How Spyfish Aotearoa Is Automating Marine Protected Area Monitoring

Marine Protected Areas (MPAs) maintain the health of our oceans — but how do we measure their effectiveness?

MPAs play an important role in maintaining our marine biodiversity and ecosystem services for commercial, customary use and recreational use. The establishment of a representative network of MPAs is recognised internationally to be a critical management tool that supports and contributes to the restoration and sustainable health of the oceans. MPAs provide opportunities for understanding of marine biodiversity and the effects of human activities on this; this role is integral to the contribution MPAs can make to the broader sustainable management of ocean ecosystems.

New Zealand’s marine waters are unusually diverse. More than half of the country’s 17,000 known marine species are found nowhere else. A 6˚C water temperature difference across the country leads to very diverse coastal conditions, such as the subtropical Northland and the subantarctic Auckland Islands in the far south.

New Zealand has 44 marine reserves covering about 17,400 km², roughly 10% of its territorial sea (out to 12 nautical miles). Most of that area lies in two large offshore reserves (Kermadecs and four Auckland Islands), leaving less than 1% of mainland coastal waters under full protection. In October 2025 the Hauraki Gulf / Tikapa Moana Marine Protection Bill passed, creating 19 new marine protection areas, including extensions to Cape Rodney to Okakari Point (Leigh) Marine Reserve and Te Whanganui-o-Hei / Cathedral Cove Marine Reserve, 12 High Protection Areas and five Seafloor Protection Areas, increasing the protection from 0.3% to 5.9%.  Additionally, 6 new marine protections are proposed for Otago.

Given the increases in the number of MPA’s, managers need more efficient tools to track their impacts. One approach marine rangers use is Baited Underwater Video (BUV). Each setup consists of a camera mounted above a bait box and scale bar, placed at predetermined locations in protected areas and at comparable control sites.

Once the camera reaches the seabed, fish are attracted to the food, and the real action begins: a feeding frenzy like the one below:

This clip comes from a deployment inside the Kāpiti Marine Reserve. As you might expect, videos are not always this lively. Both types of footage matter because together they show how fish communities are doing across space and time, capturing a snapshot of local marine life. Over a full survey season, rangers collect hundreds of these clips from different marine reserves, each video contributing data points in the long-term record of fish abundance and diversity across Aotearoa’s coastal waters.

Each BUV captures 30 minutes of activity on the seafloor, but the real challenge begins once the footage is back on land. To understand what’s happening beneath the surface, DOC rangers and scientists review each video and record the maximum number of individuals of each species seen within the 30 minutes. This method helps avoid double-counting fish that swim in and out of view, giving a consistent estimate of relative abundance.

It’s a time-consuming process that requires both species expertise and patience. Each survey involves multiple deployments, and across many MPAs, quickly adding up to hundreds of videos every year. In the 2024-25 season, 19 surveys were completed, collecting over 500 videos across 20 MPA’s.

Reviewing thousands of hours of underwater footage takes time, time that DOC rangers could otherwise spend in the field. Spyfish Aotearoa was created to help with that challenge. The project combines the efforts of rangers, scientists, volunteers, and artificial intelligence (AI) to make video analysis faster and more efficient.

This year, we focused on the groundwork: cleaning up data, building tools and puzzle pieces necessary for the fully automated analysis system.

Bringing the data together

Before anything could be automated, the existing information had to be organised. We gathered more than 1,500 historical BUV deployments from scattered spreadsheets and drives around the country into one standardised database.

Supporting rangers in the field

Together with DOC rangers, we co-designed simple mobile and desktop tools for recording new survey details. From the moment a camera goes in the water, data is captured consistently and synced automatically, reducing paperwork and making later analysis smoother.

Training the first AI model

Machine learning is a great asset in helping scientists around the world with conservation work and population monitoring, such as analysis of video, images or audio collected in the field. We trained an initial machine learning model to recognise two key species: snapper and blue cod. This proof-of-concept will be one of the core components of our full analysis pipeline.

Community contribution

We could not have done this without you!  Clips from the videos are uploaded onto Zooniverse, and similar to the experts, citizen scientists are asked to count the maximum occurrence of a certain species within a video. So far, we have more than 240,000 classifications on Zooniverse. These classifications speed up processing and provide the examples needed to improve our AI models over time.

By bringing together machine learning, citizen scientists, and marine experts, we’re building a monitoring system we can trust. Each group plays a different role in the same process.

Automation and AI handle most of the data processing and video analysis, while low-confidence cases, where the system is unsure, are flagged for review by citizen scientists. When several volunteers analyse the same clip and their counts differ, it signals that something might be confusing or complex. That’s when the experts step in. Instead of reviewing every video, they can now focus on the cases too complex for both AI and our citizen scientist volunteers. Over time, as more data and feedback flow into the system, we will retrain the AI models, making the entire process faster, more consistent, and easier to maintain.

Our focus for the coming year is to connect all these pieces and move from prototypes to a complete, reliable workflow.

Over the next year, we’ll:

• Launch the full pipeline.
• Complete our expert toolkit, giving specialists an efficient way to label and review footage.
• Finish migrating the remaining historical data into the new system.

By streamlining how underwater video is collected and analysed, we not only get to the insights faster but also free up rangers’ time to focus on impactful fieldwork that computers simply can’t do.

The data and reports produced through Spyfish Aotearoa give DOC clearer, faster evidence to guide marine protection and track the health of our reserves. Each improvement, whether it’s an organised dataset, a new AI model, or one more volunteer classification, adds to a growing body of knowledge about Aotearoa’s marine ecosystems.

Step by step, these efforts are turning video into evidence, and evidence into action, helping DOC understand the health of our oceans and showing how small improvements add up to real progress.