A SPATIAL ANALYSIS OF GREATER BILBY (MACROTIS LAGOTIS) HABITAT IN SOUTH-WEST QUEENSLAND

Abstract

Greater Bilbies (Macrotis lagotis) once occupied 70% of Australia but are now an endangered species under the Environment Protection and Biodiversity Conservation Act (C'wlth) 1999. A dedicated 29 km 2 enclosure to protect reintroduced bilbies from predators was built in Currawinya National Park in southwest Queensland in 2003. Ten bilbies (three male and seven female) were released in the enclosure during the period of December 2005 to September 2006. The objective of this research was to develop a method to identify suitable Greater Bilby habitat from remote sensing imagery. A related objective was to spatially characterize how bilbies used their environment for feeding and resting. Aerial photographs (1:40,000) were used to classify the vegetation and land cover. Soil samples were used to construct a detailed soils map. Radio tracking (2005-06) and field tracking data (2008) were used to identify spatial associations between bilby activities and land cover and soils features in order to spatially characterize bilby micro-habitats. These results formed the basis of a Weighted Sum model that accurately identified potential bilby micro-habitats within the enclosure.

Full text

E. Dunwoody, X. Liu and K. McDougall
A SPATIAL ANALYSIS OF GREATER BILBY (MACROTIS
LAGOTIS) HABITAT IN SOUTH-WEST QUEENSLAND
Ernest. Dunwoody1, Xiaoye. Liu1, and Kevin. McDougall1
1Faculty of Engineering and Surveying, University of Southern Queensland
Toowoomba, QLD, 4350, Australia
Email: edunwoody1@bigpond.com
KEYWORDS: Greater Bilby, marsupial habitat, spatial analysis, remote sensing
ABSTRACT
Greater Bilbies (Macrotis lagotis) once occupied 70% of Australia but are now an
endangered species under the Environment Protection and Biodiversity Conservation
Act (C’wlth) 1999. A dedicated 29 km2 enclosure to protect reintroduced bilbies from
predators was built in Currawinya National Park in south-west Queensland in 2003.
Ten bilbies (three male and seven female) were released in the enclosure during the
period of December 2005 to September 2006.
The objective of this research was to develop a method to identify suitable Greater
Bilby habitat from remote sensing imagery. A related objective was to spatially
characterize how bilbies used their environment for feeding and resting. Aerial
photographs (1:40,000) were used to classify the vegetation and land cover. Soil
samples were used to construct a detailed soils map. Radio tracking (2005-06) and
field tracking data (2008) were used to identify spatial associations between bilby
activities and land cover and soils features in order to spatially characterize bilby
micro-habitats. These results formed the basis of a Weighted Sum model that
accurately identified potential bilby micro-habitats within the enclosure.
The analysis showed that bilbies prefer to dig burrows in Acidic Rudosol soils with
Shrubland with Dead Wood landcover. Their feed sites occur fairly evenly on Acidic,
Basic and Salic Rudosol soils but they preferred Shrubland landcover in which to
feed.
The modelling results showed that; (i) bilby feeding and resting micro-habitat could be
accurately predicted within the confines of the enclosure, (ii) bilbies were active in
only a small part of the larger area available to them, (iii) bilbies exhibited distinct
preferences for specific soil and landcover types for constructing burrows and feeding
and (iv) micro-habitats suitable for bilbies represent only a small percentage of the
enclosure.

E. Dunwoody, X. Liu and K. McDougall
2
INTRODUCTION
The Greater Bilby (Macrotis lagotis) is a small, cryptic, nocturnal, fossorial marsupial
in the Peramelidae Family (Figure 1). It is one of Australia’s most endangered
marsupials
1
, its range having declined 99.7 % since 1836 (McRae 2004, p. 111). It
remains in the wild in
only small areas of arid
inland Australia
(McRae 2004, p. 106;
Southgate 1990a, pp.
295-298). Populations
are maintained in
protected areas in
South Australia
(Moseby & Donnell
2003), New South
Wales (Finlayson,
Vieira, Priddel et al.
2008, p. 320) and
Queensland (Mayhew
2006, p. 5).
Figure 1. Greater Bilby (Macrotis lagotis) (© QEPA)
The Greater Bilby is a generalist omnivore that can live in a wide range of habitats as
evidenced by its historically broad distribution. Vegetation types range from Acacia
rich woodlands through shrub steppe communities to tussock and forbs grasslands
(Southgate 1990b, p. 111). Lavery and Kirkpatrick (1997, p. 274) reported bilbys
burrowing in “stone free Cretaceous sediments of cracking clays” that occur in
grassland downs adjacent to the watercourses in Davenport Downs (SW QLD). McRae
(2004, pp. 10, 71-72) reported that the bilbies in Astrebla Downs National Park built
burrows in areas of suitable soil structure and stability that included “Compacted
sandstone outcrops in deep cracking clay soils of the ashy plains --- along drainage
depressions”. Bilbies in the Tanami Desert (NW of Alice Springs) were reported to
preferentially occupy palaeodrainage habitats (Paltridge & Southgate 2001, p. 255).
Bilbys released at the Arid Recovery Reserve north of Roxby Downs (SA) were found
to prefer clay swale habitat, dunes and sand plains (Moseby & Donnell 2003, p. 19).
The reduction in their population and range has been caused primarily by predation
from red foxes (Vulpes vulpes) and cats (Felis catus), habitat destruction by domestic
and feral herbivores, increased provision of watering points and by past government
bounty policies (Hrdina 1997). Recolonisation of bilbies into “wild like” protected
areas from which predators have been excluded has been successful at Thistle Island
and Yookamurra Sanctuary in South Australia (Moseby & Donnell 2003), Scotia
Sanctuary in New South Wales (Finlayson et al. 2008, p. 320) and in the Bilby
Enclosure (BE) in Currawinya National Park (CNP) in Queensland (Mayhew 2006, p.
5).
1
Vulnerable under Schedule 1 of the Environment Protection and Biodiversity Conservation Act (Cwlth) of 1999
(Pavey, C. 2006, National recovery plan for the Greater Bilby Macrotis lagotis., Darwin)

E. Dunwoody, X. Liu and K. McDougall
3
Small animals cannot be detected directly by aerial or satellite photography. Their
habitats can only be determined indirectly by detecting and mapping combinations of
vegetation, soil and landscape elements and developing associations between these
features and the target animal. Southgate et al (2005) assessed bilby abundance in the
Tanami Desert by looking for their footprints, scats and diggings. Their ability to
identify bilby tracks was not obscured by vegetation or rabbits. McRae (2004, p. 77)
reported that “bilby burrows were readily visible from the air” due to their excavation
of different coloured subsoil in the open treeless area between the Diamantina River
and the Simpson Desert. Such knowledge must be able to be summarised and
represented spatially before it can be used in habitat mapping.
GIS systems allow interaction between data at different spatial and temporal scales
(Store & Jokimäki 2003). Multi-Criteria Evaluation (MCE) in a GIS environment
allows cartographic combination of multiple habitat factors so as to detect their
influence on the suitability of the habitat for a species. Store and Kangas (2001, p. 79)
demonstrated this approach to predicting habitat suitability for the polypore fungal
species Skeletocutis odora and subsequently with three species simultaneously, the
Redstart (Phoenicurus phoenicurus), Pied Flycatcher (Ficedula hypoleuca ) and S.
odora (Store & Jokimäki 2003). Apan et al (2004, p. 812) used a knowledge based
MCE GIS method to identify areas of high priority for revegetation to ameliorate
dryland salinity in the Hodgson Creek watershed (Darling Downs, Qld).
The absence of a geographical knowledge base about bilby activity at CNP
necessitated the development of an empirical data set. This required obtaining detailed
evidence of bilby activity in a wooded environment containing a significant rabbit
population and the extraction of indicator relationships for bilby activity.
The objective of this research was to test the use of aerial photography for detecting
suitable landscape elements with which to predict the suitability of habitat for Greater
Bilbys in the BE at CNP. Imagery captured prior to release of bilbys was used to
develop a model of suitable burrowing and feeding sites based on activity records of
the bilbies that were subsequently released. The accuracy of the model for detecting
bilby activity elsewhere in the enclosure was tested on separate validation areas.
METHODS
Study Areas
The study area was the 29 km2 BE in CNP built in 2003-2004 (Figure 2). It is
designed to exclude all ground based predators of bilbys while retaining the released
bilbys and their offspring.
Field Data Collection
Detailed data about soil characteristics and bilby activity were collected at eight
investigation areas selected at random within the BE (Figure 2). Trimble GeoXT hand
held GPS units loaded with TerraSynch software were used to collect both base station
and field observations. All GPS data were differentially corrected.

E. Dunwoody, X. Liu and K. McDougall
4
Figure 2. Study Area
The key GPS settings were: observation interval, 5 sec.; observations per feature, 20
(min); and HDOP, 6 (max). These values resulted in data with the following precision.
Differentially corrected feature
Non-differentially corrected feature
Mean = 2.18 m
Mean = 10.32 m
SD (95%) = 0.12 m
SD (95%) = 0.11 m
Signs of bilbies were recorded in a GPS data dictionary by a trained observer walking
in a series of roughly concentric circles through each vegetation zone in each of the
eight investigation areas.

E. Dunwoody, X. Liu and K. McDougall
5
Data Pre-processing
Six aerial photographs acquired in June 2003 by a private contractor (Jacobs 2003)
were used based on their availability, clarity and resolution. The images were captured
during the period 18-23 May 2003 at a flying height of 3,229 m with a calibrated 80
mm focal length Hasselblad lens resulting in imagery with a scale of 1:38,800. The
pixel size was 0.6682 m. Each image was georeferenced and the images were
mosaiced with ERDAS Imagine software (Ver.9.3). The mosaiced image was linearly
stretched and dehazed prior to Supervised Classification using the maximum
likelihood classifier. The resulting raster was reclassified into 5 Landcover Classes and
a final Accuracy Assessment performed using 50 randomly stratified sites. A soils
map was created by heads-up digitising based on an observed association between the
soil types and broad vegetation zones.
DATA ANALYSIS
The spatial association of burrows and feed sites with soil and landcover types was
determined by spatially joining the data sets and extracting the positive joins. The area
of soil types and landcover classes that were inspected was obtained by buffering each
track path (5m either side) and extracting the area of each soil and landcover type.
The frequencies of burrows and feed sites were used as class weights in multiple
Weighted Sum Overlay (WSO) calculations to determine the combination of layer
weights and classification break points that maximised the number of bilby signs in
high priority areas. The predictive accuracy of these values was determined by
vectorising the WSO rasters, spatially joining them with field records and analysing
the distribution of the joins between priority areas.
RESULTS
When averaged, normalised, and expressed on a common basis (0-10) the burrow and
feed site weights had the following values (Tables 1 and 2).
Table 1. Burrow Site Weights
Soil Class
Weight
L’cover Class
Weight
Hydrosol
0
Claypan
0
Salic Rudosol
3.58
Red Soil
1.54
Basic Rudosol
0
Shrubland
0
Acidic Rudosol
6.42
S’land with dead wood
6.83
Thick Vegetation
1.63
Total
10.00
Total
10.00
Table 2. Feed Site Weights
Soil Class
Weight
L’cover Class
Weight
Hydrosol
0
Claypan
0
Salic Rudosol
3.24
Red Soil
1.65
Basic Rudosol
3.09
Shrubland
3.73
Acidic Rudosol
3.66
S’land with dead wood
2.22
Thick Vegetation
2.39
Total
10.00
Total
10.00

E. Dunwoody, X. Liu and K. McDougall
6
The most accurate prediction of burrow sites and feed sites was obtained by weighting
the landcover layer as twice as important as the soil layer. Manual classification break
values that yielded the highest accuracy are shown in Table 3.
Table 3. Classification Break Values
Priority
Burrow Site Break
Value
Feed Site Break
Value
Low
0 – 4
0 – 6
Medium
4 – 14
6 – 7
High
14 – 21
7 - 10
These values were used to produce maps of Burrow and Feed Site Priority areas for the
whole of the BE. Figure 3 shows an illustration of such mapping for Study Area 4.
The model predicted high priority locations for burrow sites and feed sites with a 67%
and 84% accuracy respectively in the areas in which it was developed (Areas 2, 4 and
7) (Table 4).
Table 4. Study Area Accuracy
Parameter
Feature
Predicted priority of sites
High
SD
Medium
SD
Low
SD
Total
Totals
Burrows
4
1.53
2
0.58
0
0
6
Scrapes
64
5.77
7
2.31
5
1.53
76
Percent
Burrows
67%
33%
0%
100%
Scrapes
84%
9%
7%
100%
When validated in independent areas (Areas 1, 3, 5, 6, and 8) the model predicted high
priority locations for burrow sites and feed sites with 84% and 80.5% accuracy
respectively (Table 5).
Table 5. Validation Area Accuracy
Parameter
Feature
Predicted priority of sites
High
SD
Medium
SD
Low
SD
Total
Totals
Burrows
16
4.60
1
0.45
2
0.65
19
Scrapes
70
9.08
10
2.35
7
1.52
87
Percent
Burrows
84%
5%
11%
100%
Scrapes
80.5%
11.5%
8
5
100%
Combining the burrow and feed site priorities into an overall Habitat Suitability Map
identified areas in which both or either criteria was a high priority. This is illustrated in
Figure 4.
DISCUSSION
The model is based on a Weighted Sum Overlay approach to predicting suitable
burrow and feeding locations for bilbys in the BE at CNP. It uses data derived from
aerial photography, soil sampling and ground presence tracking. The model was
validated by using a subset of the ground truth records. It predicted high priority
burrow and feed sites with 84% and 80.5% accuracy respectively. No reports of other

E. Dunwoody, X. Liu and K. McDougall
7
studies on predicting specific habitats for small marsupials were found. These results
need to be considered in the light of the limitations inherent in the procedures.
Figure 3. Burrow and Feed Site Priority Locations for Area 4
Figure 4. Combined Habitat Suitability detail for Test Area 4

E. Dunwoody, X. Liu and K. McDougall
8
Burrow and Feed Sites
Previous studies have established the importance of soil type in habitat selection by
bilbies. This study found that bilbies dug burrows in Acidic Rudosol soils twice as
frequently (6.42) as in Salic Rudosol soils (3.58). There was no evidence of bilby
burrows in Basic Rudosol soils or in Hydrosols. They chose Shrubland with Dead
Wood landcover twice as frequently (6.83) in which to burrow compared to exposed
Red Soil (1.54) and Thick Vegetation landcover (1.63). These findings are consistent
with Mayhew’s (2006, p. 79) more generalised findings for the BE.
The bilby burrow data used in this study were an amalgamation of burrow location
data collected by previous investigators by radio-tracking in 2005-2006 (McRae 2008
pers. comms.) and visual tracking data collected for this study. Prediction of burrow
habitats was based on 6 burrows in the three study areas and the results were validated
using a further 19 independent burrow locations. The small sample size of the
predictor burrow habitat is recognised as being less than optimal, however the
validation on a further 19 burrow locations supports the model outcomes.
Feed sites selected by bilbies were relatively evenly distributed between all three
Rudosol soil types but there were none in Hydrosols. Shrubland landcover was
preferred for feed sites (3.73) over Shrubland with Dead Wood (2.22) and Thick
Vegetation landcover (2.39). These findings were based on 76 clusters of scrapes in
the study areas and a further 87 clusters in the areas used for validation.
Analysis of the whole enclosure found that there were more areas of high priority for
feeding than for burrowing (Figure 3). This finding was anticipated. Combining the
burrow and feed site priority areas defines the areas suitable for both resting and
feeding. The combined map of burrow and feed sites (Figure 4) shows the areas in
which both are a high priority, or either one of the two is a high priority. McRae (2004,
p. sec. 2.3.2) documented the capacity of bilbies to forage at distances of hundreds of
meters from their refugia. It is therefore not surprising to find some record of feed sites
outside the high and medium priority habitat areas.
Precision and Accuracy
The Overall Classification Accuracy of 80% and Kappa Index of Agreement of 0.76
were considered satisfactory for classification of aerial imagery without an NIR band
into 5 landcover classes. The model accuracy results (Tables 4 and 5) are comparable
to or better than accuracies obtained by other models. Pasher, King and Lindsay
(2007), using Landsat 5 and Ikonos imagery, reported model accuracy of 70% for nest
sites for the Hooded Warbler (Wilsonia citrinia) in the validation area. Allowing for a
10 m error zone increased their accuracy to 87%. The bilby model’s use of a 95%
probability level resulted in a more stringent accuracy test than if a 99% probability
level had been used. The accuracy was scored according to the highest priority
polygon that lay partially or wholly within the 95% horizontal precision zone (Figure
4, subset).

E. Dunwoody, X. Liu and K. McDougall
9
CONCLUSION
This study demonstrated a procedure by which aerial photography can be used to
identify areas suitable for Greater Bilby burrow and feed sites with a high level of
accuracy. Physical tracking and GPS recording of bilby signs was shown to be a
practical and effective method of collecting geographically referenced information
about bilby habitat. Bilbys exhibited strong preference for specific types of soil and
landcover in which to burrow. Application of the model to the 29 km2 enclosure
established that there are extensive feeding sites within foraging range of all potential
burrow sites in the Enclosure. The applicability of these results is limited by their
dependence on aerial photography. Use of more widely available imagery with greater
spectral and less spatial resolution would allow (i) application to a wider area, (ii)
more accurate soil type classification, and (iii) integration with other land management
issues.
ACKNOWLEDGEMENTS
Special appreciation is acknowledged to Mr Peter McRae, Senior Zoologist,
Threatened Species Unit, QEPA for his assistance with bilby burrow data, aerial
photography and tracking and for the use of the Save The Bilby Facilities, Mr Brett
Ford, USQ student in land studies and photogrammetry, Maclean, NSW, for his very
capable research assistance in GPS recording of bilby tracks, maintaining the GPS
base station and photography and to Dr. Manda Page, Australian Wildlife
Conservancy, Perth, for suggesting the topic. Financially assistance from the Save the
Bilby Fund is gratefully acknowledged.
REFERENCES
Apan, A. A., Raine, S. R., Brocque, A. L. & Cockfield, G. 2004, 'Spatial prioritization
of revegetation sites for dryland salinity management: an analytical framework
using GIS', Journal of Environmental Planning and Management, vol. 47, no.
6, pp. 811 - 825.
Finlayson, G. R., Vieira, E. M., Priddel, D., Wheeler, R., Bentley, J. & Dickman, C. R.
2008, 'Multi-scale patterns of habitat use by re-introduced mammals: A case
study using medium-sized marsupials', Biological Conservation, vol. 141, no.
1, pp. 320-331.
Hrdina, F. 1997, 'Marsupial Destruction in Queensland 1877-1930', Australian
Zoologist, vol. 30, pp. 272-286.
Jacobs, I. O. 2003, Aerial Photographs of the Bilby Enclosure at Currawinya National
Park, I Melbourne.
Lavery, H. J. & Kirkpatrick, T. H. 1997, 'Field management of the bilby Macrotis
lagotis in an area of south-western Queensland', Biological Conservation, vol.
79, no. 2-3, pp. 271-281.

E. Dunwoody, X. Liu and K. McDougall
10
Mayhew, M. A. 2006, 'Habitat Preference and Burrow Use of Reintroduced Bilbies
(Macrotis lagotis) in Semi-Arid Mulga Lands', University of Queensland,
Gatton.
McRae, P. 2004, 'Aspects of the ecology of the Greater Bilby, Macrotis lagotis, in
Queensland', Masters thesis, University of Sydney.
Moseby, K. E. & Donnell, E. 2003, 'Reintroduction of the greater bilby, (Macrotis
lagotis) (Reid) (Marsupialia: Thylacomyidae), to northern South Australia:
survival, ecology and notes on reintroduction protocols', Wildlife Research,
vol. 30, no. 1, pp. 15-27.
Paltridge, R. & Southgate, R. 2001, 'The effect of habitat type and seasonal conditions
on fauna in two areas of the Tanami Desert', Wildlife Research, vol. 28, no. 3,
pp. 247-260.
Pasher, J., King, D. & Lindsay, K. 2007, 'Modelling and mapping potential hooded
warbler (Wilsonia citrina) habitat using remotely sensed imagery', Remote
Sensing of Environment, vol. 107, no. 3, pp. 471-483.
Southgate, R. I. 1990a, 'Distribution and Abundance of the Greater Bilby Macrotis
lagotis Reid (Marsupialia: Peramelidae)', in J. H. Seebeck, P. R. Brown, R. I.
Wallis & C. M. Kemper (eds), Bandicoots and Bilbies, Surrey beatty and Sons,
Sydney, pp. 293-302.
---- 1990b, 'Habitats and diet of the Greater Bilby, Macrotis lagotis Reid (Marsupialia:
Peramelidae)', in J. H. Seebeck, Brown, P.R., Wallis, R.L. and C.M. Kemper
(ed.), Bandicoots and Bilbies, Surrey Beatty and Sons, Sydney, pp. 303-309.
Southgate, R. I., Paltridge, R., Masters, P. & Nano, T. 2005, 'An evaluation of transect,
plot and aerial survey techniques to monitor the spatial pattern and status of the
bilby (Macrotis lagotis) in the Tanami Desert', Wildlife Research, vol. 32, no.
1, pp. 43-52.
Store, R. & Kangas, J. 2001, 'Integrating spatial multi-criteria evaluation and expert
knowledge for GIS-based habitat suitability modelling', Landscape and Urban
Planning, vol. 55, no. 2, pp. 79-93.
Store, R. & Jokimäki, J. 2003, 'A GIS-based multi-scale approach to habitat suitability
modelling', Ecological Modelling, vol. 169, no. 1, pp. 1-15.
BIOGRAPHY OF PRESENTER
Ernest Dunwoody is a post-graduate student at the University of Southern Queensland
studying for a Masters Degree in Spatial Science Technology. He holds a Bachelor’s
Degree in Agricultural Science (Entomology) from the University of Queensland, and
a Graduate Diploma in Geomatic Studies from the University of Southern Queensland.