Flood mapping using satellite data poses significant challenges for disaster response teams. In addressing these challenges, our study introduces a deep learning framework designed to map flood extents in turbid water bodies, thereby advancing practical applicability. The proposed model is a semi-supervised architecture based on ResNet-18, fine-tuned for binary classification of satellite images into flooded or non-flooded categories. Acknowledging the low availability of training samples, our approach leverages a semi-supervised methodology, capitalizing on the benefits of unlabeled data.We explore the challenges of detecting images flooded with turbid water on the model. For the initial training phase, the model undergoes rigorous training on distinct datasets comprising Sentinel-2A images with varying dimensions, specifically \(128\,\times \,128\,\times \,3\) and \(256\,\times \,256\,\times \,3\) pixels. Subsequently, the pre-trained model undergoes transfer learning on a flood dataset from Louisiana, enhancing its adaptability to diverse flood scenarios and satellite sources.The capacity to automatically map floods solves significant problems for emergency response teams.