Assessing the detectability of a cryptic arboreal marsupial by using a novel survey approach

Introduction

Accurate detection of presence or absence is important for the conservation and management of fauna species (Burns et al. 2019). Environmental impact assessments rely on accurate species detection to ensure that species are properly assessed, appropriate mitigation measures are applied, and any biodiversity losses resulting from development are effectively offset (Brownlie et al. 2013). In such instances, methods need to be species-specific, with a high level of accuracy, and appropriate to employ in a cost-efficient manner. The level of survey effort required to determine the presence or true absences at a location can be significant for some species, particularly for those that are considered to be cryptic (Burns et al. 2019; Harley and Eyre 2024). Small arboreal mammals are often difficult to survey by using conventional techniques such as trapping and spotlighting (Moore et al. 2021). These methods have known sampling biases (Tasker and Dickman 2001), even for terrestrial fauna (Johnstone et al. 2021), and high rates of false absences for arboreal mammals (Wintle et al. 2005).

The arboreal eastern pygmy possum (Cercartetus nanus) is a threatened species native to eastern Australia and is commonly considered to be cryptic and hard to survey (Bladon et al. 2002; Harris and Goldingay 2005b; Harris et al. 2007b). It generally occurs in low-density populations (Bowen and Goldingay 2000), which may be missed and assumed absent because of inappropriate survey techniques or survey effort. Despite its seemingly widespread occurrence, the eastern pygmy possum is rarely recorded during fauna surveys, particularly using standard mammal survey techniques (e.g. Elliott trapping and spotlighting) (Bowen and Goldingay 2000). For example, Suckling (1978) used spotlighting, ground-based Elliott trapping and trialled the use of baited hair tubes (30 mm wide PVC pipe open at both ends) at a survey site where the species was known to occur; yet, no individuals were recorded. Bowen and Goldingay (2000) reviewed previous surveys in New South Wales (NSW) and found at that time, that only five extensive surveys recorded more than 10 individuals. Similarly, other burramyids such as the little pygmy possum (Cercartetus lepidus) and long-tailed pygmy possum (Cercartetus caudatus) appear difficult to detect, owing to the lack of studies recording them; however, pitfall traps have shown success for surveying the western pygmy possum (Cercartetus concinnus) (Pestell and Petit 2007) and the little pygmy possum (Ward 1992; Duncan and and Taylor 2001).

Traditional methods for detecting eastern pygmy possums, such as Elliott and pitfall trapping, have shown limited and inconsistent success. Elliott trapping appears more effective when used repeatedly in the same location (Ward 1990; Laidlaw and Wilson 1996), particularly near flowering banksias (Bowen and Goldingay 2000; Harris and Goldingay 2005a), although this increases effort and the standard bait mix may be inadequate because it competes with natural foods. Pitfall traps have reported higher capture rates than spotlighting or Elliott trapping in some studies (Bennett et al. 1989; Bowen and Goldingay 2000; Tasker and Dickman 2001; Tulloch and Dickman 2006); yet, large-scale surveys in NSW State Forests recorded only one capture from over 10,000 pitfall trap nights (Bowen and Goldingay 2000). Spotlighting, although common for arboreal mammals, is considered less reliable for this species (Davey 1990; Kavanagh and Webb 1998). The literature offers little clear guidance on the most effective survey methods, and the species is regarded as ‘very difficult to detect, especially via spotlighting’ (eastern pygmy possum; Ecological Data, BioNet Atlas, NSW). Traditional trapping techniques are also labour-intensive and potentially harmful to animal welfare (Garden et al. 2007). Given this and the low detectability of the eastern pygmy possum with these methods, alternative reliable approaches are needed.

There has been an increase in the use of, and it has been suggested that the species is most reliably detected in nest boxes (Bowen and Goldingay 2000; Goldingay 2023). Nest boxes have been successfully applied to study the effects of fragmentation on demography of the eastern pygmy possum (Bladon et al. 2002), life history (Ward 1990) and the effects of habitat, movements and social organisation (Harris and Goldingay 2005b; Law et al. 2013; Harris et al. 2014; Goldingay and Keohan 2017; Goldingay and Rueegger 2018; Law et al. 2018; Goldingay 2019, 2023). Previous studies have also investigated optimal box design to record eastern pygmy possums (Beyer and Goldingay 2006; Rueegger et al. 2012). A recent study found a detection probability of 0.35 per survey visit for a cluster of five nest boxes (Chew et al. 2024). However, like conventional techniques, nest boxes can represent a labour-intensive survey method because of the ongoing need for physical checks at a site, ongoing maintenance, and these have not always proven effective for detection, particularly in areas where hollows are abundant (Harris and Goldingay 2005a).

Camera traps are rapidly becoming a preferred tool for recording mammal species (Bowler et al. 2017). Camera traps have been found to be more effective than live trapping (Bondi et al. 2010) and hair tunnels (Paull et al. 2011) for detecting small mammals, and despite high initial costs, camera traps have been found to be less costly than labour-intensive conventional methods in the long term (Welbourne et al. 2015). Even though camera trapping can be as effective and efficient in determining occupancy for some arboreal mammals as it is for terrestrial species (Harley and Eyre 2024), few studies have specifically used camera traps to detect the eastern pygmy possum. Cameras have been used to identify behaviour and assess nest box preferences of eastern pygmy possums (Rueegger et al. 2012). Another study in Jervis Bay (NSW) used a range of standard methods (Elliott and cage traps) but only detected the eastern pygmy possum on camera traps on the ground by using drift netting (Harris et al. 2007). Cameras with infrared triggers have also been used to record pollinators on banksia inflorescence and successfully recorded eastern pygmy possums (Carthew 1993). Infrared cameras also found that eastern pygmy possums were the most frequent small mammal visitor spending significant periods at inflorescence of heath-leaved banksia (Banksia ericifolia) in bushland reserves in northern Sydney (Saul 2013; O’Rourke et al. 2020), suggesting cameras focussed at resources may optimise detections.

In this study, we assess the efficacy of camera traps for detecting the eastern pygmy possum at nest boxes and flowering food resources, in all seasons and over a 5-year period that spanned a range of climatic conditions, including a significant drought that resulted in the 2019/2020 megafires across eastern Australia, although the study area was not burnt. The use of cameras has the potential to reduce field effort to detect cryptic species, given that these can be set up to collect several months of data from one installation. We modelled detection probability by using camera traps focused on banksia inflorescences and nest boxes (artificial hollows), to assess which method is best to maximise the likelihood of detection for eastern pygmy possums when using cameras. We also assessed how detection probability varied among seasons and years. Cumulative detection probability curves were generated to guide survey effort required to achieve a high degree of confidence in absences for the species at surveyed sites.

Materials and methods

Our study area comprised seven bushland remnant localities on the urban edge in northern Sydney, Australia, located within the Northern Beaches local government area on the Ingleside escarpment (Fig. 1). This area is dominated by a mixture of urban and semi-rural land uses interspersed with remnant vegetation. Vegetation at the seven localities comprised heathland or heathy woodland on sandstone dominated by flowering Myrtaceae and Proteaceae species, with an abundance of Banksia spp. Cameras were installed at or near nest boxes that had successfully captured eastern pygmy possums as part of an ongoing population study (C. Thompson, unpubl. data) (Table 1). Nest boxes were made from salvaged hollows of varying sizes (but with a minimum internal diameter of 10 cm), end caps made of steel, and a drilled opening of 3 cm near the top of the box. These were installed between 0.5 and 1.8 m from the ground on banksia or eucalypt trees, to allow for checking without a ladder. Groups of 5 or 10 nest boxes were installed at each bushland remnant, along a transect at a spacing of 50–100 m apart. Given that a typical home range for an eastern pygmy possum is approximately 3–4 ha (Harris et al. 2007a; Law et al. 2013), each transect was separated by at least 500 m, where there was no impermeable barrier (e.g. large roads, development and fences). Not all nest boxes were used for this study, because these were installed and are being monitored for an ongoing population study (C. Thompson et al., unpubl. data) associated with road crossing structures (‘underpass’ and ‘overpass’). The crossing structure study had not recorded any tagged eastern pygmy possums crossing the road over the 8 years of survey, thus the ‘north’ and ‘south’ localities associated with the crossing structure locations were considered independent for this study. Similarly, Goldingay (2023) recorded only two road crossings (road width of approximately 10 m) between nest boxes from >100 tagged individuals.

Fig. 1.

The study area and the location of nest boxes in northern Sydney.

Reconyx hyperfire (HC600 and PC800) covert infrared cameras were set to detect eastern pygmy possums by directing them at either flowering banksias or nest boxes. In total, 16 camera locations at nest boxes, and 12 camera locations at banksia inflorescence in proximity to nest box locations were established. Cameras sampling flowering banksia were installed in proximity to installed nest boxes (within 20 m); however, both camera types were not used concurrently. The cameras have inbuilt passive infrared LED motion sensors and were installed ~0.5 m in front of the boxes or banksia inflorescence on trees and secured with straps, with a high passive infrared (PIR) sensitivity. Cameras were not changed for close-proximity photos. On the basis of pilot testing of the methods, this appeared to be a sufficient distance to avoid overexposed images and to allow for identification of all mammals to species level. Cameras were set to record a sequence of 10 photos; however, this was reduced to five photos after the first year of survey given the number of false triggers and number of images to process. A quiet period of 1 min was set before they could be triggered again. For each site (nest box/banksia inflorescence), a detection of at least one eastern pygmy possum per night was scored as ‘presence’. This was done to reduce counting the same individual multiple times and to provide a ‘detection’ or ‘non-detection’ score per survey night per camera for detection modelling. Capture rates at each camera location were calculated on the basis of the number of detections (i.e. one image of an eastern pygmy possum per camera per night) divided by the total camera nights and expressed as a percentage.

Heath-leaved banksia was the dominant banksia species in the study area. This species is a known important local food resource for eastern pygmy possums (Tulloch and Dickman 2007; Goldingay and Keohan 2017; Goldingay 2023) and its inflorescences produce copious amounts of sugary nectar at night (Carpenter 1978). Heath-leaved banksia flowers in autumn through winter and sometimes into spring (Copland and Whelan 1989; Goldingay 2023); thus, cameras were installed only within these seasons for the banksia cameras. Flowering duration of individual inflorescences varies considerably, but may be less than 4 weeks (Copland and Whelan 1989). Banksia inflorescences were targeted for survey on the basis of flower maturity and nectar secretion, and to allow a minimum recording period of 30 days, so as to allow for the highest nectar availability to be sampled over the camera deployment across autumn, winter and spring (Fig. 2). Nest box cameras were not limited to the flowering season of heath-leaved banksia and recorded across several seasons. Images were individually processed using an image viewer and tagged. Detected fauna were identified to species level where possible, using reference images and texts (Van Dyck and Strahan 2008). Data were then collated as a detection history (presence/absence) for input into the modelling.

Fig. 2.

Covert infrared camera trained on a flowering heath-leaved banksia (Banksia ericifolia) inflorescence in northern Sydney (left). Covert infrared camera trained on a salvaged log nest box in northern Sydney (right).

An occupancy model was used to assess detection probabilities by using data generated by cameras at the seven bushland localities, with data collected over 5 years. Because camera numbers for the study were limited, cameras were not installed consistently within localities over the survey period, but shifted among nest boxes and banksia flowers across the survey period, resulting in detection/non-detection data and periods of no survey for each camera location (Table 2). To ensure independence, detection histories were prepared for each camera location for input into the model. A ‘site’ was included for each camera type (banksia or nest box) within a locality (e.g. Laurel Road Reserve), resulting in one record (detection/non-detection) or NA if the camera was moved to another site, per night. In total, 32 sites were sampled. Data analyses were run using the RPresence package (ver. 2.15.17; http://www.mbr-pwrc.usgs.gov/software/presence.html; MacKenzie and Hines 2023), by using a single season occupancy modelling framework, because we were modelling only detection probabilities. Site occupancy modelling accounts for the potential for imperfect detectability (i.e. false absences that result when a species is present but not detected), by using information from repeated observations at sample sites occupied to estimate detection probabilities (Mackenzie 2005).

Detection probability (p) was modelled with occupancy (psi) held constant. Detection was modelled to compare covariate models with a null model. We assessed single covariates and built on the most supported single covariate model by adding an additional covariate until there was no improvement (i.e. adding an additional variable did not improve the top model by >2 AIC points). We included year as a covariate in all detection models to account for any changes in detection probability associated with year of sampling. Variables assessed were season (austral spring, summer, autumn and winter) and survey method (camera on nest box or camera on banksia inflorescence). Banksia inflorescence could be sampled only during particular seasons, i.e. when flowering; however, nest boxes were sampled across a range of seasons. Detection models were ranked from the lowest to highest on the basis of Akaike’s information criterion (AIC) (Burnham and Anderson 2004), with the difference in AIC (ΔAIC) calculated between each model and the top-ranked model. Models with ΔAIC of <2 were considered equally plausible to explain the data (Burnham and Anderson 2004).

To evaluate the minimum survey effort required to confidently detect the eastern pygmy possum at a site when present, we calculated the cumulative detection probability (P) by using the standard formula from MacKenzie and Royle (2005), as follows:

where

This approach allowed us to estimate the number of repeated surveys required to achieve a desired cumulative detection probability (typically ≥0.95) under different conditions and detection methods, in this case for different detection methods (e.g. cameras trained on banksia inflorescences vs nest boxes), stratified by season and year to reflect observed variability. Given the strong effect of year of survey on detection probability, estimates from 2020 were chosen to guide minimum survey duration to achieve high detection probability (95%) for eastern pygmy possum. This year was chosen because it provided an intermediate detection probability value (i.e. not the extreme ends) and represented a year where both banksia and nest box cameras were used.

Results

During 2018–2022, 133 eastern pygmy possum detection nights were recorded by using cameras at nest boxes (during 2147 camera nights), whereas cameras at banksia inflorescences recorded 30 eastern pygmy possum detection nights (703 camera nights) (Fig. 3). Detection nights decreased across the survey period for nest box cameras and banksia cameras (Fig. 4).

Fig. 3.

Example images of the eastern pygmy possum (Cercartetus nanus) by using covert infrared cameras trained on nest boxes (left) and flowering banksia inflorescence (right).

Fig. 4.

Nightly eastern pygmy possum (Cercartetus nanus) activity (total detections/total camera type × 100 per year) across years for both banksia and nest box cameras in northern Sydney. NA, no banksia cameras were used in 2018 and 2019 and only one nest box camera was used in 2022, but did not get any detections in the period deployed.

Modelling showed that a single model explained the data better than any other (Table 3). This additive model allowed detection probability to vary among years of survey, among seasons and between surveys by using cameras trained on banksia inflorescences or nest boxes. The detection probability was highest in winter and lowest in spring (Fig. 5a). Modelled detection probability was more than 2.7 times higher for cameras on banksia inflorescences than for cameras on nest boxes (Fig. 5b). This trend was found for all years in which both survey methods were employed (Fig. 6). Detection probability for banksia cameras in winter ranged from 0.23 in 2000, to 0.03 in 2022, and for nest box cameras in winter from 0.29 in 2018 to 0.01 in 2022 (Fig. 6).

Fig. 5.

(a) Detection probability versus season for eastern pygmy possum (Cercartetus nanus) in northern Sydney by using cameras (assuming sampling was undertaken in 2020 and with camera on banksia). (b) Detection probability versus camera type for eastern pygmy possum in northern Sydney (assuming sampling was undertaken in 2020 and in winter).

Fig. 6.

Influence of survey year on detection probabilities for eastern pygmy possum (Cercartetus nanus) in northern Sydney by using banksia and nest box cameras.

Minimum camera night requirements at an occupied site to be 95% confident of detecting eastern pygmy possum with a camera trained on a banksia inflorescence in winter was 28 nights (assuming detection probabilities recorded in 2020) or, on average, 117 nights (when detection probability was averaged across the 3 years of surveys) for one camera. The minimum level of sampling needed to achieve a similar detection probability for a camera on a nest box was 85 nights in winter (assuming detection probabilities recorded in 2020) or 237 nights, when averaging across the 5 years of survey for one camera (Fig. 7).

Fig. 7.

Survey duration versus cumulative detection probability (2020 + season + banksia, with occupancy held constant) for the eastern pygmy possum (Cercartetus nanus) in northern Sydney bushland remnants on the basis of single-visit detection by using nest box cameras or banksia inflorescence cameras in winter.

Discussion