Back to search

National Technical University of Athens (NTUA)

Wave input reduction methods combining numerical modelling & machine learning for the prediction of the annual coastal bed morphological evolution

Abstract

dc:description

The numerical prediction of coastal bed evolution has been at the forefront of coastal engineering research for several decades. Even though numerical models are capable of resolving in great detail the coastal processes driving the morphological bed evolution, they are associated with staggering computational burden, rendering a long-term prediction a tedious task. The burden further increases considering the vast amount of input wave characteristics that are used to force the models. To counterbalance this, various wave input reduction methods have been developed and employed in coastal engineering practice. Despite their widespread usage, a need to further accelerate the morphological simulations, while simultaneously improving the reliability of the wave input reduction methods was identified. This thesis undertakes a systematic effort to thoroughly evaluate various types of wave input reduction methods and provide practical guidelines on the optimal method selection and configuration. Two new wave binning wave input reduction methods were conceptualized and implemented in the coastal area of Rethymno, in Greece. These methods focus on thee limination of sea-states considered unable to initiate sediment motion, effectively reducing the length of the forcing timeseries. The newly developed methods produced reliable results compared to the widely used energy flux input reduction method but also achieved a significant model run time reduction of up to 62%, compared to a brute force simulation containing the full forcing timeseries. Additionally, the K-Means clustering algorithm was thoroughly evaluated as a viable alternative to the classical binning input reduction methods and several enhancements were proposed aiming to overturn the unsupervised nature of the clustering algorithm. Implementation of the configurations in the coastal area of Rethymno showcased that the default parametrization of the algorithm can produce satisfactory predictions of annual coastal bed evolution. All the proposed enhancements, exhibited a performance increase compared to the default parametrization, however they added a degree of complexity in the algorithm implementation. A systematic evaluation of the Representative Morphological Wave Height selection approaches was also carried out, with the development of three new alternative configurations. Notable was the training and validation of an Artificial Neural Network that is included in one of the methods and is tasked with the elimination of lowly energetic sea-states that have little to no effect in the morphological bed evolution. The three proposed enhancements provided a noteworthy performance increase compared to the traditional method, with the best performing one being the method incorporating the Artificial Neural Network. Last but not least, selected input reduction methods were compared with available field measurements at the coastal area of Eresos, Lesvos in Greece, in an attempt to investigate the reliability of wave input reduction methods in real-life settings. The numerical model forced with three selected input reduction methods reproduced the morphological bed evolution in a very satisfying manner, with the best performing being once again the one incorporating the Artificial Neural Network. This thesis provides a thorough evaluation of wave input reduction methods, testing several configurations and enhancements, validating their use against both benchmark numerical predictions and field measurements. The incorporation of machine learning in wave input reduction can further increase the reliability of model predictions, leading to a significant reduction of the required computational effort.

Degree

thesis:*
Grantor dc:publisher
National Technical University of Athens (NTUA)
Year dc:date
2023

Author and committee

dc:creator, dc:contributor.*
Authors dc:creator
  • Papadimitriou, Andreas
  • Παπαδημητρίου, Ανδρέας

Subjects

dc:subject × 13

Rights

Language dc:language
eng

Identifiers

dc:identifier.*
Identifier
10.12681/eadd/54958

Chain of custody

source
Harvested from
Greek National Archive of PhD Theses
Base URL
phdtheses.ekt.gr/eadd_oai/request
Last updated
2026-11-18
Source record
OAI-PMH GetRecord
citation

Papadimitriou, Andreas; Παπαδημητρίου, Ανδρέας. Wave input reduction methods combining numerical modelling & machine learning for the prediction of the annual coastal bed morphological evolution. National Technical University of Athens (NTUA), 2023. http://hdl.handle.net/10442/hedi/54958