You are on page 1of 95

69EG4218: M.Sc.

Dissertation

Delineation of Landcover Boundaries in Areas Utilized or Avoided by Female Caribou


during Calving and Post-Calving Using Publicly Available Spatial Datasets

Paul Warren Saunders

A dissertation submitted in partial fulfilment of the requirements for the degree of Master of
Science in GIS, The Manchester Metropolitan University.

Department of Environmental and Geographical Sciences


The Manchester Metropolitan University

May 2010
Declaration of originality

This is to certify that the work is entirely my own and not of any other person, unless
explicitly acknowledged (including citation of published and unpublished sources). The work
has not previously been submitted in any form to the Manchester Metropolitan University or
to any other institution for assessment for any other purpose.

Signed _________________________________________________

Date ___________________________________________________

i
Abstract

The availability and utility of spatial datasets, at no cost to the end user, directly impacts the
ability of government and non-governmental wildlife management agencies to delineate
landcover utilization or avoidance for targeted wildlife species. The availability and utility of
four datasets; Canada Land Inventory for Ungulates, Earth Observation for Sustainable
Development of Forests, Provincial Forest Inventory for the Island of Newfoundland, and the
Landsat 7 ETM+ were evaluated for their usefulness in delineating landcover boundaries in
areas utilized by caribou during calving and post-calving. To perform this evaluation a
representative sample of landcover features, in both utilized and avoided areas, were selected
through the use of space-time scan statistics and maximum step length calculations, then
landcover boundaries were recorded using spiral transects based on the Fibonacci sequence.
The location of all land cover boundaries intersected during the completion of ground based
transects were recorded and provided a baseline dataset for comparison to those depicted
using the selected datasets. Object-oriented segmentation had to be completed for the
Landsat ETM+ dataset before a comparison could be conducted. Root mean square error
(RMSE) values were calculated for all datasets and compared with the ground based results.
In addition, RMSE values were also calculated for a set of randomly generated boundary
locations for each completed transect. For all datasets errors of omission were taken into
account on an independent basis. Upon completion of the evaluation it was determined that
all datasets, except the Earth Observation for Sustainable Development of Forests, where
both the RMSE for random (r) and actual (a) boundary points (r=22.89, a=14.93, error
25meters (m)) was below the associated positional error of the dataset, would be useful for
the delineation of landcover boundaries. The Canada Land Inventory (r=86.60, a=30.43,
error 35m) was deemed useful only for its ability to provide information on historical location
and permanence of boundaries at the landscape scale. To provide landcover delineation for
the island of Newfoundland a combination of both the forest inventory (r=64.71, a=39.47,
error 35m) and landsat datasets (r=37.02, a=27.92, error 30m) must be utilized along with a
variety of ancillary data sources.

ii
Word Count

Number of Pages: 86
Number of Words: 16,02

iii
Table of Contents
Table of Contents _______________________________________________________________ iv

CHAPTER 1 ________________________________________________________________ 1

INTRODUCTION ____________________________________________________________ 1

1.1: Study Background ___________________________________________________________ 1

1.2: Study Area _________________________________________________________________ 4

1.3: Aim and Objectives __________________________________________________________ 6

1.4: Dissertation Structure________________________________________________________ 7

CHAPTER 2 ________________________________________________________________ 8

The Importance of Scale in Ecology and Remote Sensing ___________________________ 8

2.1: Introduction _______________________________________________________________ 8

2.2: Spatial and Temporal Scales and the Delineation of Landcover Features _______________ 9
2.2.1 The Concept of Scale _______________________________________________________________ 9
2.2.2 The Effect of Sensor Resolution on Landcover Identification _______________________________ 11

2.3: Landcover Boundaries, Animal Behaviour and Remote Sensor Resolution _____________ 15

2.4: Woodland Caribou and Scale _________________________________________________ 16

CHAPTER 3 _______________________________________________________________ 20

DATA AND METHODS _______________________________________________________ 20

3.1: Introduction ______________________________________________________________ 20

3.2: Data Used ________________________________________________________________ 21


3.2.1: Female Caribou Telemetry Data _____________________________________________________ 22
3.2.2: Landsat-7 Orthorectified Enhanced Thematic Mapper (ETM+) Imagery _____________________ 22
3.2.3: Earth Observation for Sustainable Development of Forests (EOSD) _________________________ 25
3.2.4: Canada Land Inventory (CLI), Land Capability Classification for Wildlife, Ungulates ____________ 26
3.2.5: Newfoundland and Labrador Provincial Forest Inventory _________________________________ 27

3.3: Methods and Techniques ____________________________________________________ 28


3.3.1: The Selection of Study Parameters and Implementation of a Study Design __________________ 28
3.3.1.1 The Identification of Areas Used or Avoided by Female Caribou __________________________ 28
3.3.1.2 The Selection of Attributes to be recorded During Sampling and Sampling Protocol Design ____ 32

iv
3.3.1.3 The Selection and Utilization of an Unbiased Sampling Design ___________________________ 38
3.3.2: The Comparison of Spatial Datasets __________________________________________________ 40
3.3.3: The Selection of Boundaries and Boundary Location for the Evaluation of Spatial Datasets _____ 42
3.3.4: Landsat 7 ETM+ Dataset Preparation _________________________________________________ 42
3.3.4.1 ENVI Feature Extraction Module Workflow ___________________________________________ 45
3.3.4: Landsat ETM+ Band Selection and Segmentation _______________________________________ 46
3.3.5: Temporal Currency and the Need for Ancillary Data _____________________________________ 52

3.4: The Evaluation of Selected Datasets. ___________________________________________ 52

CHAPTER 4 _______________________________________________________________ 53

RESULTS and DISCUSSION ___________________________________________________ 53

4.1: Introduction ______________________________________________________________ 53

4.2: Observational and Statistical Evaluation of Individual Datasets _____________________ 53


4.2.1: Canada Land Inventory ____________________________________________________________ 53
4.2.2: Earth Observation for Sustainable Development of Forests _______________________________ 56
4.2.3: Provincial Forest Inventory _________________________________________________________ 60
4.2.4 Landsat ETM+ Segmentation File Evaluation ___________________________________________ 63

4.3: Fulfilment of Aims and Objectives _____________________________________________ 64

CHAPTER 5 _______________________________________________________________ 67

CONCLUSIONS _____________________________________________________________ 67

5.1: Conclusions _______________________________________________________________ 67

5.2: Recommendations _________________________________________________________ 68

APPENDICES ______________________________________________________________ 69

REFERENCES ______________________________________________________________ 72

v
List of Tables
Table 2.1: Spatial and temporal scales utilized in previous caribou studies. There has not
been a consistent approach to the study of caribou within the context of scale. Earlier studies
were often confined to a single spatial or temporal scale but more recent studies have started
to look at caribou behaviour and landscape interaction at multiple
scales........................................................................................................................................17
Table 2.2: Caribou landcover associations vary with the scale of the life history trait being
investigated. As the temporal and spatial extent of the trait decreases the resolution of the
data required to determine landcover utilization must become finer.......................................18
Table3.1: Specifications for the Landsat series of Earth observation satellites.......................23
Table 3.2: Bands and bandwidths associated with satellites in the Landsat series..................24
Table 3.3: Characteristics of the Landsat ETM+ image utilized in this study.........................25
Table 3.4: The Canada Land Inventory for Ungulates utilized capability classes to rank the
ability of land to support ungulates. These classes are based on limitations that affect the
quantity and quality of habitat for the target species................................................................27
Table 3.5: Female caribou activity patterns during study period............................................28
Table 3.6: Covariance Matrix for Landsat ETM+ bands 1 - 5 and 7......................................47
Table 3.7: Correlation Matrix for Landsat ETM+ bands 1 - 5 and 7.......................................47
Table 3.8: Optimum Factor Index for the 6 highest ranked band combinations......................48
Table 4.1: RMSE values for datasets evaluated in this study. All RMSE values should be
interpreted within the context of the positional error of individual datasets............................65

vi
List of Figures
Figure 1.1: Adult female caribou and calf (courtesy John Neville)...........................................1
Figure 1.2: Canadian distributions of Woodland and Barren-ground caribou (Rangifer
tarandus) including the various boreal dwelling ecotypes (Map compiled by the Canadian
BEACONs Project, University of Alberta)...............................................................................2
Figure 1.3: Study area, The Topsails, Newfoundland, Canada..................................................5
Figure 1.4: Elevation values for the study area (meters), scale 1:50,000...................................6
Figure 2.1: An hypothetical illustration of caribou neighbourhoods as defined by Addicott et
al. (1987). The different neighbourhoods are a result of the interaction between the spatial
and temporal behaviour of both individual caribou and their actions as a group. (A)
Represents the frame within which the neighbourhoods were measured. (B) Represents the
complete range of the population over its lifetime. (C) Represents the area utilized by caribou
in the winter. (D) Represents the neighbourhood of caribou during calving..........................10
Figure 2.2: An increase in pixel size leads to an increase in the number of landcover features
represented leading to a reduction in pixel variance................................................................13
Figure 2.3: Temporal and spatial scales associated with various ecological processes
influencing caribou in the study area.......................................................................................16
Figure 3.1: Landcover features associated with transect sc2007095h2 as depicted by utilized
spatial datasets..........................................................................................................................22
Figure 3.2: Classification scheme for the EOSD dataset.........................................................26
Figure 3.3: The construction of cylinders during the use of space-time scan statistics involves
the setting of: (a) the spatial extent which dictates the cylinder diameter thus the area over
which points are included, and (b) the temporal extent which is represented by the height of
the cylinder with increasing height signifying the inclusion of points over a larger temporal
period........................................................................................................................................30
Figure 3.4: Transect centered on the maximum step length for caribou sc2006026. Blue dots
indicate the two endpoints that were used to calculate the centroid of the maximum step
length........................................................................................................................................32
Figure 3.5: Transect line based on Fibonacci spiral................................................................39
Figure 3.6: Sampling location identification and transect creation.........................................40
Figure 3.7: Positional Differences between Actual and Mapped Locations............................41

vii
Figure 3.8: An example of the well defined ecotone that exists between bog and forest in the
study area. Ecotones of this type are often driven by the water content of the soil and tend to
be highly stable over time........................................................................................................44
Figure 3.9: An example of a classification process that could be utilized for the Landsat
imagery evaluated in this study. In this study only the boundaries of features were extracted,
classification of the features contained was not completed.....................................................45
Figure 3.10: The ENVI feature extraction module workflow diagram (ENVI 2008)..............45
Figure 3.11: Landsat bands 1, 4 and 5 represented at: (a) 30m pixel size, (b) pansharpened
15m pixel size and (c) segmented landcover features created using ENVI software. Image is
centered on transect line sc2007096H5....................................................................................50
Figure 3.12: The effects of merge value selection on the number of segments created using
ENVI software: (a) scale factor 1, merge factor 0; (b) scale factor 1, merge factor 85...........51
Figure 3.13: Temporal lags associated with the datasets utilized in this study........................52
Figure 4.1: Landscape limitations associated with sites avoided or used by female caribou
based on the CLI dataset..........................................................................................................54
Figure 4.2: Primary species designation for each polygon intersected by individual transect
lines..........................................................................................................................................55
Figure 4.3: Location of the CLI polygon boundary, detected boundaries, and random points
for transect sc2006082H3.........................................................................................................56
Figure 4.4: The location of ground survey and EOSD boundaries for transect
SC2007051L3_07.....................................................................................................................57
Figure 4.5: The number of boundaries detected for ground-based and EOSD-based datasets
based on individual transects....................................................................................................58
Figure 4.6: The difference between the number of boundaries detected for EOSD data based
on the number of boundaries detected during ground-based surveys for individual transect
lines..........................................................................................................................................59
Figure 4.7: Boundaries detected during a ground based survey of transect SC2007095L1
overlain on the boundaries found in the Provincial Forest Inventory......................................62
Figure 4.8: Forest Inventory coverage of the area included in this study................................63

viii
CHAPTER 1

INTRODUCTION

1.1: Study Background


Caribou, Rangifer tarandus, have a circumpolar distribution, restricted to the Northern
Hemisphere. They are a medium sized representative of the deer family, being recognized as
one of its most primitive members (Banfield 1974, Figure 1.1). Five sub-species are
currently recognized in Canada with Woodland Caribou being the largest, reaching up to
400kg. Only the woodland subspecies, Rangifer tarandus caribou, is found on the island of
Newfoundland, which is represented in the figure 1.2 as Woodland caribou.

Figure 1.1: Adult female caribou and calf (courtesy John Neville).

1
Island of Newfoundland

Figure 1.2: Canadian distributions of Woodland and Barren-ground caribou (Rangifer tarandus)
including the various boreal dwelling ecotypes (Map compiled by the Canadian BEACONs Project,
University of Alberta)

Caribou have been present on the island of Newfoundland for approximately 5000 years (Bell
and Renouf 2004). Historical trends in population numbers are not documented however a
drastic decline in numbers was noted in the early 1900s leading to a closure of all caribou
hunting on the island in 1925(Bergerud 1971). In response to high poaching levels, hunting
seasons were reinstated in 1935, albeit at extremely low licence numbers. By mid century,
numbers had started to increase and licence sales were allowed to increase from 33 in 1933 to

2
7200 in 1950 (Bergerud 1971). In the mid to late 1990s caribou numbers peaked at
approximately 95,000 animals and then declined markedly to an estimated 37,000 by 2008 as
a result of increased adult mortality and reduced calf recruitment (Chris Callahan per.
comm.).

Responding to these dramatic declines, the Government of Newfoundland and Labrador


initiated a caribou strategy in 2006. The objectives were to identify possible causes for the
decline, and development of recommendations to halt or reverse the populations negative
trajectory. A component was the evaluation of existing and historical habitat utilized by
caribou thereby identify habitat attributes deemed important to caribou. This research
represents a sub-component of this habitat evaluation (Environment and Conservation 2008).
The yearly association of Newfoundland caribou with habitat types during specific periods in
their lifecycle has been well documented (Davis 1895, Ware 1903, Sclater 1905, Dugmore
1913). Historical yearly migrations to calving and post calving grounds have been delineated
(Mahoney 2000). It has been postulated that migration to specific calving and post-calving
rearing areas were based on the nutritional needs of the female or an attempt to reduce calf
mortality from predation (Bergerud 1985, Bergerud et al. 1984, Rettie and Messier 2001,
Tamstorf et al. 2005, Servheen and Lyon 1989, Mahoney and Virgl 2003, Barten et al. 2001,
Crete et al. 1990, Stuart-Smith et al. 1997, Johnson et al. 2002, ). This places a critical
importance on areas utilized by female caribou for calving and post-calving rearing. Given
the accumulated evidence regarding the impacts of human development, such as mine
development, forest harvesting, and linear development, on woodland caribou (Schaefer and
Mahoney 2007, Mahoney and Schaefer 2002, Chubbs et al. 1992, Weladji 2002, Vistnes and
Nellemann 2001, Weir et al. 2007, McLoughlin et al. 2003, Courtois et al. 2007, Bergerud
1974a), there is an increased need for baseline data on all areas within the woodland
caribous range. This study has been designed to provide an evaluation of the ability to
delineate landcover boundaries, and their occurrence in areas used or avoided by woodland
caribou on the island of Newfoundland during calving and post-calving using spatial datasets
available at no cost. This need is based on the premise that without the ability to accurately
identify and delineate the boundaries of landcover features it would not be possible to
quantify the relationship between caribou and specific lancover types, or the spatial patterns
exhibited by those landcover features across the landscape. Both, the degree of relationships
between a species and landcover features, and the spatial occurrence of those features in areas

3
utilized or avoided, is required to formulate effective habitat management plans (Rolstad
2005, Loveland et al. 2005, Griffith 2004).

1.2: Study Area


The study area is comprised of three separate sites totalling 2394km2 located in the Central
Newfoundland Forest Ecoregion, North-central subregion as shown in Figure 1.3 (Parks
Division 2000). A complex of coniferous forests and wetlands characterizes this area (Parks
2000, Damman 1964). Wetlands are represented by mire complexes, as defined by Rydin
and Jeglum (2006), often as mixture of bogs and fens. Raised bogs are a common feature in
this area. Forests are predominately coniferous and represented by Black Spruce (Picea
marianna) and Balsam Fir (Abies balsamea). Fire plays an important role in the occurrence
of specific forest types allowing for the establishment of Black Spruce forest in areas
previously dominated by Balsam Fir, as well as the establishment of localized stands of
White Birch (Betula papyrifera), Trembling Aspen (Populus tremuloides) and Pin Cherry
(Prunus pensylcanica) (Damman 1964). Alder (Alnus rugosa) is also abundant along the
edges of waterways and waterbodies or the transition zones between mires and forests.

Based on data from Environment Canada (2004) this region experiences a more continental
climate than other areas of the island with an average yearly temperature of 3.5 degrees
Celsius and 1200 mm of annual precipitation, approximately 30 percent which falls as snow.
Warmest temperatures are in July, average 16.2C and the coldest month is February, average
-9.1C. The region experiences 140 -160 growing days with green up beginning around mid-
May. Evapotranspiration rates range from 450 500 mm leading to a moisture surplus of
380 630 mm per year (Damman 1964).

The study area contains approximately 800 km of manmade linear features comprised of
roadways, transmission corridors and an old railway bed, all of which have the potential to
negatively affect caribou (Forestry 2008, Vors 2006 et al.). The majority of these features are
unpaved forest access roads used for pulpwood harvesting activities. Forest harvesting has
occurred in this area on a regular basis since the 1980s and has resulted in a mosaic of
cutovers in various stages of regeneration (Forestry 2000). Three large forest fires (over 200
ha) have been recorded in the study area occurring in 1964 (393 ha, location 49.0 -56.07),

4
1986 (1399 ha, location 49.031 -56.095), and 1999 (3675 ha, location 49.3 -56.23) (Canadian
Forest Service 2002).

Figure 1.3: Study area, The Topsails, Newfoundland, Canada.

Topography of the area is characterized by rolling terrain with elevation ranging from 76 -
647 meters (Wildlife Division 2007, Figure 1.4). Extreme values are represented by river
valleys and rock outcrops, and forests are restricted to higher terrain or areas where the
terrain rises above the surrounding mires.

5
Figure 1.4: Elevation values for the study area (meters), scale 1:50,000.

1.3: Aim and Objectives


The aim of this investigation is to evaluate freely available spatial datasets (available at no
cost) for the identification of boundaries between landcover features in areas either utilized or
avoided by female caribou during calving and post calving periods.

To reach this aim the objective will be to:

Identify publicly available (cost free) spatial data sets covering the study area.
Apply a sampling protocol for the identification of areas utilized or avoided by
caribou.
Delineate boundaries between landcover features and record associated attributes
occurring along transects representing areas utilized and avoided by caribou.

6
Quantify the accuracy of selected spatial datasets, when compared to recorded
transect data, using RMSE calculations.
Evaluate spatial error rates associated with selected spatial datasets and compare to
RMSE values obtained for each dataset to determine suitability of those datasets for
caribou landcover utilization studies.
Determine if a publicly available spatial dataset (available at no cost) can be used for
the identification and delineation of landcover features associated with areas utilized
or avoided by caribou.

The evaluation of selected datasets will rely on an understanding of scale as it applies to


animal ecology, the recording and utilization of spatial data, and the variability inherent in the
spatial and temporal scales selected for the study of woodland caribou. These topics will be
discussed in the following chapter.

1.4: Dissertation Structure


The following section outlines the aims and objectives of this study. Chapter Two presents
the concept of scale, and its relationship to remote sensing, ecology and caribou behaviour is
explored. This exploration includes recognition of the interplay between scale, caribou, and
boundaries. Chapter Three begins with the outline of selected datasets. This is followed by
methods for the unbiased selection of sample sites used for the collection of ground truth
data. Sampling design and the selection of attributes to be recorded was discussed next. The
use of root mean square error as a means for the comparison of datasets is then introduced.
Chapter 4 is a combined results and discussion section since this provided the best
mechanism to present graphical results and interpretative information in a sequence that was
easy to follow. This chapter ends with conclusions based on the presented results and
associated discussion. Chapter 5 highlights important components of the dissertation and re-
emphasizes conclusions based on the results obtained in the study. The final section outlines
recommendations that will allow continuation of this work and methods for utilization of the
datasets included in this study.

7
CHAPTER 2

The Importance of Scale in Ecology and Remote Sensing

2.1: Introduction
The use of remotely sensed data for the classification and delineation of wildlife habitat has a
long and varied history of adaptation to advances in technology and techniques used in its
collection (Glenn and Ripple 2004). The extent of these advances and methods of adaption
have been extensively reviewed during the past three decades (Gustafson 1998, Tueller 1980,
Gottschalk et al. 2005, Glenn and Ripple 2004). The development of desktop computing
systems allowed for the effective and efficient incorporation of remote sensing data into
research studies aimed at determining landcover utilization by wildlife (George et al. 1978).
Additionally, enabling technologies, such as Global Positioning and Geographical
Information systems, provided a cost effective means for management agencies to monitor
changes in wildlife habitat (Rogan and Chen 2004).

Tueller (1980) recognized the potential of remote sensing for use in the delineation of
wildlife habitat over large areas. This potential was reemphasized by Leyequien et al. (2007)
who concluded that remote sensing products, in conjugation with ancillary data, is the most
promising approach for the monitoring and management of biodiversity provided validation
is conducted using traditional observation techniques. Compared to traditional ground based
surveys of wildlife habitat, remote sensing products facilitate the use of more economical
techniques for habitat delineation (Kushwaha and Roy 2002).

Classification and delineation of landcover features using remotely sensed data has to be
conducted within the bounds set by the available data, and the ecology of the species to
which the results of the classification and delineation are to apply. All remotely sensed data
have inherent properties that influence the final use to which it can be applied. These
properties take the form of positional uncertainty, radiometric and spatial resolution of
sensors, and categorical uncertainty resulting from inadequate training data or generalization
procedures used in classification (Castilla and Hay 2006, Frank and Tweddale 2006).

8
The concept of scale, from both the temporal and spatial perspective, provides the primary
link between available remotely sensed data and its ability to delineate landcover utilized for
a specific species. If animals are interacting with landcover features on scales below the
resolution of the remotely sensed data then the resulting classification may actually
underestimate the amount of habitat available. Ritters et al. (1997) found a 50% reduction
for habitat delineated as the resolution of scanning windows became coarser. The issue of
scale permeates the complete process of the classification and delineation of landcover
utilization. It is influenced by; (1)the scale at which species locations are collected; (2) the
scale at which used and unused sites are indentified, (3) the scale at which ground truthing
data is collected; and (4)the minimum mappable area possible based on sensors used in the
collection of selected spatial datasets.

2.2: Spatial and Temporal Scales and the Delineation of Landcover


Features

2.2.1 The Concept of Scale


The concept of scale has been in existence for half a century with its roots traceable to work
conducted in the field of plant ecology. Pioneers in the field, such as Watt (1947), Greig-
Smith (1952) and Hutchinson (1953), started to realize that methods for the detection of
random or non-random distributions of plants provided conflicting or erroneous results
depending on the size and distribution of sample plots used. It was noted that the comparison
of ecological features across varying scales could pose problems (Turrill 1954). Over the
next three decades, the concept of scale became prominent in ecological literature reaching a
peak in importance by the mid to late 1980s. Throughout this period researchers attempted
to define scale in an ecological context and establish methods for selecting or setting
associated boundaries (Elton and Miller 1954, Goodall 1963, Van Dyne et al. 1963, Meade
1974, Mack and Harper 1977, Delcourt et al. 1983, Addicott et al. 1987)

Two overarching concepts prevailed during this period; the presence of patterns in the
distribution of an organism and the existence of a neighbourhood in which an organism
operates (Figure 2.1). These concepts were expanded to incorporate not only the spatial
aspects in intra and inter species relationships but also interactions across temporal spans.
Scale was seen as important due to conflicting results being obtained when research was
conducted that utilized different size plots or study areas. It was not until the late 1980s that
9
the concept of scale had matured and entered mainstream research as an overarching axiom
(Levin 1992, Fortin and Dale 2005).

Figure 2.1: A hypothetical illustration of caribou neighbourhoods as defined by Addicott et al. (1987).
The different neighbourhoods are a result of the interaction between the spatial and temporal behaviour
of both individual caribou and their actions as a group. (A) Represents the frame within which the
neighbourhoods were measured. (B) Represents the complete range of the population over its lifetime. (C)
Represents the area utilized by caribou in the winter. (D) Represents the neighbourhood of caribou
during calving.

During the later part of the 1980s the issues surrounding scale had reached a point where
Golley (1989) announced that a paradigm shift, as defined by Kuhn (1980), had occurred in
ecology (Schneider 2001). This narrowing of focus was evident with the creation of a branch
of ecology concerned with landscapes and the appearance of journals and papers dealing
specifically with the subject. It was at this time that landscape ecology was defined and its
place in Holistic theory described (Weins and Milne 1989, Zonneveld 1989). Research began
to address space and time issues that were identified in the previous decades, in an attempt to
develop methods and procedures to mitigate their affect on experimental results.

10
As research progressed it seemed that no sector of ecology was immune from the effects of
scale. This led to the formulation of a series of propositions that could be used to guide
research in the field:
1. Scale is species specific (Wiens 1989, Wiens and Milne 1989, Angelstam et al. 2003,
Baldi and McCollin 2003, Nams 2005).
2. The scale selected for a study influences the results obtained (Miline et al. 1989,
Jentsch et al 2002, Denny et al. 2004).
3. There is no set scale at which a study can be performed (Levin 1992, Denny et al.
2004).
4. A change in the temporal or spatial scale of a study causes a corresponding change in
the variance of variables being measured (Wiens 1989).
5. The level of heterogeneity observed for a given landscape depends on the scale at
which it is viewed (Chen et al. 2002).
6. Results obtained at one scale are not readily transferrable to another scale (Edmunds
and Bruno 1996).

Work in these areas is still progressing within the context of two paradigms; the scale-
invariance inherent in ecological patterns, and the hierarchical and distinct nature of
ecological patterns at various scales; both, which still have a role to play in individualized
studies (McMahon and Diez 2007). With the advent of studies at a broadening range of both
spatial and temporal scales and the increasing use of satellite obtained imagery, a call has
been made for the establishment of new interdisciplinary approach to the study of the
landscape at multiple scales named Satellite Ecology (Muraoka and Koizumi 2009, Table
2.2).

2.2.2 The Effect of Sensor Resolution on Landcover Identification


The issue of scale has been a part of remote sensing since the first camera was used to get an
aerial view of the landscape (Coops et al. 2007). Photography continued to be the standard
until 1959 saw the first satellite launched to provide coverage of the Earths surface. In 1972,
the first member of the Landsat series was launched, dedicated to the delineation and
monitoring of landcover (Cohen and Goward 2004). Additional satellites have been
launched, some to take the place of aging precursors, and others targeted at the collection of
data at other locations along the spectral spectrum (Rogan and Chen 2004). Consistent

11
throughout this period was the launch of satellites with the capability to provide images at a
finer spatial and/or spectral scale.

Scale is interchangeable with resolution, which in turn dictates the size of objects that can be
delineated using remote sensing imagery. Several aspects of remote sensing are affected by
scale, such as spatial, spectral, radiometric or temporal resolution and the extent of coverage.
All can vary in their implication for feature identification, dependent on their interaction
with, and the size of target objects (Coops et al. 2007, Renick and Grebner 2002). The
influences of these aspects are outlined below:

1. Spatial Scale
The spatial scale is most often representative of the size of the instantaneous field of
view of the imaging sensor (Tueller 1989). This is often stated as the effective pixel
or grain size in produced images and is sensor dependent (Innes and Koch 1998).
Currently available remote sensing data is available with a range in pixel size from
several centimetres to several kilometres. Pixel size has a direct impact on the
smallest land based object visible, with resolution decreasing as pixel size increases
(Liam et al. 2004). This is referred to as the smallest object mappable and it imposes
strict limitations on the use of satellite imagery (Anderson et al. 1976). Resolution of
satellite imagery has a direct effect on the area occupied by a specific feature (Saura
2004, Figure 2.2). Topan et al. (2009) found that the effective resolution of a given
satellite system may be greater than the stated nominal size of objects that can be
identified. It was found that as resolution got coarser the size of contiguous features
increased and small isolated features tended to disappear (Benson and MacKenzie
1995, Ponzoni et al. 2002). In addition, a disparity between the spatial resolution of
the image and the size of specific landcover features could lead to a reduction in pixel
variance reducing the ability to distinguish unique landcover features (Woodcock
1987). This leads to the occurrence of mixed-pixels which could pose a problem for
conventional classification techniques (Tiwari et al. 1999). Tawari et al. (1999)
provided an overview of the four prevailing techniques in use at the time for dealing
with this issue:
1. Maximum Likelihood Classifier
2. Linear Mixture Modelling

12
3. Fuzzy Sets
4. Neural Networks
New algorithms and techniques are still being developed to address this issue (Uttam
et al. 2008, Ge et al. 2006).

Figure 2.2: An increase in pixel size leads to an increase in the number of landcover features represented
leading to a reduction in pixel variance.

2. Spectral Scale
Spectral scale refers to the portion of the electromagnetic spectrum from which the
satellite-based sensors are able to collect data. These portions are commonly referred
to as bands and can encompass radiation from the ultraviolet to the infrared range.
The first satellites launched were restricted in the number of bands they could record
but as technology advanced hyperspectral based receivers, capable of recording
hundreds of bands, were launched (Coops et al. 2007). The addition of these bands
allowed for the separation of landcover features where there existed only small
differences in reflective profiles. This allowed for the mediation of spectral
13
swamping of adjacent bands caused by highly reflective landcover features which is
important in areas with high levels of landcover heterogeneity. (Ferguson 1990).

3. Temporal Scale
The identification of landcover features can benefit from remotely sensed data that
has been collected at various times of the year (Lefsky et al. 2001). Seasonal
availability of imagery for a given location is controlled by the repeat period of the
satellite and prevailing weather conditions, especially cloud and haze. As an
example, the Landsat 7 satellite has a repeat period of 16 days but it could take from
3-5 years to obtain a follow-up image for a specific location (Cihlar et al. 2003,
Ranson et al. 2003). Temporal scale become even more important when applied to
the activity patterns of the species under investigation (Boyce 2006). Landcover
utilization can vary seasonally resulting in the requirement for imagery from several
periods throughout the year which is not available for the datasets included in this
study.

4. Radiometric Scale
Coops et al. (2007) refers to radiometric resolution as the information contained in an
image expressed as a number representing the intensity recorded by the sensor.
Information is stored in the form of bits where a higher bit count corresponds to an
increasing number of recorded intensity levels. Current satellites, such as Landsat-7,
record intensities with 8 bits or 28 =256 intensity levels. Newer satellites are being
launched that use 11 bits. This is deemed as the least critical factor in selecting
satellite image data because the 8 bit level used by most satellites is more than
adequate for forest cover identification (Coops et al. 2007). Of important note, is the
impact that increasing temporal and spatial scales can have on the radiometric
properties of remotely sensed images. The scale of temporal impacts are influenced
by changes in the ambient conditions at the various times that images are obtained,
whereas, the scale of spatial impacts are dictated by the spatial separation of
landcover features and the effective pixel size of the recorded image (Tuominen and
Pekkarinen 2004, Chen et al. 2005).

14
5. Extent
Extent refers to the aerial coverage of the selected satellite imagery data. A
relationship exists between the spatial scale of the imagery and the extent covered,
whereby, as resolution increases the area covered by the image is reduced (Edwards et
al. 2001). This relationship also applies to the temporal resolution whereby an
increase in extent leads to increased temporal resolution (Cihlar 2000). Increases in
spectral and radiometric resolution have an indirect affect on extent through the
increase in data file sizes. The file size for a static extent would increase as the
spectral and/or radiometric resolution is increased. Any increase in extent would
compound the increase in file size limiting possible users. This limitation is being
overcome by the introduction of quad core computers and the lifting of random access
memory restrictions through the increased adoption of 64 bit based operating systems.

Issues related to scale are often project specific and vary depending on the scale at which
landscape features exist and the patterns they exhibit. This variability in landscape features
led Ju et al. (2005) to conclude that no single scale could represent a feature or classification.
Bock et al. (2005) and Andrefouet et al. (2003) extended on this concept by stating that the
observer and methods employed contribute to the selection of scale. It can therefore be stated
that the selection of the appropriate scale is dependent on the geographical phenomenon
under study, the scale at which it interacts with its surroundings and the type of analysis to be
performed (Makido and Shortridge 2005). Given the fine scale at which some species
interact with their habitat a mismatch currently exists between readily available remotely
sensed data and the scale required to evaluate these interactions (Kerr and Ostrovsky 2003).

2.3: Landcover Boundaries, Animal Behaviour and Remote Sensor


Resolution
Advances in animal telemetry technology allowed the collection of animal movement data at
temporal and spatial resolutions far finer than the readily available remotely sensed imagery
(Ropert-Coudert and Wilson 2005). When this telemetry data is plotted on coarse resolution
remotely sensed imagery it could show an exaggerated affinity, by the species being studied,
for larger landcover features. This is a direct consequence of the relationship between the
apparent size of landcover features and the resolution at which they were recorded (Benson

15
and MacKenzie 1995, and Ponzoni et al. 2002). This issue limits the ability to determine the
relationship between landcover features (especially those covering a small extent) and the
utilization or avoidance of those landcover features by a specific species.

2.4: Woodland Caribou and Scale


Woodland caribou are large, highly mobile ungulates that interact with the landscape at
multiple scales (Johnson et al. 2001, Figure 2.3). These scales are the result of life history
traits, such as fall and spring migration, selection of calving and post-calving areas, the fall
rut, occupation of a winter range, and the inherent heterogeneity of the landscape they
inhabit. The degree of landscape utilization various by an order of magnitude across its
range, from the selection of individual plants to historical shifts in distributions and has
necessitated the selection of remotely sensed datasets that are compatible with the scale under
investigation (Table 2.2). This has led to the completion of studies that vary widely in scale
in both the spatial and temporal domain (Table 2.1).

Figure 2.3: Temporal and spatial scales associated with various ecological processes influencing caribou
in the study area

16
Table 2.1: Spatial and temporal scales utilized in previous caribou studies. There has not been a
consistent approach to the study of caribou within the context of scale. Earlier studies were often
confined to a single spatial or temporal scale but more recent studies have started to look at caribou
behaviour and landscape interaction at multiple scales.
Location
Study Collar Fix Temporal
Spatial Fix Temporal
Extent Data Type Period Span References
Scale(s) Resolution Error Periods
(km2) (years)
(meters)
landsat 80
16,000 1 N/A N/A 1 2 George et al. 1978
field plots N/A+
25,000 6
1 N/A N/A 3 or 4 days <1000 12 Schaefer et al. 2000
700,000 11
forest cover 1 or 2
9,000 5 100 350 5 8 Apps et al. 2001
map weeks
field plot
100
2100 1 transect daily 100 3 2 Barten et al. 2001
50/10*
points
Rettie and Messier
N/A 1 N/A N/A 2 or 4 days <1000 5 5
2001
landsat
25
N/A 1 forest stand N/A N/A 2 23 Hansen et al. 2001
1:20000
map
20 min 16
5100 3 landsat 25 3-8 3 1 Johnson et al. 2002a
hours
forest fire 7h -
12,000 2 636 N/A 1 4 Joly et al. 2003
map monthly
Landsat
25
sample sites Teterukovskiy
N/A 3 30 1000 N/A N/A 1 1
9 plots/60m and Edenius 2003
10
grid
5100 1 Landsat 25 N/A N/A 1 1 Johnson et al. 2003
Ferguson and Elkie
160,000 1 N/A N/A 2 or 7 days <1000 5 6
2004
provincial Ferguson and Elkie
160,000 1 25 2 or 7 days <1000 5 6
landcover 2004a
38,830 2 DEM 25 33 days N/A 1 1 Johnson et al. 2004
forest cover
data 30 30 min, 2h, Saher and
5000 3 <30 2 2
environmental 30 6h Schmiegelow 2004
data
raster 25
5100 3 transects 100 3 4 hours 3-8 1 2 Johnson et al. 2004a
field plots 50
Landsat 25
34,225 2 N/A N/A 1 1 Theau et al. 2005
Field plots 5.4723.23
Landsat 75
7,410 2 6 hours N/A 3 2 Gustine et al. 2006
DEM 1:20000
1 week,
150
8000 2 VRI 1:20000 bimonthly 5 4 Jones et al. 2006
N/A
4h and 20h
forest
50,000 2 inventory 250 1 2 weeks 100 1 10 Wittmer et al. 2007
maps
870/10, 20,
600 3 transects 40,and N/A N/A 1 1 Mayor et al. 2009
130*
+ Not Available or Not Applicable
* Transect line length/Sample location along line

There has not been a consistent approach developed for the study of woodland caribou (Table
2.1). The spatial extent of completed studies varies from 600 700,000 km2, with a mean of

17
32,000 km2 (with the 700,000 km2 study removed). Several factors have contributed to the
variability in extents between studies including; population size, temporal period selected,
and the number of years over which the study was completed. It has been recognized that the
mapping scale can have a large impact on the results obtained in wildlife habitat or
behavioural studies (Anderson et al. 1999). This has led to the completion of studies at
multiple spatial scales with two distinct scales being used for most studies, these being the
population and the individual. A similar situation exists for temporal scales and has resulted
in the division of caribou life history into 6 temporal periods; calving, post-calving, summer,
fall, winter, spring/pre-calving with some subdivisions occurring in these catagories
(Bergerud 1974b, Rettie and Messier 2000). Although some studies have addressed all
periods, the majority have been concentrated on only a select portion of the complete yearly
life cycle. This requires researchers to be cognizant of the domain(s), as described by Ford
(2004), to which the results apply and the fact that it is inappropriate to apply findings outside
those domains without proper justification.

Table 2.2: Caribou landcover associations vary with the scale of the life history trait being investigated.
As the temporal and spatial extent of the trait decreases the resolution of the data required to determine
landcover utilization must become finer.
Required
Caribou Ecological Satellite Resolution
Resolution Mission
Period Sensor (meters)
(meters)
Spot-4 Vegetation 1150
Historical Distribution 1000 Spot-5
Envisat AATSR 1000
Aqua Modis 250
Range Shifts
< 1000 Terra
Historical
Envisat MERIS 300
Generational Home Terra ASTER 15
<100
Ranges RESOURCESAT-2 AWiFS 55
Landsat 7 ETM+ 30
Feeding and Resting <50
RESOURCESAT -2 LISS-III 23.5
Resurs DK 1 Geoton-L1 1-3
Plant Selection <10 Spot-5 HRG 5-10
RapidEye MSI 6.5

Landcover data used in the reviewed studies have been restricted primarily to available
Landsat imagery and government owned forest inventory maps. The resolution of utilized
landcover data is highly variable with 25 meters representing the lower limit for most studies.

18
Data on landcover below this limit has, in most cases, been collected through the use of plots
or transect and later digitized for use in a GIS. The use of higher resolution satellite imagery
was possibly limited by the high cost of data per km2 and the larger areas covered by
migratory ungulates like caribou (Table 2.1).

Also important is the high level of variability, both spatially and temporally, associated with
recorded caribou telemetry locations. The cause for most of this variability can be attributed
to the improvements in technology which translated into more frequent locations with a lower
positional error rate. The quality of mammal collar technology has matured to a stage where
limitations involved in obtaining fine scale data on movements are now only limited by
logistical and financial considerations

It was only in the last decade that works on the relationship between caribou and the
landscape in which they inhabit has received direct attention. This was the result of advances
in computer and satellite technology, which allowed for the processing of the large datasets
required for the study of a highly mobile and migratory species. Most of the newer satellite
datasets are only available at a substantial cost limiting most scientists to the use of publicly
available data. This study will evaluate several of these datasets an attempt to determine their
usefulness in providing a province wide map of landcover resources available to woodland
caribou. The next section will provide a brief overview of these selected datasets.

19
CHAPTER 3

DATA AND METHODS

3.1: Introduction
This research will be guided by two postulates:
1. Female caribou are responding to small scale (small aerial extent) landcover features due
to increased nutritional requirements and the need to reduce the chance of losing a calf
through predation, therefore there should be a significant difference between used and
unused areas (Gustine et al. 2006, Rettie and Messier 2000).
2. Landcover selection is determined by the ability of adult female caribou and calves to
move through the landscape, therefore mobility should be restricted in areas avoided by
female caribou (Recovery Implementation Group, Hart and Cariboo Mountains 2005,
Luick and White 1986).

Landcover selection has been evaluated using three distinct experimental designs (Garshelis
2000 p. 115). These designs are characterized as:
1. Use/Availability: compares the time an animal spends in each landcover type to the
percentage of availability.

2. Use/Demographic: compares the demographic response (recruitment, survival, etc.) for


populations utilizing different landcover types.

3. Site Attribute: compares a selection of site attributes in selected landcover types to


random or avoided sites.

Given the postulate that female caribou are reacting to fine scale landcover features, the site
attribute experimental design was utilized because of its ability to provide more detailed
information about landcover (Garshelis 2000 p.150) This allowed for greater precision in areas
where landcover features are clumped or rare (Buckland et al. 2007). The study design is based
on two fundamental assumptions: used sites are in appropriate habitat and unused sites are in
inappropriate habitat (Garshelis 2000 p. 143).

20
Data selected for the completion of this study is outlined below followed by a description of
methods for the selection of sites used or avoided by caribou. This is followed by a description
of methods used to determine the suitability of selected datasets. Suitability can be interpreted as
being the ability to use the selected datasets for the identification and delineation of landcover
boundaries in areas identified as being utilized or avoided by female caribou. To increase
validity and eliminate or reduce bias, the study was designed such that individual caribou
dictated the study area boundaries and the areas utilized or avoided. This was coupled with a
ground based landcover boundary detection protocol that eliminated or reduced bias that could
be introduced by any inherent directionality of landcover features.

3.2: Data Used


One caribou telemetry dataset and five spatial datasets were utilized in this study. The caribou
telemetry dataset was collected under the Newfoundland and Labrador Caribou Strategy, a 5-
year research program initiated by the Government of Newfoundland and Labrador. Four spatial
datasets were selected based on their accessibility, being available on the internet for download
by the public, and their potential for use in completing landcover delineation for the island of
Newfoundland (Figure 3.1). The selected datasets are represented by; Landsat 7 ETM+ imagery,
Earth Observation for Sustainable Development of Forests (EOSD), Newfoundland and
Labrador Forest Inventory Dataset, and the Canada Land Inventory for Ungulates. In addition,
landcover data was collected via ground transects in areas used or unused by collared caribou.
The characteristics of each of these datasets are outlined below.

21
Figure 3.1: Landcover features associated with transect sc2007095H2 as depicted by utilized spatial datasets.
(all images are displayed at the same scale)

3.2.1: Female Caribou Telemetry Data


A dataset was obtained from the Wildlife Division, Government of Newfoundland and Labrador,
containing telemetry data obtained from GPS collars on 11 individual caribou. Locations were
recorded at 2-hour intervals for the complete study period from May 15 to September 10, 2007.
Collars were comprised of both differentially corrected and uncorrected units with a
corresponding location error of <5m and <8m respectively.

3.2.2: Landsat-7 Orthorectified Enhanced Thematic Mapper (ETM+) Imagery


Landsat 7 is the latest in a series of Earth observation satellites operated by the National
Aeronautics and Space Administration Agency, with the first lunch in 1972 (Table 3.1).
Resolution and the number of bands have increased over this time period (Table 3.2). Cost
restricted use of collected data by the scientific community prompting the United States

22
government to enact the Land Remote Sensing Policy which was design to make imagery
available based solely on delivery cost. With the advent of the internet and geoportals this cost
has been reduced to nil.

Table3.1: Specifications for the Landsat series of Earth observation satellites.

Resolution Return
Launch Date Altitude
System Sensors* Range Period
(Decommissioned) (kilometers)
(meters) (days)

Landsat 23-Jul-72 RBV


6-Jan-78 79 - 80 917 18
1 MSS

Landsat 22-Jan-75 RBV


25-Feb-82 79 - 80 917 18
2 MSS

Landsat 5-Mar-78 RBV


31-Mar-83 40 - 240 917 18
3 MSS
Landsat MSS
16-Jul-83 30 - 120 705 16
4 TM
Landsat MSS
1-Mar-84 30 - 120 705 16
5 TM
Landsat 5-Oct-93
5-Oct-93 ETM 15 - 60 705 16
6
Landsat
15-Apr-99 ETM+ 15 - 60 705 16
7
*Sensor details can be found in Table 3.2

Imagery was made available to the public through a variety of internet based geoportals such as
the Canadian Geogratis website. Documentation supplied with this imagery describes
orthorectified as An image derived from a conventional image by simple or differential
rectification to remove image displacements caused by sensor tilt and relief of terrain. (GeoBase
2008). The result of orthorectification is an image with a planimetric accuracy of 20 30
meters. A secondary consequence is the reduction in feature mismatch during the mosaicing of
multiple images. Characterics of the Landsat image used in this study can be found in figure
Table 3.3.

23
Table 3.2: Bands and bandwidths associated with satellites in the Landsat series.
Bandwidths Resolution
Satellite Sensor
(micrometers) (meters)
RBV 0.48 0.57 80
0.58 0.68 80
0.70 0.83 80
Landsats 1 and 2
MSS 0.5 - 0.6 79
0.6 0.7 79
0.7 0.8 79
0.8 1.1 79
RBV 0.505 0.75 40

MSS 0.5 - 0.6 79


Landsat 3 0.6 0.7 79
0.7 0.8 79
0.8 1.1 79
10.4 12.6 240
MSS 0.5 0.6 82
0.6 0.7 82
0.7 0.8 82
0.8 1.1 82

TM 0.45 0.52 30
Landsat 4 and 5
0.52 0.60 30
0.63 0.69 30
0.76 0.90 30
1.55 1.75 30
10.4 12.5 120
2.08 2.35 30
ETM+ 0.45 0.52 30
0.52 0.60 30
0.63 0.69 30
0.76 0.90 30
Landsat 7
1.55 1.75 30
10.4 12.5 60
2.08 2.35 30
0.50 0.90* 15
*Panchromatic

24
Table 3.3 Characteristics of the Lansat ETM+ image utilized in this study (NAD 83 UTM 21)
Path Row Date Altitude Inclination Bands Resolution Corrections Cloud Cover
Taken (km) (degrees) (m) (%)
04 026 Aug. 05, 705 98.2 6 multispectral 30 geometic <10
2005 near polar 1 panchromatic 15 and
orbit 1 high-gain 60 radiometric
1 low-gain 60

3.2.3: Earth Observation for Sustainable Development of Forests (EOSD)


The EOSD is a pre-classified product created utilizing both Landsat 5 and Landsat 7 satellite
data. In specific cases ancillary data (i.e. forest inventory, aerial photography, etc.) was also used
to distinguish areas with large spectral overlaps. The EOSD was classified according to a
predefined classification scheme and legend (Wulder and Nelson 2003, Figure 3.2). The
product is available as an 8 bit paletted geotiff file at a resolution of 25 meters, using the NAD
83 (NTv2) datum and a Transverse Mercator (UTM) projection. Calibration and validation was
completed using ancillary data such as Provincial forest inventory or National Inventory Plot
data.

25
Figure 3.2: Classification scheme for the EOSD dataset

3.2.4: Canada Land Inventory (CLI), Land Capability Classification for Wildlife,
Ungulates
The compilation of a digital classification of the suitability of selected sections of the Canadian
landmass for ungulates was initiated in 1961 (Environment Canada 1970). By 1965 a
classification system was agreed upon by the federal, provincial and territorial governments.
26.5% of the Canadian landmass was classified which included 100% of the island of
Newfoundland (Environment Canada 1980). Areas were classified at a scale of 1:50,000 and
classifications were based on the best available data derived from multiple sources and in
multiple formats. Errors associated with the location of polygonal features were not specified in
the documentation. Primary land classification scheme is outlined in Table 3.4. It should be
noted that these primary classes are used in conjunction with a series of subclasses related to
climate and land categories plus identification of the target species. The reader is directed
towards the referenced documentation for more information on the classification scheme and
selected categories.
26
Table 3.4: The Canada Land Inventory for Ungulates utilized capability classes to rank the ability of land to
support ungulates. These classes are based on limitations that affect the quantity and quality of habitat for
the target species.
Landscape
Capability Description
Classification
Lands having no significant limitations to the production of ungulates. Capability on
CLASS 1 these lands is very high. They provide a wide variety and abundance of food plants and
other habitat elements.
CLASS 1W Lands in this special class are CLASS 1 areas which are winter ranges upon which
animals from surrounding areas depend.
Lands having very slight limitations to the production of ungulates. Capability on these
CLASS 2 lands is high but less than CLASS 1. Slight limitations are due to climatic or other
factors which have a slight adverse effect on the habitat.
CLASS 2W These are CLASS 2 lands which serve as necessary winter ranges for animals from
surrounding areas.
Lands having slight limitations to the production of ungulates. Capability on these
lands is moderately high, although productivity may be reduced in some years. Slight
CLASS 3 limitations are due to characteristics of the land which affect the quantity and quality of
habitat or to climatic factors which limit the mobility of ungulates or the availability of
food and cover.
CLASS 3W These are CLASS 3 lands which serve as necessary winter ranges for animals from
surrounding areas.
Lands having moderate limitations to the production of ungulates. Capability on these
CLASS 4 lands is moderate. Limitations are similar to those in CLASS 3, but the degree of
limitation is greater.
Lands having moderately severe limitations to the production of ungulates. Capability
CLASS 5 on these lands is moderately low. Limitations are usually a combination of two or more
of climate, soil moisture, fertility, soil depth to bedrock or other impervious layer,
topography, flooding, exposure, or adverse soil characteristics.
Lands having severe limitations to the production of ungulates. Capability on these
CLASS 6 lands is very low. Limitations are so severe that they are easily recognized. For
example, soil depth may be negligible or climatic factors so extreme that ungulate
populations are severely reduced.
Lands having limitations so severe that there is little or no ungulate production.
CLASS 7 Capability on these lands is negligible or non-existent. Limitations are so severe that
ungulate production is precluded or nearly precluded.

3.2.5: Newfoundland and Labrador Provincial Forest Inventory


The digital Provincial Forest Inventory is updated and maintained by the Newfoundland and
Labrador Forest Service. The databases extent covers most of the island of Newfoundland with
the exception of areas deemed as having little or no timber of any commercial value. The vector
based dataset is a compilation of spatial data form a wide variety of sources; including, ground
and helicopter based surveys and aerial photography interpretation. Positional error rates are

27
below 30 meters (Carl Marks, GIS Analyst, Wildlife Division, Newfoundland and Labrador,
personal communications 2008). Attribute classification is per the Data Dictionary for District
Library as published by the Provincial Department of Natural Resources (2007).

3.3: Methods and Techniques


Wulder et al. (2006) outlined the steps necessary to establish an effective accuracy assessment
framework for the evaluation of large-area landcover classifications. Emphasized in this paper
was the need to select the proper sampling unit and sampling design. Given the life history
traits of woodland caribou and the unique landcover associations related to these traits, this study
was bounded by both temporal and spatial constraints. Given these constraints, completion of
this study involves:
1. The identification of areas utilized and avoided by female caribou during the
specified time period (Table 3.5)
2. The selection of attributes to be recorded for selected areas.
3. The selection of a study design that will allow for the recording of attribute values
and the elimination or reduction of bias.

Table 3.5: Female caribou activity patterns during study period.

Activity Period From To


Calving Area Selection and Calving 15-May 14-Jun
Post-Calving Migration 15-Jun 7-Jul
Post-Calving Rearing 8-Jul 10-Sep

Methods used in the identification and selection of the above noted study components are
outlined below.

3.3.1: The Selection of Study Parameters and Implementation of a Study Design

3.3.1.1 The Identification of Areas Used or Avoided by Female Caribou


The delineation of used areas was conducted with a space-time permutation scan statistic
(STPSS) (Kulldorff et al. 2005). Developed for the detection of disease outbreaks, the STPSS
has seen only limited use in the field of ecology (Marj et al. 2006).
28
Uwt = cwt x tcwt / C (1)

Equation 1 represents the general form of the statistic where Uwt is the expected number of
caribou locations at location w and time t. Cwt is the number of caribou locations at location w at
time t , and C is the total number of caribou locations. The value of Uwt is used to calculate the
test statistic via equation 2.

Twt = [cwt / Uwt ]cwt X [(C-cwt)/(C- Uwt)] C- cwt if cwt > Uwt = 1 (2)

otherwise the test statistic is

T = maxwt Twt (3)

The statistic involves the use of millions of overlapping cylinders to define the spatial extent of
the study, and whose maximum diameter the user has defined (figure 3.3). Cylinder size varies
via a set of concentric circles, since we do not know the size of existing clusters a priori,
allowing for the identification of multiple sized clusters. Cylinder sizes range from 0 to the
maximum size specified and are centered on each of the points contained in the sample. The
temporal component of the statistics is represented by the height of the cylinder, which again
varies up to a maximum set by the user, and an increase in height translates to the inclusion of
points over a longer time period. This leads to the utilization of cylinders that can vary from
short and wide or tall and skinny, depending on the unique space-time variable combination of
spatial extent and time period used in their creation. Use of the statistic requires only
information on the location of the events and the time of occurrence (Kulldorff 2009). The
output of the statistic is a value that ranks the importance of the identified cluster for a specific
location bounded by a given spatial extent and time period. p-values for identified clusters can
then be compared using Monte Carlo testing methods (Dwass 1957) as per the following
equation:

p = rank(Tcluster) / (1 + # replicants) (4)


29
Figure 3.3: The construction of cylinders during the use of space-time scan statistics involves the setting of:
(a) the spatial extent which dictates the cylinder diameter thus the area over which points are included, and
(b) the temporal extent which is represented by the height of the cylinder with increasing height signifying the
inclusion of points over a larger temporal period

The STPSS is included as an extension in the software program SatScan that was originally
designed for the identification of disease clusters or outbreaks (Kulldorff and Information
Management Services, Inc 2006). Telemetry locations for collared caribou, collected under the
Newfoundland and Labrador Caribou Strategy, was utilized for the identification of used sites.

30
The use of the STPSS required input of both the spatial and temporal scale at which clusters
were to be identified. Based on visual analysis of locations, plus calculation of average step
length and daily dispersion, a maximum scan window of 0.5 km was used for the spatial
component. When selecting a temporal window the occurrence of three distinct behavior
regimes had been taken into account for the study period (Table 3.5). Through trial and error, it
was determined that a temporal window of 30 days gave an acceptable distribution of identified
clusters. The use of STPSS is warranted when the location of individual clusters is of concern as
opposed to a global measure of clustering within the study area.

Evaluation of landcover utilization also requires the identification of a study area from which a
sample of avoided sites can be drawn. Delineation of the study area has been recognized as a
possible source of bias in habitat related studies, especially when unused or avoided areas have
to be delineated (Manly et al. 2004 p. 5, Morrison et al. 2001 pp. 103-104). Bias occurs because
of influences from factors outside the boundaries of the study area, or selection of an area that
was identified as avoided because the temporal duration of animal telemetry was too short to
allow for the complete delineation of the utilized range. Bias was eliminated by allowing the
individual caribou to select both used and unused areas. While used areas are indentified using
the STPSS, an alternate method was developed to identify unused sites. This was accomplished
by calculating the three largest step lengths, for each caribou, with a software program called
Hawth Tools v3.27, an extension available for Esri ArcGis 9.2 (Beyer 2004, ESRI 2006).
Avoided areas were represented by calculating centroids, based on the endpoints of the three step
lengths selected above, using an ArcGis script created by Pete Aniello (2003) (Figure 3.4).

31
Figure 3.4: Transect centered on the maximum step length for caribou sc2006026. Blue dots indicate the two
endpoints that were used to calculate the centroid of the maximum step length.

3.3.1.2 The Selection of Attributes to be recorded During Sampling and Sampling


Protocol Design
The sampling protocol is designed to complement habitat evaluations conducted at the landscape
scale using GIS with information on mid and micro scale features of habitat utilized by female
caribou during calving and post calving. Data collected will also be used to determine the
accuracy of the provincial forestry inventory database and other selected dataset with the goal of
evaluating its usefulness for the identification of suitable caribou habitat on the island of
Newfoundland.

The landscape scale describes the spatial distribution of habitat features within the general area
of the calving or post calving telemetry location clusters. This will be used to identify broad
scale relationships of habitat features at selected telemetry clusters with the overall

32
characteristics of the study area. At this scale topography, site coordinates, and heterogeneity of
the landscape are deemed important. This analysis will be conducted through the use of existing
Landsat and forest inventory data.

This data will be complemented by measurement of variables at the mid scale level. At this level
parasitic fly relief, terrain roughness, presence of escape areas and the variability of habitat
features are deemed important. A categorical measurement of terrain roughness, as it relates to
the movement of calves, will be conducted. One measure of topography is the terrain profile (as
outlined in II a below) which has a major impact on the moisture content of the soil and directly
and indirectly influences vegetation cover. This data will allow for the development of a
categorical scale for the measurement of habitat variability.

At the micro scale (the scale at which a species interacts with the environment on a daily basis to
fulfil the need for food, water, and shelter) an evaluation of forest stands and other habitat types
intersected along Fibonacci spiral transect lines will be carried out. Forest composition, height,
age and canopy cover will be measured as per existing forest inventory guidelines. Forest stand
evaluation will be conducted for stands intersected by transect lines and measurements taken
50m from the stand edge or at the center of the stand if stand size does not permit this distance.
Micro scale data will be used to evaluate the suitability of the four spatial datasets included in
this study.

The following list identifies the variables to be measured and variables are placed according to
their hierarchical level. Variables are located under one classification only.

I. Landscape Scale (measured for study area comparison)

a. Geographic relationships

a. Coordinates (WGS 1984)

b. Elevation

c. Slope (%)

d. Aspect (degrees)
33
II. Mid Scale (measured at the sample site level)

a. Terrain Profile

1. Concave surface profile is mainly hollow in one or several


directions
2. Convex surface profile is mainly rounded like the exterior
of a sphere
3. Straight surface profile is linear, either flat or sloping in one
direction

b. Site Classification (category)


1. Rock Barren barren rock land without sufficient soil for trees
(<10 tree cover)
2. Soil Barren barren land other than rock which does not
support tree vegetation usually due to adverse climatic
conditions of soil factors (<10% tree cover)
3. Organic Bog wet areas of bog or marsh that may include
patches of open water wetness may be seasonal or close
enough to the surface to prevent tree growth
4. Treed Bog wet areas of bog or marsh with < 10% tree cover
5. Wet Bog wet areas of bog or marsh with large amount of
open water patches
6. Agriculture land cleared for agriculture purposes
7. Residential land cleared for residential, commercial or cabin
development
8. Right-of-way land cleared for roads, railways, cutlines or
power lines
9. Cleared Land land cleared for other purposes not listed above
10. Forested all area within a 10m radius of the sample point is
treed
11. Cutover any area that was subjected to clear cutting and has
not reached the age of 80+
12. Shrub
13. Grasses
14. Stream
15. River
16. Pond
17. Lake

c. Major feature Orientation (N-S, E-W, NE-SW, etc.)

34
III. Micro Scale (measured at the forest stand level)

a. Transition Zones (measured for intersection with habitat changes)

1. For each intersection

1. Transect lag (section)

2. Distance from origin

3. Coordinates of intersection

4. As per site classification (category)

5. Terrain Roughness

1.micro low relief features (< 0.3 m high) with


minimal effect on calf movements
2.slightly prominent features (0.31m high)
spaced > 7 m apart
3.moderately prominent features (0.31m high)
spaced 37 m apart
4.strongly prominent features (0.31m high)
spaced 13 m apart
5.severely prominent features (0.31m high)
spaced < 1 m apart
6.extremely very prominent features 100%
ground cover, movement possible with large
effort
7.ultra very prominent features, movement
impossible or at an extremely low rate
Note: the inverse of these criteria are to be used to evaluate
water based features.

6. Repeat

b. Forest Stand Evaluation

a. Species Composition

1. Balsam Fir (bF)

2. Black Spruce (bS)

3. Balsam Popular (bP)


35
4. Englemann Spruce (eS)

5. European Larch (eL)

6. Jack Pine (jP)

7. Japanese Birch (jB)

8. Lodgepole Pine (iP)

9. Norway Spruce (nS)

10. Red Maple (rM)

11. Red Pine (rP)

12. Scots Pine (sP)

13. Sitka Spruce (sP)

14. Tamarack/Larch (tL)

15. Trembling Aspen (tA)

16. White Birch (wB)

17. White Pine (wP)

18. White Spruce (wS)

19. Yellow Birch (yB)

20. Alder Species (aS)

Note: > 75% coverage one species listed


Two species listed one >50% other makes up remainder
Three species 40% -30%-30%
c. Age Class

1. 1-20 (Regenerating)

2. 21-40 (Immature)

3. 41-60 (Semi-mature)

4. 61-80 (Mature)

36
5. 81-100 (Over mature)

6. 101-120 (Over-mature)

7. 121+ (Over-mature)

9. All age stands

Note: Age to be determined by core sampling of over story, visual


observation or reference to forest inventory data.
d. Height Class

1. 0-3.5 meters

2. 3.6-6.5 meters

3. 6.6-9.5 meters

4. 9.6-12.5 meters

5. 12.6-15.5 meters

6. 15.6-18.5 meters

7. 18.6-21.5 meters

8. 21.6+ meters

e. Crown Closure

1. >75%

2. 51-75%

3. 26-50%

4. <= 25%

5. Open

f. Site Disturbance

1. Cutover

2. Fire

3. Insect

37
4. Wind

5. Vegetation

6. Other

g. Understory

1. Absent (moss/lichen/grasses only)

2. Herbs <0.5m high

3. Shrubs/herbs 0.6-1.0m high

4. Shrubs/regeneration >1m high

Sampling of attributes (variables) along transect line was conducted using the line intercept
(intersect) method (Newton 2007 p. 95, Morrison et al. 2001 pp. 69-72, 109, Dale 2004 pp. 41-
42, Van Wagner 1982, Canfield 1941, Gregoire et al. 2004, Lutes 2002, Buckland et al. 2007,
Affleck et al. 2005, Manly 2002). Attributes were selected based on landcover observations in
areas where female caribou were known to have occupied during the study period. To make data
compatible with existing data on forest cover, attribute selection and associated categories
followed the Data Dictionary used by Provincial Forestry personnel as closely as possible (Dept.
of Natural Resources 2007). Site classification codes were adopted from the data dictionary with
the required addition of codes related to water bodies and waterways, as well as codes pertaining
to alternate cover types, such as grasses. When the landcover type intercepted was a forest stand
standard classification codes for species, age class, height class and canopy cover were used with
additional codes for site disturbance and understory being added. All data was recorded on the
field sheets in Appendix A.

3.3.1.3 The Selection and Utilization of an Unbiased Sampling Design


After identification of used and avoided sites, selection of an unbiased sampling design was
required. Bias in sampling can be introduced by the spatial orientation (directionality) and
distribution of features (trend or heterogeneity) (Kalikham I. 2007, Fortin and Dale 2005). It was
noted during caribou collaring work and while conducting a point sampling pilot project in 2006
that landcover features exhibit both a high degree of directionality and heterogeneity. The

38
sampling design was thus selected to eliminate or reduce bias that could be introduced by spatial
distribution of landcover features in the study area.

Fortin and Dale (2005) describe the use of Fibonacci spirals as a means of avoiding error that
may be introduced by directionality and trend. A modified version of the Fibonacci spiral was
constructed from straight line segments for use in this study(Figure 3.5). The adequacy of a
spiral sampling design was emphasized by Kalikhman (2007) who also noted that shorter spirals
provided the same results as straight or zig-zag lines. Construction of spirals centered on the
centriods of location clusters and maximum step lengths utilized an AML script developed by
Carl Marks (2008) for use in ArcInfo 9.1 (ESRI 2005).

Figure 3.5: Transect line based on Fibonacci spiral.

A flow chart representing all steps taken in the creation of a sampling transect is represented in
Figure 3.6. The methods outlined allowed for the creation of sampling transects that could be
completed in half a day and, for most caribou, resulted in the creation of transects that
represented the three behavioural periods listed in Table 3.5.

39
Figure 3.6: Sampling location identification and transect creation.

3.3.2: The Comparison of Spatial Datasets


To allow the evaluation of all spatial datasets utilized in this study, line intersect sampling was
used for the compilation of a ground truth dataset. Given the diverse techniques used in the
creation of each spatial dataset and differences in the final product, they were evaluated on an
individual basis. This evaluation was based on the ability to identify or quantify positional
accuracy of boundaries between landcover features identified during collection of the ground
truth dataset. A statistical method for the quantification of positional accuracy is the calculation
of the root mean square error (RSME) (Worboys and Duckham 2004).

The RMSE provides a measure of the error between the actual location of a feature and the map
location (Figure 3.7). Differences between the x and y coordinates are calculated as per equation
5 and 6:

x (actual) x (mapped) = difference in x (5)

40
y (actual) y (mapped) = difference in y (6)
Equations 5 and 6 then allow for the calculation of the (error radius)2 as per equation 7:

(difference in x)2 + (difference in y)2 = (error radius)2 (7)

All (error radius)2 values are summed and divided by the number of point pair comparisons (n):

( )2 + ( )2
=( )2 (8)

The RMSE is then calculated per equation 9:

( )2 = (9)

Figure 3.7: Positional differences between actual and mapped locations.

41
The allowable RMSE is based on a relationship between the absolute positional accuracy of a
given dataset and the Z score at a selected confidence interval, therefore the allowable RMSE is
dependent on the spatial dataset being evaluated and the level of confidence the users requires.

3.3.3: The Selection of Boundaries and Boundary Location for the Evaluation of Spatial
Datasets
All currently existing landcover boundaries were identified and delineated using on the ground,
transect based, surveys. Boundary identification was based on an observed change from one
landcover type to another, provided that the new landcover type encountered along the transect
was > 10m in width, otherwise the transition to an alternate landcover feature was not recorded.
This was required to avoid recording small patches included in otherwise contiguous landcover,
or the transitions zones between distinct landcover features, both of which would not be
discernable from utilized datasets due to their limited resolution. All boundary locations
observed during the completion of survey transects were recorded as a point using a handheld
GPS. GPS units used in this study had an associated positional error of 3-10 m.

Boundaries for an evaluated spatial dataset were selected based on those identified during
completion of ground based survey lines. Boundaries consisted of either vector based entities or
the edge of raster cells which depicted different landcover features. Both types of boundaries
were extracted and saved in new shapefiles using ArcGIS. The extraction of boundaries from the
Landsat 7 data required additional steps before extraction consisting of pansharpening,
segmentation and vector extraction all of which are described in more detail below. RMSE
values were then calculated using distances between the actual boundary location (from ground
based surveys) and the location depicted by each dataset.

3.3.4: Landsat 7 ETM+ Dataset Preparation


The study area is represented by landcover features from three distinct ecosystems, boreal forest,
bog and fen complexes, and alpine/upland barrens. This creates the need for the selection of a
band, band combination, or combination of band composites, that provides the best separation of

42
landcover features that are commonly found within the same spatial extent. For this component
of the study methods were developed for the identification of boundaries between landcover
features. This is based on the premise that the classification of landcover features can only be of
benefit if the extent of those features can be determined through the accurate delineation of their
boundaries.

To facilitate the delineation of boundaries associated with landcover features a mechanism needs
to be developed for the demarcation of ecotones based on landsat imagery. Ecotones can be
defined as an identifiable transition between landscape features where one feature changes to
another due to underlying biotic or abiotic factors (Kark and van Rensburg 2006). During the
completion of transects it was noted that most ecotones in the study area were narrow, often
falling below the 10m limit for feature recording, which has created sharp, will defined,
boundaries between landcover features, a factor that could be beneficial for completion of the
boundary delineation exercises (Figure 3.8).

Lu and Weng (2007) recognized that the path to a successful classification scheme was
dependent on the development of a suitable image processing procedure. It was noted that
classification algorithms and procedures were continually being developed but information on
these developments were spread across the literature. This prompted Lu and Weng (2007) to
complete an in-depth review which could be used to guide practitioners in the selection of a
suitable process for completion of their classification needs.

The delineation of feature boundaries was completed using feature extraction which was
available as an add-on extension module for the image analysis software ENVI. Details of the
ENVI Feature Extraction Model can be found in the associated users guide (ENVI 2008). This
software package was selected because it provides an object-orientated (OO) segmentation
method that could reduce the effects of high landscape heterogeneity, which was identified as a
possible reason for the discrepancy between the ground-based survey and the EOSD dataset.

Santos et al. (2006) noted that object-based classification was inferior to pixel-based techniques
for low resolution images having a pixel resolution of 300m. Lewinski (2006) performed a
43
similar exercise using Landsat imagery with a pixel resolution of 30m and found a substantial
improvement in classification accuracy over traditional pixel based methods. Both of these
conclusions were supported in a study completed by Gao and Mas (2008) with its usefulness for
the classification of Landsat ETM+ imagery being further supported in a study completed by
Oruc et al. (2004). This forms the basis for selecting the ENVI OO based extraction of
landcover boundaries using segmentation. A flow chart on the steps for such a segmentation and
classification exercise is outlined in Figure 3.9.

Figure 3.8: An example of the well defined ecotone that exists between bog and forest in the study area.
Ecotones of this type are often driven by the water content of the soil and tend to be highly stable over time.

44
Figure 3.9: An example of a classification process that could be utilized for the Landsat imagery evaluated in
this study. In this study only the boundaries of features were extracted, classification of the features
contained was not completed.

3.3.4.1 ENVI Feature Extraction Module Workflow


The ENVI Feature Extraction Module Users Guide provides a workflow diagram on page 7
which has been reproduced in figure 3.10.

Figure 3.10: The ENVI feature extraction module workflow diagram (taken from ENVI 2008).
45
The extraction of features using OO classification can be divided into two processes; image
segmentation and image classification (ENVI 2008, Gao and Mas 2008, Lewinski 2006).
Segmentation is defined as a process of dividing an image into segments by grouping adjacent
pixels with similar feature values (brightness, texture, colour, etc). The second step is
classification which has been divided into two pathways, rule based and supervised. Supervised
classification, is the process of using training data (samples of known identity) to assign objects
of known identity to one or more feature classes and then placing all remaining features into
these classes based on their similarity. Rule based classification requires the building of rules
based on object attributes and then using those rules to classify all unclassified features in the
dataset. The final step in the process is the export of all objects to a shapefile as polygons, lines
or points. It is at this stage that vector smoothing can be undertaken.

3.3.4: Landsat ETM+ Band Selection and Segmentation


Landsat ETM+ imagery is composed of 8 bands, each of which can be used for the identification
and delineation of landcover features based of their ability to reflect or emit energy at a specific
band width. Landsat spectral bands have been used individually or in combination, as composite
images, for landcover classification of forested and other vegetated areas throughout the world
(Jakubauskas and Price 1994, Boyd and Danson 2005, Wulder et al. 2004, Ramsey et al. 2004,
Syed and Abdulla 2002). Several band combinations (composites) have been used to aid in the
visualization of distinct surface features. Two of the most common are a true colour composite
combining the red, blue and green bands and the false colour near infrared composite made up of
bands red, green and the near infrared (Richards and Jia 2006, Mather 2004, Canada centre for
Remote Sensing 2007, US Army Corps of Engineers 2003). The selection of specific bands for
the completion of a landcover segmentation and\or classification exercise is dependent on the
landcover features in the target area. The study area covered by this research is composed of a
mosaic of boreal forest, scrub and shrub combinations, mires, and barren alpine\tundra\taiga
landscapes and the reflective properties of each of these landcover types varies, creating the need
for the selection of a band or band combination that will provide the highest degree of separation
between landcover types recorded during the ground based survey.

46
A representative band combination was selected through an evaluation of the covariance and
correlation matrices for all bands in the landsat image except bands 6a and 6b. Both matrices
were calculated using ArcGIS and can be seen in Table 3.6 and 3.7.
Table 3.6: Covariance Matrix for Landsat ETM+ bands 1 - 5 and 7.
ETM+ Layer 1 2 3 4 5 7
1 34.81255 30.12941 38.68843 11.73514 35.53750 26.82675
2 30.12641 29.72593 36.42610 24.68066 45.76299 29.54172
3 38.68843 36.42610 52.33810 17.72576 69.24177 46.57965
4 11.73514 24.68066 17.72576 105.34611 76.79231 27.91986
5 35.53570 45.76299 69.24177 76.79231 199.14202 108.97634
7 26.82675 29.54172 46.57965 27.91986 108.97634 66.38234

Table 3.7: Correlation Matrix for Landsat ETM+ bands 1 - 5 and 7.


ETM+ Layer 1 2 3 4 5 7
1 1.00000 0.93660 0.90637 0.19378 0.42681 0.55805
2 0.93660 1.00000 0.92350 0.44104 0.59479 0.66503
3 0.90637 0.92350 1.0000 0.23872 0.67823 0.79024
4 0.19378 0.44104 0.23872 1.0000 0.53018 0.33387
5 0.42681 0.59479 0.67823 0.53018 1.00000 0.94782
7 0.55805 0.66503 0.79024 0.33387 0.94782 1.00000

Utilizing the covariance and correlation matrix, and the standard deviation associated with each
band, the Optimum Index Factor (OIF) was calculated for all band combinations. The OIF
calculation is dependent of the standard deviation of the pixel value for individual bands and the
value of the correlation between band pairs, and was developed to identify the 3 band
combination that provides the highest amount of information with the lowest amount of overlap
(Chavez et. al.1984, Dwivedi and Rao 1992). OIF values are calculated using the following
equation:

(10)
Where:
Stdi standard deviation of band i
Stdj standard deviation of band j
Stdk standard deviation of band k
Corrij correlation coefficient of band i and band j
Corrik correlation coefficient of band i and band k
Corrjk correlation coefficient of band j and band k

47
For the six bands included in this study the band combination 1, 4, 5 was ranked as the highest
for use in classification activities (Table 3.8).

Table 3.8: Optimum Factor Index for the 6 highest ranked band combinations.
ETM+ Band Number Band Combination Optimum Index Factor
1 1,4,5 33.55
2 3,4,5 32.76
3 2,4,5 31.84
4 4,5,7 31.40
5 1,4,7 29.06
7 1,5,7 28.88

Fox et al. (2002) demonstrated the use of pan-sharpening an image to increase its usefulness in
delineating landcover features utilized by wildlife. The pan-sharpening process involves the
fusion of selected 30m landsat bands with the panchromatic band to produce an effective pixel
size of 15m. Pan-sharpening activities were conducted using the Create Pan-Sharpened Raster
Dataset tool under Data Management in ArcGis 9.3., utilizing the ESRI algorithm based on
spectral modelling (Figure 3.11).

The next step involved the identification and delineation of boundaries, based on the pan-
sharpened image, for all landcover features in the study area. This was completed using the
Feature Extraction Module included in the ENVI software package described above which
utilizes a technique called segmentation to identify candidate boundaries. The benefit of using
segmentation for the delineation of landcover features and landscape ecological analysis has
been outlined by Devereux et al. (2004). Segmentation involves the selection of the appropriate
scale at which pixels are to be viewed during the aggregation process and the degree to which
resulting segments are to be merged. ENVI employs an edge-based detection algorithm
designed by Xiaoying (2009) for segmentation and the Full Lambda-Schedule algorithm
developed by Robinson et al. (2002) for the merging of segments. Scale selection for landscape
analysis was addressed by Burnett and Blaschke (2003) who proposed reducing landscape
objects to their smallest most basic unit which they called a holon. They developed the concept
further by describing the hierarchal nature of landscape features which emphasized that any scale

48
selected which is greater than the mean holon size will delineate objects containing multiple
features.

The hierarchical nature of forest structure and the effects of scale on segmentation were also
noted by Lamonaca et al. (2008). Utilizing three different segmentation scales they
demonstrated that a decomposition of forest structure was possible. The results from these
studies show that segmentation scale selected directly influences the number of landcover objects
delineated by the segmentation process. For this study, segmentation was conducted at a scale
level of 1 based on a range of 0 to 100, 100 being the largest. This was done to insure that all
landcover features detectable within the constraints imposed by the image pixel size, were
actually detected.

The selection of such a small scale will create around 1,000,000 separate objects for an area with
similar landcover types and size as the study area. In this study 987,264 objects were created
2
having a mean size of 2230m during the scale selection step for the ENVI feature extraction
module. The size of these objects will be limited by the pixel size of the image creating a lower
threshold beyond which segmentation cannot occur. Depending on study objectives, such a large
number of objects will have to be merged into units representing contiguous landcover features.
Merging represents the second step in the ENVI feature extraction process.

As with the selection of scale, the level of merging within the ENVI program must be selected
from a scale of 0 to 100, with 0 representing the least amount of merging. To determine the best
merge level for segments created in first step, results from multiple segmentation exercises were
compared to actual transect boundary locations. This resulted in the selection of a merge level of
85 which gave the best agreement with boundaries indentified in select transects. Upon
completion of this step the number of objects were reduce to 258,858 and the mean size
increased to 21,895m2. Results from this exercise were exported to a vector file for use in
ArcGis (Figure 3.12).

49
(a)

(b)

(c)
Figure 3.11: Landsat bands 1, 4 and 5 represented at: (a) 30m pixel size, (b) pansharpened 15m pixel size and
(c) segmented landcover features created using ENVI software. Image is centered on transect line
sc2007096H5.
50
(a)

(b)
Figure 3.12: The effects of merge value selection on the number of segments created using ENVI software: (a)
scale factor 1, merge factor 0; (b) scale factor 1, merge factor 85.
51
3.3.5: Temporal Currency and the Need for Ancillary Data
All utilized datasets were prone to some degree of temporal incongruence (Figure 3.13).

Figure 3.13: Temporal lags associated with the datasets utilized in this study.

This incongruence had to be taken into account during the calculation of RMSE values for
individual datasets. Ancillary datasets, such as forest cutover and road layers, were used to allow
for the identification of boundaries that would have been present during the creation of a specific
dataset. Failure to take into account features, such as roads or cutovers, which were not present
when the selected datasets were created, could bias the results of the RMSE calculation

3.4: The Evaluation of Selected Datasets.


In the next section RMSE values will be calculated for each dataset based on the included
boundary data. Results will be presented in graphical, tabular or written format, dependent of the
characteristics of the original dataset.

52
CHAPTER 4

RESULTS and DISCUSSION

4.1: Introduction
There is a great cost differential between coarse, medium and high resolution remotely sensed
data, which can differ by an order of magnitude in price as one moves between the three dataset
types (Rogan and Chen 2004). This has restricted many government agencies, or
nongovernmental organizations, to the use of low resolution, or dated datasets that can be
obtained free of cost. This often means using information that was developed for alternate
purposes, such as forestry inventory data collected by state or provincial agencies. The use of
these datasets often requires a preliminary analysis of their appropriateness for use in a given
task. Work presented in this thesis involved the evaluation of four datasets for use in the
identification and/or delineation of landcover boundaries in areas utilized or avoided by female
caribou. Success in delineating landcover boundaries was used as a proxy to determine the
suitability of the dataset for subsequent future classification of landcover features.

4.2: Observational and Statistical Evaluation of Individual Datasets

4.2.1: Canada Land Inventory


Five primary landscape limitations were extracted from the CLI data associated with sites
avoided or used by female caribou and have been displayed in Figure 4.1. There was no
significant difference between primary landscape limitations occurring at used or avoided sites
(X2=6.392, df=4, p=0.172).

Only two ungulate species, moose and caribou, exist on the island of Newfoundland with data on
both species being incorporated into the CLI dataset. All areas of the island were designated as
both moose and caribou habitat with an interchange between primary species occurring on a
polygon-by-polygon basis. The occurrence of either moose or caribou as the primary species
had no effect on whether a site was avoided or used (X2=2.922, df=1, p=0.087). A review of
Figure 4.2 does show a difference between primary species designations and site utilization,
which is significant at the 10% interval, and indicates the need for continued investigation.
53
12.5

Limitation
Landform
Soil Moisture
10.0 Fertility
Exposure/Aspect
Climate

Count
7.5

5.0

2.5

0.0
Avoided Used
Site Classification

Figure 4.1: Landscape limitations associated with sites avoided or used by female caribou based on the CLI
dataset.

The CLI for ungulates was developed from a variety of ancillary data, which included ground
surveys and air photo interpretation. Given the use of fine resolution data in the development of
the CLI, a positional error rate of 30m was selected as a buffer around ground based GPS
locations. Map boundaries (polygon perimeters) were considered as matching ground based
survey data if they fell within a 30m buffer zone around plotted landcover boundaries. Boundary
detection was considered successful if points were within 30m of the intersection between the
boundary being evaluated and the transect. 17 of the transects used were crossed (one transect
passed within 5m of the boundary) by polygon boundaries in the CLI dataset. Boundary
detection was deemed successful for 16 of these transects.

54
20

SPECODE
Moose
Caribou
15
Count

10

0
Avoided Used
Site Classification

Figure 4.2: Primary species designation summary for each polygon intersected by individual transect lines.

To confirm boundary detection the RMSE was calculated for all transects that crossed a CLI
feature boundary. The accuracy of detection was evaluated through comparison of the obtained
RMSE with that obtained from a set of randomly distributed points (figure 4.3). The calculated
RMSE values were 30.4322 (C.I. + 2.75m, 95%) for transect boundary locations and 86.6044
(C.I. 8.49m, 95%) for randomly selected boundary locations along individual transect lines.
Mapping of the CLI data was conducted at a scale of 1:50,000, based on data obtained from
multiple sources for which positional error rates have not been stated, leading to the adoption of
an arbitrary positional error rate of 33 38m which takes into account GPS based locational
error. The RMSE of 30.4322 is reflective of this rate of positional error.

55
Figure 4.3: Location of the Canada Land Inventory polygon boundary, detected boundaries, and random
points for transect sc2006082H3.

4.2.2: Earth Observation for Sustainable Development of Forests

The number of feature boundaries was on average 67% greater for the EOSD data when
compared to ground based transect surveys and was significant at the .01 level (t=-8.08, df=34,
P<0.01). The crossing of two boundaries within a distance less than the effective pixel size was
counted as only one boundary. For the EOSD the effective pixel size was 25m. After
accounting for temporal inconsistencies, the RMSE using ground based transect data was 14.93m
(C.I. 1.51m, 95%) and 22.89m (C.I. 2.25m, 95%) for a set of randomly generated boundary
points.

56
Figure 4.4: The location of ground survey and EOSD boundaries for transect SC2007051L3_07

Both RMSE values are below the raster data pixel size of 25m; therefore, it can be stated that
EOSD data is no better at identifying actual landcover boundaries then it is at identifying a
randomly generated set of boundary locations.

To evaluate possible causes for this outcome the relationship between the numbers of ground
based boundaries identified was compared to the number of boundaries indicated in the EOSD
data for individual transects. The relationship between both boundary types can be seen in
Figure 4.5. Using the Pearson product-moment correlation it was indicated that there is a
significant positive association between the number of ground-based and EOSD-based
boundaries identified along transect lines (r = 0.64, d.f. = 35, P < 0.01). This correlation was
further evaluated through the comparison of the number of boundaries detected during ground-
based survey activities and the size of the difference between the number of boundaries
calculated from the EOSD dataset (figure 4.6). The result of a Pearson product-moment

57
correlation indicates a significant positive association between the number of boundaries
detected during ground-based transect surveys and the difference between this number and the
corresponding number of boundaries derived from the EOSD dataset for individual transects (r =
0.42, d.f. = 35, P < 0.05). Thus, the difference between the number of ground-based survey and
EOSD dataset boundaries is dependent on the number of boundaries detected during the ground
based survey of a given transect. One explanation for the observed difference between ground-
based and EOSD boundary detection could be the effect of landscape variability on the
classification of landcover features from remotely sensed data.

Figure 4.5: The number of boundaries detected for ground-based and EOSD-based datasets based on
individual transects.

58
Figure 4.6: The difference between the number of boundaries detected in the EOSD data based on the
number of boundaries detected during ground-based surveys for individual transect lines

The classification of landcover features from remotely sensed data can be influenced by the
scale at which individual features occur and the spectral properties of those features (Fisher
1998, Cracknel 1998). The resolution of available datasets was taken into account by restricting
the recording of landcover features, during ground-based surveys, only if they were greater than
10m in width. This would eliminate the recording of landcover features which would have been
impossible to detect with datasets used in this study. The 25m resolution of the EOSD dataset
could pose problem during classification of landcover features in areas exhibiting high
variability. High landcover variability coupled with low resolution can lead to the inclusion of
multiple landcover features in individual pixels, as shown in Figure 2.2. The occurrence of
mixed pixels can lead to a high degree of spectral swamping, especially when landcover features
differ widely in their spectral signatures and the transition between features are abrupt.

During the completion of ground based surveys it was noted that ecotones occurring between
landcover features were narrow, in most cases will below the 10m recording cut-off width. This
was coupled with transitions between landcover features with very distinct spectral signatures,

59
such as, forest-bog, bog-water, or shrub-lichen. The combination of high variability in both
landcover features and their spectral signatures can produce a wide range of pixel values. This in
turn would result in both a high number of pixels with distinct pixel values over short distances
and the classification of landcover features restricted to one pixel in size. Both of these
conditions exist in the EOSD data and are evident in Figure 4.4. With the existence of such
features the degree of data smoothing becomes critical, especially where single or small sets of
pixels, often referred to noise, are created through misclassification (Bradshaw and Fortin 2000).

4.2.3: Provincial Forest Inventory


The Provincial forest inventory represents the most temporally current dataset available for the
island of Newfoundland. It is maintained by the Department of Natural Resources and was
compiled from air photo interpretation, air and ground based surveys and available ancillary
data. Classification is based on the Data Dictionary for District Library as published by the
Provincial Department of Natural Resources (2007).

There was a significant difference between the number of landcover boundaries detected during
the ground based surveys and those occurring in the forest inventory dataset (t=-2.344, df=27,
P<0.05). This does not imply that the forest inventory dataset failed to identify boundaries
plotted during the land based surveys. Overall there was a 11% increase in the number of
boundaries over those found during the ground based surveys. The relationship between the
forest inventory data and the boundaries identified during completion of a ground based transect
line can be seen in figure 4.7.

Given that boundary detection was different between ground based surveys and those portrayed
on forest inventory maps an evaluation of the positional accuracy of those boundaries was
undertaken. This accuracy was compared to a set of randomly generated boundary points,
equivalent to the number of boundaries detected during the ground based survey of individual
transects. It should also be noted that boundaries for features not included in the forest inventory
dataset, such as roads, were removed before the analysis was completed.

60
The results of the evaluation produced a RMSE of 39.47 (C.I. 1.82m, 95%) and 64.71 (C.I.
2.88m, 95%) for the ground based survey and random points respectively. These results must be
viewed within the context that the positional error inherent in the inventory dataset is 30m plus
a positional error of 5m for handheld GPS units thus it would not be uncommon to have a
boundary positional error of 35m. This number corresponds to the error of 39.47m calculated
based on the ground based transect data. The error of 64.71m derived from the evaluation of
randomly placed points along the transect represents a 64% increase in positional error over that
achieved using the transect dataset. Thus it can be concluded that the forest inventory dataset
can be used for the delineation and identification of landcover features utilized by female
caribou.

A review of the provincial forest inventory dataset completed by McLaren and Mahoney (2001)
identified limitations in the delineation of landcover features in areas that have a non-commercial
potential. These limitations involve the inclusion of features in a specific classification even
though it may form a substantial component of another landcover type. Issues of this nature
were also noted for the classification of scrub, where bogs had a tendency to be placed within
this category in areas with a large ericaceous cover. This was confirmed during a visual
assessment directed at identifying areas, along with the underlying landcover classifications,
where the ground truth data and the forest inventory dataset differed. The evaluation suggested
that landcover features, other than commercial forests, were not delineated on the scale use for
forest stands resulting in a blending of features in these areas. The opposite effect was seen in
areas comprised of commercial stands where delineation was resulted in the differentiation of
individual components of contiguous forest stands based on composition, size, density, or age.
Ground truth data would have produced a listing of fewer landcover features for these areas since
the differences in forest stand type was not recorded with this level of detail.

61
A

Figure 4.7: Boundaries detected during a ground based survey of transect SC2007095L1 overlain on
the boundaries found in the Provincial Forest Inventory.

A review of figure 4.8 will show another limitation involved in using Provincial forest inventory
for the identification and delineation of caribou landcover usage and concerns the spatial
coverage of the dataset. The lack of coverage is not confined to the study area but occurs at
various sites across the province. The areas excluded are often void of commercial forests but
nonetheless represent important areas for caribou. This has led to the need for the identification
of a spatial data set that could be used to fill the gaps inherent in the forest inventory, thus the
inclusion of the Landsat 7 ETM+ dataset in this study. An evaluation of this dataset can be
found in the next section.

62
Figure 4.8: Forest Inventory coverage of the area included in this study.

4.2.4 Landsat ETM+ Segmentation File Evaluation

The number of feature boundaries was on average 40% greater for the segmented landsat image
when compared to ground based transect data and was significant at the .01 level (t=-9.05,
df=27, P<0.01). During enumeration the crossing of two boundaries within a distance less than
the effective pixel size was counted as only one boundary. For the pan-sharpened landsat
imagery the effective pixel size was 15m. A portion of the boundary discrepancy was a result of
the difficulty to create adequate segmentation in areas covered by bogs and mires which required
a reduction in the size of the scaling factor and a small reduction in the merge value utilized. A
consequence of this was the creation of additional segments in forested areas leading to an
increase in the number delineated boundaries. Most of these additional segments would be
eliminated upon completion of either the rule or supervised based classification of the objects
identified. Given that the aim of this study was the identification of boundaries, originally
63
identified during ground-based surveys, using selected datasets, the final step of classification
was not required.

After accounting for temporal inconsistencies and errors of omission, the RMSE using ground
based transect data was 27.92m (C.I. 1.14m, 95%) and 37.02m (C.I. 1.52m, 95%) for a set of
randomly generated boundary points. The landsat image had a positional error of 25m and
when combined with the GPS error of 5m creates an error rate comparable to the RMSE error
obtained for the segmented image. The RMSE obtained during the comparison with ground
based survey results is within the 30m combined positional error rate thus represents a
confirmation that landsat imagery and segmentation can be used to delineate existing and former
landscape boundaries. This fact is further supported by the 24.6% difference between the
RMSE obtained for ground based survey points and the set of randomly generated boundary
points.

The implementation of a rule or supervised classification scheme would reduce the number of
boundaries in the segmented Landsat image. This would increase the difference between the
ground based and random RMSE results, since most of the segments occurring inside contiguous
landcover features would be eliminated, thereby increasing the discrepancy between the location
of random point and segments. Given these results, segmented, pansharpened, Landsat ETM+
imagery can be used to complement the Provincial Forestry Inventory dataset for the delineation
and classification of caribou habitat on the island of Newfoundland.

4.3: Fulfilment of Aims and Objectives


The aims and objectives of this study were centered on three main tasks; the identification and
selection of spatial datasets, the selection of methods for the evaluation of selected datasets that
will allow for the identification of landcover boundaries, and the evaluation of these datasets
utilizing a representative sample of boundaries occurring in areas utilized or avoided by caribou.
Given the financial constraints imposed on many government and non-government organization
only datasets that were freely available were selected for testing.

64
Upon selection of the four datasets included in this study methods for their evaluation were
explored. This consisted of selecting a method to obtain a representative sample of landcover
boundaries in areas utilized or avoided by caribou and a mechanism to compare this dataset to
those selected. The use of the space-time scan statistics provided an unbiased means of
identifying representative areas utilized by caribou with avoided areas being represented by
maximum step length calculations. The comparison of the representative sample and selected
datasets was completed through the calculation of RMSE values.

The outcome of this evaluation is summarized in Table 4.1.

Table 4.1: RMSE values for datasets evaluated in this study. All RMSE values should be interpreted within
the context of the positional error of individual datasets.
Forest Landsat
Dataset CLI EOSD
Inventory ETM+
Transect 30.43 14.93 39.47 27.92
Random 86.60 22.89 64.71 37.02
Positional
35 25 35 30
Error(m)

The CLI, forest inventory and Landsat datasets have been shown as useful for the identification
and delineation of landcover boundaries associated with areas utilized by caribou. Each of these
data sources must be used within the constraints inherent in each of the datasets. All datasets
suffer from temporal inconsistency and as a result errors of omission or commission. This leads
to the need for the use of ancillary data to insure a complete delineation of all features in a
specific area.

The CLI dataset is prone to a high degree of generalization and as such should only be used for
delineation of areas utilized by caribou at the provincial scale and only if other datasets are not
available. One important result obtained from the evaluation of the CLI dataset was the fact that
landcover boundaries were detected during ground based surveys that were evident during the
construction of the dataset. This outcome indicates the permanence of landcover boundaries and
could provide a baseline dataset for landscape change studies relating to caribou habitat.

65
Results of the RMSE evaluation for the forest inventory dataset indicate that it can and should be
used for the delineation of landcover features. In some cases, such as the delineation of forest
stands, some level of generalization may be warranted. The incompleteness of the forest
inventory dataset precludes its usage across the province and demonstrates the need for a
complementary data source such as classified Landsat ETM+ imagery.

The Landsat ETM+ dataset can be used to complement the forest inventory dataset in those areas
currently not covered by the inventory. Although RMSE values indicate that the landsat data can
be used to identify existing landcover boundaries at an acceptable level of accuracy, the issue of
a 40% difference in the number boundaries when compared to the ground based survey must be
addressed. During the segmentation exercise it was noted that scaling and merging levels that
produce the best results for forested areas often failed to produce adequate segmentation in areas
represented by bogs and mires. In this study a compromise was made during the selection of
scaling and merging values such that the level of accuracy within forested areas was sacrificed
for a better delineation in bogs and mires. It is recognized that the completion of either rule
based of supervised based classification will be able to compensate for this discrepancy.

It is not possible to recommend the use of the EOSD dataset for the delineation of caribou habitat
unless it is subjected to reclassification or generalization. Since the dataset is supplied only as a
platted geotiff file, reclassification will not be possible without access to the original pre-
classified files. Generalization of the existing files may produce acceptable results for restricted
areas of the island.

66
CHAPTER 5

CONCLUSIONS

5.1: Conclusions

Through the use of space-time scan statistics, fibonnaci sampling, segmentation and RMSE
evaluations it was determined that the Provincial Forest Inventory and Landsat ETM+ data can
be used to delineate boundaries associated with landcover features in areas utilized or avoided
during calving and post-calving on the island of Newfoundland. The CLI dataset can be used to
provide information on the location of permanent boundaries occurring at the landscape scale
within the range utilized by caribou. Use of the EOSD dataset cannot be recommended based on
the inability to delineate ground based boundaries at the scale utilized by this study. An attempt
was made to generalize the dataset such that it would better reflect the ground based location of
boundaries but satisfactory results were not achieved.

The methods developed to provide representative sampling have eliminated bias that could be
introduced by both the selection of a study area, and the influence of directionality inherent in
some landcover features. This was accomplished by allowing individual caribou to select the
area of both utilization and avoidance through the use of space-time scan statistics, and non-
linear transect sampling.

This study represents only a preliminary step in the identification and delineation of landcover
features important to caribou. The results of this study and data obtained during the ground
based transect survey must now be used to provide a habitat based landcover map for caribou on
the island of Newfoundland. This can be achieved through completion of the items outlined in
the next section.

67
5.2: Recommendations
To ensure the completion of knowledge based habitat mapping for caribou during calving and
post-calving the following activities are recommended:

1. A combination of both forest inventory and landsat data should be utilized to provide
complete coverage of the island.
2. Ancillary data (roads, waterways, water bodies, rightaways, etc) must be incorporated in
all delineation activities
3. Logistic regression must be completed for data collected during the ground based survey
to identify landcover features utilized or avoided by caribou.
4. Reclassification of the forest inventory and classification of the segmented landsat
datasets must be based on the logistic regression results.
5. The completed knowledge based habitat map is to be used for the identification of
landcover features utilized by caribou during other periods in the yearly caribou lifecycle.
6. The spatial relationship between landcover features in areas utilized or avoided by
caribou must be quantified and the results applied to classification activates conducted on
spatial datasets representing the island of Newfoundland.

68
APPENDICES
Appendix A

Cluster Assessment Form


Cluster No._____ Coordinates (WGS84):_______________
________________ Date:_____________ Observer:__________________
Animal Id:____________ Elevation:________ Slope:_____ Radius:_______
Terrain Profile:______ Site Classification:_____ Feature
Orientation:__________

Comments:_________________________________________________________
__________________________________________________________________
__________________________________________________________________
__________________________________________________________________
__________________________________________________________________

69
Transect Line Intersections:
Lag: Dis: N: W: Site Class: TR: FS No.:
Lag: Dis: N: W: Site Class: TR: FS No.:
Lag: Dis: N: W: Site Class: TR: FS No.:
Lag: Dis: N: W: Site Class: TR: FS No.:
Lag: Dis: N: W: Site Class: TR: FS No.:
Lag: Dis: N: W: Site Class: TR: FS No.:
Lag: Dis: N: W: Site Class: TR: FS No.:
Lag: Dis: N: W: Site Class: TR: FS No.:
Lag: Dis: N: W: Site Class: TR: FS No.:
Lag: Dis: N: W: Site Class: TR: FS No.:
Lag: Dis: N: W: Site Class: TR: FS No.:
Lag: Dis: N: W: Site Class: TR: FS No.:
Lag: Dis: N: W: Site Class: TR: FS No.:
Lag: Dis: N: W: Site Class: TR: FS No.:
Lag: Dis: N: W: Site Class: TR: FS No.:
Lag: Dis: N: W: Site Class: TR: FS No.:
Lag: Dis: N: W: Site Class: TR: FS No.:
Lag: Dis: N: W: Site Class: TR: FS No.:
Lag: Dis: N: W: Site Class: TR: FS No.:
Lag: Dis: N: W: Site Class: TR: FS No.:
Lag: Dis: N: W: Site Class: TR: FS No.:
Lag: Dis: N: W: Site Class: TR: FS No.:
Lag: Dis: N: W: Site Class: TR: FS No.:
Lag: Dis: N: W: Site Class: TR: FS No.:
Lag: Dis: N: W: Site Class: TR: FS No.:
Lag: Dis: N: W: Site Class: TR: FS No.:
Lag: Dis: N: W: Site Class: TR: FS No.:
Lag: Dis: N: W: Site Class: TR: FS No.:
Lag: Dis: N: W: Site Class: TR: FS No.:
Lag: Dis: N: W: Site Class: TR: FS No.:

70
Forest Stand Evaluation:

Species Comp.: Age Class: Height Class: Crown Closure: Site


Disturbance: Understory:

FS No.: SC: AC: HC: CC: SD: US:


FS No.: SC: AC: HC: CC: SD: US:
FS No.: SC: AC: HC: CC: SD: US:
FS No.: SC: AC: HC: CC: SD: US:
FS No.: SC: AC: HC: CC: SD: US:
FS No.: SC: AC: HC: CC: SD: US:
FS No.: SC: AC: HC: CC: SD: US:
FS No.: SC: AC: HC: CC: SD: US:
FS No.: SC: AC: HC: CC: SD: US:
FS No.: SC: AC: HC: CC: SD: US:
FS No.: SC: AC: HC: CC: SD: US:
FS No.: SC: AC: HC: CC: SD: US:
FS No.: SC: AC: HC: CC: SD: US:
FS No.: SC: AC: HC: CC: SD: US:
FS No.: SC: AC: HC: CC: SD: US:
FS No.: SC: AC: HC: CC: SD: US:
FS No.: SC: AC: HC: CC: SD: US:
FS No.: SC: AC: HC: CC: SD: US:
FS No.: SC: AC: HC: CC: SD: US:
FS No.: SC: AC: HC: CC: SD: US:
Comments:_______________________________________________________
________________________________________________________________
________________________________________________________________
________________________________________________________________
________________________________________________________________
______________

71
REFERENCES

Addicott J.F., Aho J.M., Antolin M.F., Padilla D.K., Richardson J.S. and Soluk D.A. (1987)
Ecological Neighborhoods: Scaling Environmental Patterns. Oikos, Vol. 49, pp. 340-346.

Affleck D.L.R., Gregoire T.G. and Valentine H.T. (2005) Design Unbiased Estimation in
Line Intersect Sampling Using Segmented Transects. Environmental and Ecological
Statistics, Vol. 12, pp. 139-154.

Anderson J.R., Hardy E.E., Roach J.T and Witmer R.E. (1976) A Land Use and Landcover
Classification System for Use with Remote Sensor Data. Geological Survey Professional
Paper 964.

Anderson R.B., Dzus E. and Boutin S. (1999) Using Selection to Define Crucial Caribou
Habitat: A Case for Caution. In: Proceedings Biology and Management of Species and
Habitats at Risk, Kamloops, British Columbia, Canada, February 15-19, 1999.

Andrefouet S., Robinson J.A., Hu C., Feldman G.C., Salvat B., Payri C, and Muller-Karger
F.E. (2003) Influence of the Spatial Resolution of SeaWiFS, Landsat-7, Spot, and
International Space Station Data in Estimates of Landscape Parameters of Pacific Ocean
Atolls. Canadian journal of Remote Sensing, Vol. 29, No.2, pp. 210-218.

Angelstam P.K., Butler R., Lazdinis M., Mikusinski G. and Roberge J. (2003) Habitat
Thresholds for Focal Species at Multiple Scales and Forest Biodiversity Conservation
Dead Wood as an Example. Annales Zooligical Fennici, Vol. 40, pp. 473-482.

Aniello P. (2003) Get Centroid ArcScript. [Internet] ESRI Support Center, Your Online
Technical Resource. Available at: < http://arcscripts.esri.com/details.asp?dbid=12781>
[Accessed November 25, 2008]

Apps C.D., McLellan B.N., Kinley T.A. and Flaa J.P. (2001) Scale-Dependent Habitat
Selection by Mountain Caribou, Columbia Mountains, British Columbia. The Journal of
Wildlife Management, Vol. 65, No. 1, pp. 65-77.

Baldi A. and McCollin D. (2003) Island Ecology and Contingent Theory: The Role of Spatial
Scale and Taxonomic Bias. Global Ecology and Biogeography, Vol. 12, pp. 1-3.

Banfield A.W.F. (1974) The Mammals of Canada. University of Toronto Press, Toronto,
Canada, 438 pages.

Barten N.L., Bowyer R.T. and Jenkins K.J. (2001) Habitat Use by Female Caribou:
Tradeoffs Associated with Parturition. Journal of wildlife Management, Vol. 65, No. 1, pp.
77-92.

Bell T. and Renouf M.A.P. (2004) Prehistoric Cultures, Reconstructed Coasts: Maritime
Archaic Indian Site Distribution in Newfoundland. World Archaeology, Vol. 35, No. 3, pp.
350-370.

72
Benson B.J. and Mackenzie M.D. (1995) Effects of Sensor Resolution on Landscape
Structure Parameters. Landscape ecology, Vol. 10, No. 2, pp. 113-120.

Bergerud A.T. (1971) The Population Dynamics of Newfoundland Caribou. Wildlife


Monographs, No. 25, pp. 3-55.

Bergerud A.T. (1974a) Decline of Caribou in North America Following Settlement. The
Journal of Wildlife Management, Vol. 38, No. 4, pp 757-770.

Bergerud A.T. (1974b). The Role of the Environment in the Aggregation, Movement and
Disturbance Behaviour of Caribou. Pages 552-584 in V. Geist and F. Walther, eds. The
Behaviour of Ungulates and its Relation to Management. Int. Union Conserv. Naturea nd
Nat. Resour. Publ.N ew Ser.2 4, Morges, Switz

Bergerud A.T. (1985) Antipredator Strategies of Caribou: Dispersion along Shorelines.


Canadian Journal of Zoology, Vol.63, pp. 1324-1329.

Bergerud A.T., Butler H.E. and Miller D.R. (1984) Antipredator strategies of caribou:
Dispersion in mountains. Canadian Journal of Zoology, Vol. 62, pp. 1566-1575.

Beyer, H. L. 2004. Hawth's Analysis Tools for ArcGIS. [Internet] SpatialEcology.Com.


Available at: http://www.spatialecology.com/htools [Accessed November 25, 2008]

Bock M., Xofis P., Mitchley J., Rossner G. and Wissen M. (2005) Object-oriented Methods
for Habitat Mapping at Multiple Scales Case Studies from Northern Germany and Wye
Downs, UK. Journal for nature Conservation, Vol. 13, pp. 75-89.

Boyce M.S. (2006) Scale for Resource Selection Functions. Diversity and Distributions, Vol.
12, pp 269-276.

Boyd D.S. and Danson F. M. (2005) Satellite Remote Sensing of Forest Resources: Three
Decades of Research Development. Progress in Physical Geography, Vol. 29, No. 1, pp. 1-
26.

Bradshaw G.A. and Fortin M.J. (2000) Landscape Heterogeneity Effects on Scaling and
Monitoring Large Areas Using Remote Sensing Data. Geographic Information Sciences,
Vol. 6, No. 1, pp. 61-68.

Buckland S.T., Borchers D.L., Johnston A., Henrys P.A. and Marques T.A. (2007) Line
Transect Methods for Plant Surveys. Biometrics, Vol. 63, pp. 989-998.

Burnett C. and Blaschke T. (2003) A Multi-Scale Segmentation/Object Relationship


Modelling Methodology for landscape Analysis. Ecological Modelling, Vol. 168, pp. 233-
249.

Canada Centre for Remote Sensing (2007) Fundamentals of Remote Sensing: A Canada
Centre for Remote Sensing Remote Sensing Tutorial. Natural Resources Canada, Canada,
258 pages.

73
Canadian Forest Service (2002) Canadian Large Fire Database (LFDB). [Internet] CFS Fire
Research The Canadian Large Fire Data Base, Natural Resources Canada. Available from:
< http://fire.cfs.nrcan.gc.ca/research/climate_change/lfdb/lfdb_download_e.htm> [Accessed
19 February 2008].

Canfield, R.H. (1941) Application of the line interception method in sampling range
vegetation. Journal of Forestryy, Vol.39, pp. 388-394.

Castilla G. and Hay G.J. (2006) Uncertainties in Land Use Data. Hydrology and Earth
System Science Discussions, Vol. 3, pp. 3439-3472.

Chavez, P.A., Guptill, S.C. and Bowell, J.A. (1984), Image Processing Technique for
Thematic Mapper Data. Proceedings of ASPRS-ACSM Technical Papers, Vol. 2,
pp. 728-42.

Chen Y.F., Yu F.H. and Dong M. (2002) Scale-Dependent Spatial Heterogeneity of


Vegetation in Mu Us Sandy Land, a Semi-Arid Area of China. Plant Ecology, Vol. 162, pp.
135-142.

Chen X., Vierling L. and Deering D. (2005) A Simple and Effective Radiometric Correction
Method to Improve Landscape Change Detection Across Sensors and Across Time. Remote
Sensing of Environment, Vol. 98, pp. 63-79.

Chubbs T.E., Keith L.B., Mahoney S.P. and McGrath M.J. (1992) Responses of Woodland
Caribou (Rangifer tarandus caribou) to Clear-cutting in East-Central Newfoundland.
Canadian journal of Zoology, Vol. 71, pp. 487-493.

Cihlar J. (2000) Landcover Mapping of large Areas from Satellites and Research Priorities.
International Journal of Remote Sensing, Vol. 21, No. 6 and 7, pp. 1093-1114.

Cihlar J., Guindon B., Beaubien J., Latifovic R., Peddle D., Wulder M., Fernandes R. and
Kerr J. (2003) From Need to Product: A Methodology for Completing a landcover Map of
Canada with Landsat Data. Canadian Journal of Remote Sensing, Vol. 29, No. 2, pp. 171-
186.

Cohen W.B. and Goward S.N. (2004) Landsats Role in Ecological Applications of Remote
Sensing. BioScience, Vol. 54, No. 6, pp. 535-545.

Coops N.C., Wulder M.A. and White J.C. (2007) Identifying and Describing Forest
Disturbance and Spatial Pattern: Data Selection Issues and Methodological Implications.
In: Understanding Forest Disturbance and Spatial Pattern: Remote sensing and GIS
Approaches. Editors, Michael Wulder and Steven E. Franklin, CRC Press, USA.

Courtois R., Ouellt J., Breton L., Gingras A. and Dussault C. (2007) Effects of Forest
Disturbance on Density, Space Use, and Mortality of Woodland Caribou. Ecoscience, Vol.
14, No. 4, pp. 491-498.

Cracknel A.P. (1998) Synergy in Remote Sensing Whats in a Pixel? International Journal
of Remote Sensing, Vol. 19, No. 11, pp. 2025-2047.

74
Crete M., Huot J. and Gautheir L. (1990) Food Selection during Early lactation by Caribou
Calving on the Tundra in Quebec. Arctic, Vol. 43, No. 1, pp. 60-65.

Dale M.R.T. (2004) Spatial Pattern Analysis in Plant Ecology. Cambridge University Press,
New York, NY, USA.

Damman A.W.H. (1964) Some Forest Types of Central Newfoundland and their Relation to
Environmental Factors. Dept. of Forestry, Canada, (Society of American Foresters). Forest
Research Branch Contribution No. 596, Forest Science Monograph No. 8, 62 pp.

Davis S.T. (1895) Caribou Shooting in Newfoundland, With a History of Englands Oldest
Colony from 1001 to 1895. The New Era Printing House, Lancaster, PA., 212 pages.

Delcourt H.R., Delcourt P.A. and Webb T. (1983) Dynamic Plant Ecology: The Spectrum of
Vegetational Change in Space and Time. Quaternary Science Reviews, Vol. 1, pp. 153-175.

Denny M.W., Helmuth B., Leonard G.H., Harley D.G., Hunt L.J.H. and Nelson E.K. (2004)
Quantifying Scale in Ecology: Lessons from a Wave-Swept Shore. Ecological Monographs,
Vol. 74, No. 3, pp. 513-532.

Dept. of Natural Resources (2007) Forest Inventory Data Dictionary. Dept. of Natural
Resources, Forestry Division, Newfoundland and Labrador, Canada.

Devereux B.J., Amable G.S. and Costa Posada C. (2004) An Efficient Image Segmentation
Algorithm for Landscape Analysis. International Journal of Applied Earth Observation and
Geoinformation, Vol. 6, pp. 47-61.

Dytham C. (2003) Choosing and Using Statistics, A Biologists Guide. Blackwell Science,
Malden, MA, USA, 248 pages.

Dugmore A.A.R. (1913) The Romance of the Newfoundland Caribou. J. P. Lippincott Co.,
Philadelphia, Pa, USA, 186 pages.

Dwass M. (1957) Modified Randomization Tests for Nonparametric Hypotheses. The Annals
of Mathematical Statistics, Vol. 28, No.1, pp. 181-187.

Dwivedi R.S. and Rao B.R. M. (1992) The Selection of the Best Possible Landsat TM Band
Combination for Delineating Salt-Affected Soils. International Journal of Remote Sensing,
Vol. 13, No. 11, pp. 2051-2058

Edmunds P.J. and Bruno J.F. (1996) The Importance of Sampling Scale in Ecology:
Kilometer-Wide Variation in Coral Reef Communities. Marine Ecology Progress Series, Vol.
143, pp. 165-171.

Edwards T.C., Moisen G.G., Frescino T.S. and Lawler J.J. (2001) Use of Forest Inventory
and Analysis Information in Wildlife Habitat Modeling: A Process for Linking Multiple
Scales. In: Proceedings of the Third Annual Forest Inventory and Analysis Symposium,
Editors, R.E. McRoberts, G.A. Reams, P.C. Van Deusen and W. John. General Technical
Report NC-230. St. Paul, MN: U.S. Department of agriculture, Forest Service, North Central
research Station, pp. 188-194.
75
Elton C.S. and Miller R.S. (1954) The Ecological Survey of Animal Communities: With
Practical System of Classifying Habitats by Structural Characters. The Journal of Ecology,
Vol. 42, No.2, pp. 460-496.

ENVI (2008) ENVI Feature Extration Module Users Guide. Feature Extraction Module
Version 4.6, December, 2008 Edition, ITT Visual Information Solutions.

Environment and Conservation (2008) Five-Year Caribou Strategy Seeks to Address


Declining Populations. [Internet] Government of Newfoundland and Labrador Website.
Available at: http://www.releases.gov.nl.ca/releases/2008/env/0207n06.htm [Accessed April
06,2009].

Environment Canada (1970) The Canada Land Inventory Land Capability Classification
for Wildlife. The Canada Land Inventory, Report No. 7, 30 pages.

Environment Canada (1980) Canada Land Inventory Land Capability Classification for
Wildlife Ungulates Summary Report. The Canada Land Inventory, Report No. 17, 39
pages.

Environment Canada (2004) Canadian Climate Normals 1971 -2000, Buchans


Newfoundland. [Internet] Canadian Climate Normals 1971-2000, The Green Lane,
Environment Canada. Available from:
<http://www.climate.weatheroffice.ec.gc.ca/climate_normals/results_e.html?Province=NFLD
&StationName=&SearchType=&LocateBy=Province&Proximity=25&ProximityFrom=City
&StationNumber=&IDType=MSC&CityName=&ParkName=&LatitudeDegrees=&Latitude
Minutes=&LongitudeDegrees=&LongitudeMinutes=&NormalsClass=A&SelNormals=&StnI
d=6579&> . [Accessed 19 February 2008].
ESRI (2004) ArcGIS 9, Linear Referencing in ArcGIS. Environmental Systems Research
Institute, Inc., 380 New York Street, Redlands, CA, USA.

ESRI (2005) ArcInfo v. 9.1. Environmental Systems Research Institute, Inc., 380 NewYork
Street, Redlands, CA, USA.

ESRI (2006) ArcGis v. 9.2. Environmental Systems Research Institute, Inc., 380 NewYork
Street, Redlands, CA, USA.

Ferguson S.H. and Elkie P.C. (2004) Seasonal Movement Patterns of Woodland Caribou
(Rangifer tarandus caribou). Journal of Zoology London, Vol. 262, pp. 125-134.

Ferguson S.H. and Elkie P.C. (2004a) Habitat Requirements of Boreal Forest Caribou
During the Travel Seasons. Basic and Applied Ecology, Vol. 5, pp. 465-474.

Ferguson R. (1990) Detection and Classification of Muskox Habitat on banks Island,


Northwest Teritories, Canada, Using Landsat Thematic Mapper Data. Arctic, Vol. 44, Supp.
1, pp. 66-74.

76
Fisher P. (1998) The Pixel: A Snare and a Delusion. International Journal of Remote
Sensing, Vol. 18, No. 3, pp. 679-685.

Ford E.D. (2004) Scientific Method for Ecological Research. Cambridge University Press,
Cambridge, UK.

Forestry (2008) Roads 2004 Shapefile. Dept. Natural Resources, Government of


Newfoundland and Labrador, Canada.

Forestry (2000) Cuts_d9 Shapefile. Dept. Natural Resources, Government of Newfoundland


and Labrador, Canada.

Fortin M. and Dale M.R.T. (2005) Spatial Analysis, A Guide for Ecologists. Cambridge
University Press, Cambridge, UK, 365 pages.

Fox L., Garrett M.L., Heasty R. And Torres E. (2002) Classifying Wildlife Habitat with Pan-
Sharpened Landsat 7 Imagery. Proceedings: ISPRS Comission I Mid-Term Symposium in
conjunction with Pecora 15/Land Satellite Information IV Conference10-15 November 2002,
Denver, CO USA

Frank T.D. and Tweddale S.A. (2006) The Effect of Spatial Resolution on the Measurement
of Vegetation Cover in Three Mojave Desert Shrub Communities. Jopurnal of Arid
Environments, Vol. 67, pp. 88-99.
Gao Y and Mas J.F. (2008) A Comparison of the Performance of Pixel Based and Object
Based Classification over Images with Various Spatial Resolutions. Online Journal of Earth
Science, Vol. 2, No, 1, pp. 25-35.
Garshelis D.L. (2000) Delusions in Habitat Evaluation: Measuring Use, Selection and
Importance. In: Research Techniques in Animal Ecology: Controversies and Consequences,
Luigi Boitani and Todd K. Fuller , Eds, Columbia University Press, New York, pp. 111-153.
Ge Y., Li S. and Li D. (2006) New Algorithm for Sub-Pixel Boundary Mapping.
Proceedings: ISPRS Technical Commission II Symposium, Vienna July 12-14, 2006.
GeoBase (2008) Landsat 7 Orthorectified Imagery over Canada, Level 1Product
Specifications. Edition 1.1. Government of Canada, Natural Resources Canada, Centre for
Topographic Information, Quebec, Canada.
George T.H., Stringer J.W. and Baldridge J.N. (1978) Reindeer Range Inventory in Western
Alaska from Computer-aided Digital Classification of Landsat Data. In: Proceedings 11th
International Symposium of Remote Sensing of the Environment, pp. 671-682.
Glenn E.M. and Ripple W. J. (2004) On Using Digital Maps to Assess Wildlife Habitat.
Wildlife Society Bulletin, Vol. 32, No. 3, pp. 852-860.
Golley F.B. (1989) Paradigm Shift. Landscape Ecology, Vol. 3, No. 2, pp. 65-66.
Goodall D.W. (1963) Pattern Analysis and Minimal Area-Some Further Comments. The
Journal of Ecology, Vol. 51, No. 3, pp. 705-710.

77
Gottschalk T.K., Huettmann F. and Ehlers M. (2005) Thirty Years of Analysing and
Modelling Avian Habitat Relationships Using Satellite Imagery Data: A Review.
International Journal of Remote Sensing, Vol. 26, No. 12, pp. 2631-2656.
Gregoire T.G., Affleck D.L.R. and Valentine H.T. (2004) Conditioning Inference on Line
Orientation in Line Intersect Sampling, SamNordisk Skogsforskning, Inventory Working
Group, Forest Inventory, Planning, and Modelling, Sjusjoun, Norway, September 6-8, 2004.
Greig-Smith P. (1952) The Use of Random and Contiguous Quadrats in the Study of the
Structure of Plant Communities. Annals of Botany, N.S., Vol. XVI, No. 62, pp. 293-316.
Griffith J.A. (2004) The Role of Landscape Pattern Analysis in Understanding Concepts of
Land Cover Change. Journal of geographical Sciences, Vol. 14, No. 1, pp. 3-17.
Gustafson E.J (1998) Quantifying Landscape Spatial Pattern: What is the State of the Art?
Ecosystems, Vol. 1, pp 143-156.
Gustine D.D., Parker K.L., Lay R.J., Gillingham M.P. and Heard D.C. (2006) Interpreting
Resource Selection at Different Scales for Woodland Caribou in Winter. The Journal of
Wildlife Management, Vol. 70, No. 6, pp. 1501-1614.
Hansen M.J., Franklin S.E., Woudsma C.G. and Peterson M. (2001) Caribou Habitat
Mapping and Fragmentation Analysis Using Landsat MSS, TM, and GIS Data in the North
Columbia Mountains, British Columbia, Canada. Remote Sensing of Environment, Vol. 77,
pp. 50-65.
Hutchinson G.E. (1953) The Concept of Pattern in Ecology. Proceedings ot the Academy of
Natural Sciences of Philadelphia, Vol. 105, pp. 1-12.
Innes J.L. and Koch B. (1998) Forest Biodiversity and Its Assessment by Remote Sensing.
Global Ecology and Biogeography Letters, Vol. 7, No.6, pp. 397-419.

Jakubauskas M.E. and Price K. P. (1994) Landsat Thematic Mapper Characterization of


Coniferous Forest Succession. Proceedings: 1994 ASPRS-ACSM Annual Meeting, Reno,
Nevada, USA.

Jentsch A., Beierkuhnlein C. and White P.S. (2002) Scale, the Dynamic Stability of Forest
Ecosystems, and the Persistence of Biodiversity. Silva Fennica, Vol. 36, No. 1, pp. 393-400.

Johnson C.J., Parker K.L. and Heard D.C. (2001) Foraging Across a Variable Landscape:
Behavioral Decisions Made by Woodland Caribou at Multiple Scales. Oecologia, Vol. 127,
pp. 590-602.
Johnson C.J., Parker K.L., Heard D.C. and Gillingham M.P. (2002a) Movement Parameters
of Ungulates and Scale-Specific Responses to the Environment. Journal of Animal Ecology,
Vol. 71, pp. 225-235.
Johnson C.J., Parker K.L., Heard D.C. and Gillingham M.P. (2002) A Multiscale Approach to
Understanding the Movements of Woodland Caribou. Ecological Applications, Vol. 12, No.
6, pp. 1840-1860.
Johnson C.J., Alexander N.D, Wheate R.D. and Parker K.L. (2003) Characterizing Woodland
Caribou Habitat in Sub-Boreal and Boreal Forests. Forest ecology and Management, Vol.
180, pp. 241-248.

78
Johnson C.J., Seip D.R. and Boyce M.S. (2004) A Quantitative Approach to Conservation
Planning: Using Resource Selection Functions to Map the Distribution of Mountain Caribou
at Multiple Scales. Journal of Applied Ecology, Vol. 41, pp. 238-251.
Johnson C.J., Parker K.L., Heard D.C. and Seip D.R. (2004a) Movements, Foraging Habits,
and Habitat Use Strategies of Northern Woodland caribou During Winter: Implications for
Forest Practices in British Columbia. BC Journal of Ecosystems and Management, Vol. 5,
No. 1, pp. 22-35.
Joly K., Dale B.W., Collins W.B. and Adams L.G. (2003) Winter Habitat Use by Female
Caribou in Relation to Wildland Fires in Interior Alaska. Canadian Journal of Zoology, Vol.
81, pp. 1192-1201.
Jones E.S., Gillingham M.P., Seip D.R. and Heard D.C. (2006) Comparison of Seasonal
Habitat Selection Between Threatened Woodland Caribou Ecotypes in Central British
Columbia. Proceedings: The Eleventh North American Caribou Workshop, Jasper, Alberta,
Canada, April 24-27, 2006.
Ju J., Gopal S. and Kolaczyk E.D. (2005) On the Choice of Spatial and Catagorical Scale in
Remote Sensing Landcover Classification. Remote Sensing of Environment, Vol. 96, pp. 62-
77.
Kalikhman I. (2007) Patchi Distribution Fields: ASpiral Design and Reconstruction
Adequacy. Environmental Monitoring and Assessment, Vol. 124, pp.243-252.
Kark S. and van Rensburg B.J. (2006) Ecotones: Marginal or Central Areas of Transition?
Israel Journal of Ecology and Evolution, Vol. 52, pp. 29-53.
Kerr J.T. and Ostrovsky M. (2003) From Space to Species: Ecological Applications for
Remote Sensing. Trends in Ecology and Evolution, Vol. 18, No.6, pp. 299-305.
Khun T.S. (1962) The Structure of Scientific Revolutions. Second Edition 1970, University
of Chicago Press,
Kulldorff M. and Information Management Services, Inc (2006). SaTScanTM v7.0: Software
for the spatial and space-time scan statistics. [Internet] SaTScan - Software for the spatial,
temporal, and space-time scan statistics, Available at:< http://www.satscan.org/> , <Accessed
November 25,2008>.

Kulldorff M., Heffernan R., Hartman J., Assuncao R. and Mostashari F. (2005) A Space-Time
Permutation Scan Statistic for Disease Outbreak Detection. PLoS Medicine, Vol. 2 No. 3,
pp. 216-224.

Kulldorff M. (2009) SaTSanTM User Guide for Version 8.0. [Internet] SaTScan - Software for
the spatial, temporal, and space-time scan statistics, Available at :< http://www.satscan.org/>
, <Accessed March 20,2009>.

Kushwaha S.P.S. and Roy P.S. (2002) Geospatial Technology for Wildlife Habitat
Evaluation. Tropical Ecology, Vol. 43, No. 1, pp. 137-150.

Lamonaca A., Corona P. And Barbati A. (2008) Exploring Forest Structure Complexity by
Multi-Scale Segmentation of VHR Imagery. Remote Sensing of Environment, Vol. 112, pp.
2839-2849.

79
Lefsky M.A., Cohen W.B. and Spies T.A. (2001) An Evaluation of Alternate Remote Sensing
Products for Forest Inventory, Monitoring, and Mapping of Douglas-fir Forests in Western
Oregon. Canadian Journal of Forest Research, Vol. 31, No. 1, pp. 78-87.

Levin S.A. (1992) The Problem of Pattern and Scale in Ecology: The Robert H. MacArthur
Award Lecture. Ecology, Vol. 73, No. 6, pp. 1943-1967.

Lewinski S, (2006) Object-Orientated Classification of Landsat ETM+ Satellite Image.


Journal of Water and Land Development, No. 10, pp. 91-106.

Liam N., Catts D., Quattrochi D., Brown D. and McMaster R. (2004) Scale. In: A Research
Agenda for geographic information Science. Editors, Eddy Lynn Usery and Robert Brainerd
McMaster, CRC Press, USA, 402 pages.
Loveland T.R., Gallant A.L. and Vogelmann J.E. (2005) Perspectives on the Use of Land-
Cover Data for Ecological Studies. In: Wiens J. A. and Moss M.R. (Ed.) Issues and
Perspectives in Landscape Ecology, pp. 120-128, Cambridge University Press, New York.

Lu D. And Weng Q. (2007) A survey of Image Classification Methods and Techniques for
Improving Classification Performance. International journal of Remote Sensing, Vol. 28,
No. 5, pp. 823-870.

Luick B.R. and White R.G. (1986) Oxygen Consumption for Locomotion by Caribou Calves.
Journal of Wildlife Management, Vol. 50, No. 1, pp. 148-152.

Lutes D.C. (2002) Assessment of the Line Transect Method: An Examination of the Spatial
Patterns of Down and Standing Dead Wood. USDA Forest Service General Technical
Report, PSW-GTR-181.

Leyequien E., Verrelst J., Slot M., Schaepman-Strub G., Heitkonig I. M. A. and Skidmore A.
(2007) Capturing the Fugitive: Applying Remote Sensing to Terrestrial Animal Distribution
and Diversity. International Journal of Applied Earth Observation and Geoinformation, Vol
9, pp. 1-20.

Mack R.N. and Harper J.L. (1977) Interference in Dune Annuals: Spatial Pattern and
Neighbourhood Effects. The Journal of Ecology, Vol. 65, No.2, pp. 345-363.

Mahoney S.P. (2000) A Synthesis and Interpretation of the Biology of Woodland Caribou on
the Island of Newfoundland. Wildlife Division, Government of Newfoundland and Labrador.

Mahoney S.P. and Schaefer J.A. (2002) Hydroelectric development and the Disruption of
Migration in Caribou. Biological Conservation, Vol 107, pp. 147-153.

Mahoney S.P. and Virgl J.A. (2003) Habitat Selection and Demography of a Nonmigratory
Woodland Caribou Population in Newfoundland. Canadian journal of Zoology, Vol. 81, pp.
321-334.

Makido Y. and Shortridge A. (2005) Landcover Mapping at Sub-Pixel Scales: Unravelling


the Mixed Pixel. In: Preceedings of the 8th International Conference on GeoComputation,
University of Michigan, USA, July 31 August 3, 2005.
80
Manly B.F.J. (2002) Estimating a Resource Selection Function with Line Transect Sampling.
Journal of Applied Mathematics and Decision Sciences, Vol. 6, No. 4, pp. 213-228.3

Manly B.F.J., McDonald L.L., Thomas D.L., McDonald T.L. and Erickson W.P. (2004)
Resource Selection by Animals, STATISTICAL Design and Analysis of Field Studies. Kluwer
Academic Publishers, New York, USA, 221 pages.

Marj T., Devis T. Rafael T. and Frederic R. (2006) Forest Fires Cluster Detection with
Space-Time Scan Statistics. 4th Swiss Geoscience Meeting, Bern, Switzerland, November 24
25, 2006.

Marks C. (2008) AML Script fot the Creation of Fibonacci Spirals Based on Point Locations.
Department of Environment and Conservation, Wildlife Division, Corner Brook, NL,
Canada.

Mather P.M. (2004) Computer Processing of remotely-Sensed Images An Introduction. 3rd


Edition. John Wiley and Sons Ltd., The Atrium, West Sussex, England, 339 pages.

Mayor S.J., Schaefer J.A., Schneider D.C. and Mahoney S.P. (2009) The Spatial Structure of
Habitat Selection: A Caribous-Eye-View. Acta Oecologica, Vol. 35, pp. 253-260.

McLaren B.E. and Mahoney S.P. (2001) Comparison of Forestry-Based Remote Sensing
Methodologies to Evaluate Woodland Caribou Habitat in Non-Forested Areas of
Newfoundland. The Forest Chronicle, Vol. 778, No. 5, pp. 866-873.

McLoughlin P.D., Dzus E., Wynes B. and Boutin S. (2003) Declines in Populations of
Woodland Caribou. The Journal of Wildlife Management, Vol. 67, No. 4, pp. 755-761.

Meade R. (1974) A Test for Spatial Pattern at Several Scales Using Data from a Grid of
Contiguous Quadrats. Biometrics, Vol. 30, pp. 295-307.

Miline B.T., Johnston K.M. and Forman R.T.T. (1989) Scale-Dependent Proximity of
Wildlife Habiata in a Spatially-Neutral Bayesian Model. Landscape Ecology, Vol. 2, No. 2,
pp. 101-110.

Morrison M.L., Block W.M. Strickland M.D. and Kendall W.L. (2001) Wildlife Study
Design. Springer, New York, USA, 210 pages.

Muraoka H. and Koizumi H. (2009) Satellite Ecology (SATECO)-Linking Ecology, Remote


Sensing and Micrometeorology, from Plot to regional Scale, for the Study of Ecosystem
Structure and Function. Journal of Plant Research, Vol. 122, pp. 3-20.

Nams V.O. (2005) Using Animal Movement Paths to Measure Response to Spatial Scale.
Oecologia Vol. 143, pp. 179-188.

Newton A.C. (2007) Forest Ecology and Conservation, A Handbook of techniques. Oxford
University Press Inc., New York, NY, USA, 454 pages.

81
Oruc M., Marangoz A.M. and Buyuksaliha G. (2004) Comparison of Pixel-based and
Object-Oriented Approaches using LANDSAT-7 ETM Spectral Bands. ZKU, Engineering
Faculty, Zonguldak, Turkey.

Parks Division (2000) Central Newfoundland Forest, North-central Subregion, 2a. [Internet]
Dept. Environment and Conservation, Parks Division, Newfoundland, Canada,4pp. Avaiable
from:
< http://www.env.gov.nl.ca/parks/library/pdf/Ecoregions/Island_2a_north-central%20-
%202007.pdf> [Accessed 17 February 2008].

Ponzoni F.J, Galvao L.S. and Epiphanio C.N. (2002) Spatial Resolution Influence on the
Identification of Landcover Classes in the Amazon Environment. Annals of the Brazilian
Academy of Sciences, Vol. 74, No. 4, pp. 717-725.

Ramsey R. D., Wright Jr. D. L. and McGinty C. (2004) Evaluating the Use of Landsat 30m
Enhanced Thematic Mapper to Monitor Vegetation Cover in Shrub-Steppe Environments.
Geocarto International, Vol. 19, No. 2, pp. 39-47

Ranson K.J., Kovacs K., Sun G. and Kharuk V.I. (2003) Disturbance Recognition in the
Boreal Forest Using Radar and Landsat-7. Canadian journal of Remote Sensing, Vol. 29,
No. 2, pp. 271-285.

Recovery Implementation Group, Hart and Cariboo Mountains (2005) Recovery


Implementation Plan for Threatened Woodland Caribou (Rangifer tarandus caribou) in the
Hart and Cariboo Mountains Recovery Area, British Columbia. British Columbia, Canada.

Renick N.J. and Grebner D.L. (2002) Preliminary Study on the Effects Inaccurate Area
Estimation has on Harvest Scheduling Using Different Image Resolutions. Journal Article
No. FO212, Mississippi State University, USA.

Rettie W.J. and Messier F. (2001) Range Use and Movement Rates of Woodland Caribou in
Saskatchwan. Canadian Journal of Zoology Vol. 79, pp. 1933-1940.

Rettie W.J and Messier F. (2000) Hierarchical Habitat Selection by Woodland Caribou: Its
Relationship to Limiting Factors. Ecography, Vol. 23, No. 4, pp. 466-478.

Richards J. A. and Jia X. (2006) Remote sensing Digital Image Analysis. 4th Edition,
Springer Berlin Heidelberg, New Yory, USA, 439 pages.

Ritters K.H., ONeill R.V. and Jones K.B. (1997) Assessing Habitat Suitability at Multiple
Scales: A Landscape-Level Approach. Biological Conservation, Vol. 81, pp. 191-202.

Robinson D.J., Redding N.J. and Crisp D.J. (2002) Implementation of a Fast Algorithm for
Segmenting SAR Imagery. Scientific and Technical Report, 01 January 2002. Australia:
Defense Science and Technology Organization.

Rogan J. and Chen D. (2004) Remote Sensing Technology for Mapping and Monitoring
Land-cover and Land-use Change. Progress in Planning, Vol. 61, pp. 301-325.

82
Ropert-Coudert Y. and Wilson R.P. (2005) Trends and Perspectives in Animal-Attached
remote Sensing. Frontiers in Ecology and Environment, Vol. 3, No. 8, pp. 437-444.

Rolstad J. (2005) Landscape Ecology and Wildlife Management. In: Wiens J. A. and Moss
M.R. (Ed.) Issues and Perspectives in Landscape Ecology, pp. 208-216, Cambridge
University Press, New York.

Rydin H. and Jeglum J.K. (2006) Biology of Peatlands. Oxford University Press, USA, 343p

Saher D.J. and Schmiegelow K.A. (2004) Movement Pathways and Habitat Selection by
Woodland Caribou During Spring Migration. Proceedings: The Tenth American Caribou
Workshop, Girdwood, Alaska, USA, May 4-6, 2004.

Santos T., Tenedorio J.A., Encarnacao S. and Rocha J. (2006) Comparing Pixel vs. Object
Based Classifiers for Landcover Mapping with Envisat-MERIS Data. 26th EARSeL
Symposium, Maio, Varsovia.

Saura S. (2004) Effects of Remote Sensor Spatial Resolution and Data Aggregation on
Selected Fragmentation Indices. Landscape Ecology Vol. 19, pp. 197-209.

Schaefer J.A., Bergman C.M. and Luttich S.N. (2000) Site Fidelity of Female Caribou at
Multiple Scales. Landscape Ecology, Vol. 15, pp. 731-739.

Schaefer J.A. and Mahoney S.P. (2007) Effects of Progressive Clearcut Logging on
Newfoundland Caribou. Journal of Wildlife Mangament, Vol.71, No. 6, pp 1753-1757.

Schneider D.C. (2001) The Rise of the Concept of Scale in Ecology. BioScience, Vol. 51,
No. 7, pp. 545-553.

Sclater W.A.B. (1905) The Woodland Caribou. The Newfoundland Quarterly, Vol. 5, No. 1.

Servheen G. and Lyon L.J. (1989) Habitat Use by Woodland Caribou in the Selkirk
Mountains. The Journal of Wildlife Management, Vol. 53, No. 1, pp. 230-237.

Stuart-Smith A.K., Bradshaw C.J.A., Boutin S., Hebert D.M. and Rippin A.B. (1997)
Woodland Caribou Relative to Landscape Patterns in Northestern Alberta. The Journal of
Wildlife Management, Vol. 61, No. 3, pp. 622-633.

Syed A. and Abdulla A. M. S. (2002) Assessing Desert Vegetation Cover Using Remotely
Sensed Data: A Case Study from the State of Qatar. Proceedings: 23rd Asian Conference on
Remote Sensing, Kathmandu, Nepal, November 25-29, 2002.

Tamstorf M.P., Aastrup P. and Cuyler L.C. (2005) Modelling Critical Caribou Summer
Ranges in Western Greenland. Polar Biology, Vol. 28, pp. 714-724.

Teterukovskiy A. and Edenius L. (2003) Effective Field Sampling for Predicting the Spatial
Distribution of Reindeer (Rangifer tarandus) with Help of the Gibbs Sampler. Ambio, Vol.
32, No. 8, pp. 568-572.

83
Theau J., Peddle D.R. and Duguay C.R. (2005) Mapping Lichen in a Caribou Habitat of
Northern Quebec, Canada, Using an Enhancement-Classification Method and Spectral
Mixture Analysis. Remote Sensing of Environment, Vol. 94, pp. 232-243.

Tiwari R.S., Arora M.K. and Kailash T. (1999) Soft Classification for Sup-Pixel Land Cover
Extraction. Potonirvachak, Journal of the Indian Society of Remote Sensing, Vol. 27, No.4,
pp. 225-234.

Topan H., Maktav D., Jacobsen K. and Buyuksalih G. (2009) Information Content of Optical
satellite images for Topographis Mapping. Journal of Remote Sensing, Vol. 30, No. 7, pp.
1819-1827.

Tueller P.T. (1980) Remote Sensing Applications for Wildlife Habitat Management.
California-Nevada Wildlife Transactions.

Tueller P.T. (1989) Remote Sensing for Rangeland Management. Journal of Range
Management, Vol. 42, No. 6, pp. 442-453.

Tuominen S. and Pekkarinen A. (2004) Local Radiometric Correction of Digital Aerial


Photographs for Multi Source Forest Inventory. Remote Sensing of Environment, Vol. 89,
pp. 72-82.

Turrill W.B. (1954) Mapping the Ranges and Distribution of Taxonomic Groups of Plants.
An appeal to Geographers. Kew Bulletin. Vol. 9, No. 1, pp. 59-64.

US Army Corps of Engineers (2003) Engineering and Design, Remote Sensing, Engineer
Manual, EM 1110-2-2907. Department of the Army, US Army Corps of Engineers,
Washington, D.C., USA,

Uttam K., Mukhopadhyay C., Kumar R. S. and Ramachandra T.V. (2008) Soft Classification
Based Sub-Pixel Allocation Model. Proceedings: International Conference on Operations
Research for a Growing Nation, Decxember 15-17, 2008, Sri Venkateswara University,
Tirupati 517 502.

Van Dyne G.M., Vogel W.G. and Fisser H.G. (1963) Influence of Small Plot Size and Shape
on Range Herbage Production Estimates. Ecology, Vol. 44, No. 4, pp. 746-759.

Van Wagner (1982) Practical Aspects of the Line Intersect Method. Information Report PI-
X-12, Petawawa National Forestry Institute, Canadian Forestry Service, Chalk River,
Ontario, Canada.

Vistnes I. and Nellemann (2001) Avoidance of Cabins, Roads, and Power Lines by Reindeer
during Calving. The Journal of Wildlife Management, Vol. 65, No. 4, pp. 915-925.

Vors L.V., Schaefer J.A., Pond B.A., Rodgers A.R. and Patterson B.R. (2006) Woodland
caribou Extirpation and Anthropogenic Landscape Disturbance in Ontario. The Journal of
Wildlife Management, Vol. 71, No. 4, pp. 1249-1256.

Ware R.D. (1903) Newfoundland Caribou. The Outing, Vol. 43, No. 1, pp. 24-28.

84
Watt A.S. (1947) Pattern and Process in the Plant Community. Journal of Ecology, Vol. 35,
Nos. 1 and 2, pp. 1-22.

Weir J.N., Mahoney S.P., McLaren B. and Ferguson S.H. (2007) Effects of Mine
Development on Woodland Caribou Ranifer tarandus Distribution. Wildlife Biology, Vol.
13, No. 1, pp. 66-74.

Weladji R.B. (2002) Disturbance Effects of Human Activities on Rangifer tarandus Habitat:
Implications for Life History and Population Dynamics. Polar Geography, Vol. 26, No. 3,
pp. 191-206.

Wiens J.A. (1989) Spatial Scaling in Ecology. Functional Ecology, Vol. 3, No. 4, pp. 385-
397.

Wiens J.A. and Milne B.T. (1989) Scaling of Landscapesin Landscape Ecology, or,
Landscape Ecology from a Beetles Perspective. Landscape Ecology, Vol. 3, No. 2, pp. 87-
96.

Wildlife Division (2007) Tin Dataset Derived from 10 Meter Contour Data. Dept.
Environment and Conservation, Wildlife Division, Newfoundland and Labrador, Canada.

Wittmer H.U., McLellan B.N., Serrouya R. and Apps C.D. (2007) Changes in Landscape
Composition Influence the Decline of a Threatened Woodland Caribou Population. Journal
of Animal Ecology, Vol. 76, pp. 568-579.

Woodcock C.E. (1987) The Factor of Scale in Remote Sensing. Remote Sensing of
Environment, Vol. 21, pp. 311-332.

Worboys M. and Duckham M. (2004) GIS A Computing Perspective. CRC Press,


Washington D.C. USA, 426 pages.

Wulder M.A., Franklin S.E., White J.C., Linkes J. and Magnussen S. (2006) An Accuracy
Assessment Framework for Large-Area Cover Classification Products Derived from
Medium-Resolution Satellite Data. International Journal of Remote Sensing, Vol. 27, No. 4,
pp. 663-683.

Wulder M., Cranny M., Dechka J. and White J. (2004) An Illustrated Methodology for
Landcover Mapping of Forests with Landsat-7 ETM+ Data: Methods in Support of EOSD
Landcover, Version 3. Natural Resources Canada, Canadian Forest Service, Pacific Forestry
Centre, Victoria, BC, Canada, March 2004, 35p.

Wulder M.A. and Nelson T. (2003) EOSD Landcover Classification Legend Report- Version
2. Canadian Forest Service, Victoria, British Columbia, Canada.

Xiaoying J. (2009) Segmentation-Based Image Processing System. [Internet] United States


Patent Application 0090123070. Available at: http://appft.uspto.gov/netacgi/nph-
Parser?Sect1=PTO2&Sect2=HITOFF&p=1&u=%2Fnetahtml%2FPTO%2Fsearch-
bool.html&r=1&f=G&l=50&co1=AND&d=PG01&s1=20090123070.PGNR.&OS=DN/2009
0123070&RS=DN/20090123070 [Accessed 27 October 2009].

85
Zonneveld I.S. (1989) The Land Unit A Fundamental Concept in Landscape Ecology, and
its Applications. Landscape Ecology, Vol. 3, No. 2, pp. 67-86.

86

You might also like