Can AI help detect marine oil spills

Can artificial intelligence (AI) help protect Alaska's coastal waters from marine oil spills? Mohamed Elsheref, Ph.D., a postdoctoral researcher in UAA's Department of Chemistry, is the lead author on a scientific paper published last month in the Journal of Environmental Chemical Engineering. In the article, titled "OilSpillNet: A Deep Learning Framework for Accurate Detection and Segmentation of Marine Oil Spills from Imagery," Elsheref and his co-authors presented a deep learning framework, OilSpillNet, which offers possibilities for automated detection of marine oil spills. Rapid and accurate spill detection is critical for protecting marine ecosystems, fisheries and coastal communities, where early response can significantly reduce environmental and economic damage.

The traditional method for monitoring marine oil spills involves humans manually reviewing images from synthetic aperture radar (SAR). Elsheref noted that this process can be "slow, labor-intensive and error-prone" as natural phenomena like algal blooms or low-wind areas can "mimic oil signatures" and create false positives. While trained specialists can discern the difference, this manual review "cannot scale to the vast data volumes generated by frequent satellite overpasses, nor do they enable the rapid alerts necessary for emergency response," Elsheref wrote.

"By integrating environmental science expertise with artificial intelligence and computer vision methods, OilSpillNet enables automated, large-scale monitoring of satellite imagery-supporting faster, data-driven decision-making for coastal management and spill response," said Elsheref.

After training OilSpillNet on a "curated [National Oceanic and Atmospheric Administration] dataset of 206 SAR images," Elsheref and his co-authors found that the framework was more accurate than other models currently available and was better able to distinguish between "oil and background under challenging conditions."

The project was developed in collaboration with researchers at the University of New Orleans (UNO) in a multidisciplinary team led by UNO's Professor Md Tamjidul Hoque, integrating expertise in computer science and environmental analytical chemistry. The model's code and a working software version of OilSpillNet are freely available online via GitHub.

In 2025, prior to coming to UAA, Elsheref completed a Ph.D. in environmental analytical chemistry and a master's degree in computer science in parallel at the University of New Orleans. Elsheref now works in UAA's Analytical Science for Environmental Toxicology (ASET) Lab with his advisor, Patrick Tomco, Ph.D., associate professor.

Elsheref and Tomco are currently working on a project funded through Fisheries and Oceans Canada, titled "Hydrocarbon Oxidation Products of the Unresolved Complex Mixture (UCM): Advancing New Approaches in Chemical Characterization and Biological Effects."