Analysis
All analyses were undertaken in R v.4.1.0 (R Core Team, 2021) and where
required, specialised packages used are referenced below.
Camera data – Video metadata tagging was undertaken in ‘digiKam’ v.7.0.0 (digiKam Team, 2022) and we used R packages ‘camtrapR’ v.2.2.0 (Niedballa et al. , 2016) and ‘exiftoolr’ v.0.2.0
(Harvey, 2016) to extract and prepare the camera data to create a record
table and spatial detection history of sugar glider observations.
Capture-history, camera operation and trap location data were prepared
for spatially explicit capture-recapture analysis. Sampling occasions
were defined as commencing from midday to centre detections on nights as
sugar gliders are nocturnal. Capture histories were filtered to a single
detection per individual per night at each detector. Varying effort
(e.g., camera malfunctions) were accounted for by including camera usage
in our capture history.
To evaluate how our capture and trapping results compared to published
literature, we calculated log-transformed incidence rates for our data
per the steps outlined in the meta-analysis (Supporting Information).
SECR models – SECR analyses were performed using the ‘secr’ package v.4.5.8 (Efford, 2023). The suggested buffer width (439m) from
‘secr’ was revised upwards to 532m (4 × σ ) based on
exploratory modelling and confirmed against an effective sampling area
(ESA) plot. The habitat mask was created using this buffer distance with
a 40m spacing, and any non-forest (detected from satellite imagery)
boundaries were excluded. We assumed the population was closed as the
total sampling period (30 nights per survey; 60 nights total) was brief
compared to sugar glider longevity (5-7 years) and recruitment of den
young (3-4 months) (Jackson, 2000b, Smith, 1973, Suckling, 1984).
Factors influencing probability of detection - Using data from
the entire survey, we contrasted the impact of: bait type (i.e., effect
of bait-type was constant over time, but different between bait types)
and rebaiting period (i.e., effect of bait-type was constant within each
rebaiting period, but different between bait type and rebaiting period)
with other environmental variables that might influence the probability
of detection (g0 ) and movement parameters (σ) (i.e., min and max
temperature, tree species and tree height). In total, 12 models were fit
(plus the null model, Table 1). Models of a linear time trend failed to
converge and were not considered further. We compared model results to
the null model and selected the best model based on Akaike’s Information
Criterion corrected for small sample size (AICc, Buckland et al. 1979).
Models with a difference in AICc <2 were considered equivalent
(Burnham and Anderson, 2002).
Estimating density - Data were subset to the period when only the
fish-bait was used to explore behavioural models. We held σ (σ )
parameters constant and investigated the influence of the following
factors on g0 : b , learned response; B; transient
response; k , site response; bk , animal x site learned
response, and; Bk , animal x site transient response. We estimated
density with our preferred parameterization (based on AICc model
selection).
Home range statistics were estimated using all data. We calculated the
mean maximum distance moved pooled across individuals (MMDM) and core
home ranges (i.e., root pooled spatial variance; RPSV) (Borchers &
Efford, 2008).