Neural networks for spatial and temporal ocean wave height prediction considering coastal morphodynamics in the East Frisian North Sea / vorgelegt von Christoph Jörges. Wuppertal, Dezember 2022
Content
List of Figures
List of Tables
Nomenclature
Abstract
Kurzfassung
Introduction
Introduction and motivation
Emerging climate-related hazards to the North Sea coast
State of the art and basic related work
What waves are and how they are measured and modeled
Neural networks and their application for wave height prediction
Study area and ETD sandbanks
Research questions, aims, and objectives
Research Study I
Introduction
Research area and data
Morphodynamic analysis of the ETD sandbanks by GIS
Analysis methods of ETD sea state damping
Buoy data
Statistical methods of natural sea state variability
Numerical modeling
Model validation and transfer to test cases
Sea state damping effect
Results and discussion of ETD sea state damping
Natural wave climate variability
Sea state damping effect – impact of ETD sandbanks on sea state
General findings of the buoy-measured damping effect
Conclusions
Research Study II
Introduction
Methods
Research area and dataset
Morphodynamic influences and bathymetry preprocessing
Long Short-Term Memory neural networks
Model settings
Loss function, optimizer, and error metrics
Hyperparameter tuning
Results and discussion
Model comparison – nearshore excluding bathymetry features
Model comparison – validation on offshore data
Impact of bathymetry features
Reconstruction including bathymetry
Prediction including bathymetry
Limitations of including bathymetric data
Further feature selection analysis
Conclusions
Research Study III
Introduction
Material and methods
Study area
Metocean data
ETD sandbank characterization and geostatistical bathymetry simulation
SWAN modeling
Neural network data preprocessing
Convolutional neural networks (CNN)
Neural network model settings and architecture
Loss function, optimizer, and error metrics
Hyperparameter tuning
Results and discussion
Model performance of spatial prediction
CNN neural networks vs. SWAN numerical model
High-resolution uncertainty estimation on the coastline
Application of the proposed method for a climate change scenario
Conclusions
Synopsis
Discussion
Coastal morphodynamics, ETD variability, and damping effect
Bathymetric data and observation methods for improving predictive sea state modeling
ML vs. numerical modeling – Spatial wave predictions using synthetic and real bathymetry data
Conclusions
Outlook
Summary
Bibliography
Appendix