The frequency and severity of coastal inundation events are increasing along most of the world’s shorelines. Global sea levels were remarkably stable for thousands of years but climate change is driving rising sea levels, presenting significant challenges for coastal management and conservation efforts. Other factors such as land subsidence and shifting weather patterns further increase the vulnerability of coastal regions to inundation. Understanding the complex interactions among these factors is necessary for developing effective strategies to mitigate the impacts of coastal inundation and improve coastal resilience. More accurate and actionable predictive models and better monitoring systems are essential in assessing and managing the risks associated with coastal inundation. This dissertation explores and assesses the potential of machine learning techniques to predict coastal inundation for short-term, i.e., hours to days, predictions on the sandy beach adjacent to the instrumented pier of Horace Caldwell Pier in Port Aransas, Texas. Traditionally, short-term water level predictions have been used to alert the public of the potential for inundation events, providing stakeholders with valuable lead time to prepare and implement mitigation measures. However, in cases where longer preparation periods are required, seasonal to multi-year water level prediction models offer extended planning for stakeholders and beach managers. This dissertation contributes to the field of coastal inundation forecasts and assessment in several ways: (1) by assessing and enhancing the performance of deep learning architectures for short-term water level predictions. Sequence-to-sequence was identified as the most appropriate architecture for the problem as it significantly improves upon existing methodologies and pushes the boundaries of reliable predictability from 48 hours to 96 hours and more for inland stations. (2) by exploring the development of machine learning models for seasonal to multi-year water level predictions in the Texas coast region, offering valuable insights for longer-term planning and adaptation strategies with lead times of three months up to three years. (3) Finally, this dissertation introduces the first predictive machine learning model for total water levels, including wave runup, while incorporating metocean variables. The development of this model was possible through the installation of a fixed camera system, resulting in a unique dataset containing 30-minute imagery of the study area for over a year, which, along with bimonthly surveys and further processing, resulted in one of the first total water level time-series data sets. This allowed for a morphological analysis that determined that for the study area, inundation events are triggered by increases in dominant wave period, average wave period, significant wave height, and average water levels. It was also found that a dominant wave period of 9.5 seconds or more leads to temporal beach erosion, with a recovery period under two weeks. Overall, this research contributes to advancing our understanding of predictive capabilities in managing coastal inundation, thereby assisting stakeholders and policymakers in developing proactive measures to safeguard coastal communities and ecosystems.