Introduction

Landscape modification due to urbanisation, agriculture and other human activity is one of the leading causes of biodiversity decline and species range contractions globally (Pereira et al. 2012; Venter et al. 2016). In modified landscapes, native vegetation is reduced to smaller and more isolated patches (Fahrig 2003; Hobbs and Yates 2003) that may be degraded due to changes in age structure or species composition, and invasion by exotic species (McKinney 2002; Holland and Bennett 2010; Fischer and Schröder 2014). The loss, degradation and fragmentation of native vegetation can make landscapes less suitable to native wildlife (Fahrig 2003; Dargan et al. 2019). Furthermore, higher levels of anthropogenic noise and light pollution associated with modified landscapes can alter behaviour and be a cause of chronic stress for wildlife species. This can result in reduced breeding rates or an increase in disease (Blickley and Patricelli 2010; Taylor-Brown et al. 2019).

Although some native species may thrive in modified landscapes (Chace and Walsh 2006; Holland and Bennett 2010), many undergo population declines or become locally extinct (Fischer and Schröder 2014). Species that persist, typically reduce their ranges to the highest quality or largest remaining patches, and limit movements between them (e.g., Holland and Bennett 2010). Populations of agile antechinus (Antechinus agilis) for example, require habitat patches greater than 15 ha, rather than several isolated patches (Holland and Bennett 2010). When landscape modification is coupled with an increase in human activity, animals may modify their movement behaviour or exhibit increased vigilance (Frid and Dill 2002; Strasser and Heath 2013). In Canada, wolverines showed heightened vigilance and were less likely to linger near tourism centres, than at remote undisturbed sites (Stewart et al. 2016). Vigilance behaviour can result in a trade-off of other important behaviours such as feeding or mating (Frid and Dill 2002; Strasser and Heath 2013). The resulting increase in stress, if long-term, may then have direct negative effects on an individual’s health (Tkaczynski et al. 2014).

The koala (Phascolarctos cinereus), an arboreal marsupial with a specialised diet and limited energy budget, is especially susceptible to landscape modification. In the eastern states of Queensland, New South Wales and the Australian Capital Territory, loss and degradation of habitat is an ongoing threat to koala populations. Combined with climate change impacts, this has resulted in the species being listed as endangered across these states (Department of Agriculture, Water and the Environment 2022). In contrast, koala populations in Victoria and South Australia are thought to be secure; however, there has been little monitoring and in many locations populations are probably declining similarly to those elsewhere (Whisson and Ashman 2020). Habitat loss, fragmentation and degradation can result in koalas spending more time on the ground moving between trees and habitat patches (Rus et al. 2021). This may have high metabolic cost, and result in an increase in ectoparasite load (Kwak and Reed 2017) and risk of dog attack and vehicular collision (Schlagloth et al. 2021). Fragmentation of habitat can also disrupt movement between populations resulting in reduced genetic and demographic diversity (Lee et al. 2010) and increasing the potential for either localised extinction or overabundance (Whisson and Ashman 2020). Determining the influence of landscape modification on the distribution of koalas in a region is critical for developing conservation programs that protect important habitats and that create links to permit safe movement of koalas between habitat patches.

Several studies have been undertaken in attempts to identify factors influencing the distribution of koalas at regional and local scales (McAlpine et al. 2006b, 2008; Rhodes et al. 2006; Januchowski et al. 2008; Sequeira et al. 2014; Predavec et al. 2015; Law et al. 2018; Ashman et al. 2020b). At large regional scales, elevation, climate factors, fire history, vegetation types and forest cover perform well as predictors of distribution (Sequeira et al. 2014; Predavec et al. 2015; Law et al. 2017; Ashman et al. 2020b; Heard and Ramsey 2020), whereas at local scales, the presence of preferred food trees, patch size, habitat connectivity and road density become more important (McAlpine et al. 2006b; Rhodes et al. 2006; Januchowski et al. 2008). However, the influence of these local-scale factors on koala populations varies between regions and different landscape types such that a regional approach to developing distribution models and conservation strategies is recommended (McAlpine et al. 2008).

Until recently, developing predictive models for koalas was hindered by the high labour cost and low detection probability (especially in areas of low koala abundance) of traditional koala survey methods (scat searches, observer surveys). With the development of passive acoustic methods for determining site occupancy by koalas, it is now possible to rapidly survey multiple sites to determine with high confidence, site occupancy by koalas (Hagens et al. 2018; Law et al. 2020). In a study of koala distribution in the north-east forests of New South Wales, Law et al. (2020) found that passive acoustic surveys were three times more effective than scat surveys in detecting koalas. Although passive acoustic surveys have some limitations (e.g., surveys are restricted to the breeding season; Law et al. 2020), their use may facilitate a better understanding of the distribution of koalas, especially across regions where population densities may be low.

In southern Australia, koalas occur at highly variable population densities across landscapes that have been highly modified due to urbanisation, rural land uses and the establishment of commercial hardwood plantations (Sequeira et al. 2014; Heard and Ramsey 2020; Whisson and Ashman 2020). Most research on factors influencing distribution of koalas has focused on those populations that occur at high densities (for example, Ashman et al. 2020b), with only one study of a low-density koala population in a rural landscape (Januchowski et al. 2008; McAlpine et al. 2008; Rhodes et al. 2008). There remains a significant gap in our knowledge of koala populations and their distribution in landscapes modified by urbanisation and rural land uses.

We therefore investigated koala distribution on the Mornington Peninsula, Victoria, a region that comprises a mosaic of highly urbanised and rural land uses. We used passive acoustic surveys to determine the relationships between koala presence and site and landscape variables, and subsequently to derive a predictive model of koala distribution for the region. By identifying areas where koalas have a high probability of presence, conservation strategies that aim to conserve and potentially connect populations can be developed. Such strategies may have substantive additional benefits for other animal communities threatened by habitat modification.

Methods

The Mornington Peninsula (~ 73,000 ha) is a region that has been subject to extensive landscape modification for more than 150 years (Antos and White 2004). It comprises areas of intensive urban development (primarily along the western coast), small townships, agricultural land, bushland reserves and coastal environments. Approximately 30% of native vegetation remains. Reserves cover ~ 10% of the Peninsula, supporting ~ 27% of the total area of native vegetation (Schmidt et al. 2016). Approximately 80% of land is privately owned. Much of the region is flat, with the highest point being 305 m at Arthur’s Seat.

The region has a mild and temperate climate with a mean monthly temperature between 12.2 and 25.1 °C and mean annual rainfall of 740.2 mm (Bureau of Meteorology, http://www.bom.gov.au/climate/averages/tables/cw_086079.html). Koalas are indigenous to the region, although the current low-density population is thought to be derived from koalas translocated from French Island in the mid-1900s as part of a broad scale conservation program to re-establish koalas on the Victorian mainland (Department of Sustainability and Environment 2004; Menkhorst 2008).

We surveyed koalas using passive acoustic monitoring (sensu Hagens et al. 2018) at 123 sites across the Mornington Peninsula (Fig. 1). We used a systematic approach to select sites, ensuring that they were at least 1 km apart to minimise the potential of a koala being detected simultaneously at more than one site. Koala bellows are detected by audio recorders for between 100 and 300 m; (Law et al. 2017; Hagens et al. 2018) and in Victoria, koalas typically have small home ranges of less than nine hectares (Whisson et al. 2016, 2020; Ashman et al. 2020a). Each site was sampled once between September and December 2019 during the species breeding season when male koalas are vocalising (Hagens et al. 2018). At each site we deployed a single autonomous recording unit (ARU; Song Meter SM3 & SM4; Wildlife Acoustics, Maynard, Massachusetts, USA; https://www.wildlifeacoustics.com/). Each unit was attached to a tree ~ 1.5 m above ground and programmed to record continuously from 20:00 to 02:00 h each day for a seven-night period as recommended by Hagens et al. (2018). We visually scanned spectrograms of each 1 h recording in Audacity (Version 2.3.2, https://www.audacityteam.org/) and determined the presence or absence of koala vocalisations in each recording. We scored inclement weather (wind or rain) for each 1 h recording on a scale of 0—3 (0 = calm conditions with 0% spectrogram containing ambient noise, 1 = wind and rain causing ambient noise in 1–20% of spectrogram, 2 = wind and rain causing high levels of ambient noise in 21–50% of spectrograms; and 3 = wind and rain causing high levels of ambient noise in > 50% of spectrograms (sensu Whisson et al. 2021).

Fig. 1
figure 1

Study sites at which passive acoustic surveys for koalas were conducted on the Mornington Peninsula, Victoria, Australia

Full size image

We considered factors that might influence presence of koalas in a location at two spatial scales: site (within 100 m of the ARU) and landscape (within 1 km of the ARU). We chose 100 m for the site scale as the distance at which bellows are detected by ARUs (Hagens et al. 2018) and 1 km for the landscape scale to encompass the area of a koala’s home range in Victorian landscapes (Whisson et al. 2016, 2020; Ashman et al. 2020a).

Site-scale factors included tree cover (STC, %), area of ‘disturbed’ land uses (DLU, %), and daytime anthropogenic sound (Normalised Difference Sound Index [NDSI]). We also recorded percentage composition of koala food tree species but did not consider this in analysis due to its low variation amongst sites; 93% of sites comprised known food tree species (E. ovata, E. viminalis, E. radiata). Tree cover was estimated from aerial photos (50 cm resolution). Pixels were classified as trees or ‘other’ using the visible green band, and the percentage of the site that was ‘tree’ pixels was calculated. Disturbance was estimated by intersecting each site buffer with a land use layer (Department of Jobs, Precincts and Regions 2016) and calculating the percentage of the buffer area classified as residential, commercial or infrastructure. NDSI is a measure of anthropogenic disturbance on a soundscape and computed as the ratio of human-generated (anthrophony) to biological (biophony) acoustic components of sound. It ranges from -1 (high anthrophony) to + 1 (high biophony) (Kasten et al. 2012). We programmed ARUs to record for five minutes per hour between 06:00 and 20:00 h. Using these recordings, we then estimated NDSI (Kasten et al. 2012) using the ‘Soundecology’ package in R (Villanueva-Rivera et al. 2011; R Development Core Team 2019).

Landscape-scale factors considered included tree cover (LTC, %), sealed road density (SRD), watercourse length (WL), average Euclidean nearest neighbour distance (ENN) between tree patches, and the mean core area index (CAI_MN) of tree patches. The core area index (CAI) is the percentage of a patch that is not within a user-specified distance of the patch edge and therefore provides a measure of patch spatial configuration. CAI is zero when a patch does not have a core area and approaches 100 for large patches with large interior areas (McGarigal et al. 2002). CAI_MN was chosen as it does not correlate with habitat abundance nor configuration metrics and is highly recommended as a measure of landscape fragmentation (Wang et al. 2014; McGarigal 2015; Garcia and Ballester 2016). We specified an edge of 30 m, based on the 30-m resolution of our tree cover layer. Site tree cover and area of disturbed land uses, and all landscape-scale factors were derived in a Geographic Information System (ESRI 2011. ArcGIS Desktop: Release 10.2. Redlands, CA: Environmental Systems Research Institute), using digital spatial layers for native vegetation, roads, watercourses, and land use (https://datashare.maps.vic.gov.au/). Fragstats was used to derive estimates of ENN and CAI_MN (McGarigal et al. 2002).

Statistical analysis

We first checked for collinearity between variables; all correlations were ≤ 0.5 so all variables were retained. We then used single-season single-species occupancy models to identify factors influencing site occupancy by koalas while accounting for imperfect detection (MacKenzie et al. 2002). Inclement weather was considered as a factor influencing detection probability; and all site- and landscape-scale factors considered as factors influencing site occupancy. We also considered interactions between fragmentation indices and NDSI. We used a sequential approach where we first considered factors influencing detection probability and then used the best supported model to consider factors influencing site occupancy. We then considered all possible combinations of variables influencing site occupancy. Akaike Information Criterion corrected for small sample sizes (AICc) was used to determine the most parsimonious model. Models with a delta AIC < 2 were considered supported. Where model selection uncertainty existed, we calculated model averaged parameter estimates to determine which variables were considered to have the greatest influence. Nagelkerke's (1991) R-squared index was calculated to quantify the proportion of variation explained by each model. Models were validated using the Mackenzie-Bailey goodness-of-fit statistic (MacKenzie and Bailey 2004). Occupancy models were run in the ‘unmarked’ package in R (Fiske and Chandler 2011; R Development Core Team 2019). We used our strongest predictors of occupancy to predict probability of occurrence (and Standard Error) at 100 m resolution across the Mornington Peninsula.

Results

Male koala bellows were detected at 31% (39/123) sites with bellows present on an average of 3 ± 0.29SE (range: 1—7) of the seven sampling nights at those sites. Nightly detection probability was 0.363 with 95% confidence in a site absence being achieved after 6.6 nights of sampling, thereby supporting our use of a seven-day sampling period.

Factors influencing koala presence

Our best supported model included the influence of inclement weather on detection (AIC ω = 1.0) and the influence of tree cover, road density and CAI_MN on koala site occupancy (AIC ω = 0.084). There was model selection uncertainty with support also given to an influence of ENN, NDSI and an interaction between these on-site occupancy (∆AIC = 1.63). Model averaging identified three variables that had the greatest influence on koala distribution: inclement weather influencing detection probability, and tree cover and road density (within 1 km of a site) influencing koala site occupancy. Koala detection probability declined with increasing inclement weather (β = − 0.561, 95%CI: − 0.842 to − 0.281). Site occupancy increased with increasing tree cover (β = 1.007, 0.315 to 1.698; Fig. 2) and declined with increasing road density (β = − 0.896, − 1.709 to − 0.083; Fig. 2). The rate of change in site occupancy probability was greatest for road densities greater than 1 m per hectare (β < − 1.25) and for forest cover between 30 and 68% (β > 1.0). Model averaged parameter estimates showed limited support for an influence of CAI, ENN and NDSI (Table 1). Our models were not overdispersed (X2 = 135.700, p = 0.249, ĉ = 1.08).

Fig. 2
figure 2

Relationship between koala site occupancy and road density (A) and tree cover (B) across the Mornington Peninsula, Victoria Australia

Full size image
Table 1 Model coefficients, standard errors (SE) and 95% confidence intervals based on full-model model averaging
Full size table

Predicting koala distribution

We used landscape tree cover and road density (i.e., the best predictors of koala occupancy) to predict the probability of koala presence (with Standard Error [SE]) in one-hectare grid cells across the Mornington Peninsula (Fig. 3). High probability of presence (≥ 75%) occurred across 51km2 (7.08%) with a further 131km2 (18.10%) predicted to have 50 to 75% probability of presence.

Fig. 3
figure 3

Predicted site occupancy and standard error of prediction of koalas across the Mornington Peninsula, Victoria, Australia

Full size image

The standard error of prediction probabilities averaged 0.058 (range: 0.005 to 0.22), with the least uncertainty (< 0.05 SE) associated with areas of high tree cover or urban areas, and highest uncertainty (> 0.1 SE) at the interface of highly urbanised and treed areas.

The proportion of land predicted to have high probability of koala presence was more than three times higher for public land (17.38%; including transport and communication infrastructure) than land under private tenure (4.75%). High probability (≥ 75% probability of presence) grid cells occurred in 11 distinct clusters, nine of which were greater than 100 ha (range 17 to 1840 ha; Fig. 4). Six of these were associated with reserves: the Mornington Peninsula National Park (2540 ha), Main Ridge Nature Conservation Reserve (64 ha), Arthur’s Seat State Park (617 ha), Devil Bend Natural Features Reserve (1005 ha), Mt Martha Nature Conservation Reserve (53 ha); and another was associated with the HMAS Cerberus Commonwealth property (1469 ha). Of the land under private tenure for which we predicted high probability of koala presence, 44.51% was classified as grazing, 23.31% was rural residential, 13.88% was cropping, and the remainder comprised numerous other diverse land uses (e.g., quarries, horse studs, poultry farms).

Fig. 4
figure 4

Areas having high (≥ 75%) predicted probability of koala presence relative to parks, reserves, and crown land on the Mornington Peninsula

Full size image

Identifying areas for conservation actions

Occupancy modelling indicated that the probability of site occupancy by koalas declined with increasing road density from 1 m road length per hectare in the landscape (within 1 km) around the site. Conversely, the probability of site occupancy increased with increasing tree cover in the landscape, with the greatest increase occurring where forests and woodlands already cover 30% to 68% of the landscape around a site. Based on these relationships, increasing tree cover in landscapes where there already is 30% to 68% tree cover and a road density of < 1 m per hectare would have the greatest positive influence on site occupancy by koalas (Fig. 5). Approximately 85 km2 (11%) of the Mornington Peninsula meets these criteria, with 87% of that land under private tenure.

Fig. 5
figure 5

Locations on the Mornington Peninsula where road density is low (< 1 m per hectare) and where tree planting to increase the amount of tree cover in the landscape may have the greatest influence on increasing the probability of site occupancy by koalas (existing tree cover 30 to 68%)

Full size image

Discussion

Our results confirm the overarching importance of landscape-scale factors in determining koala site occupancy and distribution in modified landscapes. We found that regardless of tree cover, land use, and soundscape at a site, the probability of koalas being present was positively influenced by landscape scale tree cover and negatively influenced by road density. In modified landscapes where significant habitat clearance has occurred, habitat on public land (e.g., treed parks and woodland or forest reserves) are therefore critical habitat for koalas, and a source of koalas to smaller habitat patches in the surrounding landscape. On the Mornington Peninsula, there are a number of large reserves (e.g. Arthur’s Seat State Park, Mornington Peninsula National Park) with extensive tree cover providing important koala habitat. Although conservation of koala populations in these protected areas is important, actions to increase tree cover on private land which typically accounts for a large proportion of the total land area in modified landscapes, may be critical for regional conservation of koalas. For example, habitat created on private land may provide important connectivity between larger patches, and be less susceptible to large scale fire. On the Mornington Peninsula, our predictions show that probability of koala occupancy is highest in protected areas and on a small proportion of land used for grazing but that opportunities exist for increasing tree cover on rural properties.

The influence of landscape-scale tree cover on koala populations has been observed in other regions where landscapes have been modified due to urbanisation or rural activities (McAlpine et al. 2006b, a; Rhodes et al. 2006; Dargan et al. 2019; Rus et al. 2021). An increasing extent of tree cover in the landscape increases food and habitat resources for koalas and may also result in greater patch connectivity (McAlpine et al. 2006b, a; Dargan et al. 2019; Rus et al. 2021). In our study, fragmentation only had a weak influence on site occupancy by koalas. However, this may be due to the consistently high level of habitat fragmentation across the Mornington Peninsula, resulting in low variation in this factor across our sites. Most (76%) sites were within landscapes having a mean core area index of between 20 and 30%, with tree cover primarily in small blocks or along roadsides. Distances between patches (measured by average distance between patches [ENN]) tended to be short with most (70%) sites having a mean ENN of less than 60 m. However, where habitat is fragmented, koalas have been observed to move more frequently across open ground to access resources (Rus et al. 2021). In addition to increasing energy costs, this places them at a high risk of mortality due to vehicle collisions and predation by dogs and misadventure (Schlagloth et al. 2021). Schlagloth et al. (2021) reported dog attacks and vehicular collisions as the most frequent reasons for koalas to be admitted to wildlife shelters in Victoria. Of particular relevance to our study, they analysed 495 records of koalas admitted to a Mornington Peninsula shelter from 1987 to 2006. Of those, almost 70% had suffered a vehicular collision and a further 28% had been attacked by dogs.

A negative influence of road density on koala distribution also has been observed elsewhere (Rhodes et al. 2006, 2014; Sequeira et al. 2014). High road density is generally associated with urbanised areas where suitable habitat for koalas may be sparse and the risk of mortality is high (Rhodes et al. 2006; De Oliveira et al. 2014; Whisson et al. 2020; Schlagloth et al. 2021). Although conservation actions to increase tree cover and encourage koalas into urban areas are often undertaken, these may have poor outcomes for koalas (De Oliveira et al. 2014). High mortality may also occur where important koala habitats abut urbanised areas. Actions (e.g. fencing along major roads) to mitigate koala mortality are therefore warranted in these areas. On the Mornington Peninsula, an area of concern is the western boundary of Arthur’s Seat State Park which is bounded by a highly urbanised area as well as the Mornington Peninsula Freeway, a road that carries considerable traffic.

The absence of influence of any site-scale factors is not surprising given that all sites contained some Eucalyptus cover and food trees for koalas, and that land use classification may not necessarily reflect levels of disturbance at a site. Koalas are known to utilise a continuum of tree cover from scattered paddock trees and roadside vegetation to large, contiguous forest and woodland patches (Barth et al. 2020; Rus et al. 2021). Whisson et al. (2020) reported that roadside trees provide important habitat to koalas in a small township on the Mornington Peninsula. The fact that koalas occur in modified landscapes and urban areas indicates that they can tolerate anthropogenic disturbance although the impact of factors such as noise and human activity on their behaviour and health have not been examined. We calculated a bioacoustic index (Normalised Difference Soundscape Index [Kasten et al. 2012]) from our daytime recordings for each site in an attempt to better understand the influence of anthropogenic sound on koala site occupancy. High values of NDSI reflect low levels of human-generated sound (Kasten et al. 2012). Although NDSI had little explanatory power in our model, we recommend further research into the influence of soundscape on koalas. As a species that relies heavily on vocal cues for communicating, anthropogenic noise may disrupt communication and the potential for social interactions.

Our study suggests that on the Mornington Peninsula, koalas are relatively widespread; however, based on the low average nightly detection probability we recorded in our passive acoustic surveys, they may occur at low densities. Our average nightly detection probability (0.363) was lower than that reported by (Hagens et al. 2018) for koalas at densities of < 1 koala/hectare and at any stage of the breeding season (0.385 in early breeding season to 0.793 in the late breeding season). Low density koala populations are probably typical across most modified landscapes where extensive habitat clearance has occurred, and where remaining habitat may be degraded and highly fragmented. However, localised extinction is a risk for such populations given that koalas have low reproductive output coupled with a heightened risk of mortality in modified landscapes (Cardillo et al. 2005).

Mapping predictions of occupancy across a region can inform the spatial allocation of resources for conservation (Rhodes et al. 2006). Based on our strongest model using landscape tree cover and road density, on the Mornington Peninsula we predict that koalas have the highest probability of occurrence in and around parks and reserves and on Crown land (i.e., HMAS Cerberus). Populations in these areas may function as sources of koalas to surrounding smaller habitat patches. However, parks and reserve areas only comprise a small proportion (18%) of the Mornington Peninsula, and risks to koalas moving between habitat patches may be high. Persistence of koalas on the Mornington Peninsula may therefore rely on actions to increase forest cover on private land. Considering the high cost and effort associated with tree planting, it is critical that actions are targeted to those areas where there will be the most benefit (Brancalion and Holl 2020). Analysis of the relationship between koala site occupancy and landscape-scale forest cover suggest that tree planting will have the most benefit to koalas at a site when forest cover within 1 km is already between 30 and 68% of the landscape. On the Mornington Peninsula, almost 85km2 (11% of the total region) falls into this category, suggesting that there is considerable scope to improve conservation outcomes for koalas.