Digital twins and data on the water: where AI meets Australian aquaculture

When we last wrote about AI in Australian fisheries, the focus was on cameras and catch data in the wild-catch sector. That story centred on AFMA's electronic monitoring program and the slow, practical work of getting AI to count fish on wet decks at four in the morning. That side of the industry is still progressing. But aquaculture, the farming side of Australian seafood, is where some of the more interesting AI work is now happening.

Australia's aquaculture sector produced an estimated $2.31 billion in real gross value in 2024 and 2025, making it one of the larger and faster-growing segments of the country's primary industries. The sector covers a wide range of species and operations: Atlantic salmon in Tasmania, barramundi in Queensland and Victoria, prawns in northern Australia, oysters along multiple coastlines, and a growing number of smaller-scale operations farming everything from abalone to yellowtail kingfish.

The AI applications emerging across these operations are varied, but they cluster around a few practical problems: monitoring fish health, optimising feeding, managing water quality, and making better breeding decisions. None of this is science fiction. The question is how much of it is genuinely useful today, and how much is still closer to research than reality.

Digital twins on the barramundi farm

One of the more concrete examples is coming out of a collaboration between the University of Queensland, James Cook University's ARC Research Hub for Supercharging Tropical Aquaculture, and MainStream Aquaculture Group. A PhD researcher, Jessica Hintzsche from UQ's Queensland Alliance for Agriculture and Food Innovation, is testing software that creates a three-dimensional digital replica of a barramundi farm in Victoria.

The concept is straightforward in principle. A digital twin is a virtual model of a real operation that uses actual data from the farm to simulate outcomes. What makes it different from a standard spreadsheet model or a simple simulation is that the digital twin can test multiple changes at once, using real-time data from the specific farm rather than generic assumptions. Want to know what happens if you change your parental selection strategy, adjust harvesting timing, and trial a new feeding regime all at the same time? The digital twin can model that.

For barramundi producers, the potential value is clear. Barramundi farming involves significant upfront investment and long production cycles. A wrong call on genetics or management can cost months of growth and substantial revenue. Being able to test scenarios before committing real resources and real fish is the kind of practical advantage that makes sense to operators, not because it sounds impressive, but because it reduces risk.

The honest picture, though, is that this work is still in the research phase. The researchers themselves note that nobody yet has the capacity to apply these digital twin techniques at commercial scale in aquaculture. The software is being customised for individual farms, which means it is not yet something you can buy off the shelf and plug into your operation next week.

What is working now

Where AI is making the most tangible difference in Australian aquaculture today is in monitoring and feeding, the unglamorous but critical daily operations that determine whether fish grow well and stay healthy.

Automated feeding systems that use sensors and machine learning to adjust feed delivery based on fish behaviour and environmental conditions are in use at larger salmon and prawn operations. The economics are compelling: feed is typically the single largest operating cost in aquaculture, often accounting for 40 to 60 per cent of production costs. Even a modest improvement in feed conversion ratios can have a significant impact on profitability.

Underwater camera systems with AI-powered analysis are being deployed to monitor fish health, detect early signs of disease or stress, and estimate biomass. Tidal, an AI and underwater robotics company, has been expanding in Australia's aquaculture market with camera systems and software that track and monitor fish growth in real time. For operations running pens in deep water or managing multiple sites, this kind of continuous monitoring replaces what used to require divers or periodic manual sampling.

Water quality monitoring is another area where AI adds genuine value. Dissolved oxygen, temperature, pH, ammonia levels, and salinity all affect fish health and growth. AI systems that continuously monitor these parameters and predict problems before they become critical can prevent stock losses. In prawn farming in tropical Australia, where water conditions can shift quickly with weather events, this kind of early warning has real practical value.

AFMA's expanding electronic monitoring

On the wild-catch side, AFMA's electronic monitoring program continues to grow. In 2025, the AFMA Commission approved expanding mandatory electronic monitoring to the Great Australian Bight Trawl sector, the Commonwealth Trawl sector, the North West Slope sector, the Western Deepwater Trawl Fishery, and the Northern Prawn Fishery. These fisheries will be required to have electronic monitoring between July 2026 and April 2027.

A significant development was AFMA bringing its EM review work in-house during 2025, moving away from third-party providers. This gives AFMA more direct control over data quality and, importantly, creates a pathway to implement AI-based video analysis more rapidly. When a government regulator decides to build internal capability rather than outsourcing, it usually signals that they see the technology as core infrastructure, not a novelty.

The investment here is substantial. The federal government committed $20 million to the electronic monitoring expansion, and AFMA is explicitly investing in artificial intelligence and machine learning as part of the program. The goal is to move from human reviewers watching hours of footage to AI systems that can flag events of interest and reduce the review burden.

Where the gaps are

For all the progress, several practical barriers remain. Connectivity is the most obvious. Many aquaculture operations in northern Australia, where barramundi and prawn farming is concentrated, have limited or unreliable internet access. AI systems that depend on cloud processing or real-time data transfer struggle in these environments. Edge computing, where the AI processing happens on-site rather than in a distant data centre, is part of the solution, but it adds complexity and cost.

Cost is the second barrier. The larger salmon companies in Tasmania have the scale and the margins to invest in sophisticated monitoring and AI systems. A smaller oyster operation on the South Australian coast or a family-run prawn farm in the Gulf of Carpentaria faces a very different calculation. The technology needs to be affordable at the scale these businesses actually operate, not just at the scale the vendor's business model requires.

Skills and trust are the third barrier. Aquaculture operators are, by and large, practical people who trust what they can see and measure themselves. Adopting an AI-based feeding system or a digital twin requires a willingness to trust the model's recommendations, at least partially, over decades of accumulated experience. That trust is earned slowly, through demonstrated results on their specific operation, not through conference presentations or vendor demos.

What this means for operators

If you run an aquaculture operation in Australia, the practical picture in mid-2026 looks like this. AI-powered monitoring for water quality and fish health is commercially available and genuinely useful, particularly at larger scale. Automated feeding systems are proven and deliver measurable returns on feed costs. Digital twins and advanced genomic modelling are promising but still in research. And if you fish commercially in Commonwealth waters, electronic monitoring with AI-assisted review is coming to your vessel within the next twelve months whether you are ready or not.

The broader pattern is consistent with what we see across regional industries. AI is most useful when it solves a specific, well-defined operational problem with clear data inputs and measurable outcomes. It is least useful when it is sold as a general capability without a clear connection to the daily realities of running the operation.

For aquaculture businesses looking to start somewhere, the most practical entry point remains the same as it is across most regional industries: use the generative AI tools that are already available to help with the administrative side of the business. Compliance documentation, biosecurity plans, export paperwork, grant applications. These are tasks where tools like Claude or ChatGPT can save real hours every week, without requiring any hardware installation or connectivity upgrades. The sector-specific AI, the cameras and the digital twins, will continue to mature. In the meantime, the paperwork is not going away.