Optimising camera trap survey effort to reliably detect a …60869/Thesis_CDU_60869_Risle… ·...
Transcript of Optimising camera trap survey effort to reliably detect a …60869/Thesis_CDU_60869_Risle… ·...
MTEM Thesis - Jenni RISLER s910718
1
Optimising camera trap survey effort
to reliably detect a threatened species,
the black-footed tree-rat, Mesembriomys gouldii,
in open forest and woodland of tropical savannas
of the Top End, Northern Territory
Jennifer Anne Risler BSc
(also published as J. Low Choy)
Research Institute for the Environment and Livelihoods
School of Environment Charles Darwin University
Submitted in partial fulfilment of the Masters of Tropical Environmental Management
19 June 2017
MTEM Thesis - Jenni RISLER s910718
2
Statement of authorship
I declare that this thesis is my own work and has not been submitted in any form for any
other degree or diploma at any university or other institute of tertiary education. Information
derived from the published and unpublished work of others has been acknowledged in the
text and list of references.
19 June 2017
Me checking images from a camera trap at Fish River Station, 2015
The remote camera trapping component of this research was conducted under the Charles
Darwin University Animal Ethics Committee approved project A13026 Ongoing monitoring
and biodiversity surveys in the Northern Territory held by the Flora and Fauna Division, NT
Department of Environment and Natural Resources.
MTEM Thesis - Jenni RISLER s910718
3
Acknowledgements
Firstly I would like to thank Penny Wurm for her constant support, enthusiasm and
guidance through my Masters of Tropical Environmental Management (MTEM)
journey. I would also like to thank all the lecturers who have contributed to my
knowledge and experience in the research realm; in particular I thank Karen Joyce,
Natalie Rossitor-Rachor and Keith McGuinness. Thank you to Bernie from the Charles
Darwin University Library for your constant patience and knowledge with Endnote.
A huge thank you to Mum and Dad for reading my papers throughout all my studies,
and for making dinners for the kids and for helping financially those times when
needed. Thank you to Alaric Fisher for financial and moral support at critical times
through my studies. Thank you to Daniel Low Choy for the valuable child free time
required during fieldwork and close to submission deadlines.
Thank you to my thesis supervisors. A massive thank you to Christine Schlesinger who
has made this thesis possible with your patience, guidance and questions in addition
to the support material you have provided me. Thanks to Dani Stokeld for the
invaluable statistical advice and R code, often while conducting arduous fieldwork.
Thank you to Graeme Gillespie for access to corporate data, valuable comments and
for keeping me focused. Thank you Brydie Hill for your support and valuable
comments on the final run.
Thank you to everyone who set camera traps, identified animals and managed the
datasets that formed the basis for my research. Thanks to the Indigenous rangers,
Parks and Wildlife rangers, Kakadu National Park staff, Flora and Fauna scientists and
field staff; and the friends and volunteers who have spent many hours and days both
on country and in the office to capture the high quality data I used.
Thanks to David Lindsay for valuable guidance in writing my thesis through his
thoroughly readable book “Scientific Writing = Thinking in Words” (Lindsay 2011). The
tools and inspiration I have gleaned from this book have made my writing more
concise and structured and even enjoyable.
Thanks to Hugh Davies and Jayne Brimbox who read and commented on drafts, an
amazing process in itself.
MTEM Thesis - Jenni RISLER s910718
4
Table of Contents
Statement of authorship .......................................................................................................... 2
Acknowledgements .................................................................................................................. 3
List of Tables ........................................................................................................................ 5
List of Figures ....................................................................................................................... 6
Abstract .................................................................................................................................... 7
Introduction ............................................................................................................................. 8
Camera traps ........................................................................................................................ 9
Occupancy analyses ............................................................................................................. 9
Survey effort and sampling design..................................................................................... 11
My research ....................................................................................................................... 13
Methods ................................................................................................................................. 15
Study area and data acquisition......................................................................................... 15
Camera trap survey design ................................................................................................ 17
Mammal Data .................................................................................................................... 19
Analyses ............................................................................................................................. 20
Results .................................................................................................................................... 22
Discussion............................................................................................................................... 26
Management implications ................................................................................................. 29
References ............................................................................................................................. 31
Appendix 1. R code developed by Danielle Stokeld 2016 ..................................................... 38
Appendix 2. Camera trap orientation to bait station ............................................................ 44
Appendix 3. Variation in detection due to camera trap set up ............................................ 46
Distance between camera trap and bait station ......................................................... 46
Model of camera (Infra-red flash and White light flash) ............................................. 46
Camera Trap setting (1080 Pixels versus 3.1 Mega Pixels) .......................................... 46
MTEM Thesis - Jenni RISLER s910718
5
List of Tables
Table 1. Regions for which camera trap data were available with the number of sites
sampled, my (JR) contribution to surveys and sites at which black-footed tree-rats
(BFTR) were present. Only regions 1-8 were used in final analyses. ..................... 16
Table 2. Detection probability (p) and occupancy (ψ) estimates with standard error (SE) for
the black-footed tree-rat, listed by region in descending order of occurrence. The
number of study sites per region is represented by n. .......................................... 25
MTEM Thesis - Jenni RISLER s910718
6
List of Figures
Figure 1. Distribution of the black-footed tree-rat represented by, pre 2000 (red dots) and
from 2000 onwards (green dots), atlas records overlayed on the study regions. . 14
Figure 2. Eight study regions ................................................................................................. 16
Figure 3. Camera trap placement of the five camera array deployed at each site. .............. 17
Figure 4. Set up of an individual camera trap and bait station with cleared understory
vegetation Fish River Station 2015 ......................................................................... 18
Figure 5. Camera trap images from the six regions in which they were detected. ............... 20
Figure 6. Estimates of detection probability (p) per number of camera traps (1-5) deployed
per site. Dots represent mean estimates and bars represent the 95% confidence
interval. ................................................................................................................... 22
Figure 7. Estimates of occupancy (ψ) per number of camera traps (1-5) deployed per site.
Dots represent mean estimates and bars represent the 95% confidence interval.
................................................................................................................................ 23
Figure 8. The probability of detecting a BFTR at least once at an occupied site when using 1-
5 cameras with increasing numbers of cameras from top to bottom (numbers of
cameras shown in grey boxes) over four weeks (28 nights). The shaded area
represents the 95% confidence interval around the detection probability estimate.
The red lines indicate the time at which a reliable detection probability estimate is
reached (lower limit of 95% confidence interval = 0.85). Numerical values are
presented in Appendix 1. ....................................................................................... 24
Figure 9. Camera score v1 template showing scoring zones (0-4) used to score Fish River
data by JR.. .............................................................................................................. 44
Figure 10. Camera trap images showing the difference between a camera trap set at 2.5 m
and 1.5 m from the bait station ............................................................................. 47
Figure 11. Camera trap images showing the difference between flash types used by
different camera trap models. ............................................................................... 48
Figure 12. Camera trap images showing the difference between the two resolution settings
available by the same camera trap model. ............................................................ 49
MTEM Thesis - Jenni RISLER s910718
7
Abstract
Context. Imperfect detection, where a species remains undetected in surveys even though
it is present, occurs in all wildlife surveys. The probability of detecting a species with different
sampling designs can be assessed using occupancy modelling, then improved through
employing the most appropriate survey effort and methods.
Aims. My research applies analytical methods to determine optimal survey effort, by
varying the number of camera traps per site and deployment time, for the detection of the
threatened black-footed tree-rat, Mesembriomys gouldii.
Methods. My analysis was applied to camera trap data collected by myself and others for
a range of research objectives from 224 sites across eight regions in the Top End, Northern
Territory, between 2013 and 2016. All data were collected using a five camera array and over
a minimum six week period. Single-season occupancy models were applied to daily detection
data over 42 days to assess the effect of altering the number of camera traps set per site on
detection probability. Cumulative detection curves with 95% confidence intervals were
calculated to determine the optimal length of deployment for each scenario of deploying one
to five camera traps per site.
Key Results. The detection probability for black-footed tree-rats was low (p = 0.15) for the
one camera scenario, but increased with the number of cameras set per site, as did the
precision around the detection probability estimate. The optimal length of deployment with a
predefined precision (p* > 0.85) dropped from 21 days to 11 days with the addition of a
second camera trap. The optimal survey design, that should reliably determine the presence
of black-footed tree-rats if they are present, was identified as two camera traps per site for
two weeks.
Conclusions. The optimal survey effort required to reliably detect a target species can be
derived from pre-existing data. Additional variation due to environmental, ecological and
detection method, in this case using camera traps, will also influence detection probability
and needs to be considered.
Implications. The outcomes of this research will be incorporated into impact assessment
guidelines to improve detectability and provide definitive and achievable methods for
detecting the black-footed tree-rat prior to development activities. The guidelines will serve
to standardise survey design and methods, providing a greater knowledge base for natural
resource management and species conservation. The analytical methods outlined can also be
applied to determine optimal survey design for other species using pre-existing data from
past surveys.
MTEM Thesis - Jenni RISLER s910718
8
Introduction
Wildlife surveys make a key contribution to biodiversity conservation worldwide by
providing valuable data for biodiversity inventory (Smith et al. 2016), habitat modelling
(Tyre et al. 2003; Gu and Swihart 2004; Stephens and Anderson 2014), ecological research
(Hamel et al. 2012) and environmental monitoring and management studies (Pellet and
Schmidt 2005; Woinarski et al. 2010; Durso et al. 2011; Dostine et al. 2013; Ziembicki et al.
2015). The value of these data relies on surveys being conducted with clear objectives
(MacKenzie and Royle 2005) and appropriate sampling design and methods. The most
effective survey methods to detect wildlife will vary depending on the target species.
Evaluating and optimising sampling design and survey effort is particularly important when
target species are rare or elusive and competing priorities for funding and resources exist
(Durso et al. 2011; Shannon et al. 2014; Gálvez et al. 2016). Rare and elusive animals are
often classified as threatened and so are afforded protection through state and national
legislation (TPWCA 2000; EPBCA 1999). Significant resources are invested from both
public and private sectors to improve the information regarding the occurrence of
threatened species prior to either conservation or development activities. It is important
to be able to accurately determine the presence of threatened species in an area to
ensure appropriate environmental management is applied. This is especially important
when natural resources, such as native vegetation, experience significant impact by
agricultural and mining development. The consequence of non-detection of threatened
species in an area may lead to inappropriate environmental management decisions which
result in local extinction events.
Australia has suffered from the highest rate of modern mammal extinctions in the world,
with 30 mammal species becoming extinct in the past 200 years (Woinarski et al. 2010;
Woinarski et al. 2011). Multiple threats have been identified with little known about how
they operate together or vary regionally (Ziembicki et al. 2015). As a result of these threats
traditional inventory methods such as live trapping, return too few captures to effectively
assess or monitor the presence of these species. Camera trapping has been widely
adopted as an alternative method for surveying threatened species because it enables
data collection over long periods, with minimum disturbance to the target species and can
optimise species detection.
MTEM Thesis - Jenni RISLER s910718
9
Camera traps
Camera trapping has fast become the most widely used, cost effective, low impact means
to reliably detect terrestrial mammal species because it can provide systematic and
accurate data over prolonged survey periods without the requirement of live trapping
(Gálvez et al. 2016). The advantages of camera trapping over conventional methods for
species inventory, ecological and monitoring studies are well recognised (Meek et al.
2014; Smith et al. 2016). One very important advantage is that camera traps allow efficient
survey of remote locations with minimal disturbance (Rovero et al. 2013; Smith et al.
2016). Camera traps can continuously operate for weeks at a time and, following the initial
effort of site preparation, require no subsequent attention. For longer deployments
potential disturbance to animals and the need to access study areas is limited to
occasional changing of batteries and SD cards, and minor clearing where vegetation has
grown. However, camera trap data can be compromised by poor sampling design and or
misdirected survey effort. Conversely appropriate guidelines aimed to standardise
methods and optimise the detection of target species can serve to enhance the value of
data and their application to management outcomes (Meek et al. 2014; Pease et al. 2016).
Occupancy analyses
The presence/absence data, hereafter detection/non-detection data, acquired through
camera trapping is particularly suited to estimating and modelling occupancy. Occupancy
is defined as the proportion of sampling units where a target species is present
(MacKenzie et al. 2002), and being able to determine occupancy with reasonable accuracy
is integral to monitoring studies and environmental management. Occupancy analysis or
occupancy modelling are terms used to describe the analysis of detection/non-detection
data to investigate occupancy in relation to ecological and environmental variation. Two
critical considerations of occupancy analyses are the spatial variation in a species’
distribution and the detectability of the species (MacKenzie et al. 2006); the latter is the
primary focus of my research. The concept of imperfect detection is key to occupancy
modelling, and recognises that even when a species occupies an area it may not be
detected during a particular survey (MacKenzie et al. 2002; Nichols et al. 2008; Shannon et
al. 2014). Imperfect detection results in ‘false absences’ which can lead to
underestimation of occupancy (Wintle et al. 2005; Guillera-Arroita et al. 2010). Therefore
to increase the accuracy of occupancy estimates it is important to incorporate the
probability of detecting a species (Royle 2006). Furthermore, if probability of detection
MTEM Thesis - Jenni RISLER s910718
10
differs between sites, failure to account for this heterogeneity can bias estimates of
species occurrence and cause misinterpretation of the factors and their influence on
patterns of occurrence (Tyre et al. 2003; MacKenzie et al. 2006; Stephens and Anderson
2014; Pease et al. 2016).
Detection probabilities are also likely to change over time, by region, for different species,
and may also vary according to the detection method used. For example, Dostine et al.
(2013) showed that detection of multiple frog species in tropical northern Australia varied
over time with increased detection when water temperature was lower during rainfall
events. Shannon et al. (2014) conducted multispecies assessment of mammals in
Colorado, USA, with camera trap data. They truncated the daily detection history for black
bears by inserting ‘missing data’ for the time when bears began hibernation during the
survey period, to avoid skewing their detection probability estimates.
Regional variation in detection may exist between locations depending on vegetation
structure or ecological adaptations of broad ranging species (O’Connell Jr. et al. 2006;
Pease et al. 2016). Durso et al. (2011) identified interspecific variability in detection
probabilities of North American aquatic snakes and provide many more examples of
interspecific differences in the probability of detection for a range of taxa from around the
world. Detection probabilities have also been shown to vary greatly among different
methods of detection (Nichols et al. 2008; Durso et al. 2011). To account for so many
sources of variability in detection probability MacKenzie et al. (2006) and Durso et al.
(2011) emphasize the importance of estimating detection probabilities at every
opportunity.
Occupancy modelling explicitly accounts for imperfect detection by using data from repeat
surveys at sites calculate detection probability (MacKenzie et al. 2004; MacKenzie and
Royle 2005). Detection probability may be expressed either as constant (p) or cumulative
(p*) over multiple surveys at each site. Cumulative detection probability incorporates
knowledge from previous surveys in repeat sampling so that the cumulative detection
probability value increases exponentially with each survey. Acceptable probability values
vary depending on the type of detection probability and analyses being applied. By current
convention, confidence in results is achieved where p > 0.2 for constant detection
(MacKenzie et al. 2006) or p > 0.3 when allowing for constant detection with additional
covariate information and missing observations (MacKenzie et al. 2002). When considering
cumulative detection a p* > 0.85 is desirable (MacKenzie and Royle 2005; Tyre et al. 2003).
MTEM Thesis - Jenni RISLER s910718
11
Occupancy metrics and modelling are valuable to ecological and management studies
because they account for imperfect detection (MacKenzie et al. 2006; Nichols et al. 2008).
However, the increase in use of occupancy modelling for a wide range of ecological and
management studies has highlighted the importance of guidance towards improved
sampling design to optimise detection probability (Hamel et al. 2012; Guillera-Arroita et al.
2014; Shannon et al. 2014; Gálvez et al. 2016; Pease et al. 2016; Smith 2016).
Existing data are valuable for informing future sampling design and survey effort because
they can provide preliminary estimates of detection probability and occupancy and can be
used as the basis for simulations that optimise accuracy and precision of these estimates
for future sampling designs (Hamel et al. 2012; Shannon et al. 2014; Geyle 2015).
However, to be useful for determining detection probability and occupancy estimates that
can be generalised over an entire study area, the pre-existing data must meet several
principles and assumptions. These principles and assumptions aim to reduce variability in
the data set and are: (i) sites must be randomly selected if representing a subset of the
study area, (ii) sites are closed with respect to changes in occupancy within the sample
period, (iii) model parameters are constant across sites, and (iv) species detections within
and between sites are independent. The consequences of relaxing these principles and the
subsequent resolutions required are provided and discussed in MacKenzie et al. (2004;
2005).
Survey effort and sampling design
The fundamental reasons for considering occupancy modelling to investigate occurrence
of rare and elusive animals derived from wildlife surveys, are cost and efficiency
(Mackenzie et al. 2004). To optimise the value of wildlife surveys for monitoring and
environmental management, it is paramount to consider the trade-offs between expense,
survey effort and the accuracy and precision of results. One of the most important trade-
off in relation to survey effort is how many sites to survey versus the level of effort per site
(Shannon et al. 2014; Pease et al. 2016; Smith et al. 2016). Following this, decisions around
what sampling methods to use, how many devices per site and how long to survey are
necessary. MacKenzie and Royle (2005) used a theoretical approach to determine the
survey effort required to either optimise the precision of occupancy for a given level of
survey effort, or to achieve a desired level of precision with minimal effort. Assuming
detection probability and occupancy were constant, they found that for rare species, in
general, it may be more efficient to survey more sites less intensively, while for common
MTEM Thesis - Jenni RISLER s910718
12
species fewer sites should be surveyed with more repeat surveys. Shannon et al. (2014)
used a simulation approach based on camera trap data from Colorado, USA, and
considered both occupancy and detection probability to explore trade-offs in sampling
design and survey effort. Shannon et al. (2014) found that increasing the total survey
effort generally decreases the error associated with occupancy estimates, as would be
expected. However, changing the number of sites or camera trap deployment length had
very different outcomes depending on whether a species was common or rare (its
occupancy) and how easy or hard to detect it was (its detection probability). Shannon et
al. (2014) provided recommended survey designs for ten mammal species and three
virtual species with varying occupancy and detection probability estimates, based on
simulations with three levels of acceptable error. They demonstrated that for rare species
with low detection probability an unfeasible number of sites and length of deployment
would be required to achieve sufficient levels of overall detectability. However, for
common species with low detection probability, increasing the length of deployment was
most efficient.
Smith et al. (2016) assessed survey effort for a finite set of resources, in this case cameras,
for biodiversity inventory surveys. They used data they collected from north-western
Australia to evaluate the number of sites and camera traps per site required to optimise
multiple species detection. They found that increasing the number of sites increased
species richness estimates in a shorter time than sampling fewer sites, which required
longer deployments. Increasing the number of camera traps per site also increased species
richness estimates, but not to the same extent, indicating that with a limited number of
camera traps, increasing the number of sites rather than camera traps per site was
favourable. They also found that fewer sites or fewer camera traps per site required more
sample occasions (trap nights) to achieve similar estimates. This study showed that most
detections of mammals occurred within the first seven days with limited value in leaving
camera traps deployed beyond two weeks (Smith et al. 2016). The value of optimising
sampling design to wildlife surveys lies in formulating a clear research objective,
identifying target species and detection methods, selecting sampling units and appropriate
survey effort within the available budget and acceptable error rate (MacKenzie et al. 2005;
Shannon et al. 2014).
MTEM Thesis - Jenni RISLER s910718
13
My research
At present there are no guidance statements available that standardise and optimise the
ability to detect the presence of individual threatened fauna species in the Northern
Territory. My research had two main aims: The first was to outline analytical methods for
determining optimal survey effort to reliably detect a particular target species, when pre-
existing data are available. My second aim was to contribute to the information required
to prepare a guidance statement for the target species, the threatened black-footed tree-
rat, Mesembriomys gouldii.
Specifically, my objective was to determine how to optimise detection through survey
effort, with the number of camera traps per site and the length of time camera traps should
be deployed as the key variables considered. I also discuss some considerations of sampling
design and survey methods to improve detection in a study area in the context of the
dataset used for analyses.
The black-footed tree-rat is classified as Vulnerable in the Northern Territory by the
Territory Parks and Wildlife Conservation Act (TPWCA) 2000 and nationally as Endangered
by the Environment Protection and Biodiversity Conservation Act (EPBCA) 1999. These
classifications represent International Union for Conservation of Nature (IUCN) categories
and the Northern Territory classification is based upon the IUCN criterion which recognises
“a population reduction of greater than 30 per cent over the last ten years where the cause
of reduction may not have ceased” (Hill 2012). Potential sources for the decline of the
black-footed tree-rat include habitat modification through increased fire frequency and
fragmentation (Hill 2012).
Black-footed tree-rats occupy open forest and woodland of Australian tropical savannas
where they den in tree hollows or Pandanus during the day and forage on the ground and
in trees at night (Pittman 2003; Rankmore 2006). Three disjunct subspecies are recognised
across northern Australia; M. gouldii gouldii on the northwest Kimberley and Northern
Territory mainland, M. gouldii melvillensis on Melville Island and M. gouldii rattoides on
Cape York Peninsula. While black-footed tree-rats are considered uncommon to rare in the
Kimberley and Queensland they are still common but patchily distributed across the Top
End of the Northern Territory (Figure 1). As a result of the decline of this species across its
range coupled with increasing habitat modification through small and large scale clearing, it
is imperative that the presence of the black-footed tree-rat can be accurately determined
MTEM Thesis - Jenni RISLER s910718
14
prior to the approval of development projects that could impact the native vegetation in
tropical woodlands.
Figure 1. Distribution of the black-footed tree-rat represented by, pre 2000 (red dots) and from 2000 onwards,
(green dots) atlas records overlayed on the study regions.
In the Northern Territory the impact of development on the environment is assessed by
the Environmental Protection Authority (EPA) through the environmental impact
assessment process (EIA). Consideration of impact on threatened species and the
environmental management required to mitigate these impacts is required by national
and local government legislation (EPA 2006; DEWHA 2010; EPA 2017). Guidance
statements generally provide advice and guidelines to proponents, consultants and the
general public regarding information requirements for an environmental impact
assessment. This context provides the scope for my second aim, to contribute to the
information required to prepare the guidance statement for the detection and ultimate
protection of the black-footed tree-rat in the Northern Territory.
I predicted that by using pre-existing camera trap data and a precision around detection
probability of p* > 0.85, I would be able to identify the number of camera traps per site
and the length of deployment required to reliably detect the black-footed tree-rat, my
test case, in tropical savanna woodland of the Top End.
MTEM Thesis - Jenni RISLER s910718
15
Methods
Study area and data acquisition
The camera trap data used in this study were collected in the central north mainland of
the Top End in the Northern Territory (NT). This area represents the western and southern
range of the black-footed tree-rat in the NT, excluding two known isolated populations, on
Melville Island and in north east Arnhem Land (NT Department of Environment and
Natural Resources Fauna Atlas). The study area experiences a tropical monsoonal climate
with highly seasonal rainfall. Rainfall is almost exclusively limited to the humid wet-season
months between November and April (Friend and Taylor 1985; Price et al. 2005; Firth et al.
2006; Murphy et al. 2014; Davies et al. 2016). The study area experiences an annual
rainfall gradient from c. 1700 mm in the north-west (Greater Darwin Region) to c. 600 mm
in the south-east (Nitmiluk National Park) (Bureau of Meteorology 2017). Temperatures
are high throughout the year with mean monthly minima and maxima during the survey
period ranging from 25.4 to 33.3 oC in November in the north and 13.2 to 30.9 oC in July in
the south (Bureau of Meteorology 2017). The most widespread vegetation type in
northern Australia landscapes is savanna; eucalypt dominated (Eucalyptus and Corymbia
spp.) forests and woodlands with a grassy understorey. Within the study area savanna
vegetation varies subtly across the rainfall gradient from north to south with a reduction
of tree height, canopy cover, structural complexity and shift in species composition with
decreasing rainfall (Woinarski et al. 2000; Ziembicki et al. 2015). Eucalyptus tetrodonta
and E. miniata (Darwin Stringybark and Darwin Woolybutt) are the most common trees in
savanna forests and woodlands, however, large areas comprising mixed eucalypt
woodlands and sandstone specific woodlands also occur in the Top End (Dunlop and Webb
1991; Woinarski et al. 2007).
Camera trap data available for this study had been collected for a range of research
objectives between 2013 and 2016, across 11 regions (Table 1). These data were collected
and managed by staff of the Flora and Fauna Division including myself. I was responsible
for all fieldwork and data management for the Fish River Project which included field
deployment and retrieval of camera traps; live trapping; site, fauna and habitat metadata
collection and management; downloading of images and species identification. I was also
lead scientist for managing field sites on a number of other surveys as noted in Table 1.
MTEM Thesis - Jenni RISLER s910718
16
Table 1. Regions for which camera trap data were available with the number of sites sampled, my (JR)
contribution to surveys and sites at which black-footed tree-rats (BFTR) were present. Only regions 1-
8 were used in final analyses.
Region Location Total # Sites
Number of sites surveyed by JR
Sites with BFTR
1 Garig Gunak Barlu (Cobourg) National Park (2014) 25 - 22
2 Djelk IPA (2013 - 2015) 30 - 0
3 Warddeken IPA (2013 - 2015) 46 - 0
4 Greater Darwin Region (2014) 31 8 12
5 Kakadu National Park (2014) 19 - 5
6 Nitmiluk National Park - Fire Plots (2015) 34 4 1
7 Litchfield National Park - Fire Plots (2016) 20 2 1
8 Fish River Station (2014) 19 19 7
9 Judbarra (Gregory) National Park - Bush Blitz 15 - 0
10 Wardaman IPA 9 - 0
11 Groote Eylandt 20 - 0
Data from three regions were excluded from analyses because they were outside the
current known range of the black-footed tree-rat and no black-footed tree-rats were
detected on images (Table 1). The data analysed in this study comprise detection/non-
detection data of the black-footed tree-rat derived from images from camera traps set at
224 sites across the eight remaining regions (Figure 2). All sites are broadly classified as
woodland and open forest (Wilson et al. 1990; Brocklehurst et al. 2007; DENR 2016).
Figure 2. Eight study regions; 1 = Garig Gunak Barlu National Park, 2 =Djelk Indigenous Protected Area,
3 = Warddeken Indigenous Protected Area, 4 = Greater Darwin Region, 5 = Kakadu National Park,
6 = Nitmiluk National Park, 7 = Litchfield National Park and 8 = Fish River Station.
MTEM Thesis - Jenni RISLER s910718
17
Camera trap survey design
Five cameras were deployed per site in a diamond configuration with one camera placed
in the centre. The four remaining cameras were placed with a minimum and maximum
spacing of 30 and 100 m and a maximum distance of 200 m between the outer camera
traps (Figure 3). The variation in location and distance between cameras is due to camera
traps being attached to trees, which do not grow uniformly in the environment. However
the placement of camera traps is standardised to maintain a sample footprint of
approximately 0.5 ha. The broad aim of the surveys was to maximise detection of small
native mammals and larger native and feral mammals. Three models of Reconyx infrared
motion triggered camera traps were used (HC600 - infra-red flash, HC550 and PC850 -
white light flash, http://www.reconyx.com/) Different camera models were spread
amongst sites such that one infra-red flash camera was used per site with four white light
flash cameras to ensure that trapping effort would be equal between sites with respect to
any differences in detection efficiency among camera types. All cameras were set to
record three still images per trigger, one second apart with no delay between subsequent
triggers and sensitivity was set to high. Deployment periods varied in length but were at
minimum six weeks long across all data sets. Camera traps were active during January to
September inclusive.
Figure 3. Camera trap placement of the five camera array deployed at each site.
Min distance 30 m
Distance 30-100 m (ideal 50 m)
Max distance between outside camera traps 200 m
MTEM Thesis - Jenni RISLER s910718
18
Cameras were attached to trees greater than 20 cm diameter at breast height (DBH) at a
height of 70 cm to the top of the camera trap housing. In areas with no suitable trees,
cameras were mounted to star pickets. Cameras were set south facing to enhance the
temperature differential required to detect animal movements by the passive infrared
sensor, especially during dawn and dusk. Camera orientation was angled downward from
horizontal so the bait station was central to the field of view. All cameras used with bait
stations comprising 50 mm polyvinyl chloride (PVC) pipe with removable vent caps at
either end and mounted by cable ties on a post at a height of 30 cm. Bait comprising oats,
peanut butter and honey was placed in each bait station. At each site, bait stations were
placed 2.5 m from three of the cameras and at 1.5 m from the other two cameras. The
vegetation surrounding the camera and bait station was cleared to a perimeter distance of
1 m and 1.5 m respectively (Figure 4).
Figure 4. Set up of an individual camera trap and bait station with cleared understory vegetation Fish River
Station 2015
1 m 1.5 m
MTEM Thesis - Jenni RISLER s910718
19
Mammal Data
As part of the original data collection images from camera traps were downloaded and
viewed using a variety of image processing programs. The presence of all vertebrate fauna
species was identified independently by two observers and recorded in Microsoft Excel or
Microsoft Access for each project area. Upon sourcing this data for 11 regions and
confirming the absence of black-footed tree-rat records in the three data sets that were
subsequently excluded, data for black-footed tree-rat detections and absences were
collated for the remaining eight regions. Figure 5 shows example images of black-footed
tree-rats from each of the six regions where they were detected. Detection periods (day to
night), site names and camera trap number (1-5) were standardised across the regions,
where differences had occurred between projects. Camera trap failures were identified
and recorded and all information was consolidated into a single excel spreadsheet. Some
detection data were recorded by date (midnight to midnight) rather than the active,
overnight, period for the black-footed tree-rat (midday to midday). These data were
converted to ensure detection events from the same night were not split over two dates.
Individual detections were converted into daily detection histories over a consecutive six
week survey period (42 nights) for each camera trap. When camera trap failures occurred
these affected either part of the deployment period (e.g. camera trap moved by animal,
SD card filled with false triggers) or the entire deployment period (e.g. camera trap left off
at set up). Occupancy modelling can account for missing data so all data from all camera
traps at all sites, including where cameras had failed, were used in analyses to account for
these eventualities in future sampling. The resultant detection histories for each camera
trap (1120 camera traps) at every site (224 sites) were represented by 1 when black-
footed tree-rats were present for the trap night, 0 when black-footed tree-rats were
absent for the trap night and NA where there was no data due to camera trap failure or no
survey undertaken that day.
MTEM Thesis - Jenni RISLER s910718
20
Figure 5. Camera trap images from the six regions in which they were detected. No detections were recorded for two regions
Analyses
To assess the effect of the number of camera traps per site on the detectability of black-
footed tree-rats, single-season occupancy models were run using the package “unmarked”
(Fiske and Chandler 2011 and 2012) in R version 3.1.0 ( R Core Team 2014) using code
developed by Danielle Stokeld (Appendix 1). Analyses used the entire dataset, with
detection histories from all camera traps from each site within the eight regions,
Garig Gunak Barlu National Park Greater Darwin Region
Kakadu National Park Nitmiluk National Park
Litchfield National Park Fish River Station
MTEM Thesis - Jenni RISLER s910718
21
irrespective of whether black-footed tree-rats were detected at an individual site. The
estimate of occupancy and overall probability of detecting a black-footed tree-rat from at
least one of the five camera traps set at a site, during a single sampling occasion, in this
case a night, was determined for the entire dataset. Detection probability estimates were
then calculated for different numbers of camera traps per site by subsequently dropping
one camera trap randomly from the five camera array down to a single camera trap (5
cams, 4 cams, 3 cams, 2 cams, 1 cam) using 100 bootstrap iterations. The output file
contained estimates of occupancy and detection probability for each scenario with
standard errors and upper and lower 95% confidence limits (Appendix 1). The mean
estimated detection probability and upper and lower confidence limits for each scenario
were then calculated.
To investigate the effect of the length of deployment on the probability of detection,
cumulative detection probabilities were calculated using the mean estimated detection
probability (p) and the number of survey nights (K) using the equation 1 - (1 - p)K
(MacKenzie and Royle 2005). Cumulative detection represents the probability of detecting
the target species on at least one of K nights if the site is known to be occupied.
Cumulative detection curves with 95% confidence intervals were then graphed in R for
each scenario of using one to five camera traps per site using code developed by Danielle
Stokeld (Appendix 1). The number of nights required to obtain reliable probability of
overall detection (p* = 0.85), reflecting an acceptable false negative rate of 0.15 (Tyre et
al. 2003), was determined for each scenario of one, two, three, four and five camera traps
deployed per site. The detection curves were truncated from the six week deployment
length, for which the data were available to four weeks when it became evident that the
reliable detection estimates were reached well within this time frame.
As regions had been pooled for previous analyses, a final analysis was conducted to
determine whether there were regional variations in detection probability and occupancy
estimates. Data from the five camera traps at a site were collapsed to derive a single
detection history per site to maintain the independence assumption. Then a single-season
model, using region as a covariate, was run in Presence version 12.1 (Hines 2006; US
Geological Survey 2017) to obtain regional detection probability and occupancy estimates.
MTEM Thesis - Jenni RISLER s910718
22
Results
Black-footed tree-rats were detected at 48 of the 224 sites, with more than 974 detections
recorded from 163 camera traps over 47 040 trap-nights (naïve occupancy = 0.21). Some
camera trap failures occurred at set up (32 of 1120 camera traps) and a further 271
camera traps were moved or damaged during the six week deployment through
disturbance by animals or fire.
The probability of detecting a black-footed tree-rat at any site increased with the number
of camera traps deployed at a site (Figure 6). The accuracy of the estimate relative to the
true value, and precision, demonstrating the confidence in the detection probability
estimate, also increased. The precision around the probability of detecting a black-footed
tree-rat at any site is represented by a 95% confidence interval. Deploying five camera
traps per site (p5 = 0.31, range: 0.29-0.33) doubled the probability of detection compared
to deploying a single camera trap per site (p1 = 0.15, range: 0.09-0.22). Deploying more
than two cameras per site provided a reliable detection probability (p > 0.2) based on
analysis with no covariate information (MacKenzie et al. 2006). However if consideration is
given to the lower confidence limit, three camera traps per site provides a more reliable
estimate (p3 = 0.25, range: 0.20-0.30). The precision around detection estimates also
increased dramatically with the inclusion of the fifth camera trap.
Figure 6. Estimates of detection probability (p) per number of camera traps (1-5) deployed per site. Dots represent mean estimates and bars represent the 95% confidence interval.
Det
ecti
on
pro
bab
ility
est
imat
es (
95
% c
on
fid
ence
inte
rval
)
Number of camera traps per site
MTEM Thesis - Jenni RISLER s910718
23
Estimates of occupancy and their accuracy also increased with the number of camera traps
per site. Using one and two camera traps per site resulted in low occupancy estimates (ψ1
= 0.15, range: 0.09-0.24 and ψ2 = 0.19, range: 0.12-0.28) which increased as more cameras
were added to the model (ψ3 = 0.21, range: 0.13-0.29; ψ4 = 0.22, range: 0.15-0.30 and ψ5 =
0.24, range: 0.18-0.30) (Figure 7). The precision around the occupancy estimates also
increased with increasing number of cameras per site, although not as dramatically as with
the detection probability estimates (Figures 6 and 7).
Figure 7. Estimates of occupancy (ψ) per number of camera traps (1-5) deployed per site. Dots represent
mean estimates and bars represent the 95% confidence interval.
The precision of cumulative detection probability estimates increased considerably with
each additional camera trap deployed per site (Figure 8). The time required to reliably
detect (p* ≥ 0.85) the presence of black-footed tree-rats at occupied sites was reduced
with each additional camera trap deployed per site (red line in Figure 8). The modelling
showed that deploying a single camera trap for 21 nights (3 weeks) would result in an 85%
chance of detecting a black-footed tree-rat at least once. If using a second camera trap,
the deployment time required to achieve the same precision is nearly halved at 11 nights.
If four and five camera traps are used, the time of deployment required to maintain
precision is further reduced to one week and 6 days respectively.
Occ
up
ancy
est
imat
es (
95
% c
on
fid
en
ce in
terv
al)
Number of camera traps per site
MTEM Thesis - Jenni RISLER s910718
24
Figure 8. The probability of detecting a BFTR at least once at an occupied site when using 1-5 cameras with
increasing numbers of cameras from top to bottom (numbers of cameras shown in grey boxes)
over four weeks (28 nights). The shaded area represents the 95% confidence interval around the
detection probability estimate. The red lines indicate the time at which a reliable detection
probability estimate is reached (lower limit of 95% confidence interval = 0.85). Numerical values
are presented in Appendix 1.
Based on the analyses conducted, reliable detection (p* ≥ 0.85) of black-footed tree-rats, if
they are present in the landscape, can be achieved by using two camera traps for 11 nights
or three cameras over 9 nights (Figure 8), provided future sampling is undertaken in
similar conditions to those under which pre-existing data used in these analyses were
collected.
Number of trap nights
Pro
bab
ility
of
det
ecti
ng
a B
lack
-fo
ote
d t
ree
-rat
MTEM Thesis - Jenni RISLER s910718
25
The core analyses for this project used data pooled across eight regions, however data
were also analysed separately for each region using a single detection history per site
based on five cameras per site over 42 nights. There was notable variation in both
occupancy and detectability of the black-footed tree-rat across regions (Table 2). The
highest occupancy was recorded at Garig Gunak Barlu National Park (ψ = 0.88 ± 0.06) with
moderate detectability (p = 0.37 ± 0.02). Litchfield National Park exhibited the lowest
occupancy (ψ = 0.03 ± 0.03) of all regions that detected black-footed tree-rats, but with a
relatively high probability of detection (p = 0.29 ± 0.07). Kakadu National Park showed
moderately low occupancy (ψ = 0.26 ± 0.10) with the highest rate of detection (p = 0.43 ±
0.03).
Table 2. Detection probability (p) and occupancy (ψ) estimates with standard error (SE) for the black-footed
tree-rat, listed by region in descending order of occurrence. The number of study sites per region is represented
by n.
Regions sampled n p SE ψ SE
Garig Gunak Barlu NP 25 0.37 ± 0.02 0.88 ± 0.06
Greater Darwin Region 31 0.13 ± 0.02 0.39 ± 0.09
Fish River Station 19 0.26 ± 0.03 0.37 ± 0.11
Kakadu NP 19 0.43 ± 0.03 0.26 ± 0.10
Nitmiluk NP 34 0.07 ± 0.04 0.05 ± 0.05
Litchfield NP 20 0.29 ± 0.07 0.03 ± 0.03
Djelk IPA* 30 0.50 ± 2.50 0.00 ± 0.00
Warddeken IPA* 46 0.50 ± 2.50 0.00 ± 0.00
*These regions had no detections of black-footed tree-rats
MTEM Thesis - Jenni RISLER s910718
26
Discussion
My analyses demonstrate the effects of modifying survey effort on the ability to obtain
reliable detection of a single target species. I showed that the sampling design of a target
species can be effectively informed by modelling detection probabilities and occupancy
that result from different survey efforts. The data used for modelling were collected at
survey sites within the known distribution range of the black-footed tree-rat and provided
the information from which detection of this threatened species could be optimised in the
future. Multi-method occupancy modelling, incorporating imperfect detection, provided
the framework to investigate variation in the detection probability for differing numbers
of camera traps per site and the length of deployment. Deploying a second camera trap
per site reduced the required survey time to 11 days, a reduction of 10 days from a single
camera trap deployment. Increasing the number of camera traps per site beyond two only
reduced the deployment time by two days for each additional camera, which is not
enough to offset the increase in deployment effort and camera traps required. Based on
these results, I recommend that surveys targeting black-footed tree-rats should use two
camera traps per site for a minimum of two weeks, a survey effort that should reliably
detect the presence of this species.
The five-camera array described by Gillespie et al. (2015) is useful for monitoring and
evaluating the diversity of mammals present within an area, as it optimises detection of
multiple species. For this reason it has become standard for many wildlife surveys in the
Top End of the Northern Territory. The detection probability and occupancy estimates for
the research presented here were derived from wildlife surveys that used the five-camera
array. However the resources required to deploy five camera traps per survey site are
considerable, and the results of my analyses suggest that this level of survey effort is not
necessary if the purpose is solely to determine the presence or absence of the black-
footed tree-rat. The use of more than one camera trap per site, resulted in increases in
detection probability, which is consistent with other recent work that has demonstrated
that multiple cameras increase detection of single and multiple species (Stokeld et al.
2015; Pease 2016). The use of multiple camera traps per site also provides a contingency
for camera failure. The failure of one camera due to animal interference, environmental
damage or operator error will not result in complete loss of data for a site when multiple
cameras are used (Stokeld et al. 2015). For black-footed tree-rats, reliable overall
detection probability estimates (p > 0.20) were also higher for three camera traps per site
MTEM Thesis - Jenni RISLER s910718
27
(0.25, range 0.20-0.30) than two camera traps per site (0.21, range 0.16-0.26), particularly
considering the lower confidence limit. However, the cumulative detection probability
curves, which also account for length of deployment show that deploying two cameras per
site rather than three should still enable reliable occupancy estimates.
Other mammals of northern Australia that are less conspicuous, more elusive, or wider
ranging than the black-footed tree-rat may require a greater effort in the number of
cameras or length of deployed per site. For example Stokeld et al. (2016) showed that five
to six camera traps per site coupled with deployed for a minimum of eight weeks were
required to adequately detect the presence of highly mobile feral cats, Felis catus. Pease
et al. (2016) found that detection probabilities of individual species varied as the number
of camera traps increased per site, which is not surprising given the variability in how
individual species utilise and move through the environment. Habitat use varies for
different species both seasonally and spatially resulting in variation in detection through
time and space (Shannon et al. 2014; Stokeld et al. 2015; Pease et al. 2016). Given the
variation in detectability between species, there is a need to investigate and refine survey
effort on a case-by-case basis, using similar analyses to those presented here for the black-
footed tree-rat, to optimise detection when single species are being targeted.
My analyses were limited to investigating the variation around the overall and cumulative
detection probability for survey effort. However detectability of target species can also be
affected by timing of surveys, regional variation, environmental and habitat factors, inter-
specific responses to camera trap set up and different camera models. For example, black-
footed tree-rat detection probabilities and occupancy estimates were shown to differ
across regions (Table 2). While occupancy modelling can incorporate these sources of
variability in multivariate analyses, this was outside the scope of the current study.
However, the implications of regional variation in detection probability and other sources
of variation that existed in the dataset used for my analyses, and that are likely to affect
detection probabilities of black-footed tree-rats, are discussed below.
The variation in dominant habitat type sampled per region and structural variation across
the rainfall, productivity gradient are the most likely factors that could account for the
observed regional variation in both occupancy and detection probability estimates (Table
2). An attempt to classify habitat at the survey sites in a way that could be incorporated
into analyses was made using two vegetation datasets. These datasets were broad enough
to encompass all sites, but resulted in the classification of all sites as various forms of
MTEM Thesis - Jenni RISLER s910718
28
woodland and open forest (Wilson et al. 1990; Brocklehurst et al. 2007 in NR Maps 2016).
This broad classification indicates that all sites occur in suitable habitat for black-footed
tree-rats. However, some sites occurred in small-scale habitat types that occur patchily
within the savanna matrix, such as monsoon vine thicket, riparian, sandstone heath,
wetlands or open grasslands (Woinarski et al. 2000). With the exception of the riparian
areas it is unlikely that these habitats contain the target species. Previous research has
shown that the presence of species specific den trees and the patchy distribution of shrub
species that provide food resources are important determinants of habitat selection by
black-footed tree-rats (Friend and Taylor 1985; Friend 1987; Morton 1992; Rankmore
2006; Firth et al. 2006). Fine scale vegetation data were collected at survey sites, but
accessing and analysing these data was beyond the scope of my research. These data,
however, would enable the survey sites and regions to be classified according to the
predominant vegetation type in which camera traps were placed. For example, the low
detection and occupancy estimates of Nitmiluk National Park are likely to be a result of the
high proportion of sites sampled in sandstone heath or sandstone woodlands that lack the
mid-storey shrubs identified as important to black-footed tree-rats. Removing sites
located in unsuitable habitat from analyses, may enable more accurate estimates of
detectability. The implication of considering a subset of a larger study area is that
estimates and inferences regarding occupancy and detection will only be valid for the
subset of the landscape sampled (MacKenzie et al. 2006).
The timing of surveys can also affect the likelihood of detecting a species and has
implications for modelling the data. The timing of surveys to maximise detection should be
informed by the activity cycles of the target species, including breeding, aestivation,
hibernation or activity based on resource availability (Dostine et al. 2013; Shannon et al.
2014; Geyle 2015). The data used in the current study were obtained from camera trap
detections of black-footed tree-rats between January and September. However Friend
(1987), observed peak activity and trapability when live trapping black-footed tree-rats in
May to June and in October to November, due to the influx of young animals at these
times. More recently Geyle (2015) also identified bias in detectability due to seasonal
activity patterns of the closely related brush-tailed rabbit-rat, Conilurus penicillatus, on the
Tiwi Islands in northern Australia, and stressed the importance of timing surveys to
optimise detectability of this species. Further investigation of the current dataset may
provide direction to the optimal timing for camera trap surveys to improve detectability of
black-footed tree-rats.
MTEM Thesis - Jenni RISLER s910718
29
Finally, detection of target species may be further optimised by considering the camera
trap set-up. Variations in placement of camera traps in the landscape, orientation of the
camera trap to bait station, type and or model of camera trap used, height of the camera
trap from the ground, distance between the camera trap and bait station, camera trap
settings, and flash type used by the camera trap have been shown to affect detection of
target species (Hamel et al. 2013; Meek et al. 2014; Gálvez 2016; Pease et al: 2016).
Preliminary exploratory analyses from Fish River Station indicated that detectability of
black-footed tree-rats and northern brown bandicoots, but not dingoes, may decrease
when the bait station was not centralised in the camera field of view (J. Risler unpublished
data presented in Appendix 2). Some sources of variation may be mitigated or managed
prior to data acquisition through clearly defined survey protocols to optimise detection of
target species. Where variation cannot be avoided occupancy modelling can account for
these sources of variation through the inclusion of covariate data in analyses. Further
guidance on survey design to optimise detection of black-footed tree-rats would be
achieved by incorporating these other factors in subsequent analyses.
Management implications
The recommended survey effort for black-footed tree-rats, of two traps deployed for a
minimum of two weeks, represents the most efficient survey effort identified in my
analyses. Optimal detectability for this species should be achievable with this design
provided that future sampling is conducted with the same methods and environmental
conditions as the original dataset. Along with deliberate consideration of other factors
that are likely to influence detectability of the species, the results presented here will be
incorporated into environmental impact assessment guidelines for the black-footed tree-
rat, to be endorsed by the Northern Territory Environmental Protection Authority. The
development of these guidelines will adopt a conservative approach to ensure that the
final survey design will optimise detection and account for the multiple sources of
variation to detectability. The guidelines will be available to proponents of development,
environmental consultants and the general public to guide and enable them to effectively
assess an area for the presence of black-footed tree-rat.
The ability to reliably determine the presence or absence of this species will become ever
more crucial. This research and the ensuing impact assessment guidelines are timely as
there is a high risk that future small and large scale clearing will further modify or reduce
the availability of suitable habitat for the black-footed tree-rat. Standardisation of survey
MTEM Thesis - Jenni RISLER s910718
30
effort as proposed by this study will ensure that there is reliable and rigorous information
to guide future management decisions and protect the black-footed tree-rat from further
decline. Furthermore, the systematic stepwise approach to analysing pre-existing data to
determine optimal survey design, presented here, may be applied more broadly to other
species and detection methods, to improve the efficacy of surveys for other threatened
species.
MTEM Thesis - Jenni RISLER s910718
31
References
Albano, P.G., Strazzari, G., D’Occhio, P., and Succetti, F. (2015). Field estimates of
detectability and site occupancy show that northern Italy forest molluscs are spatially
rare and poorly detectable. Italian Journal of Zoology, 82(4), 592-608.
Bureau of Meteorology (2017). Climate data on line. Available at
http://www.bom.gov.au/climate/data/ [Verified 12 June 2017]
Brocklehurst, P., Lewis D., Napier, D. and Lynch, D. (2007). Northern Territory and Field
Methodology for Vegetation Survey and Mapping. Survey and Mapping. Department of
Natural Resources, Environment and the Art Technical Report No. 02/2007D.
(Palmerston, Northern Territory)
Davies, H.F., McCarthy, M.A., Firth, R.S.C., Woinarski, J.C.Z., Gillespie, G.R., Andersen, A.N.,
Geyle, H.M., Nicholson, E. and Murphy, B.P. (2016). Top-down control of species
distributions: feral cats driving the regional extinction of a threatened rodent in
northern Australia. Diversity and Distribution 1-12.
Department of Environment and Natural Resources, DENR (2016). NR Maps. Available at
http://nrmaps.nt.gov.au/;jsessionid=1225vl3u54kw51i1e1h7ymdmsn [Verified 16 June
2017]
Department of Environment, Water, Heritage and Arts, DEWHA (2010). Survey guidelines
for Australia’s threatened frogs. Guidelines for detecting frogs listed as threatened
under the Environment Protection and Biodiversity Conservation Act 1999. Department
of Environment, Water, Heritage and Arts. (Canberra)
Dostine, P.L., Reynolds, S.J., Griffiths, A.D. and Gillespie, G.R. (2013). Factors influencing
detection probabilities of frogs in the monsoonal tropics of northern Australia:
implications for the design of monitoring studies. Wildlife Research 40 (5) 393-402.
Dunlop, C.R., and Webb, L.J. (1991). Flora and vegetation. In 'Monsoonal Australia:
Landscape, Ecology and Man in the Northern Lowlands'. (Eds C.D. Haynes, M.G. Ridpath
and M.A.J. Williams.) pp. 41-60 (Balkema: Rotterdam.)
MTEM Thesis - Jenni RISLER s910718
32
Durso, A.M., Willson, J.D. and Winne, C.T. (2011). Needles in haystacks: Estimating
detection probability and occupancy of rare and cryptic snakes. Biological Conservation
144 1508-1515.
Environmental Protection Authority (2006) Guidance for the Assessment of Environmental
Factors (in accordance with the Environmental Protection Act 1986) Protection of the
western swamp tortoise habitat, Upper Swan/Bullsbrook. Environmental Protection
Authority Guidance No. 7 (Perth, Western Australia)
Environmental Protection Authority (2017). Guidelines for assessment of impacts on
terrestrial biodiversity. Available at https://ntepa.nt.gov.au/environmental-
assessments/env-assessment-guidelines [Verified 16 June 2017]
Firth, R.S.C., Woinarski, J.C.Z., Brennan, K.G. and Hempel, C. (2006). Environmental
relationships of the brush-tailed rabbit-rat, Conilurus penicillatus, and other small
mammals on the Tiwi Islands, northern Australia. Journal of Biogeography 33 1820-
1837.
Fiske, I. and Chandler, R. (2011). Unmarked: An R package for fitting hierarchical models of
wildlife occurrence and abundance. Journal of Statistical Software 43, 1-23.
Fiske, I. and Chandler, R. (2012). Overview of Unmarked: An R package for the analysis of data from unmarked animals. Available at http://cran.r-project.org/web/packages/ unmarked/vignettes/unmarked.pdf [Verified 16 June 2017]
Friend, G.R. and Taylor, J.A (1985). Habitat preference of small mammals in tropical open-
forest of the Northern Territory. Australian Journal of Ecology 10 173-185.
Friend, G.R. (1987). Population ecology of Mesembriomys gouldii (Rodentia: Muridae) in
the wet-dry tropics of the Northern Territory. Australian Wildlife Research 14 293-303.
Gálvez, N., Guillera-Arroita, G., Morgan, B.J.T. and Davies Z.G. (2016). Cost-efficient effort
allocation for camera-trap occupancy surveys of mammals. Biological Conservation
Available at http://dx.doi.org/10.1016/j.biocon.2016.10.019 [Verified 13 June 2017]
Geyle, H. (2015). Survey design for detecting declines in the threatened brush-tailed rabbit-
rat Conilurus penicillatus on the Tiwi Islands of the Northern Territory. Honours Thesis
(Deakin University, Melbourne)
MTEM Thesis - Jenni RISLER s910718
33
Gillespie, G.R., Brennan, K., Gentles, T., Hill, B., Low Choy, J., Mahney, T., Stevens, A. and
Stokeld, D. (2015). A guide for the use of remote cameras for wildlife survey in northern
Australia. National Environmental Research Program, Northern Australia Hub, Charles
Darwin University (Casuarina, Northern Territory)
Gillespie, G.R., Risler, J.A., Einoder, L. and Stokeld, D. (2016). Factors Influencing Patterns of
Native Mammal Diversity at Fish River Station. Unpublished Report to the Indigenous
Land Corporation and The Nature Conservancy. Department of Land Resource
Management (Darwin, Northern Territory)
Gu, W. and Swihart, R.K. (2004). Absent of undetected? Effects of non-detection of species
occurrence on wildlife-habitat models. Biological Conversation 116 195-203.
Guillera-Arroita, G., Ridout, M.S. and Morgan, B.J.T. (2010). Design of occupancy studies
with imperfect detection. Methods in Ecology and Evolution 1 131-139.
Guillera-Arroita, G., Lahoz-Monfort, J.J., MacKenzie, D.I., Wintle, B.A. and McCarthy, M.A.
(2014). Ignoring imperfect detection in biological surveys is dangerous: A response to
“Fitting and interpreting occupancy models”. PLoS One 9 (7) Available at
https://doi.org/10.1371/journal.pone.0099571 [Verified 13 June 2017]
Hamel, S., Killengreen, S.T., Henden, J. Eide, N.E., Roed-Eriksosen, L., Ims, R.A. and Yoccoz,
N.G. (2013). Towards good practice guidance in using camera-traps in ecology: influence
of sampling design on validity of ecological inferences. Methods in Ecology and Evolution
4 105-113.
Hill, B (2012). Threatened species of the Northern Territory Factsheet – Black-footed tree-
rat Mesembriomys gouldii. Available at
https://nt.gov.au/environment/animals/threatened-animals [Verified 16 June 2017]
Hines, J.E. (2006). ‘PRESENCE: Software to Estimate Patch Occupancy and Related
Parameters.’ (USGS–PWRC.) Available at https://www.mbr-
pwrc.usgs.gov/software/presence.html [Verified 13 June 2017]
Lindsay, D. (2011). Scientific writing = thinking in words. (CSIRO Publishing: Victoria)
MTEM Thesis - Jenni RISLER s910718
34
MacKenzie, D.I., Nichols, J.D., Lachiman, G.B., Droege, S., Royle, J.A and Langtimm, C.A.
(2002). Estimating site occupancy rates when detection probabilities are less than one.
Ecology 83 (8) 2248-2255.
MacKenzie, D.I., Royle, J.A., Brown, J.A. and Nichols, J.D. (2004). Occupancy estimation and
modelling for rare and elusive populations. In ‘Sampling Rare or Elusive Species’ (Ed. W.I.
Thompson) pp. 149-172 (Island Press: Washington DC)
MacKenzie, D.I. and Royle, J.A. (2005). Designing occupancy studies: general advice and
allocating survey effort. Journal of Applied Ecology 42 1105-1114.
MacKenzie, D.I., Nichols, J. D., Royle, J.A., Pollock, K.H., Bailey, L.L. and Hines, J.E. (2006)
Occupancy estimation and modelling: Inferring patterns and dynamics of species
occurrence. (Academic Press: San Diego)
Meek, P.D., Ballard, G., Claridge, A., Kays, R., Moseby, K, O’Brien, T., O’Çonnell, A.,
Sanderson, J., Swann, D.E., Tobler, M. and Townsend, S. (2014). Recommended guiding
principles for reporting on camera trapping research. Biodiversity Conservation 23 (9)
2321–2343. doi:10.1007/s10531-014-0712-8
Murphy, B.P., Lehmann, C.E.R., Russell-Smith, J. and Lawes, M.J. (2014). Fire regimes and
woody biomass dynamics in Australian savannas. Journal of Biogeography 41 133–144.
Nichols, J.D., Bailey, L.L., O’Connell Jr., A.F., Talancy, N.W., Campbell Grant, E.H., Gilbert,
A.T., Annand, E.M., Husband, T.P. and Hines, J.E. (2008). Multi-scale occupancy
estimation and modelling using multiple detection methods. Journal of Applied Ecology
45 1321-1329.
O’Connell Jr., A.F., Talancy, N.W., Bailey, L.L., Sauer, J.R., Cook, R. and Gilbert, A.T. (2006).
Estimating site occupancy and detection probability parameters for meso- and large
mammals in a coastal ecosystem. Journal of Wildlife Management 70 (6) 1625-1633.
Olea, P.P. and Mateo-Tomás, P. (2011). Spatially explicit estimation of occupancy, detection
probability and survey effort needed to inform conservation planning. Diversity and
Distribution 17 714-724.
Pease, B.S., Nielsen, C.K. and Holtzmueller, E.J. (2016). Single-camera trap survey design
miss detections: Impacts on estimates of occupancy and community metric. Plos One 11
MTEM Thesis - Jenni RISLER s910718
35
(11) Available at http://dx.plos.org/10.1371/journal.pone.0166689 [Verified 13 June
2017]
Pellet, J. and Schmidt, B.R. (2005). Monitoring distribution using call surveys: estimating site
occupancy, detection probabilities and inferring absence. Biological Conservation 123
27-35.
Price, O., Rankmore, B., Milne, D., Brock, C., Tynan, C., Kean, L. and Roeger, L. (2005).
Regional patterns of mammal abundance and their relationship to landscape variables in
eucalypt woodlands near Darwin, northern Australia. Wildlife Research 32 435-446.
Rankmore, B.R. and Price, O.F. (2004). Effects of habitat fragmentation on the vertebrate
fauna of tropical woodlands, Northern Territory. In ‘The Conservation of Australia’s
Forest Fauna (second edition)’ (Ed. Daniel Lunney) pp. 452-473. (Royal Zoological Society
of New South Wales: Mosman)
R Core Team (2014). R version 3.1.0 The R Project for statistical computing. Available at
http://www.R-project.org/. [Verified 13 June 2017]
Rovero, F., Zimmermann F., Berzi, D. and Meek, P. (2013). “Which camera trap type and
how many do I need?” A review of camera features and study designs for a range of
wildlife research applications. Hystrix Italian Journal of Mammalogy 24 9-17.
Royle, J.A. (2006). Site occupancy models with heterogenous detection probabilities.
Biometrics 62 97-102.
Shannon, G., Lewis, J.S. and Gerber, B.D. (2014). Recommended survey designs for
occupancy modelling using motion-activated cameras: insights from empirical wildlife
data. PeerJ 2:e532 https://doi.org/10.7717/peerj.532 [Verified 13 June 2017]
Smith, J., Legge, S., James, A. and Tuft, K. (2016). Optimising camera trap deployment
design across multiple sites for species inventory surveys. Pacific Conservation Biology
23(1) 43-51. https://doi.org/10.1071/PC16017 [Verified 13 June 2017]
Stephens, R.B. and Anderson, E.M. (2014). Habitat association and assemblages of small
mammals in natural plant communities of Wisconsin. Journal of Mammology 95(2) 404-
420.
MTEM Thesis - Jenni RISLER s910718
36
Stokeld, D., Frank, A.S.K., Hill, B. Low Choy, J., Mahney, T., Stevens, A., Young, S., Djelk
Rangers, Warddeken Rangers and Gillespie, G.R. (2015). Multiple cameras required to
reliably detect feral cats in northern Australian tropical savanna: an evaluation of
sampling design when using camera traps. Wildlife Research 42(8) 642-649.
https://doi.org/10.1071/WR15083 [Verified 13 June 2017]
Tyre, A.J., Tenhumberg, B., Field, S.A., Niegalke, D., Parris, K. and Possingham, H.P. (2003).
Improving precision and reducing bias in biological surveys estimating false-negative
error rates. Ecological Applications 13(6) 1790-1801.
U.S. Geological Survey (2017). Presence v12.1 Software to Estimate Patch Occupancy and
Related Parameters. Available at
https://www.mbr-pwrc.usgs.gov/software/presence.html [Verified 13 June 2017]
Wilson, B.A., Brocklehurst, P.S., Clarke, M.J. and Dickinson, K.J.M. (1990). Vegetation Survey
of the Northern Territory, Australia. Conservation Commission of the Northern Territory
Technical Report No. 49. (Northern Territory)
Wintle, B.A., Kavanagh, R.P., McCarthy, M.A. and Burgman, M.A. (2005). Estimating and
dealing with detectability in occupancy surveys for forest owls and arboreal marsupials.
Journal of Wildlife Management 69 (3) 905-917.
Woinarski, J.C.Z., Fisher, A. and Fensham, R. (2000). Conservation of the grassy landscapes
of northern Australia. In 'Northern Grassy Landscapes Conference' pp. 5-15 (Savannas
Cooperative Research Centre, Darwin, Northern Territory.)
Woinarski, J.C.Z., Mackay, B., Nix, H. and Traill, B. (2007). The nature of northern Australia
Natural values, ecological processes and future prospects. (ANU Press, Canberra)
Woinarski, J.C.Z., Armstrong, M., Brennan, K., Fisher, A., Griffiths, A.D., Hill, B., Milne, D.J.,
Palmer, C., Ward, S., Watson, M., Winderlich, S. and Young, S. (2010). Monitoring
indicates rapid and severe decline of native small mammals in Kakadu National Park,
northern Australia. Wildlife Research 37 116-126.
Woinarski, J.C.Z., Legge, S., Fitzsimons, J.A., Traill, B.J., Burbidge, A.A., Fisher, A., Firth,
R.S.C., Gordon, I. J., Griffiths, A.D., Johnson, C.N., McKenzie, N.L., Palmer, C., Radford, I.,
Rankmore, B., Ritchie, E.G., Ward, S. and Ziembicki, M. (2011). The disappearing
MTEM Thesis - Jenni RISLER s910718
37
mammal fauna of northern Australia: Context, cause, and response. Conservation Letters
4 192-201.
Ziembicki, M.R., Woinarski, J.C., Webb, J.K., Vanderduys, E., Tuft, K., Smith, J., Ritchie, E.G.,
Reardon, T.B., Radford, I.J., Preece, N., Perry, J., Murphy, B.P., McGregor, H.W., Legge,
S., Leahy, L., Lawes, M.J., Kanowski, J., Johnson, C.N., James, A., Griffiths, A.D., Gillespie
G.R., Frank, A.S.K., Fisher, A. and Burbridge, A.A. (2015). Stemming the tide: Progress
towards resolving the causes of decline and implementing management responses for
the disappearing mammal fauna of northern Australia. THERYA 6 169-225.
MTEM Thesis - Jenni RISLER s910718
38
Appendix 1. R code developed by Danielle Stokeld 2016
Single-season occupancy models were implemented using the package “unmarked” in R version 3.1.0 (R Core Team 2014). The following code ran the occupancy models to;
determine detection probability (p) and occupancy (ψ) estimates for each scenario of one to five cameras per site,
create five output files, one for each scenario of one to five cameras, with ψ and p and their associated standard error, lower and upper confidence limit for each of 100 bootstrap iterations
graph ψ and p with upper and lower confidence intervals based on mean values of from the output files
#Set the working file directory
setwd("D:\MASTERS 2016-17\Data\rcode version2 and final dataset")
#Install required packages
#Install <- TRUE
#toInstall <- c("ggplot2","plyr","dplyr","unmarked")
#if(Install){
# install.packages(toInstall,
# dependencies = TRUE,
# repos = "http://cran.us.r-project.org")
#}
#Load packages
library(plyr)
library(dplyr)
library(unmarked)
library(ggplot2)
##Load csv file
dat <- read.csv("BFTR20160928.csv") #change file name
head(dat)
str(dat)
#S <- unique(dat$Site)
#S
#D <- dat[ sample( which( dat$Site == "CADELL_RTC_10"), 4, replace=F), ]
#D %>% summarise_each(funs(sum),3:44)
#######################################
#4 Cameras
#set up matrix for data to be entered into
output <- matrix(data=NA,nrow=100,ncol=8)
colnames(output) <- c("psi","psi.SE","LCL.psi","UCL.psi","p","p.SE","LCL.p","UCL.p")
#Run random selection of data 100 times
for(i in 1:100){
D <- dat[unlist(tapply(1:nrow(dat),dat$Site,function(x) sample(x,4))),] ## Replace 4 to 3, 2, 1 for random
choosing
DH<- D %>%
MTEM Thesis - Jenni RISLER s910718
39
group_by(Site) %>%
summarise_each(funs(sum),3:44)
Sp <- DH[,2:43]
umf <- unmarkedFrameOccu(y = Sp)
md <- occu(~1 ~1, umf)
psi <- backTransform(md, 'state')
p <- backTransform(md, type="det")
output[i,1] <- coef(psi)
output[i,2] <- SE(psi)
output[i,3] <- min(confint(psi, level = 0.95))
output[i,4] <- max(confint(psi, level = 0.95))
output[i,5] <- coef(p)
output[i,6] <- SE(p)
output[i,7] <- min(confint(p, level = 0.95))
output[i,8] <- max(confint(p, level = 0.95))
}
#save output
head(output)
output2 <- data.frame(output)
#Name file relevant of number of cameras randomly chosen
write.csv(output2, "BFTR4.csv") #rename file for number of cameras randomly chosen
(bftr4 <- c(4,mean(output2$psi),min(output2$LCL.psi),max(output2$UCL.psi),
+ mean(output2$p),min(output2$LCL.p),max(output2$UCL.p)))
###########################################################################
# 3 cameras
#set up matrix for data to be entered into
output <- matrix(data=NA,nrow=100,ncol=8)
colnames(output) <- c("psi","psi.SE","LCL.psi","UCL.psi","p","p.SE","LCL.p","UCL.p")
#Run random selection of data 100 times - 3 cams
for(i in 1:100){
D <- dat[unlist(tapply(1:nrow(dat),dat$Site,function(x) sample(x,3))),] ## Replace 4 to 3, 2, 1 for random
choosing
DH<- D %>%
group_by(Site) %>%
summarise_each(funs(sum),3:44)
Sp <- DH[,2:43]
umf <- unmarkedFrameOccu(y = Sp)
md <- occu(~1 ~1, umf)
psi <- backTransform(md, 'state')
p <- backTransform(md, type="det")
output[i,1] <- coef(psi)
output[i,2] <- SE(psi)
output[i,3] <- min(confint(psi, level = 0.95))
output[i,4] <- max(confint(psi, level = 0.95))
output[i,5] <- coef(p)
output[i,6] <- SE(p)
output[i,7] <- min(confint(p, level = 0.95))
output[i,8] <- max(confint(p, level = 0.95))
}
#save output
head(output)
output2 <- data.frame(output)
MTEM Thesis - Jenni RISLER s910718
40
#Name file relevant of number of cameras randomly chosen
write.csv(output2, "BFTR3.csv") #rename file for cameras randomly chose
(bftr3 <- c(3,mean(output2$psi),min(output2$LCL.psi),max(output2$UCL.psi),
+ mean(output2$p),min(output2$LCL.p),max(output2$UCL.p)))
###########################################################################
#2 cameras
#set up matrix for data to be entered into
output <- matrix(data=NA,nrow=100,ncol=8)
colnames(output) <- c("psi","psi.SE","LCL.psi","UCL.psi","p","p.SE","LCL.p","UCL.p")
#Run random selection of data 100 times - 3 cams
for(i in 1:100){
D <- dat[unlist(tapply(1:nrow(dat),dat$Site,function(x) sample(x,2))),] ## Replace 4 to 3, 2, 1 for random
choosing
DH<- D %>%
group_by(Site) %>%
summarise_each(funs(sum),3:44)
Sp <- DH[,2:43]
umf <- unmarkedFrameOccu(y = Sp)
md <- occu(~1 ~1, umf)
psi <- backTransform(md, 'state')
p <- backTransform(md, type="det")
output[i,1] <- coef(psi)
output[i,2] <- SE(psi)
output[i,3] <- min(confint(psi, level = 0.95))
output[i,4] <- max(confint(psi, level = 0.95))
output[i,5] <- coef(p)
output[i,6] <- SE(p)
output[i,7] <- min(confint(p, level = 0.95))
output[i,8] <- max(confint(p, level = 0.95))
}
#save output for each species
head(output)
output2 <- data.frame(output)
#Name file relevant of number of cameras randomly chosen
write.csv(output2, "BFTR2.csv") #rename file for cameras randomly chose
head(output2)
str(output2)
(bftr2 <- c(2,mean(output2$psi),min(output2$LCL.psi),max(output2$UCL.psi),
mean(output2$p),min(output2$LCL.p),max(output2$UCL.p)))
###########################################################################
#1 camera
#set up matrix for data to be entered into
output <- matrix(data=NA,nrow=100,ncol=8)
colnames(output) <- c("psi","psi.SE","LCL.psi","UCL.psi","p","p.SE","LCL.p","UCL.p")
#Run random selection of data 100 times
for(i in 1:100){
D <- dat[unlist(tapply(1:nrow(dat),dat$Site,function(x) sample(x,1))),] ## Replace 4 to 3, 2, 1 for random
choosing
DH<- D %>%
MTEM Thesis - Jenni RISLER s910718
41
group_by(Site) %>%
summarise_each(funs(sum),3:44)
Sp <- DH[,2:43]
umf <- unmarkedFrameOccu(y = Sp)
md <- occu(~1 ~1, umf)
psi <- backTransform(md, 'state')
p <- backTransform(md, type="det")
output[i,1] <- coef(psi)
output[i,2] <- SE(psi)
output[i,3] <- min(confint(psi, level = 0.95))
output[i,4] <- max(confint(psi, level = 0.95))
output[i,5] <- coef(p)
output[i,6] <- SE(p)
output[i,7] <- min(confint(p, level = 0.95))
output[i,8] <- max(confint(p, level = 0.95))
}
#save output
head(output)
output2 <- data.frame(output)
#Name file relevant of number of cameras randomly chosen
write.csv(output2, "BFTR1.csv") #rename file for number of cameras randomly chosen
(bftr1 <- c(1,mean(output2$psi),min(output2$LCL.psi),max(output2$UCL.psi),
mean(output2$p),min(output2$LCL.p),max(output2$UCL.p)))
###########################################################################
#5 cameras
#set up matrix for data to be entered into
output <- matrix(data=NA,nrow=100,ncol=8)
colnames(output) <- c("psi","psi.SE","LCL.psi","UCL.psi","p","p.SE","LCL.p","UCL.p")
#Run random selection of data 100 times
for(i in 1:100){
D <- dat[unlist(tapply(1:nrow(dat),dat$Site,function(x) sample(x,5))),] ## Replace 4 to 3, 2, 1 for random
choosing
DH<- D %>%
group_by(Site) %>%
summarise_each(funs(sum),3:44)
Sp <- DH[,2:43]
umf <- unmarkedFrameOccu(y = Sp)
md <- occu(~1 ~1, umf)
psi <- backTransform(md, 'state')
p <- backTransform(md, type="det")
output[i,1] <- coef(psi)
output[i,2] <- SE(psi)
output[i,3] <- min(confint(psi, level = 0.95))
output[i,4] <- max(confint(psi, level = 0.95))
output[i,5] <- coef(p)
output[i,6] <- SE(p)
output[i,7] <- min(confint(p, level = 0.95))
output[i,8] <- max(confint(p, level = 0.95))
}
#save output
head(output)
output2 <- data.frame(output)
MTEM Thesis - Jenni RISLER s910718
42
#Name file relevant of number of cameras randomly chosen
write.csv(output2, "BFTR5.csv") #rename file for number of cameras randomly chosen
(bftr5 <- c(5,mean(output2$psi),min(output2$LCL.psi),max(output2$UCL.psi),
mean(output2$p),min(output2$LCL.p),max(output2$UCL.p)))
######################################################
######################################################
#GRAPH THE RESULTS
#############################
#Create matrix
names <- c("cameras","m.psi","LCL.psi","UCL.psi","m.p","LCL.p","UCL.p")
results <- rbind(names,bftr5,bftr4,bftr3,bftr2,bftr1)
write.csv(results,"SamplingResults.csv")
######################################################
######################################################
#Plot estimates
r <- read.csv("BFTR_SamplingResults.csv")
str(r)
r$CT <- factor(r$cameras) #set the num of cameras to a factor
#Plot PSI and 95% confidence interval
p1 <- ggplot(r, aes(x=CT,y=psi)) + geom_point(size = 2.5) + theme_bw()
p1 +
facet_grid(~ Species) +
geom_errorbar(aes(ymin=LCL.psi, ymax=UCL.psi),alpha=0.5) +
xlab("Number of Cameras") + ylab("psi (95% confidence interval)") +
theme(axis.title.y=element_text(face="italic"))
#Plot P and 95% confidence interval
p2 <- ggplot(r, aes(x=CT,y=p)) + geom_point(size = 2.5) + theme_bw()
p2 +
facet_grid(~ Species) +
geom_errorbar(aes(ymin=LCL.p, ymax=UCL.p),alpha=0.5) +
xlab("Number of Cameras") + ylab("p (95% confidence interval)") +
theme(axis.title.y=element_text(face="italic"))
MTEM Thesis - Jenni RISLER s910718
43
I then calculated a single mean detection probability and mean upper and lower 95%
confidence limits from the 100 output values for each scenario of one to five camera traps
per site. I converted these to cumulative detection values for days one to 28 days. I then
transposed the values in the following format to graph in R with the following code
developed by Danielle Stokeld. Highlighted cells show the days after which the precision of
p* > 0.85 was reached by the lower limit of the 95% confidence interval.
Q = read.csv("pCurves.csv")
head(Q)
names(Q)
str(Q)
###########################################
#Plot estimates
Q$Camera <- factor(Q$Camera) #set the num of cameras to a factor
p1 <- ggplot(Q, aes(x=Day,y=Mean)) + geom_line(size = 1) + theme_bw()
p1 +
facet_grid(Camera ~ .) +
geom_errorbar(aes(ymin=LL, ymax=UL),alpha=0.3) +
xlab("Number of Days") + ylab("p (95% confidence interval)") +
theme(axis.title.y=element_text(face="italic"))
p1 +
facet_grid(Camera ~ .) +
geom_ribbon(aes(ymin=LL, ymax=UL),alpha=0.3) +
xlab("Number of Days") + ylab("p (95% confidence interval)") +
theme(axis.title.y=element_text(face="italic")) +
scale_x_continuous(limits=c(1,28), breaks = c(1,7,14,21,28)) +
scale_y_continuous(limits=c(0,1.00), breaks= c(.2,.4,.6,.8,1))
MTEM Thesis - Jenni RISLER s910718
44
Appendix 2. Camera trap orientation to bait station
Figure 9. Camera score v1 template showing scoring zones (0-4) used to score Fish River Data from 2014 and 2015 deployments by JR. This version was used in Gillespie et al. (2016) to account for variation in camera trap orientation to bait station.
Single-season occupancy models were implemented using Presence version 6.2 (Hines 2006) prior to field deployment of the five camera array at Fish River Station in 2015 by Jenni Risler. Results are shown on the following page. The sample size favours the more optimal camera trap orientation to the bait station. And while there is a perceived effect on camera score to detectability of the target species the precision around the detection probability estimates are great and show that the variation is not statistical significant.
MTEM Thesis - Jenni RISLER s910718
45
Black-footed tree-rat
score # sites n = 154 p SE
1 10 0.1014 0.1111
2 5 0.3094 0.1995
3 54 0.4941 0.0665
4 78 0.5687 0.0697
0 10 - -
Northern brown bandicoot
score # sites n = 154 p SE
1 10 0.1000 0.1100
2 5 0.1622 0.1653
3 54 0.6372 0.0812
4 78 0.4746 0.0647
0 10 - -
Dingo
score # sites n = 154 p SE
0 10 0.1791 0.0791
1 10 0.2800 0.1205
2 5 0.3723 0.1671
3 54 0.2078 0.0538
4 78 0.2312 0.0398
Northern brown
bandicoot
Black-footed tree-rat
Northern brown bandico
ot
Dingo
MTEM Thesis - Jenni RISLER s910718
46
Appendix 3. Variation in detection due to camera trap
set up
Distance between camera trap and bait station
Model of camera (Infra-red flash and White light flash)
Camera Trap setting (1080 Pixels versus 3.1 Mega Pixels)
MTEM Thesis - Jenni RISLER s910718
47
Figure 10. Camera trap images from Fish River Station showing the difference between a camera trap set 2.5 m from the bait station (TOP) and 1.5 m from the bait station (BOTTOM).
MTEM Thesis - Jenni RISLER s910718
48
Figure 11. Camera trap images showing the difference between flash types used by different camera trap models. In the infra-red (IR) flash image from Fish River Station (TOP) it is much harder to see the definition of the animal compared to the white led flash used at Garig Gunak Barlu National Park (BOTTOM).
MTEM Thesis - Jenni RISLER s910718
49
Figure 12. Camera trap images showing the difference between the two resolution settings available by the same camera trap model. The 1080 Pixel setting returns a ‘cropped’ image as seen in the image from Fish River Station (TOP) where the 3.1 Mega Pixel setting provides a greater field of view as seen in the image from Litchfield National Park (BOTTOM).