Storm surge early warning with machine learning methods including dynamic bathymetries in the East Frisian North Sea
Christoph JÖRGES, University of Wuppertal, Institute of Geography, Human-Environment Research, Germany
Britta STUMPE, University of Wuppertal, Institute of Geography, Human-Environment Research, Germany
Storm surges have always caused great danger to people living along the North Sea coast. Rising sea levels due to climate change are increasingly intensifying the risk. Accurate predictions of sea state conditions caused by storm surges are therefore essential for warning the affected population. Currently, however, storm surge forecasts on the Lower Saxony North Sea coast cannot predict small-scale wave height impacts which are important for location-specific installation of e.g. sandbags and flood protection walls. For the East Frisian Islands, the Ebb Tidal Delta (ETD) sandbanks provide a natural coastal protection. However, the ETD sandbanks are constantly in morphological change and can already alter their form by single storm surges. Therefore, we developed a new method to predict wave heights on small spatial scales of 100m along the coast of Norderney island. The proposed method additionally takes into account the morphodynamic change of the ETD sandbanks off the Norderney island. We therefore simulated different potential ETD bathymetries by geostatistical methods as input to our predictive model. This machine learning prediction model is based on a mixed-data CNN-LSTM neural network that is designed to process both images (bathymetry maps) as well as time series. The neural network also received numerically simulated sea state data (simulations using the SWAN wave model), water level data, and wind data as input. The output of the neural network enables predictions and warnings of wave heights based on latest water level and wind forecasts, as well as possible morphodynamic changes in the ETD sandbanks. The proposed model reduces the prediction time for a single event by a factor of the order of 100. The influence of sea level rise and potential future storm surge events can also be included with this method.
Mots clés : storm surge early warning|machine learning|wave height prediction
A104034CJ