Abstract
This study presents a novel multi-modal methodology for detecting oil spills in the Caspian Sea and combines remote sensing, deep learning and natural language processing (NLP) of media content. We developed an accurate and comprehensive oil spill database covering incidents from 2002 to 2023 by integrating satellite synthetic aperture radar imagery with deep learning segmentation models. A key innovation of our approach is cross-referencing satellite-detected spills with media reports, enhancing detection accuracy while revealing significant underreporting of spills in media outlets. Our approach demonstrates the potential of merging technological innovations with media analytics to improve environmental monitoring effectiveness and inform policy-making for sustainable marine ecosystems.