Shallow water equations are extensively considered in the domains of oceans, atmospheric modelling, and engineering research (Franca et al., 2022), which play significant roles in floods and tsunami governance. Nonetheless, the accurate prediction of shallow water behaviours is regarded as an arduous undertaking, particularly when confronted with multi-dimensional data and potential errors within the model. To address these challenges and improve the accuracy of forecasts, this study employs an integrated approach, involving dimensionality reduction methods, deep learning architectures, and data assimilation techniques. Indeed, Reduced-order modelling facilitates the conversions of high-dimensional data, extracting important features and attenuating the complexity of problems (Zhong et al., 2023). Subsequently, three different predictive models are utilized to prognosticate shallow water data in the reduced latent space, followed by comparisons of their prediction performance. Moreover, Bach and Ghil (2023) propose that through the amalgamation of model forecasts with observational metrics, the data assimilation algorithm can rectify their discrepancies, thereby enhancing the model's predictive prowess. Finally, the experimental results demonstrate that prediction values are congruent with actual observations, which accentuates the resilience and effectiveness of this comprehensive methodology. Its potential to accurately forecast shallow water data holds the applicability and referential significance in preventing storm swell and other meteorological events.