EGU General Assembly 2020
© Author(s) 2020. This work is distributed under
the Creative Commons Attribution 4.0 License.

Application of an artificial neural network to generate wave projections at southern African coasts

Felix Soltau1, Matthias Hirt1, Jessica Kelln1, Sara Santamaria-Aguilar2, Sönke Dangendorf3, and Jürgen Jensen1
Felix Soltau et al.
  • 1University of Siegen, Research Institute for Water and Environment, Civil Engineering, Siegen, Germany (
  • 2Christian-Albrechts-University, Coastal Risks and Sea-Level Rise | Research Group, Institute of Geography, Kiel, Germany
  • 3Old Dominion University, Norfolk, VA, USA

In the past decades, severe so called ‚compound events‘ led to critical high water levels at the coasts of southern Africa and as a consequence to property damage and loss of human life. The co-occurrence of storm surges, wind waves, heavy precipitation and resulting runoff increases the risk of coastal flooding and exacerbates the impacts along the vulnerable southern African coasts (e.g. Couasnon et al. 2019). To mitigate these high-impacts, it is essential to understand the underlying processes and driving factors (Wahl et al. 2015). As compound flooding events at southern African coasts are dominated by wind waves, it is of great importance to investigate the regional wave climate to understand the wave forcing as well as the origin of the wave energy.

Wind waves around southern African coasts are affected by the complex interactions between the Agulhas current and the atmosphere. In the research project CASISAC* we analyse the present evolution of the Agulhas Current system and quantify its impact on the future regional wave climate. Ocean waves contributing to high sea levels can be generated offshore resulting in swell or closer to the coasts by strong onshore winds. To identify responsible atmospheric pressure fields that force high wind wave events we apply a hybrid approach: (1) linking south hemispheric pressure fields with offshore wave data using an artificial neural network and (2) determine the prevailing nearshore wave conditions by regional numerical wave propagation models (SWAN). By validating the modelled nearshore wave data from hindcast runs with wave buoy records, this approach allows us to predict future extreme wind wave events and thus potential flooding. In a next step, extreme value analysis is used to determine future return periods of extreme wave events. These results can contribute to the development of more reliable adaptive protection strategies for southern African coast.

*CASISAC (Changes in the Agulhas System and its Impact on Southern African Coasts: Sea level and coastal extremes) is funded by the German Federal Ministry of Education and Research (BMBF) through the project management of Projektträger Jülich PTJ under the grant number 03F0796C


Couasnon, Eilander, Muis, Veldkamp, Haigh, Wahl, Winsemius, Ward (2019): Measuring compound flood potential from river discharge and storm surge extremes at the global scale and its implications for flood hazard. In: Natural Hazards and Earth System Sciences, Discussion Paper, S. 1–24. DOI: 10.5194/nhess-2019-205, in review.
Wahl, Jain, Bender, Meyers, Luther (2015): Increasing risk of compound flooding from storm surge and rainfall for major US cities. In: Nature Climate Change 5 (12), S. 1093–1097. DOI: 10.1038/nclimate2736.

How to cite: Soltau, F., Hirt, M., Kelln, J., Santamaria-Aguilar, S., Dangendorf, S., and Jensen, J.: Application of an artificial neural network to generate wave projections at southern African coasts, EGU General Assembly 2020, Online, 4–8 May 2020, EGU2020-19940,, 2020

Comments on the presentation

AC: Author Comment | CC: Community Comment | Report abuse

Presentation version 1 – uploaded on 04 May 2020
  • CC1: Comment on EGU2020-19940, Theodore Shepherd, 05 May 2020

    If you are just using SLP as a predictor, training the network on historical data, there is a potential problem. Some time ago a paper was published claiming that extratropical storms would become more intense under climate change. (It was even highlighted as a bullet by IPCC.)  This was based on the observed relationship between storms and minimum SLP. But the climate change prediction followed from a hemispheric-wide decrease in SLP at midlatitudes. And it's not SLP that matters for storms, it is the gradient in SLP (which is related to winds). Hence the index was non-stationary. The IPCC has now had to back down from that result. So be careful here.

    • AC1: Reply to CC1, Felix Soltau, 05 May 2020

      Thanks for your comment! So you mean we should prepare more of the physical background of wave generation by calculating the gradients before we start to train the ANN? Would you then even go further and calculate geostrophic winds for example? How far would you go explaining the physics before using a variable as predictor?

      I would be very interested in your answer. These were questions we asked ourselves before we decided to start at the beginning with the SLP.

      Thank you.


      • CC2: Reply to AC1, Theodore Shepherd, 05 May 2020

        It's not my field, but I believe that wave forecasting in NWP is quite mature, so I would think there is fairly good knowledge of what would be the best wind index. I suspect it would not be geostrophic, though, since it's in the boundary layer. And then for the future: look out for a paper by J. Mindlin et al. (I am a co-author) which is in press in Climate Dynamics, on storylines of SH midlatitude circulation change. You could consider different storylines of wind changes. (An example for Europe for wind extremes over land is in Zappa and Shepherd 2017 J. Clim.)