Problem statement

Tasmania is a known hotspot for plant biodiversity, due to the relatively cool, wet, and stable environment over many long-range climate cycles Harrison & Noss, 2017Macphail, 1979Read & Hill, 1988. However, rapid changes to the environment are eroding these refugial habitats faster than ever before, posing an existential threat to the ancient species persisting within Mokany et al., 2017. The threat to these vegetation communities generally is well understood Bliss et al., 2021Fletcher et al., 2021Worth et al., 2016, but there is little insight into the vulnerabilities at a population level needed to inform targeted conservation planning. Targeted and adaptive conservation planning is non-trivial, as it requires comprehensive species inventories, identifying and quantifying vulnerabilities, and ongoing monitoring at temporal intervals appropriate for the target McCarthy & Possingham, 2007Williams & Brown, 2016. This is a challenge that extends beyond Tasmania, with research showing that adequate conservation assessments for endemic flora species fall short of requirements globally Gallagher et al., 2023.

Population inventories are a fundamental parameter for effective conservation planning, but the immense complexity of natural terrestrial ecosystems can make it challenging to gather data that accurately represents a population and the changes over time Burg et al., 2015Good et al., 2006. The approaches employed by the Department of Natural Resources and Environment for mapping vegetation are centred around manual aerial photo interpretation Department of Natural Resources and Environment Tasmania, 2020. While this method is effective for mapping vegetation at the community level at a regional scale, it is insufficient for monitoring threats to individual trees and changes to populations over time. The need for detailed and timely data of these trees was made obvious in February 2025, when bushfires burned within metres of Huon pines estimated to be 3000 years old Lohberger, 2025. These objectives are particularly important to the conservation outcomes of long-lived tree species with multi-century generations Leonard, 2021Leonard, 2021Lindenmayer et al., 2012.

Remote sensing and deep learning technologies have the potential to revolutionise automated species inventories at a regional scale, but the substantial training data requirements combined with variably in sensor hardware, configuration, conditions, and target definitions make the approaches challenging to deploy in real world scenarios Brandt et al., 2025. Many of these challenges have well established solutions and are already integrated into systems we use daily Safonova et al., 2023. Knowledge distillation is one approach offers a promising pathway to overcome these limitations by transferring learned representations from high-fidelity datasets to low-fidelity datasets, yet this approach remains largely unexplored in vegetation mapping applications Zhong et al., 2024. Similarly, while multimodal approaches that integrate environmental and ecological contextual information have demonstrated improved performance in species distribution modelling, their application to dense segmentation tasks for flora remains limited, with virtually no research examining how models utilise this domain knowledge Brodrick et al., 2019Harmon et al., 2023Tiel et al., 2025.

Yet the complexity of both remote sensing data and the target/objective (i.e. vegetation), mean these techniques remain inaccessible to decision makers. The convergence of conservation needs for flora in Tasmania and worldwide, technological maturity of RS with DL approaches, and critical knowledge gaps in operational implementation creates a unique research opportunity to develop scalable, species-level monitoring frameworks that can provide the spatial detail, temporal frequency, and demographic information required for adaptive conservation planning of these vulnerable trees.