Floods present significant threats to ecosystems, human settlements, and infrastructure, making rapid and accurate flood mapping crucial for effective disaster response. This project employs satellite image processing techniques alongside deep learning models to detect flood extents and classify affected land cover types. Specifically, a U-Net convolutional neural network (CNN) model is trained on pre-flood and post-flood optical satellite images to segment areas into water bodies, human-made structures, and vegetation. The U-Net model, known for its contracting and expansive paths, enables precise localization of flood-affected regions by extracting multi-scale features from the optical imagery. Data augmentation techniques, such as rotation, scaling, and translation, along with transfer learning, are employed to improve model performance, especially in flood event analysis using Sentinel-2 imagery from the Caprivi floodplain. Once trained, the model generates segmented images where flood-affected areas are visually marked water bodies in blue, structures in red, and vegetation in green. The model achieved an Intersection over Union (IoU) score of 0.76, precision of 0.80, recall of 0.72, and an F1 Score of 0.76, highlighting its segmentation accuracy. This segmentation provides clear and accurate flood maps, supporting environmental monitoring and aiding agencies in disaster management and relief efforts. The generated maps enable better-informed decisions during flood events, ensuring a more efficient response. Future work will focus on refining model accuracy and exploring real-time applications to improve the effectiveness of disaster response capabilities.