Digital twins are moving beyond buildings and machines into reefs, cities and farms, where complex environmental data is turned into detailed schematics to improve decision-making.
On the Great Barrier Reef, change can be hard to see clearly, even when it is happening quickly. After bleaching events, cyclones or marine heatwaves, researchers need to know more than how much coral remains. They need to understand which colonies survived, where new corals are recruiting, which parts of the reef are growing and how the structure of the habitat is shifting over time.
To capture that detail, engineers are turning sections of reef into digital models. Divers, boat-towed cameras and underwater robots photograph coral from multiple angles, while scale markers fix the measurements. Software then stitches thousands of overlapping images into a 3D reconstruction.
That reconstruction shows the reef at one point in time, but a digital twin goes further. By adding new observations and connecting them with data and modelling systems, it becomes a living model that can help researchers ask what might happen next.
“The defining feature of a digital twin is its dynamic nature,” said Dr Renata Ferrari, Senior Research Scientist at the Australian Institute of Marine Science (AIMS), where she leads research into coral demographics and spatial ecology.
“While traditional 3D models and maps offer only a single snapshot in time, a digital twin continuously evolves.”
Not so natural
For many engineers, digital twins are already part of the language of infrastructure, manufacturing and the built environment. A digital twin of a bridge, building or machine can help monitor performance, simulate stress and test changes before they are made in the real world.
Nature is less obliging. Reefs, catchments, oceans, farms and cities are shaped by climate, weather, water, biology, genetics, land use, human behaviour and chance. They can’t be fully instrumented or neatly controlled. Yet as climate pressure grows, researchers and land managers need better ways to see what is changing, update their understanding and make decisions under uncertainty.
In the Netherlands, Professor Femke Vossepoel works on the problem of bringing models and observations together in ways people can use. As Professor of Earth System Simulation at Delft University of Technology, her expertise is data assimilation: combining numerical models with observations. This is a familiar idea in weather forecasting, where models are continually updated with new data to improve predictions.
The challenge now is applying that logic to other systems, from cities and farms to reefs and catchments.
“A digital twin is not just a model,” Vossepoel said. “It’s a tool that helps a decision-maker evaluate different criteria, run what-if scenarios and decide what to do.”
A dashboard can tell users what has been measured, a map can show where change is happening, and a model can simulate how a system behaves. A digital twin, at its strongest, brings those functions together, combining observations and models so users can test interventions, compare risks and understand uncertainty before acting.
READ: How digital twin technology is cutting costs and emissions
Image to intervention
At AIMS, reef digital twins are being used across restoration, monitoring and ecological research. Ferrari describes the workflow in five broad stages: data collection, processing, extraction, integration and analysis.
High-performance computing can help scale the work, but Ferrari said accessibility matters just as much as technical power. Her team focuses on off-the-shelf tools and regularly updated standard operating procedures so the methods can be adopted beyond specialist research environments.
“Global impact requires accessibility,” she said.
Through the Ecological Intelligence for Reef Restoration and Adaptation Program, known as EcoRRAP, AIMS and its partners use digital twins to track coral recruitment, growth and survival across the Great Barrier Reef and Torres Strait. The aim is not simply to make reefs visible in 3D, but to turn repeated observations into evidence that can feed models, guide restoration and test where heat-tolerant corals may have the best chance of thriving.
Ferrari said the technology allows researchers to follow tens of thousands of individual corals at once. The same models can show how reef structure influences fish diversity and abundance, with insights feeding into engineered habitat design, including “fish-smart” seawalls designed to support biodiversity or attract target species.
AI is helping AIMS researchers extract useful information from the large image sets behind digital twins. RapidBenthos, an automated tool developed by AIMS, analyses reef imagery to help quantify what is living on the seafloor. AIMS has also worked with La Trobe University to test 3D Gaussian Splatting, a visualisation method that can create highly realistic reef representations.
Ferrari said traditional photogrammetry remains stronger for precise measurement, while newer 3D approaches can help people see and understand the reef more easily. For digital twins, both functions matter: the system needs reliable measurements to feed models, and clear visualisations to help decision-makers understand what those models are showing.
READ: Why Australian engineers need to jump aboard the digital twin bandwagon
Beyond past data
Vossepoel is currently working on UrbanAIR, a $23.6 million Horizon Europe project developing urban digital twins for air quality and heat resilience. Horizon Europe is the European Union’s main research and innovation funding program, and UrbanAIR brings together 18 partners across 11 countries to help cities plan for heat, air pollution and climate risk.
The project connects to Destination Earth, a European Commission initiative to build a highly detailed digital model of the planet. UrbanAIR draws on that larger infrastructure, then zooms down from climate-scale modelling to the street-level questions cities need to answer: where heat will build, how air pollution will move and what might change if trees, buildings or evacuation plans are altered. It shows how the digital twin concept can connect large-scale climate modelling with the local decisions made by councils and city managers.
Vossepoel said one newer challenge is integrating human behaviour with physics-based models. Another is computational cost – fine-scale probabilistic modelling can require thousands of simulations, and the observations needed at street scale are not always available.
AI can help digital twins process large datasets, but Vossepoel said it can’t replace physical modelling. A model trained only on past data may struggle when climate change pushes natural systems into conditions not previously observed.
“If you stick to the physical equations, then you know that whatever your model or twin is going to predict is something that is physically consistent and sensible,” she said.
The same issue applies to reefs, catchments and farms. A system under repeated bleaching, altered rainfall or hotter and drier conditions may not behave as it did in the past. Digital twins need data, but they also need models that can work when history is no longer a reliable template for the future.
READ: Transforming urban planning using spatial digital twins
Start with a decision
In agriculture, the same principle applies at landscape scale. Dr Roger Lawes leads CSIRO’s Spatial Temporal Decisions Team, which operates from Canberra and Floreat in Western Australia, developing digital agriculture technologies that draw on satellites, crop models, climate data, soils data, Australian Bureau of Statistics data, remote sensing and machine learning.
The team works across climate response, land-use change, resilience, nitrogen fertiliser decisions, genetics, crop rotations and farm economics. The questions vary, but Lawes said the starting point should always be the same: who is using the tool, and what decision do they need to make?
“You have to focus on understanding the end user and how they will engage with the platform, not how you as a scientist or a software engineer might interact or design it,” Lawes said.
According to Lawes, CSIRO has built an internal proof-of-concept agricultural digital twin for Australia, bringing together datasets on wheat, cotton, land use, soils, climate and other agricultural variables. Through a web-based interface, users can explore questions about farming systems and landscapes without needing to assemble the data themselves.
A farmer-facing question may sound simple: should a land use change, an input be adjusted, or a crop decision be made differently? Behind the interface, the answer may depend on crop, pasture or livestock data, soils, satellite imagery, farm boundaries, cadastral data, climate records, machine learning, and in some cases logistics or transport networks. The user doesn’t need to see every layer, but the tool still has to turn those layers into something usable.
For Lawes, that is what separates a digital twin from a dashboard: A digital twin should be able to perform complex analysis, replicate a real-world system and inform a decision. A dashboard, he said, “throws a suite of numbers at you in pretty graphs” but doesn’t necessarily help someone decide what to do.
READ: How dynamic digital modelling is helping engineers optimise electrical systems
Central test
For all their promise, ecosystem digital twins are still at different stages of maturity. Some are already helping researchers measure change, validate models and guide restoration planning. Others remain closer to prototypes, research infrastructure or sophisticated visualisations.
The central test is whether they can move beyond representing a living system to helping people decide what to do.
In reef restoration, Ferrari said the next hurdles are user-friendly simulation platforms, automated 3D data extraction and broader accessibility. AI is streamlining some two-dimensional data extraction, but pulling reliable information from 3D digital twins remains difficult. New tools also need rigorous validation to ensure they produce data that are reliable and reproducible.
Ferrari imagines a future in which practitioners could use an intuitive virtual-reality environment to run virtual experiments and predict restoration outcomes at their own sites. For now, she said, accessible simulation platforms remain an aspiration.
Vossepoel sees the same gap between technical capability and practical use. “At the moment I see that as the biggest risk: that we develop a computational engine that is very heavy, but may never be used in practice,” she said.
Uncertainty is also part of the decision-making process that digital twins will inform. Vossepoel’s work uses uncertainty quantification to explore likelihoods, probabilities, and best-case and worst-case outcomes. Lawes said the challenge is not only to produce those measures, but to communicate what they mean for the person making the decision.
“The point is just trying to inform that decision,” Lawes said. “You can’t make it for them.”
That may be the clearest test for ecosystem digital twins. The technology can make reefs, cities and farms more visible, but visibility alone is not enough. The useful twin is the one that helps people ask a sharper question, understand the limits of the answer and make a better decision in the real world.
READ: Digital twins in the rail sector: Avoiding downtime, improving safety and reducing costs
The stack behind ecosystem twins
An ecosystem digital twin is not a single technology. It is a chain of tools and processes that turns observations into usable insight.
It starts with observation. Depending on the ecosystem, that might mean satellite imagery, drones, underwater cameras, photogrammetry, environmental sensors, field surveys, climate records or farm-level data.
Next comes processing. Raw images, sensor readings and spatial datasets need to be cleaned, quality checked, aligned and made usable. In reef work, that might mean turning thousands of overlapping underwater images into a scaled 3D reconstruction. In agriculture, it can involve combining satellite imagery with soils, weather and crop data.
Then information needs to be extracted. AI and machine learning can help classify features such as reef composition, crop type or land use. For more complex systems, extracting reliable information from 3D models remains difficult and needs careful validation.
The next step is integration. A useful twin does not simply store information. It connects observations with physical, ecological, hydrological, climate or crop models, so the system can be updated and used to explore change over time.
Simulation adds the “what if?” layer. Users might test what happens if a heatwave hits, a restoration site is chosen, a crop input changes or a land use shifts.
The final piece is the interface. This might be a map, dashboard, visualisation, virtual environment or decision tool. Its job is to make the system usable by the people who need to interpret it.
The stack only works if those layers serve a clear purpose. Before engineers choose the sensors, models or interface, they need to know what question the twin is being built to answer.
Watch the EA OnDemand webinar Simulating the great barrier reef from our Industry Partners Series.
