How Satellites And Machine Learning Are Being Used To Detect Plastic In The Ocean

While we know plastic is terrible for marine life, detecting plastic pollution in the ocean is notoriously challenging. Plastics come in many colors, break down to microscopic sizes, and are made of a variety of chemicals. Adding to the problem is the vast size of the ocean, to which millions of tons of plastic are added each year.

It is essential to identify which parts of the ocean collect the most plastic to effectively target cleanup and pollution prevention efforts. Might satellites bolstered with machine learning be up for the oceanic task of tracking plastic pollution? According to research recently published in Scientific Reports, yes.

A team of scientists at the Plymouth Marine Laboratory in the United Kingdom tested whether data from two satellites operated by the European Space Agency could be analyzed using a machine-learning algorithm trained to detect plastic. The two Sentinel-2 satellites used in this research are each equipped with 12-band Multi-Spectral instrument (MSI) sensors that allow for 10-meter resolution in the data they collect. With the efforts of the two satellites combined, data is repeatedly collected from all coastal locations around the world every 2 to 5 days. In other words, every part of the world where land meets the sea is re-imaged between 6 and 15 times every month - that’s a lot of data!

Satellites collect data on light signals, among other things. Materials can be distinguished using light signals based on which wavelengths of light they reflect. While clear water efficiently absorbs light in the near-infrared (NIR) to shortwave infrared (SWIR) light range, floating materials like plastic and natural debris reflect NIR instead. These differences in light absorption allow satellites to detect floating objects from space.

The NIR signals of various floating objects vary. Using the satellite data, researchers trained a machine-learning algorithm to identify the light signal of floating plastic by releasing a plastic float off the coast of Greece and obtaining the associated light signal data from the satellites. The researched used this light data to teach the algorithm to associate certain NIR light signals with floating plastic debris. Similarly, they also taught the algorithm to distinguish between plastic and natural materials such as seaweed, driftwood, and seafoam.

Once the algorithm was up-and-running, the researchers put it to the test against satellite data from coastal waters in four places around the world: Accra (Ghana), the Gulf Islands (Canada), Da Nanng (Vietnam) and Scotland (United Kingdom). Overall, the algorithm detected plastic with 86% accuracy. Better yet, the algorithm was 100% accurate in its analysis of the data from the Gulf Islands. Not too bad for data collected from thousands of miles above!

Importantly, this algorithm is only equipped to locate large floating plastic pieces. However, it is from these floating ‘macroplastics’ that many harmful microplastics form. These results show that satellite data combined with machine-learning algorithms could aid in the tracking, and subsequent clean-up, of plastic pollution around the world.