Marine pollution, including Marine Debris and Oil Spills, poses a serious environmental threat that requires systematic monitoring. Satellite observations from both passive and active sensors, combined with established machine learning techniques, have been widely used for mapping marine pollution. However, the application of cutting-edge deep learning approaches specifically tailored to this task remains limited. In this study, we use the MADOS Sentinel-2 (S2) marine pollution dataset to construct a new Sentinel-1 (S1) Synthetic Aperture Radar (SAR) dataset containing annotations for oil spills, sea surface, look-alikes (e.g., low-wind areas and internal waves), ships, and offshore oil platforms. We then train deep learning models on this Sentinel-1 dataset, including well-established architectures such as U-NET, specialized frameworks for marine pollution segmentation such as MARINEXT, and state-of-the-art approaches like SEGNEXT, and we evaluate their performance both quantitatively and qualitatively. Our findings show that MARINEXT achieves the highest F1-macro score at 92.7%, outperforming U-NET at 70.6% and SEGNEXT at 75.9%. Qualitative evaluation using the corresponding multispectral Sentinel-2 imagery further supports these results. Finally, our analysis shows that mapping Marine Debris in SAR imagery remains particularly challenging, especially in the absence of corresponding optical observations.