Automatic improvement of unstructured grids for coastal simulations

Abstract

The generation of unstructured triangular grids for coastal applications requires extensive manual tunning, in order to improve the stability, accuracy and efficiency of the model simulations. As current grids reach 105-106 nodes, this manual tunning is no longer feasible. This paper presents a new post-processor that aims at optimizing the local distribution of nodes and elements by adding and deleting nodes and by changing the connections between nodes. The algorithms target a smooth transition between elements sizes by making the number of connections of each internal node as close to 6 as possible. An application of the post-processor to various grids shows that the grid roughness is reduced by 10-30%, while decreasing marginally the number of nodes and maintaining the global nodal distribution. To illustrate the benefits of the post-processor, a grid with 130.000 nodes of the maritime zone under Portuguese jurisdiction is used to simulate tides with a good accuracy (root mean square errors below 10 cm) and without numerical oscillations. The model results provide an extensive database of tidal elevations that can feed boundary conditions to local estuarine models.