Examples from Machine learning algorithm to detect plastic accumulations on beaches and at sea developed by Pixalytics
Last week, the European Space Agency (ESA) issued an article about the MARLISAT project tracking marine plastic litter using a combination of in-situ and satellite data together with a computer model. We’re delighted as this is a project we’ve been doing the Earth Observation element for the last few years .
MARLISAT Project
MARLISAT aims to track how marine plastic litter move around the world’s ocean and other water bodies of the world, and in identifying how this occurs it hopes to help countries reduce marine plastic litter. The project is led by the French organisation CLS, Collecte Localization Satellites, and was funded as part of ESA’s Open Space Innovation Platform (OSIP) portfolio of ideas on marine plastic litter. Pixalytics joined CLS in the successful bid and we focus on the Earth Observation part of the project.
CLS developed a set of wooden buoys that could be deployed in the ocean, fitted with transmitters so their journey through the water could be tracked. The buoys were deployed in Indonesia as they have a target within their national marine pollution plan to reduce plastic waste by 70% by the end of 2025, coupled with the fact that the currents around the area are complex.
The buoys, which had a battery life of 100-days for their transmitters, were released in May and were tracked in real time. The data gained from tracking these buoys is being used as an input into an existing CLS ocean drift computer model, known as MOBIDRIFT, to enable marine plastic litter routes to be tracked and areas of litter accumulation highlighted.
Pixalytics’ Earth Observation Role
Our part of the project has been to develop a machine-learning based classifier using Sentinel-1 and Sentinel-2 data to identify plastic accumulation along beaches and ocean hotspots. The project idea was originally developed by one of our former employees, Robert Page, during his time at Pixalytics, and the project work was undertaken by Pixalytics’ staff Dr Samantha Lavender, Dr Ashley Smith and Sarah Lappin.
To support the training and validation of the algorithm, a dataset was created with various examples of both terrestrial and aquatic plastics in various land covers. It got to the point where we could barely listen to the news, go on a journey, or simply watch a TV programme without noting potential training sites! The trained classifier, including an Artificial Neural Network and post-processing decision tree, was verified using five locations encompassing different forms of plastic. Although exact matchups are challenging to digitize, the performance has generated high accuracy statistics, and the resulting land cover classifications have been used to map the occurrence of plastic waste in both aquatic and terrestrial environments. We’ve also written our work up into a paper, currently in preprint, as it goes through peer review.
The really interesting aspect of the project is that has combined satellite data, in situ data from the buoys and numerical modelling. It has also been great being involved in a project focused on delivering a tangible improvement and benefit to the world’s ocean.
