Ocean Wave Prediction and Characterization for Intelligent Maritime Transportation

• Main Author: Pokhrel, Pujan
• Format: Text
• Language: unknown
• Published: ScholarWorks@UNO 2022
• Subjects: data assimilation; machine learning; ocean waves; weather forecasting; vessel routing; Artificial Intelligence and Robotics; Data Science; Dynamic Systems; Environmental Monitoring; Numerical Analysis and Scientific Computing; Partial Differential Equations; Robotics; Sea ice
• Online Access: https://scholarworks.uno.edu/td/3025; https://scholarworks.uno.edu/context/td/article/4274/viewcontent/thesis_pujan_pokhrel.pdf; https://scholarworks.uno.edu/context/td/article/4274/filename/0/type/additional/viewcontent/ThesisandDissertationApprovalForm_Data__83_.pdf

The national Earth System Prediction (ESPC) initiative aims to develop the predictionsfor the next generation predictions of atmosphere, ocean, and sea-ice interactions in the scale of days to decades. This dissertation seeks to demonstrate the methods we can use to improve the ESPC models, especially the ocean prediction model. In the application side of the weather forecasts, this dissertation explores imitation learning with constraints to solve combinatorial optimization problems, focusing on the weather routing of surface vessels. Prediction of ocean waves is essential for various purposes, including vessel routing, ocean energy harvesting, agriculture, etc. Since the machine learning approaches cannot forecast ocean waves with sufficient accuracy for longer forecast horizons and the numerical methods are not flexible due to being expert-designed, there is a need to study both methods to improve forecasts. One popular way to improve forecasts is to perform data assimilation, which fails to improve the numerical model in the model space. In this dissertation, we explore different ways to improve wave forecasts. We combine data assimilation and machine learning methods to improve predictions from the numerical model WaveWatch III. We have also explored rogue ocean waves, which are not predicted using traditional numerical methods. Moreover, using imitation learning to guide combinatorial optimization problems should allow fast training of reinforcement learning algorithms while satisfying the constraints.