Satellite-Based Detection and Assessment of Marine Oil Spills Using Synthetic Aperture Radar (SAR) Imaging

Marine oil spills pose a significant threat to the marine ecosystem and coastal environments. This in turn creates a need for immediate and reliable detection and assessment techniques. This paper presents a remote sensing framework for marine oil spill monitoring using Sentinel-1 Synthetic Aperture Radar (SAR) imagery, specifically Level-1 Ground Range Detected (GRD) dual-polarization (VV/VH) acquisitions, integrated with sea surface temperature (SST), atmospheric wind fields, and Automatic Identification System (AIS) vessel information. SAR imagery is used as the primary observation source for detecting oil spill signatures, while auxiliary environmental datasets are incorporated to improve discrimination between actual spills and naturally occurring look-alike phenomena. A convolutional neural network (CNN-OSD) classifies 256×256 SAR patches, using VV-polarized backscatter as the primary detection channel, owing to its sensitivity to oil-induced dampening of sea-surface capillary waves, while VH polarization is retained for contextual feature analysis, with outputs modulated by an atmospheric wind-regime filter, a sea surface temperature (SST) thermal-anomaly engine. The proposed framework was evaluated using observations from four study regions in the Bay of Bengal, Eastern Mediterranean, and Gulf of Aden. Results demonstrate the capability of the system to identify oil spill features across diverse oceanographic conditions while reducing false detections associated with low-wind areas and other dark formations. Validation experiments achieved an overall detection accuracy of approximately 85% under 5-fold cross-validation. The framework generates geospatial oil spill maps, calibrated SAR backscatter products, SST trend visualizations, and event-based assessment reports, demonstrating its potential for operational marine environmental monitoring and rapid response applications.