Abstract
Flooding poses significant challenges globally, particularly in vulnerable regions such as Nigeria. Accurate detection and segmentation of flood-prone areas are critical for effective disaster management, risk mitigation, and urban planning. This research utilized Sentinel-1 satellite images and a deep learning-based segmentation approach to identify and delineate flood-prone regions. The study employed a U-Net convolutional neural network, optimized for binary segmentation, to process pre- and post-flood Sentinel-1 images. To enhance transparency and interpretability in model predictions, Explainable Artificial Intelligence (XAI) techniques were integrated into the methodology. XAI tools provide insights into the model's decision-making process, ensuring stakeholders can trust and understand the system's outputs. The dataset comprised satellite imagery of flood-affected areas, segmented into training, validation, and testing subsets. The model's performance was evaluated using metrics such as accuracy, Intersection over Union (IoU), Dice coefficient, and confusion matrix. The model achieved remarkable performance at an optimized threshold of 0.3. With a precision of 0.8705, the model correctly identifies flood regions in over 87% of its predictions, while a recall of 0.8774 indicates that it successfully captures nearly 88% of all actual flood areas. An AUC of 0.887 further confirms its robust ability to distinguish between flood and non-flood regions. Additionally, the Intersection over Union (IoU) of 0.7762 and a Dice score of 0.8740 demonstrate excellent spatial overlap between the predicted segmentation masks and the ground truth. The inclusion of XAI techniques ensures that the model outputs are not only accurate but also interpretable and actionable. This research provides a cost-effective and scalable solution for flood detection, addressing the need for trust and interpretability in AI-driven solutions. The findings highlight the potential for integration into real-world applications, including early flood warning systems, urban development planning, and disaster response frameworks.

Key words: Sentinel-1 Satellite Imaginery, Deep learning, Explainable Artificial Intelligence, Flood Detection, semantic segmentation.