SWIPE: Detecting oil spills in ports

#Webstory, #STEREO, #Seas & Oceans, #Water, #Pollution

Published on 12 December 2024

We all know what an oil spill is. The images of huge oil slicks flowing into the sea and approaching precious coastal areas are etched in our memories. Earth observation has long been used with increasing success to improve detection. Oil spills also occur in ports. There, although smaller in size, they are much more frequent. The SWIPE project has tried to find the best techniques to detect them.

Sneaky pollution

The past decade has seen intensive research into detecting oil spills at sea using Earth observation. Different sensors, methods and spectral bands proved suitable for this purpose. However, the vast majority of scientific papers published on the subject have dealt with accidents, some of them highly publicised, on the open sea, either on oil extraction platforms or on tankers.

Oil spills in ports generally do not attract public attention. Although they are on a smaller scale they are still relatively common.

Besides their harmful effects on ecosystems, these oil spills disrupt port operations, resulting in financial losses. In addition, they expose the public to volatile petroleum compounds with negative effects on human health. It is therefore essential to have specific tools for early detection. Nevertheless, there is no in-depth study on this in the literature.

Tools adapted to the port environment

To respond quickly to an oil spill, port authorities need reliable methods to monitor the water surface for the presence of an oil spill.
Unfortunately, many of the available methods for detecting oil spills are not suitable for port environments. This is because:

the oil dumped in ports is usually treated oil and not crude oil;
commonly used SWIR sensors are less sensitive to thin oil layers, which are most common in ports;
the applicability of low-resolution (satellite) data is limited due to the multitude of objects and activities in ports;
laser-based active detection is risky in ports due to their weight and size.

The team behind the SWIPE project (SWIR and drones for early detection of oil spills in ports) set out to find safe lightweight sensors for mobile platforms such as drones and ships and methods adapted to the local port scale. The research focused on SWIR, RGB and UV sensors.

Oil spill at the port of Antwerp-Bruges observed by the RGB camera (left) and the UV camera (right). The green rectangle indicates a reference panel used for data quality purposes.

RGB or UV?

The team first developed a physical model to estimate the thickness and volume of oil slicks based on reflectance in the visible and infrared range. This model, suitable for layer thickness greater than 200 µm, gives the best results for hydraulic oil. In addition, a specific deep learning model was developed to infer the size of oil slicks. This gave good results for RGB video images of hydraulic oil, lubricating oil and fuels. However, for oil thicknesses below 500 µm, the model performed poorly. Yet, oil spills in port environments are often much thinner.

Therefore, a new approach was investigated: the use of ultraviolet (UV) cameras. UV imaging has been shown to provide a relevant contrast between oil and water for crude oil spills, but it had not yet been tested for processed oil.

A UV camera sensitive to wavelengths below 400 nm was compared with an RGB camera for:

controlled experiments ;
acquisitions of real data from an active ship in the port of Antwerp-Bruges.

The controlled experiments showed that UV cameras distinguish better between oil and water than RGB cameras for different types of oil, viewing angles and atmospheric conditions. Moreover, UV images are much less affected by bottom influences and oil spots can be observed from any viewing angle, while RGB images are strongly affected by acquisition conditions.

After the controlled experiments, the camera was mounted on a ship in the port of Antwerp-Bruges and images were taken for three months (March to May 2024). These confirmed that the UV camera was more effective than the RGB camera in distinguishing between oil and water, and allowed an automated procedure to be established to adjust camera settings to environmental conditions in order to obtain optimal results.

Illustration of class separability. The gap between the water and oil modes is larger in the UV image (right) than in the 'Blue' image (left). A class of dark pixels is introduced to deal with low intensity values and only water and oil areas are considered.

A pioneering project with positive impact

SWIPE proved to be a pioneering project that demonstrated the ability of lightweight passive UV cameras to detect oil spills in port environments. The project's advances in methods to detect and monitor oil spills have positive impacts in a number of areas:

financial: by reducing the response time and therefore the financial impact of interrupting port operations;
environmental: by reducing the harmful impact of a leak on ecosystems;
social: by reducing the exposure of people living and working in the port to harmful petroleum compounds that have been shown to affect human health in the long term.