Ecological forecasting plays a critical role in supporting adaptive management for conservation. However, the complexity of ecosystems, coupled with data limitations and unpredictable environmental drivers, pose significant challenges. Drawing on insights from long-term small mammal monitoring in arid Australia, we propose a five-step framework to improve forecast reliability: 1) align ecological assumptions with statistical models, 2) manage missing data effectively, 3) select models balancing complexity and interpretability, 4) assess model performance across training lengths and forecast horizons, and 5) incorporate projected environmental covariates. Tailoring statistical models to species-specific dynamics, rather than applying a one-size-fits-all approach, improves interpretability and reliability. While zero-filling missing values distorts trends, we show that multivariate imputation by chained equations (MICE) preserves population signals and improves forecast performance across varying data gap patterns. We highlight the use of multivariate generalized additive models (MVGAMs) in capturing nonlinear species–environment relationships, site-level effects, and temporal dynamics. Comparing Gaussian process and vector autoregressive (VAR) latent trends reveals trade-offs between accuracy and computational feasibility in data-limited settings. Bayesian approaches enable probabilistic forecasts, showing that even short time series can capture boom–bust dynamics, though uncertainty increases with forecast horizon and species variability, highlighting the importance of understanding the optimal amount of data required for model training and the forecast horizon that can provide reliable predictions. The ecological significance of covariates can shift across forecast horizons, and proxies like dominant vegetation cover, can improve model stability. Forecast reliability depends on training length, forecast horizon, and climate scenarios. Lastly, using projected rainfall, we found how the interpretation of population trends varies between different forecast horizons. Species exhibit different responses to climate change, some resilient and others highly sensitive. Our framework strengthens the integration between ecological understanding and forecasting tools to inform robust, evidence-based ecological management.

The National Ecosystem Assessment System for Australia (NEASA) is a collaborative initiative between TERN and CSIRO that aims to establish the scientific foundations for a national-scale ecosystem condition assessment framework. A core component of Phase 1 of the initiative is the development of nationally consistent ecosystem archetype and state-and-transition model (STM) templates that describe the structure, dynamics, and condition of Australia’s native ecosystems, both in their reference state and under modification.
These templates offer a standardised, scalable method for understanding and communicating ecosystem change. Key benefits include improved consistency and comparability across ecosystems and regions, more efficient and structured expert engagement, and enhanced utility for national initiatives such as natural capital accounting and the Nature Repair Market. The templates support meaningful interpretation of ecological change and decision-making by distinguishing between states that are similar in condition but differ in ecological processes or drivers.
In this presentation, we will demonstrate how these national templates can be downscaled and adapted to regional contexts through a case study example. This case study will highlight how empirical data and expert knowledge can be used to refine and validate conceptual models. It will also illustrate the flexibility of the template framework and its applicability to diverse landscapes and management challenges.
By synthesising empirical data and expert knowledge into structured templates, NEASA is building the infrastructure foundation for a robust and repeatable national ecosystem assessment system. The resulting STM templates can be used to assess ecosystem condition, guide restoration, and inform management, while remaining adaptable to regional ecological contexts.

The National Koala Monitoring Program is a collaboration between CSIRO, DCCEEW & State Governments to deliver a robust estimation of Koala populations and build long-lasting capability to assess trends in those populations. Using a network of government, community and scientific koala data providers the program will contribute to trusted local, regional and national koala population assessments through a range of complimentary monitoring methods.
The NKMP utilises a wide range of approaches to monitoring koalas which enables us to use existing data and suit our ongoing data collection methods to the specific needs of each site. While multiple approaches are harder to integrate into a single model, they have the advantage of fostering partnerships to share responsibility for koala monitoring and management in different parts of the species range. These partnerships are vital to ensuring that regular, robust data continues to be collected to enable informed management of the species.
Given the range of techniques utilised by the model it is critical to compare the accuracy and precision of the different survey methods across different habitats, allowing calibrations and corrections to be applied to the model. These calibration experiments provide crucial estimates of relative abundance between the methods. They provide a link that will enable broader-scale analysis, and do so with more confidence and ultimately less assumptions.
Here we present results from calibration studies using techniques including Daytime walk transects, Spotlight transects and Thermal Drone transects conducted at “Tier 1” monitoring sites.

Evidence-based decisions to effectively manage the impacts of threats on imperilled biodiversity rely on knowledge of which threats have the greatest impact on threatened species, and where these impacts are most severe. However, current information on the spatial distribution of key threats in Australia, and their impacts on threatened biodiversity is incomplete. In a partnership between the Australian Government and Australian ecologists, funded from the Australian Government’s Saving Native Species Program, we aim to develop and apply a framework that systematically integrates information on spatial distribution of threats, their potential overlap with threatened species distributions, and traits and attributes from which the likely severity of the impacts can be inferred. To do so, we model the spatial distribution of selected threats (initially focussed on invasive species) using species distribution models informed by published evidence and expert elicitation, and develop a novel traits-based framework to estimate species susceptibility to a given threat. The traits-based framework categorises native species’ susceptibility to invasive species, based on the native species’ traits (morphological, behavioural, life history) and other attributes including ecological variables (e.g., preferred habitat, specialist/generalist), or contextual variables (e.g., range size, population decline). Combining the spatial distribution of threats and native species with the species susceptibility will show where the threats are having the most impact in Australia’s landscape. The approach is intended to inform a range of regulatory, policy and investment decisions, including spatial prioritisation of threat management to mitigate current impacts, and where biosecurity and incursion management is required to reduce potential future impacts. The goal is to prioritise threat management in areas where the risk of impact on biodiversity is most significant. Furthermore, the approach is designed to be explicit and transparent, and iterative in nature so that new threats and species can be assessed as new information becomes available.

A rapid transition to renewable energy is essential for climate change mitigation, yet poorly planned developments risk contributing to the biodiversity extinction crisis and losing social licence through impacts on sensitive species. We provide the first quantitative analysis that demonstrates how renewable energy can be strategically deployed to effectively erase biodiversity impacts while meeting energy needs cost-effectively. Focusing on Queensland, Australia - a global biodiversity hotspot experiencing rapid renewable energy expansion - we quantify how increasing levels of biodiversity protection only modestly increases costs of renewable energy infrastructure projects required to meet electricity demand in a net-zero emission energy system in 2050. By avoiding 30% of the most important areas for threatened species when developing renewable energy infrastructure, we avoid 90% of species’ distributions, with 77% of species distributions completely avoided. Avoiding infrastructure in these areas adds just 1-2% to electricity bills in 2050. Increasing protection to 50% of lands protects 96% of species distributions, adding just 2-4% to electricity bills in 2050. These cost increases are likely much smaller than the uncertainty of the planning task, rendering them effectively unobservable. We show how other Australian states also have high land use flexibility relative to energy demand, indicating the generality of our results in Australia and the global transferability of our method and analysis. Our approach reveals opportunities for biodiversity protection within energy transition planning and challenges the perception that global renewable energy and biodiversity protection targets are irreconcilable.

We use the outputs of species distribution models to make conservation decisions such as where money and resources will be allocated, which areas need to be protected and how to manage invasive species. In my honours project, I aimed to quantify the accuracy of MaxEnt in predicting species’ future distributions. I fit MaxEnt models to species’ occurrence data from 1958-1990 and asked where each species would occur in the 2019-2023 time-period. I then quantified the difference between the MaxEnt predictions, and species’ actual 2019-2023 distribution data. In building and running the models for 34 plant species within North America, I evaluated the models using both the widely-used area under the curve (AUC) metric, as well as more novel spatial metrics including the distance between centroids (DBC) of predicted and observed occurrences. The models performed well in AUC across species (Mean AUC = 0.94 ). However, distance between centroids was significant with values between 52 and 803km. This study raises concern for the reliance on species distribution models when making conservation decisions and highlights the importance of strengthening our current evaluation methods.

The Habitat Condition Assessment System (HCAS) has been collaboratively developed over the past ten years between CSIRO and the Department of Climate Change Energy the Environment and Water (DCCCEW). At the outset the HCAS was recognised as a novel application of remote sensing data with potential to support more evidence-based conservation decisions and reporting. Its key offering is an ability to report in a consistent, transparent and repeatable way on ecosystem and landscape condition at a fine spatial scale, with completeness across continental Australia.

Prior to the HCAS, reference-based condition assessment could only be done at field sites using local benchmarks, which could then be used to inform coarse-scale reporting based largely on expert opinion informed by available site data and available Earth observation imagery. Such approaches, while valid, are compromised by a general lack of consistency, granularity, currency, repeatability and uncertainty quantification.

The HCAS has been used as a key input to several decision support tools and reporting processes within DCCEEW. Recent improvements in its resolution, model-based uncertainty quantification, time-series of monitoring and currency have improved its application in species conservation action planning, corporate reporting and the Nature Repair Market.

Parental coordination is quantified by two metrics: (1) alternation, the level of turn-taking between partners, and (2) synchrony, the temporal organisation of each parent's behaviour with the other. High coordination levels may enhance nest success by reducing the risk of nest predation and ensuring that offspring consistently receive warmth and food from their parents. Coordination requires both parents to invest in the clutch. Birds might be more motivated to increase investment in the clutch and to coordinate with their partner if they are paired with a high-quality partner. As a case study, we examined parental coordination in the Spotted Pardalote (Pardalotus punctatus), a socially monogamous passerine in which both sexes incubate the eggs and feed the chicks. We investigated whether pairs coordinate their activities at the nest, whether higher coordination correlates with greater nest success, and the effect of individual quality on pair coordination. We placed cameras at 87 nests during incubation and chick rearing, and tested if partners coordinate their visits more than what would be expected by chance with a randomisation approach. Individual quality was quantified using scaled body mass index. This study will provide insights into coordination during incubation and chick rearing in a passerine bird with biparental care at all breeding stages, a rare focus given that male passerines in the Northern Hemisphere typically do not incubate, and into the interaction between individual quality and coordination in a socially monogamous species.