Predicting the introduction of avian influenza remains a sensitive issue, with a direct impact on humans and animals’ health, which requires considerable efforts to prevent and limit its spread within the Moroccan territory. This is due to the inability to interrupt the natural phenomenon of wild birds’ migration, which carries the disease and consequently contaminates other species, including poultry. Since the spread of this disease primarily occurs in wetland areas, experts are keen to identify and border these regions. To address this, our study provides a decision aid tool based on segmentation of water bodies within these ecosystems, which are susceptible sites for contamination. This is achieved through three deep learning-based semantic segmentation algorithms, namely the UNet, Attention UNet, and TransUNet models, trained on Sentinel-2 satellite imagery. The aim is to achieve optimal predictions and select the most adept model among these three with the best ability to recognize and extract water bodies. The process includes data preprocessing, hyperparameter tuning, qualitative and quantitative validation, and model performance enhancement through augmentation techniques using the Segment Anything model. Finally, the resulting model is deployed through a local web application. The results achieved by the most performant model are as follows: An F1 score of 87.21%, the average of the Intersection over Union (mIoU) of 74.31%, and an accuracy of 69.68%, which proved competitive with previous approaches to water body extraction.