The Terrestrial Ecosystem Research Network (TERN) is Australia’s national ecological research infrastructure, funded through NCRIS. It supports long-term ecosystem monitoring and data delivery across a range of biomes and land tenures. As part of its regional-scale ecosystem surveillance program, TERN systematically collects vegetation and soil observations and an extensive range of physical samples, including plant tissue, herbarium specimens, soils, and metagenomic material from sites across Australia's diverse landscapes.
These material samples are critical for advancing ecological understanding, supporting studies in biodiversity, plant-soil interactions, climate change impacts, and ecosystem processes. However, despite their scientific value, many of these samples remain underutilised due to limited visibility and disconnected data systems.
To address this, TERN has established a material sample repository within its EcoPlots platform, enabling open access to curated physical collections. Each sample is assigned a globally unique identifier (IGSN) and linked to detailed metadata, including site information, collection protocols and related digital artefacts. Researchers can search and filter samples by region, site, year, and sample type, and submit expressions of interest directly through the platform to initiate access and collaboration.
With more than 130,000 samples published, this initiative represents a significant step toward bridging ecological field collections with digital infrastructure, advancing Australia's capacity in long-term management of material sample for ecosystem science research.
The presentation will outline digitising and publishing ecological samples as linked data, including metadata harmonisation, persistent identifiers, and integration with observational records.

The presentation covers recent innovations in GNSS technology (formerly GPS), emphasizing developments in receiver design, infrastructure, services and government policy.

Key learnings include hardware and its now, ease of use and the significance of four Global GNSS constellations.

Importantly, we will cover the regional SBAS (Satellite Based Augmentation System) called SouthPAN (Southern Pacific Augmentation Network), which provides free sub-meter corrections without mobile wireless data (connection to the internet) and without a local CORSNet base station in all regions of Australia and New Zealand. What this means is that subject to environmental considerations (blockages to the sky), it is now easy and cost effective to have sub metre (down to 300 mm) right across Australia and New Zealand.

We also cover the federal governments NPI (National Positioning Infrastructure) policy that allows free access to more than 750 CORSNet base stations for sub metre (L1 receiver) and centimeter (Dual/Multi frequency RTK receivers) across Australia and New Zealand.

We also highlight the importance of understanding datums when using different differential correction services.

The implementation of GNSS technology has significantly boosted field effectiveness leading to improved outcomes including operational efficiency, completeness and accuracy of the data collected through a structured field form. In some cases, it makes compliance much easier and cheaper than in the past.

Through innovation, GNSS is no longer the realm of surveyors, but is used successfully by subject matter experts in other field such as environment, arboriculture, utilities, cultural heritage, councils, forestry and so on. It is cheaper lighter, easier to use with advance services that were once either not available or very expensive. This leverages the value of your investment in GIS by providing accurate location to field attribute data.

The ARDC Planet Research Data Commons (Planet RDC) delivers national-scale digital infrastructure to support researchers, policy makers, and environmental managers tackling Australia’s most pressing environmental challenges. In response to calls for improved data coordination from the 2021 State of the Environment report and the EPBC Act Review, the Planet RDC enables trusted, interoperable environmental data to be shared and reused across sectors. Built through long-term partnerships with research, government, and industry, the Planet RDC integrates diverse data sources and accelerates model and analytics development.
To deliver on its mission, the Planet RDC supports four national platforms. Open EcoAcoustics enables the storage, processing, and sharing of ecoacoustic data using open standards and automated species recognisers, integrated with facilities like TERN and ALA. EcoCommons offers an accessible modelling environment with trusted datasets, species distribution tools, Jupyter notebooks, and training resources. WildObs is building national-scale infrastructure for processing and sharing wildlife camera data, while Biosecurity Commons provides a secure platform for modelling biosecurity risk and surveillance strategies.
These platforms are supported by ARDC’s core infrastructure. The Nectar Research Cloud delivers scalable computing for workflows and machine learning. Nectar’s research services include access to virtual desktops, Jupyter Notebooks, BinderHub, National GPU services and Preemptible Instances. Research Link Australia enhances project visibility and collaboration, while Research Data Australia facilitates data discovery and access.
Together, these capabilities form a coordinated, national ecosystem that advances environmental research, supports FAIR data practices, and enables informed decision-making for a more resilient Australia.

Australia is a global leader in the deployment of wildlife cameras to monitor terrestrial wildlife. However, infrastructure to store, integrate, analyse, and report on this vast data stream remains underdeveloped. The Wildlife Observatory of Australia (WildObs) is a purpose-built analysis infrastructure and wildlife camera data commons that overcomes key barriers to analysing disparate datasets. Historically, collaboration was hindered by manual image processing, offline data storage, and limited access to advanced analytics. WildObs addresses these challenges by providing (i) a tagged image repository for AI research, (ii) an intuitive image management platform with AI-assisted species identification, (iii) connections for data sharing, and (iv) a data standardization pipeline for integrating diverse survey methods into a unified statistical framework. This initiative positions Australia to fully leverage machine-based observations for biodiversity monitoring, climate change responses, and evidence-based conservation strategies.

As a case study, we highlight WildObs' collaboration with WWF’s Eyes on Recovery project, which collected eight million images from 17 surveys in nine landscapes, involving 24 organizations. All images were catalogued in Wildlife Insights, with WildObs providing analytical support to estimate occupancy of 56 species and assess wildlife recovery after the 2019–2020 bushfires.

Prawn trawl fisheries in tropical Australia catch large quantities of bycatch as part of routine operations including ETP (Endangered, threatened and protected) listed species. By far the most abundant of these are sea snakes, whose size and body shape make them particularly hard to exclude using bycatch reduction devices. By law all sea snake interactions must be reported, but their abundance, diversity as well as dangerous nature can make it hard for crews to reliably report all interactions.
Our project aims to improve the reliability of reporting by Western Australian prawn trawl fisheries, which overlaps with a particularly diverse and endemic sea snake fauna. We did this by running workshops and spending time onboard with commercial crews, with particularly engaged crew collecting photo verified data throughout the year. For 2023-24 we obtained ~1600 photo verified sea snake records across three crews in Shark Bay and Exmouth Gulf. Rates of reporting and successful identifications have improved markedly, and species compositions of each fishery differ greatly from previous surveys. Analysis of body length shows that most caught sea snakes are sexually mature, suggesting juveniles might more readily escape from the net. Population genetic analysis shows that sea snakes in Exmouth Gulf are broadly connected with the rest of the North-west shelf. In contrast sea snakes across four species are more isolated in Shark Bay and a much greater proportion of these populations overlap with trawl fisheries.

Effective wildlife conservation relies on accurate and scalable monitoring of keystone species in complex ecosystems. However, it is a challenging task using traditional methods because of dense vegetation, variable illumination, and extensive manual efforts. We present the Multi-Task Detection and Recognition Framework (MDRF), which combines Grounding DINO (vision-language grounding) with ConvNeXt V2 (advanced instance segmentation) into a single end-to-end pipeline. The novel model is pretrained on large-scale benchmarks (COCO, LVIS) and then fine-tuned on a custom camera trap images dataset with extensive data augmentation, including multi-angle rotations, flips, light and brightness variations, to make our model more resilient to changes in the environment. Standardised cross-domain methods are employed to evaluate the performance in both supervised and zero-shot settings. MDRF gets a closed-set Average Precision (AP) of 59.4%, AP for bounding box detection (APbox) up to 91.8%, exhibiting substantial quantitative gains over state-of-the-art (SOTA) methods. Instance segmentation predictions exhibit precise boundary delineation up to (IoU > 0.93) under severe occlusion and illumination distortions. This was achieved through Global Response Normalisation to reduce background noise and focus on discriminative features, making it much easier to find tiny objects under challenging conditions. Our framework establishes a new standard for automated ecological monitoring, offering a scalable and deployable tool for real-time biodiversity assessment and conservation. The design can work for various species and may easily be adaptable for diverse species conservation goals. MDRF makes it possible for bold, data-driven initiatives towards a more resilient future by greatly decreasing the need for human observers and making it easier to find objects in difficult tropical rainforest dynamics.

Passive acoustic monitoring and machine learning are increasingly being used to survey threatened species. When automated detection models are applied to large novel datasets, false-positive detections are likely even for high-performing models. Manual validation of outputs can be time consuming, and additional fine-scale annotation of individual notes is impractical for large datasets and difficult to automate when using noisy field data. This research presents an acoustic monitoring pipeline which employs a multi-stage hybrid approach: initial detection using a transfer learning classifier, followed by segmentation and iterative unsupervised clustering of extracted acoustic features using UMAP and HDBSCAN to remove label noise. This process essentially allows for threshold-agnostic validation as well as rapid 'harvesting' of individual note annotations from large field datasets.
Our method begins by identifying potential vocalisations in environmental recordings using a neural network classifier. Segmentation and feature extraction are then applied to the outputs, and refined using iterative unsupervised clustering methods. The unsupervised approach significantly improved detection accuracy, successfully separating true detections from false positives and substantially reducing label noise. Additionally, this approach generated detailed note-level annotations, offering insights into vocal individuality, song structure, and geographic variations within and between species. The clustering step also facilitated semi-supervised learning by rapidly highlighting uncertain cases to further enhance model accuracy. This scalable pipeline addresses key bioacoustic monitoring challenges: minimizing manual validation efforts and ensuring accurate annotations in noisy field data. We demonstrate the utility of this approach on large datasets and calls of several cryptic species, including the Powerful Owl (Ninox strenua).

Globally, natural disturbance events are becoming increasingly extreme, highlighting the need for effective, large-scale biodiversity monitoring. While in-person surveys are valuable, they are often costly, time-consuming, and disruptive to wildlife. Advances in bioacoustics offer scalable, less-invasive alternatives, yet efficiently processing extensive and complex recordings into interpretable datasets remains a major challenge.
Artificial Intelligence offers a promising solution, reducing the time and effort required to identify vocalising species. This study evaluates BirdNET, an AI-powered platform for species identification, as a tool for post-disturbance monitoring of birds. Since March 2023, 90 autonomous recording units have been deployed across the Greater Blue Mountains, a Key Biodiversity Area significantly impacted by the 2019/20 mega-fires. Each device records the first 10 minutes of each hour, generating a dataset of 249,387 recordings. Using BirdNET, 7,826,514 detections from 180 species have been documented, including 17 threatened species.
To assess BirdNET’s performance, we manually verified 444 recordings and evaluated precision, recall, F1 scores, and Receiver Operating Characteristic and Precision-Recall curves. Species-specific confidence thresholds significantly improved accuracy (precision: 0.957, recall: 0.573, F1: 0.717) compared to a universal threshold of 0.90 (precision: 0.961, recall: 0.056, F1: 0.106). When considering only species presence, the universal threshold yielded a precision of 0.956, recall of 0.479, and F1 of 0.256. Higher sensitivity reduced recall, contrasting with European studies. Despite consistently high precision, our results highlight the importance of species-specific thresholds to minimise false negatives and ensure that target species are not overlooked.
While BirdNET shows strong potential for monitoring bird diversity, its effectiveness depends on thoughtful implementation, ecological context, and continued human verification to ensure reliable performance. When applied, BirdNET can reveal patterns in species richness and activity following recent disturbances. By understanding these patterns, we can inform proactive conservation programs and help ensure their effectiveness in the face of future environmental change.

With rapid advances in technology, ecology researchers are presented with new opportunities to collect richer types of data, conduct more complex forms of analyses and display data in unique ways. To work with these new technologies often requires some knowledge of code; most technology uses code as a foundation for ingestion, processing or downloading, and users are often required to use code to wrangle and interpret these new types of data. However, many scientists lack formal training in best practice coding, despite spending much of their time working with data. This makes it more difficult to adopt new technologies, especially if they wish to make them reproducible in the long term, because poorly formatted code is prone to break. The consequence is not only that researchers may shy away from the high barrier-to-entry of new technology, but that even if they do commit to new workflows, their work may not run again in a year’s time. This talk will provide tips for creating readable and reproducible workflows in R, based on our own learnings working with and supporting users of Australia’s largest biodiversity data infrastructure, the Atlas of Living Australia. These tips are intended not only to help people write clearer code, but to learn which steps to prioritise when deadlines are tight to ensure their code is readable and will reliably run again. We hope that this talk encourages both beginners and experts to write code that is easier to share, improving the utility and longevity of their work.