As biodiversity faces unprecedented challenges, bold and innovative approaches are essential to secure a resilient future for ecosystems and the communities that depend on them. At the Ol Pejeta Conservancy in Kenya, I serve as a volunteer and operations manager supporting the safe deployment of drones, AI, and computer vision in conservation practice. These cutting-edge tools, harnessed through the global WildDrone initiative (https://wilddrone.eu/), are transforming how we monitor, protect, and understand biodiversity.

WildDrone is an international training network funded by the EU’s Marie Skłodowska Curie Actions to research and develop the use of drones for wildlife conservation. Its mission is to revolutionise conservation practices by using autonomous drone technology to monitor wildlife populations, track their movements, and manage human-wildlife conflicts, addressing the urgent need for effective and scalable nature conservation solutions. The project trains thirteen doctoral candidates in multidisciplinary skills, working together in a global network to realise the full potential of drone-based ecology and conservation.

Our work at Ol Pejeta is strengthened by WildDrone’s dynamic collaborations and accomplishments, including hosting the first WildDrone Hackathon in Kenya—a pioneering event that united technologists and conservationists to tackle biodiversity challenges through AI and computer vision. The World Economic Forum has also highlighted WildDrone’s achievements, underscoring the global importance of these efforts.

This presentation will share field experiences and lessons learned from using drones, computer vision - including thermal cameras, and AI analytics to track wildlife, assess habitat integrity, and detect threats such as poaching and habitat encroachment. By placing safety and operational risk management at the forefront—minimising conflicts with other airspace users and ensuring the well-being of people and wildlife—we demonstrate how technology can be responsibly and effectively integrated into conservation efforts. We invite the ecological community to embrace these innovative approaches and safeguard biodiversity on the brink.

NatureScan aims to advance biodiversity monitoring in Australia by leveraging consumer grade drones, remote sensing analysis and expertise in field ecology. Funded by the Department of Climate Change, Energy, the Environment and Water, NatureScan aims to develop innovative biodiversity monitoring techniques in the context of a Nature Repair Market. The project will develop robust and scientifically sound techniques to inform biodiversity across diverse ecosystems, validated with field data, and aims to place drone remote sensing technology for national-scale monitoring in the hands of land managers.

This presentation will provide insight into the impact of collection and processing parameters for drone ecosystem monitoring, and the development of scientifically sound protocols to inform on biodiversity. Using a variety of analysis techniques, remotely sensed biodiversity products can be extracted from orthomosaics and photogrammetric point clouds and validated with ecological field data, providing information into key biodiversity metrics including canopy cover, growth form, health and structural variability. To measure true ecological change in drone data products, it is important to understand the impact of user defined parameters to ensure that analysis-ready data products (a RGB and 4-band multispectral orthomosaic, and Structure-from Motion 3D point cloud) best represent ecosystem structure, function and composition.

This research has included developing clear, easily accessible standards for data collection, and has defined processing parameters that return analysis-ready data products that best represent ecosystem condition. It highlights methodological developments such as sensor alignment, orthorectification surface selection, and the generation of spatially accurate analysis-ready data products. These refinements are critical for delivering consistent, repeatable, and ecologically meaningful monitoring – essential for on-going monitoring of ecosystem condition and supporting restoration outcomes.

Australia is experiencing rapid growth in wind farm development driven by renewable energy targets. Although windfarms have many environmental positives, turbines present a risk to birds and bats, particularly raptor species which have documented mortality caused by turbine strikes.
To avoid avian casualties, a standard recommendation is for windfarm projects to conduct monitoring and implement mitigation strategies to both understand collision frequencies and to decrease the likelihood of turbine strike. One of these strategies is to identify nests and ensure measures are implemented to reduce the likelihood of strike, such as turbine exclusion buffers around nests. However, the identification and continued monitoring of raptor nests can be a difficult task as they typically build their nests in hard to access areas resulting in active nests not being identified and a subsequent lack of an exclusion buffer. As technology has evolved, new survey methods to identify nests are available, including drone surveys.
Drone surveys for Wedge-tailed Eagle nests were undertaken as part of the mitigation measures for a proposed windfarm in South Australia. The drone surveys provided a “birds-eye” view of the project site, allowing ecologists to identify nests that could not be accessed during ground surveys. Drones also provided a view of the contents of the nest to determine whether it was active or not. Still, with new technology comes new risk such as how raptors will react to the drone and whether it could have adverse impacts on nesting success. The development of drone methodology and mitigation measures illustrates the potential of drone surveys to increase identification of nests and provide greater protection for raptors as well as ongoing monitoring potential.

Kangaroo population monitoring is an environmental challenge where thermal drone technology could make substantial contributions based on the need to frequently survey large areas. Despite potential, thermal drones require further development before being used as an optimised kangaroo monitoring technique, in particular the development of machine learning and automated detection capabilities to make use of large thermal datasets. This talk will present protocols for surveying kangaroo populations using thermal drones, considering timing, flight heights, camera angles, field-of-view and drone/camera combinations. This talk will also detail our post-hoc footage review and data annotation process, as well as tests of an automated counting and tracking system for kangaroos underpinned by machine learning models. This talk will present the testing and training approach for this automated system and provide comparisons of machine-learning assisted counts against ground-truth data and conventional monitoring approaches such as walked transects and helicopter surveys. Further optimisation of these tools and user-friendly protocols for deploying them will greatly support kangaroo monitoring.

Obtaining accurate estimates of body length (BL) is critical for assessing population structure and health of marine mammals. Existing drone methods have been restricted to images taken during surfacing events when the animal’s body is nearly straight, and orientated parallel to the water surface. As marine mammals’ bodies are typically curved with considerable individual variation, this approach may result in significant error or substantially reduced sample size, especially for animals lacking definite landmarks and with short surface intervals. The aim of this study was to examine the need for, and accuracy of, a new method to estimate BL of wild dugongs, that accounts for body curvature. Ten-minute videos were collected from 98 dugongs using two drones with cameras angled 90° downward or 30°–45° obliquely: body measurements of four dugongs were ground-truthed directly, and the other 94 dugongs were filmed at random to examine variation in surfacing posture. Three-dimensional coordinates of the dorsal bodyline were estimated using triangulation. Curvature ratio was defined as BL estimated using vertical only, versus vertical and oblique shots. Surfacing postures were classified into three types: horizontal (23.4%), nose-up (35.1%), and curved (42.6%). Curvature ratios were 97.3 ± 0.1%, 93.6 ± 2.7%, and 87.5 ± 6.3% for each posture respectively. Ground-truthed dugongs showed curvature ratio of 97.0 ± 1.5% (n=32 dives). Root mean square error of the method considering body curvature was 6.4 ± 3.8 cm (2.4 ± 1.1% of BL, n=4 dugongs), which was comparable with or better than previous studies on related species. A large proportion of dugongs showed high curvature rates, and the proposed new method yielded estimation of errors small enough to compensate for differences in BL caused by curvature. BL estimation via drone that accounts for body curvature could improve the number of sampling opportunities and measurement accuracy for marine mammals.

Uranium mining of Pit 1 at Ranger in the Northern Territory of Australia concluded in 1994. Pit 1 was revegetated in 2022 after it was backfilled with tailings and capped with waste rock. The size of Pit 1 (~40 ha) presents challenges for revegetation monitoring. Our aim was to develop a rapid and cost-effective method for landscape-scale monitoring of the revegetation progress and assessment of vegetation health using very-high resolution satellite imagery. Since May 2023, Airbus Pleiades Neo satellite imagery (6-band multispectral, 30 cm pansharpened spatial resolution) across dry and wet seasons were routinely acquired for the Ranger mine, with the revegetated Pit 1 as our area of study. The satellite platform allowed acquisition of critical data during wet/monsoon seasons which were not possible by ground survey or drone capture. Vegetation indices were evaluated to characterise dry-wet seasonal growth dynamics for revegetated areas on Pit 1, with mean Normalised Difference Vegetation Index (NDVI) found to be suitable to track vegetation growth and health over time across different management zones on Pit 1. Results generated from satellite data were validated with corresponding drone multispectral data. NDVI was further combined with vegetation cover to indicate both plant health and density. A combined analysis of Pit 1 vegetation cover based on NDVI from 2023 to 2025 showed areas of consistent poor plant growth, earmarking these areas for further investigation. High resolution satellite data provides a rapid method to monitor the revegetation progress at the Ranger mine across a large spatiotemporal scale.

Camera trap surveys are increasingly used for wildlife population estimates but analysis is still impeded by the number of images produced. This reduces the utility of camera trap surveys for conservation managers who need timely population estimates to inform adaptive management actions. There is potential to address this delay by more fully exploiting technological advances commonly used in other fields for ecological study. These include automated image processing, deep learning ensembles, an expanding range of remote sensors, and more accessible spatial analysis tools.
To date, there is an absence of studies that are truly exploiting the potential of these technologies in combination. This presentation introduces a method for creating and evaluating a bespoke deep learning ensemble designed to process a camera trap dataset of >200,000 images. A case study is presented in which the automated output is combined with data extracted from GIS spatial proxies via a fully-scripted workflow to create a series of hierarchical models, using the unmarked R package to produce a population estimate without any field-collected data or distance measurements. The semi-automated workflow is repeatable and can be used to unlock the potential of datasets that are otherwise limited by size and / or an absence of field observations. The workflow has high potential for use with pre-existing camera trap datasets that have not been fully exploited for knowledge generation.

Spatiotemporal variation in environmental moisture availability drives many landscape processes, such plant growth, fuel accumulation and fire activity. Land managers and researchers need reliable environmental moisture data to understand vegetation drought stress and associated fire risk, amongst other processes. Accurate measurements of the moisture content of the air, soil and fuels are easily obtained through direct measurement\. However, the challenge is to produce accurate estimates across the entire landscape at a high spatiotemporal resolution. The growing use of in-situ sensors represents attempts to address the temporal component of this challenge, but fails to produce data that is reliably extrapolated across the landscape. Landscape scale moisture estimates produced from satellite remote sensing and meteorological products address the temporal and spatial challenge, but may be less sensitive to fine-scale microclimatic variation and therefore less accurate. We used in-situ measurements of environmental moisture at over 90 forest sites in south-eastern Australia, to quantify the uncertainty of several widely used remote estimate products. Environmental moisture estimates included fuel moisture, vapor pressure deficit (atmospheric moisture) and soil moisture content. We accounted for drivers of fine scale microclimatic variation such as terrain and vegetation cover, to understand the causes of this uncertainty. Uncertainty in remote estimates varied between products, but was generally greatest for drier observations. Vegetation cover was associated with the greatest discrepancies between remote and in-situ moisture measurements. This research highlights the limitations of landscape scale environmental moisture estimates. The most extreme drying events, which lead to large, high severity bushfires, may not be accurately represented by remote moisture estimates Further, the accuracy of remote estimates may also differ at a sub-pixel scale due to variation in vegetation cover and terrain. However, these results can be used as a basis of interpreting landscape scale environmental moisture data and designing more accurate products.

Habitat loss, degradation and fragmentation are leading anthropogenic drivers of declining biodiversity, such that they are key criteria in assessing species’ conservation status. The Adelaide Mount Lofty Ranges (AMLR) was one of the first regions in Australia to undergo extensive land clearing and remains among the most impacted via its extinction debt. In response, bold conservation initiatives, including implementation of artificial habitat, are being implemented to mitigate the loss of habitat threatening species within the region.
However, when such implementation lacks empirical evidence to support species survivorship, it may waste valuable conservation resources or even worsen species’ decline by creating ecological traps. Even where broad-scale analysis identifies habitat as suitable, fine-scale ecological requirements can differ substantially between natural and artificial habitat, creating a divide between ostensive suitability and ecological reality. Understanding species-specific microhabitat preferences is therefore critical for effective conservation planning.
We identified Egernia cunninghami (Cunningham’s skink), a saxicolous habitat specialist under local threat in the AMLR, as a candidate for artificial habitat provision. This study aimed to quantify the fine-scale rocky outcrop habitat attributes associated with E. cunninghami occupancy by surveying occupied and unoccupied outcrops across its AMLR distribution.
We used this case study to evaluate photogrammetry-based 3D modelling as a non-invasive, cost-effective tool for surveying microhabitat preferences. To our knowledge, this is the first study to create digital reconstructions of rocky crevices for ecological research.
We identified several key variables influencing occupancy of crevices by E. cunninghami and provided dimension estimates for appropriate shelter sites (crevices). Concurrently, we identified other attributes for which E. cunninghami displayed surprising plasticity.
Our findings highlight both the value of photogrammetry for applied conservation and the importance of quantifying fine-scale habitat preferences before implementing conservation solutions. This research supports a widely applicable, scientifically grounded approach to habitat restoration for specialists in fragmented landscapes.

VegMachine.net is a free online platform for analysing long-term vegetation trends across Australian landscapes by providing a time series drill of medium-high resolution (10-30 m) satellite imagery. Of particular interest to ecologists, VegMachine offers an easy way to extract and summarise decades of vegetation cover and rainfall data for user-defined locations. Several ecological applications using VegMachine-derived datasets will be presented, including recent work on the greater bilby (Macrotis lagotis) within its current Queensland range. This study used landscape features as proxies for predation, competition, and productivity in a series of multivariable binary logistic models, with the best-supported model identified as the combined predation-productivity model. Use of the VegMachine-derived fractional ground cover time series allowed for nuanced analysis of vegetation productivity, including relative greenness over time, mean greenness through time and actual greenness during the bilby survey period. Satellite imagery is very useful for this kind of analysis, placing current and previous ecological surveys into their spatial and temporal context, but accessing these datasets can require specialist knowledge or extensive processing. VegMachine offers simple, easy to use tools to explore and extract these datasets, while recent major upgrades to VegMachine have prioritised mobile accessibility and user experience, including new analysis tools, and data persistence for seamless use across sessions and when offline.

Life cycle assessment (LCA) is increasingly used by businesses to quantify their impacts on biodiversity. However, current LCA methods remain limited in their capacity for spatially explicit biodiversity assessments, especially for midpoint impacts. This review examines how biodiversity impacts have been spatialised within and beyond the LCA framework, and explores methodological opportunities to enhance spatial resolution and ecological relevance.
We systematically reviewed 354 peer-reviewed articles and retained 150 studies that assessed biodiversity impacts in a spatially explicit manner. Of these, 58 were LCA-based and 92 employed non-LCA approaches. Among the LCA studies, only 27 quantified spatialised midpoint impacts, and the most studied one is land use, followed by ecotoxicity and climate change. A total of 49 studies assessed endpoint impacts, typically using the potentially disappeared fraction of species (PDF) as an indicator. While widely applied, PDF indicators often lack ecological specificity. Spatialisation in LCA was primarily achieved through regionally differentiated life cycle inventories or spatially explicit characterisation factors.
Non-LCA studies were grouped into five categories: state-based, exposure-based, response-based, interaction-based, and qualitative assessment. Spatialisation methods varied across categories. Ecological models like species distribution models were commonly used in response-based studies, spatial overlays in exposure-based studies, and empirical field data supported the rest.
Overall, spatialisation in LCA remains concentrated on land use and weakly connected to species-level outcomes. This review highlights opportunities to expand the spatialised modelling of various midpoint impacts and to incorporate ecological methods from non-LCA research, thereby supporting more spatially accurate and biologically meaningful biodiversity assessments.