Group spacing and cohesion are fundamental aspects of social structure in group-living species, shaping interactions and ecological dynamics. However, defining what constitutes a group can be problematic in highly mobile group-living species such as delphinids. Spatial proximity among individuals is often used to delimit group membership, yet threshold distances lack justification. In this study, we used drone-based observations, artificial intelligence, and photogrammetry to investigate how sighting size, age composition, and behaviour influence nearest neighbour distance (NND) and inter-individual distance (IID) among Indo-Pacific bottlenose dolphins (Tursiops aduncus) in Coffin Bay, South Australia. Our aim was to establish an objective and ecologically meaningful definition of group membership based on spatial proximity. We identified threshold distances of 12 m for NND and 25 m for IID, defined by the 95th percentile of the distributions across all sightings. Mixed-effects models showed statistically significant effect of sighting size, age composition and behaviour on both NND and IID, although the effect were generally small (<1–3.5 m). Behaviour had a strongest impact, with dolphins exhibiting closer spacing during socialising and travelling compared to foraging. Larger sighting size significantly resulted in an increase in NND and IID, while higher proportions of juveniles and calves influenced spacing differently depending on sighting size. These results highlight the flexibility of dolphin spatial structure in response to various factors, providing insights into the mechanisms underlying cohesion and coordination. Improved definitions of group membership based on these results will support more accurate assessments of social structure in marine mammals and enhance the effectiveness of conservation and management strategies.

Artificial Intelligence (AI) is transforming the capacity of conservation scientists to analyse and interpret the very large datasets of images or recordings necessary for comprehensive monitoring programs. In combination with sophisticated user interfaces, AI is also being used to alert conservation managers of trap captures, detections of biosecurity threats, or instrument diagnostics within seconds of events occurring, thus improving welfare outcomes and monitoring efficacy and cost effectiveness.
EdgeAI - AI algorithms running directly on local devices - takes these advances to a new level by enabling not just monitoring but instantaneous, autonomous triggering of conservation tools. Vision and sound based EdgeAI detectors can automatically activate specific traps, gates, feeders, nest boxes, deterrents and poisoning devices, dramatically reducing off-target risks and improving management of both threatened species and pests.
We present initial results from several pilot studies of autonomously-triggered conservation tools to inspire conservation ecologists to dream of a more informed, efficient and successful future aided by bespoke tools triggered by generic edgeAI detectors.

Wildlife ecologists often face the onerous task of manually analysing hours of audio recordings to detect species presence and activity. While deep learning has demonstrated exceptional capability in complex audio analysis, these state-of-the-art models remain inaccessible to most of Australia’s ecologist community. An implementation of a state-of-the-art deep learning model inside existing audio analysis software was designed and evaluated. This system aims to reduce ecologists’ workloads by automatically curating snippets of interest for target species, laying the foundation for nearly fully automatic species detection—all within a familiar, user-friendly interface. As an extension, a second system was developed to further leverage modern deep learning. This custom-trainable model enables ecologists to detect any species or call using user-provided reference audio as a training dataset. Both systems were evaluated on real-world ecological data using confusion matrices and class-wise F1-scores, precision, and recall. The models successfully detected numerous species—including several threatened and endangered ones—achieving F1-scores between 0.7 and 1.0 for many targets. We also outline practical data preparation steps to maximise the custom model’s performance.

Machine learning is widely used to classify fine-scale behaviours from animal-borne accelerometers, enabling ecologists to quantify behavioural budgets and environment interactions across hundreds of species. Despite advances in model design, however, classification remains imperfect, particularly for rare, highly-variable, or context-dependent behaviours. Ecologists thus face the persistent challenge of interpreting behaviour while accounting for this uncertainty.

Post-processing refers to techniques that refine machine learning predictions through secondary inference and smoothing after classification. Unlike altering model architecture or gathering additional training data, post-processing can be applied directly to existing predictions, offering a computationally inexpensive, quick, and flexible approach to improve classification quality. Despite extensive use in other machine learning domains, post-processing is yet to be explored in the animal behaviour classification literature.

In this study, we compare seven post-processing techniques across five animal behaviour datasets. These methods include rule-based smoothers (e.g., rolling averages), domain-informed logic (e.g., minimum duration thresholds), and secondary machine learning models (e.g., Hidden Markov Models). We assess each post-processing technique’s performance relative to the original, unprocessed, predictions, focusing on classification performance as well as changes to ecological interpretation of behavioural budgets. Our results demonstrate that even simple post-processing techniques improve performance by enforcing ecologically plausible behavioural sequences.

In this ecologist-accessible guide to interpreting the output of accelerometer-based behaviour research, we present machine learning as just one step in a broader interpretive pipeline, shifting focus away from maximising base algorithm accuracy towards optimising for overall ecological interpretability.

Globally biodiversity declines are occurring at an alarming rate. Including rampant declines across Australia’s unique mammals. Prompting the critical need for detailed understanding of the occurrence and distribution of species. Additionally, animal ethics concerns promote the need for such data to be gained in a minimally invasive manner. Recently invertebrate-derived (iDNA) has arisen as a sensitive, non-invasive approach allowing for the detection of birds and mammals through the ingested DNA of blood or carrion feeding invertebrates. Furthermore, iDNA samplers such as mosquitoes offer a high level of temporal and spatial resolution, allowing for community patterns to be explored on a fine-scale. Here we applied mosquito iDNA to examine the avian and mammalian community of six sites in Kakadu National Park, across two occasions, representing the wet and dry season. Additionally, the relationship between the detected communities and the vegetation structure and recent fire regimes was explored. In total 50 avian and 19 mammalian taxa were detected. The detected taxa included those from difficult to monitor groups such as arboreal and small-bodied mammals and threatened species such as the White-throated grasswren (Amytornis woodwardi) and Ghost Bat (Macroderma gigas). Mammalian richness showed to be significantly different between sites, while avian richness changing between sampling periods, with higher richness observed during the sampled dry season. Community composition was also shown to be significantly different between sampling events. With composition driven by recent fire regimes during the dry season and vegetation structure during the wet. Highlighting, the capability of mosquito iDNA as an approach to gain fine-scale understanding of the responses of bird and mammal communities to fire regimes and vegetation structure.

High-throughput DNA metabarcoding is rapidly supplanting morphological identifications in ecological studies, yet transforming sequence read counts into meaningful, quantitative estimates of organismal abundance remains a major challenge. This is particularly relevant for monitoring insect communities, where timely, data-driven measurements of native and invasive population trends are required to better understand how anthropogenic factors impact insect abundance and biomass. Standard metabarcoding protocols applied to bulk invertebrate samples only provide relative abundance information, only capable of providing counts of DNA molecules which are not easily transferable to accurate insect counts, and are biased by species-specific morphological and molecular characteristics which distort the proportion of metabarcoding reads produced for each species relative to the true proportion of individual organisms in the community. Here, we present our work towards identifying and characterising these taxonomic biases, including previously under-appreciated factors such as mitochondrial copy-number variation between individuals and life-stages within the same species. We will also outline complementary strategies to improve the quantitative outputs and interpretability of metabarcoding datasets, including novel molecular laboratory methods, and taxon-specific correction factors, which will be explored further throughout my PhD project. Through development of quantitative metabarcoding approaches for monitoring entire insect communities we aim to bridge the gap between DNA-based identification and ecological monitoring to develop a better understanding of insect diversity under man-made pressures.

We are experiencing a global mass extinction crisis and amphibians are the most impacted vertebrate taxa. Over 40% of species are threatened, including the once common green and golden bell frog (Litoria aurea). L.aurea faces multiple threats from climate change, predation, and habitat disturbance, however, the fungal skin infection chytridiomycosis is considered a major driving force behind its decline.
Batrachochytrium dendrobatidis (Bd), the fungal pathogen responsible for chytridiomycosis, is an aquatic organism that infects keratinised skin. The resulting epidermal dysfunction disrupts electrolyte balance, ultimately leading to death by cardiac arrest. Bd infects over 50% of amphibian species worldwide with disease severity and host responses varying widely across species. These differences are in part explained by variation in constitutive skin and immune defences and immune responses to infection. L. aurea is highly susceptible to chytridiomycosis, yet its immune response and constitutive defenses remain poorly understood. Understanding host response to infection is vital for developing effective interventions including captive breeding and environmental manipulation.
This study utilised NovaSeq high-throughput sequencing on foot webbing biopsies to examine immune gene expression in L. aurea infected with Bd. Non-lethal sampling improves animal welfare and allows frogs to contribute to future breeding and population genetics following successful treatment. Repeat sampling could assess long-term impacts of infection, while experimental reinfection studies may reveal how immune responses change with repeat infection. Studies performed under differing environmental conditions may also improve predictions of disease outcome, identify conserved host responses and help inform effective conservation interventions designed to limit the impact of chytridiomycosis
Future work will focus on developing primers to target identified immune genes of interest and performing gene expression with quantitative polymerase chain reaction (qPCR) using non-invasive skin swabs. These methods can be utilised in both captive breeding colonies and wild populations and adapted to other species and taxa.

The Flinders Ranges Purple-spotted Gudgeon (Mogurnda clivicola) is a small species of fish endemic to South Australia. It originally only inhabited Weetootla Gorge and Nepouie Springs in the Flinders Ranges, South Australia. The species is Critically Endangered in South Australia and Nationally Endangered under the Environment Protection and Biodiversity Conservation Act. Concerned about a number of factors including the chances of a catastrophic event and climate-change, a decision was made to translocate 600 individuals to two populations well south of these sites in 2021, namely Spring Creek in the Ikara Flinders Ranges National Park and Hookina Creek in the Yappala Indigenous Protected Area as insurance populations under an approved translocation plan and funded by the Australian Government’s National Landcare program, delivered through the South Australian Arid Lands Landscape Board and managed as part of the award-winning Bounceback project.
Healthy and viable populations usually have high genetic diversity, required for evolving and adapting to change. Augmenting or rescuing depleted genetic diversity can improve genetic fitness and requires the development of effective management approaches. The Translocation Plan had built into its ongoing monitoring, genetic sampling at the release sites to better understand changes in genetic drift and heterozygosity, and to determine if genetic supplementation (intervention) was required.
We describe the logistics of translocating the fish and the on-going monitoring. In 2025 genetic samples from 96 fish at the two release sites were used to understand their genetic viability and a simulation modelling was used to better understand what interventions are required in the future to reduce genetic drift and maintain genetic heterozygosity. This modelling found it is currently not necessary to supplement from the original sources sites but there was value in intra-site translocations, particularly from downstream to upstream ponds. This has also been identified as important in other freshwater fish studies.

Air is increasingly being recognised as a biologically rich source of taxonomically diverse environmental DNA (eDNA). Multiple proof-of-concept studies have now been published that explore air as a medium for the detection of single species and even whole communities of terrestrial species, including mammals, non-anemophilous plants and insects. Airborne eDNA has been sampled using various stationary devices but if we can access it using mobile collection methods, we can rapidly assess biodiversity at large spatial scales. We compared passive eDNA filters deployed for short (30 mins) and long periods (24 hrs) with 3D printed filters attached to cars that carried out rapid transects through different types of Australian environments. Across all three DNA air sampling approaches our metabarcoding procedure detected 49 vertebrate taxa (33 birds, 15 mammals and 1 amphibian), with 29 of these detected in samples collected using the mobile method. All sampling methods detected differences in vertebrate communities between woodland and agricultural landscapes whereas no significant differences in taxa richness or community composition were observed between the different sampling methods. We therefore propose that our car-mounted airDNA sampler could be a game changer for large scale, rapid biomonitoring efforts.

Environmental DNA (eDNA) metabarcoding is an emerging biomonitoring tool that provides a non-invasive, high-throughput, and cost-effective alternative to manual methods for generating biodiversity inventories across broad spatial and temporal scales. However, the high dimensionality and large volume of eDNA data requires dedicated analytical pipelines, with specialist bioinformatics packages used to process and extract information from amplicon libraries. Existing bioinformatics solutions for amplicon analysis are typically implemented as command-line tools, which present significant technical barriers for their adoption by non-expert users. We present Qiimera (pronounced “Chimera”), a Python-based extensible bioinformatics toolbox with graphical user interface for end-to-end eDNA amplicon analysis. Qiimera offers an intuitive web interface that can be hosted locally or on-cloud premises. Core functionalities of Qiimera are derived from popular R and Python bioinformatics libraries, with the customised ability to ingest and maintain internal (custom) and external (online, e.g. BOLD) databases for taxonomic annotation. In addition, Qiimera allows the user to create and share pipeline recipes, with full data visualization enabling rapid assessment of key results throughout the entire workflow. Qiimera can be easily extended or chained with R, Java or Python packages, including machine learning frameworks for more advanced data analytics. Using a freshwater macroinvertebrate eDNA dataset collected in our routine biomonitoring program as an example, we demonstrate the functionalities of Qiimera for amplicon analysis. Qiimera provides a user-friendly, efficient and flexible amplicon analysis solution for both beginner and expert users and should facilitate a greater adoption of eDNA in biomonitoring and research programs.