDF

Using Landsat and a Bayesian hard classifier to study forest change in the Salmon Creek Watershed area from 1972–2013

PDF

Spatial distribution of the Nile crocodile (Crocodylus niloticus) in the Mariarano River system, Northwestern Madagascar

PDF

Spatial analysis of human activities and wildfires in the Willamette National Forest

PDF

Quantifying changes in monarch butterfly habitat in California

PDF

A critical assessment of the green sea turtle central west Pacific distinct population segment utilizing maxent modeling on nesting site locations

PDF

Investigation and analysis of land use/land cover to determine the impact of policy developments in cities

Asset Metadata

Creator Dowling, Caitlan R. (author)

Core Title Using Maxent modeling to predict habitat of mountain pine beetle in response to climate change

School College of Letters, Arts and Sciences

Degree Master of Science

Degree Program Geographic Information Science and Technology

Publication Date 09/24/2015

Defense Date 09/08/2015

Publisher University of Southern California (original), University of Southern California. Libraries (digital)

Tag Maxent,mountain pine beetle,OAI-PMH Harvest

Format application/pdf (imt)

Language English

Contributor Electronically uploaded by the author (provenance)

Advisor Longcore, Travis (committee chair), Ruddell, Darren (committee member), Swift, Jennifer (committee member)

Creator Email caitlanryan@gmail.com,crdowlin@usc.edu

Permanent Link (DOI) https://doi.org/10.25549/usctheses-c40-186456

Unique identifier UC11276342

Identifier etd-DowlingCai-3949.pdf (filename),usctheses-c40-186456 (legacy record id)

Legacy Identifier etd-DowlingCai-3949.pdf

Dmrecord 186456

Document Type Thesis

Format application/pdf (imt)

Rights Dowling, Caitlan R.

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 a...

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