+44 (0)1752 874289 mhennen@argans.co.uk

EO Processor For Marine Debris Detection

 

Marine Litter (ML) is a major issue in our global environment, and it’s presence in the Mediterranean Sea is no different. It is estimated that the sea contains as many as 24.7×1010 plastic pieces (Eriksen, M., et al. 2014), with many of these coming from seasonal tourist sources and washing up on beaches throughout the year (Poeta, G., et al. 2016). In-situ observations have been made of marine plastic in the Mediterranean previously (Pham, C. K., et al., 2014), but to observe the issue on large spatial and temporal scales, satellite Earth Observation (EO) techniques must be utilised. ARGANS Ltd. is addressing this with the EO Tracking of Marine Debris in the Mediterranean Sea from Public Satellites project. The project aims to develop an EO processor capable of detecting ML in Sentinel-2 data captured over the Mediterranean Sea, and incorporate this into a litter beaching forecast for the region’s coastal areas. To learn more about marine litter click here.

 

Through analysis of 10m resolution satellite images under ideal weather conditions captured by the Sentinel-2 Multispectral Imager, the EO Processor determines if marine debris is present in the image, with a focus on detecting plastic spectral signatures. It does this by completing the following steps:

1. Pre-processing the raw Level 1 or Level 2 data, it’s main function shall be to improve the quality of the input data.

2. Retrieval of marine litter information from the image through spectral analysis and the application of optical indices.

 3. Generation of a report detailing the location of the marine litter in a user-friendly format.

 

 

As part of the EU Copernicus programme, the European Space Agency (ESA) launched the Sentinel-2 missions, yielding optical imagery of high spatial resolution (up to 10 m) and frequent revisit times (5 days or less).

Operating in a sun-synchronous, low earth orbit (786 km), the Sentinel-2 missions provide EO data in 290 km swaths, covering the entire Earth’s land surface, including coastal waters within 20 km of the land and the entire Mediterranean Basin. The Multispectral Imager (MSI) provides reflectance data in 13 spectral bands from Visible to short-wave infrared (SWIR).

Sentinel-2
Marine litter mixes with other marine debris to create large floating rafts.

 

The EO processor for ML detection applies a variety of corrections to Sentinel-2 images, they include cloud screening, land masking and sun glint correction to remove false positives. The presence of ML is then determined through the application of optical indices. Finally a detailed report is generated, including a false colour composite (FCC) image, that identifies potential ML targets within the Sentinel-2 data.

Case studies gathered from around the world have been utilised for this project, not just those identified in the Mediterranean Sea. This has been done to provide the greatest number of marine litter cases possible for utilisation during the development and training of the processor.

 

Research completed during this project and associated projects will enable identification of a unique ML spectral signature. Upon conclusion of the validation campaign, the Sentinel-2 EO Processor for ML identification will be able to classify multiple surface types in the marine environment, identifying the location and size of ML rafts.

The results from the processor can be input into a drift model to predict ML transport routes. The relatively high temporal resolution of Sentinel-2 will allow regular tracking of marine litter rafts, validating the predictions of the drift model, and aiding mitigation.

The life cycle of plastics © WWF-Aus / Stef Mercurio

EO Processor Results

This EO Processor example (Right), first shows a feature identified of an output image of a Sentinel-2 tile that was captured in the coastal waters off the Po River Delta, Italy. The feature includes pixels classified as vegetation or algae (green and yellow), and potentially marine debris (yellow and red). The black boxes are areas that were subsequently classified as an objects after meeting the required characteristics and information from those black boxes were extracted for further analysis. After analysis of the same area of the corresponding S2 RGB image, it was clear that there was a feature present in that area of the image. 

The feature shown in this example was found in a EO Processor output image depicting an area off of the coast of Crete, Greece. The output image shows a feature classified as vegetation, algae, or potentially marine debris (green and yellow pixels). Sections of this feature passed the object classification requirements, shown by the areas of the processor output image masked by the black squares. Upon viewing the same region in the corresponding RGB image, this feature was a coagulation of some sort of material at the edge of a fluvial sediment plume. 

The feature shown in this example was found in a EO Processor output image depicting an area south of Calabria, Italy. The output image shows a feature classified as vegetation, algae, or potentially marine debris (green and yellow pixels). Sections of this feature passed the object classification requirements, shown by the areas of the processor output image masked by the black squares. The Sentinel-2 RGB image shows that there was a floating feature on the sea surface at the time of image capture. 

This EO Processor output image shows the correct classification of the plastic targets released off of the coast of Tsamakia Beach, Lesvos Island as part of the validation campaign. The targets are identified by the green and yellow pixels. The targets were classed as objects and data was extracted for further analysis. The targets can be viewed in the Sentinel-2 image, each target visible in at least one pixel.

The feature shown in this example was found in a EO Processor output image depicting an area north of Sicily, Italy. The output image shows a feature classified as vegetation, algae, or potentially marine debris (green, yellow and red pixels). Sections of this feature passed the object classification requirements, shown by the areas of the processor output image masked by the black squares. The Sentinel-2 RGB image shows that there was a floating feature on the sea surface at the time of image capture. 

References

Eriksen, M., et al,  (2014). Plastic Pollution in the World’s Oceans:  More than 5 Trillion Plastic Pieces   Weighing over 250,000 Tons Afloat at Sea   PLoS ONE. 9: 1-15.

Pham, C. K., et al, (2014)  Marine litter distribution and density in European Seas, from Shelves to deep basins, PLoS ONE. 9.

Poeta, G., et al (2016). The cotton buds beach: Marine litter assessment along the Tyrrhenian coast of central  Italy following the marine strategy framework directive criteria   Marine Pollution Bulletin. 113  266-270.