In this paper, we address two critical challenges in the domain of flood detection: the absence of a comprehensive flood detection framework and the lack of interpretable decision-making processes in explainable AI (XAI). To overcome these challenges, an Interpretable Multi-stage Approach to Flood Detection, IMAFD, has been proposed. The proposed IMAFD provides a comprehensive, efficient and interpretable solution suitable for large-scale remote sensing tasks and offers insight into decision-making. The proposed IMAFD approach combines the analysis of the dynamic time series image sequences to identify images with possible flooding with the static, within-image semantic segmentation. It combines anomaly detection (at both image and pixel level) with semantic segmentation. The flood detection problem is addressed through four stages: (1) at a sequence level: identifying the suspected images (2) at a multi-image level: detecting change within suspected images (3) at an image level: semantic segmentation of images into Land, Water or Cloud class (4) decision making. Our contributions are twofold. First, we provide a multi-stage holistic approach to flood detection, which efficiently reduces the number of images to be processed for semantic change detection in the later stage and reduces the processing time, which is critical for rapid disaster management such as flood. Secondly, the proposed semantic change detection method (stage 3) offers human users an interpretable decision-making process, while most explainable AI (XAI) methods provide post hoc explanations. The evaluation of the proposed IMAFD framework was performed on two datasets, WorldFloods and RavAEn. For both datasets, the proposed framework demonstrates a competitive performance compared to other methods while also providing interpretability. Specifically, the proposed IDSS+ outperformed U-Net on the Worldfloods dataset by 1.57% for IoU water and 0.19% for mIoU. On the Raven dataset, the proposed IMAFD first stage outperformed DINO by 0.16 for average precision, 0.25 for average recall and 0.21 for average F1. For the flood detection task, the proposed IMAFD achieved 100% accuracy, precision recall and F1 score by setting threshold at 4. Furthermore, our framework significantly reduces the time required to process the image sequences on the Raven dataset by 212.66 s in total compared to the state-of-the-art framework.