The protection of wetlands is a critical concern in environmental science, as these ecosystems serve as vital habitats for a wide range of valuable species. Remote sensing technologies, particularly satellite imagery, provide practical means for continuous monitoring of wetland conditions, including water level and spatial changes. In this study, a nationwide wetland dataset for Iran was generated using Sentinel-2 imagery and image fusion techniques, achieving a spatial resolution of 10 meters. This dataset was designed to support the training and validation of a deep learning model across various seasons and geographical regions. The proposed deep network is characterized by high depth and a reduced number of trainable hyperparameters to enhance computational efficiency while maintaining robust feature extraction. The model was applied to the Miankaleh Wetland, located in Iran's Mazandaran Province, for three consecutive years. Validation against ground truth data demonstrated high performance, with an average overall accuracy of 98%, a Kappa coefficient of 94%, and an F1-score of 99%. Finally, the temporal changes in water area over the three-year period were quantified, fulfilling the main objective of this research.