- A new study shows that more than 75% of industrial fishing activity and almost 30% of transport and energy activity in the oceans has not been tracked by public systems, revealing a significant gap in global observational data.
- The study, led by Global Fishing Watch, used AI to analyze 2 petabytes of satellite imagery collected between 2017 and 2021, providing unprecedented insights into hidden fishing hotspots and offshore energy infrastructure development.
- The research highlighted the potential of combining AI with Earth observation data to gain a more comprehensive understanding of ocean activities, which is needed to manage and improve the sustainability of the $2.5 trillion blue economy.
- The open-source code developed during the study can help inform policy for safeguarding ocean ecosystems, enforcing laws and identifying renewable energy expansion sites, the study authors said.
We’ve been largely flying blind when it comes to fishing, shipping and energy production intel on our oceans, according to a new study that combines artificial intelligence with satellite technology. Researchers found that more than three-quarters of industrial fishing activity and almost one-third of transport and energy activity is missing from public tracking systems, and therefore largely invisible globally.
The study, from Global Fishing Watch (GFW), a vessel-tracking platform created by Google and the NGOs Oceana and SkyTruth, was published in the journal Nature on Jan. 3. Analysts used deep learning models and 2 petabytes of satellite imagery to map industrial activities spanning more than 15% of the ocean between 2017 and 2021 and compared the results with activity indicated by publicly available tracking data.
“We can’t manage what we can’t see,” co-lead author Fernando Paolo, senior machine learning engineer at GFW, told Mongabay. “In this data void, it is all too easy to do harm to the environment, mismanage marine resources or disregard the law — and get away with it.”
Global observational data are essential to managing the emerging $2.5 trillion blue economy, which encompasses fisheries, offshore oil and renewable energy, aquaculture and mining activities, the research suggests. The study’s deep learning models enable increased mapping accuracy that can show what’s really going on in our oceans. This can help decision-makers safeguard ecosystems and communities, enforce fishing and labor laws and pinpoint suitable renewable energy expansion sites, the study authors said. They also said they hope it will help call out transgressions in marine protected areas and assist with efforts to protect 30% of the global oceans by 2030, as agreed by 190 U.N. member states in 2022.
“This project highlights the larger potential of combining modern machine learning methods with Earth observation data,” Konstantin Klemmer, a geospatial phenomena and machine learning specialist with Microsoft Research in Massachusetts, U.S., who was not involved in the study, told Mongabay. The ambition and scope are “outstanding,” Klemmer said, adding that it’s the most comprehensive global-scale mapping of its kind and “one of the first to combine these two incredible tools.”
What can we see now that we couldn’t before?
Until now, public mapping of oceangoing vessels has relied mainly on broadcasts of vessel coordinates from Automatic Identification Systems (AISs), which transmit a ship’s location, course, identity and speed. But those data are incomplete because regulations on which vessels must use AIS vary from country to country, and according to vessel activity and size. Crucially, vessels can switch off their AIS. Such mapping also often relies on satellite reception, which has blind spots in some coastal areas, and data can be restricted by governments. Information on fixed offshore infrastructure, like oil rigs and wind turbines, is not collated in one place and is often restricted for commercial or bureaucratic reasons, according to the study.
So, until now, we’ve been a few ingredients short of a recipe for a comprehensive handle on ocean activity.
Paolo’s team designed three deep convolutional neural networks (DCNNs). These models learn directly from data and were taught to identify ships and infrastructure with very high accuracy. They used specially adapted algorithms and deep learning on radar and optical imagery from European Space Agency satellites Sentinel-1 and Sentinel-2.
“These deep learning models are especially well-suited for big data projects,” Klemmer said. Global-scale mapping efforts like this one require processing a huge amount of data: The data in this study could cover around a trillion pages of standard typed text if one letter represented 1 byte. DCNNs are also used in facial recognition, self-driving cars and medical imaging analysis.
GFW has been running an AIS-based fishing vessel mapping platform since 2012. “We had an idea that our maps of industrial activity at sea, particularly industrial fishing, were missing a great deal of activity,” Paolo said. “What we didn’t know is how much and where.”
To find out, they used the specially designed DCNNs to detect and classify vessels in satellite imagery, mapped their locations and movement, then cross-referenced them with AIS data for 53 billion vessel GPS positions to work out which vessels were missing from public tracking.
They were right to question. The intricate tracery of our hitherto hidden oceanic perambulations is both strangely beautiful and revealing.
Large areas of South Asia, Southeast Asia and the north and west coasts of Africa, where previous public tracking systems had shown “little to no activity,” were actually a-bustle with aquatic traffic, according to the new paper.
It also overturned AIS-only indications that the levels of fishing activity in Europe and Asia were comparable. Instead, it found around 70% of all fishing vessel detections in Asia — they had just been going undetected by AIS.
Similarly, AIS data alone indicated the number of fishing hours occurring in European Mediterranean countries’ exclusive economic zones was 10 times higher than the number in African Mediterranean countries’ waters. But the new analysis showed they were almost even.
“Having an incomplete view of the fishing that is actually taking place is like trying to complete a puzzle while missing its pieces,” John Weber, marine spatial planning director at the Waitt Institute, a California, U.S.-based sustainable ocean planning consultancy, told Mongabay. The new data and analysis from GWF can be used to “strengthen the management plans that each country is able to create so that they can better integrate ocean health and economic activity,” he said.
Another revelation from the study: Fishing activity dropped by 12% globally during the COVID-19 pandemic and had not returned to previous levels by late 2023, surprising Paolo. “Either the pandemic had a long-lasting effect and is still ongoing, or this reflects a long-term decline in the activity of fishing vessels,” he said.
The study also found more than 20 untracked fishing vessels per week inside the Great Barrier Reef Marine Park and five inside the Galapagos Marine Reserve, two of the most biologically important and well-monitored marine protected areas on the planet. “We have likely found some fishing within marine protected areas or restricted areas that will require further investigation and protection” by authorities, Paolo said.
Policymakers need such information to “adequately manage, protect and restore their ocean ecosystem,” Weber said.
Asia also had the largest concentration of energy-related and transport vessels, 65% of the global total, including the majority of those untracked by AIS. Many vessels in Asian waters may have been untrackable even if they broadcast their positions because they frequented areas with poor satellite AIS reception, the study noted.
As for the fixed offshore infrastructure, the researchers were stunned by the speed with which they saw offshore wind energy installations developing, as China took the lead over Europe via a 900% increase in turbines between 2017 and 2021. There are now more wind turbines than oil rigs in the ocean, the study showed. However, offshore oil structures also increased, by about 16% in five years.
Code developed during the study is open source and mapping is public and updated daily on the GFW platform, meaning everyone can access it and anyone can use it to develop additional tools to bring industrial activity in our oceans out of the shadows and into public view. GFW aims to expand their mapping into a new open data platform covering all human activities at sea, using a $60 million investment from TED’s The Audacious Project.
“Satellite technology and artificial intelligence have the potential to democratize access to data on human use of the ocean,” Paolo said.
Banner image: A school of snappers under a pier in French Polynesia. Image by Hannes Klostermann / Ocean Image Bank.
Paolo, F., Kroodsma, D., Raynor, J., Hochberg, T., Davis, P., Cleary, J., … Halpin, P. (2024). Satellite mapping reveals extensive industrial activity at sea. Nature, 625(7993), 85-91. doi:10.1038/s41586-023-06825-8
FEEDBACK: Use this form to send a message to the editor of this post. If you want to post a public comment, you can do that at the bottom of the page.