University of Southern California Dissertations and Theses

Exploring remote sensing and geographic information systems technologies to understand vegetation changes in response to land management practices at Finke Gorge National Park, Australia Between ...

(USC Thesis Other)

Exploring remote sensing and geographic information systems technologies to understand vegetation changes in response to land management practices at Finke Gorge National Park, Australia Between ...

PDF

Download

Open document

Flip pages

Download a page range

Download transcript

Copy asset link

Request this asset

Transcript (if available)

Content

Exploring Remote Sensing and Geographic Information Systems Technologies to Understand
Vegetation Changes in Response to Land Management Practices at Finke Gorge National
Park, Australia Between 1989 and 1999

by

Adlin Noelia Botkin

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 2018

Copyright © 2018 by Adlin Botkin
All rights reserved
ii

This thesis is dedicated to every scientist whose imagination and creativity were standardized
in a test, labelled, and placed in a symmetrical box; there is always time to open the box.

iii

Table of Contents

List of Figures ............................................................................................................................ v

List of Tables ........................................................................................................................... vii

Acknowledgements ................................................................................................................ viii

List of Abbreviations ................................................................................................................. x

Abstract ..................................................................................................................................... xi

1 Introduction ........................................................................................................................ 1

1.1 Thesis Evolution and Research Design ............................................................................. 3
1.2 Definition of Research Question and Project Stages ....................................................... 5

2 Literature Review and Related Work ................................................................................. 7

2.1 Early Remote Sensing .......................................................................................................... 7
2.2 Australian Arid Rangelands .............................................................................................. 10

2.2.1 Central Australia ................................................................................................ 12

2.3 Finke Gorge National Park................................................................................................ 16

2.3.1 Climate and Ecology .......................................................................................... 20
2.3.2 Vegetation and Soil ............................................................................................ 20
2.3.3 Feral Horses ....................................................................................................... 23
2.3.4 Long-Term Vegetation Monitoring Program .................................................... 24

2.4 Remote Sensing and Geographic Information Systems to Monitor and Manage
Natural Resources ....................................................................................................................... 28

2.4.1 Bitemporal Image Analysis................................................................................ 29
2.4.2 Multitemporal Image Analysis .......................................................................... 30

2.5 Principal Components Analysis ....................................................................................... 31

3 Data and Methods ............................................................................................................. 34

3.1 Seasonal Fractional Vegetation Cover Data ................................................................... 34

3.1.1 Model ................................................................................................................. 36
3.1.2 Pre-processing .................................................................................................... 37
3.1.3 Possible problems with the model ..................................................................... 38
iv

3.2 Rainfall and Temperature Data......................................................................................... 39
3.3 Field Survey Data from LVMP ........................................................................................ 41

3.3.1 Field Sites........................................................................................................... 41
3.3.2 Soil unit Data ..................................................................................................... 43
3.3.3 Plant Species Data.............................................................................................. 44

3.4 Methods ............................................................................................................................... 45

3.4.1 Classification...................................................................................................... 46
3.4.2 Principal Components Analysis ......................................................................... 48

4 Results .............................................................................................................................. 51

4.1 Supervised classification ................................................................................................... 51

4.1.1 Soil and Vegetation Classified Maps ................................................................. 52
4.1.2 Combined vegetation and soil classes................................................................ 56
4.1.3 Variability of Classes through Time .................................................................. 59

4.2 PCA Results ........................................................................................................................ 62

4.2.1 Principal Component 1 (PC1) – 76.1%.............................................................. 65
4.2.2 Principal Component 2 (PC2) – 6.2%................................................................ 66
4.2.3 Principal Component 3 (PC3) – 2.8%................................................................ 67
4.2.4 Principal Component 4 (PC4) – 1.7%................................................................ 68
4.2.5 Area of Interest – Palm Paddock ....................................................................... 69

5 Discussion and Conclusions ............................................................................................. 72

6 References ........................................................................................................................ 79

v

List of Figures

Figure 1 Earthrise photograph by William A. Anders, December 24, 1968 from Apollo 8...... 8
Figure 2 Map depicting IBRA boundaries in Australia. Source: Department of the Environment
and Energy, 2017 .............................................................................................................. 11
Figure 3 MacDonnell Ranges bioregion an its three subregions. Source: Department of the
Environment and Energy, 2017 ........................................................................................ 13
Figure 4a Hill near Heavitree Gap February 2013. Photo credit: Adlin Botkin...................... 14
Figure 4b Hill near Heavitree Gap February 2017. Photo credit: Adlin Botkin ..................... 15
Figure 5a Finke Gorge National Park is located near the geographic centre of Australia. Photo
credit: Adlin Botkin .......................................................................................................... 17
Figure 5b FGNP signage. Photo credit: Adlin Botkin ............................................................. 17
Figure 6 FGNP boundary and visible fence line. Source: ESRI Images ................................. 19
Figure 7 Rainfall data for Palm Valley Weather Station, located within the boundaries of
FGNP. Source: Bureau of Meteorology, 2016 ................................................................. 20
Figure 8 Livistona palms. Figure 8a in tropical area and Figure 8b in arid area. Photo credit:
Adlin Botkin ..................................................................................................................... 22
Figure 9 Wild horses in small numbers at FGNP, January 2016. Photo credit: Adlin Botkin 23
Figure 10 Location of LVMP survey sites ............................................................................... 27
Figure 11 Example of a pixel in a SFVC composite ............................................................... 36
Figure 12a Average Rainfall chart, Palm Valley Station, 1989 – 1999. Source: Bureau of
Meteorology, 2016 ........................................................................................................... 40
Figure 12b Average Maximum Temperature chart, Alice Springs Airport station, 1989 – 1999.
Source: Bureau of Meteorology, 2016 ............................................................................. 40
vi

Figure 12c Average Minimum Temperature chart, Alice Springs Airport Station, 1989 – 1999.
Source: Bureau of Meteorology, 2016 ............................................................................. 40
Figure 13 Soil Unit map. Source: Low, Foster and Berman, 1991 .......................................... 43
Figure 14 Field survey map for Site 3. Source: Low, Foster and Berman, 1991 .................... 45
Figure 15 Steps used in the supervised classification .............................................................. 47
Figure 16 Steps used in the PCA ............................................................................................. 50
Figure 17 Soil classification map ............................................................................................. 53
Figure 18 Vegetation classification map .................................................................................. 54
Figure 19 Vegetation and soil map .......................................................................................... 57
Figure 20 Classified pixels with significant vegetation cover over time ................................. 61
Figure 21 Image and Loadings for Principal Component 1 ..................................................... 65
Figure 22 Image and Loadings for Principal Component 2 ..................................................... 66
Figure 23 Image and Loadings for Principal Component 3 ..................................................... 67
Figure 24 Image and Loadings for Principal Component 4 ..................................................... 68
Figure 25a-d Images for Principal Components 1-4 for the area of Palm Paddock ................ 71

vii

List of Tables

Table 1 Description of LVMP sites ......................................................................................... 27
Table 2 Landsat 5 Satellite and Image Characteristics. Source: Terrestrial Ecosystem Research
Network, 2013 .................................................................................................................. 34
Table 3 Radiometric characteristics of TM sensor. Source: United States Geological Survey,
2017 .................................................................................................................................. 35
Table 4 Description of the four bands in the SFVC composites .............................................. 36
Table 5 SFVC Composites available from TERN with dates covered and quality of image .. 37
Table 6 Description of nine survey sites of LVMP. ................................................................ 41
Table 7 Soil units in the immediate vicinity of each site. Source: Low, Foster and Berman,
1991 .................................................................................................................................. 44
Table 8 General characteristics, soil preferences and root-type of dominant vegetation ........ 54
Table 9 Description of vegetation and soil classes .................................................................. 57
Table 10 Data for each significant vegetation cover class ....................................................... 62
Table 11 Data for each significant vegetation cover class ....................................................... 63

viii

Acknowledgements

I am grateful to my daughters Isabel and Camila. I am humbled at your patience and
consideration every time I entered my office to work on this project. You are the reason for
everything I do and I plan to do, because I love you and want the world to be good for you. I
will thank my husband Dustin, the best human being I know, for taking on many of my
responsibilities, never doubting me and supporting every single idea I have. You truly are my
rock. I am grateful to my mom Nilda for imparting in me the inability to ever give up. I want to
express gratitude to my military family for past and continued support without question. I want
to thank my mob in Alice Springs, my family “down under”, because it takes a town to write a
thesis.
I want to thank all my professors and staff at SSI at USC. Each one impacted my work
and my life in one way or another. Thanks to Dr. Karen Kemp who is an excellent teacher, an
inspiration, and everything I hope to become as a professor. I am very grateful to Dr. Katsuhiko
Oda, Dr. Su Jin Lee, Dr. Jordan Hastings, and Dr. Robert Vos for sharing all your knowledge
and guidance. I am grateful to Dr. Fleming, a fellow Army veteran, who was essential to the
successful completion of this project. He was able to understand my chaotic military life with
little more than a short call and allowed for more extensions than I can count. I also want to
thank Richard Tsung, who provided IT support across the world, most times after hours, over
the course of three years and three failed hard drives.
I want to thank the Joint Remote Sensing Research Project and Terrestrial Ecosystem
Research Network in Australia for making available the Landsat data I required to complete
this project. I want to express gratitude to Glenn Edwards at the Department of Resources and
ix

Land Management (DLRM) in Alice Springs, for accepting me as a volunteer and providing
me with office space and supplies for almost four years, during which time this thesis
developed. I want to thank Dr. W. A. “Bill” Low and his colleagues for his time and the
exceptionally comprehensive work conducted in the 1990’s, detailed in the Long-term
Vegetation Monitoring Program documents, which enabled my thesis. I am perpetually grateful
to Dr. Catherine Nano, Dr. Paul Box, and Dr. Jayne Brim Box with the Flora and Fauna
Division at the Department of Land Resources and Management, Northern Territory who
shared as much knowledge and expertise as I could retain and believed in me every step of the
way. Without their preceding work to build upon, and the work of others in the organization, I
would still be knee-deep in research. Thank you for your guidance, your support and, most
importantly, your friendship.
And finally, I want to acknowledge central Australia, my inspiration for this project. I
close my eyes and see the perfect blue skies, the magnificent trees and the fine, red dirt. May
our paths cross again soon because my heart is still in the Red Centre.

x

List of Abbreviations

BOM Bureau of Meteorology
BTEC Brucellosis and Tuberculosis Eradication Programme
DN Digital Number
ENSO El Niño - Southern Oscillation
ERTS Earth Resources Technology Satellite
FGNP Finke Gorge National Park
GIS Geographic information system
IBRA interim biogeographic regionalisation for Australia
JRSRP Joint Remote Sensing Research Program
LULC Land Use and Land Cover
LVMP Long-term Vegetation Management Project
NASA National Aeronautics and Space Administration
NT Northern Territory
PC principal Components
PCA Principal Components Analysis
RMSE Root-Mean-Square Error
RS Remote Sensing
SSI Spatial Sciences Institute
TERN Terrestrial Ecosystem Research Network
TM Thematic Mapper
USC University of Southern California
WWF World Wildlife Fund
xi

Abstract

This project aims to increase knowledge of vegetation changes in arid and semi-arid areas in
central Australia. Most of these zones are located across remote, sparsely-populated, large and
geographically diverse regions, making them difficult to study (Burns et al., 2014). Satellite
imagery and geographic information systems (GIS) are viable options to decrease the
knowledge gap in time- and cost-effective ways and to understand how vegetation changes in
areas with atypical annual seasons. The main goal of this thesis is to use modern techniques to
understand vegetation dynamics occurring during 1989 – 1999 in Finke Gorge National Park
(FGNP). During this time, land managers placed a fence around some park boundaries and
removed a significant number of wild horses to enable the vulnerable vegetation to recover. An
ensuing eight-year field study observed and documented changes. This thesis intends to do the
same, using remote sensing (RS) and GIS techniques. A supervised classification of soils and
plants is done using data collected during field surveys. Principal components analysis (PCA),
a data reduction technique, is used on multitemporal images to enhance continuous spatial and
temporal changes and to extract factors that can be attributed to land management efforts at
FGNP. Visual interpretation of components and analysis of classification information allowed
for exploration of vegetation dynamics at an appropriate spatial and temporal resolution to
understand variation and trends across time. The resulting components are compared to results
of previous field surveys conducted at the time. The principal components indicate there are
natural and human-derived sources of variation. Rainfall and other environmental factors play a
major role on vegetation recovery of areas inside the fence, however, components also indicate
that other sources of variation, such as land management practices conducted in the area, are
contributors to variation. The field survey results are comparable to the thesis results; however,
modern technique use provides a different perspective of trends and variation.
1

1 Introduction

An increasing number of areas in arid and semi-arid central Australia have been identified as
culturally and ecologically significant, requiring detailed information about how their condition
is changing through time. Most of these sites are located across remote, sparsely-populated, large
and geographically diverse regions, making them difficult to study (Burns et al., 2014).
Australia’s government has recognized the importance of a healthy environment to maintain its
way of life and has developed, and successfully implemented, several management policies at
national, regional and local levels to understand vegetation changes, protect the land and
conserve existing natural resources (Department of the Environment and Energy, 2016).
In arid central Australia, multiple forms of human-induced disturbance influence the
structure and composition of endemic vegetation (Bradshaw, 2012). Some land management
options are readily devised and implemented (e.g. fences for eroded areas, feral herbivore culls
to reduce over-grazing, and management burning to minimize the risk of uniformly-severe
wildfire). Management-related trends are difficult to distinguish due to central Australia’s
extremely variable climate (Bastin et al., 2012). The effects of these human-introduced factors
can be measured and partially quantified, but no one factor works in isolation. For example,
vegetation responses to rainfall are known to differ according to season, implicating factors such
as temperature and day length, as well as available plant soil moisture per se. Further, negative
trends such as an increase in bare soil may result from drought, over-grazing, increased fire or a
combination of all factors. Thus, temporal vegetation trends need to be examined in the context
of the prevailing climate and disturbance regimes, and at appropriate spatial and temporal scales.
2

To thoroughly understand landscape changes and the effectiveness of land management
efforts, multiple factors must be simultaneously considered, making remote sensing (RS) and
geographic information systems (GIS) valuable technologies to use. Further, Principal
Components Analysis (PCA) is a technique typically used to compress data from a large number
of variables to a smaller set, effectively reducing redundancy, minimizing data loss and
maximizing variation between variables. The application of this technique seems a good fit for
digital image processing and geographic time-series analysis. Notably, however, the technique to
date has been applied to a limited set of study systems, and its potential is therefore under-
explored.
Townshend, Goff, and Tucker (1985) for example, used PCA as an exploratory tool to
understand the relationships between multitemporal images in the continents of Africa and North
America. A decade later, Piwowar and Ledrew (1995) discussed PCA as one of three useful
techniques for hypertemporal image analysis to identify connections between variables and to
identify redundancy. In a contemporary study, Piwowar (1996) used PCA to isolate temporal and
spatial variations to identify patterns in ice concentration over a nine-year time-series. Most
recently, Henderson (2010) used PCA to understand vegetation productivity changes and normal
variation in areas of Grasslands National Park, Saskatchewan, Canada. To summarise their
contribution, these studies were successful in defining spatial and temporal localized anomalies
and highlighting the predictive capacity of the technique in relation to landscape changes across
a diverse set of landscapes in the continents of Africa, North America and the Arctic. It is
possible therefore, that PCA as applied to a set of time-series RS images will be similarly useful
for exploring patterns of vegetation change in remote and isolated areas of Australia.
3

1.1 Thesis Evolution and Research Design
A vegetation monitoring project, known as the Long-Term Vegetation Management
Project (LVMP), was conducted in central Australia in the 1990’s to assess changes after land
management efforts. This project monitored Finke Gorge National Park (FGNP), Northern
Territory (NT) and occurred between 1991 and 1998. The LVMP documented the vegetation
response in the years after large, grazing animals (mainly feral horses) were removed from the
FGNP and a fence was placed around some of the boundaries to minimise further incursion. The
intention of these land management efforts was to aid the recovery of native vegetation at this
important site (Low, Foster, and Berman, 1991). One main objective of the survey was to
document long-term vegetation response to these specific efforts, inside and outside of the fence
(Low, Foster, and Berman, 1991). During that time, satellite images were not readily available
and GIS tools were not specialized, nor accessible enough, to conduct advanced analysis or
calculations (Low, 2016).
After examining the survey documents and reports available, including an extensive
photographic record of each site inside and outside of the park boundaries, the LVMP was
selected as a model case-study for the thesis, using similar scope, spatial and temporal scales. In
this thesis, however, GIS techniques and RS image composites collected over the park in the
1990’s are used. The availability of the field survey documentation allows: (1) the establishment
of a baseline of how the area looked in 1991 and changed during that decade; to (2) compare and
validate the results from this analysis with the conclusions from the field survey, to (3) evaluate
the efficacy modern remote imagery techniques as a low-cost alternative to field surveys by way
of evaluating land management practices; and to (4) assess changes in central Australian remote
arid zones using GIS techniques and satellite imagery, beyond that which the field survey could
4

provide. Overall, the aim is to use modern techniques (RS and GIS components) to evaluate
vegetation responses to land management at FGNP during the same time frame as the long-term
vegetation field survey in the 1990’s. Although other studies have used PCA to explore changes
in vegetation cover, the technique has not been applied to central Australian landscapes. This
region is marked by highly variable and unpredictable inter-annual rainfall patterns (Box et al.,
2008), introducing a level of complexity not found in many other arid areas.
The potentially confounding effects of environmental (viz climate, atmospheric CO2,
wildfire) and anthropogenic (specifically feral herbivore herbage offtake, soil impacts, etc)
factors at FGNP is acknowledged from the outset of this study. Specifically, increasing shrub
cover through time may result from changed rainfall, decreased browsing or both. This makes
the task of assessing the efficacy of land management implementation somewhat challenging.
This was motivation to attempt to identify contributing factors, underlying variations and their
sources, and to quantify them. PCA was used to partition the influencing factors, to distinguish
the sources of variation, and to explore and summarize temporal and spatial change within the
park. Time-series image analysis, also referred to in the literature as hypertemporal or
multitemporal image analysis, is the basis to represent each component resulting from the PCA.
For instance, healthy and well managed areas may fail to respond due to seedbank loss, top-soil
loss and/or excessive runoff, while degraded sites may instead exhibit a pronounced response to
well-above-average rainfall. This means that short-term changes that are detected by remote
imagery need to be understood within a longer temporal context.
To address the above issues, it was best to use seasonal fractional vegetation cover
(SFVC) digital composites derived from geometrically and radiometrically corrected Landsat
images, between 1989 to 1999. Each composite represents the summary of land cover over a
5

three-month period and each pixel represents the percentage of cover of three continuous values:
bare ground, living vegetation and dry vegetation. The images were clipped to include FGNP and
surrounding areas. Out of the possible 40 composites (one for each season over a 10-year
period), only 27 displayed complete, cloud-free images of the park and were suitable for use. The
method used in this thesis is sensitive to missing data, and the exclusion of incomplete images
was a better option than their inclusion.
A supervised classification was completed for soil and vegetation using data from the
surveys as training samples. The classification method allows for decomposition and the
labelling of vegetation cover and soil units. The two classified maps were combined to make a
soil /vegetation map. A total of 16 classes were derived from this classification. Charts for each
class were constructed, resulting in an analysis of the vegetation and soil classes over time.
Following the classification, a PCA was performed on all the SFVC composites. The
resulting principal components (PC), as well as the information from the classification, were
used to explore spatial and temporal localized anomalies, based on knowledge of vegetation
composition and soil units. The use of time-series composites, versus a two-image analysis,
helped explain the response and variability of the vegetation community within the park at a
more appropriate time scale. This is a more gradual approach to land cover change analysis
because it considers all the changes that occur internally. The results of this project can provide
data to facilitate more efficient implementation of land management practices in vulnerable and
threatened bioregions.
1.2 Definition of Research Question and Project Stages
It is crucial to recognize the complexity and interdependence of all factors affecting the
vegetation communities in arid zones before enforcing a land management plan or determining
6

its success. This thesis presents an opportunity to use existing techniques and technologies to:
understand changes in vegetation in an area that has been observed and well-documented, their
connection to land management practices and to each other, as well as the opportunity to use
them as an exploratory technique.
The research question in this thesis was: Can similar vegetation changes be identified at
FGNP using modern techniques, as they were identified during the field survey in the 1990’s?
Specific objectives were:
- To evaluate the use of principal components analysis as a suitable technique to
understand sources of variation through time
- To determine which areas had the most and least variability in and around FGNP
- To assess the suitability of Landsat imagery temporal resolution from 1989 to 1999
- To correlate droughts and rainfall events with significant vegetation changes
- To compare the thesis findings to the findings of LVMP conducted in the area at the time
FGNP and surrounding areas encompass the spatial scale of this thesis.
The temporal scale is between 1989 and 1999, including the years when the Long-Term
Vegetation Management Project took place.

7

2 Literature Review and Related Work

Chapter 2 discusses where this thesis exists among other work. Section 2.1 discusses early
history of RS and GIS with a focus on Landsat use in arid zones, and what others have done to
study these areas. Section 2.2 examines Australian arid rangelands and the NT in terms of
environmental and land management factors. It describes the interim biogeographic
regionalisation for Australia (IBRA) framework, used to manage resources. Section 2.3 describes
FGNP in detail, since it was established and presents main issues the vegetation in the park has
endured, and the efforts by land managers and ecologists to understand the area and preserve.
This section includes detailed explanation of the long-term vegetation management project.
Section 2.4 touches on the use of RS and GIS in arid zones to monitor and manage natural
resources, to include change detection methods. Section 2.5 presents the possibility that a
mathematical technique such as PCA can provide information about the localized variations in
the area, to improve the understanding of complex processes, as well as vegetation response to
land management. RS, GIS, vegetation cover, overgrazing, invasive species, land management,
sources of variation and arid zones are topics contributing to the research question for this thesis
and the topics driving the literature review.
2.1 Early Remote Sensing
In December 1968, the crew of the Apollo 8 took and shared a remarkable colour image of
the Earth rising over the moon, known as Earthrise (Kluger, 2013) shown in Figure 1. It is
argued that this image, revealing a solitary, brightly-coloured sphere rising over the ashy lunar
landscape, inspired a new global perspective that set-in motion the environmental movement as a
response to furthermore understand our planet (Kluger, 2013). By the time Earthrise made
8

headlines, National Aeronautics and Space Administration (NASA) was well into the
development of the Landsat program, originally called Earth Resources Technology Satellite
(ERTS) program.

Figure 1 Earthrise photograph by William A. Anders, December 24, 1968 from Apollo 8.

The Landsat program was a RS satellite effort inspired by the early Moon-bound Apollo
missions, where pictures of Earth’s land surface were taken for the first time (NASA, 2017). The
experimental program had as a main purpose to study and monitor Earth landmasses and its
resources (NASA, 2017). In 1970, the Landsat program was officially approved and in 1972 the
first of its satellites, Landsat 1, was launched into orbit around the Earth (NASA, 2017). Digital
images from satellite sensors were preferred over aerial photographs because they could be
processed with computers, automatically enhanced, required less manual processing, cover a
larger area, and ultimately were more cost effective (Woodcock, Strahler, and Franklin, 1983.)
9

By 1974, the British Antarctic Survey used Landsat images to create and publish several maps of
Antarctica at the fraction of the cost (Fischer, Hemphill, and Kover, 1976). Story, Yapp, and
Dunn (1976) used Landsat imagery to successfully recreate about half of the patterns in
topographic maps of central Australia that had been conventionally surveyed between 1956 and
1957.
Over the following decades, largely due to Landsat’s success in providing quality
imagery of all land surface, a realization that Earth’s resources are finite ensued, highlighting the
importance of careful resource management with the aid of new technology (Strahler,
Woodcock, and Smith, 1986). GIS were concurrently developed as a complementary technology
to RS where spatially-referenced images were stored, processed, retrieved, and displayed
(Woodcock, Strahler, and Franklin, 1983.) As RS and GIS technologies improved, satellite
images, processing software and data analysis became increasingly available to the public and,
naturally, their application to previously hard-to-access, arid areas increased exponentially.
Prior to satellite digital imagery becoming accessible to the public, ground and aerial
photography was used to derive topographic, soil, vegetation, and environmental geology maps,
and was most useful to monitor surface conditions and land management planning (Woodcock,
Strahler, and Franklin, 1983.) Field surveys also served to collect data in remote or inaccessible
areas, however they lacked the temporal scale necessary to represent the dynamics of the study
area. Remotely sensed digital images are collected at large spatial scales and small enough
temporal scales to detect slower processes (Tueller, 1987). The ability of using RS and GIS for
image and data processing over large areas, without costly and time-consuming aerial
photography or field surveys is a significant advantage to land management efforts, especially in
arid, remote or scarcely populated zones.
10

2.2 Australian Arid Rangelands
The island continent of Australia covers an area of 7.69 million square kilometres with a
coastline almost 60,000 kilometres long (Geoscience Australia, 2016) and population of under
23.5 million, as of 2014 (Australian Bureau of Statistics, 2016). Arid rangelands make up 70% of
the continent, including arid and semi-arid inland areas (Smyth and James, 2004). This thesis is
concerned with those regions, referred to as central Australia hereafter.
Thackway and Cresswell (1995) described an interim biogeographic regionalisation for
Australia (IBRA) to establish common criteria for the conservation and management of
biodiversity. A number of regions were classified as distinct ecosystems, or ecoregions, based on
biogeographic regionalisation, geology, geomorphology, climate, present and natural vegetation,
and biogeographic knowledge about flora and fauna (Thackway and Cresswell, 1995). The
boundaries of bioregions and sub-bioregions, delineated as IBRAs, fluctuate as they are updated
when protected areas are identified (Department of the Environment and Energy, 2017). IBRAs
are not exactly bound by the limits on a map, but they are geographically distinct with common
characteristics such as climate, ecology, and geology (Department of the Environment and
Energy, 2017). Figure 2 shows the most recent IBRA boundaries. Desert and Xeric Shrublands
(in beige) is the dominating ecoregion in central Australia, where FGNP is located. This
ecoregion is further classified into smaller bioregions, and even smaller eco-regions.

11

Figure 2 Map depicting IBRA boundaries in Australia. Source: Department of the Environment
and Energy, 2017

The words ‘ecoregion’ and ‘bioregion’ seem to have a similar meaning in the literature
found; ‘ecoregion’ is used by the World Wildlife Fund (WWF) to refer to distinct flora, fauna
and environmental conditions within a geographical area, whilst ‘bioregion’ is used by the
Department of the Environment and Energy in Australia to describe the same concept, as well as
common geology. Bioregions place loose boundaries around areas with similar characteristics,
whereas ecoregions describe the life and conditions within these dynamic boundaries. These two
similar, yet different, categorizations are a significant point to mention because they reflect the
complications in depicting a dynamic environment in a static form and the difficulties in
managing these areas. Further complications arise when bioregions or ecoregions change,
requiring quick and deliberate changes to management plans.
12

2.2.1 Central Australia
Central Australia encompasses more than 175,000 sq. km, including FGNP. Its
dominating ecoregion, as well as for the NT, is Deserts and Xeric Shrublands (Department of the
Environment and Energy, 2017) (Figure 2). These habitats are generally characterized by
extreme temperatures, with evaporation exceeding rainfall, although each bioregion within is
distinctly classified (WWF, 2016).
The NT has an area of 1.35 million sq. km, equivalent to twice the size of Texas, and a
population of 200,000 (Australian Bureau of Statistics, 2016). Having such low population
density, the territory does not experience the same rates of urbanization or over-population as
coastal regions in Australia, resulting in advantages and disadvantages, ecologically speaking.
An advantage is that natural processes and resources are generally not under great pressure, and
human interaction with vegetation and wildlife is relatively low. In this respect, an area can be
studied without placing much weight on complex human variables. However, large and scarcely
populated areas tend to not be monitored frequently (Burns et al., 2014) and issues affecting
ecosystems can go unnoticed for extended periods of time.
The MacDonnell Ranges bioregion, with an area of 39,290 square kilometres, is
described as high relief ranges and foothills covered with spinifex hummock grassland, sparse
acacia shrublands and woodlands along watercourses (Thackway and Cresswell, 1995). Figure 3
shows that the bioregion is further divided into three smaller sub-bioregions (Hartz Range,
MacDonnell and Watarrka), to provide a more detailed description of the landscape (Department
of the Environment and Energy, 2017). FGNP is within the boundaries of the Watarrka sub-
bioregion (Department of the Environment and Energy, 2017). The specified boundaries support
administration and protection efforts in the area. However, it is understood that the processes that
13

affect, and are affected by, ecological factors may not have defined boundaries. Each of these
political, geographical and ecoregion boundaries influence FGNP in the way the park is used,
conserved, and managed. Research and surveys on land management effects aim to increase the
knowledge and understanding of complex ecological systems.

Figure 3 MacDonnell Ranges bioregion an its three subregions. Source: Department of the
Environment and Energy, 2017

2.2.1.1 Climate and Vegetation
The climate in Australia’s rangelands is exceedingly variable and unpredictable, with
several high rainfall events occurring in one year followed by several years without any
rain (Morton et al., 2011). Consequently, regional-scale biological dynamics are tightly
14

structured by a direct relationship between sporadic, large rainfall events and pulses of
plant growth and reproduction (Morton et al., 2011; Smyth and James, 2004). This relation
motivates the abrupt, and often striking transformation of usually red-brown landscapes
into green areas through the arid Australian centre (Wardle, Pavey, and Dickman, 2013).
Figures 4a and 4b show hills at Heavitree Gap in Alice Springs, east of FGNP. Both
pictures were taken mid-February, towards the end of the summer. Figure 4a was taken in
February 2013, during a dry summer. Figure 4b was taken in February 2017 after an
unusually wet year. In many places, vegetation follows a predictable pattern of growth,
however, central Australia’s rainfall patterns are unpredictable and variable leaving a stark
contrast between these two photographs taken almost four years apart.

Figure 4a Hill at Heavitree Gap February 2013. Photo credit: Adlin Botkin
15

Figure 4b Hill at Heavitree Gap February 2017. Photo credit: Adlin Botkin

Quick vegetation growth followed by long periods of drought can also increase the fire
fuel load, affecting the frequency, intensity and extent of bushfires (Griffin and Friedel, 1984).
Domestic and feral large animals have also placed pressure by significantly overgrazing areas,
thus displacing native perennial grasses and palatable shrubs (Smyth and James, 2004), and
degrading and eroding areas that serve as habitat to smaller animals (Morton et al., 2011).
Central Australia’s deserts and xeric shrublands, inclusive of FGNP and its surroundings,
are climatically dominated by high variability, generally very low rainfall, prolonged droughts,
and marked by occasional periods of high rainfall. Vegetation change monitoring requires the
consideration of the inconsistent and erratic rainfall patterns that mark the area (Amiraslani and
Dragovich, 2013). While the wet to dry seasons can be described as somewhat cyclical, the
cycles are irregular and can take decades to complete. Temperatures routinely exceed 40° Celsius
for most the summer and below freezing in the winter (Box, 2014-2016). Many plant species
16

remain dormant during periods of drought and harsh conditions, but become very active during
the infrequent rain episodes (Nano, 2014-2016). Field surveys conducted during droughts can
give the impression that few species live in the area. However, after periods of heavy rainfall it is
not uncommon for botanists and wildlife biologists to express surprise at the reappearance of
species presumed to be extinct from the region (Box, 2014-2016).
When areas are designated as vulnerable, land managers proceed with conservation
efforts. Management-related trends are difficult to distinguish due to central Australia’s
exceedingly irregular climate (Bastin et al., 2012). There may also be a substantial lag between
land management action and observable events in the region. For example, conservation
measures conducted during a dry period may not produce observable effects until several years
later when the next significant rainfall event occurs. So, it can be very difficult to correlate
ecosystem conditions with land management decisions, even when detailed records exist.
2.3 Finke Gorge National Park
FGNP is located 138 kilometres west of Alice Springs, near the geographical centre of
Australia, and occupies an area of 42,253 hectares of arid rangelands, shrublands and desert in
the NT (Figure 5a and 5b). The park contains a wide range of land forms, flora, fauna, cultural
and recreational areas. It is home to the rare and endangered Red Cabbage Palm (Livistona
Mariae), Western Arrernte people sacred sites and the Finke River, thought to be one of the
oldest rivers in the world sites (Parks and Wildlife Commission, 2011).

17

(5a)

(5b)

Figure 5a Finke Gorge National Park is located near the geographic centre of Australia. Photo
credit: Adlin Botkin. Figure 5b FGNP signage. Photo credit: Adlin Botkin

The first portions of the park were proclaimed a conservation reserve (Parks and Wildlife
Commission, 2011) in 1966 after Henbury Station, a private cattle operation, and the Finke River
Mission in Hermannsburg (Parks and Wildlife Commission, 2011) surrendered the lands to the
18

NT. The landscape was over-grazed and eroded for years prior to the incorporation of the park.
The FGNP was established in 1978 and more land portions were subsequently included. The
boundaries of FGNP, as we know them today, were officially recognized in 2004 (Parks and
Wildlife Commission, 2011). Since 2011, the park has been jointly managed between Arrernte
Traditional Owners and the NT Government, and recognized as being ecologically and culturally
important. FGNP and surrounding areas encompass the spatial range of this thesis.
Before major land management measures were taken in the 1980’s and 1990’s, feral
horses and cattle roamed freely, over-grazing, denuding the area, contaminating the scarce water
resources and eroding and compacting the soil (Box 2014-2016). Although cattle, camels,
donkeys, and other animals were present in the park, horses were the main concern because of
their larger population in the park (Graham and Johnson, 1986). Overgrazing by feral horses
(Equus caballus) systematically reduced native vegetation cover as animal numbers increased,
which in turn enabled soil erosion and compaction, especially in areas near water bodies (Box et
al., 2008). In arid zones, water resources are precious and vital to the sustainment of ecosystems
and the livelihood of people. A decrease in water quality and availability can result in
competition, species displacement or disease propagation (Nano, 2014-2016). These problems
are magnified in FGNP where invasive flora and fauna species cause significant stress to the
native wildlife (Parks and Wildlife Commission, 2011).
Many attributes of biodiversity can be used as measurable indicators to monitor natural
resources in arid zones (Smyth and James, 2004). Biodiversity indicators can be environmental
or biotic, indicating changes caused by pressure (Landsberg and Crowley, 2004). Landsberg and
Crowley (2004) argue that plants offer substantial information about their surroundings and are
valuable indicators of the pressure associated with land use. Landsberg and Crowley’s (2004)
19

focus on the national and regional scales limits the utility of these indicators to changes
occurring over smaller areas such as FGNP. The use of plants as indicators of change is not
novel but it is key to land managers and ecologists when studying complex and variable
ecosystems at a smaller, local spatial scale.
A fence placed around some areas at FGNP, in combination with drastic horse-removal
efforts, have allowed the vegetation of the area to recuperate (Brim Box, 2014-2016). As seen in
Figure 6, Landsat images show a clear boundary where the fence stands. The fence boundary has
become increasingly visible over time as the land management efforts have translated into
increased vegetation cover. FGNP has been routinely monitored and actively managed since
becoming a national park, making it a suitable candidate for hypertemporal image analysis using
RS and GIS.

Figure 6 FGNP boundary and visible fence line. Source: ESRI Images

20

2.3.1 Climate and Ecology
The regional climate in FGNP is semi-arid and hot, with temperatures routinely
exceeding 40° Celsius during the summer and falling below freezing during the winter (Box
2014-2016). Precipitation has high inter-annual variability, low predictability, is
characteristically low (< 250 mm per annum) and it is driven by El Niño - Southern Oscillation
(ENSO) cycle (Van Etten, 2009; Morton et al., 2011). Broadly speaking, the ecosystem in the
park typically exists in two states: a prolonged drought, or more rarely, under non-arid conditions
(> 350 mm per annum) underpinned by a concentration of discrete, but temporally connected
summer rain pulses (Nano et al., 2012). Figure 7 shows rainfall data for Palm Valley weather
station from 1989 to 1999, where five out of the ten years were under drought conditions.

Figure 7 Rainfall data for Palm Valley Weather Station, located within the boundaries of FGNP.
Source: Bureau of Meteorology, 2016

2.3.2 Vegetation and Soil
FGNP has over 680 plant species and a wide variety of habitats, soils and geographical features.
At finer scales, soil texture and relief are important to understanding productivity responses, as
21

well as the seasonal growth constraints of plant species (i.e. cool-season forbs) and attributes of
their root structures (i.e. deep-rooted perennials over annuals) (Nano and Pavey, 2013). For
example, relative frequent, small rainfall events can trigger a rapid and marked response in
shallow-rooted species on course-textures (water-yielding) soils, whereas plant responses on
clay-rich soils are comparatively rare and delayed. (Nano and Pavey, 2013).
Palm Valley, the most popular area at FGNP, lies in the Amadeus Basin aquifer and has a
large population of red cabbage palms (Livistona mariae), a threatened species (Wischusen,
Fifield, and Cresswell, 2004). The Hermannsburg sandstone beneath is a reliable source of water
(Lau and Jacobson, 1991) for Palm Valley with bores over 50 m deep. Underground, low salinity
water moves slowly through the sandstone below, causing favourable conditions (Wischusen,
Fifield, and Cresswell, 2004) for the red cabbage palms in the arid zone. Red cabbage palm’s
closest relative, Mataranka palm (Livistona rigida), exists 1000 miles north in the tropical
Mataranka region, where rainfall is abundant and other tropical plants are common (Wischusen,
Fifield, and Cresswell, 2004). The unique hydrogeology of Palm Valley suggests that red
cabbage palms have found a flora refuge at Palm Valley allowing these tropical plants to thrive
in such arid conditions (Wischusen, Fifield, and Cresswell, 2004). This indicates that the
response of these deep-rooted plants to the rainfall would be a stark contrast to shallow-rooted
plant species. Figures 8a (Livistona rigida) and 8b (Livistona mariae) show mature Livistona
palms thriving in two different areas. Figure 8a depicts Mataranka palms in tropical Rainbow
Springs, Mataranka and Figure 8b depicts red cabbage palms in arid Palm Valley, FGNP.
22

(8a)

(8b)

Figure 8 Livistona palms. Figure 8a in tropical area and Figure 8b in arid area. Photo credit:
Adlin Botkin

These broad-scale and fine-scale environmental drivers of primary productivity interact
in complex ways on the vegetation to determine ground-cover dynamics in the park. In addition
to the vegetation, geography and hydrogeology, land management efforts contribute to the
23

complexity of assessments of relationship between plant species and a series of connected rain
events.
2.3.3 Feral Horses
Feral horses are considered an environmental threat in central Australia and populations
can increase by 20% each year if not managed (Csurhes, Paroz, and Markula, 2009). They can
travel up to 50 kilometres from water sources in search of food, displacing native wildlife by
competing for already scarce resources (Csurhes, Paroz, and Markula, 2009). Feral horse
population in the NT was estimated at 50,000 in 1976 and not deemed a pest at such low
population density (Graham and Johnson, 1986). However, when the brucellosis and tuberculosis
eradication programme (BTEC) was developed to eliminate bovine tuberculosis from Australia
(Gee, 1986), horses were viewed as a potential hindrance to the goal of the program (Graham and
Johnson, 1986). A water buffalo survey conducted in 1982 (Graham and Johnson, 1986),
collaterally estimated the population of feral horses in the NT at 76,000, a number considerably
higher than accepted before.

Figure 9 Wild horses in small numbers at FGNP, January 2016. Photo credit: Adlin Botkin
24

An increasing concern about the economic impact of horses on the BTEC and the
conflicting reports of population totals led to further research. An ensuing aerial survey of
horses, and other large animals, completed on May 1984 documented their distribution and
abundance in known horse range areas. The survey found that the population of horses was much
higher than originally estimated, with total population numbers surrounding Alice Springs at
over 82,000 and population density at FGNP between 0.3 and 1 horse per square kilometre
(Graham and Johnson, 1986).
Continued efforts to eradicate bovine tuberculosis in the area resulted in a perimeter
fence project that began in 1986 and was completed in 1989 (Day, 1989). Through the multi-year
project, feral horses caused severe damage to the fence, taking it down in some places (Day,
1989). In the months prior to the completion of the fence, the Finke Gorge Muster and Shoot was
proposed, and approved, to mitigate further damages by removing horses and cattle from the
park and from a 10-kilometre buffer around the park (Day, 1989). Between 1986 and 2001, a
total of 32,881 horses were removed from FGNP and surrounding areas, the majority in the first
three years of the program (Low and Hewett, 1990).
The removal of horses and cattle from the park and the placement of the fence resulted in
an opportunity to monitor vegetation recovery in the area (Berman, 1990). A long-term
vegetation monitoring program was funded to understand the environmental implications of
removing the horses from the park by observing the vegetation.
2.3.4 Long-Term Vegetation Monitoring Program
After the major land management efforts were completed in 1990, the long-term
vegetation monitoring project commenced in and around FGNP. The ‘Environmental
25

Implications of Horse Removal in FGNP’ project is documented in five reports between 1991
and 1998 (Low, Foster, and Berman, 1991; Low et al., 1992; Low et al., 1993; Cook et al., 1995;
Miller, Low, and Matthews, 1998), and referred to as the Long-Term Vegetation Monitoring
Program (LVMP) for the remaining of this thesis. It was led by Dr. W. A “Bill” Low at Low
Ecological Services in Alice Springs. The project was intended to monitor the recovery of the
vegetation in areas where animals were removed and the observations compared to vegetation in
areas where they remained. LVMP was conducted in five phases (1991, 1992, 1993, 1995, and
1998) over eight years, collecting vegetation information such as herbage composition, species
frequency, number of species per site, biomass, density of juvenile trees, canopy cover and some
ecological indices.
The results of the LVMP were complex, and each site studied reflected its landscape and
vegetation attributes (Low, Foster, and Berman, 1991; Low et al., 1992; Low et al., 1993; Cook
et al., 1995; Miller, Low, and Matthews, 1998). The multi-year survey concluded that the sites
inside the fence had greater biomass than the sites outside the fence when comparing the yield
from 1991 to 1998 (Miller, Low, and Matthews, 1998). However, environmental effects, and
more specifically rainfall, profoundly affected the vegetation changes at intervals between those
years (Miller, Low, and Matthews, 1998). In addition to the seemingly recovery of native plants
inside the park boundary, invasive grasses, such as couch and buffel grasses, increased in
coverage (Miller, Low, and Matthews, 1998).
Dr. Low, who continues to provide services in Alice Springs, agreed to discuss the
LVMP he and his colleagues conducted between 1991 and 1998. Dr. Low was enthusiastic about
the use of GIS and satellite images looking back at the LVMP and provided as many details as
possible.
26

The scope of the LVMP was extensive and included the examination of species
composition, total biomass and trends for herbage, trees and shrubs, soil stability, water quality,
rabbit activity, and mapping of land systems. The main goal of the LVMP was to monitor
vegetation response and trends after the horses and cattle were nearly eradicated from FGNP and
a strategic fence was emplaced to keep any remaining large animals entering the park. The
LVMP is a loose model to define scope and scale in this thesis.
The original proposed schedule divided the LVMP into five stages, with the initial survey
and site selection taking place July – August 1990, and the remaining four surveys occurring 12
months after the previous one. The project did occur in five stages, however, the time of
sampling was altered due to drought, poor rainfall, and lack of funds. Surveys 1-5 were
conducted on May 1991, May 1992, September/October 1993, February 1995, and March 1998,
respectively. It is important to consider the implications of comparing data that has been
collected in different seasons. Rainfall is another factor to be considered when comparing data in
the surveys; if rainfall is lacking the vegetation can dry quickly in high temperatures, on the
other hand, even a small amount of rain can quickly turn dry vegetation green. Seasonality and
precipitation are the two most important elements affecting plant composition and productivity,
followed by soil composition and grazing pressure (Foran, 1986).
Nine sites, shown in Figure 10, representing a wide range of landforms were selected for
monitoring across the FGNP. Three sets of paired sites (1&2, 3&4, and 7&8) were selected to
compare similar areas, inside and outside the park. The other three sites (5, 6, and 9) were
representative of other landforms in areas known to have been heavily grazed within the park,
but no comparable paired sites were found for any of them (Low, Foster, and Berman, 1991).
27

Figure 10 Location of LVMP survey sites

Table 1 Description of LVMP sites
Site Number Location Grazing
1 (Paired) Palm Paddock – South Inside Park Light
2 (Paired) Palm Paddock – South Outside Park Moderate
3 (Paired) Palm Paddock – West Inside Park Light
4 (Paired) Palm Paddock – West Outside Park Moderate
5 Junction of Palm Creek and Finke River Heavy
6 Boggy Hole Waterhole Heavy
7 (Paired) Illbilla Dune Field - Inside Park Light
8 (Paired) Illbilla Dune Field - Outside Park Light
9 Circle Gully Heavy

28

2.4 Remote Sensing and Geographic Information Systems to Monitor and Manage
Natural Resources
A study by Burns et al. (2014) investigating the extent of monitoring of Australia’s
ecosystems found that arid areas are insufficiently sampled and monitored, although they make
up the majority of the country. Arid zones have limited resources, so the effects that fire,
overgrazing, and introduced animals and plants can have on ecosystems may be felt for long
periods of time after they occur, or even permanently.
Remotely sensed imagery is an important source of data in the monitoring of natural
resources due to its practicality in Land Use and Land Cover (LULC) applications (Hussain et
al., 2013). The differences in the spectral signatures between areas of LULC change and areas
that have not changed can be identified with multi-band RS images. These differences may
provide insight into the ongoing processes and how they behave and affect the area. Pixel-based,
object-based, and spatial data mining change-detection techniques have been developed and
continue to be developed (Hussain et al., 2013), as satellite images become finer and more
accessible.
Most Landsat data became available free of charge to the public January 2009,
facilitating user access to a retroactive archive of multi-spectral images with an 18-day or less
coverage cycle (NASA, 2017). Single images of natural resources such as waterbodies,
vegetation and mountain ranges, although important to understand spatial differences within the
area imaged, do not provide comprehensive information about the ongoing processes when
looked at individually. However, the analysis and comparison of two images of the same area at
different times can yield comparative data to help understand the changes transpired in that
period. Moreover, multiple images analysed over time can uncover trends or slower processes.
29

2.4.1 Bitemporal Image Analysis
Pairs of images are commonly used to detect changes on the landscape. The process, also
known as bitemporal change detection, compares two images of the same geographic area taken
at different times to detect and measure changes. Several RS methods have been established to
identify bitemporal changes.
Celik (2009b) describes an unsupervised change detection method to extract feature
vectors for each pixel so that it automatically considers the contextual information of the
neighbourhood for each block. A related study also by Celik (2009a), outlines a similar method
for paired images using k-mean clustering to calculate final change detection and PCA to reduce
dimensionality and extract features. This method identifies significant changes irrespective of the
nature of the image used as input (Celik, 2009a).
Collins and Woodcock (1996) describe several techniques for comparing pairs of images,
including simple digital number (DN) matching, where two images are stacked one above the
other and PCA is performed on the DN. Byrne, Crapper, and Mayo (1980) described a similar
method overlaying a pair of Landsat MSS images of a coastal town to detect changes near the
shore.
Bovolo, Marchsi and Bruzzone (2012) describes a number of techniques including
Bayesian changes and PCA for quantifying the differences in two images of the same area. The
study accounts for the difficulties of both, the collection of ground truth data and the possible
loss of change information when using unsupervised methods; issues that continually arise when
attempting to detect change in vast and unpopulated arid zones (Bovolo, Marchsi and Bruzzone,
2012). The technique presented is advantageous because it does not require previous knowledge
of the changes occurred (Bovolo, Marchsi and Bruzzone, 2012).
30

Other bitemporal vegetation change detection studies have developed methodologies
appropriate to the ecoregion of interest, such as coastal environment (Weismiller et al., 1977),
wetland (Howarth and Wickware, 2007), and desert (Abuelgasim et al., 1999). Using bitemporal
analysis is advantageous because it minimizes the atmospheric, sensor and environmental
differences between multi-temporal images (Sun et al., 2016).
2.4.2 Multitemporal Image Analysis
At a lower temporal resolution, ecological dynamics requiring frequent measurements
may be overlooked, requiring time-series image analysis. When images are collected at small
enough intervals they can be used to detect changes and understand trends in highly variable
ecoregions (Lawley, Lewis, and Ostendorf, 2016).
A study by Deng et al. (2008) uses PCA as a pre-classification spectral change-detection
technique to transform a stack of hypertemporal images. The combined bands of the resulting
image are transformed into components. It is further processed using a classified analytical
method to produce labelled change-detection output. This is a type of “hybrid” analysis where
more than one technique is used to process the images. Amiraslani and Dragovich (2013)
presented a method to investigate vegetation changes in a degraded rangeland area. They
considered a 42-year rainfall period and acquired images based on major rainfall or drought
events. This method allowed enough time to understand the dynamics at a more appropriate
temporal scale, minimizing drastic periods of change.
Sparrow, Friedel, and Stafford Smith (1997) developed a model to predict changes of
chenopod vegetation in an area south of Alice Springs in Australia, considering the small amount
of data available, climatic variability and vegetation heterogeneity. The changes in vegetation in
the area of their study are most dependent on grazing effects and soil erosion. The study uses a
31

plant cover index derived from Landsat images to test the model, resulting in a successful
association of the index with the field data, except in highly eroded areas. Bastin et al. (2012)
developed a successful method using 89 Landsat images over a tropical savannah to detect
changes in ground cover relating to land management. Their study demonstrates the utility of this
type of analysis at a paddock scale for those trying to monitor land cover changes.
In addition to the typical visualization of extent of herbage cover, Landsat’s capability to
record images in different bands of the electromagnetic spectrum provides information about the
type and health of the vegetation. Graetz et al. (1976) used multi date and multi spectral imagery
to assess rangeland type, condition and response to rainfall in arid central Australia. They found
that the method was most useful where fenceline contrast in the vegetation condition were visible
(Graetz et al., 1976). Matheson (1994) used multiple Landsat infrared images to analyse green
vegetation cover relative to the background in arid area in the NT of Australia. Vegetation
darkening enabled spectral separability of green vegetation in semi-arid rangelands and arid
sandy areas.
2.5 Principal Components Analysis
Principal Components Analysis is a mathematical technique that reduces the number of
correlated variables into a lesser number of uncorrelated factors. Each resulting component
explains the variability independent of other components. That is, principal component 1, or
PC1, will create a new variable in terms of a new axis and will have no correlation to PC2, which
will present the next highest source of variability.
In earth and environmental sciences, PCA has traditionally been used for land cover
change detection (Byrne, Crapper, and Mayo, 1980; Ingebritsen and Lyon, 1985), image
enhancement (Deng et al., 2008; Alavi, 2012), and climate data analysis (Kelly et al., 1982;
32

Farhangfar et al., 2016). PCA has also been used in the search of less discrete changes that may
provide more details of underlying factors responsible for localized variation across space and
time.
Townshend, Goff, and Tucker (1985) used PCA as an exploratory tool to understand the
relationships between multitemporal images in the continents of Africa and North America. They
compared the resulting components from both continents and found significant similarities in the
variation structures of underlying relationships. The first two principal components (PC) for each
continent explain over 90% of the variation, with the first one following rain patterns and the
second one seasonal variation.
Piwowar and Ledrew (1995) discuss PCA as one of three useful techniques for
hypertemporal image analysis to identify connections between variables and to identify
redundancy. The authors argue that PCA can assist in understanding polar climate changes due
to the technique’s capability to generalize a time-series of RS images. In his thesis, Piwowar
(1996) uses PCA to isolate temporal and spatial variations to identify patterns in ice
concentration over a nine-year time-series. Despite influence from interannual, regional and
seasonal changes, Piwowar (1996) identified three new phase-shifted regionalisms in the area,
evidencing the suitability of PCA to highlight underlying relationships and trends in melting ice
that may begin and dissipate within the studied timeframe.
Bengraine and Marhaba (2003) used PCA to extract sources of variation and understand
these in a temporal and spatial context to understand water pollution in the Passaic River, New
Jersey. PCA was useful in explaining opposing patterns of organic, biological and chemical
pollution. The authors were able to identify stations that affected the water quality, negatively
and non-negatively.
33

A thesis written by Henderson (2010) used PCA to understand vegetation productivity
changes and normal variation in areas of Grasslands National Park, Saskatchewan, Canada. She
interpreted the components spatially by looking at each resulting image and temporally by
plotting the loadings. Henderson (2010) chose to explore pixels with high spatial and temporal
definition, resulting in the identification of types of changes in areas showing high variation. In
her conclusion, she explains the possibility of predicting how the vegetation in the park will
respond to climate change.
It is common to use change detection methods to measure how much the land has
changed from time A to time B. But to understand patterns and trending over time, a multi-
temporal set of images may be more appropriate. PCA can isolate the sources of variation (each
as an orthogonal component), to assist in detecting patterns and trends, that are uncorrelated to
other sources of variation. Although most of the variation can be explained with the first
component, the subsequent higher components further identify other sources of variation that
may be related to plant species, soil texture, specific rainfall events, or other indicators. The
importance of the higher PCs, presenting increasingly smaller percentages of variation, may
seem subjective because it is integrally dependent on the knowledge and ability of those
interpreting the images (Davis, 2002).

34

3 Data and Methods

Chapter 3 describes the framework of this thesis, including the data used and the processes
followed. Section 3.1 describes the seasonal fractional cover composites derived from Landsat
images that were the principal source of remotely sensed data. Section 3.2 describes other data
used, mainly rainfall data and information derived from the LVMP. Section 3.3 explains the
methods used to process the data, including supervised classification and principal components
analysis (PCA).
3.1 Seasonal Fractional Vegetation Cover Data
The main source of satellite images for this thesis were seasonal fractional vegetation
cover (SFVC) composites with path number 103 and row number 77, published by the
Terrestrial Ecosystem Research Network (TERN) with data from the Joint Remote Sensing
Research Program (JRSRP) at the University of Queensland in Brisbane, Australia between the
years 1989 and 1999. SFCV composites for those years were derived from images captured by
Landsat 5 Thematic Mapper (TM) sensor (Terrestrial Ecosystem Research Network, 2013).
Landsat 5 satellite and image characteristics and the radiometric characteristics of TM sensor are
listed in Table 2 and Table 3.

Table 2 Landsat 5 Satellite and Image Characteristics. Source: Terrestrial Ecosystem Research
Network, 2013
Characteristics Description
Ground Sampling Interval (GSI) 30 x 30 m – for bands 1 – 5 & 7
120 x 120 m – for band 6
Swath width 185 km
Repeat coverage interval 16 days, 233 orbits in each 16-day interval
Altitude 705 km
Quantisation B bits (256 levels)
On-board data storage Magnetic tape failed
35

Orbit type Sun-synchronous
Inclination 98.2
Equatorial crossing Descending node at 10:10 AM
Image size 185 x 172 km
Number of bands 7

Table 3 Radiometric characteristics of TM sensor. Source: United States Geological Survey,
2017
Band Spectral Range Electromagnetic Region
1 0.45 ~ 0.52 Visible blue
2 0.52 ~ 0.60 Visible green
3 0.63 ~ 0.69 Visible red
4 0.76 ~ 0.90 Near infrared
5 1.55 ~ 1.75 Middle infrared
6 10.40 ~ 12.5 Thermal infrared
7 2.08 ~ 2.35 Middle infrared

The SFVC images are time-serial composites of fractional, sub-pixel vegetation cover
over Australia. They are regular time-series representative of each typical atmospheric season
and capture the variability, whilst minimizing data gaps that may occur in single images (Flood,
2013). Each one is created by selecting representative pixels using the medoid (three-
dimensional median) of a three-month period (seasons) of fractional cover of satellite images.
Each year has four seasons; Summer (December to February), Autumn (March to May), Winter
(June to August), and Spring (September to November), resulting in a maximum of four
composites per year. The fractions for each pixel represent the portions of bare, green, and non-
green cover. The Landsat images used to build the SFVC composites have been corrected for
atmospheric effects, bi-directional reflectance and topographic effects using the methods
described in Flood et al (2013). Images with over 80% cloud cover were excluded to reduce the
possibility of extra noise (Terrestrial Ecosystem Research Network, 2013).
36

3.1.1 Model

The fractional cover of the bare soil, green and dry vegetation cover was assessed using
models based on substantial sampling of more than 1,500 sites covering a wide variety of
vegetation, climate, and soils in Australia (Terrestrial Ecosystem Research Network, 2013),
following methods described in Muir et al. (2011). The values are calculated by inverting
multiple linear regression estimates and a least squares non-mixing.
Values present a percentage of cover and range from 0 to 100 and, increased by 100 to
allow for values under 0 and values over 100 (undershoots or overshoots of modelled values)
(Terrestrial Ecosystem Research Network, 2013). Each pixel is categorized into three continuous
variables representing the fraction of cover stored over three bands or layers; see Table 4 for
band descriptions. A very simple example of such pixel is shown in Figure 11. A fourth band
contains the root-mean-square error (RMSE) between the predicted and actual pixel value. For
this thesis, 100 was subtracted from each value, and represented each pixel as a point in the
percentage scale.

Figure 11 Example of a pixel in a SFVC composite

Table 4 Description of the four bands in the SFVC composites
Band Category Description
1 Bare (red) Bare ground, rock face, disturbed soil
2 Green vegetation
(green)
Live vegetation, green vegetation
3 Non-green vegetation
(blue)
Dead plants, dead leaves, dry vegetation, dormant
vegetation, branches, trash
4 Model fitting error Error layer representing root-mean-square error
37

The construction of the perimeter fence and the initial large-scale horse removal effort at
FGNP was completed by May 1990. The LVMP’s last survey was completed in April 1998. So,
the images considered for this thesis cover the timeframe between December 1989 and
November 1999, slightly before and slightly after these two dates.
3.1.2 Pre-processing
The composites were derived from geometrically and radiometrically corrected Landsat
images. Fractional cover of vegetation and bare soil were estimated using models based on
extensive sampling of several hundred sites covering the variety of vegetation and climate types
throughout Australia (Terrestrial Ecosystem Research Network, 2013) and applied retroactively
to historical Landsat imagery to provide consistent classification of land cover over time. Each
layer used in this analysis represents a summary (composite) of the land cover over a three-
month period. As the methods used are sensitive to missing data, only complete or almost
complete, cloud free composites were used. This gave a total of 27 images between the years of
1989 and 1999, out of the possible 40. List of available composites is shown in Table 5.
Table 5 SFVC Composites available from TERN with dates covered and quality of image
Composite Dates covered Data Quality
1 December 1989 to February 1990 Full Image
2 March 1990 to May 1990 Missing some data
3 June 1990 to August 1990 Missing some data
4 September 1990 to November 1990 Full Image
5 December 1990 to February 1991 Full Image
6 March 1991 to May 1991 Missing some data
7 June 1991 to August 1991 Missing some data
8 September 1991 to November 1991 Full Image
9 December 1991 to February 1992 Not enough coverage
10 March 1992 to May 1992 Full Image
11 June 1992 to August 1992 Not enough coverage
38

12 September 1992 to November 1992 Not enough coverage
13 December 1992 to February 1993 Not enough coverage
14 March 1993 to May 1993 Missing some data
15 June 1993 to August 1993 Missing some data
16 September 1993 to November 1993 Not enough coverage
17 December 1993 to February 1994 Not enough coverage
18 March 1994 to May 1994 Missing some data
19 June 1994 to August 1994 Missing some data
20 September 1994 to November 1994 Missing some data
21 December 1994 to February 1995 Full Image
22 March 1995 to May 1995 Not enough coverage
23 June 1995 to August 1995 Not enough coverage
24 September 1995 to November 1995 Missing some data
25 December 1995 to February 1996 Full Image
26 March 1996 to May 1996 Not enough coverage
27 June 1996 to August 1996 Missing some data
28 September 1996 to November 1996 Full Image
29 December 1996 to February 1997 Not enough coverage
30 March 1997 to May 1997 Not enough coverage
31 June 1997 to August 1997 Not enough coverage
32 September 1997 to November 1997 Missing some data
33 December 1997 to February 1998 Missing some data
34 March 1998 to May 1998 Missing some data
35 June 1998 to August 1998 Missing some data
36 September 1998 to November 1998 Not enough coverage
37 December 1998 to February 1999 Full Image
38 March 1999 to May 1995 Missing some data
39 June 1999 to August 1999 Missing some data
40 September 1999 to November 1999 Full Image

3.1.3 Possible problems with the model
The calibration and models were originally developed for agricultural land and
rangelands in temperate and subtropical regions of Australia, with relatively little representation
of calibration points in the Australian desert. So, the existing model is not perfectly suited for
extrapolation into remote rangeland areas. For example, some fractional values were slightly
smaller than 0%, or slightly larger than 100%.
39

The months covered in each composite align with typical atmospheric seasons. However,
as discussed in section 2.2.1.1, the climate in the central Australian arid areas is unpredictable
and highly variable causing anomalies such as the seemingly “skip over” a typical wet, summer
season when there is a drought, possibly lasting years. Using composites that summarize each
season, may introduce errors in the application of any of the methods as well as in the analysis
part. However, due to the consistent classification criteria across time for the study area and
given that they are still reasonably indicative of vegetation cover values, the model and images
were suitable for this thesis.
The methods used to analyse the images are sensitive to missing data. 40 composite
images were available for use, but only 27 had complete, or almost complete coverage over
FGNP. The remaining images were not included in the analysis, introducing a level of error that
may be difficult to measure as there is no way the data can be re-collected, and there are no other
known sources of data, except for the LVMP. However, as one of the points of this thesis is to
use what was available at the time, this is acceptable.
3.2 Rainfall and Temperature Data
Rainfall data was derived from the Palm Valley Bureau of Meteorology (BOM) Station,
located within the boundaries of FGNP. Temperature data was not available at Palm Valley
Station, so data from the Alice Springs Airport station was used, located about 200 km east of
FGNP. The rainfall, minimum and maximum temperature has been recorded daily since the early
1980’s and is available as bulk downloads from the BOM website. They are presented as
averages in Figures 12a, 12b and 12c, each covering the three-month timeframe of a single
SFVC composite.
40

Figure 12a Average Rainfall chart, Palm Valley Station, 1989 – 1999. Source: Bureau of
Meteorology, 2016

Figure 12b Average Maximum Temperature chart, Alice Springs Airport station, 1989 – 1999.
Source: Bureau of Meteorology, 2016

Figure 12c Average Minimum Temperature chart, Alice Springs Airport Station, 1989 – 1999.
Source: Bureau of Meteorology, 2016
41

3.3 Field Survey Data from LVMP
The long-term vegetation monitoring project (LVMP) was conducted in five stages
between June 1991 and April 1998 over nine field sites in FGNP. Paper maps from the initial
survey’s documentation were scanned, converted to Tiff file and georeferenced using land
feature association. The paper maps provided soil unit, plant cover, and prominent features of the
nine field sites surveyed in LVMP and surrounding areas. Additionally, conversations with
ecologist Dr. Catherine Nano and Dr. Paul Box (Nano, 2014-2016; Box, 2014-2016) about the
dominant plant species further enriched the classification.
3.3.1 Field Sites
The nine field sites were selected to represent distinct vegetation, topography, and soil
composition, inside and outside of the park. In Table 6, vegetation, soil and landform data for
each site was tabulated to understand what the area looks like in more detail. This data was
derived from the field observation made by Low, Foster and Berman (1991).

Table 6 Description of nine survey sites of LVMP.

Location Vegetation Soil Landform
1 – South Palm
Paddock (Paired
with 1, Located
inside park)
Open Witchetty bush
shrubland.
Light sandy clay
loam, overlies
weak, kaolinized
sandstone.
Gently sloping
hillslope plain of
low relief.
2 - South Palm
Paddock (Paired
with 2, Located
outside park)
Sparse Witchetty bush
shrubland over Five
Minute grass.
Light sandy clay
loam, overlies
weak, kaolinized
sandstone.
Gently sloping
hillslope plain of
low relief.
3 - West Palm
Paddock (Paired
with 4, Located
inside park)
Sparse Witchetty bush
shrubland over Five
Minute grass.
Platy, kaolinized
sandstone, red
sandy loam.
Hillslope within
plain of low
relief.
42

4 - West Palm
Paddock (Paired
with 3, Located
outside park)
Sparse, witchetty bush
shrubland. Other shrubs
and grasses.
Calcrete
substrate, sandy
claim loam.
Hillslope within
plain of low
relief.
5 - Palm Creek
(Not paired)
Sparse Senna with
Witchetty bush over buffel
and other grasses.
Light sandy clay
loam. River
gravel below.
Within pediment
extending into
flood plain.
6 - Boggy Hole
Waterhole (Not
paired)
Sparse Mulga shrubland
over Senna, Buffel grass
and other grasses. Sparse
River Red gum woodland
over Couch grass.
Yellowish red
sandy loam and
red light sandy
clay loam.
Alluvial landform
on channel bench
within alluvial
terrace. Bar plain.
7 - Illbilla
Dunefield
(Paired with 8,
Located inside
park)
Open Hummock grassland
of Spinifex. Wattle,
Ironwood and Blue
Mallee.
Red sandy loam. Wind-eroded
sparse dunefield.
No drainage
features.
8 - Illbilla
Dunefield
(Paired with 7,
Located outside
park)
Open Hummock grassland
of Spinifex. Mulga,
Wattle, Ironwood and
Blue Mallee.
Red sandy loam. Wind-eroded
sparse dunefield.
9 - Circle Gully
(Not paired)
Sparse Acacia shrubland
with Whitewood. Mulga,
Fuschia bush and Senna.
Strong, brown
sandy loam over
strong brown
sandy clay loam.
Gently sloping
with low relief,
within pediment.
Erosion gullies
present.

The location of each field site was confirmed with GPS in 2014 and an MXD file was
created to denote each corner of each site. Although the location of each field site was identified
in 1991 with markers in each corner, not all remain. Most points of the GPS-confirmed corners
correspond closely to the georeferenced corners. However, sites 1 and 2, which have a fence as
their common boundary, did not match exactly when georeferenced but were within
approximately 15 meters. Site 5’s markers were not found in 2014 and its exact location was
also in question. It was found to have two possible, equally likely sites. After reviewing the
images, the most likely location for Site 5 was identified using geographical features visible in
the ArcMap World Imagery Basemap and slightly shifted it north-east.
43

3.3.2 Soil unit Data
The soil unit maps, like the one in Figure 13, were drawn by hand (exact date unknown),
to 1: 50,000 scale, prior to the first survey in 1991. The land features used to classify the soil unit
maps are described in Table 7. These soil units can vary slightly over time, especially in the
areas where the Finke River can flood during heavy rainfalls and cause accumulation of sand in
new areas or a wash away of previously existing sand bars. The soil unit data provides further
insight into what the area looked during the LVMP. These maps were also used to identify the
soil units in areas in the park.

Figure 13 Soil Unit map. Source: Low, Foster and Berman, 1991
44

Table 7 Soil units in the immediate vicinity of each site. Source: Low, Foster and Berman, 1991
Abbreviation Soil Unit
A Alluvial plain
B Bench
B+C Bar plain plus channel in montane gully
BP Back plain
CBE Channel bench in floodplain
CBL Calcareous badlands
CH Calcrete hills
CRP Calcareous rolling plain
CUP Calcrete Undulating plain with spinifex
D Dunefield with sparse dunes
EC Creek valley
H Hills
LSH Low sandstone hills with sand veneer
MP Meander plain
P Pediment in EC, surrounding dunefield or montane valley
S Sandhill or ridge
SO Sandstone outcrop
SP Sand plain
T Terrace
U Upper hill slope of eroded plain

3.3.3 Plant Species Data
The plant species present in each field site are depicted in site maps, like the one in
Figure 14 for field survey site 3. These maps were also hand-drawn as the sites were initially
surveyed in 1991. A significant photo record was found, one photo taken from each corner of
each site looking in. Although these photos were not central in digitally identifying specific
plants on the sites, it provided a visual of the area in terms of vegetation. Once georeferenced,
these maps served to identify the vegetation/soil combinations. A complete list of plant species
found during the LVMP can be found in Low, Foster and Berman (1991).

45

Figure 14 Field survey map for Site 3. Source: Low, Foster and Berman, 1991

3.4 Methods
Spatio-temporal analysis of satellite images is a common way to detect changes in the
landscape. Satellite sensors record reflectance values on objects in the landscape (Lu et al.,
2003). Changes in the reflectance indicate changes in the objects themselves, independent of
changes in the atmosphere, illumination and viewing angles, and moisture in the soil (Lu et al.,
2003). Change detection techniques help calculate these changes and assist in the understanding
of the causes or factors in reflectance differences. The most frequently used detection techniques
46

are described in Lu et al. (2003). Some methods, such as image differencing, provide
mathematically simple but significant answers that can be used immediately. Other techniques,
such as spectral mixture models, provide results that must be further analysed and examined to
provide significant answers. Supervised classification and principal components analysis (PCA)
are the main methods used in this thesis.
The main goal is to understand how the vegetation changed in FGNP during the years
after large grazing animals were removed and a fence was placed in some areas to prevent them
from returning. Supervised classification and PCA were used on a set of SFVC composites to
interpret these changes. A supervised classification will help understand the vegetation and soil
relationships and determine which combinations have the least and most variability. PCA is used
as a method to enhance dissociation between the factors (or variables) with the intent to
distinguish sources of variation. Lastly, the results are compared to the LVMP conclusions.
3.4.1 Classification
The first three bands of the seasonal fractional vegetation cover (SFVC) composites
describe bare soil (B1), green vegetation (B2) and dry/dormant/dead vegetation (B3). These
values are continuous, a function of each other and add up to 100. When B2 (green vegetation)
values increase in a pixel, the values of B1 and/or B3 necessarily decrease. Due to this, it seemed
acceptable to do a supervised classification only using green vegetation, with the understanding
that as the values of B2 decrease the vegetation dries, dies or disappears from the area, leading to
an increase in one or both other bands.
Training samples were created using the LVMP georeferenced maps (including the nine
sites surveyed in the LVMP), knowledge of the area, and ArcMap World Imagery Basemap. B2
47

was separated from each of the final 27 SFVC composites. Using only these 27 B2 bands, a
signature file was created and used to do a Maximum Likelihood Classification resulting in a
classified vegetation map. Due to the harsh climatic conditions in the area, most plants are hardy
and resilient, with the ability to flourish in multiple soil conditions, however unlikely. This
means that the same plant species may appear dominant in two, or more different soil types.
The classification of these vegetation type/soil unit combinations allows analysis of the
vegetation changes in the area after the land management efforts at FGNP and provide a baseline
for the principal components analysis to examine possible sources of variation. Calculation for
the mean. median, standard deviation, number of pixels per class, and percentage of class cover
were calculated to further understand the vegetation in the area. Figure 15 show the process
followed for the supervised classification.

Figure 15 Steps used in the supervised classification
▪ SFVC Composites - Download, Clip, Check for completeness
▪ Scan and georeferenced of paper maps
▪ Supervised classification
 Create training samples
 Create sig file for veg and for soil
 Perform MLC for veg and soil
▪ Combine veg and soil classes to create combined class map
▪ Data export for each class to create 16 individual class layers
▪ Separate bands
▪ Test which B2 rasters have pixels with values > 25
▪ Using 16 layers as mask file extract all B2 values >25 (extract by
mask)
▪ Convert raster to point
▪ Create summary statistics
▪ Append to main table
▪ Export to excel
▪ Create charts
48

3.4.2 Principal Components Analysis
Principal Components Analysis (PCA) is generally described as a data-reduction
technique used to detect internal processes and significant events, as opposed to other
hypertemporal analysis techniques that are concerned with larger, more general processes
(Piwowar, 1996). The PCA algorithm works by looking at properties (or variables) of a larger set
of variables (e.g. set of pixels across images over time) and finds the best way to describe them
by reducing them into a smaller set of new variables. The reduction occurs when the properties
are combined to enhance maximum variability between each item in the set. For example,
properties such as temperature and evaporation may not enhance variability in arid zones such as
FGNP because high temperatures typically deliver high evaporation. Thus, explaining the
variability of vegetation growth in terms of temperature and evaporation as variables may yield
similar results. In this case, PCA creates new property permutations, in the form of linear
combinations, to describe the sets, effectively removing this type of redundancy.
New properties combinations are called principal components (PC), also found in the
literature as eigenvalues. PCs show the significance of each relationships. A PCA will result in
as many components as there are variables describing the set. The SFVC composites have four
bands, with band 1 (red), band 2 (green) and band 3 (blue), each representing bare soil, green
vegetation and dry/dead/dormant vegetation respectively. Band 4 represents the root-mean-
squared error between predicted and actual pixel value and was not included in the PCA. A PCA
analysis will result in three PCs for each composite used. This is because each composite is
defined in three variables; red, green, and blue bands. The first principal component, or PC1, will
show the most variance and least error, and if graphed, the linear combination representative of
the new PC1 will best fit the data. The second principal component, or PC2, will show the next
49

most variance, independent (or orthogonal) from PC1, and so on for the rest of the PCs. Each
new component describes one “artificial” property created specifically to describe the set. In a
multitemporal analysis, the new properties or PCs can be presented in a spatial relation (mapping
the values), as well as a temporal relation (charting the PC loadings).
PCA may be calculated manually if the number of variables is low, but in this project the
number of variables is high and the PCA function in ArcMap 10.5.1 was a better option. A
tutorial on PCA by Smith (2002) was used to understand the mathematics of PCA.
Like standard deviation, variance is a measure of the spread of a data set in one
dimension. Their equations, shown in Equation 1 and Equation 2, are very similar and expressed
independently from other dimensions.
𝑆𝑡𝑎𝑛𝑑𝑎𝑟𝑑 𝐷𝑒𝑣𝑖𝑎𝑡𝑖𝑜𝑛 = 𝑆 = √∑
(𝑥 𝑖 − 𝑥 ̅ )
2
(𝑛 − 1)
𝑛 𝑖 =1

𝑉𝑎𝑟𝑖𝑎𝑛𝑐𝑒 = 𝑆 2
= 𝑣𝑎𝑟 (𝑥 ) = ∑
(𝑥 𝑖 − 𝑥 ̅ )
2
(𝑛 − 1)
𝑛 𝑖 =1

Covariance is the spread of a data set from one dimension with respect to other
dimensions. The covariance is always measured between two dimensions. Positive values
indicate that the both dimensions increase together and are positively correlated (green
vegetation and rainfall). Negative values indicate that one dimension increases as the other
decreases and are negatively correlated (green vegetation and heavy grazing). Covariance values
of zero indicate that the dimensions are independent of each other and uncorrelated (heavy
grazing and rainfall). The formula for covariance for n dimensions, in this case two dimensions
(x and y), is shown in Equation 3.
(2)
(1)
50

𝑐𝑜𝑣 (𝑥 , 𝑦 ) = 𝑐𝑜𝑣 (𝑦 , 𝑥 ) = ∑
(𝑥 𝑖 − 𝑥 ̅ ) (𝑦 𝑖 − 𝑦 ̅)
(𝑛 − 1)
𝑛 𝑖 =1

If there are more than two dimensions, more than one covariance will be calculated. For
𝑛 dimensions, there are
𝑛 !
(𝑛 −2)! × 2
different covariance values. For convenience, covariance
matrices can be created to view every possible covariance between every pair of dimensions. The
equation for the covariance matrix with a data set of 𝑛 dimensions is shown in Equation 4.
𝐶 𝑚 ×𝑛 = (𝑐 𝑖 ,𝑗 , 𝑐 𝑖 ,𝑗 = 𝑐𝑜𝑣 (𝐷𝑖𝑚 𝑖 , 𝐷𝑖𝑚 𝑗 ))
The resulting covariances fed the loading for each PC in the charts. The loadings present
relationships through time and help determine the strength between them (Piwowar, 1996). The
PCA process followed is described in Figure 16. This process was automated using the ArcMap
user interface, but it can also be coded with Python. Model Builder was used to automate some
of the processes.

Figure 16 Steps used in the PCA

▪ Separate each band from all 27 SFVC Composites
▪ Subtract 100 from each band to display as percentage
▪ Combine all bands again
▪ Run PCA on all components using ArcMap 10.5.1
▪ Present each spatial association with a graphic and each
temporal association with a chart.

(4)
(3)
51

4 Results

Chapter 4 describes the thesis results after the data was processed and analysed. Section 4.1
presents what was discovered about the vegetation and soil in and around FGNP after doing a
supervised classification. The classification provides a baseline to understand the vegetation and
soil, in terms of variability, representative of the park. Section 4.2 describes the results of a
principal components analysis on the multitemporal set of composites. Selected components are
described and analysed, spatially, temporally, and in terms of trends. In section 4.3, the results
are compared to those in the LVMP.
4.1 Supervised classification
The supervised classification begins with the digitization of all the data available from the
LVMP. The paper maps drawn at the initial stages of the LVMP surveys were scanned and
georeferenced to provide a starting point. The composites have four bands, with Band 2
representative of the vegetation cover percentage for each pixel. During the classification
process, only the values of Band 2 were used. The maps and charts show the green vegetation
cover values (Band 2) over the 10-year period for each vegetation class. As stated in section
3.1.1, the values for all three bands in each pixel are correlated and add to 100. This means that
when green vegetation increases in a pixel from one date to the next, the value of bare soil (Band
1) and/or the value of dry vegetation (Band 3) decreases because they will sum to approximately
100. For the purpose of showing variation in vegetation, only green vegetation was used.
Training samples of plant species and soil types were created using the georeferenced
maps and recent personal conversations (Nano, 2014-2016; Box, 2014-2016; Brim Box, 2014-
52

2016) with experts and ecologists in the area. Using these training samples, two supervised
classifications (using ArcMap 10.5.1) were done with the 27 SFVC composites, one for soil and
one for vegetation. The classes reflect defined boundaries on the map, however they are the
result of a Maximum Likelihood Classification based on the training samples. The soil and
vegetation classes were derived from Low, Foster and Berman (1991). Both rasters were
combined into a joint vegetation/soil classified image to understand the predominant classes.
4.1.1 Soil and Vegetation Classified Maps
The Soil classification map yielded 10 different classes of soil units, shown in Figure 17.
The soil classes in the area represent a variety of land types. A full list of all the soil types used
in the classification can be found in Table 7 (section 3.2.2.2). Meander plains and alluvial plains
are visible, as well as ridges in between eroded hills. The class identified as calcrete undulating
plains with Spinifex is a special class because it is not only a soil class. In this case, Spinifex
grass covers such as significant amount of the area that it is listed as occurring with this type of
soil. Spinifex is the most extensive vegetation type in australia, covering 22% of the continent
(Alice Springs Desert Park, 2017). Despite its short, hummock appearance, Spinifex is a
perennial grass that has deep roots and can survive in the most difficult areas of Australia’s
deserts. There were no samples available for this plant to train the classification, however, its
presence is dominant over the southern area of FGNP.
The vegetation classification map yielded 11 different classes, shown in Figure 18.
Although, there are only 11 vegetation classes, there is field record of dozens more. General
characteristics, soil preferences and root-type are described for the 11 classes in Table 8. Some
of these classes co-exist with several other dominant plant species, such as spinifex and invasive
grasses like couch grass and buffel grass that spread and burn easily. However, the map
53

produced is a good representation of plant coverage, and more specifically plant location among
geographic features.

Figure 17 Soil classification map

54

Figure 18 Vegetation classification map

Table 8 General characteristics, soil preferences and root-type of dominant vegetation.
Class Scientific Name/
Abbreviation
Description
1 Acacia Kempeana
(AcKe)
 Acacia Kempeana is a straggling shrub common in arid
zones.
 Common name: Witchetty Bush.
 Soil: Stony loam or clay loam on hills, along seasonal
watercourses, on decomposing granitic sandy soils.
2 Capparis Mitchellii
(CaMi)
 Capparis Mitchellii is a thorny, slow-growing small tree.
Frost and drought resistant. Requires good drainage.
 Common name: Wild orange plant.
 Soil: Wide range of soil types, including sand, sandy loam,
clay loam and poor soil.
3 Atalaya
Hemiglauca (AtHe)
 Atalaya Hemiglauca is a flowering, small tree. Drought
tolerant.
 Common name: Whitewood tree.
55

 Soil: It occurs on sandy and clay soils, on flood plains, and
sandy ridges.
4 Hakea Leucoptera
(HaLe)
 Hakea Leucoptera is a small shrub or tree that re-sprouts
from base. Occurs in arid and semi-arid Australia, grasslands,
shrublands and woodlands.
 Common name: Silver needlewood.
 Soil: sandy to clay soil
5 Sclerolaena / Open
Shrub (Scle/Os)
 Sclerolaena is a genus of annuals or short-lived perennials
shrubs. Abundant Species in the area of FGNP include
Sclerolaena convexula and Sclerolaena spinose.
 Common name: Bindieyes.
6 Hakea Suberea
(HaSu)
 Hakea Suberea is a small tree with thick, corky bark.
Tolerant of arid conditions and frost.
 Common name: Corkwood nectar tree.
 Soil: Prefers moist, well-drained soils. Occurs in outcrops,
rocky or stony soils, sand plains or flood plains.
7 Typha Domingensis
(TyDo)
 Typha Domingensis is an erect, glasslike, perennial
herbaceous plant. Grows along drains and edges of
watercourses in slow moving water.
 Common name: Bulrush, cumbungi or cattail.
 Soil: Clay or sand substrate. Freshwater swamps, creeks or
rivers.
8 Eucalyptus
Camaldulensis
(EuCa)
 Eucalyptus Camaldulensis is a perennial, single-stemmed,
medium-sized to tall tree that could reach ages of 1000 years.
deep sinker roots making them very effective in conducting
water
 Common name: River red gum.
 Soil: Commonly grows on riverine sites, along river banks, on
floodplains, preferring deep moist subsoils with clay content.
9 Acacia / Senna
(AS)
 Acacia is a genus of shrubs and trees well adapted to hot
climate and droughts, particularly prevalent in arid and semi-
arid areas of Australia. Abundant Species in the area of FGNP
include Acacia tetragonophylla (kurara), Acacia victoriae
(gundabluie), Acacia kempeana (witchetty bush) and Acacia
aneura (mulga).
 Common name: Wattles or acacias.
 Soil: Dry soil prone to bushfires, where they quickly grow
over burned areas.
 Senna is a species of flowering shrub adapted to a wide range
of climatic conditions, although it is susceptible to frost,
particularly when young. Common subspecies are Senna
filifolia and Senna quadrifolia.
 Common name: Silver senna, silver cassia or feathery cassia
 Soil: Dry, well drained sites.
56

10 Acacia
Estrophiolata
(AcEs)
 Acacia Estrophiolata is a tall tree, usually found in areas
with about 220–350 mm/year of average rainfall.
 Common name: Ironwood or southern ironwood.
 Soil: Sandy alluvial flats as scattered trees, but also in tall
open shrubland and open woodland.
11 Acacia Murrayana /
Eremophila
Longifolia
(AcMu/ErLo)
 Acacia Murrayana is an adaptable, fast-growing
large shrub or tree. It is highly fire-tolerant and drought-
adapted.
 Common name: Murray's wattle.
 Soil: Deep red sands but it may also occur on clay loams. It
favours well-drained sites with access to run-on water such as
the base of dunes, road verges and stream levees.
 Eremophila Longifolia is an evergreen, small, rounded shrub
distributed in arid and semi-arid regions suited to dry climates.
It is a very hardy, drought tolerant plant and will tolerate
moderate frost.
 Common name: Berrigan emubush, dogwood, long-leaved
eremophila.
 Soil: Wide range of soil types and habitats. It generally grows
in acacia or eucalyptus woodland but is also common on
rocky hills, sand plains and sand dunes. It needs well-draining
soil, not heavy clay.

4.1.2 Combined vegetation and soil classes
Both rasters in figures X and Y were combined to create a vegetation and soil map,
shown in Figure 19. Each one of the 16 resulting classes was exported to create individual
polygon features, and those used as a mask to extract the raster values from band 2 of the SFVC
composite. B2 values represent the green vegetation, and can range from 0 to 100. Only values
>= 25 were used. This ensured that each pixel used in the analysis had a significant amount (25%
or more) of green vegetation for each class. Table 9 has description of these combined vegetation
and soil classes.
57

Figure 19 Vegetation and soil map

Table 9 Description of vegetation and soil classes
Class Vegetation and Soil Abbreviation
1 Acacia/Senna on
Upper hill slope of
eroded plain
AS on U Shrubs and trees growing in a slope of
eroded plain.
2 Acacia Estrophiolata
on Calcrete Hills
AcEs on CH Tall tree preferring arid conditions in
calcrete hills.
3 Acacia Murrayana /
Eremophila Longifolia
on Calcrete Hills
AcMu/ErLo
on CH
Very hardy and drought-resistant shrubs
in calcrete hills.
4 Atalaya Hemiglauca
on Bar Plain plus
Channel in Montane
Gully
AtHe on BC

Drought-tolerant, small tree growing in
elevated areas of a riverbed or a ditch in
a hill.
58

5 Capparis Mitchellii on
Upper Hill Slope of
Eroded Plain
CaMi on U Slow-growing, small tree growing in a
slope of eroded plain.
6 Eucalyptus
Camaldulensis on
Channel Bench in
Floodplain
EuCa on
CBE
Deep-rooted trees in areas that flood and
can collect and store water even during
dry periods.
7 Eucalyptus
Camaldulensis on
Sandhill or Ridge
EuCa on S Deep-rooted trees in areas that can
collect and store water even during dry
periods.
8 Hakea Leucoptera on
Upper Hill Slope of
Eroded Plain
HaLe on U Small tree that likes sandy, clay soil
growing in a slope of eroded plain.
9 Hakea Suberea on
Alluvial Plain
HaSu on A Small tree with thick, corky bark,
drought and frost tolerant growing in
alluvial, and subject to floods.
10 Hakea Suberea on
Calcareous Badlands
HaSu on
CBL
Small tree with corky bark, drought and
frost tolerant, growing in badlands.
11 Hakea Suberea on
Calcrete Undulating
Plain with Spinifex
HaSu on
CUP
Small tree with corky bark, drought and
frost tolerant, in undulating plains with
spinifex.
12 Sclerolaena/Open
Shrub on Bar Plain
plus Channel in
Montane Gully
Scle/Os on
BC

Shrubs of the family sclerolaena are
annuals or short-lived growing in
elevated areas of a riverbed or a ditch in
a hill.
13 Sclerolaena/Open
Shrub on Meander
Plain
Scle/Os on
MP
Shrubs of the family sclerolaena are
annuals or short-lived in winding plain.
14 Sclerolaena/Open
Shrub on Upper Hill
Slope of Eroded Plain
Scle/Os on
U
Shrubs of the family sclerolaena are
annuals or short-lived growing in a slope
of eroded plain.
15 TyDo on Back Plain Typha
Domingensis
on BP
On the plan between a riverbed and a
hill.
16 TyDo on Channel
Bench in Floodplain
Typha
Domingensis
on CBE
Shallow-rooted, drought tolerant and
typically found in areas that flood and
can collect and store water even during
dry periods.

Soil unit and vegetation cover regions can be recognized in Figure 19. The nine sites of
the LVMP were selected to present distinct vegetation, topography, and soil composition so their
location can be used as a representation of composition.
59

Short-lived tussock shrubs and grasses can be seen in the northwest of the park as well as
some parts in the southeast and northeast (blue and green). These grasses grow in clumps or
bunches and in Figure 19 exist in calcrete and sandstone hills and eroded plains. Sparse witchetty
bush shrubland and five-minute grass are also found in the area. Sites 3 and 4 are in this area.
Annual short-lived grasses and forbs are a dominant species in the centre area of the park
(pink and purple), including bindieyes, a favourite among native birds. Annual grasses can
appear between hummock grasses, especially after rains but can quickly dry. Some of these
softer plants are palatable and prone to being grazed, which is another reason they quickly
disappear after rain. Sites 6 and 9 are located in this area. Site 6 sits on the alluvial plain next to
the Finke river and has a good population of River Red gum trees as well as a significant
population of Mulga shrubs over grasses. Site 9 is on a gentle slope with erosion gullies and its
dominant vegetation appears to be sparse Acacia shrubs with other grasses.
Spinifex grasslands and hummock grasses are found in the south. Hummock grasses
grow in mounds. Spinifex cover is as fundamental to the area as the land features because it is so
prevalent. Sites 7 and 8 sit on wind-eroded dunefields within hummock grasslands. Other plant
species found in these grasslands are mulga, wattle, ironwood and blue mallee.
Sparse witchetty bush areas appear in the north centre of FGNP. Witchetty bushes are
common in arid areas and provide cover for many grass species. Sites 1, 2, and 5 are located in
this area. Sites 1 and 2 are in a gentle slope over sandstone and Site 5 is on a flood plain.
4.1.3 Variability of Classes through Time
Figure 20 shows the numbers of pixels with significant green vegetation cover in each
class for each composite time frame. Out of the original 40 composites, each one covering an
60

annual season over 10 years, only 27 were used because they showed complete, or almost
complete, coverage over the park. The remaining composites were excluded. The temporal gaps
left by the exclusion of incomplete composites can be seen in the chart, with the largest gaps
between May 1992 and March 1993 and between November 1996 and September 1997.
Although there are only 16 classes, it is understood that there are dozens more plant species that
may not have occurred in significant numbers or that may be too similar to others.
Classes 5, 6, 7, 8, and 11 have the largest coverage of green vegetation. These classes
consist of almost 74% of all significant green vegetation pixels. Most of the vegetation in these
classes are frost and drought tolerant or deep-rooted trees. Classes 7 and 8 have the most
consistent cover, only slightly increasing after rains and slightly decreasing during dry periods,
suggesting a looser connection to rainfall than other plants. This type of vegetation, river red
gum trees, have deep roots and can maintain green canopy cover even during droughts. Classes
1-4, 9-10, and 12-16 are classified as short-lived grasses or shrubs with short roots on floodplains
or sloped areas. In these classes, rainfall is a more influencing factor of change. Some of the
pixels may appear misidentified due to atypical plant behaviour after rain or during a drought.
The composites covering the spring seasons of 1990, 1993 and 1998, present the largest
number of pixels with significant green vegetation. There is also a slight increase in the summer
of 1994-1995. These higher values correspond to the increase in rainfall during the months of
May 1990, May 1993, January 1995 and February 1997. The mean, median, standard deviation,
number of pixels with significant (> 25% green vegetation cover), and the percentage of cover
for each class are shown in Table 10.
61

Figure 20 Classified pixels with significant vegetation cover over time
62

Table 10 Data for each significant vegetation cover class
Class
Number of
Pixels
Standard
Deviation Median Mean
Percentage of Cover
per Class
1 51952 5686.833117 95 1924.148148 5.40%
2 45455 5162.661392 76 1683.518519 4.73%
3 36209 4222.6943 44 1341.074074 3.76%
4 55789 5733.583787 139 2066.259259 5.80%
5 168457 10281.01543 1251 6239.148148 17.52%
6 105760 9564.434269 399 3917.037037 11.00%
7 149589 2695.591886 6088 5540.333333 15.55%
8 97302 3595.660048 2332 3603.777778 10.12%
9 27047 3187.339831 38 1001.740741 2.81%
10 18840 2306.906392 26 697.7777778 1.96%
11 168457 10281.01543 1251 6239.148148 17.52%
12 8049 983.9013918 13 298.1111111 0.84%
13 7255 883.1243407 16 268.7037037 0.75%
14 19375 121.5012691 762 717.5925926 2.01%
15 1237 151.9997563 4 45.81481481 0.13%
16 1012 111.3719929 8 37.48148148 0.11%
961785

100.00%

4.2 PCA Results
A PCA performed on the multitemporal set of SFVC composites yielded 81 principal
components (PC). Each of the 27 composites has 3 bands, representing bare soil (Band 1), green
vegetation (Band 2) and dry vegetation (Band 3); 27 x 3 = 81. The first PCs account for most of
the variance, with the first three representing more than 85%. The higher the percentage the more
variation it explains and the lower the percentage, the less variation it explains. Table 11 shows
all 81 PCs and the percentages associated with each.

63

Table 11 Data for each significant vegetation cover class
PC1 76.0964% PC18 0.3222% PC35 0.1329% PC52 0.0373% PC69 0.0019%
PC2 6.2308% PC19 0.2985% PC36 0.1238% PC53 0.031% PC70 0.0019%
PC3 2.7607% PC20 0.2912% PC37 0.1092% PC54 0.0278% PC71 0.0019%
PC4 1.7174% PC21 0.2794% PC38 0.1063% PC55 0.0026% PC72 0.0019%
PC5 1.4412% PC22 0.2716% PC39 0.1% PC56 0.002% PC73 0.0019%
PC6 0.8736% PC23 0.2548% PC40 0.0952% PC57 0.002% PC74 0.0019%
PC7 0.7635% PC24 0.2451% PC41 0.0859% PC58 0.0019% PC75 0.0019%
PC8 0.7171% PC25 0.2376% PC42 0.0791% PC59 0.0019% PC76 0.0019%
PC9 0.6698% PC26 0.2282% PC43 0.0749% PC60 0.0019% PC77 0.0019%
PC10 0.5481% PC27 0.2188% PC44 0.07% PC61 0.0019% PC78 0.0019%
PC11 0.5049% PC28 0.1951% PC45 0.0666% PC62 0.0019% PC79 0.0019%
PC12 0.4651% PC29 0.1885% PC46 0.0628% PC63 0.0019% PC80 0.0018%
PC13 0.4315% PC30 0.1747% PC47 0.0614% PC64 0.0019% PC81 0.0016%
PC14 0.4032% PC31 0.1662% PC48 0.054% PC65 0.0019%
PC15 0.3831% PC32 0.1505% PC49 0.0456% PC66 0.0019%
PC16 0.3472% PC33 0.1469% PC50 0.0435% PC67 0.0019%
PC17 0.3387% PC34 0.1368% PC51 0.0421% PC68 0.0019%

A visual inspection of the components was performed to identify significant PCs. Most of
the higher PCs are included and most of the PCs with lower variation are excluded. Although
this decision seems arbitrary, the inclusion/exclusion of PCs is based on thorough visual analysis
of the components significance. This seemingly subjective process requires knowledge of the
area and an understanding of the processes, soils and vegetation composition.
Each component shows processes happening to the entire image area. They are presented
spatially and temporally using map images and component loading charts. The component map
images present spatial relationships graphically for each PC and are shown in the map images for
64

Figures 21 – 24. The red areas (more negative values) and blue areas (more positive values)
represent areas of strong association. The beige areas (values closer to zero) represent weaker
relationships and less significant variation.
The PC loadings graphs, shown in the charts for Figures 21 - 24, represent the temporal
relationship. The Y-axis plots the covariance between each input band and each component. The
three lines denote bare soil (red), green vegetation (green), and dry vegetation (blue). The X-axis
represents the time in three-month periods corresponding to each SFVC composite. The positive
values indicate that they increase together or decrease together. The negative values indicate that
the dimensions are inversely correlated. Values of zero indicate variables are independent.
The first four principal components were selected for review an analysis. The remaining
components did not seem to have had visible patterns, however they may explain other smaller
sources of variation.
65

4.2.1 Principal Component 1 (PC1) – 76.1%
Principal Component 1 (PC1) in Figure 21 represents most of the variation through time
at 76.1% and shows broad-scale processes in the area. The high value of PC1 can be explained
by the regional processes that often dominate ecological dynamics. The darker red areas
highlight river beds, meander and sloped plains, and sandy plains, typical areas that could collect
and hold water. The darker blue areas represent hilltops, rocky outcrops, and other areas where
rainfall run off is common. As the colours lighten and get closer to yellow, the areas vary the
least, with some obvious depressions and outcrops seen in the images.
PC1 is indicative of processes that occur over the area simultaneously, such as a drought,
a freeze, or any other environmental affects. The loadings for PC1 show consistent values, a
characteristic of typical conditions. These typical differences are very representative of the
geography and vegetation and map out different ecological zones in the area. PC1 explains how
they share the same rates of variation; how they rise and fall through time.

Figure 21 Image and Loadings for Principal Component 1. The figures represent spatial variation
of PC1 at 76.1% over FGNP, and surrounding areas.
66

4.2.2 Principal Component 2 (PC2) – 6.2%
Principal Component 2 (PC2) in Figure 22 represents the second most variation through
time at 6.2%, independent of variation caused by broad-scale proccesses accounted for by PC1.
The dark blue areas support persistently green, photosynthesizing vegetation over time, such as
riverbeds or channel benches. Deep-rooted trees in these areas can access underground water and
maintain leaf canopies even during periods of drought. Most of the dark red areas are classified
as rocky hills, calcareous rises or sloped plains with perennial, short-lived grasses and sparse
woody vegetation. These plants with shallow root systems have limited or no access to
underground water. Their growth and reproduction are closely linked to sporadic rainfall events.
PC2 shows the difference between persistently green vegetation and rainfall-dependent
vegetation. The loadings present higher values of dry cover and bare ground when compared to
green coverage. There is a large gap in the input composites between the winter of 1992 and the
winter of 1993, where a shift in the composites can be seen between dry vegetation cover and
bare soil.

Figure 22 Image and Loadings for Principal Component 2. The figures represent spatial variation
of PC2 at 6.2% over FGNP, and surrounding areas.
67

4.2.3 Principal Component 3 (PC3) – 2.8%
Principal Component 3 (PC3) in Figure 23 represents 2.8% of the variation in the area
and shows some fine-scale processes. There is a visible change corresponding to a fenceline on
the west side of the park (Palm Paddock) bisecting a red area. The dark red colour above the
fence correspond to an increase in vegetation. The lighter red colour below the fenceline
corresponds to a slower increase in the vegetation, indicating a difference between growth inside
and outside the fence. The same colour pattern is seen towards the center of the east boundary,
where the sandy portions near the river with a sigificant population of deep-rooted trees show
dark red and the surrounding meandering plain with small trees and shrubs show a lighter shade
of red.
PC3 components represents localized changes over time. In the loadings, Band 2 showing
green cover increases just after the first significant high rainfall periods, in winters of 1990
and1993, and summer 1994, following land management efforts. Band 1 bare soil cover loadings
show a decline as band 3 dry vegetation cover shows a steady increase.

Figure 23 Image and Loadings for Principal Component 3. The figures represent spatial variation
of PC3 at 2.8% over FGNP, and surrounding areas.
68

4.2.4 Principal Component 4 (PC4) – 1.7%
Principal Component 4 (PC4) in Figure 24 represents 1.7% of the variation in the area
and shows other finer-scale processes. The dominant spatial pattern on the map, shows intense
red colours in the area where the Finke River splits into Palm Creek to the the west and Ellery
Creek towards the north. This area is mostly surrounded by blue zones made up of sandy hills
and calcareous rolling plains. Many of these blue zones may reflect abrupt changes in the area’s
vegetation cover. These abrupt changes can represent fires burning quick through dry foliage or
floods clearing the river vegetation as it floods. Another area of dark red colour appears on
calcrete undullating hills at Palm Paddock, predominantly covered by hummock grasses like
spinifex, with some blue areas representing the riverbed and floodplain.
The loadings for the PC4 explain some of the spatial pattern where localized changes
occurred. The rains in 1990 and 1993 are represented with high values for the green vegetation
cover and low value for the dry vegetation cover. The rains of 1995 and 1997 were not visible,
however the rainfall during those years was consistently low.

Figure 24 Image and Loadings for Principal Component 4. The figures represent spatial variation
of PC4 at 1.7% over FGNP, and surrounding areas.
69

4.2.5 Area of Interest – Palm Paddock
During a close analysis of all the components, the area of Palm Paddock seemed to
provide interesting patterns and were used it as a case study to further understand the changes in
vegetation at a more appropriate scale. Palm Paddock is located at the far northwest section of
FGNP and was an area fenced during the land management efforts in the late 1980’s and early
1990’s. Principal components show the strength of the spatial association with the really high
and really low values represented as dark blue and dark red.
PC1 (Figure 25a) depicts broad scale processes under average conditions in the area. Arid
zones like FGNP are strongly dependent on environmental effects like rain, temperature, fire,
etc. Some features of the landscape are defined spatially and visible but this image and its
loadings provide a picture of that the average conditions look like. A PCA with the spatial scale
of only Palm Paddock would yield different results and the spatial relationship strength would be
more obvious.
A significant portion of the blue areas in PC2 (Figure 25b) are in a dry river beds, which
typically collect moisture and favour persistent ground water collection long after rainfall. One
dominant plant species present in the dry riverbeds in the Finke River and its tributaries is River
red gum tree. These and other deep-rooted trees can access water underground, maintaining the
green leaf canopies, even during dry periods. A second type of dominant vegetation in these
areas is cattail. Cattail is shallow-rooted, fairly tolerant to drought and typically found near water
sources. Their growth and reproduction are closely synched to sporadic rainfall events, resulting
in quick green vegetation bursts following even a small amount of rain. Both types of vegetation
respond differently to rainfall, but dominate the same area. The red areas in PC2 are classified as
hilly and sloped areas with perennial, short-lived grasses and sparse woody vegetation. These
70

shallow-rooted grasses and forbes (Acacia, Senna, Ironwood, Corkwood) have restricted access
to underground water sources and quickly respond to drought or any amount of rainfall.
PC3 (Figure 25c) represents an increase in woody vegetation in areas where horses were
removed to cease grazing and a fence was emplaced to prevent more large herbivores to enter the
area; most of the work was completed by 1989. This type of land management was drastic and
encouraged quick recovery of the vegetation. Other environmental threats remained, however,
the effects of the horse removal can be seen in PC3 as the fence can be outlined in the image.
The localized variation in PC4 (Figure 25d) highlights some areas in the south part of
Palm Paddock that may have underwent fires and areas where river currents can clear the
vegetation with its force after a flood. Controlled fires are set to prevent wild fires from
spreading out of control. Additionally, wild fires are difficult to document in isolated areas
because they can start and die down without notice. Some of the remaining components could
represent significant features of the landscape or meaningful spatial patterns but, their
increasingly insignificant values make their importance difficult to assess.

(a)
71

(b)

(c)

(d)
Figure 25a-d Images for Principal Components 1-4 for the area of Palm Paddock
72

5 Discussion and Conclusions

This chapter discusses classification and PCA results in the context of previous work
conducted as well as how the methods were applied to FGNP. The research questions are also
addressed and conclusions are presented, based on analysis results. Limitations encountered
during the thesis process are considered and possible future research options area identified.
The two main methods used to conduct the thesis were supervised classification and PCA.
The supervised classification used LVMP field data collected in the 1990’s at FGNP. This was
mainly due to not having the capability to collect training samples. It also made sense to use data
matching temporal range to the SFVC composites used in the PCA. The classification helped in
creating a baseline to understand the vegetation and soil of the park.
Many pixels in the 16 resulting classes representing combinations of soil forms and
vegetation were identifying plants that were similar or behaved similarly as one class.
Environmental effects could cause some of the pixels to be misrepresented in a supervised
classification, particularly rainfall. Rainfall in this arid area is unpredictable and vegetation may
behave atypically. Some invasive or drought-resisting plants can invade an area occupied by a
dry and dormant species. Non-native invasive species like couch grass and buffel grass are
notorious for spreading quickly in central Australia and thus changing the plant composition.
Upon closer inspection, a general pattern of plants with similar characteristics was
seemingly present. The patterns were tightly connected with the land forms in the areas but
groups of short-lived tussock shrubs and grasses in the northwest, annual short-lived grasses and
forbs in the centre, spinifex grasslands and other hummock grasses in the south, and sparse
73

witchetty bushland in the north centre were apparent. The general characteristics for each
vegetation class represent a large number of other plants with a similar footprint. In addition, it
was common to see similar plants behaving differently as they grow in different soil forms (i.e.
acacia short bush growing on the riverbed as well as on the eroded side of a hill). Heavy and
consistent rains during the year 1993, and the summers of 1994 and 1996 are reflected as an
increase of green vegetation cover for most classes. The supervised classification results
provided a greater understanding of the vegetation changes in the area based on rainfall,
vegetation characteristics and land forms. These findings were essential to the understanding of
the resulting PCs of the PCA.
The first four PCs were selected for further analysis. PC1 accounts for most of the
variation at 76% and represents broad-scale processes, typical of the area. Areas that can
typically collect water, either due to soil composition or land form, have reverse values to areas
where water typically runs off, such as hilly area or areas where the soil is sandy and water can
quickly seep through. Common conditions at FGNP are a persistent drought with periodic
rainfall that may be significant or in short bursts. Rainfall occurs mostly during the summer
months but also during the other seasons and can take years to fall at a rate that can significantly
affect the area. Studies by Piwowar (1996) and Henderson (2010) had similar results. Piwowar
(1996) found that PC1 showed “a generalized perspective of the northern ocean’s ice cover.” His
study of the area is pertaining to ice and lacks the vegetation component, however, much of the
literature on PCA agrees that PC1 will represent the most variation and therefore, when looking
at a large enough temporal range will show average, or the most common conditions. In his
study, Piwowar (1996) compared the differences of areas with constant ice cover and areas with
no ice cover. In her study, Henderson (2010) found that PC1 reflects conditions that would
74

typically occur in the area, with dry areas showing opposite values to wet areas, as well as
protected areas showing opposite values to cultivated areas.
PC2 accounts for 6.2% of the variation and presents a more defined pattern in vegetation.
It shows the difference between persistently green vegetation and rainfall-dependent vegetation.
Areas that are persistently green, like deep-rooted trees that can access underground water tables,
have opposite values of areas that are not persistently green, such as annual grasses or shallow-
rooted plants that have no access to underground water and can turn green or dry quickly. It is
understandable that PC2 corresponds to vegetation attributes because most plants in central
Australia have adapted to live in extremely harsh conditions. In Henderson’s study of Grasslands
National Park in Saskatchewan, Canada, the second most important source of variation seems to
be tightly related to temperature. She found more localized changes showing strong spatial
patterns in years that had either very high or very low temperature readings (Henderson 2010).
This shows that it is possible to have very different results when PCA is applied to two similar
areas. Grasslands National Park and FGNP are both protected arid and semi-arid zones with
drastic temperature changes between seasons.
PC3 shows 2.8% of the vegetation and shows finer-scale, localized processes. It is
apparent that this component highlights differences between areas inside and outside the fence
placed around the Palm Paddock area in the northwest portion of FGNP. The difference in the
colour of the areas above and below the fence indicate a difference in growth between vegetation
inside and vegetation outside the fence. The relatively small Palm Paddock area is affected
similarly by rainfall, temperature and other environmental factors. Human-introduced effects,
however, have allowed the vegetation inside the fence to recover while the area outside this
boundary remain available to grazing and trampling by horses and other large herbivores. The
75

lighter red colour below the fenceline suggests a slower increase in the vegetation, corresponding
to the land management efforts.
PC4 shows 1.7% of the vegetation and shows other fine-scale processes. The dominant
spatial pattern on the map shows blue colours over dry areas, specially dry riverbeds and creeks.
The spatial pattern can be further scrutinized by looking at the Palm Paddock area alone where
blue zones could represent areas of rapid vegetation changes, such as the burning of dry plants or
clearing of vegetation by flooding river waters.
As expected, rainfall was a pivotal factor that drove much of the variation occurred at
FGNP. High rainfall years directly influenced some of the vegetation growth in the area, but did
not explain all. PCA showed that some of the variation was independent from rain. This is
particularly visible in the PC3 with differences inside and outside of the park fence. The
supervised classification and PCA in this thesis support the findings of Miller, Low and
Matthews (1998) that there was a general vegetation recovery in areas of FGNP that underwent
land management methods.
Miller, Low and Matthews (1998) also concluded that a number of invasive species
benefited from these changes and increased in the area. This conclusion was not apparent in the
PCA, and may require a PCA of a smaller spatial scale and/or large temporal scale, or a further
review of the smaller components over the entire park after field work determines the extent of
cover of these plant species.
The use of classification using existing non-digital maps was necessary for this thesis.
General knowledge of vegetation cover and land forms in the area was necessary but no recent
field samples were available. The data collected during the LVMP and documented in Low,
Foster, and Berman, (1991); Low et al., (1992); Low et al. (1993); Cook et al., (1995); Miller,
76

Low, and Matthews, (1998) filled a portion of this knowledge gap, however, it was a limiting
factor which introduced errors in pixel categorization. Uncharacteristic plant response to rainfall
and unpredictable rainfall patterns also accounted for some of the pixel misclassification. Future
research in this topic or this area will improve with the use of more current information about
vegetation type, vegetation cover, soil composition, erosion and grazing animals in the area. This
will allow for a more reliable quantification of change and therefore a better classification.
The PCA rested on the use of seasonal fractional vegetation cover composites that
represented typical annual seasons. These composites were derived from geometrically and
radiometrically corrected Landsat images. Composites minimize data loss from individual
images, however, when looking at large areas these same composites can introduce gaps as well.
Only 27 composites out of the possible 40 were used due to completeness, leaving large gaps in
the temporal scale. Landsat 5 was the only source of satellite images available during that time.
These issues limited trend analysis because some annual season composites were not included in
the PCA. Future research using PCA in this area should include a more complete set of regular
time-series images or composites. Multiple sources of data are now available and the possibility
of using one or more of these sources will minimize the gaps in the coverage and maximize the
detection of trends.
Future work should also include a study of smaller, paddock-sized areas. When doing a
PCA at a smaller scale the variations are more apparent, and so are trends. Preliminary work
conducted over only the area of Palm Paddock shows stronger trends, not visible when doing a
PCA over the entire park. The methodology created to process the data in this thesis should be
revised to ensure that it is most effective in determining how sources of variation have affected
the vegetation.
77

This thesis presents a novel use of PCA as a tool to understand and explore changes in
managed areas in remote central Australia. The method described in this thesis can be used as an
exploratory way to identify areas that may not be obviously threatened, but due to its particular
signature should be evaluated. This kind of research is significant because it uses an established
and accepted, relatively simple and cost-effective method to understand changes occurred in
remote areas. Land managers can use this kind of analysis as a viable alternative to initially
assess the value of their efforts.
This thesis supports the claim that the vegetation in managed areas at FGNP has seen
some recovery, as the field survey has stated. The conclusion was reached through the analysis
of the results of a supervised classification and PCA whereas previous work concluded the same
with the collection of field data over a period of eight years. This thesis did not have a fine
enough temporal scale to identify trends in individual plants species as the LVMP was able to
capture. However, at the larger spatial scale used, this research showed how some dissimilar land
forms in the areas presented similar trends in variation and how some comparable land forms
presented variation in different ways. With currently available, finer images, the PCA would
have been more efficient at recognizing some trends, as the ground sampling interval is smaller.
Products from any modern suite of satellite systems can provide better resolution than the
Landsat 5. Another option would be the use of unmanned aerial vehicles or drones to regulate
and improve the spectral and temporal resolution. This technology allows full control of image
collection, far outweighing its marginal cost. These differences occur due to either natural
effects, such as vegetation species or soil composition, or due to human or animal action. The
natural effects were associated with the classification of plants and soil in the area. The human
and animal effects were due to land management practices and heavy grazing and soil
78

compacting. PCA was used as a tool to help separate and understand the variation. Of interest in
the PCA results, was the area of Palm Paddock, where most land management efforts were
focused at. This method provides an alternative for land managers to study other remote parts of
central Australia, where field surveys may not be cost-effective, and to more appropriately
allocate efforts to protect vulnerable areas.

79

6 References

Abuelgasim, A., W. D. Ross, S. Gopal, and C. Woodcock. "Change Detection Using Adaptive
Fuzzy Neural Networks: Environmental Damage Assessment after the Gulf War."
Remote Sensing of Environment 70, no. 2 (May): 208-223.

Alavi, N. 2012. “Land Use and Land Cover Change Detection in Isfahan, Iran Using Remote
Sensing Techniques.” Master’s thesis, University of Ottawa. Accessed June 10, 2016.
ProQuest Dissertations and Theses.

Alice Springs Desert Park. 2017. “Nature Notes: Spinifex.” Accessed March 8.
http://www.alicespringsdesertpark.com.au/kids/nature/plants/spinifex.shtml

Al-Jawarneh, R. 2004. “Spatial Analysis of Land Cover and Land Use in Evaluating Land
Degradation in Northwestern Al-Mafraq City, Jordan.” Master’s thesis, Yarmouk
University. Accessed September 11, 2016. ProQuest Dissertations and Theses.

Amiraslani, F., and D. Dragovich. 2013."Image Acquisition for Detection of Vegetation Change
Based on Long-term Rainfall in an Arid Rangeland in Western NSW, Australia."
Environmental Earth Sciences 70, no. 1 (November): 83-95.

Australian Bureau of Statistics. 2016. “Population.” ABS Statistics. Last modified December 12.
Accessed February 15, 2017. http://www.abs.gov.au/Population.
Bastin, G, P. F. Scarth, V. Chewings, A. Sparrow, R. J. Denham, M. D. Schmidt, P. J.
O’Reagain, R. Shepherd, and B. N. Abbott. 2012. "Separating Grazing and Rainfall
Effects at Regional Scale Using Remote Sensing Imagery: A Dynamic Reference-cover
Method." Remote Sensing of Environment 121 (March): 443-457.
Bengraine, K., and T. F. Marhaba. 2003. "Using Principal Component Analysis to Monitor
Spatial and Temporal Changes in Water Quality." Journal of Hazardous Materials 100,
no. 1 (March): 179-195.
Berman, D. 1990. “Monitoring Vegetation and Rabbits at FGNP by Rangers After Removal of
feral Horses. File A/89/200.” Minute Paper, Alice Springs, December 19.

Bovolo, F., S. Marchesi, and L. Bruzzone. 2012. "A Framework for Automatic and Unsupervised
Detection of Multiple Changes in Multitemporal Images." IEEE Transactions on
Geoscience and Remote Sensing 50, no. 6 (June): 2196-2212.

Box, J. B., A. W. Duguid, R. E. Read, R. G. Kimber, A. Knapton, J. A. Davis, and A. E.
Bowland. 2008. "Central Australian Waterbodies: The Importance of Permanence in a
Desert Landscape." Journal of Arid Environments 72, no. 8 (August): 1395-1413.

80

Box, P. 2014-2016. Interviews by Adlin Botkin. Alice Springs.

Brim Box, J. 2014-2016. Interviews by Adlin Botkin. Alice Springs.

Bradshaw, C. J. A. 2012. "Little Left to Lose: Deforestation and Forest Degradation in Australia
since European Colonization." Journal of Plant Ecology 5, no. 1 (March): 109-20.

Bureau of Meteorology. 2016. “Climate Data Online.” Accessed February 10.
http://www.bom.gov.au/climate/data.

Burns, E., D. Lindenmayer, A. Lowe, and N. Thurgate. 2014. Biodiversity and Environmental
Change: Monitoring, Challenges and Direction. Victoria: CSIRO Publishing.

Byrne, G., P. F. Crapper, and K. Mayo. 1980. "Monitoring Land-cover Change by Principal
Component Analysis of Multitemporal Landsat Data." Remote Sensing of Environment
10 (November): 175-184.

Celik, T. 2009a. "Multiscale Change Detection in Multitemporal Satellite Images." IEEE
Geoscience and Remote Sensing Letters 6, no. 4 (October): 820-824.

------. 2009b. "Unsupervised Change Detection in Satellite Images Using Principal Component
Analysis and K-Means Clustering." IEEE Geoscience and Remote Sensing Letters 6, no.
4 (October): 772-776.

Collins, J. B., and C. E. Woodcock. 1996. "An Assessment of Several Linear Change Detection
Techniques for Mapping Forest Mortality Using Multitemporal Landsat TM Data."
Remote Sensing of Environment 56, no. 1 (April): 66-77.

Cook, J., A. Grattidge, W. A. Low, D. Schunke, and D. Matthews. 1995. The Environmental
Implications of Removal of Horses from Finke Gorge National Park. Stage 4: Fourth
Vegetation Monitoring and Progress Assessment. Alice Springs, NT: Arid Zone Research
Institute Library.

Csurhes, S., G. Paroz, and A. Markula. 2009. “Pest Animal Risk Assessment: Feral Horse Equus
Caballus.” Queensland Primary Industries and Fisheries, Queensland.

Davis, J. C. 2002. Statistics and Data Analysis in Geology. 3rd ed. New York: Chichester:
Wiley.

Day, C. 1989. “Feral Animal Muster and Cull – FGNP. File A/89/186.” Minute Paper, Alice
Springs, May.

Deng, J. S., K. Wang, Y. H. Deng and G. J. Qi. 2008. “PCA-Based Land-Use Change Detection
and Analysis Using Multitemporal and Multisensor Satellite Data.” International Journal
of Remote Sensing 29, no. 16 (July): 4823-4838.

81

Department of the Environment and Energy. 2016. “Plan for a Cleaner Environment.” Accessed
December 16. http://www.environment.gov.au/cleaner-environment/plan-2016.

Department of the Environment and Energy. 2017. “Australia’s Bioregions (IBRA).” Accessed
January 20. http://www.environment.gov.au/land/nrs/science/ibra.
Farhangfar, S., J. Kambouzia, R. Deihimfard, S. Soufizadeh, and B. Mirbagheri. 2016.
"Introduction of an Index for Drought Evaluation Using Principle Components Analysis."
Agricultural Ecology 5, no. 4 (2016): 452-461.
Fischer, W. A., W. R. Hemphill, and A. Kover. 1976. "Progress in Remote Sensing /1972-
1976/." Photogrammetria 32 (August): 33-72.
Flood, N. 2013. "Seasonal Composite Landsat TM/ETM plus Images Using the Medoid (a
Multi-Dimensional Median)." Remote Sensing 5, no. 12 (December): 6481-6500.
Flood, N., T. Danaher, T. Gill, and S. Gillingham. 2013. "An Operational Scheme for Deriving
Standardised Surface Reflectance from Landsat TM/ETM plus and SPOT HRG Imagery
for Eastern Australia." Remote Sensing 5, no. 1 (January): 83-109.
Foran, B. D. 1986. "The Impact of Rabbits and Cattle on an Arid Calcareous Shrubby Grassland
in Central Australia." Vegetatio 66, no. 1 (April): 49-59.
Gee, R.W. 1986. “Bovine Tuberculosis Eradication in Australia.” World Organization for
Animal Health, 1985. Accessed November 22, 2016.
http://www.oie.int/doc/ged/d8570.pdf.
Geoscience Australia. 2016. “National Local Information.” Australian Government Geoscience
Australia. Accessed November 15, 2016. http://www.ga.gov.au/scientific-topics/national-
location-information.
Graetz, R. D., D. M. Carneggie, R. Hacker, C. Lendon, and D. G. Wilcox. 1976. “A Quantitative
Evaluation of Landsat Imagery of Australian Rangelands.” The Rangeland Journal 1, no.
1 (1976): 53-59.

Graham, A., and K. Johnson. 1986. “An Aerial Survey of Horses and Other Large Animals in the
Alice Springs and Gulf Regions.” Alice Springs, NT: Arid Zone Research Institute
Library.

Griffin, G. F., and M. H. Friedel. 1984. “Effects of Fire on Central Australian Rangelands. II
Changes in Tree and Shrub Populations.” Australian Journal of Ecology 9, no. 4
(December): 395-403.

Henderson, J. A. 2010. “A Study of the Response of the Northern Mixed-grass Prairie in
Grasslands National Park, Saskatchewan to a Changing Climate.” Master’s thesis,
University of Regina. Accessed January 8, 2016. ProQuest Dissertations and Theses.
82

Howarth, P. J., and G. M. Wickware. 2007. "Procedures for Change Detection Using Landsat
Digital Data." International Journal of Remote Sensing 2, no. 3 (April): 277-291.

Hussain, M., D. Chen, A. Cheng, H. Wei, and D. Stanley. 2013. "Change Detection from
Remotely Sensed Images: From Pixel-based to Object-based Approaches." ISPRS
Journal of Photogrammetry and Remote Sensing 80 (June): 91-106.

Ingebritsen, S. E., and R. J. P. Lyon. 1985. "Principal Components Analysis of Multitemporal
Image Pairs." International Journal of Remote Sensing 6, no. 5 (1985): 687-696.
Kelly, P. M., P. D. Jones, C. B. Sear, B. S. G. Cherry, and R. K. Tavakol. 1982. "Variations in
Surface Air Temperatures: Part 2, Arctic Regions, 1881-1980." Monthly Weather Review
110, no. 2 (February): 71-83.
Kluger, J. 2013. “Earthrise on Christmas Eve: The Picture That Changed the World.” Time,
December 24. Accessed November 15, 2016.
http://science.time.com/2013/12/24/earthrise-on-christmas-eve-the-picture-that-changed-
the-world/photo/earthrise1/.
Landsberg, J., and G. Crowley. 2004. "Monitoring Rangeland Biodiversity: Plants as Indicators."
Austral Ecology 29, no. 1 (February): 59-77.
Lau, J. E., and G. Jacobson. 1991 “Aquifer characteristics and groundwater resources of the
Amadeus Basin.” Bureau of Mineral Resources, Australia, Bulletin 236 (1991): 563-580.
Lawley, E. F., M. M. Lewis, and B. Ostendorf. 2016. "A Remote Sensing Spatio-temporal
Framework for Interpreting Sparse Indicators in Highly Variable Arid Landscapes."
Ecological Indicators 60 (January): 1284-1297.
Low, W. A. and M. R. Hewett. 1990. “Aerial Survey of Horses and Other Large Animals in the
Central Ranges of the Alice Springs District, October 1988.” Consultancy Report to the
Conservation Commission of the Northern Territory.
Low, W. A., E. Foster, and D. McK. Berman. 1991. The Environmental Implications of Removal
of Horses from Finke Gorge National Park. Stage 1: Site Selection, Ranger Training and
First Vegetation Monitoring. Alice Springs, NT: Arid Zone Research Institute Library.
Low, W. A., N. de Preu, D. Schunke, D. Matthews, and D. McK. Berman. 1992. The
Environmental Implications of Removal of Horses from Finke Gorge National Park.
Stage 2: Second Vegetation Monitoring and Continued Ranger Training. Alice Springs,
NT: Arid Zone Research Institute Library.
Low, W. A., R. Dunlop, M. Kanter, D. Matthews, and D. McK. Berman. 1993. The
Environmental Implications of Removal of Horses from Finke Gorge National Park.
83

Stage 3: Third Vegetation Monitoring and Progress Assessment. Alice Springs, NT: Arid
Zone Research Institute Library.
Low, W. A. Bill. 2016. Interview by Adlin Botkin. Alice Springs. March.
Lu, D., P. Mausel, E. Brondízio, and E. Moran. 2003. “Change Detection Techniques.”
International Journal of Remote sensing. 25, no. 12 (April): 2365-2401.
Matheson, W. 1994. "The Use of Landsat MSS Imagery to Determine Plant Characteristics and
Community Spectral Separability from Sub‐tropical to Arid Zones in Northern
Australia." Geocarto International 9, no. 2 (June): 5-13.

Morton, S., D. M. Stafford Smith, C. R. Dickman, D. L. Dunkerley, M. H. Friedel, R. R. J.
Mcallister, J. R.W. Reid, D. A. Roshier, M. A. Smith, F. J. Walsh, G. M. Wardle, I. W.
Watson, and M. Westoby. 2011. "A Fresh Framework for the Ecology of Arid Australia."
Journal of Arid Environments 75, no. 4 (April): 313-329.

Miller, C., W. A. Low, and D. Matthews. 1998. The Environmental Implications of Removal of
Horses from Finke Gorge National Park. Stage 5: Fifth Vegetation Monitoring and
Progress Assessment. Alice Springs, NT: Arid Zone Research Institute Library.

Muir, J., M. Schmidt, D. Tindall, R. Trevithick, P. Scarth, and J. Stewart. 2011. “Guidelines for
Field Measurement of Fractional Ground Cover: A Technical Handbook Supporting the
Australian Collaborative Land Use and Management Program.” Technical Report,
Queensland Department of Environment and Resource Management for the Australian
Bureau of Agricultural and Resource Economics and Sciences, Canberra.

Nano, C. E. M., A. E. Bowland, T. J. Nano, S. Raghu, and C. R. Pavey. 2012. "Demographic
Hurdles to Persistence in Acacia Peuce (F. Muell.): Effects of Resources, Fire and
Browsing on a Threatened Keystone Tree Species from Arid Australia." Journal of Arid
Environments 80 (May): 17-26.

Nano, C. E. M., and C. R. Pavey. 2013. "Refining the 'pulse-reserve' Model for Arid Central
Australia: Seasonal Rainfall, Soil Moisture and Plant Productivity in Sand Ridge and
Stony Plain Habitats of the Simpson Desert." Austral Ecology 38, no. 7 (February): 741-
753.

Nano, C. 2014-2016. Interviews by Adlin Botkin. Alice Springs.

NASA. 2017. “Landsat.” Last modified August 3. Accessed November 15, 2016.
https://www.nasa.gov/mission_pages/landsat/main/index.html.

Parks and Wildlife Commission. 2011. “Finke Gorge National Park Joint Management Plan.”
Last modified October. Accessed May 15, 2016.
https://dtc.nt.gov.au/__data/assets/pdf_file/0010/249148/finke-gorge-national-park.pdf.
84

Piwowar, J. M. 1996. “Hypertemporal Image Analysis of Arctic Sea Ice Concentrations as an
Index of Climate Change and Variability.” PhD thesis, University of Waterloo. Accessed
January 8, 2016. ProQuest Dissertations and Theses.

Piwowar, J. M., and E. F. LeDrew. 1995. "Hypertemporal Analysis of Remotely Sensed Sea-ice
Data for Climate Change Studies." Progress in Physical Geography 19, no. 2 (June):
216-242.

Smith, L. I. 2002. “A Tutorial on Principal Components Analysis.” Last Modified February 26.
Accessed January 25, 2016.
https://klevas.mif.vu.lt/~tomukas/Knygos/principal_components.pdf.

Smyth, A. K., and C. D. James. 2004. "Characteristics of Australia's Rangelands and Key Design
Issues for Monitoring Biodiversity." Austral Ecology 29, no. 1 (2004): 3-15.

Sparrow, A. D., M. H. Friedel, and D. M. Stafford Smith. 1997. "A Landscape-scale Model of
Shrub and Herbage Dynamics in Central Australia, Validated by Satellite Data."
Ecological Modelling 97, no. 3 (April): 197-216.

Story, R., G. A. Yapp, and A. T. Dunn. 1976. "Landsat Patterns Considered in Relation to
Australian Resources Surveys." Remote Sensing of Environment 4, no. C (1976): 281-
303.

Strahler, A. H., C. E. Woodcock, and J. A. Smith. 1986. "On the Nature of Models in Remote
Sensing." Remote Sensing of Environment 20, no. 2 (October): Sun, Danfeng, Xiaoqian
Yu, Xiaona Liu, and Baoguo Li. 2016. "A New Artificial Oasis Landscape Dynamics in
Semi-arid Hongsipu Region with Decadal Agricultural Irrigation Development in Ning
Xia, China." Earth Science Informatics 9, no. 1 (March): 21-33.

Terrestrial Ecosystem Research Network. 2013. “Seasonal fractional cover - Landsat, JRSRP
algorithm, Australia coverage.” Last Modified April 2017. Accessed April 15, 2016.
http://data.auscover.org.au/xwiki/bin/view/Product+pages/Landsat+Seasonal+Fractional+
Cover.

Thackway, R., and I. D. Cresswell, eds. 1995. An Interim Biogeographic Regionalisation for
Australia: a framework for establishing the national system of reserves, Version 4.0.
Australian Nature Conservation Agency, Canberra.

Townshend, J. R.G., T. E. Goff, and C. J. Tucker. 1985. "Multitemporal Dimensionality of
Images of Normalized Difference Vegetation Index at Continental Scales." IEEE
Transactions on Geoscience and Remote Sensing Ge-23, no. 6 (November): 888-895.

Tueller, P. T. "Remote Sensing Science Applications in Arid Environments.” 1987. Remote
Sensing of Environment 23, no. 2 (November): 143-154.

85

United States Geological Survey. 2017. “Landsat 5 History.” Last Modified June 22. Accessed
September 20, 2016. https://landsat.usgs.gov/landsat-5-history.

Van Etten, E. J. B. 2009. "Inter-annual Rainfall Variability of Arid Australia: Greater than
Elsewhere?" Australian Geographer 40, no. 1 (March): 109-120.

Wardle, Glenda M., Chris R. Pavey, and Chris R. Dickman. 2013. "Greening of Arid Australia:
New Insights from Extreme Years." Austral Ecology 38, no. 7 (July): 731-740.

Weismiller, R. A., S. J. Kristof, D. K. Scholz, P. E. Anuta, and S. A. Momin. 1977. "Change
Detection in Coastal Zone Environments." Photogrammetric Engineering and Remote
Sensing 43, no. 12 (December): 1533-1539.

Wischusen, J. D. H., L. K. Fifield, and R. G. Cresswell. 2004. "Hydrogeology of Palm Valley,
Central Australia; a Pleistocene Flora Refuge?" Journal of Hydrology 293, no. 1 (June):
20-46.

Woodcock, C. E., A. H. Strahler, and J. Franklin. 1983. "Remote Sensing for Land Management
and Planning." Environmental Management 7, no. 3 (May): 223-237.

World Wildlife Fund. 2016. “Ecoregions.” Accessed July 15.
https://www.worldwildlife.org/biomes

Abstract (if available)

Abstract This project aims to increase knowledge of vegetation changes in arid and semi‐arid areas in central Australia. Most of these zones are located across remote, sparsely‐populated, large and geographically diverse regions, making them difficult to study (Burns et al., 2014). Satellite imagery and geographic information systems (GIS) are viable options to decrease the knowledge gap in time‐ and cost‐effective ways and to understand how vegetation changes in areas with atypical annual seasons. The main goal of this thesis is to use modern techniques to understand vegetation dynamics occurring during 1989–1999 in Finke Gorge National Park (FGNP). During this time, land managers placed a fence around some park boundaries and removed a significant number of wild horses to enable the vulnerable vegetation to recover. An ensuing eight‐year field study observed and documented changes. This thesis intends to do the same, using remote sensing (RS) and GIS techniques. A supervised classification of soils and plants is done using data collected during field surveys. Principal components analysis (PCA), a data reduction technique, is used on multitemporal images to enhance continuous spatial and temporal changes and to extract factors that can be attributed to land management efforts at FGNP. Visual interpretation of components and analysis of classification information allowed for exploration of vegetation dynamics at an appropriate spatial and temporal resolution to understand variation and trends across time. The resulting components are compared to results of previous field surveys conducted at the time. The principal components indicate there are natural and human-derived sources of variation. Rainfall and other environmental factors play a major role on vegetation recovery of areas inside the fence, however, components also indicate that other sources of variation, such as land management practices conducted in the area, are contributors to variation. The field survey results are comparable to the thesis results

Linked assets

University of Southern California Dissertations and Theses

Conceptually similar

PDF

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

PDF

Classifying the alpines: developing a methodology to track environmental changes in the alpines utilizing remote sensing and ecozone vegetation patterns

PDF

Impacts of vegetation management on wildfire severity: a study of the 2021 Caldor fire

PDF

Development and evaluation of a USV based mapping system for remote sensing of eelgrass extent in southern California

PDF

Assessing modern conflict to monitor human rights with remote sensing: Russia's war in Ukraine

PDF

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

PDF

Questioning the cause of calamity: using remotely sensed data to assess successive fire events

PDF

Risk analysis and assessment of non‐ductile concrete buildings in Los Angeles County using HAZUS‐MH

PDF

Using VHSR multispectral imagery and object-based extraction to discover vernal meadows through vegetative persistence at Fort Ord, California

PDF

Mapping future population impacts caused by sea level rise in Huntington Beach and Newport Beach: comparing the cadastral-based dasymetric system to past dasymetric mapping methods

PDF

Satellite derived bathymetry in the Canadian Archipelago using multispectral and LiDAR space-based remote sensing

PDF

Spatial and temporal patterns of long-term temperature change in Southern California from 1935 to 2014

Asset Metadata

Creator Botkin, Adlin Noelia (author)

Core Title Exploring remote sensing and geographic information systems technologies to understand vegetation changes in response to land management practices at Finke Gorge National Park, Australia Between ...

School College of Letters, Arts and Sciences

Degree Master of Science

Degree Program Geographic Information Science and Technology

Publication Date 03/07/2018

Defense Date 01/22/2018

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

Tag arid environment,geographic information systems,land management,multi temporal imagery,OAI-PMH Harvest,principal components analysis,remote sensing,seasonal fractional vegetation cover,time series analysis,unsupervised classification,vegetation trends

Language English

Contributor Electronically uploaded by the author (provenance)

Advisor Fleming, Steven (committee chair), Lee, Su Jin (committee member), Ruddell, Darren (committee member)

Creator Email abotkin@usc.edu,adlin.botkin@gmail.com

Permanent Link (DOI) https://doi.org/10.25549/usctheses-c40-484361

Unique identifier UC11266249

Identifier etd-BotkinAdli-6085.pdf (filename),usctheses-c40-484361 (legacy record id)

Legacy Identifier etd-BotkinAdli-6085.pdf

Dmrecord 484361

Document Type Thesis

Rights Botkin, Adlin Noelia

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