The Next Frontier: How AI and Underwater Robots Can Help Us Better Understand Sharks
SHOW NOTES
Martina has spent a lot of time in the ocean, and says she feels lucky to have encountered sharks many times [6.58]. “It just feels so special…like you’ve been chosen, and you’ve been granted this privilege of seeing something so out of this world.” One particular stand-out memory comes from Martina’s time working on the Crown of Thorns Control Programme for the Great Barrier Reef Park Authority. She was diving at a site that she describes as an underwater “cathedral”, with great structures that rise from the seabed. Because she was looking for crown of thorns starfish, Martina usually had her head stuck in the tiny nooks and crannies of the reef, immersed in detail. But one particular day, Martina looked up and saw a large tiger shark gliding overhead. “The sun was shining through the water, and the reef was sparkling, and she was just so slow and perfect…you’re like wow, how did evolution make something so perfect? How is that even possible?” She smiles.
Martina grew up an ocean lover, and remembers sailing in the Mediterranean sea during her childhood [11.40]. But she didn’t come across sharks until an opportunity came up to volunteer for a project in Fiji, led by shark scientist Dr Diego Cardeñosa. Nine months later, Martina came away knowing that she wanted to work with sharks. During this time, she connected with many students and researchers from James Cook University (JCU) in Australia [14.10]. They encouraged Martina to apply for a marine science degree at the university, and a year later she arrived in Queensland, ready to start.
Martina completed her master’s degree at JCU, and now works as a research assistant for the Biopixel Oceans Foundation and a research associate for the Fish and Fisheries Lab at JCU. Her research focuses on the capabilities of new technologies, like AI and ocean robotics, in shark science, especially their performance compared to more traditional research methods [15.54]. This is a field that has advanced substantially over the years, and continues to move at a rapid pace. “New technologies have naturally made their way into our traditional methods, as soon as underwater cameras became a thing, and they became more accessible to researchers.” Explains Martina. For example, technologies like BRUVs (Baited Remote Underwater Video systems), have really furthered our understanding of sharks to date and helped to overcome the limitations of using only dive surveys. These are camera systems that can be lowered onto the seabed, and are baited to attract certain species. They can go deeper than divers can, and are less intrusive, allowing scientists to observe sharks in their natural environment. But up until recently, BRUVs have only had forward-facing cameras, meaning that their field of view was limited. “But now we’re seeing 360-degree cameras come through, as a way to counteract that front view only camera, which is a reasonably big spatial limitation of this technology.” Says Martina. Similarly, drones have become increasingly more accessible, and are being used more and more for science. Drones can be our ‘eyes in the skies’, helping to locate, observe and track animals from above. And ROVs, or Remotely Operated Vehicles (ROVs) operate like underwater drones, and can be our ‘eyes in the sea’. These are being used more and more as well, to try and get deeper and obtain knowledge of sharks in areas where divers can’t go.
More specifically, Martina has been trying to understand how new technologies can address the spatial and temporal limitations of traditional methods [17.20]. For example, there are safety concerns with conducting dive surveys of sharks at night, making it difficult to obtain a risk assessment or permit. Divers are also limited by depth. BRUVs and fishing surveys go some way to counteract this, but there are still biases to these methods. And this leaves a glaring knowledge gap. Some sharks may be more active at night, or different species come out as the sun fades. Equally, some species undertake vertical migrations, moving into deeper waters during the day or night. It is therefore important that we address these spatial and temporal limitations to gather a more comprehensive understanding of sharks and rays – information that can be used towards their protection.
Martina’s work so far has focussed on two technologies in particular, the first of which is Artificial Intelligence, or AI [20.20]. “There are many, many different types of AI, and of course the type of AI that ChatGPT is, is very different from the one that I’ve been using and developing.” Explains Martina. “But one of the main applications of AI or machine learning in shark research is through photo identification.” Many organisations around the world have full databases of photographs to identify individual animals, but largely rely on volunteers to sift through them. This work is time and resource heavy, and many times these valuable, large datasets are left unattended. But, researchers have started to develop AI models that are capable of identifying species, or individuals, extremely quickly and reliably. “It’s a hugely impactful support for researchers who have these massive databases. It’s such a waste of data not to have anybody that can go through them, and if AI can do it for you, that’s so much easier.” Says Martina. AI can also allow us to obtain fine-scale and long-term data, as many of these datasets span multiple years, sometimes decades. This even enables researchers to carry out growth studies, and observe an individual over time – incredibly valuable information for species that don’t cope well with other methods, like tagging. Photographs are less invasive, and put less stress on the animal.
For Martina’s masters research, she worked on developing an AI model to recognise epaulette sharks (sometimes known as “walking sharks”) [24.12]. This has been trialled before on other species – including whale sharks and manta rays. The key is that these species have patterns on their bodies (for example spots or stripes) that are present throughout their lifetime and that are unique to an individual, like a fingerprint. AI can learn these individual patterns to identify specific sharks. In the case of epaulettes, it’s a special combination of dots and patches that can identify them, which Martina used for her own model. In species that don’t have funky patterns, the trailing edge of the dorsal fin can sometimes be used – which can have a distinctive shape, or notches and nicks, that can identify individuals. The important thing with this is that the shape stays consistent throughout their lifetime, and the only species that is known to have this to date is the white shark. Although AI isn’t perfect, it can sift through thousands of photos much more efficiently and reliably than a human can, and provide large quantities of valuable data – a much better application for machine learning than creating strange videos and images on TikTok and Instagram!
Another key part of Martina’s research has been looking at the capabilities of Remotely Operated Vehicles, or ROVs [29.20]. “ROVs are like aerial drones, but they go underwater…imagine an aerial drone, make it much bigger and heavier, and put it underwater. That’s an ROV.” Says Martina. They are bigger because they require a larger battery, which allows them to survey for hours. Few people have written about using ROVs to study sharks, or how to use them. Martina’s focus was to try and find out if ROVs could be used as a tool to study sharks in more challenging spatial and temporal conditions, such as at depth or at night, and compare their performance against more traditional methods, like diver surveys, BRUVs, and drumline catches [34.48]. Because there is so little information about how to use ROVs in shark science, Martina basically had to start from scratch: “it was a lot of problem-solving, troubleshooting and trial and errors!”. This required a lot of creativity and thinking outside the box, such as using circles cut out from coloured carrier bags to replace red light filters that kept falling off the ROV’s lights. “You try whatever you can!” Says Martina. “The things that we did to try and get data. Because when you have new tech, that’s what you have to do – to try.”
But all the hard work and sleepless nights paid off [37.45]. “I looked at the data, and actually, the little thing surprised me. It was really good!”. Firstly, they came across more species and individuals than if there had just been divers in the water – even at night. Secondly, when Martina looked at data from collaborators working with the same ROV model and BRUVs in the coral sea, things got even more interesting. “BRUVs are notoriously good tools for sharks. It’s one of the golden hammers of shark research – BRUVs work. They attract the sharks, and you can see what’s out there. So I was sure that the ROV wasn’t going to be as good.” Martina says. “But then, in some of the reefs, when I compared the two methods by standardising by effort – which in our case was soak time of the method – the ROV actually outperformed the BRUVs.” Martina hypothesises that this is because while BRUVs are stationary, ROVs drive through the environment, and may pick up group-like, swimming species who are usually found at shallower depths, and don’t come all the way down to the seabed to visit the BRUV. “So you really get a lot of benefit from having an actual mobile type of survey method, which technically would be divers, but as divers can’t be at 80 metres…that’s where the ROV comes in.” She explains. Similar results came from comparing the ROVs to drumlines, another tried and tested shark research method often used to capture larger species. “The ROV outperformed the drumlines almost by ten times. We saw so many more sharks and rays with the ROV, and quite an array of species in terms of rays and reef sharks which you wouldn’t normally catch with a drumline.”
So, what does all this mean for shark science and conservation as a whole [42.00]? As these technologies become more advanced, and more accessible, they can really help fill the gaps left by other methods, and be used in conjunction with other methodologies to provide a more comprehensive picture. For example, AI can help to analyse large datasets for long-term population studies, especially for ‘no-catch’ species where photo ID is sometimes the only data we have. And as our use and understanding of AI improves, there is potential in the future for this work to be more instantaneous – AI may even be able to identify species on the spot. “This technology is out there, and it’s happening.” Says Martina. “I would love to see the research world investing in and bringing in these new technologies – even if it can be a bit scary at times.” ROVs, too, are an exciting new frontier, even if at the moment “they aren’t perfect.” “I am expecting the technology to get better and better,” Martina says, “it would be really great to see other research groups adding this in to their other traditional, normally used methods, because I think it definitely has a space there. And especially when it comes to doing research at night time and at depth, which are one, a temporal, and the other one, a spatial limitation of current methods.”
If you want to find out more about Martina’s projects, you can follow her on social media (@martipermare), or by heading to these links:
https://aims.jcu.edu.au/our-people/students/martina-lonati.737/
https://www.fishandfisheries.com/
https://saveourseas.com/project-leader/martina-lonati/
You can also read her latest publication, ‘Novel use of deep neural networks on photographic identification of epaulette sharks across life stages’ here: https://onlinelibrary.wiley.com/doi/pdf/10.1111/jfb.15887
ABOUT OUR GUEST
MARTINA LONATI
Martina is a research assistant for Biopixel Ocean Foundation and a research associate at the Fish and Fisheries Lab, focusing on innovative technologies to enhance methods in shark research. Martina is collaborating with software engineers and leading AI organizations to co-design a multi-species deep-learning model for photographic identification of shark and ray species. Alongside her work in AI, Martina has been experimenting with and fine-tuning remotely operated vehicles (ROVs) and comparing their performance with traditional methods like diver surveys (UVS), baited underwater video stations (BRUVS), and drumlines.
Martina’s happy place is in the field, searching, observing, and following sharks and rays. Over the past eight years, she has gathered unforgettable memories and invaluable skills through her fieldwork experiences. She has deployed hundreds of BRUVs in marine protected areas in Fiji, spent nights longline fishing for baby sharks in Fijian rivers, photo-IDed great white sharks in South Africa, completed thousands of hours of dive surveys on the Great Barrier Reef, tagged sharks along Australia’s coast, and established a research breeding colony for epaulette sharks, performing all sorts of physiological assessments. More recently, she has become an ROV driver, using this technology to search for sharks and rays at night and in deep waters.
While Martina is fascinated by the creativity and problem-solving nature of new tech applications, her interests extend across fisheries science, species conservation, ocean governance, and the social science dimensions of shark conservation.
Instagram: @martipermare
X/twitter: @Martina0Lonati
