It’s universally agreed that oil spills and marine ecosystems do not mix. An oil spill can have devastating environmental consequences, take a long time to clean up, and is economically costly.
The Great Barrier Reef is a region recognised by the Australian Maritime Safety Authority (AMSA) as a high-risk area for oil spills, partly due to increasing ship traffic in the marine park. Ship collisions are not frequent, but there is always the potential for unreported illegal oil discharges from passing ships. In addition to the Great Barrier Reef, the North West Shelf off Western Australia is also identified as a high-risk region due to operating oil and gas platforms.
Satellites provide a bird’s eye view of the coastal ocean
Imagery from the Sentinel satellites is providing scientists, like David Blondeau-Patissier and Thomas Schroeder, from our Coastal Ocean Colour and Radar Sensing team, with the much-needed intel to better detect oil spills.
“Never before in Australia have Earth Observation techniques routinely been used to detect oil spills. However, over the last three years, we have developed an approach, in collaboration with the Department of Environment and Science (DES) Queensland and AMSA, to process all satellite imagery acquired over Australian waters by satellites from the European Space Agency Sentinel missions and analyse them for oil spills,” explains David.
“Thankfully, none of the detected events were found in the Great Barrier Reef, but some were found in the North West coast of Australia.”
For this research, the team mostly uses Sentinel-1 SAR, a synthetic aperture radar. Images are pre-processed to mask out the land and smooth out the sea surface to provide better visualisation of possible presence of oil.
“An oil slick will typically appear black against a clear grey background in a Sentinel-1 SAR image, which is the oil-free seawater. But many features can look like oil spills, such as wind shadows. Several key descriptors, such as the width, length and the shape itself, are signature elements we look for to discriminate what is an oil slick, and what isn’t,” says Thomas.
Earth observation snapshots
In Australia, Sentinel-1 SAR images are generally acquired every 12 days over the same location. Around 80 Sentinel-1 SAR images are acquired over the Great Barrier Reef every month, with the region extending up to Cape York.
Processing satellite imagery for the Great Barrier Reef is a computational intensive task, as the marine park spans 348,000 square kilometres and is 2,300 kilometres in length. In comparison, the portion of the North West Shelf where most of the oil and gas platforms are located has an estimated coastline of 800 kilometres.
“The Great Barrier Reef is a complex environment, and oil detection can be a challenge. The width of a Sentinel-1-SAR image covers approximately 250 kilometres, so to cover the whole of the Great Barrier Reef region, approximately 10 Sentinel-1 SAR images are needed to be processed each time, resulting in excess of 3,000 images over the last three years,” explains David.
“The Great Barrier Reef also has reef structures that add complexity to the images and may make it more difficult to detect an oil spill in comparison to other marine regions. However, our processing can mask these features to improve accuracy and avoid misclassification.”
Processing satellite imagery
Once an image is acquired by the satellite, it is transferred from the satellite’s ground segment to CSIRO’s archive. It can take 12 hours from when the image is taken to when the team can process it and assess if an oil spill is present in the satellite scene.
“Using advances in high performance computing and machine learning, we are aiming to identify hotspots of illegal discharges along a coastline, or oil platforms that regularly discharge oil more than others. This detection method can be vital for oil spill management response teams as it provides a map of the spill, independently of any weather or light conditions,” explains Thomas.
An automated workflow of Sentinel-1 SAR imagery for oil spills analysis is in place for the region of the Great Barrier Reef, processing any new image acquired over the region. The next step is to fine-tune a machine learning detection algorithm to develop a fully automated alert system.
Real-world impacts through collaboration
For this research, CSIRO consulted with project partners DES and AMSA but also the Great Barrier Reef Marine Park Authority (GBRMPA), and the National Offshore Petroleum Safety and Environmental Management Authority (NOPSEMA) to ensure the tool can provide real value for the real world.
“This research gives AMSA and other environmental agencies information that we haven’t had access to before. It could help us better assess regulatory compliance options and improve our ability to assess the extent of any oil spill. We will be able to respond in a more informed and cost-effective manner. It is all about reducing potential threats to the marine environment,” says Paul Irving, AMSA Senior Advisor, Science and Technical Response.
“This would be particularly helpful when conditions mean aircraft and ships cannot usefully be deployed to obtain the required intelligence imagery.”
New generation research using NovaSAR
The developed spill detection methods are in the process to be adapted to the NovaSAR-1 satellite, which provides an Automatic Identification System (AIS), combined with NovaSAR-1’s maritime surveillance mode for ship identification. This can help to identify specific ships linked to possible illegal discharges.
In addition to the detection of oil slicks, the team are also exploring the use of SAR data for the detection and monitoring of river runoff during the wet season, and how it can be integrated with other CSIRO modelling platforms like eReefs to manage water quality.
The oil spill detection method can be redeployed under less than an hour for any region of the world. This novel approach is not only beneficial for Australia, it can also be used for South-East Asia or the Pacific Islands where CSIRO is actively undertaking environmental research.
This research received funding from Advance Queensland. The research report is available here.
We acknowledge the European Space Agency for the use of the satellite imagery for this research.