Generally, urbanization is a major threat to biodiversity; however, urban areas also provide habitats that some species can exploit. Flying-foxes (Pteropus spp.) are becoming increasingly urbanized; which is thought to be a result of increased availability and temporal stability of urban food resources, diminished natural food resources, or both. Previous research has shown that urban-roosting grey-headed flying-foxes (Pteropus poliocephalus) preferentially forage in human-modified landscapes. However, which land-use areas and food plants support its presence in urban areas is unknown. We tracked nine P. poliocephalus roosting in Adelaide, South Australia, between December 2019 and May 2020, using global positioning systems (GPS), to investigate how individuals used the urban landscape mosaic for feeding. The most frequently visited land-use category was “residential” (40% of fixes) followed by “road-side,” “reserves” and “primary production” (13–14% each). However, “reserves” were visited four times more frequently than expected from their areal availability, followed by the “residential” and “road-side” categories that were visited approximately twice more than expected each; in contrast, the “primary production” category was visited approximately five times less than expected. These results suggest that while residential areas provide most foraging resources supporting Adelaide’s flying-fox population, reserves contain foraging resources that are particularly attractive to P. poliocephalus. Primary production land was relatively less utilized, presumably because it contains few food resources. Throughout, flying-foxes visited an eclectic mixture of diet plants (49 unique species), with a majority of feeding fixes (63%) to locally indigenous Australian native species; however, in residential areas 53% of feeding visits were to non-locally indigenous species, vs only 13% in reserves. Flowering and fruiting phenology records of the food plants visited further indicated that non-locally indigenous species increase the temporal availability of foraging resources for P. poliocephalus in urban Adelaide. Our findings demonstrate the importance of residential areas for urban-roosting P. poliocephalus, and suggest that the anthropogenic mixture of food resources available in the urban landscape mosaic supports the species’ year-round presence in urban areas. Our results further highlight the importance of conserving natural habitats within the urban landscape mosaic, and stress the need for accounting for wildlife responses to urban greening initiatives.

Pteropus poliocephalus individuals were captured at the Adelaide Botanic Park on 10 and 11 December 2019. Flying-foxes returning to the roost before dawn were captured using two double banked mist nets (18m × 5m and 12m × 5m, 38mm mesh, Ecotone Telemetry, Poland) suspended 15 m high in the canopy of the colony. Nets were run on pulley systems that were continuously monitored by volunteers and two or three trained researchers. Each flying-fox was removed from the net by researchers and placed into a pillowcase hung from a horizontal pole. Females that were lactating, pregnant or carrying a pup were released immediately upon capture.

Ten focal individuals (five adult males and five non-reproductively active adult females) were transported to Adelaide Zoo where they were processed. Individuals were anaesthetized using 5% vaporized isoflurane via facemask and maintained at 2% isoflurane until processing was complete. While anaesthetized, morphometrics were taken and all bats were banded with a single stainless-steel band (Australian Bird and Bat Banding Scheme). The ten individuals were fitted with a collar supporting a GPS and accelerometry unit (CREX GPS Logger, Ecotone Telemetry, Poland, hereafter: transmitter). The five females had a mass of 725 g ± 93.1 (568 - 806) and the five males had a body mass of 818 g ± 127.0 (685 - 954). The transmitter and collar weighed 10 g and 3 g, respectively, giving a total weight of 13 g, representing 1.6-2.3% and 1.4-1.9% of the body weight of females and males, respectively. After animals recovered from anesthesia, they were placed in animal holding facilities at the Adelaide Zoo for recovery. The five male flying-foxes required surgery to implant a temperature-sensitive VHF FM radio transmitter (model PD-2THX, 3.9 g; battery life: 5 months; Holohil) as part of another study (Walker, unpublished). Tracking devices represented < 3.0 % of the body mass of the lightest individual. Individuals that did not require surgery (n = 5 females) were released back into the colony within 6 h of capture; the five males were released back into the colony the next morning following an assessment by a wildlife veterinarian (Author WB).

Research was conducted under Animal Research Authority no. A12217, issued by Western Sydney University.

Data collection

Global positioning systems and accelerometer data were collected for 5 months from 13 December 2019 to 23 May 2020 (Austral Summer and Autumn). Accelerometer data were recorded on three orthogonal axes and were used to identify the GPS fixes associated with feeding, as opposed to other behaviors such as flying (see below). Transmitters were programmed to collect accelerometer data in three burst types, dependent on battery voltage (solar recharge): 12 s at 5 Hz every 15 mins, 2 s at 30 Hz every 30 mins, or 3 s at 10 Hz every 30 mins. Transmitters were set to record GPS data every 30 mins during the night when there was sufficient solar recharge. Accelerometer data were linked with the GPS duty cycle. The duty cycles of the transmitters were monitored and changed remotely via Global System for Mobile (GSM) network using the web-panel depending upon the solar recharge of the batteries.

Data were collected via a GSM link from GPS trackers to 3G-enabled mobile phone towers that then reported the data to a File Transfer Protocol (FTP) server, accessed through the GPS data processing software package “NGA Analyzer” (Ecotone Telemetry, Poland).

The amount of time that the trackers produced usable data varied from 3 to 154 days (Mean = 78.4; Supplementary Table 1). Some of the collared individuals left the Adelaide region (defined as ≥75 km from the center of the Adelaide Botanic Garden roost) during our study (Supplementary Table 1). One female (FFOX05) left the Adelaide region before the transmitter began collecting data and hence her data were excluded from analyses herein. Another female (FFOX02) left the Adelaide region on 18 January 2020, and returned on 2 April 2020. Two males (FFOX07 and FFOX09) left the Adelaide region on 31 December 2019 and on 19 March 2020, respectively, and did not return to the region for the duration of the study. We removed GPS fixes associated with travel outside of the Adelaide region (GPS fixes greater than 75 km from the colony). GPS fixes less than 500 m from the center of the colony were also removed to exclude resting time within the colony. As the boundary of the roost varies with the number of flying-foxes present, this conservative approach means that we are potentially excluding some nearby foraging fixes. For those individuals that left the Adelaide region, GPS fixes recorded on the day of departure or day of return were excluded as any food resources associated with these fixes were likely supporting the individual’s journey to or from another roost rather than their stay in Adelaide.

Identifying feeding fixes

Global positioning systems locations were designated “feeding fixes” if they were temporally aligned with moderate levels of activity, as determined by variance in the acceleration data. We applied a principal component analysis (PCA) using the “prcomp” function from the “stats” package in R to the three recorded axes of acceleration forces to maximize the amount of variation caused by movement that is expressed in a single vector (i.e., principal component 1). We did this also to account for variation in the spatial orientation of the transmitter between individuals, which could have influenced the distribution of acceleration forces caused by movement across the three axes. We plotted a frequency histogram of the standard deviation (SD) of the PC1 scores over each burst of acceleration data for each individual and used troughs between primary modes in the distribution of SD values as thresholds for designating among broad categories of acceleration intensity (Collins et al., 2015). Firstly, we identified a mode of greatest SD values, most likely associated with wing flapping during flight, and assigned data above a trough threshold defining the lower limit of this mode to high-level activity (Supplementary Figure 1A). The remaining data were subject to another PCA. We identified two modes in the frequency distribution of SD values of the new PC1 vector: a lower mode clearly associated with inactivity (i.e., little, if any, and body movement), and an upper mode that included values of greater acceleration intensity, but excluded the highest values previously assigned to high-level activity. We assigned data above the trough between these two modes as moderate activity that likely pertains to tree-based movements, including feeding (i.e., the behavior of eating a dietary component and any associated movement within a foraging tree; Supplementary Figure 1B).

Date-time stamps associated with moderate activity were rounded to the nearest second. We then aligned the closest activity date-time stamps to the GPS date-time stamp. To minimize the possibility of misinterpreting the level of activity at each GPS coordinate, we calculated the time discrepancy between each pair of activity and GPS data, using the “difftime” function from the “lubridate” package in R. We excluded the data to include only data pairs that were within a ±60 s discrepancy buffer. We used the “suncalc” package in R to calculate local sunrise and sunset times and define night (after sunset and before sunrise) and included GPS data for the moderate level of activity during the night only, which resulted in 489 GPS fixes associated with moderate activity during the night.

Land-use categories

To investigate which land-use categories P. poliocephalus used for feeding we used the South Australian Government’s “Generalized Land Use 2020” shapefile of the land-use categories of South Australia (Government of South Australia, 2020). Land-use polygons were grouped into 10 categories: “primary production” (i.e., areas that are used for agriculture, horticulture, forestry, and livestock), “residential” (i.e., urban and rural residential, hotel/motel accommodation, institutional accommodation, and orchards nestled within peri-urban residential areas), “reserve” (i.e., national park, median strips, and road reserves), “vacant” (i.e., urban and non-urban vacant land, and steep/rocky land), “utilities” (i.e., gas, electricity, water/sewage/waste disposal, public transportation, and telecommunications), “institution” (i.e., government, and education), “mining” (i.e., mines, open workings, wells, and quarries), “recreation” (i.e., ovals, golf courses, camping grounds, and stadiums), “commercial” (i.e., wholesale trade, retail, and finance), and “industrial” (i.e., food manufacture). The shapefile has gaps between polygons that pertain to linear landscape features including roads and watercourses and thus, we assigned these areas a new land-use category called “road/river.”

Land-use by foraging P. poliocephalus

The land-use shapefile was clipped by a circle with a 75 km radius (Figure 1), and a geoBoundaries shapefile of Australia (Runfola et al., 2020), using the function “st_buffer” in the R package “st,” to obtain the total area where P. poliocephalus foraging could have occurred. A 75 km radius was selected since the maximum distance any of the tracked individuals traveled and returned to the Botanic Park colony was 70.8 km.

Land-use categories were extracted for all feeding fixes in the study area (n = 489). The areal extent of each land-use category within the study area was calculated. To calculate the area pertaining to the “road/river” land-use category, the sum of the area of all the land-use categories was subtracted from the clipped study area.

To examine whether P. poliocephalus foraged at random across the landscape, we compared the proportion of feeding fixes in each land-use category to the proportion that would be expected based on the area of each land-use category in the study area, using a Chi-squared test for given probabilities.

Food plant species

To investigate the food plant species visited by foraging individuals, we first read fixes pertaining to moderate levels of activity at night into Google Earth Pro and confirmed that all likely feeding fixes were in trees. This resulted in 489 feeding tree locations.

A random sample (n = 322) of the total 489 feeding tree locations, were approached in person. Of these, some (n = 35) could not be visited due to dangerous terrain or formal access restrictions; of which 60% were in residential areas, 20% in reserves, 9% in agricultural areas, 6% in mining and road/river areas, and 3% in utilities (Supplementary Table 2). The frequency of inaccessible fixes in each land-use category was not significantly different to the observed frequency of feeding fixes in each land-use category overall (Pearson’s Chi-squared test n = 489; χ 2 = 55, df = 50, p = 0.29), and thus, exclusion of these fixes from the dataset is unlikely to bias the results.

Each accessible feeding tree (n = 287) was photographed, and samples of bark, leaves, buds, flowers and fruit were taken, where possible, to aid in species identification. In addition, close-up photographs of plant characteristics using a telephoto zoom lens were captured, particularly where direct samples were unable to be obtained (see Figure 2A-F). Tree characteristics including details on the bark, leaves, buds, fruits and inflorescences were recorded for each tree following the EUCLID key, and an attempt at identifying the species was made using tree identification guides (Nicolle, 2013; Centre for Australian National Biodiversity Research [CANBR], 2015; Brooker and Kleinig, 2016; Lucid, 2016). Author DN provided subsequent validations on each food plant species from photographs and detailed field notes. Of the 287 food plants assessed, 22 lacked sufficient detail and thus could not be confidently identified. Therefore, a total of 265 individual food plants were identified.

We used known distributions from plant identification guides (Nicolle, 2013; Centre for Australian National Biodiversity Research [CANBR], 2015; Brooker and Kleinig, 2016; Lucid, 2016) to classify each identified tree (n = 265) as “locally indigenous” to the study region, “non-indigenous Australian native,” or “non-Australian” species, this is referred to as their “geographic origin.” Where plants could not be identified to species level, their origin was conservatively classified as “unknown.”

Here, two Pearson’s Chi-squared tests were conducted, respectively, to examine (i) whether the relative number of unique species that occurred in each of the four “geographic origin” categories (i.e., locally indigenous, non-indigenous Australian native, non-Australian, and unknown) varied across land-use categories (Supplementary Table 3); and (ii) whether the relative number of visits (i.e., feeding fixes) to trees of varying “geographic origin” (i.e., locally indigenous, non-indigenous Australian native, non-Australian, and unknown) varied across the visited land-use categories (Supplementary Table 4).

Food plant species flowering and fruiting phenologies

Phenology tables were constructed by compiling flowering and fruiting data collected from a range of published articles, online databases, and apiary flowering records. Where data were available, the peak time of flowering/fruiting, duration of flowering/fruiting in months, and annual reliability of flowering/fruiting, was recorded and standardized. Where records from the Northern Hemisphere were used, flowering/fruiting phenology was standardized by stating in which season flowering was recorded. 

Data filtering, compilation and analyses were performed in the R environment for statistical computing (version 4.1.2; R Core Team, 2017) interfaced with RStudio (version 2021.09.0; RStudio Team, 2015).