Bangladesh has been a nation deeply affected by a growing concern for environmental degradation, with riverbank erosion as the most prevalent concern that also has a strong socio-economic implication. The process of erosion is responsible for an annual loss of fertile land that is home to not just houses and rural households, but also the country's vital social and infrastructure like roads, schools, and markets. Consequently, several thousands people living along river banks are displaced due to a constant loss of land. For people living along the riverbanks, river erosion, besides being a natural disaster, is also posing a risk to the ownership and ownership of their property and life. Being one of the most unstable river in Bangladesh, the Jamuna is a major source of instability with high discharge and sediment load in the monsoon. Its many braids can split, join and even move from place to place. Thus, it becomes difficult to assess the changes in its physical and spatial parameters and as a result it has remained highly challenging to perform and develop monitoring. Conventional methods of riverbank erosion monitoring have several limitations. Field surveys can provide useful and accurate information, but they are expensive, time consuming, and difficult to conduct frequently over such a large and dynamic river system. As a result, satellite-based monitoring provides an effective alternative for observing riverbank changes over large areas and across different time periods. This study uses time series Sentinel-1 Synthetic Aperture Radar (SAR) data to analyze riverbank erosion along the Jamuna River. Sentinel-1 GRD images with VV and VH polarizations are processed to distinguish water and land areas. A threshold based classification method is applied to generate dry season water masks. These masks are used to extract river boundaries, examine bankline movement and identify erosion and deposition patterns over time. Apart from the study of historical erosion, this work also considers the prediction of future riverbank erosion. The Earthformer spatiotemporal deep learning technique is used for the learning of the historical movement patterns of the Jamuna River using watermask sequences. Once trained, the technique is used for prediction of the river extent until the year 2040. Using comparison of water mask predictions between two successive years, the areas which will be changed to water can be detected as future erosion sites. The primary aim of this research work is to create a deep learning-based predictive model from satellite images for the prediction of riverbank erosion and river dynamics. It involves creating historical dry season water masks, analyzing the shifting of river channels, developing a deep learning-based predictive model, predicting future river extent by using Earthformer, detecting the areas that are prone to erosion, and calculating the year-wise erosion area. This can help the planners, engineers, disaster managers, and local administrators in locating the vulnerable areas.