Summary:
Tracking floating plastic in the ocean remains a major challenge for cleanup operations. A research team led by the Swiss Federal Institute of Technology Lausanne (EPFL) has developed a system that uses artificial intelligence and satellite imagery to detect large debris patches and predict where they will drift within the next 24 hours.
The project, called AI for Detecting Ocean Plastic Pollution with Tracking (ADOPT), combines two tools. The first analyzes satellite images to identify concentrations of floating debris such as plastic-rich windrows that can extend hundreds of meters across the ocean surface. The second predicts how these patches will move using machine learning to refine existing models of winds and ocean currents.
Researchers trained the detection system using images from the European Space Agency’s Sentinel-2 satellites together with higher-resolution daily imagery from PlanetScope nanosatellites. Drift predictions are calibrated using data from GPS-equipped ocean drifters deployed through the Global Drifter Program.
The technology has now reached the proof-of-concept stage and is ready for field testing. Even as the two-year ADOPT project nears completion, partners including the nonprofit The Ocean Cleanup plan to continue refining the approach.
— Press Release —
AI helps marine scientists track floating debris from space
Being able to identify and track floating masses of debris is critical to ocean clean-up efforts. Yet effective systems for doing that remain elusive, despite the wealth of satellite images and weather data currently available. This could soon change, however, thanks to technology being developed under a project called AI for Detecting Ocean Plastic Pollution with Tracking (ADOPT), which started two years ago.
ADOPT is being carried out by EPFL’s Environmental Computational Science and Earth Observation Laboratory (ECEO) and the Swiss Data Science Center (SDSC), a joint initiative from EPFL, ETH Zurich and PSI, in association with Wageningen University in the Netherlands. It aims to develop two types of systems: “One is to identify garbage patches by analyzing satellite images, and the other is to predict where the patches will have drifted by the time clean-up teams can reach them, usually within 24 hours,” says Emanuele Dalsasso, an ECEO scientist. The idea is to meet a simple need: governments and NGOs can’t respond immediately when debris is detected, as they need time to organize and deploy clean-up operations.
The ADOPT team initially worked with the open-access data collected by Sentinel-2 satellites, which are a series of optical imaging satellites launched by the European Space Agency (ESA). But these instruments pass over a given point in the ocean only once every six days and their images have a low resolution, only 10 meters per pixel, making it difficult to track debris. To make up for those two drawbacks, the team designed an AI system that can also be trained on data from PlanetScope, a constellation of hundreds of nanosatellites that collect images every day with a resolution of 3 to 5 meters per pixel.
The result is an AI-driven detector that draws data from both sources and is updated daily with higher-resolution images, with no need for data annotation.
Given the scale at which the system operates, it can detect only large collections of plastic and other debris and not individual bottles, for instance. It tracks garbage patches – including the long rows of debris known as windrows – which can stretch hundreds of meters and are made up of mainly anthropogenic waste, especially plastic.
Once the debris has been identified, the next step is to predict where it will drift by the time clean-up teams arrive. The second system, developed by Christian Donner at the SDSC, is designed to make such near-term predictions.
“I draw on widely used models for forecasting winds and currents and then apply machine learning to correct them since the models often have biases,” he says. “The machine learning program compiles data from different sources and calibrates these biases, in order to better predict the trajectory of floating debris.” Because little field data are available on garbage patches, he trained the program using data from GPS-equipped drifters as a proxy. These drifters were rolled out under the Global Drifter Program and have been used to collect measurements since the 1990s.
Yet there’s one major catch: the system doesn’t work well in bad weather. Optical sensors don’t function across clouds. “One option could be to incorporate radar images from Sentinel-1,” says Dalsasso. “Radar signals can travel through clouds, and they work day and night. But they provide information on only the texture and geometry of debris, meaning we’d lose the key spectral signatures that are picked up by optical sensors and essential for detecting garbage patches.”
For now, the ADOPT team isn’t exploring the combined radar-optics option. Perhaps that will be done in the future – but by other research groups, as the project will officially come to an end this fall when the two-year funding program terminates. The team will leave behind a solid proof of concept along with two publications currently being finalized and the code for both systems (debris detection and drift prediction). Going forward, Dutch NGO The Ocean Cleanup will continue to work on comparing the algorithms, and university scientists will keep working together take the research further.
“The effort will continue, maybe not by us directly but by our research partners,” says Donner.
Article Source:
Press Release/Material by Gregory Wicky | EPFL
Featured image credit: Naja Bertolt Jensen | Unsplash
