Date of Award


Document Type


Degree Name

Doctor of Philosophy (PhD)


Ocean Engineering and Marine Sciences

First Advisor

Gary A. Zarillo

Second Advisor

Ashok Pandit

Third Advisor

Stephen Wood

Fourth Advisor

Robert J. Weaver


The study aims to develop a robust and adaptable real-time forecasting system for coastal and estuarine regions, considering the challenges of limited or unavailable forecast data. To address the challenge of data unavailability, this study proposes the application of a deep learning model (DLM). This research involves the development of a high-resolution numerical model nested within global models. Calibration and validation are performed by comparing the model with observed data. The study explores the effects of various parameters, such as vertical resolution, wind variability, and bottom roughness, on model performance. It finds that the 10-layer model improves vertical stratification and current components, while 2-D wind data leads to more accurate temperature and velocity profiles near surface layer. Moderate values of the Chezy coefficient yield the best outcomes for bottom roughness parameters. The influence of the Gulf Stream on water level variations is also investigated, showing a strong correlation between non-tidal forcings and Gulf Stream flow in longer time scale. The DLMs are developed for both flow and wave models. The DLM is evaluated by comparing the simulation results obtained using DLM-predicted data with the simulation results obtained using data from global models. The results demonstrate a high correlation and around 90% accuracy for up to the first 5 days of the forecast. The real-time forecast with DLM can achieve an accuracy above 70% of the Global models' forecast data, even in the absence of hindcast data up to 4 days prior to the forecast beginning time. However, direct application of the DLM to an unknown domain significantly underperforms, highlighting the importance of retraining the DLM with local data. The real-time forecast system has been designed to handle three scenarios: forecast data from Global models are available, not available, and intermittently unavailable. Overall, the study presents a comprehensive approach to developing a real-time forecasting system for coastal and estuarine regions. It addresses the challenges of limited forecast data through machine learning techniques and demonstrates the system's flexibility in handling different data availability scenarios.


Copyright held by author