A neural network super-resolution approach for the reconstruction of
coastal sea states
Abstract
In this paper, a neural-network-based super-resolution technique is
applied to the reconstruction of significant wave height and other sea
state variables calculated over coarse meshes by a spectral wave model.
The potential of the technique is demonstrated in a case study and the
efficiency of the training process as well as the requirements with
respect to data quality are analyzed. In this particular example,
reasonable accuracy is achieved using only one year of training data
with the help of traditional Machine Learning methods like Transfer
Learning and Data Augmentation. The presented method leads to up to
50-times lower computation time in comparison to an equivalent
traditional direct modeling approach at fine resolution. Overall,
incorporation of the presented method into major wave forecasting
systems has the potential to allow for the creation of “zoomed-in’
areas of interest without the requirement for supplementary calculations
at higher resolution.