Advanced AI and satellite technology are transforming the way we detect and track ocean plastic islands. A new project uses smart algorithms and integrated satellite networks to pinpoint debris, improving cleanup efforts.
Ocean plastic islands: satellite data and AI to track them at sea
Although ocean plastic islands represent a worrying and growing phenomenon of marine pollution, there is still no technology for the definitive detection of small, dispersed debris accumulations. One of the difficulties lies in tracing the debris trails that form along the convergence lines of ocean currents. Two years ago, precisely to address this problem, an ambitious project called AI for Detecting Ocean Plastic Pollution with Tracking, or more simply ADOPT, was launched by the renowned Swiss research institute EPFL.
Satellite imagery and AI
The objective was to develop two types of systems: “To identify waste islands by analyzing satellite images, and the other to predict where the islands will have moved when cleanup teams can reach them, usually within 24 hours,” says Emanuele Dalsasso, one of the scientists. In fact, even when plastic islands are identified, the time that passes from reporting to intervention leads to further displacement in the water.
Today, thanks to the ADOPT project, artificial intelligence offers a concrete solution: an algorithm is capable of scanning thousands of satellite images to locate floating plastic in real-time, distinguishing it from natural elements that can cause confusion, with unprecedented accuracy.
AI recognizes spectral signature
The problem with marine debris is its fragmentary nature. It often consists of partially submerged objects that blend in with sea foam, sun reflections, or marine vegetation. The model developed by the researchers is not limited to simple visual analysis. The key to success lies in a data-centric approach.
The system was trained on thousands of examples provided by oceanographers, learning to recognize the specific plastic’s spectral signature (how it reflects light) compared to natural materials. Thanks to a label refinement algorithm, AI can correct imprecise human annotations, distinguishing between human-made debris and sea foam.
An integrated satellite network
These instruments fly over a specific point in the ocean only once every six days, and their images have low resolution, only 10 meters per pixel, making it difficult to track debris. In fact, the innovation lies not only in the software but also in the combined use of data from space. The algorithm uses data from ESA’s Sentinel-2 satellite, which provides free images every 2-5 days with a resolution of 10 meters per pixel.
However, to fill temporal and spatial gaps, the team integrated the system with data from PlanetScope, a constellation of hundreds of nano-satellites that scan the Earth daily with higher resolution. This combination allows for continuous monitoring: while Sentinel-2 identifies weekly trends, PlanetScope’s daily data enables tracking the movement of plastic patches almost in real-time.
Challenges: inclement weather
There is, however, a significant problem: the system does not work well in bad weather. Optical sensors cannot “see” through clouds. “One option could be to incorporate radar images from Sentinel-1,” says Dalsasso, one of the researchers, adding: “Radar signals can travel through clouds and work day and night. But they only provide information on the consistency and geometry of the debris, which means we would lose the key spectral signatures that are detected by optical sensors and are essential for detecting garbage islands.”
Ocean plastic islands: from mapping to prediction
The research, recently published in Nature Communications, goes beyond mere identification. The next step of the ADOPT project is drift prediction. “The objective is twofold,” the researchers explain. “To identify garbage patches and predict where they will move in the next 24 hours by leveraging wind and current models corrected by machine learning.” This would enable organizations like The Ocean Cleanup to send collection vessels to exact points, saving fuel and increasing recovery efficiency.
With a view to global collaboration, the researchers have decided to make the system Open Source, releasing the code and models on platforms like GitHub. The goal is to allow governments and NGOs worldwide to monitor their marine ecosystems, identifying pollution hotspots, especially near river mouths, where the risk of plastic discharge into the sea is highest.
