Mangroves play important roles in the blue carbon ecosystem. Mangrove map is important data, robust and reproducible methods for mangrove mapping and monitoring are needed. Along with the freely available optical remote sensing satellite data such as Sentinel-2 and the development of deep learning fields, mangrove mapping and monitoring are more reachable. Therefore, the main goal of this study is to evaluate and utilize some state-of-the-art deep learning semantic segmentation architectures (U-Net, LinkNet, PSPNet, and FPN) for mangrove mapping and monitoring. This study will provide evidence of the ability of state-of-the-art deep learning semantic segmentation that can provide a robust and reproducible method for mangrove mapping and monitoring. The study area is the coastal zone of Rookery Bay, Florida, USA. The Sentinel-2 bottom-of-atmosphere corrected reflectance data (2016) with the target data (water body, nonmangrove, and mangrove) used for training and evaluating the capability of U-Net, LinkNet, PSPNet, and FPN for mangrove mapping. While, for the mangrove monitoring evaluation, the best trained deep learning model based on the 2016 dataset was used here to produce a new mangrove map in 2022. The reference data were collected from google earth imagery in 2022 by visual interpretation (250 points for each class) and conducted mangroves monitoring evaluation by calculating class accuracy and overall accuracy for the produced mangrove map in 2022. Based on the experiment results, all of the state-of-the-art deep learning semantic segmentation architectures have promising results and U-Net achieved the highest performance with an average intersection over union (IoU) score of 0.926. Based on the evaluation result, the trained deep learning model based on the 2016 dataset successfully produced a mangrove map using 2022 Sentinel-2 data with an overall accuracy of around 0.98. This finding indicates the ability of U-Net, LinkNet, PSPNet, and FPN for mangrove mapping and monitoring.