Dr. Desley Whisson
Deakin University, Victoria, Australia
Field work has been completed, and analysis of recordings is about to commence. The effort was undertaken in November and December 2015 when koala breeding activity and frequency of vocalizations were at their peak.
In each of four sites, a Song Meter SM3 was deployed and programmed to record continuously between 8pm and 2am each night for four weeks. A standard visual survey for koalas showed that densities ranged from 2.0 to 7.7 koalas/hectare per site.
The SM3 recordings now need to be analyzed to determine if the frequency of vocalizations is related to koala density. The Acoustic Biodiversity Monitoring Laboratory at the Queensland University of Technology (research.ecosounds.org) has generously offered to assist with this part of the project.
A sub-sample of recordings have been reviewed in Songscope to identify (and annotate) male vocalisations and male-female interactions. Recognizers were built using these annotations and tested but still need refining to eliminate many false positives and to detect distant (but still audible) calls.
The male-female interactions are proving to be especially difficult to create a recognizer for because of the high variation in the female call during the interactions (Figure 1). The recordings soon will be analysed using a detector that has been used to quantify koala vocalisations at a site in Queensland.
Preliminary results do not suggest that vocalization frequency is related to population density (Figure 2). In reviewed recordings, bellow frequency is relatively high in the site with the lowest population density.
A number of other species have been detected during review of the recordings (Figure 3). Yellow-bellied gliders have been recorded at two sites and southern boobooks at all sites. Bioacoustics may prove useful for detecting these species.
Over the last few months, the team continued its efforts towards developing accurate recognizers for vocalisations of males, and male - female interactions. The best Songscope recognizer was not entirely effective, detecting less than about 30 percent of any koala vocalisations. Desley and her colleagues are seeking assistance from the bioacoustics group at the Queensland University of Technology to improve performance of the recognizer.
This study found a correlation between the frequency of female but not male vocalisations and a visual estimate of koala population density at four sites during the koala breeding season (Figure 1). Females vocalised only during interactions with males, whereas males vocalised both spontaneously and during interactions. Lack of a relationship between spontaneous male vocalisations and population density may indicate either a density dependent relationship (i.e. more frequent vocalisations where densities are lower) or poor performance of a single visual survey in estimating population density.
Population densities ranged from 2.5 to 6.0 koalas per hectare at each of the four sites used in this study. Songmeters (SM3) programmed to record from 2000h to 0200h per night (period of greatest koala activity) were deployed at each site for one month between October and December, recording for between 27 and 32 days (total of 708h). Recordings from some days (2 to 5 per site; total of 14 days) were discarded due to poor weather that negatively influenced detection of vocalisations. From 624 hours of recordings, there were 1525 koala vocalisations, including 1251 spontaneous male bellows and 274 interactions (females and males vocalising together). Vocalisations of yellow-bellied gliders (Petaurus australis) and common brush-tailed possums (Trichosurus vulpecula) that are common in the area, were also detected.
Songscope had limited effectiveness for detecting koala vocalisations, with a high rate of false positives as well as low recall (i.e. true call detection). Recordings were therefore manually processed (i.e. spectrograms visually scanned). This was time-consuming but manageable for the recordings obtained in this study, but development of an automated recogniser will be critical to facilitate the processing of recordings from larger studies or in applying bioacoustics to regional monitoring of koala populations.
The primary goal of this study was to determine the relationship between the frequency of koala vocalisations and population density, and thus the potential for using bioacoustics for indexing koala populations. Results suggest that the frequency of female vocalisations (indicating interactions between females and males) are correlated with population density; however, given the low frequency of these vocalisations and the small number of sites sampled, sampling across more sites is necessary to confirm this.
Development of a cost-effective and accurate method of surveying koalas is important for gaining a better understanding of the distribution and conservation status of koalas throughout their range. This study has indicated that bioacoustics may have potential as a non-invasive survey method. However, testing at more sites and developing automated methods of detecting vocalisations in recordings are necessary to confirm this.
We developed a relationship with the QUT Ecoacoustics Research Group during the study and hope to continue working with them to develop improved recognisers for southern koalas. The project also has led to a collaborative project with another university to assess the effectiveness of bioacoustics for detecting other cryptic arboreal mammals.
Following the results of this project, we are now conducting a state-wide study using bioacoustics to both monitor koala populations and to examine the seasonality in their breeding behaviour. We have deployed Songmeters (SM4) at 11 sites throughout Victoria, and will operate them for one year.