Preprints
https://doi.org/10.5194/egusphere-2025-196
https://doi.org/10.5194/egusphere-2025-196
03 Feb 2025
 | 03 Feb 2025
Status: this preprint is open for discussion and under review for Natural Hazards and Earth System Sciences (NHESS).

Computing Extreme Storm Surges in Europe Using Neural Networks

Tim H. J. Hermans, Chiheb Ben Hammouda, Simon Treu, Timothy Tiggeloven, Anaïs Couasnon, Julius J. M. Busecke, and Roderik S. W. van de Wal

Abstract. Because of the computational costs of computing storm surges with hydrodynamic models, projections of changes in extreme storm surges are often based on small ensembles of climate model simulations. This may be resolved by using data-driven storm-surge models instead, which are computationally much cheaper to apply than hydrodynamic models. However, the potential performance of data-driven models at predicting extreme storm surges is unclear because previous studies did not train their models to specifically predict the extremes, which are underrepresented in observations. Here, we investigate the performance of neural networks at predicting extreme storm surges at 9 tide-gauge stations in Europe when trained with a cost-sensitive learning approach based on the density of the observed storm surges. We find that density-based weighting improves both the error and timing of predictions of exceedances of the 99th percentile made with Long-Short-Term-Memory (LSTM) models, with the optimal degree of weighting depending on the location. At most locations, the performance of the neural networks also improves by exploiting spatiotemporal patterns in the input data with a convolutional LSTM (ConvLSTM) layer. The neural networks generally outperform an existing multi-linear regression model, and at the majority of locations, the performance of especially the ConvLSTM models approximates that of the hydrodynamic Global Tide and Surge Model. While the neural networks still predominantly underestimate the highest extreme storm surges, we conclude that addressing the imbalance in the training data through density-based weighting helps to improve the performance of neural networks at predicting the extremes and forms a step forward towards their use for climate projections.

Publisher's note: Copernicus Publications remains neutral with regard to jurisdictional claims made in the text, published maps, institutional affiliations, or any other geographical representation in this preprint. The responsibility to include appropriate place names lies with the authors.
Share
Tim H. J. Hermans, Chiheb Ben Hammouda, Simon Treu, Timothy Tiggeloven, Anaïs Couasnon, Julius J. M. Busecke, and Roderik S. W. van de Wal

Status: open (until 17 Mar 2025)

Comment types: AC – author | RC – referee | CC – community | EC – editor | CEC – chief editor | : Report abuse
Tim H. J. Hermans, Chiheb Ben Hammouda, Simon Treu, Timothy Tiggeloven, Anaïs Couasnon, Julius J. M. Busecke, and Roderik S. W. van de Wal
Tim H. J. Hermans, Chiheb Ben Hammouda, Simon Treu, Timothy Tiggeloven, Anaïs Couasnon, Julius J. M. Busecke, and Roderik S. W. van de Wal

Viewed

Total article views: 98 (including HTML, PDF, and XML)
HTML PDF XML Total BibTeX EndNote
84 10 4 98 1 1
  • HTML: 84
  • PDF: 10
  • XML: 4
  • Total: 98
  • BibTeX: 1
  • EndNote: 1
Views and downloads (calculated since 03 Feb 2025)
Cumulative views and downloads (calculated since 03 Feb 2025)

Viewed (geographical distribution)

Total article views: 124 (including HTML, PDF, and XML) Thereof 124 with geography defined and 0 with unknown origin.
Country # Views %
  • 1
1
 
 
 
 
Latest update: 10 Feb 2025
Download
Short summary
We studied the performance of different types of neural networks at predicting extreme storm surges. We found that that performance improves when during model training, events with a lower density are given a higher weight. Additionally, we found that the performance of especially convolutional neural networks approaches that of a state-of-the-art hydrodynamic model. This is promising for the application of neural networks to climate model simulations.
Share