The paper, led by Alina Nitzke from the Geoinformatics group, has been accepted in Computers & Geosciences and is available as a preprint.
In the new approach, a triangulation of the tide gauge locations is found through an exact optimization method, in a way such that the resulting sea surface fits best to the satellite data in the altimeter era. For the past, where we have no altimeter data, this optimized triangulation is used for interpolating the tide gauge data. The triangulation is thus learned on the basis of dense satellite data and then transferred to the past. The groups prove in their study that this provides a better reconstruction of the sea surface as compared to existing approaches for at least twenty years backwards.
In collaboration with computer scientists at the University of Bonn (Arutyunova et al. 2021) the enormous complexity (NP-hardness) of the triangulation problem was formally proved. This result is important for algorithm development as it shows that some approaches are doomed to fail whereas others are particularly promising. Integer linear programming, which is the approach chosen by Nitzke et al., belongs to the latter category.
- Alina Nitzke, Benjamin Niedermann, Luciana Fenoglio-Marc, Jürgen Kusche, and Jan-Henrik Haunert (2021)
Reconstructing the dynamic sea surface from tide gauge records using optimal data-dependent triangulations
Computers & Geosciences, 2021. Accepted for publication.
- Anna Arutyunova, Anne Driemel, Jan-Henrik Haunert, Herman Haverkort, Petra Mutzel, and Heiko Röglin (2021)
Minimum-error triangulation is NP-hard.
In Proceedings of the 37th European Workshop on Computational Geometry (EuroCG'21).