Assessing the efficacy of passive acoustic monitoring for the swift parrot

May 2023

Background

Recent technological improvements in recording devices and associated analytical software used for passive acoustic monitoring (PAM) have substantially increased the potential applications and efficacy for wildlife monitoring and conservation. To produce tangible real-world applications from this technology a clear rationale for adopting these methods and defining specific and achievable aims and objectives are crucial (Wood et al. 2020, Wood and Peery 2022).

Applied studies utilising PAM are now providing reliable information to direct conservation planning, actions and decision-making (e.g. Gros et al. under review). The dynamic and spatially-structured breeding distribution of swift parrot due to flowering patterns of their food trees means the population is exposed to varying degrees of habitat availability and quality. There is an urgent need to better understand spatial and temporal variation in swift parrot nesting success/failure. Sugar gliders have been implicated as the major cause of nest failure (Stojanovic 2014, Heinsohn et al. 2015); however, other factors are also likely important (e.g. food availability, severe weather events). Swift parrot nesting success has not been recently monitored at a population level; however, spatial and temporal variation in nesting success has been observed by several field workers on the population monitoring program.

Previous estimates of swift parrot nest success/failure are largely derived from studies investigating the species’ breeding biology and reliant on accessing nests using tree climbing techniques (e.g. Stojanovic et al. 2014). Since 2015, the majority of survey effort in such studies has focussed on Bruny Island investigating other research questions where sugar gliders are absent (see Table 1 in Owens et al. 2023). During this period many regions occupied by a significant proportion of the swift parrot population during the breeding season (see Section 2) have not been formally sampled to estimate nest success/failure. This is a significant data and knowledge gap at the population level. Further, the interpretation of existing data for many regions is also hindered by small sample sizes (Heinsohn et al. 2015, Owens et al. 2023). Successful nests are regularly identified while undertaking the population monitoring program in most regions; however, these data were collected opportunistically and not to estimate success/failure rates.

Using tree-climbing techniques to access nests facilitates the collection of detailed data on breeding biology and other parameters at the nest-level. However, the logistical difficulties and safety issues associated with directly accessing swift parrot nests creates further sampling biases. Swift parrot nests are generally found in very old and senescent trees, regularly occurring in dead or unstable limbs and branches high in the tree (Webb et al. 2012). Possibly >80% of nest hollows identified over the past ~20 years were either extremely unsafe or not possible to climb due to their position in the nest tree (personal observations M. Webb). These difficulties have also led to a sampling bias towards locations and forest types where it is more likely that swift parrot nests are accessible by tree climbing once located (e.g. dry forest and woodland).

PAM provides a platform to collect spatially representative data to quantify swift parrot nesting success by removing the inherent sampling limitations and costs associated with tree climbing, address a significant knowledge gap for the species, and direct conservation resources to where they will be of greatest benefit.

Case study assessing the efficacy of acoustic monitoring

The purpose of this study was to trial the efficacy of using PAM to estimate swift parrot nest success using a subset of data from the 2021-22 breeding season.

Methods

PAM generally requires the analysis of very large sound files not feasible by manual listening. Developing automated methods to detect the species of interest (e.g., recognisers) is essential for PAM to be cost effective in these circumstances. During the 2020-21 swift parrot breeding season we deployed acoustic recorders (primarily Songmeters, Wildlife Acoustics) at selected locations where swift parrot nesting attempts were observed through direct observation. The primary aim was to collect training data/reference calls to develop automated methods (i.e., recognisers) to identify swift parrot calls in sound recordings. The primary vocalisations of interest were the begging calls of swift parrot chicks which are made several times a day when the adults return to feed young, and adult vocalisations commonly described as ‘warbling’ and ‘flight’ calls. Identification of these calls is based on field experience of the observers.

The begging calls of swift parrot chicks are audible for approximately 2 weeks before fledging, depending on the distance of the observer from the nest and weather conditions. Hence, detecting chick begging for ~2 weeks is a strong indicator a nest was successful, and was not predated. Nest failure was defined as when no chick calls were recorded over a period of 2 weeks.

During the 2021-22 breeding season, we again deployed acoustic recorders within occupied areas of the swift parrot breeding range, identified through the population monitoring program (see Section 2). The primary aim of deploying acoustic recorders was to determine the success or failure of nesting attempts identified through direct observation of swift parrot behaviours and vocalisations recorded during nesting surveys (see Webb et al. 2012). Recorders were also deployed in potential nesting and/or foraging habitat throughout the season to determine presence/absence of adult swift parrots, and when present the frequency of use in relation to the habitat resources at a site (e.g., food, potential nest sites).

Calls recorded using the acoustic monitors were analysed using our reference library of swift parrot vocalisations to automate detection of calls within the recordings. Call detections were automated using a variety of software applications and code developed by the authors. Detections were then manually verified by assessing spectrograms and manually listening to calls. This study was not funded, relying on volunteer time and financial resourcing, but was integrated into the current population monitoring program.

Results and discussion

Automated recognition to analyse sound files during the breeding season proved to be an invaluable tool to obtain timely information during the 2021-22 breeding season. Incorporating ‘new’ reference calls collected during the season into the recognisers continually improved their effectiveness. The primary outcomes from the PAM undertaken during the 2021-22 breeding season are summarised in Table 1 and Figure 1. Passive acoustic monitoring from a spatially representative sample of 19 nesting attempts in the Southern forests revealed a nesting success rate of 58%. The monitoring focused on the Southern forests as this area was recognised through the annual monitoring program as a key occupancy area based on habitat availability. This is far higher than current or previous estimates of nest success in the Southern forests or the Tasmanian mainland (Stojanovic et al. 2014, Owens et al. 2022). These data highlight the importance of collecting spatially representative samples to estimate this parameter over time and provide strong evidence for spatial variation in nesting success/failure, as suspected by experienced field workers undertaking population monitoring. Importantly, > 50 % of these nests were located in forest scheduled to be logged in Sustainable Timber Tasmania’s three year logging plan (https://www.sttas.com.au/forest-operations-management/our-operations/three-year- wood-production-plan).

PAM is also proving to be very effective for determining the presence - absence of adult swift parrots, and their behaviour. If the presence - absence of swift parrots is uncertain from traditional observation methods, targeted short deployments of acoustic recorders combined with immediate data analysis in the field also allows limited resources for nesting surveys (to identify nesting attempts through direct observation) to be directed more efficiently.

By reducing sampling biases and logistical problems associated with tree climbing, acoustic monitoring may better quantify spatiotemporal variation in nesting outcomes for the swift parrot. This kind of information is critical for effective conservation planning and informed decision-making, particularly considering the multiple serious threats to the survival of swift parrots. For example, as discussed above, existing data and analyses on sugar/Krefft’s glider predation rates have been hindered by small sample sizes and often not spatially representative of the swift parrot breeding population in any given year, whereas this report shows swift parrot nest success can be highly variable (Appendix B). Informed decision-making and conservation planning to address the threat of nest predation by sugar glider is crucial for the effectiveness of conservation actions addressing this threat (also see Appendix B).

Acoustic monitoring provides an efficient methodology to better understand key parameters affecting the swift parrot population. Obtaining spatially representative data on nest success/failure addresses a key knowledge gap in swift parrot conservation and will guide several conservation actions into the future. In addition, the ability to reliably establish presence-absence of swift parrots during the breeding season through rapid analyses of large sound files in the field provides real time data to inform where monitoring resources should be directed. This is particularly valuable in more remote locations which have not been logistically feasible to survey within contemporary budgets.

Further refinement of automated recognisers is currently underway in collaboration with other experts using a Convolution Neural Network approach and an extensive library of thousands of swift parrot calls.

Figure 1. Location of acoustic recorders, purpose of deployment, breeding success/failure.

(yellow crosses – confirmed nesting attempts, red squares – evidence of nest success, blue squares - nest failure, pink squares – establishing swift parrot presence)

Table 1. Location of acoustic recorders during the 2021-22 breeding season, purpose of deployment, and information acquired.

References

Gros, C., McNamara, K., Bell, P., Webb, M.H. (under review). Detection of the endangered Tasmanian masked owl Tyto novaehollandiae castanops using passive acoustic monitoring.

Heinsohn R, Webb M, Lacy R, Terauds A, Alderman R, Stojanovic D. (2015). A severe predator-induced population decline predicted for endangered, migratory swift parrots (Lathamus discolor). Biological Conservation 186, 75-82.

Owens, G., Heinsohn, R., Crates, R. & Stojanovic, D. (2023), ‘Long-term ecological data confirm and refine conservation assessment of critically endangered swift parrots’. Animal Conservation. Available at: https://zslpublications.onlinelibrary.wiley.com/doi/full/10.1111/acv.12834

Stojanovic, D., Webb, M., Alderman, R., Porfirio, L. L., and Heinsohn R. (2014). Discovery of a novel predator reveals extreme but highly variable mortality for an endangered migratory bird. Diversity and Distributions 20, 1200-1207.

Webb, M. H., Holdsworth, M. C., Webb, J. (2012). Nesting requirements of the endangered Swift Parrot (Lathamus discolor). Emu 112, 181-188.

Wood, C.M., Klinck, H., Gustafson, M., Keane, J.J., Sawyer, S.C., Gutierrez, R.J. & Peery, M.Z. 2020. Using the ecological significance of animal vocalizations to improve inference in acoustic monitoring programs. Conserv. Biol. 35: 336–345.

Wood, C.M. and Peery, M.Z. (2022). What does ‘occupancy mean in passive acoustic surveys? Ibis, doi: 10.1111/Ibi.13092

Acknowledgements

Significant contributions to this work were made by Alex Wylie, Amelia Cromb, Billy Rowe, and Charley

Gros and Dave James.