• Medientyp: E-Artikel
  • Titel: A Neural Network Approach to Predict Nearshore Morphology along the Dutch Coast
  • Beteiligte: Cser, Josef; Oosterlaan, Liesbeth M.; van der Veer, Peter; Heemink, Arnold W.; van de Graaf, Jan
  • Erschienen: Coastal Education & Research Foundation (CERF), 2001
  • Erschienen in: Journal of Coastal Research (2001), Seite 143-153
  • Sprache: Englisch
  • ISSN: 0749-0208; 1551-5036
  • Schlagwörter: Beaches and Sedimentation
  • Entstehung:
  • Anmerkungen:
  • Beschreibung: This paper presents a neural network approach to analyze beach profile change and nearshore morphology of the multiple bar system along the Dutch coast. The main objective of the research is to understand the processes that underlie the cyclic behavior of the nearshore zone and to predict its future developments. Three neural networks will be constructed in three successive phases, which together are meant to help attain these objectives. The first neural network is based on morphologic data only. The morphologic data come from annual beach profile monitoring, which have been pre-processed by principle component analysis. This analysis has been carried out to remove noisy data and obtain the main patterns of the data set. The first neural network model is used to determine the relation between the length of the input period and the beach profile changes a certain number of years ahead. In the second network, using the results of the first network, other parameters are included, to analyze the relation between profile shape and hydrodynamic processes, as well as morphologic boundary conditions. The third network includes data of other areas along the Dutch coast and aims at validating the morphodynamic relations of the second network. This paper is focused on the development of the first neural network and aims at the proper input definition and network architecture. When complete profile measurements are presented to the network, the network reasonably predicts the main trends of bar migration, although it underestimates the height of the bars. Therefore, it was decided to train the network with reduced data and use the eigenvector coefficients for simulation and prediction. This method is still under construction, but early results show improving network performances when it comes to simulation. The prediction however, yielded lesser results.
  • Zugangsstatus: Freier Zugang