Hydroinformatician Michael Siek developed a chaos-based computer model, which seems to predict storm surges better than the models currently in use. Siek would like to see his model put to the test.
High tides and northwesterly storms have always been scary circumstances for the Dutch. In February 1953, for example, water levels rose by 4.5 metres and caused major flooding in the southwest of the country.
It’s hardly surprising then that major efforts have been put in to correctly predicting water levels hours or even days in advance. The official Dutch Continental Shelf Model (DCSM) does this by combing wind speed and direction, air pressure and tides, in a mathematical model based on physical equations and knowledge about the sea floor. However complicated and refined the model is, it is not fail-safe. In another storm, on 9 November 2007, the DCSM model underestimated the sea rise by more than one metre. The water rose even higher than in 1953.
Michael Siek, MSc (Indonesia, 1974), was not surprised by this underestimation. He argues that a model is as good as its inputs, such as weather forecasts, and the fact that – like the weather – the sea level is basically a chaotic system. This implies that small differences in the starting conditions can produce wildly different outcomes.
Instead of building a physics-based model, Siek, following an approach that the Hydroinformatics group of Unesco-IHE has been developing for the past ten years, made a highly sophisticated data-driven model based on chaos theory and neural networks. His model uses huge historical data files of water levels in, say, Hoek van Holland, without solving the underlying physics.
Chaos theory can reveal hidden patterns in seemingly random phenomena, which it does by embedding the data in hyperspace – choosing the right number of dimensions and other parameters is essential. If well chosen, repeating patterns emerge from the data, as do occasional sudden jumps to other patterns. Siek predicts the storm surge by following neighbouring trajectories in hyperspace step by step.
Strange as it may seem, chaotic prediction has performed very well,
although thus far it has only been performed on historic data. Siek would like his model to be tested side-by-side with the DCSM.
Michael Siek, ‘Predicting Storm Surges: Chaos, Computational Intelligence, Data Assimilation, Ensembles’, 6 December 2011, PhD supervisor Prof. Dimitri Solomatine (TU/IHE)
