University of Southern California Dissertations and Theses

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

(USC Thesis Other)

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

PDF

Download a page range

Download transcript

Copy asset link

Request this asset

Transcript (if available)

Content

Detection and Accuracy Assessment of Mountain Pine Beetle Infestations
Using Landsat 8 OLI and WorldView02 Satellite Imagery

Lake Tahoe Basin-Nevada and California

by

Norman Nash

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)

May 2016

Copyright © 2016 by Norman Nash

This thesis is dedicated to my parents Kris and Norman Nash Sr. I could not have
completed this or any other goal in life without your support and love.
Table of Contents

List of Figures .................................................................................................................... vi
List of Tables ................................................................................................................... viii
Acknowledgements ............................................................................................................ ix
List of Abbreviations ...........................................................................................................x
Abstract ............................................................................................................................. xii
Chapter 1 Introduction .........................................................................................................1
1.1 Mountain Pine Beetle and Infestations ...................................................................1
1.2 Region 5 Forest Management Objectives ...............................................................2
1.3 Role of Remote Sensing in MPB Detection and Monitoring ................................4
1.4 Research Questions ................................................................................................6
1.5 Thesis Outline ........................................................................................................8

Chapter 2 Related Work .....................................................................................................10
2.1 Lodgepole Pine Tree ...........................................................................................10
2.2 Mountain Pine Beetle ............................................................................................11
2.2.1. Mountain Pine Beetle population groups ....................................................12
2.2.2. Mountain Pine Beetle infestation symptoms ..............................................13
2.3 Scale and Detection using RS imagery .................................................................15
2.4 Regional, Landscape and Local Scale ..................................................................15
2.5 Remote Sensing and MPB Infestations .................................................................17
2.5.1. Indices used for MPB detection ..................................................................18
2.5.2. MPB detection using Multi date Landsat TM and ETM+ imagery ............19
2.5.3. Multi date analysis using Landsat Imagery and EWDI ..............................21
2.5.4. Accuracy assessment of EWDI and Supervised Classification ................23
2.5.5. RS unsupervised classification of IKONOS imagery ...............................25
2.6 Applications of Previous Studies ..........................................................................31

Chapter 3 Data and Methodology ......................................................................................32
3.1 Study Area ...........................................................................................................32
3.2 Data ......................................................................................................................34
3.2.1. Types of data .............................................................................................35
3.2.2. Landsat 8 OLI 30m spatial resolution imagery .........................................36
3.2.3. WorldView02 2m spatial resolution imagery ...........................................37
3.2.4. Aerial Detection Survey sketch maps .......................................................40
3.2.5. Ground Truthed Validation Points ..............................................................40
3.3 Procedures and Analysis ......................................................................................41
3.3.1. Pre-processing ...........................................................................................42
3.3.2. Maximum Likelihood Supervised Classification ......................................43
3.3.3. Vegetation Classification Map ..................................................................44
3.3.4. Accuracy Assessment Methods Using Ground Truthed Data ..................45
3.3.5. TCT and EWDI Calculations ......................................................................48

3.4 Overview of Methods ..........................................................................................51

Chapter 4 Results ...............................................................................................................52
4.1 Aerial Detection Survey Data ..............................................................................54
4.2 Ground Truthed Validation Points .......................................................................58
4.3 MLSC Landsat 8 OLI 30m imagery ......................................................................61
4.4 MLSC WorldView02 2m imagery ........................................................................64
4.5 EWDI Landsat 8 OLI 30m imagery for 2014 and 2015 ........................................67

Chapter 5 Discussion and Recommendations ....................................................................73
5.1 Understanding Remote Sensing and the Mountain Pine Beetle ............................73
5.2 Accomplishments of Thesis Project .......................................................................74
5.3 Limitations of Thesis Project .................................................................................75
5.4 Recommendations for Future Work .......................................................................76
References ..........................................................................................................................79

vi
List Of Figures

Figure 1 Color stages of a MPB infestation ......................................................................14

Figure 2 Ground-based evidence of MPB infestations .....................................................15

Figure 3 Meddens et al. 2006 percent red attack stage from 1996-2011 ...........................20

Figure 4 Wulder et al. 2006 Enhanced Wetness Difference Index results map ...............23

Figure 5 Exclusionary mask of burn scars, roads, clear cuts, water bodies and other
cultural phenomena used by White et al. 2005 study .................................................27

Figure 6 Accuracy assessment of IKONOS unsupervised classification .........................28

Figure 7 Location of study region .....................................................................................33

Figure 8 WorldView02 2m spatial resolution imagery for July 2014 and August 2015 ...38

Figure 9 Workflow Diagram .............................................................................................43

Figure 10 Area flown by USDA Forest Service in August, 2015 and area tested for
reliability .....................................................................................................................55

Figure 11 Side view photograph of ADS site and attribute information of the ADS
sketch site ....................................................................................................................56

Figure 12 Evidence of healthy lodgepole pine trees in a mixed forest of other beetle
infestations, affecting other tree types ........................................................................57

Figure 13 Landsat 8 OLI 30m spatial resolution satellite imagery ground validation
point map ....................................................................................................................59

Figure 14 WV02 2m spatial resolution satellite imagery ground validation point map ...60

Figure 15 Landsat 8 OLI 30m spatial resolution satellite imagery Maximum
Likelihood Supervised Classification with 5 user-defined classes .............................61

vii
Figure 16 Section A, B, and C Landsat 8 OLI 30m spatial resolution satellite
imagery Maximum Likelihood Supervised Classification .........................................62

Figure 17 Maximum Likelihood Supervised Classification WV02 2m spatial
resolution satellite imagery with 5 user-defined classes .............................................65

Figure 18 Larger scale of sections A, B, and C Maximum Likelihood Supervised
Classification ...............................................................................................................66

Figure 19 EWDI calculation using the 2014 and 2015 TCT wetness indices for
Landsat 8 OLI 30m spatial resolution imagery ...........................................................68

Figure 20 Section A lodgepole red attack 2014 and 2015 Landsat 8 OLI 30m TCT
wetness and EWDI ......................................................................................................69

Figure 21 Section A lodgepole dead attack 2014 and 2015 Landsat 8 OLI 30m TCT
wetness and EWDI ......................................................................................................69

Figure 22 Section B lodgepole red attack 2014 and 2015 Landsat 8 OLI 30m TCT
wetness and EWDI ......................................................................................................69

Figure 23 Section C lodgepole red attack 2014 and 2015 Landsat 8 OLI 30m TCT
wetness and EWDI ......................................................................................................70

viii
List Of Tables
Table 1 Bands, wavelengths and applications ..................................................................17

Table 2 Band combinations used to create indices ...........................................................18

Table 3 User’s, Producer’s and Overall accuracy assessment of RS methods .................24

Table 4 Error matrix used for accuracy assessment of supervised classification .............25

Table 5 Accuracy assessment results of unsupervised classification methods of
IKONOS 4m spatial resolution imagery .....................................................................29

Table 6 Datasets used in this project .................................................................................35

Table 7 Landsat 8 OLI band characteristics and applications ..........................................37

Table 8 WV02 Band characteristics and applications ......................................................39

Table 9 Example of an error matrix ..................................................................................47

Table 10 TCT coefficients for Landsat 8 OLI ..................................................................50

Table 11 Maximum Likelihood Supervised Classification 2015 Landsat 8 OLI 30m
spatial resolution imagery error matrix for sections A, B, and C ...............................63

Table 12 Maximum Likelihood Supervised Classification 2015 WV02 2m spatial
resolution imagery error matrix for sections A, B, and C ...........................................66

Table 13 2014 and 2015 TCT wetness and EWDI pixel values of ground validation
points per 30m
2
pixel ..................................................................................................70

ix
Acknowledgements

Personally, I would like to thank Suzanne Lindell, Amanda Nash, Loretta Lynn Nash,
and Nick Davis for keeping me company during all the Mountain Pine Beetle infestation
hunts. Your company and support throughout this entire process means the world to me.
Suzanne Lindell, I would like to thank you for your patience with me in our relationship.
You have taught me how patience is a virtue and I do commend yours. Thank you to all
my family members (too many to list) that have provided considerate emotional support
during the last two years leading up to the completion of this project.
Professionally, I would like to begin with the two that got me to where I am now.
Dr. Kate A. Berry and Dr. Scott A. Mensing, thank you very much for the continuous
support academically. You both taught me how to appreciate my surrounding world and
to learn from it whilst expanding my knowledge within the field of Geography. I would
like to thank my committee members Dr. Robert Vos and Dr. Su Jin Lee for providing all
your feedback and support in this project. Dr. Fleming, I thank you for advising this
project and providing your expertise, recommendations and personal support all which
was valuable in the completion of this project. I am thankful to Richard Tsung for
repeatedly expanding the storage space on the USC server to accommodate exceedingly
large imagery files. A special thanks to Devon Libby at Digital Globe Foundation for
providing an imagery grant for the 2014 and 2015 WV02 imagery, I am amongst
colleagues that have and will continue to graciously accept your support.

x
List of Abbreviations
ADS Aerial Detection Surveys

B Blue

DEM Digital Elevation Model

DN Digital Number

ETM Enhanced Thematic Mapper

EWDI Enhanced Wetness Difference Index

G Green

GIS Geographic Information System(s)

GPS Global Positioning System

LTB Lake Tahoe Basin

LTBMU Lake Tahoe Basin Management Unit

m meter

MLSC Maximum Likelihood Supervised Classification

MPB Mountain Pine Beetle

NASA National Aeronautics and Space Administration

NDF Nevada Division of Forestry

NDMI Normalized Difference Moisture Index

NDVI Normalized Difference Vegetation Index

NIR Near Infrared

NPS National Park Service

OBIA Object Based Image Analysis

R Red

xi
RGI Red Green Index

RMSE Root Mean Square Error

RS Remote Sensing

SWIR Short Wave Infrared

TCT Tasseled Cap Transformation

TM Thematic Mapper

TOA Top Of Atmosphere

TRPA Tahoe Regional Planning Agency

USDA United States Department of Agriculture

USFS United States Forest Service

USGS United States Geological Survey

WV02 WorldView-2

xii
Abstract
This work evaluates and reports the accuracy assessment of Maximum Likelihood
Supervised Classification (MLSC) using the different stages of Mountain Pine Beetle
(MPB) infestations outside the Lake Tahoe Basin Management Unit (LTBMU) using
Landsat 8 OLI 30m and WorldView-02 (WV02) 2m (comparatively higher) spatial
resolution imagery. Using ArcGIS 10.3, the accuracy of satellite imagery using MLSC
and the Enhanced Wetness Difference Index (EWDI) provide a good comparison of the
imagery at dissimilar spatial resolutions.
MPB infestations at epidemic population levels can cause economic losses and
have detrimental effects ecologically in Lodgepole pine tree stands. Detecting endemic
populations of MPB can prevent epidemic infestations, preventing economic and
ecological losses. After pre-processing, using the different stages of the MPB
infestations as a control points, MLSC and the calculation of Tasseled Cap
Transformation (TCT) indices (e.g., to calculate EWDI) are used to assess the accuracy
of each type of imagery. The overall accuracy results of MLSC of Landsat 8 OLI 30m
(51.22%) supersede those of WV02 imagery (26.82%) and are shown in error matrices
within this thesis. Accomplishments of this project include the advantage to using WV02
imagery to locate MPB infestations at their endemic stage rather than relying on annual
ADS’s. Improvements in positional accuracy of Global Positioning System (GPS) data
collection devices and improved Remote Sensing (RS) software for image analysis may
improve this analysis.
1
Chapter 1 Introduction
The Mountain Pine Beetle (Dendroctonus ponderosae) exists among forested regions in
western North America, primarily in lodgepole pine (Pinus contorta) forests. Mountain
Pine Beetle (MPB) infestations are native to North America and therefore cannot be
eradicated. Knowledge of the size and location of MPB populations can help forest
management control the spread of endemic populations and prevent epidemic populations
from occurring. Remote Sensing (RS) methods have been shown to be an efficient way
of detecting and monitoring MPB infestations (Meddens et al. 2014; Wulder et al. 2006a;
Wulder et al. 2006b; White et al. 2005). Knowledge of the accuracy of RS techniques
per imagery type may improve the precision of future analyses of MPB infestations at
varying scales/resolutions. Testing the accuracy of these methods can potentially provide
forest management with reliable sources of imagery and suitable classification methods
used to detect and monitor MPB infestations saving time and money.

1.1 Mountain Pine Beetle and Infestations

Although native to North America, the MPB has had detrimental impacts on
lodgepole pine tree stands, their primary hosts. The MPB range spans from Mexico to
Canada and can be found at elevations from sea level up to 11,000 feet (NPS 2015). It is
normal for the MPB to kill off trees in their native biological range. From northern New
Mexico spreading north to British Colombia, the MPB has killed off more than 60
million acres of forest since the early 1990’s (Rosner 2015). MPB infestations rely on
dead timber and weak dying trees. The effects of the increasing temperatures and severe
drought throughout the Rockies and the Western United States has weakened pine
forests, expanding the endemic population to epidemic population levels beyond their
2
natural biological range. Weakened trees, from influences such as the drought, have
provided the MPB with plenty of food, helping them thrive and reproduce at epidemic
rates. Without mitigation techniques to maintain endemic populations, epidemic
populations occur, resulting in negative effects ecologically and economically. The
extent of ecological and economic impacts from the MPB infestations can vary
depending upon forest management objectives.
Economically, tree loss can be viewed as a detriment to local economy in regards
to tourism and timber production. Where the production of timber from logging is
dependent on mature lodgepole pine trees, MPB infestations can have detrimental
impacts on the economy (Amman and Schmitz 1988). Aside from direct negative
economic influences due to MPB infestations, the cost of maintaining forests due to dead
tree removal in forests that are visited by tourists increases (Amman and Schmitz 1988).
Ecologically, MPB infestations affect the overall tree cover, reducing the natural
habitat for both large and small mammals, indirectly affecting fisheries. Loss of
vegetation that naturally intercepts and absorbs water provides stability to soil. With the
loss of trees, water runoff increases which leads to erosion and sedimentation in
drainages where aquatic life persists. Detecting areas of MPB infestations at epidemic
levels using RS may assist forest managers in mapping out future mitigation techniques,
which can slow the spread of infestations reducing ecological and economic losses.
1.2 Region 5 Forest Service Management Objectives
According to the 2010 USDA Forest Service Management Guide for the MPB,
the main objective is to prevent economic impacts resulting from tree mortality.
3
Mitigation efforts of the USDA Forest Service include preventive and direct control,
focusing on forests and not specifically the MPB (USDA 2010).
Although it is expensive, preventive management strategies keep the population
of the MPB below epidemic levels by limiting the food supply which includes; (1) partial
cutting (stands experiencing a new attack), (2) clear cutting (already infested trees), (3)
prescribed cutting, and (4) species discrimination (cutting larger host trees in order to
prevent MPB population growth) (USDA 2010). Preventative management is applied in
stands that are already infested or that are nearby an infestation. These methods of
preventive control reduce the density of healthy host trees (larger diameter lodgepole),
thus reducing the rate of population growth.
Methods of direct control, used to control MPB population growth include; (1)
semiochemicals (baiting with a pheromone that attracts the MPB in order to contain the
population in one area), (2) spot baiting (used to eliminate small infestations of ~30 trees
or less), (3) mop-up baiting (used in previously treated areas), and (4) grid baiting
(containing infestations that are ~50 acres in size) (USDA 2010).
Mitigation efforts and strategies of forest management are dependent upon forest
management objectives. Depending on the scale (endemic, incipient epidemic or
epidemic) of MPB population, 30m and 2m spatial resolution imagery can both be used
in the detection and mapping of infestations. Epidemic populations that span larger areas
(>30m
2
) are visible using Landsat 8 OLI imagery, whereas endemic populations that are
intermittent throughout the forest that cover smaller areas (>2m
2
) can be detected using a
comparatively higher spatial resolution such as WV02 imagery. Forest management
4
objectives define the level of control applied to MPB infestations.
According to the forest management objectives of USDA Forest Service Region
5, in the LTBMU, forest health and undesirable changes (cause and extent) amongst are
of concern (USDA 2013). Once a MPB infestation is located (normally through ADS)
forest management objectives of the LTBMU attempt to maintain an endemic population,
preventing incipient-epidemic and epidemic populations from developing. Mitigation
efforts (thinning) have prevented epidemic populations from developing in the LTBMU,
preserving the aesthetic of the forest.
Tourists frequent the LTBMU year-round. Objectives of the LTBMU, aside from
aesthetics, are to reduce fire risk. Tree mortality (caused by MPB infestations) impact the
wildlife, plant succession, recreation, and increase fire hazards (Safranyik 1982). Fuel
reduction requires field crews to remove dead timber, which is time consuming and
expensive, directly impacting the annual budget allotted for forest management.
Detecting MPB populations using modern day satellite imagery can assist forest
management in reaching their objectives and provide a means of locating endemic,
incipient epidemic and epidemic populations so preventive and direct control can be
applied.
1.3 Role of Remote Sensing in MPB Detection and Monitoring

MPB infestations can be detected in aerial photography, verified through field
methods and mapped over a multiple years using RS imagery. Many of the areas that
MPB infestations occur are in remote regions that are steep and inaccessible by foot. The
areas of infestation are expansive and estimating tree mortality caused by MPB
5
infestations can be difficult or impossible to gather using ground crews (Wulder et al.
2006). Manual field data collection and aerial detection surveys (ADS) conducted using
aircraft and aerial photography are time consuming and expensive. RS imagery allows
easy seasonal/annual acquisition and has been used in estimating MPB cause tree
mortality that would normally not be possible using ground-based survey methods
(Wulder et al. 2006). Although RS imagery cannot completely replace all traditional
data collection methods, it compliments these methods with greater detail, assisting in
addressing the objectives of forest management (Wulder et al. 2006).
RS imagery has been used in the field of environmental science for monitoring
vegetation and the physical changes/phases that occur. The lodgepole pine tree goes
through different mortality phases when attacked (Coops et al. 2006). These mortality
phases are evident in RS imagery bands that are visible to the human eye. Enhanced
detection using computer technology enables analysts to capture multiple spectral bands
of the color spectrum including the near infrared band (NIR) of the color spectrum using
modern satellite imagery sensors.
In this, multispectral imagery (a type of RS imagery with multiple bands), allows
analysts to delineate different stages of the MPB infestation based on reflectance values
of the bands. This imagery contains the red (R), green (G), blue (B) and Near Infrared
(NIR) bands, which allows analysts to delineate different phases of the MPB infestation,
not apparent to the human eye. The multiple band reflectance values all serve a purpose
in the visual separation of the different colored stages of the pine needles that are
characteristic of MPB infestation stages. The Normalized Difference Vegetation Index
(NDVI), Tasseled Cap Transformation (TCT), and Enhanced Wetness Difference Index
6
(EWDI) are a few indices used in delineating healthy from unhealthy vegetation. Both
TCT and EWDI indices will be used in this research and are explained later in the
methods section of this project.
Using RS methods to map and detect MPB infestations, this study informs forest
managers of the capabilities and accuracies of MLSC of imagery to accurately detect and
measure the intensity of infestations using 30m and 2m spatial resolution multi-spectral
imagery. Using TCT and EWDI indices over a multi-date period (2014 and 2015), the
accuracy of detecting red attack MPB infestations was assessed. Based on the accuracy
of this analysis, locational and temporal characteristics of MPB infestations within the
study area were detected and accuracy was assessed over a two-year period using ground
truthed validation points. Locating the infested stands using RS imagery and integrating
Geographic Information Systems (GIS) to produce accurate maps of these areas may be
an efficient and cost effective management tool in locating MPB infestations.
Calculating the TCT Wetness index for 2014 and 2015 assists in calculating the EWDI.
EWDI is not directly related to the amount of precipitation or drought condition of each
respective year used in calculating the index, nor is it directly related to the classification
process of MPB infestations. This project tests the accuracy of classification methods
based on the resolution of the imagery available.
1.4 Research Questions
The MPB, biology, infestations, and impacts caused by these native beetles have
been studied for over a century. Information regarding the dynamics of the MPB in
central British Columbia date back to 1910, giving researchers data on the trends of
infestations throughout a long period of time (Lundquist and Reich 2014). Studies
7
conducted in the past include the detection of MPB infestations using RS imagery at
varying spatial resolutions. The spatial resolution of the imagery required is dependent
upon forest management objectives. Different population stages of MPB infestations can
be detected at varying spatial resolutions. Studies related to this thesis similarly involve
the use of many different types of RS imagery, the differing accuracies based on the
spatial resolution of the imagery, and the scale used for such analyses.
This project compares the accuracy of MLSC using both Landsat 8 OLI 30m and
WorldView-02 (WV02) 2m spatial resolution satellite imagery. The result of accuracy
assessment of 30m and 2m spatial resolution imagery suggests which type of imagery
best suits management objectives, depending upon the population stage and scale of
infestation. High spatial resolution (2m) satellite imagery is normally costly
($126.50/25km
2
) (Digital Globe 2015) and therefore accuracy assessment of lower spatial
resolution (30m) imagery that is readily available to the public (at no cost) may be
valuable to forest management. 30m-spatial resolution satellite imagery is readily
available on a consistent basis whereas ADS are scheduled annually, normally more
expensive, and not up-to-date. Calculating the accuracy of MLSC methods using two
types of RS imagery offers forest management with reliable options (rather than ADS)
for detecting and monitoring MPB infestations.
Testing the accuracy of MLSC using MPB infestations and their different stages
answers a few main questions throughout this thesis project. Questions involved in this
thesis include; (1) Will ground truthed validation points of red attack stage MPB
infestations help in establishing training areas in the supervised classification process for
both 30m and 2m spatial resolution imagery?; (2) What type of imagery (30m and 2m)
8
results in a higher overall accuracy using supervised classification?; and (3) Using 2014
and 2015 Landsat 8 OLI 30m spatial resolution satellite imagery to calculate the TCT
Wetness Index, is there a relationship between the wetness difference from 2014 to 2015
(EWDI) where red attack and dead attack stage MPB infestation ground truthed
validation points existed?
1.5 Thesis Outline

With the purpose of this project now established, four chapters follow that
address: (2) related work; (3) data and project methodologies; (4) results; and (5)
discussion and recommendations.
Chapter 2 includes the description of the habitat of the lodgepole pine tree and
why it serves as the primary host for the MPB. The MPB lifecycle, population size and
symptoms associated with an infestation of the lodgepole pine tree are also clarified.
Following this, RS applications used for the detection of MPB infestations are explained
through studies similar to this thesis project that have been conducted by others in the
past. These studies include the detection of MPB infestations using multi-date Landsat
TM and ETM+ 30m resolution imagery and the calculation of the TCT and EWDI. One
study mentioned in this chapter explains how unsupervised classification method can be
used to detect MPB infestations using high spatial resolution (2m) imagery.
Acknowledging the observation of unsupervised classification, the following chapters (3-
5) will explain how supervised classification and EWDI can improve on these previous
methods.
Chapter 3 includes the data and methodology used to complete this project. This
project utilizes multiple different data types including: raster 30m and 2m-spatial
9
resolution imagery, vector ADS sketch maps, vector boundary data, vector vegetation
classification maps, and vector ground truthed validation points. Descriptions of the
multiple bands of the imagery used and their applications offer an introductory
understanding of each band. ADS sketch maps are used to locate potential MPB
infestations to assist in locating the manually collected ground truthed validation points in
the field. The different types of data and their uses in this project are explained. MLSC,
leveraging unsupervised classification, methods and accuracy assessment while using an
error matrix to calculate the errors of commission, errors of omission, and overall
accuracy are explained as well.
Chapter 4 includes the results from using ground truthed validated points to
establish training areas used in the MLSC method. The accuracy assessment results from
this, are discovered in this project. Maps are presented based on the type of imagery
(30m or 2m spatial resolution), classification method (MLSC), and the calculation of the
EWDI.
Finally, Chapter 5 discusses conclusions and recommendations of this project,
articulating how these methods may be adopted and/or abandoned based on forest
management objectives. Scale, resolution and mitigation efforts are discussed within this
chapter. Technological improvements, RS processing techniques, and improved spatial
resolution of imagery used in the detection and mapping of the MPB are progressive and
need acknowledgement.

10
Chapter 2 Related Work
Remote Sensing and aerial photography provide analysts with an accurate and efficient
means of detecting, assessing, and monitoring the damage of MPB infestations
throughout North America. From New Mexico, northwest to British Columbia,
ecological and economic impacts have occurred due to the MPB infestations. The extent
of ecological and economic impacts caused by MPB infestations varies depending upon
forest management objectives and mitigation efforts. Detecting areas of MPB
infestations using RS imagery have and continue to assist forest managers in mapping out
future mitigation techniques, slowing the spread of infestations, and, thus reducing
ecological and economic losses.
2.1 Lodgepole Pine Tree
The MPB, although native to North America, has had detrimental impacts on pine
forests, mostly in the lodgepole pine tree stands throughout parts of western North
America. The primary host for the MPB is the lodgepole pine tree, which suffers
extensive mortality shortly after an initial infestation (Coops et al. 2006). Not all trees in
a lodgepole pine stand are directly affected by MPB infestations. The MPB selects trees
based on the diameter, phloem thickness, habitat type, and elevation (Roe and Amman
1970, Cole and Amman 1980).
Phloem thickness is of the depth of the inner bark of the lodgepole pine tree. An
older tree that has a larger diameter tends to have a larger phloem thickness, which is
conducive to successful brood productions of the MPB. The lodgepole pine tree
increases in size and diameter in an ideal habitat type. A study conducted by Phister et
11
al. in 1977 in the Gallatin canyon, just southwest of Bozeman Montana, concluded that
tree mortality was more prevalent in lodgepole pine trees than other types of trees such as
Douglas-fir (Pseudotsuga menziesii), Subalpine-fir (Abies lasiocarpa) and Spruce tree
(Picea) types which prevail in higher elevations (Cole and Amman 1980). The MPB has
been known to cause diebacks in lodgepole pine, Sugar pine (Pinus lambertiana), and
Whitebark pine (Pinus albicaulis) tree stands. Other insects such as the Jeffrey pine
beetle (Dendroctonus jeffreyi) and the Fir engraver (Scolytus ventralis) cause tree
mortality in mixed conifer forests, also prevalent in the LTB (Van Gunst 2012).
In higher elevations, above 8,000 feet, where the lodgepole pine tree is not as
prevalent, the MPB may require more than one year to complete its life cycle. These
higher elevations experience colder and often freezing temperatures, characteristic of fall
and winter. Successful brood production is dependent upon moderate temperatures and
therefore the reproduction cycle at higher elevations is curtailed due to freezing
temperatures. Consequently, the mortality of lodgepole pine trees caused by MPB
infestations is linked to elevation (Cole and Amman 1980).
2.2 Mountain Pine Beetle
The one-year lifecycle of a MPB goes through four stages of development: egg,
larva, pupa, and adult. The majority of the MPB’s lifespan is spent feeding on the tree the
inner phloem, beneath the bark. Once the MPB is mature, the adult emerges from
beneath the bark and migrates to other trees to attack (Amman et al. 1990). The MPB
usually feed on healthy lodgepole pine trees that are larger in diameter because of the
increased thickness of the inner phloem. Natural causes, such as wind assist the
migration of the MPB, normally within close proximity from the primary host trees
12
(Carroll 2007). Densely populated lodgepole pine tree stands contain a greater amount of
weakened trees, providing plenty of food and are more susceptible to MPB infestations.
An abundance of host trees provide the MPB not only a place to migrate, but to thrive
and reproduce at rates that create epidemic populations.
2.2.1. Mountain Pine Beetle population groups
MPB infestations are divided into four different population categories: endemic,
incipient epidemic, epidemic and post-epidemic. All of these are defined by the ratio of
the MPB population to the availability of host trees (Safranyik and Wilson 2007).
Endemic populations exist in healthy forests but naturally kill off small patches of host
trees. Incipient-epidemic populations occur between the endemic and epidemic stage
(outbreak phase) where the MPB population grows in size and concentrates into larger
infested patches of trees, surrounded by individual trees that bound the edge of a large
infestation (British Columbia Ministry of Forests 2003). Declining populations of the
MPB are referred to as the post-epidemic phase (after infestation).
Indigenous to North America, the MPB has spread outside of its normal
biological range into new areas, creating epidemic population levels. At the endemic
population stage, the MPB populations are relatively low and the size/density of
infestations (~1 tree per acre) are harder to detect than epidemic populations (Lundquist
and Reich 2014). Once the MPB infestations reach an epidemic stage, the population
increases rapidly, causing extensive damage to lodgepole pine tree stands, increasingly in
densely affected tree stands. As detailed earlier, therefore, detecting the extent and
severity of MPB infestations is crucial in forest management and mitigation planning.
13
2.2.2. Mountain Pine Beetle infestation symptoms
The MPB infests trees and the effects of these infestations occur in four different
color phases. During an infestation, the pigments in the pine needles break down and
chlorophyll is lost, causing the needles to change color (Hill et al. 1967). These four
phases include the green attack phase, yellow/green phase, red attack phase, and the grey
attack phase. The green attack phase is the initial infestation, usually occurring from
mid-July through August. Following this initial attack, the needles of the tree turn a
yellow/green (faders), which indicates the transition from green to red. The pine needles’
pigment gradually breaks down from a green to yellow, losing chlorophyll, and turn to
red at an advanced stage of infestation (Hill et al. 1967). During the red attack phase the
needles turn a reddish-brown about a year later (red attack phase). The grey attack (dead
attack) phase occurs two or more years after the green attack phase during which the trees
lose most of their foliage (Krause 2006). These four transitional color phases of the pine
needles are visible in RS imagery and can assist in the identification of the phases and
severity of MPB infestations. These physical changes are evident in the different
reflectance values found in RS imagery. These four-color phases are useful in identifying
the stages and severity of the MPB infestation using RS imagery and GIS classification
techniques.
In recent studies, loss of pigment in foliage has aided in classifying pixels as MPB
red attack stage in RS imagery. The symptoms related to MPB infestations have also
helped in establishing training areas that are used in supervised classification methods of
RS imagery. Changes in foliage color are evident in ADS data, which is used to
positively identify MPB infestation stages that can be classified in RS imagery (Meddens
14
2012; Wulder et al. 2006; Wulder et al. 2008; White et al. 2004; Franklin et al. 2003;
Wulder et al. 2009; Sharma 2007). The pigmentation (color) change in trees due to MPB
infestations can be seen clearly from an aerial photograph (Figure 1).

Figure 1 Color stages of a MPB infestation. Photo credit: Dezene Huber, University of
Northern British Columbia

With knowledge of the symptoms characteristic of MPB infestations, these
infestations are positively identified on the ground during field data collection.
Symptoms besides pigment loss in foliage include bore holes in the bark, dust at the base
of the tree from the bore holes, pitch tubes that form at the entrance of the bore holes, and
crevice like formations beneath the bark where the MPB lays its eggs. Once the MPB is
mature, it exits from beneath the bark and moves to another host tree leaving, a small exit
hole in the bark (Safranyik and Wilson 2007) (Figure 2).
15

Figure 2 Ground-based evidence of MPB infestation; above left, pitch tubes present from
MPB entrance; above middle, bark loss during winter months from birds preying on
larvae; above right, exit holes from mature MPB leaving host tree. Source: Nash 2015
2.3 Scale and Detection using RS Imagery
The Lake Tahoe Basin (LTB) is a national forest with almost 80% of the land in
the basin being managed by the United States Department of Agriculture (USDA) Forest
Service. The objectives of forest management and their mitigation efforts are the deciding
factor as to the methods used (ADS or supervised classification of RS imagery) in
detecting and monitoring MPB infestations. Mitigation efforts are dependent upon the
stage and size of the MPB infestation. Both high (2m) and comparatively lower (30m)
spatial resolution multi-spectral imagery has been used in detecting and mapping MPB
infestations. The magnitude of infestation (endemic or epidemic), at which MPB
infestations occur determines which type of imagery (30m or 2m) to use. Epidemic
population clusters can be detected using both types of imagery. Endemic population
clusters require a higher (2m) spatial resolution to detect and monitor.
2.4 Regional, Landscape and Local Scale
The management of forests in North America occurs on three different scales:
regional, landscape and local. Regional scale includes provinces, states, and forest
16
service regions. Landscape scale is smaller and includes forest districts and national
forests. Local scale focuses on new sites where the MPB has infested experiencing
minimal effects (British Columbia Ministry of Forests 2003). For example, local scale
requires imagery with a comparatively higher spatial resolution that has multiple spectral
bands. This higher spatial resolution imagery can be used to detect smaller (endemic and
incipient-epidemic) infestations (2m
2
in diameter) for object-based analysis of newly
infested trees. Larger infestations (greater than 30m
2
in diameter) can be detected using
multi spectral imagery that has a much lower spatial resolution (30m).
Detecting the extent and severity of MPB infestations at different scales is crucial
in forest management and inventory. The mapping of MPB infestations has been
ongoing since the 1960s. The advent of improved RS satellite instruments and quality of
images/processing has increased the capabilities of detecting and mapping MPB
infestations (Wulder et al. 2006). Technological advancements of sensors have resulted
in higher spatial resolution images with a shorter turn-around time. This facilitates the
production of imagery that is up-to-date and available without copyright restrictions
(Wulder and Franklin 2012). The technological capabilities of monitoring forests have
increased the demand for improved precision of data collection at different scales, which
is now possible using RS imagery. Improved methods of data collection using RS
imagery combined with traditional methods (ADS and field data collection) meeting this
demand. The latest version of ArcGIS 10.3 allows a range of data sets at varying spatial
resolutions to be processed together using a multitude of tools, resulting in enhanced
precision of monitoring forest health (Wulder and Franklin 2012).
17
2.5 Remote Sensing and MPB Infestations
RS is the activity of measuring, observing and monitoring phenomena that occurs
on the Earth’s surface without physically touching the phenomena (NASA 1998).
Measuring, observing, and monitoring these phenomena is possible by capturing images
using sensors that are, most often, onboard spacecraft or aircraft. In this thesis, RS
imagery includes aerial photographs, Landsat 8 Operational Land Imager (OLI) 30m
spatial resolution and WV02 2m spatial resolution imagery. Landsat 8 OLI and WV02
imagery are satellite collections that have multiple bands, where each band, or a
combination of bands is used for specific purposes in the classification of pixels
representing objects on earth (Table 1).
Table 1 Bands, wavelengths and applications

18
2.5.1 Indices used for MPB detection
There are multiple indices used in monitoring and detecting vegetation for
evaluating forest health. These indices include: Short Wave Infrared/Near Infrared
(SWIR/NIR), Normalized Difference Moisture Index (NDMI), Normalized Difference
Vegetation Index (NDVI), Red Green Index (RGI), Enhanced Wetness Difference Index
(EWDI), and Tasseled Cap Transformation (TCT). Relevant to this project, TCT and
EWDI indices are used in the detection of MPB infestations using RS imagery (White et
al. 2005; Wulder et al. 2006; Meddens et al. 2014). The calculation of the
aforementioned indices and their applications can be viewed in (Table 2).
Table 2 Band combinations used to create indices. Source: Gillis 2012

19
2.5.2. MPB detection using Multi-date Landsat TM and ETM+ imagery
There are currently no studies conducted in the LTB assessing the accuracy of
multi-date 30m vs. 2m-spatial resolution imagery. One study was recently conducted in
north central Colorado and southern Wyoming (similar in climate and vegetation patterns
found in the LTB), using multi-date RS imagery from 1996 to 2011 that included mixed
conifer forests at high elevations, Lodgepole Pine and Aspen at middle elevations, and
Ponderosa Pine and Douglas Fir at lower elevations. In 2008, an aerial survey completed
by the USDA Forest Service estimated 1.15 million hectares (~4500 square miles) MPB
caused tree mortality in Colorado and estimated the cost of mitigation in 2010 close to
$30 million (Meddens et al. 2014). This particular study documented the percent change
and spatial patterns of red attack tree mortality caused by the MPB. Detecting a pattern
in epidemic MPB populations assisted in mitigating the problem by reducing the
economic costs related to MPB infestations.
Meddens et al. 2014 used bands 1-5 and 7 from Landsat TM and ETM+ multi-
spectral imagery to calculate the TCT brightness/wetness/greenness indices as control
variables, as well as the NDVI in order to monitor the rate at which MPB infestations
spread using multi-date imagery. Multi-date classification of red attack pixels, calculated
from SWIR/NIR, was provided from previous research by Meddens et al. 2013. A
regression model with a Root Mean Square Error (RMSE) of 13.7 was applied to predict
the percent red attack pixels within the 30m spatial resolution pixels across the entire
Landsat scene and results were compared to ADS to assess accuracy of predictions
(Meddens et al. 2014). The results from using Landsat 30m spatial resolution imagery
delivered an average MPB mortality rate of 29.4% in primarily lodgepole pine areas at
20
mid elevations (2500m-3000m). Elevations where Ponderosa pine, Douglas-fir, and
Aspen were prevalent at lower elevations (2000m-2500m) resulted in a 5.2% mortality
rate, followed by mixed fir and lodgepole pine at high elevations (3000m-3500m) with a
16% mortality rate (Meddens et al. 2014) (Figure 3).

Figure 3 Percent red attack stage from 1996-2011. Source: Meddens et al. 2014
The coupling of the SWIR/NIR indicator of forest health and NDVI has shown
that red attack stage MPB infestations can be detected using the bands available in
Landsat 30m spatial resolution more accurately and completely using multi-date imagery
rather than using single date imagery.

21
2.5.3. Multi-date analysis using Landsat Imagery and EWDI
Another study was conducted by Wulder et al. 2006 describing the procedures for
using multi-date Landsat TM and ETM+ imagery to detect the location and severity of
MPB infestations in the red attack stage in southern British Columbia using the EWDI
(Wulder et al. 2006). The MPB had reached epidemic population levels in many parts of
British Columbia and caused extensive damage and mortality to forests (164,000 ha in
1999 to 8.5 million ha in 2005), especially the lodgepole pine tree (the primary host) due
to its size and density amongst other species of trees (Wulder et al. 2006). Measuring the
magnitude of infestations required sources beyond human means. Ground crews,
helicopter surveys, ADS and other traditional methods of data collection were not timely
or suitable for the detection and mapping of the epidemic populations of the MPB. RS
imagery proved more efficient and less expensive than previously mentioned methods
used for detecting MPB infestations at both landscape and provincial scales.
The use of multi-date imagery has proven to be more reliable in change-detection
than single date imagery due to the nature at which the foliage changed in color following
an infestation (green to red). The use of multi-date imagery also assists in the detection
of MPB infestations due to their progressive changes in the landscape (pre/post
infestation symptoms). For example, if a tree was attacked by the MPB in 2000, the red
attack phase would not be obvious until the following year (2001) due to the one year
lifecycle of the MPB and its noteworthy effects on trees. One year following a red attack
stage, the foliage dissipated and no existing foliage was apparent (grey attack stage). As
the tree loses pigmentation, it is deprived of the essential nutrients for survival, therefore
22
losing moisture. EWDI measures the wetness (moisture) difference between multi-date
imagery and is therefore a great indicator of forest health between years of RS imagery.
Imagery used in the study by Wulder et al. in their 2006 study included Landsat
imagery from 2002-2004 during the months of July, August and September. These
months were ideal for imagery acquisition as the MPB had completed its brood cycle and
migrated to the next host tree. Pre-processing methods (geometric correction,
radiometric correction) were applied by using ground control points, referencing the
imagery shortly after acquisition to ensure the accurate representation of objects upon the
landscape. The images from 2002-2004 were normalized, excluding; burn scars, clear
cuts, snow, clouds, roads, lakes and other phenomena using an exclusionary mask. This
mask reduced the chances of false positives occurring in the classification process by
excluding objects that were not intended for use in the analysis, ensuring that only target
phenomena were included in the classification. This process was commonly used when
utilizing indices to detect infestations and monitor forest health in RS imagery.
After pre-processing and normalization, the TCT wetness indices were calculated
by year (multiplying each band by a constant that is specific to the satellite sensor). This
process involved the conversion of the original bands to a set of bands (TCT wetness)
that were useful in calculating the EWDI, commonly used in vegetation monitoring
(ESRI ND) (Figure 4). This method was used to show the moisture difference in
vegetation from 2002 to 2004. MPB infestations caused moderate moisture loss in the
vegetation, whereas healthy vegetation did not (Wulder et al. 2006).
23

Figure 4 Section A, an area that was clear-cut with an obvious lack of moisture; section
B, MPB infestation where trees are losing moisture; section C, moisture increase in this
image. Source: Wulder et al. 2006
2.5.4. Accuracy assessment of EWDI and supervised classification
Accuracy of EWDI classification methods in Wulder et al. 2006 study consisted
of gathering 100 ground validated points of non-attack trees and another 100 ground
validated points of attacked trees. The overall accuracy of classification using the EWDI
was assessed by using an error matrix by calculating the cumulative of the correctly
classified sample points of each class and dividing this number by the total number of
sample points (e.g., 193/200=96.5%) (Wulder et al. 2006). The producer’s accuracy, or
percent of reference points collected that were classified using RS methods, equaled 96%
(96% of the attacked trees were identified by the RS method) (Wulder et al. 2006).
User’s accuracy showed 97% of the trees being correctly identified with an error of
commission of 3% (i.e., trees being erroneously included in the RS data) (Table 3).
24
Table 3 User’s, Producer’s and Overall accuracy assessment of RS methods. Source:
Wulder et al. 2006
Ground Truthed
non-attack trees
Ground Truthed
attacked trees
Sum User’s
Accuracy
RS non-
attack trees
96 4 100 96%
RS attack
trees
3 97 100 97%
Sum 99 101 200 Overall
accuracy 96.5%
Producer’s
accuracy
96.9% 96% Null Null
EWDI proved to be an accurate method in the classification of red attack trees
that have been infested by the MPB. This method, applied in this thesis project was
similar to the method used by Wulder et al. 2006. Table 4 shows an example of accuracy
assessment using supervised classification in a 1991 study by Congalton. In Table 4, the
producer’s accuracy represents is the ratio of the number of correctly classified ground
truthed MPB(R) to the total number of MPB(R) ground truthed points. In this example,
65 out of 75 (65/75=87%) MPB(R) ground truth points were correctly classified by the
computer. The user’s accuracy represents the probability that a MPB(R) ground truth
point is classified as MPB(R), or in this example, 65 of the ground truth MPB(R) points
out of 115 classified points is MPB(R), the probability that the MPB(R) ground truthed
points are classified by the computer is (65/115), or 57%. The overall accuracy is
calculated by dividing the total number of objects correctly classified in RS imagery by
the total number of samples, or in this example, (65 MPB(R)+81 MPB(G)+85
MPB(D)+90 Healthy(NI)/434 Total Points)=74%. The error matrix used to calculate the
accuracy assessment for MPB(G), MPB(D), and Healthy(NI) trees can also be seen in
(Table 4).
25
Table 4 Error matrix used for accuracy assessment of supervised classification. Source:
Congalton 1991
MPB(R) MPB(G) MPB(D) Healthy(NI) Row total
MPB(R) 65 4 22 24 115
MPB(G) 6 81 5 8 100
MPB(D) 0 11 85 19 115
Healthy(NI) 4 7 3 90 104
Column
Total
75 103 115 141 434

2.5.5. RS unsupervised classification of IKONOS imagery
Unlike the classification of imagery using indices, unsupervised classification
techniques are dependent upon the computer, without supervision and guidance of the
user. High-resolution imagery has a spatial resolution of 2m or less. Objects can be
detected more clearly and with less distortion than 30m spatial resolution imagery such as
Landsat 30m satellite imagery. High-resolution imagery allows the detection and
mapping of MPB infestations at a local scale, which includes smaller plots of infested
tree stands (2m
2
or greater). MPB infestations can be localized before incipient-epidemic
populations occur. Forest management may utilize local scale analysis to detect small
26
infestations so that field crews can remove the problem before it persists. High resolution
RS imagery is available commercially and can assist in the detection of MPB infestations
accurately, consistently and efficiently while providing coverage at local to landscape
scales (White et al. 2005). The downfall to using high-resolution imagery is the
associated costs as well as the coverage of certain phenomena, such as MPB infested
trees, over time. Digital Globe offers high-resolution (2m) imagery for ~$106 per 25km
2

(3 July 2015) unless acquired through an educational imagery grant, in which 1000km
2
may be granted per individual.
A study was conducted by White et al. 2005 near Prince George, British
Colombia using high resolution multi-spectral imagery testing the accuracy of IKONOS
multispectral imagery in the detection of MPB infestations during their low and medium
attack stages. Concerns regarding accuracy, consistency and timeliness of satellite
imagery were brought forth by White et al. 2005 regarding data acquisition. The
questions asked by White et al. 2005 were: (1) Is the imagery going to be accurate
enough for the analysis so the detection method is replicated? and (2) Is it possible to
acquire imagery at a high resolution per the study area in a timely manner? (White et al.
2005). These questions should pertain to all forest management regarding MPB
infestations and the use of RS imagery for detection.
Between 2002 and 2003, the MPB infestation had doubled from 2.0 to 4.2
million hectares with projected estimates as high as 7 million hectares in 2004 near
Prince George, B.C. (White et al. 2005). Rather than trying to control the epidemic
population growth, management and mitigation efforts have shifted to focus on the
27
source of the problem, trying to control future epidemic populations of the MPB. This
shift in management resulted in detecting smaller incipient-epidemic populations using
high-resolution RS imagery, followed by mitigation efforts, preventing incipient-
epidemic populations from progressing into epidemic populations.
A combination of ground validated red attack trees recorded with a handheld GPS
receiver (with a positional accuracy of 15m), ortho-rectified aerial photography, and
georeferenced IKONOS 4m spatial resolution imagery (captured the same day as the
aerial photography) were used in the detection and mapping of incipient epidemic MPB
infestations (White et al. 2005). Following this process, White et al. 2005 applied an
exclusionary image mask, including; burned areas, clear cuts, water bodies, roads,
buildings and other cultural objects (acquired from an existing vector GIS database).
This mask excluded all objects that could possibly be misconstrued as vegetation
considering the spectral variability, characteristic of the progression of MPB infestations
(Figure 5).

Figure 5 Exclusionary mask of burn scars, roads, clear cuts, water bodies and other
cultural phenomena. Source: White et al. 2005
28
Included in the accuracy assessment by White et al. 2005 was the comparison of
ground truthed locations (aerial GPS point data collected by helicopter) of moderate red
attack phase MPB infestations to the unsupervised classification results of IKONOS 4m
high-resolution imagery. True positive accuracy (red attack trees identified in aerial
photography that fall within the IKONOS classified pixels), errors of omission (red attack
trees that fall outside IKONOS imagery pixels), and errors of commission (pixels that
contain no red attack trees) were calculated using an error matrix, assessing the accuracy
of unsupervised classification of IKONOS imagery (Figure 6).

Figure 6 Above left, red attack tree crowns delineated from ADS; above middle, IKONOS
pixels representing red attack tree crowns delineated from unsupervised classification
methods; above right, example of errors of commission, omission and true positive
results. Source: White et al. 2005
IKONOS imagery spatial resolution differs from the spatial resolution of aerial
photography, therefore, a 4m buffer was applied to the ground control points to match
IKONOS 4m spatial resolution imagery. Based on the values of ground control points in
Figure 6, the true positive, errors of commission and errors of omission were calculated
(Table 5). Table 5 shows the total number of red attack ground control points and total
number of red attack points classified in IKONOS imagery that were used to assess the
29
accuracy of unsupervised classification. True positive results of moderate red attack
damage MPB infestations using a 4m buffer on IKONOS imagery showed ~92%
accuracy, omission results at ~8% and commission at ~2% (White et al. 2005). The
accuracy assessment of using high-resolution multi-spectral IKONOS imagery proved to
be an efficient approach to detecting moderate red attack MPB infestations in this
particular study.
Table 5 Accuracy assessment results of unsupervised classification methods of IKONOS
4m spatial resolution imagery using medium-density attack sites. Source: White et al.
2005
Buffer Size 1m 2m 3m 4m
Total red attack trees (air photos) 510 510 510 510
Total red attack pixels (IKONOS) 389 389 389 389
True positive (pixel count) 398 439 457 471
True positive (%)
a
78.04 86.08 89.61 92.35
Lower 95% confidence interval (%) 70.45 79.67 83.91 87.33
Upper 95% confidence interval (%) 85.63 92.49 95.31 97.37
Omission (tree count) 112 71 53 39
Omission (%)
b
21.96 13.92 10.39 7.65
Lower 95% confidence interval (%) 14.37 7.51 4.69 2.63
Upper 95% confidence interval (%) 29.55 20.33 16.09 12.67
Commission (buffer count) 25 16 13 8
Commission (%)
c
6.43 4.11 3.34 2.06
Lower 95% confidence interval (%) 3.86 2.01 1.43 0.52
Upper 95% confidence interval (%) 9 6.21 5.25 3.6
Forest management objectives determine the urgency and level of accuracy
necessary to detect and mitigate infested tree stands. As mentioned earlier, epidemic
MPB populations can have detrimental impacts both ecologically and economically.
Prevention of the progression of incipient-epidemic to epidemic MPB populations
requires a higher level of classification accuracy and higher spatial resolution imagery.
White et al. 2005 objective was to locate individual trees in order to quickly mitigate the
30
problem. The chosen IKONOS 4m spatial resolution imagery by White et al. 2005
proved to be an efficient way of completing this task. Accuracy assessment in this study
confirmed the quality of unsupervised classification techniques using IKONOS imagery.
MLSC, using ADS data and ground truthed validation points of healthy and MPB
infested trees in the LTBMU and surrounding areas may improve the classification
accuracy. Combining these data sources will help in establishing training areas from
known MPB infested sites, assisting in supervised classification methods of 30m and 2m
spatial resolution imagery. White et al. 2005 study showed that IKONOS 4m spatial
resolution multi-spectral imagery can be used in the accurate detection of incipient-
epidemic MPB infestations.
Both 30m and 4m spatial resolution RS imagery can be used for accurately
detecting MPB infestations at regional, landscape and local scales. The stage of MPB
population (endemic, incipient-epidemic, or epidemic) and forest management objectives
(i.e., control an incipient epidemic MPB population or prevent epidemic MPB population
from spreading) combined play a role in the required level of precision and accuracy of
classification needed to efficiently plan mitigation efforts. Both high and lower spatial
resolution imagery can be used as a cost effective data source to detect infestations at the
endemic, incipient-epidemic and epidemic population levels of the MPB. Results from
previous studies (Meddens et al. 2014; Wulder et al. 2006; White et al. 2005) show that
improved RS methods can be incorporated with traditional data acquisition methods in
the detection and monitoring of MPB infestations, improving the accuracy of detection of
MPB infestations.
31
2.6 Applications of Previous Studies
Previous studies by Meddens et al. 2014; Wulder et al. 2006; and White et al.
2005 have concluded that the accuracy of MPB detection using high and lower spatial
resolutions from multiple different satellite sensors can be tested using many different
methods (EWDI, supervised classification and unsupervised classification). In all three
studies, ground truthed validation points are used to test the accuracy of each method.
The combination of MLSC techniques (to establish training areas) and testing the
accuracy using ground truthed validation points for the detection and monitoring of MPB
infestations has proved to be the best way to assess the accuracy of classification methods
using either high or lower spatial resolution imagery.
Rather than relying on a single ADS, WV02 2m spatial resolution imagery is used
to locate areas of tree mortality in this thesis project. The quality of the image allows
analysts to detect areas of tree mortality without reliance on aerial surveys. Not all stands
of MPB infested trees were recorded in ADS sketch maps, seen throughout this thesis
project. The areas where MPB infestations were located, outside the LTBMU are
included in Region 5. These trees, recorded in this project were not recorded as
lodgepole pine tree mortality caused by the MPB in 2015 ADS. Using higher spatial
resolution imagery to locate lodgepole pine tree mortality, previously mentioned
classification techniques and known accuracy assessment methods, these previous studies
can be elaborated. Previous studies have focused on incipient-epidemic and epidemic
MPB population groups. This particular study focuses on the detection of endemic MPB
population clusters using the aforementioned classification and accuracy assessment
methods.
32
Chapter 3 Data and Methodology
In the previous chapter, methods and results from studies by Meddens et al. 2014, Wulder
et al. 2006, and White et al. 2005 were discussed. The data and methodology used to
detect MPB infestations and assess the accuracy of MLSC using Landsat 8 OLI 30m and
WV02 2m spatial resolution satellite imagery for this project is similar to the
aforementioned studies.
This chapter describes the study region and the different types of data, along with
the methodology used to test the accuracy of MLSC. The study area describes the
background of the region chosen for this analysis. The data section provides detailed
information about each type of imagery used in this analysis. Images from Landsat 8
OLI (30m) and WV02 (2m) comprise the low and high spatial resolution RS imagery.
Vector data provided from ADS, along with field collected waypoints of known MPB
infested trees were used as ground truthed validation data (control points) in the accuracy
assessment process. The data and methodology section describes the classification
methods and different types of indices used to conduct this analysis. The methodology
section of this chapter provides a detailed account of how to conduct this analysis so that
others may replicate these steps in order to conduct studies similar to this thesis project.
Results show the assessment of accuracy of 30m and 2m spatial resolution imagery using
MPB as the subject of matter.
3.1 Study Area
After speaking with Jeffrey Moore (vegetation analyst for USDA Forest Service
Region 5), he stated that there was little to no MPB caused tree mortality or infestations
33
in lodgepole pine stands within the LTBMU. From this information, the study area lies 3
miles (~5000m) just outside the LTBMU in the Northeast section of the LTBMU where
the elevation and vegetation is similar to that within the basin. The LTB itself lies
between the Sierra Nevada mountain range to the west and the Carson Mountain Range
to the east, overlapping the state line that separates California and Nevada (Figure 7).

Figure 7 Location of study region. Red Bounding box represents the location of sections
A, B, and C. Source: Nash 2015
Mountains surrounding the lake (elev. 6225 ft.) rise up to elevations of over
10,000 ft. and consist of mixed forests and vegetation (USGS 2012). Nearly 80%
(150,000+ acres) of the land in the LTB is public and managed by the USDA Forest
3 Mile Extended Study Region
Lake Tahoe Basin Management Unit Boundary
´
0 3.5 7 10.5 14 1.75
Miles
Lake Tahoe
1:400,000
Source Layer Credits: TRPA,
GLOVIS, USGS
34
Service making up the Lake Tahoe Basin Management Unit (LTBMU) (USDA 2015a).
The lodgepole pine tree (primary host to the MPB) exists among the Jeffrey Pine, Red
Fir, White Fir and mixed coniferous trees in the LTB. During a recent phone
conversation in September 2015 with Jeffery Moore mentioned that when MPB
infestations occur, the LTBMU is proactive in their mitigation procedures. The MPB
infestations in the LTB have been controlled using a mitigation technique known as
thinning, reducing the density of lodgepole pine trees, and preventing epidemic
populations from occurring. The red bounding box in Figure 7 is the boundary of three
small sections (A, B and C) of endemic clusters of MPB infestations, located ~3 miles
Northeast of the Northeast portion of the LTBMU boundary, that are included in this
analysis. In these three sections: MPB red attack, dead attack and healthy lodgepole pine
trees existed that were positively identified and recorded as ground truthed validation
points with a handheld Garmin GPS device.
3.2 Data
Data used in these analyses included raster WV02 2m (high spatial resolution)
and Landsat 8 OLI 30m (comparatively lower spatial resolution) satellite imagery, ADS
data (vector format), GPS ground collected data (vector format), study region boundary
(vector format) and vegetation classification data (vector format). Satellite imagery was
used in this project to classify MPB infestations using the MLSC method while the
ancillary data from ADS and ground truthed validation points of MPB infestations were
used to test the accuracy of this classification method. Because 30m spatial resolution
imagery has a lower spatial resolution than 2m, both types of imagery are used to
35
determine how the accuracy of MLSC varies between 30m and 2m spatial resolution
imagery.
3.2.1. Types of data
The level of precision in classification methods required is dependent upon the
population stage of the MPB (endemic, incipient-epidemic, or epidemic) and the
mitigation objectives of forest management. By comparing the accuracy of classification
using two different spatial resolutions (2m and 30m), forest management personnel may
choose the type of imagery with a suitable spatial resolution that assists mitigation
objectives effectively. Mitigation efforts are decided based on the stage of the MPB
infestation and dependent upon the population stage of the MPB. Known areas of MPB
infestations in lodgepole pine stands were used to define the study areas. Since the LTB
is a well-managed area, ADS sketch maps within and outside the LTBMU were
considered for use as ground validation points of recorded lodgepole pine tree stands that
were infested by the MPB. The data used for this analysis is summarized in (Table 6).
Table 6 Datasets used in this project
Data Type Dates Source
Landsat 8 OLI 30m
resolution imagery
Raster 2014, 2015 USGS
WV02 2m
resolution imagery
Raster 2014, 2015 Digital Globe
Foundation
ADS Sketch Maps Vector 2014, 2015 USDA
Ground Truthed
Validation GPS
waypoints
Vector July, August,
September, October
2015
Author

Vegetation
Classification Map
Vector 2009 USDA

36
3.2.2. Landsat 8 OLI 30m spatial resolution imagery
Landsat 8 OLI imagery was selected and downloaded for the months of July 2014
and August 2015 from the USGS Global Visualization Viewer at http://glovis.usgs.gov.
The months of the imagery vary across 2014 and 2015 due to the amount of cloud
coverage existing over the study region at the time of image acquisition. For example,
the cloud coverage for Landsat 8 OLI for August 2014 was 4% (mainly over the study
region) so the month of July was used. These months are ideal for image acquisition
since the effects of MPB infestations (green attack phase turns to red attack phase) occur
during this time of year, one year after the initial infestation. As mentioned previously,
the MPB lifecycle is one year so the effects from an infestation are not apparent to the
human eye until the summer after the initial attack (Wulder et al. 2006).
Shortly after Landsat 5 was decommissioned, Landsat 8 OLI was launched on 11
February 2013 (NASA 2016). Landsat 8 OLI has 11 spectral bands and the band uses are
similar to Landsat 5 TM due to having similar resolutions and spectral character capture.
Bands used in this analysis are summarized in (Table 7). Landsat imagery with a spatial
resolution of 30m allows analysts to accurately detect tree mortality (Meddens et al.
2014). A study conducted in British Columbia by Franklin et al. 2003 concluded that
using Landsat TM ETM+ imagery and supervised classification methods resulted in a
73% accuracy in the detection of red attack stage MPB infested trees (Franklin et al.
2003). Although the use of 30m imagery has been shown to be fairly accurate, using
satellite imagery that has a higher spatial resolution (2m) may increase the accuracy
results of supervised classification.
37
Table 7 Landsat 8 OLI band characteristics and applications. Source: USGS 2014
Band
number/color
Wavelength (um) Resolution
(meters)
Application
Band 1/Coastal
aerosol
.43-.45 30 Coastal and aerosol
studies, dust and
cloud detection in
atmosphere
Band 2/Blue .45-.51 30 Soil from vegetation
delineation
Band 3/Green .53-.59 30 Plant vigor and
health
Band 4/Red .64-.67 30 Vegetation slopes
Band 5/Near
Infrared (NIR)
.85-.88 30 Shorelines
Band 6/Short Wave
Infrared 1(SWIR 1)
1.57-1.65 30 Moisture content of
soil and vegetation.
Used in EWDI
Band 7/SWIR 2 2.11-2.29 30 Soil and vegetation
classification
Band
8/Panchromatic
.50-.68 15 High resolution
band

3.2.3. WorldView02 2m spatial resolution imagery
WV02 imagery was selected and downloaded from the Digital Globe FTP server
after an imagery grant was approved. Imagery captured by WV02 satellite during July
2014 and August 2015 covered mainly the eastern portion of the LTBMU (Figure 8).
WV02 is the first high-resolution 8-band multispectral satellite commercially available
since IKONOS was decommissioned in 2013 (Digital Globe 2015). As previously
mentioned, the months of July and August are ideal months for image acquisition due to
the nature of the MPB infestation and progression of tree mortality that follows the initial
infestation from the prior year.
38

Figure 8 Bounding box represents the coverage of WV02 2m spatial resolution imagery
for July 2014 and August 2015. Source: USGS 2012

WV02 was launched on 8 October 2009 with an expected mission life of 10-12
years. Table 8 shows WV02 band characteristics and applications. WV02 has one
panchromatic band (pan-sharpened capable of producing 1.85m spatial resolution) and
eight multispectral bands (capable of producing imagery with an average of 2.07m spatial
resolution) (DigitalGlobe 2015). A comparatively higher spatial resolution to Landsat 8
OLI 30m, WV02 2m gives the analyst greater detail, allowing the analysis of smaller
patches (~2m
2
) of red attack phase MPB infested trees. With a spatial resolution of 2m,
39
objects that cover 2m
2
can be viewed with WV02 imagery.
Table 8 WV02 Band characteristics and applications. Source: DigitalGlobe 2010

40
3.2.4. Aerial Detection Survey Sketch Maps
Every year, the USDA Forest Service conducts one ADS to sketch map trees that
have been either damaged or killed. The sketch map attributes include the damage type,
number of trees affected, and the tree species that were affected (USDA 2015b).
Normally, the sketch maps are validated using field crews and the data is fairly reliable.
The numbers of trees affected in each polygon, representing a sketch of the infested area
are estimated per acre. Areas containing less than one affected tree per acre are
considered a normal level of mortality (endemic MPB population), and therefore not
recorded on the sketch map (USDA 2015b).
WV02 imagery was used to detect these small patches of trees not recorded in
ADS. WV02 has a spatial resolution of 2m, therefore smaller patches of trees that are
infested by the MPB can be detected using the imagery rather than by aircraft, whereas,
ADS are used primarily for detection of infested stands for planning and allocating
mitigation procedures (British Columbia Ministry of Forests 2003; White et al. 2005).
Since ADS is not used to detect endemic (<1 tree per acre) populations, WV02 is useful.
ADS sketch maps can be used as a source of ground validation to establish training areas,
as well as to assess the accuracy of MLSC methods using RS imagery. ADS sketch
mapping is also useful in establishing an inventory of affected trees and detecting MPB
infestations at a provincial scale (White et al. 2005). ADS from 2005-present are
provided by the USDA Forest Service and are downloadable free of charge.
3.2.5. Ground Truthed Validation Points
Ground truthed validation points were collected in the field during August,
September, and October 2015 with a handheld GPSMAP© 62sc GPS recreational GPS
41
receiver. The positional accuracy of most recreational GPS devices is on average about 9
feet. These waypoints were used to establish training areas to create signature files, later
used in MLSC function in ArcGIS 10.3. The ground truthed validation points were used
to assess the accuracy of MLSC of both 30m and 2m spatial resolution imagery of 2015
Landsat 8 OLI 30m and WV02 2m spatial resolution imagery.
Single-tree coordinates were collected in locations that included healthy, red
attack and dead attack lodgepole pine trees in dense stands as well as single infested
trees. The collected ground truthed validation points served the same purpose as ADS
sketch maps, however, the total amount of infested trees could be recorded more
precisely when single trees were recorded and immediately documented. This procedure
was used to record smaller endemic patches of MPB infested trees that were not recorded
in the Region 5 ADS sketch maps.
According to Jeffrey Moore, infestations that are less than one tree per acre
(endemic) were not recorded in the sketch maps. Human errors occur when attempting to
define smaller patches of tree mortality/infestation during ADS, therefore, ground
validation points were used to establish training areas that were used in the MLSC
method.
3.3 Procedures and Analysis
This section of this paper describes the processes that occur after image
acquisition. These processes include pre-processing (completed by Glovis and
DigitalGlobe), creating natural color composite images, collection of ground truth
validation points, vegetation classification map used to establish training areas, MLSC,
42
accuracy assessment of MLSC per imagery resolution (30m and 2m) and comparison of
results. The calculation of TCT wetness indices for Landsat 8 OLI 30m spatial resolution
imagery for 2014 and 2015 were also included in the processes used to calculate the
EWDI (Figure 9).
3.3.1. Pre-processing
Glovis and Digital Globe completed orthorectification prior to image acquisition.
Orthorectification is the process of correcting the geometry of an image to reduce
distortion caused by terrain, sun angles, and the angle at which the image was taken from
the satellite sensor (ESRI n.d.). These corrections were completed by comparing the
imagery with ground control points (known landmarks) and resampled so the exact
location of a single pixel is located in the correct location on earth (ESRI n.d.). Without
orthorectification and geometric correction, the imagery can misrepresent pixels resulting
in a misalignment of the imagery, thus creating inaccuracies in the analyses. These pre-
processing steps ensured that a single pixel in the image correctly represented the precise
location on earth.
43

Figure 9 Workflow diagram. Source: Nash 2015
3.3.2. Maximum Likelihood Supervised classification
Maximum Likelihood Supervised Classification is the process of classifying
objects in an image based on the guidance of the analyst. After natural color composite
images were created, training areas (based on the vegetation classification map and
ground truthed validation points) were established and used in the MLSC process using
the spatial analyst tool in ArcGIS 10.3. A MLSC algorithm was used to classify cells that
were similar to specified pixels that represented an object based on classification
confidence. Different classes were assigned based on the level of certainty after the
44
MLSC process was executed. For example, if a signature file was created for a red attack
stage tree, a maximum likelihood classification process was executed and other pixels in
the image that match the digital number of chosen pixel (with the most certainty) was
reclassified in the image as red attack stage. Ground truthed validated points of healthy,
red attack and dead attack lodgepole pine trees assisted in creating classes and control
variables used for the MLSC.
3.3.3. Vegetation Classification Map
Insect infestations of other types, existing vegetation other than the lodgepole pine
trees and other phenomena exist that can affect the accuracy of the classification of MPB
red attack stage. These other phenomena, that affect the visual signature of the
landscape, include water bodies, clear cuts, fire scars, other vegetation, and urban
development. These phenomena must be excluded to ensure accuracy of the results.
Other types of trees/vegetation exist amongst lodgepole pine in the LTB, which may
affect the classification MPB in lodgepole pine tree stands. White et al. 2005 used an
exclusionary mask of logged areas, water bodies and cloud cover present at the time of
image acquisition. The purpose of this mask was to reduce the spectral variations that are
characteristic of a mixed forest (White et al. 2005). Minimizing the spectral variations
before executing supervised classification improved the accuracy of the results.
Dependent upon the size and location of infestations located using ground
truthing, a vegetation classification map can be helpful in creating a mask; it would be
similar to White’s exclusionary mask that excludes other types of trees and phenomena
(water bodies, clear cuts, fire scars, other vegetation and trees, and urban development)
45
besides the lodgepole pine tree. The vegetation classification map is available to the
public via the USDA Forest Service and was downloaded free of charge. Existing
lodgepole pine tree stands were queried from this data and created into a vector polygon
layer. The vegetation classification map included: (1) scrub, (2) forests and (3)
developed areas. Lodgepole pine forests were queried from the forest category and then
used to create training areas, utilized in creating signature files that were relevant to
running the MLSC algorithm in ArcGIS 10.3.
3.3.4. Accuracy Assessment Methods Using Ground Truthed Data
Data used in the accuracy assessment of MLSC included the vegetation
classification map (only lodgepole pine trees) and ground truthed validation points of
lodgepole green (LPG) and lodgepole red (LPR) trees. The overall accuracy, errors of
commission and errors of omission were calculated based on these layers. GPS
waypoints were collected in the sections A, B and C and later used in the MLSC of
objects in RS imagery.
ADS mapping (mentioned later) is a type of ground truth data collected over a
large area by aircraft. The purpose of using GPS waypoints as ground truth data was to
precisely locate points/areas in both 30m and 2m spatial resolution imagery, associating a
ground control point with a pixel of the image that occurs relatively close in time to the
acquisition of imagery, ensuring the most accurate data used in the accuracy assessment
of 2015 imagery. Ground truthed validation data acquired during the months of July,
August, September and October (approximately the same time of year as the RS imagery
acquisition) provided a sample amount of points, within pixels, that were assigned to a
46
particular category (LPG or LPR). These points represented objects with locations on
earth and assigned a Digital Number (DN), which was represented one pixel of the
image.
After all ground truthed validation points were collected from the field, they were
uploaded into ArcGIS 10.3 and overlaid upon Landsat 8 OLI and WV02 imagery post
MLSC. In order to assess the accuracy of classification, the ground truthed validation
points were assigned values (DN’s) based on their location and underlying class (5 user-
defined classes). This step was completed using the Spatial Analyst Extract Values to
Points tool in ArcGIS 10.3.
This sample data was then inputted into an error matrix, which is a square array of
numbers (sample units) in rows and columns and assessed for accuracy (Congalton
1991). The error matrix allows analysts to assess the accuracy of classification methods
by comparing the location and class of ground truthed validation points to the location
and class of the imagery. Producer’s accuracy is the total number of correct pixels
classified using supervised classification in RS imagery by the total number of reference
(ground truth) pixels. User’s accuracy is calculated by the total number of correct pixels
classified by the total number of pixels classified in the RS imagery using supervised
classification.
Using this error matrix (Table 9)., the overall accuracy, errors of commission
(misclassified as MPB(R)) and errors of omission (objects left out that are included in
sample) were calculated. The overall accuracy was calculated by dividing the total
number of objects correctly classified by the total number of samples of supervised
47
classification, which equaled 74% (Table 9). The goal of this project was to calculate
and compare the overall accuracy of supervised classification using 30m and 2m spatial
resolution imagery, which was achieved by using an error matrix similar to that of
Congalton 1991.
Table 9 Example of an error matrix. Columns represent ground truth data and rows
represent the results from supervised classification of RS imagery. Source: Congalton
1991

It is possible to use an error matrix (similar to that of Congalton 1991), in this
project with estimates of MPB infested lodgepole pine trees recorded in the attribute data
from the ADS, however, the attribute information included in the ADS data is only an
estimate and does not represent the true number of trees infested and false results would
occur. The results from using ADS attribute data may be subjective, therefore, using
48
ground truthed validation points is a more precise method of recording MPB infested
trees, later used to calculate the overall classification accuracy, errors of commission and
errors of omission of 2015 Landsat 8 OLI 30m and WV02 2m spatial resolution imagery
using an error matrix.

3.3.5. Tasseled Cap Transformation Wetness Index and Enhanced Wetness Difference
Index Calculation

EWDI enables analysts to measure the change in wetness over time using multi-
date imagery. Landsat 8 OLI data was used to measure the change in wetness from 2014
to 2015. Loss of moisture in the canopy of lodgepole pine trees is characteristic of MPB
infestations. In order to calculate the wetness difference (EWDI) between multi-date
imagery (2014 and 2015), the TCT wetness index must be calculated for each year. The
TCT wetness index exhibits the wetness for 2014 and 2015. Once the TCT wetness
index is calculated, the EWDI measures the change in moisture from one year to the
other. When a MPB infestation occurs, the foliage loses moisture and changes from
green attack to red attack, one year after the initial infestation. EWDI has proven to be an
effective tool in detecting areas of red attack MPB infestations (Wulder et al. 2006).
The calculation of the EWDI includes five steps: (1) DN values converted to
reflectance, (2) Bands 2 through 7 SUN_ELEVATION converted to radians, (3) bands 2
through 7 corrected with the SUN_ELEVATION (post conversion to radians), (4) TCT
wetness index calculation using corrected bands 2 through 7 and current Landsat 8 OLI
coefficients, and (5) EWDI calculation.
The first step (1), DN values were converted to reflectance using Equation 1
(USGS 2014) and is as follows:
49
ρλ
'
= M
ρ
Q
cal
+ A
ρ
(1)

where ρλ
'
= the Top of atmosphere reflectance (TOA), M
ρ
= the
RADIANCE_MULT_BAND_x (located in metadata of imagery where x is the band
number), Q
cal
= DN values (located in metadata), and A
ρ
=
RADIANCE_ADD_BAND_x (located in metadata where x is the band number).

Following step 1, step (2) is calculated where each individual band (2 thru 7) was
corrected with the sun elevation after the SUN_ELEVATION (located in the metadata)
was converted to radians from degrees using the following equation (2):
radians = (degrees * π)/180° (2)
where degrees = SUN_ELEVATION (located in the metadata) that is specific to each
image at the time of acquisition. The SUN_ELEVATION changes throughout the year
due to the north-south position of the sun in relation to the earth (NASA 2011). Post
conversion of the SUN_ELEVATION from degrees to radians, each band is corrected to
be used in the TCT wetness calculation. Equation 3 was used to correct each band and is
as follows:
ρλ= ρλ
'
/ sin(θ
SE
) (3)
where ρλ = TOA Corrected reflectance, ρλ
'
= uncorrected TOA reflectance (calculated
using Equation 1), and sin(θ
SE
) = sun elevation in radians.
After bands 2 through 7 were corrected, (step 3), the TCT wetness index was
calculated using the latest Landsat 8 OLI coefficients (Table 10) (Baig et al. 2014). The
50
TCT wetness index was calculated using the same equation that Wulder et al. used in
their 2006 study when they calculated the EWDI using Landsat 7 ETM+ satellite sensor.
Table 10 TCT coefficients for Landsat 8 OLI. Source: Baig et al. 2014 Note: Band
1(coastal and aerosol) and Band 8 (panchromatic) were not used in calculating the TCT
wetness index
Index Band2
(Blue)
Band 3
(Green)
Band 4
(Red)
Band 5
(NIR)
Band 6
(SWIR1)
Band 7
(SWIR2)
Wetness 0.1511 0.1973 0.3283 0.3407 -0.7117 -0.4559

Using the raster calculator in ArcGIS 10.3, Equation 4 was applied to step (4) and
used to calculate the TCT wetness index based on the current Landsat 8 OLI band
coefficients. This equation is applied for bands 2 through 7 for both 2014 and 2015
Landsat 8 OLI satellite imagery and is as follows:
TCW_x =(0.1511*(Band2_corrected))+(0.1973*(Band3_corrected))+(0.3283
*(Band4_corrected))+(0.3407*(Band5_corrected))+(-0.7117*(Band6_corrected))+(-
0.4559*(Band7_corrected)) (4)
where TCW_x = the Tasseled Cap Wetness and x is the year (2014 or 2015).
Finally, after the TCT wetness index was calculated, the EWDI was calculated
using Equation 5 as follows:
EWDI=TCW1-TCW2 (5)
where EWDI = the Enhanced Wetness Difference Index, TCW1 = the Tasseled Cap
Wetness Index for 2014, and TCW2 = the Tasseled Cap Wetness Index for 2015.
Ground truthed validation points of healthy, red attack and dead attack stage MPB
infested lodgepole pine trees assist in detecting the wetness changes from 2014 to 2015.
If the value of a single 30m
2
pixel decreased from 2014 to 2015 where known red attack
and dead attack stage lodgepole pine trees existed, the EWDI calculation will prove to be
51
another means of detecting MPB caused tree mortality by recording the wetness
difference of infested stands and correlating these observations to the stages of the MPB
infestations.

3.4 Overview of Methods
The aforementioned methods include the use of ground validation points used to
test the accuracy of MLSC for Landsat 8 OLI 30m and WV02 2m spatial resolution
satellite imagery. Normally, the computer can classify certain phenomena based on a
DN, however, ground truthed validation points are used to test the accuracy of
classification based on a user’s guidance (supervised classification). Each cluster of red
attack, grey attack and healthy lodgepole pine tree were recorded. Based on these ground
truthed validated points on earth (~9 feet positional accuracy), the accuracy of MLSC
was assessed. In addition to this method, TCT wetness for 2014 and 2015 was calculated
and used in the calculation of the EWDI in order to detect the change in wetness of
locations where the MPB red attack and dead attack trees existed.

52
Chapter 4 Results

Using RS imagery (Landsat 8 OLI and WV02) to find a better option than ADS in order
to cost efficiently detect MPB infestations of lodgepole pine trees in the LTBMU did not
meet my expectations. However, WV02 imagery and USDA Region 5 Forest Service
vegetation classification maps helped in locating red attack trees amongst stands of
healthy lodgepole pine trees, invaluable for this study. These resources (pre-existing and
new) resulted in the documentation of endemic populations of red attack stage MPB
infestations outside the intended study region (LTBMU) that have yet to be recorded or
detected within Region 5 USDA Forest Service ADS.
In previous studies, incipient-epidemic and epidemic populations have been
detected using Landsat 30m spatial resolution imagery (30m
2
or greater, or one pixel of
Landsat 8 OLI imagery). MPB infestations were located although they did not cover
30m
2
. Using WV02 2m spatial resolution imagery, I was able to locate endemic
populations of red attack stage MPB infestations outside the LTBMU in Region 5,
without the use of ADS 2015 sketch maps. Based on the ADS, observations of WV02
imagery and field investigations, only endemic populations of the MPB existed outside
the LTBMU.
The MPB infestations located in this project were much smaller than 30m
2
,
sometimes, only spanning 2m
2
in diameter (endemic stage), and in clusters of only 2 to
20 trees. Ground truthed validation points of healthy, red attack and dead attack stage
lodgepole pine tree MPB infestations were recorded. Considering the positional accuracy
of the collection device (~3m), the coordinate of each tree in this study was within 3m of
its actual location.
53
ADS data provided no assistance in locating endemic populations of the MPB in
the LTBMU based on investigations of the ADS sketch maps, however, WV02 assisted
in locating stands of red attack stage MPB infested trees, later confirmed as endemic red
attack stage MPB infestations just outside the LTBMU in Region 5. Although USDA
Forest Service sketch maps provided an estimation of tree mortality, they were not
beneficial in detecting endemic populations that were specific to the MPB and lodgepole
pine tree, which is imperative to mitigating an incipient-epidemic MPB population before
epidemic populations occur. As a result, combining WV02 imagery and vegetation
classification maps provided data sources that were useful in the detection of MPB
infestations within lodgepole pine tree stands.
MLSC of Landsat 8 OLI 30m and WV02 2m imagery was not as accurate as I
presumed. On the other hand, improved spatial resolution of WV02 satellite imagery
assisted in locating three small endemic clusters (A, B and C) of MPB populations
consisting of dead attack, red attack and healthy lodgepole pine trees. Section A
contained healthy, red attack and dead attack MPB infested lodgepole pine trees while
sections B and C contained only red attack MPB infested lodgepole pine trees.
This chapter is broken into five sections. In Section 4.1, the reliability of ADS
data is discussed. Section 4.2, ground truthed validation points and the positional
accuracy are explained. Section 4.3 reports the accuracy assessment results of MLSC of
Landsat 8 OLI 30m spatial resolution imagery, while Section 4.4 reports the accuracy
assessment results of MLSC of WV02 2m spatial resolution imagery. Finally, Section
4.5 exhibits the results from using TCT and EWDI in the detection of red and dead attack
stages of MPB infestations between 2014 and 2015 using Landsat 8 OLI imagery.
54
4.1 Aerial Detection Survey Data

After querying out MPB-caused (USDA Forest Service Region 5 ADS code
11006) mortality in lodgepole pine trees (USDA Forest Service Region 5 ADS code 108)
from the ADS sketch map attribute table, only a couple smaller areas on the western edge
of the LTBMU were identified. Based on local knowledge from hiking in this particular
area, I concluded that these two smaller areas would be very difficult to access in steep
terrain. Before taking this risk, extra time and not knowing the situation from the air, I
conducted a limited investigation as to the reliability of the ADS sketch maps and their
attributes. I chose an aerial sketch location south of the extended study region (Figure
10). Coordinates of this location were derived from a KMZ file created in ArcGIS 10.3
and imported into Google Earth Pro. The hike was nearly 12 miles round-trip south of
Tragedy Springs, located just off Highway 88, southwest of Silver Lake California
(Figure 10).
55

Figure 10 Area flown by USDA Forest Service Region 5 ADS during August 2015 Note:
inset map shows a larger scale of a southern section of Region 5 ADS sketch map section
ONLY used as an example to test reliability of ADS data. Source: Nash 2015

An elevated side view photo of the area used to test the reliability of the ADS data
and attributes associated with this polygon can be seen in (Figure 11), showing there was
Aerial Detection Survey 2015
Area Flown for ADS 2015
3 mile Extended Study Region
´
0 0.5 1 1.5 2 0.25
Miles
Source Layer Data: USDA Forest Service,
USGS, TRPA
Lake Tahoe
1:550,000
56
a lack of red attack stage MPB infestation. Using knowledge of the symptoms associated
with MPB infested lodgepole pine trees, ground truthed validation concluded there was
no MPB infestation in lodgepole pine trees in this particular ADS sketch map site. Tree
mortality occurred sporadically (about 1 tree per acre) in different types of trees other
than the lodgepole pine tree (Figure 12).

Figure 11 Above left, Side view photograph of ADS site; bottom left, attribute
information of the ADS sketch site; right, ADS sketch map. Source: USDA 2015
57

Figure 12 Above left, healthy lodgepole pine tree amongst Whitebark pine infested trees;
above right, beetle damage causing sporadic tree mortality of Whitebark pine trees
located in ADS Figure 11. Source: Nash 2015

In September 2015, Jeffrey Moore stated in a recent phone conversation that ADS
sketch maps illustrate the extent of tree mortality and infestations where at least one tree
per acre that was affected (as illustrated in the attribute table in Figure 11). He also
explained that clusters are not necessarily recorded and the attribute data is not
completely accurate because the observations are recorded from an airplane and human
errors occur. According to Jeffrey Moore, based on knowledge of infestations and what
lodgepole tree stands look like from the air, this specific aerial sketch site was labeled as
MPB caused mortality in a lodgepole pine tree stand.
From this information and visiting an ADS sketch site to validate the attributes, I
abandoned treks to the smaller ADS sketch sites on the west side of the LTBMU (basing
this decision on reliability of ADS data and safety). ADS sketch sites were not used as
ground truth data for 2015 Landsat 8 OLI 30m and WV02 2m spatial resolution imagery.
According to Jeffrey Moore, when epidemic infestations occur within the LTBMU, field
58
crews visit these sites and thin susceptible trees to maintain endemic populations. Based
on this information, it was best to extend the LTBMU study area boundary using a 3-mile
(5000m) buffer. Here, I was successful in locating endemic clusters of the MPB in
lodgepole pine tree stands near the intended study region, the LTBMU.
4.2 Ground Truthed Validation Points
Nearly 3 miles outside the LTBMU in the northeastern portion of the extended
study region, three separate areas of MPB infested lodgepole pine trees were located.
These ground truthed validation points were used for Landsat 8 OLI 30m (Figure 13) and
WV02 2m (Figure 14) spatial resolution satellite imagery.
The ground validation points recorded in these three stands (A, B, and C)
represent healthy lodgepole, red attack stage, and dead attack stage MPB infested trees.
The image was divided into three different sections: (1) A, (2) B, and (3) C. Section A
consisted of 69 waypoints representing locations of healthy lodgepole and red attack
stage MPB infested lodgepole. Section B consisted of 6 red attack stage MPB infested
lodgepole and section C consisted of 7 red attack stage MPB infested lodgepole pine
trees. Together, (A, B and C) 82 total ground truthed validation points were collected
and used in the accuracy assessment of MLSC for Landsat 8 OLI and WV02 imagery.

59

Figure 13 Landsat 8 OLI 30m spatial resolution satellite imagery ground validation point
map. Source: Nash 2015

´
0 0.1 0.2 0.3 0.4
Miles
Lodgepole Pine Green
Lodgepole Pine Red
Lodgepole Pine Dead
5000m Extended Study Region
Source Layer Data: USGS,
TRPA, Norman Nash
*All Waypoints have a 3m buffer applied
for accuracy adjustment*
1:10,000
A
B
C
60

Figure 14 WV02 2m spatial resolution satellite imagery ground validation point map.
Source: Nash 2015
Aforementioned, a Garmin GPSMAP 62sc© recreational GPS device was used to
collect all points with an accuracy of about 9 feet. With an accuracy of about 9 feet, this
meant that the true point could be within 9 feet in any direction from the center of where
´
0 0.1 0.2 0.3 0.4
Miles
Lodgepole Green
Lodgepole Red
Lodgepole Dead
5000m Extended Study Region
Source Layer Data: Digital Globe,
TRPA, Norman Nash
*All Waypoints have a 3m buffer applied
for accuracy adjustment*
1:10,000
A
B
C
61
it was collected. This positional accuracy discrepancy affects the results when
conducting an accuracy assessment using 30m and 2m spatial resolution RS imagery. A
3m buffer was applied to compensate for the lack of positional accuracy of the Garmin
GPSMap 62sc and used in the accuracy assessment of MLSC.
4.3 Maximum Likelihood Supervised Classification Landsat 8 OLI 30m
Spatial Resolution Satellite Imagery
MLSC allows the user to define classes based on ground validation points and
local knowledge of the area to determine which pixels and associated DN’s represent
certain phenomena on the earth. Using this method, 5 classes were defined based on the
DN of each class prior to classifying the image (Figure 15).

Figure 15 Landsat 8 OLI 30m spatial resolution satellite imagery Maximum Likelihood
Supervised Classification with 5 user-defined classes. Source: Nash 2015
´
0 0.1 0.2 0.3 0.4
Miles
Lodgepole Green
Lodgepole Red
Lodgepole Dead
5000m Extended Study Region
Lodgepole Green
Lodgepole Red
Granite
Meadow
Water
Source Layer Data: USGS,
TRPA, Norman Nash
*All Waypoints have a 3m buffer applied
for accuracy adjustment*
1:9,000
A
B
C
62

The five user-defined classes were based on smaller training areas that contained
a higher density of red attack stage, healthy lodgepole pine trees, and known
vegetation/soil conditions of the area. Using local knowledge, classes were defined based
on the prevalence of known soil and vegetation type in the area and also cross-referenced
with Google Earth Pro imagery. The three sections (A, B and C) can be viewed at a
larger scale in (Figure 16).

Figure 16 Above left, section A; upper middle, section B; upper right, section C Landsat
8 OLI 30m spatial resolution satellite imagery Maximum Likelihood Supervised
Classification Note: Refer to the 5 user defined classes in the legend of Figure 19
Sections B and C have a light grey fill applied for visual enhancement. Source: Nash
2015

The values of the 5 user defined classes were extracted using the “Extract Values
to Points” method mentioned earlier in this chapter. The classes for Landsat 8 OLI are;
Lodgepole Green (Value 1), Lodgepole Red (Value 6), Granite (Value 10), Meadow
(Value 12), and Water (Value 14). The ground validation points’ values were extracted
and assigned to each user-defined class using the “Extract Values to Points” tool located
in the “Extraction” tools in ArcToolbox (Table 11). This table shows the error matrix
used for accuracy assessment of the MLSC of Landsat 8 OLI 30m spatial resolution
63
imagery using ground truthed validation points in sections A, B, and C where all three
sections were combined into one error matrix.
Table 11 Error Matrix for 2015 Landsat 8 OLI 30m spatial resolution imagery of
sections A, B, and C. Columns, represent ground truthed validations points; rows,
Maximum Likelihood Supervised Classification based on 5 user-defined classes. Source:
Nash 2015
Predicted
Sections A,
B, and C
LPG
(Value 1)
LPR
(Value 6)
Granite
(Value 10)
Meadow
(Value 12)
Water
(Value 14)
Row
Total
LPG 12 2 0 0 0 14
LPR 20 30 0 0 0 50
Granite 3 1 0 0 0 4
Meadow 7 7 0 0 0 14
Water 0 0 0 0 0 0
Column
Total
42 40 0 0 0 82

The error matrix in Table 11 was used to calculate the overall accuracy, errors of
commission and errors of omission. The location and class of each ground truthed
validation point and the location and class of each Landsat 8 OLI 30m pixel were
compared, allowing the accuracy assessment of MLSC using ArcGIS 10.3.
The overall accuracy (ratio of the total number of ground truthed validation points
correctly classified to the total number of ground truthed validation points) of MLSC of
Landsat 8 OLI 30m spatial resolution imagery, using LPG and LPR as ground validation
control points was (12+30/82) or 51.22%. LPG errors of commission (ratio of the
number of pixels incorrectly classified by the total number of pixels classified in the LPG
class) was (2/14) or 14.3% of pixels were classified as LPG that do not belong in the LPG
class. LPG errors of omission (ratio of the number of incorrectly classified LPG ground
truthed validation by the total number of total LPG ground truthed validation points) was
(30/42) or 71.4% of LPG ground truthed validation points that belong in the LPG class
64
were not classified as LPG. Based on the descriptions of errors of commission and
omission, the LPR errors of commission was (20/50) or 40% of pixels were classified as
LPR that do not belong in the LPR class. LPR errors of omission was (10/40) or 25% of
LPR ground truthed validation points that belong in the LPR class were not classified as
LPR.

4.4 Maximum Likelihood Supervised Classification WV02 2m Spatial
Resolution Satellite Imagery

With higher resolution imagery, endemic clusters of lodgepole pine red attack
stage MPB infestations were visible, allowing analysts to create more precise training
areas. The use of five user-defined classes based on training sites resulted in a more
detailed and precise classification of WV02 imagery due to the comparatively higher
spatial resolution than Landsat 8 OLI. A MLSC map of WV02 2m spatial resolution
satellite imagery was created with five user-defined classes. These classes were based on
training sites created from ground truthed validation points and local knowledge of the
soil type and vegetation type within the imagery (Figure 17).
65

Figure 17 Maximum Likelihood Supervised Classification WV02 2m spatial resolution
satellite imagery with 5 user-defined classes. Source: Nash 2015
A larger scale of sections A, B, and C can be viewed in (Figure 18). Following
this figure, (Table 12) exhibits an error matrix, including ground truthed validation points
from sections A, B, and C, that were used to conduct and accuracy assessment of the
MLSC method for 2015 WV02 2m spatial resolution satellite imagery.

´
0 0.1 0.2 0.3 0.4
Miles
Lodgepole Green
Lodgepole Red
Lodgepole Dead
5000m Extended Study Region
Meadow
Dense Lodgepole
Bare Soil
Granite/Sand
Lodgepole Red
Source Layer Data: Digital Globe,
TRPA, Norman Nash
*All Waypoints have a 3m buffer applied
for accuracy adjustment*
1:10,000
A
B
C
66

Figure 18 Above left, section A; above middle, section B; above right, section C
Maximum Likelihood Supervised Classification Note: Refer to the 5 user defined classes
in the legend of Figure 24. Source: Nash 2015
The values of the 5 user-defined classes of WV02 were extracted using the same
method applied to Landsat 8 OLI imagery. The classes for WV02 were; Dense
Lodgepole or LPG (Value 3), Lodgepole Red (Value 12), Bare Soil (Value 5),
Granite/Sand (Value 8), and Meadow (Value 1). The ground validation points’ values
were extracted and assigned to each user-defined class using the same method as Landsat
8 OLI. LPR and LPG ground truthed validation points, along with the 5 user-defined
classes, were used in an error matrix (Table 12) to assess the accuracy.
Table 12 Error Matrix for 2015 WV02 2m spatial resolution satellite imagery of sections
A, B, and C. Columns, ground truthed validations points; rows, Maximum Likelihood
Supervised Classification based on 5 user-defined classes. Source: Nash 2015
Predicted
Sections A,
B, and C
LPG
(Value 3)
LPR
(Value
12)
Bare Soil
(Value 5)
Granite/Sand
(Value 8)
Meadow
(Value 1)
Row
Total
LPG 13 18 0 0 0 31
LPR 16 9 0 0 0 25
Bare Soil 4 5 0 0 0 9
Granite/Sand 2 2 0 0 0 4
Meadow 7 6 0 0 0 13
Column
Total
42 40 0 0 0 82

67
The error matrix in Table 12 was used to calculate the overall accuracy, errors of
commission and errors of omission. The location and class of each ground truthed
validation point and the location and class of each WV02 2m pixel were compared,
allowing the accuracy of MLSC to be assessed using ArcGIS 10.3.
The overall accuracy (ratio of the total number of ground truthed validation points
correctly classified to the total number of ground truthed validation points) of MLSC of
WV02 2m spatial resolution imagery, using LPG and LPR as ground validation control
points was (13+9/82) or 26.82%. LPG errors of commission (ratio of the number of
pixels incorrectly classified by the total number of pixels classified in the LPG class) was
(18/31) or 58.1% of pixels were classified as LPG did not belong in the LPG class. LPG
errors of omission (ratio of the number of incorrectly classified LPG ground truthed
validation by the total number of total LPG ground truthed validation points) was (29/42)
or 69% of LPG ground truthed validation points that belonged in the LPG class were not
classified as LPG. Based on the descriptions of errors of commission and omission, the
LPR errors of commission was (16/25) or 64% of pixels were classified as LPR did not
belong in the LPR class. LPR errors of omission was (31/40) or 77.5% of LPR ground
truthed validation points that belonged in the LPR class were not classified as LPR.

4.5 Enhanced Wetness Difference Index for Landsat 8 OLI 30m Spatial
Resolution Satellite Imagery using imagery from 2014 and 2015

In order to calculate the EWDI for 2014 and 2015, the TCT calculations were
completed (Section 3.3.5). The EWDI is an index that shows the wetness difference
between 2014 and 2015 Landsat 8 OLI imagery (Figure 19).
68

Figure 19 EWDI calculated using the TCT wetness 2014 and TCT wetness 2015 of
Landsat 8 OLI 30m spatial resolution imagery. Source: Nash 2015

EWDI was calculated using the TCT wetness values for 2014 and 2015 Landsat 8
OLI 30m spatial resolution imagery. The following figures (Figure 20-Figure 23) show
the TCT wetness and the EWDI between 2014 and 2015 in sections A, B and C of the
study area for Landsat 8 OLI 30m spatial resolution satellite imagery. Ground truthed
validation points collected for both lodgepole red attack and lodgepole dead attack trees
were included and separated by section according to the TCT wetness of each year (2014
and 2015), and the resulting EWDI.
! ! !
!
!
!
!
! ! !
! !
!
!
! !
! ! !
! ! !
! ! !
! ! ! !
!!! ! ! !! !! !!
!
!
!
!
! !
!
!
!
!
! !
! ! ! ! ! ! !
! !
!
! ! !! !! ! !
! ! ! ! ! ! ! ! !
! !
! ! ! !
!
!!
! ! ! !
EWDI
High : 7.56247
Low : -7.04392
!
Lodgepole Green
!
Lodgepole Red
!
Lodgepole Dead 0 0.2
Miles
´
A
B
C
69

Figure 20 Landsat 8 OLI 30m TCT wetness index and EWDI in section A LPR ground
truthed validation points. Above left, 2014 TCT wetness; above middle, 2015 TCT
wetness; above right, EWDI. Source: Nash 2015

Figure 21 Landsat 8 OLI 30m TCT wetness index and EWDI in section A LPD ground
truthed validation points. Above left, 2014 TCT wetness; above middle, 2015 TCT
wetness; above right, EWDI. Source: Nash 2015

Figure 22 Landsat 8 OLI 30m TCT wetness index and EWDI in section B LPR ground
truthed validation points. Above left, 2014 TCT wetness; above middle, 2015 TCT
wetness; above right, EWDI. Source: Nash 2015
70

Figure 23 Landsat 8 OLI 30m TCT wetness index and EWDI in section C LPR ground
truthed validation points. Above left, 2014 TCT wetness; above middle, 2015 TCT
wetness; above right, EWDI. Source: Nash 2015

The values of each LPR and LPD attack ground truthed validation point were
clustered based on location within each pixel. Each cluster of ground truthed validation
points that lies in a single pixel (30m
2
) were associated to a number (1-7), which displays
the TCT wetness and EWDI value per pixel for 2014 and 2015 (Table 13). The EWDI
scale is between (-7.04392 to 7.56247) shown previously in the legend of Figure 19.
Table 13 2014 and 2015 TCT wetness and EWDI pixel values of ground truthed
validation points per 30m
2
pixel. (é), represents an increase in wetness from 2014 to
2015; (ê), indicates a wetness loss from 2014 to 2015. Source: Nash 2015
Section A
Lodgepole Red
2014 TCT Wetness
Pixel Value
2015 TCT Wetness
Pixel Value
EWDI Value
(1) -0.008176 0.007907 -0.016083 é
(2) -0.021351 -0.040020 0.018669 ê
(3) -0.048578 -0.037942 -0.010636 é
(4) -0.026604 -0.017879 -0.008725 é
(5) -0.015521 -0.016065 0.000544 ê
(6) -0.076063 -0.043578 -0.032485 é

Section A
Lodgepole Dead
2014 TCT Wetness
Pixel Value
2015 TCT Wetness
Pixel Value
EWDI
(1) 0.008532 0.005571 0.002960 ê
(2) -0.022056 -0.017385 -0.004671 é
(3) -0.021351 -0.040020 0.018669 ê
(4) -0.026604 -0.017879 -0.008725 é
(5) -0.015521 -0.016065 0.000544 ê
71
(6) -0.015570 -0.011545 -0.004025 é
(7) -0.013238 -0.004617 -0.008622 é

Section B
Lodgepole Red
2014 TCT Wetness
Pixel Value
2015 TCT Wetness
Pixel Value
EWDI
(1) -0.050812 -0.042007 -0.008805 é
(2) -0.030524 -0.033019 0.002495 ê

Section C
Lodgepole Red
2014 TCT Wetness
Pixel Value
2015 TCT Wetness
Pixel Value
EWDI
(1) -0.003867 -0.002140 -0.001727 é
(2) -0.159249 -0.022355 -0.136894 é
(3) -0.237038 -0.027283 -0.209755 é

According to the pixel values derived from the 2014 and 2015 TCT wetness
index, a positive EWDI number represents a loss of moisture (ê) and a negative number
represents a moisture increase (é). Using the ground truth validation points as reference,
12 out of 18 pixels in this study region exhibited a wetness increase. MPB infested trees
decrease in moisture as pigmentation was lost with the progression of infestations. In
Table 13 EWDI was calculated using Landsat 8 OLI 30m spatial resolution imagery and
with MPB infestations at the endemic stage, 30m
2
coverage was not likely, yet the change
in moisture from 2014 to 2015 per pixel was still affected due to MPB infestations
EWDI, an index, averages the reflectance value of the phenomena within the
pixel. The DN represents the average of an entire pixel, like an index. If the majority of
a pixel is green, the red will not be strong enough to create an average. Red attack and
dead attack trees decrease in moisture as the progression of MPB infestations occur.
Therefore, red attack trees will be red in a natural color image an dead attack will be
grey. An explanation for the results of EWDI for 2014 and 2015 is that the density and
population stage of the red attack and dead attack MPB infested trees were not high
enough to affect the overall reflectance (DN), resulting in a decrease in the wetness of the
72
entire pixel. The results from EWDI showed that 3 out of 11 or 27.3% of Landsat 8 OLI
30m pixels containing LPR MPB infested trees decreased in wetness from 2014 to 2015.
3 out of 7 or 42.8% of the pixels included in the analysis containing LPD MPB infested
trees decreased in wetness from 2014 to 2015. An average lodgepole pine tree spans
nearly 2m in diameter. Considering the area covered per pixel, the diameter of lodgepole
pine trees and lack of number of trees prevented a viable accuracy assessment using
EWDI could not be properly performed in this analysis (not a bad thing). According to
the results of the EWDI, there is not enough ground truthed validation points (LPR and
LPD) to create a reliable relationship between wetness decrease/increase associated with
MPB infestations due to a lack of sufficient sample points considering the 30m spatial
resolution.
Originally, I was expecting a lot higher accuracy assessment results of MLSC of
MPB stages using Landsat 8 OLI and WV02 imagery. Using ArcGIS 10.3’s MLSC
method, accuracy assessment results using the different stages of the MPB infestations as
control points was 50% accurate using imagery that cannot be used to visibly locate
infestations, therefore cannot be used to detect endemic stages of the MPB. In order to
prevent epidemic population levels of the MPB, endemic clusters must be detected first.
WV02 can be used aside ADS to detect endemic infestations, yet comes with a price.
The accuracy assessment results of Landsat 8 OLI 30m, WV02 2m and EWDI of Landsat
8 OLI showed that endemic stages of MPB can accurately be detected less than 50% of
the time using MLSC techniques with ArcGIS 10.3.

73
Chapter 5 Discussion and Recommendations

The main focus of this project was to test the accuracy of MLSC of MPB red attack stage
infestations in the LTBMU using both Landsat 8 OLI 30m and WV02 2m spatial
resolution imagery; however, there was a scarcity of both endemic and epidemic MPB
populations in the LTBMU. This resulted in a smaller study region within close
proximity of the LTBMU where endemic clusters of MPB infestations occurred. This
area was thoroughly hiked and the stages of MPB infestations documented. With an
absence of MPB infestations located in the ADS sketch survey maps, WV02
(comparatively higher spatial resolution than Landsat 8 OLI) assisted in locating endemic
clusters of MPB infestations that were used in this thesis project.

5.1 Understanding Remote Sensing and the Mountain Pine Beetle

Throughout this thesis, like White et al. 2005, I have found that high resolution
RS imagery is available commercially and can assist in the detection of MPB infestations
without the use of ADS. Although this thesis did not prove the classification methods to
be accurate according to the error matrix, the consistency of WV02 imagery supersedes
ADS due to the availability of imagery throughout the year. Forest management can use
WV02 high-resolution imagery to detect tree mortality rather than relying on ADS.
Rather than trying to control the epidemic population growth, management and
mitigation efforts have shifted to focus on the source of the problem, trying to control
incipient-epidemic populations of the MPB. One of the objectives of Region 5 Forest
Service is to maintain the aesthetics of forest that are frequented by tourists. Being able
to address an endemic MPB population fulfills this objective through mitigation efforts
74
like thinning (mentioned in Chapter 1). This shift in management resulted in detecting
smaller incipient-epidemic populations using high-resolution RS imagery, followed by
mitigation efforts to prevent these incipient-epidemic populations from progressing into
epidemic populations. Forest management may utilize local scale analysis to detect small
infestations so that field crews can remove the problem before it persists.

5.2 Accomplishments of Thesis Project
ADS are conducted annually. According to Jeffrey Moore (who is in charge of
the ADS), the budget for this service is $100,000 annually (does not include the salary of
employees who process the aerial photography). Landsat 8 OLI 30m satellite imagery is
available for download at no cost whereas, WV02 2m satellite imagery costs
$126.50/25km
2
. This imagery would need to be purchased annually to map MPB and
other beetle-caused infestations. The LTBMU covers nearly 832 km
2
(not including the
area of Lake Tahoe itself). Therefore, the cost of WV02 2m spatial resolution satellite
imagery would amount to just over $4,200. Compared to ADS at $100,000/yr., it is
evident that WV02 2m-satellite imagery serves a purpose for the price. Throughout this
thesis project, WV02 has assisted in locating endemic population clusters of MPB that
were not recorded in ADS of 2015. The ground truthed validation points collected, that
did not exist in the ADS database, may easily be added to the 2015 ADS database for
future research and records.
WV02 accuracy assessment results using MLSC, although inferior to Landsat 8
OLI imagery, has its benefits. Sketch maps defining the areas of tree mortality, whether
endemic or epidemic, can be drawn from using a combination of vegetation classification
maps and WV02 imagery rather than having to gather this information from a plane. It
75
would be beneficial to forest management to purchase WV02 imagery, covering their
intended study region, to be referenced with ADS data collected per year.

5.3 Limitations of Thesis Project
Problems occur with accuracy assessment when trying to classify a 30m
2
pixel when only
a few objects (red attack stage lodgepole pine tree) existed within a single pixel,
especially if the infested trees were amongst healthy lodgepole pine trees in that pixel.
The DN was calculated as an average of all objects within that 30m
2
. Object Based
Image Analysis (OBIA) is another method used to detect single objects and segment
these objects into relatively homogenous groups of pixels (Blaschke 2009). This method
allows the analyst to segment WV02 (or imagery with a similar spatial resolution)
imagery into “like” pixels that have relatively similar DN’s. While most clusters of red
attack stage MPB infested lodgepole pine trees were found in clusters of 2-20 trees and
were scattered amongst a primarily healthy lodgepole pine tree stand, this method would
improve the classification of MPB infestation stages (red, yellow or grey). The LTBMU
is well managed so the population of the MPB population is kept at an endemic level.
MLSC techniques have proven to be more useful in detecting epidemic MPB populations
that cover large areas (1 acre or more), using 30m spatial resolution imagery. MLSC can
be used with both 30m and 2m spatial resolution imagery depending on the MPB
population (endemic, incipient-epidemic, or epidemic), while OBIA is reserved more for
2m or less (high) spatial resolution imagery.

76
5.4 Recommendations for Future Work
A few recommendation for completing a project similar to this include: (1) Using
WV02 2m spatial resolution imagery rather than ADS to locate tree mortality, (2) Using a
GPS data collection device with the highest positional accuracy, (3) Exploring the newest
RS software programs to process and analyze the imagery, (4) Defining more classes
would increase the spectral variability and therefore segment the image into more pixels
that represent other phenomena amongst MPB infested Lodgepole pine trees, and (5)
Considering the forest management objectives to decide which population stage of MPB
infestations need to be studied.
As mentioned earlier in the results chapter, a lot of time was spent assessing the
reliability of ADS sketch maps. WV02 2m spatial resolution and vegetation classification
maps proved to be reliable sources to locate tree mortality in lodgepole pine tree stands.
Errors occur while collecting data, and are more likely to happen in an airplane. This
study was not intended to discount ADS sketch surveys, rather to find a more efficient
and reliable way to survey tree mortality (specifically MPB caused tree mortality).
A GPS device that has sub-meter accuracy, rather than a recreational GPS device
(roughly 3m positional accuracy at best), would improve the accuracy of this thesis
project. I found that using a Garmin GPSMap 62sc created issues with positional
accuracy when creating training sites for the WV02 high-resolution imagery
classification. A 3m buffer applied to ground truth points included 1.5 pixels, 360
degrees around the center of the waypoint. With this, a healthy lodgepole pine tree
adjacent to a red attack stage lodgepole pine tree was included in the 3m buffers, which
77
was applied for positional accuracy and slightly degraded the accuracy assessment
process.
ArcGIS 10.3 has been commonly used to perform MLSC, however, there are
other programs that are easier to use and more efficient. A few known RS software
programs such as ERDAS (Earth Resources Data Analysis System) Imagine, eCognition,
ENVI (Environment for Visualizing Images), and TerrSet (formally IDRISI), can be used
for projects similar to this thesis. As of August 3, 2015, the Office of Surface Mining
Reclamation and Enforcement mentioned on its website (http://www.tips.osmre.gov),
ERDAS Imagine is used for classification, modeling, image analysis, multispectral
classification and map production. Software features and benefits of Trimble’s
ECognition 2016 show the possibilities of OBIA and extracting groups of pixels for
classification, commonly used for change detection and tree classification in forestry
(http://www.ecognition.com). ENVI is also used for classification of images and
compatible with ArcGIS (used for map production). Finally, according to Clark Labs in
2015, TerrSet is commonly used for surface analysis, change detection in vegetation and
time series analysis (https://clarklabs.org/terrset/). These aforementioned software
programs can be used for the detection and monitoring of MPB infestations, some more
applicable than others depending on the population stage of the MPB infestation.
This work defined five classes that represented the five most prevalent types of
soil and vegetation in the generalized study area sections A, B, and C. Gathering ground
truthed validation points of different types of vegetation and soil types would allow the
an increasingly amount of classes, creating a more segmented image (similar to OBIA
mentioned earlier) so the red attack stage MPB infestations may be more visible.
78
Forest management objectives can be located through the USDA website and
should be thoroughly read prior to study design. In forests experiencing epidemic MPB
infestations, management objectives are likely focused on slowing the progression of
MPB infestations by locating incipient-epidemic populations surrounding the infested
areas and applying mitigation efforts. Locating smaller endemic and incipient epidemic
MPB infestations requires the use of high-resolution imagery (2m or less) to detect
smaller patches of trees that are newly infested. On the other hand, forest management
mitigation efforts to control an epidemic MPB population can use Landsat 30m spatial
resolution imagery to detect larger MPB infestations spanning more than 30m
2
.
Above are the main recommendations that I would suggest when approaching a
project similar to this. Forest management objectives and mitigation efforts are
dependent upon the population stage of the MPB infestation (endemic, incipient-
epidemic, or epidemic), which determines the spatial resolution of the RS imagery
needed for detection and monitoring of MPB infestations.

79
REFERENCES

Amman, Gene D., and Richard F. Schmitz. 1988. “Mountain Pine Beetle: Lodgepole Pine
Interactions and Strategies for Reducing Tree Losses”. Ambio 17, no. 1: 62–68.

Amman, Gene D., Mark D. McGregor, and Robert E. Dolph. 1990. Mountain Pine
Beetle. Forest Insect & Disease Leaflet 2. Portland: USDA Forest Service, Pacific
Northwest Region, Natural Resources.

Baig, Muhammad H.A., Lifu Zhang, Tong Shuai, and Qingxi Tong. 2014. Derivation of a
tasseled cap transformation based on Landsat 8 at-satellite reflectance.” Remote
Sensing Letters 5, no. 5: 423-431.

Blaschke, Thomas. 2010. “Object based image analysis for remote sensing.” ISPRS
Journal of Photogrammetry and Remote Sensing 65, no. 1: 2-16.

British Columbia Ministry of Forests. 2003. Provincial Bark Beetle Management
Technical Implementation Guidelines. Victoria, BC: Forest Practices Branch.

Carroll, Allan L. 2007. "The mountain pine beetle Dendroctonus ponderosae in Western
North America: Potential for area-wide integrated management." In Area-wide
control of insect pests, 297-307. Netherlands: Springer.
Cole, Walter E., and Gene D. Amman. 1980. "Mountain pine beetle dynamics in
lodgepole pine forests, Part 1: course of an infestation." In The Bark Beetles,
Fuels, and Fire Bibliography, 1-52. Ogden, Utah: USDA Forest Service,
Intermountain Forest and Range Experiment Station.
Congalton, Russell G. 1991. "A review of assessing the accuracy of classifications of
remotely sensed data." Remote sensing of environment 37, no. 1: 35-46.
Coops, Nicholas C., Matt Johnson, Michael A. Wulder, and Joanne C. White. 2006.
"Assessment of QuickBird high spatial resolution imagery to detect red attack
damage due to mountain pine beetle infestation." Remote Sensing of Environment
103, no. 1: 67-80.
DigitalGlobe. 2010. “The Benefits of Eight Spectral Bands of WorldView-2.” Accessed
October 5, 2015. http://global.digitalglobe.com/sites/default/files/DG-
8SPECTRAL-WP_0.pdf
DigitalGlobe. 2015. “WorldView-2.” Accessed October 1, 2015.
https://www.digitalglobe.com/about-us/content-collection

80
ESRI. 2014. “How Maximum Likelihood Classification works.” Last modified April 10,
2014. Accessed July 23, 2015.
http://resources.arcgis.com/en/help/main/10.2/index.html#/How_Maximum_Likel
ihood_Classification_works/009z000000q9000000/
ESRI. ND. “Landsat Imagery Enhancements.” Accessed August 7, 2015.
http://www.esri.com/software/landsat-imagery/enhancements
Franklin, S. E., M. A. Wulder, R. S. Skakun, and Allan L. Carroll. 2003. "Mountain pine
beetle red-attack forest damage classification using stratified Landsat TM data in
British Columbia, Canada." Photogrammetric Engineering & Remote Sensing 69,
no. 3: 283-288.
Gillis, Trinka. 2012. “Use of Remotely Sensed Imagery to Map Sudden Oak Death
(Phytophthora Ramorum) in the Santa Cruz Mountains.” Master’s thesis,
University of Southern California.
Hill, John, B. Grove, R. Alvin, and Henry W. Popp. 1967. Botany: A textbook for
colleges. Toronto: McGraw-Hill Book Co.
Krause, Alexander. 2006. "Identifying Pine Beetle Infestations in the TNRD." The
Journal, August 15. Accessed June 2015.
http://search.proquest.com.libproxy1.usc.edu/docview/374737239?accountid=147
49.

Lundquist, John E., and Robin M. Reich. 2014. "Landscape dynamics of mountain pine
beetles." Forest Science 60, no. 3: 464-475.

Meddens, Arjan JH., and Jeffrey A. Hicke. 2014. "Spatial and temporal patterns of
Landsat-based detection of tree mortality caused by a mountain pine beetle
outbreak in Colorado, USA." Forest Ecology and Management 322: 78-88.

NASA. 1998. “Fact Sheets.” Last modified April 22, 2008. Accessed July 10, 2015.
http://www.nasa.gov/centers/langley/news/factsheets/RemoteSensing.html

NASA. 2011. “Landsat 7: Science Data Users Handbook.” Last modified March 11,
2011. Accessed December 15, 2015.
http://landsathandbook.gsfc.nasa.gov/data_properties/prog_sect6_3.html

NASA. 2016. “Landsat Science: Landsat 8.” Last modified January 21, 2016. Accessed
December 15, 2015. http://landsat.gsfc.nasa.gov/?p=3186

National Park Service. 2015. “Forest Health: Mountain Pine Beetle.” Last modified
January 3, 2016. Accessed June 26, 2015.
http://www.nps.gov/romo/learn/nature/mtn_pine_beetle_background.htm

Roe, Arthur L., and Gene D. Amman. 1970. "The Mountain pine beetle in lodgepole pine
81
forests." In The Bark Beetles, Fuels, and Fire Bibliography, 1-22. Ogden, Utah:
USDA Forest Service, Intermountain Forest and Range Experiment Station.
Rosner, Hilllary. 2015. “Pine Beetle Epidemic: The Bug That’s Eating the Woods.”
Accessed June 3, 2015. http://ngm.nationalgeographic.com/2015/04/pine-
beetles/rosner-text

Safranyik, L. 1982. “Alternative solutions: Preventive management and direct control.”
In Proceedings, Joint Canada/USA Workshop on MPB related problems in
western North America. Victoria, B.C.: Canadian Forestry Service, Pacific Forest
Research Centre: 29-32.

Safranyik, L., and Bill Wilson. 2007. The mountain pine beetle: a synthesis of biology,
management and impacts on lodgepole pine. Canadian Forest Service.

Sharma, Rajeev. 2007. "Using multispectral and hyperspectral satellite data for early
detection of mountain pine beetle damage." PhD diss., University of British
Columbia.
Skakun, Robert S., Michael A. Wulder, and Steven E. Franklin. 2003 "Sensitivity of the
thematic mapper enhanced wetness difference index to detect mountain pine
beetle red-attack damage." Remote Sensing of Environment 86, no. 4: 433-443.
USDA. 2010. “Forest Health Protection and State Forestry Organizations: Management
Guide for Mountain Pine Beetle.” Accessed December 15, 2015.
http://www.fs.usda.gov/Internet/FSE_DOCUMENTS/stelprdb5187520.pdf
USDA. 2012. “Mountain Pine Beetle” In Major Forest Insect and Disease Conditions in
the United States: 2011. USDA Forest Service: 2-5.
USDA. 2013. “Forest Health Monitoring.” Last modified December 16, 2013. Accessed
December 12, 2015. http://www.fs.usda.gov/detail/r5/forest-
grasslandhealth/?cid=fsbdev3_046691
USDA. 2015a. “Aerial Detection Survey: Methodology.” Last modified December 16,
2013. Accessed July 24, 2015. http://www.fs.usda.gov/detail/r5/forest-
grasslandhealth/?cid=stelprdb5429568
USDA. 2015b. “Lake Tahoe Basin Management Unit: Who We Are and What We Do.”
Last modified December 16, 2013. Accessed July 23, 2015.
http://www.fs.usda.gov/detail/ltbmu/home/?cid=fsm9_046755
USGS. 2012. “Facts About Lake Tahoe.” Last modified December 12, 2012. Accessed
on June 11, 2015. http://tahoe.usgs.gov/facts.html
82
USGS. 2014. “Landsat Missions: Frequently Asked Questions about the Landsat
Missions.” Last modified June 19, 2014. Accessed July 23 2015.
http://landsat.usgs.gov/band_designations_landsat_satellites.php
Van Gunst, Kristin Jane. 2012. "Forest Mortality in Lake Tahoe Basin from 1985-2010:
Influences of Forest Type, Stand Density, Topography and Climate." PhD diss.,
University of Nevada, Reno.
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." Remote Sensing of Environment
96, no. 3: 340-351.

Wulder, Michael A., Caren C. Dymond, Joanne C. White, Donald G. Leckie, and Allan
L. Carroll. 2006a. "Surveying mountain pine beetle damage of forests: A review
of remote sensing opportunities." Forest Ecology and Management 221, no. 1:
27-41.

Wulder, Michael A., Joanne C. White, Nicholas C. Coops, T. Han, M. F. Alvarez, C. R.
Butson, and X. Yuan. 2006b. "A Procedure for Mapping and Monitoring
Mountain Pine Beetle Red Attack Forest Damage using Landsat Imagery."
Accessed July 11, 2015.
http://www.cfs.nrcan.gc.ca/pubwarehouse/pdfs/26209.pdf

Wulder, Michael A., Stephanie M. Ortlepp, Joanne C. White, Nicholas C. Coops, and
Sam B. Coggins. 2009. "Monitoring the impacts of mountain pine beetle
mitigation." Forest ecology and management 258, no. 7: 1181-1187.
Wulder, Michael A., and Steven E. Franklin. 2012. Remote sensing of forest
environments: concepts and case studies. Springer Science & Business Media.

Abstract (if available)

Abstract This work evaluates and reports the accuracy assessment of Maximum Likelihood Supervised Classification (MLSC) using the different stages of Mountain Pine Beetle (MPB) infestations outside the Lake Tahoe Basin Management Unit (LTBMU) using Landsat 8 OLI 30m and WorldView-02 (WV02) 2m (comparatively higher) spatial resolution imagery. Using ArcGIS 10.3, the accuracy of satellite imagery using MLSC and the Enhanced Wetness Difference Index (EWDI) provide a good comparison of the imagery at dissimilar spatial resolutions. ❧ MPB infestations at epidemic population levels can cause economic losses and have detrimental effects ecologically in Lodgepole pine tree stands. Detecting endemic populations of MPB can prevent epidemic infestations, preventing economic and ecological losses. After pre-processing, using the different stages of the MPB infestations as a control points, MLSC and the calculation of Tasseled Cap Transformation (TCT) indices (e.g., to calculate EWDI) are used to assess the accuracy of each type of imagery. The overall accuracy results of MLSC of Landsat 8 OLI 30m (51.22%) supersede those of WV02 imagery (26.82%) and are shown in error matrices within this thesis. Accomplishments of this project include the advantage to using WV02 imagery to locate MPB infestations at their endemic stage rather than relying on annual ADS’s. Improvements in positional accuracy of Global Positioning System (GPS) data collection devices and improved Remote Sensing (RS) software for image analysis may improve this analysis.

Linked assets

University of Southern California Dissertations and Theses

Conceptually similar

PDF

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

PDF

Comparing Landsat7 ETM+ and NAIP imagery for precision agriculture application in small scale farming: a case study in the south eastern part of Pittsylvania County, VA

PDF

Assessing the use of normalized difference chlorophyll index to estimate chlorophyll-A concentrations using Landsat 5 TM and Landsat 8 OLI imagery in the Salton Sea, California

PDF

Use of remotely sensed imagery to map sudden oak death (Phytophthora ramorum) in the Santa Cruz Mountains

Asset Metadata

Creator Nash, Norman Henry Biltz, Jr. (author)

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

School College of Letters, Arts and Sciences

Degree Master of Science

Degree Program Geographic Information Science and Technology

Defense Date 12/01/2015

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

Tag accuracy,assessment,GIS,Landsat 8 OLI,mountain pine beetle,OAI-PMH Harvest,remote sensing,WorldView02

Format application/pdf (imt)

Language English

Contributor Electronically uploaded by the author (provenance)

Advisor Fleming, Steven (committee chair), Vos, Robert (committee member)

Creator Email nnash@usc.edu

Permanent Link (DOI) https://doi.org/10.25549/usctheses-c40-218166

Unique identifier UC11278344

Identifier etd-NashNorman-4175.pdf (filename),usctheses-c40-218166 (legacy record id)

Legacy Identifier etd-NashNorman-4175.pdf

Dmrecord 218166

Document Type Thesis

Format application/pdf (imt)

Rights Nash, Norman Henry Biltz, Jr.

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