Block-structured grids for finite element models of coastal ocean
- 1Mathematical Institute, University of Bayreuth, Bayreuth, Germany
- 2European Centre for Medium-Range Weather Forecasts, Bonn, Germany
- 3Courant Institute of Mathematical Sciences, New York University, New York, United States
For numerical models of the ocean, the choice of the underlying grid is a crucial technical decision affecting the accuracy, stability, computational performance, and, ultimately, modeling skill. This is a particularly important issue for high-resolution simulations of coastal and regional oceans, where a precise representation of irregular land boundaries and geometric features such as islands, channels, rivers, etc. is a key requirement. For these types of applications, unstructured triangular meshes are often the preferred type of horizontal mesh due to their adaptability and ease of construction. However, optimizing the computational performance of unstructured mesh codes on many modern hardware architectures is quite challenging compared to doing the same for stencil-based numerical schemes used in structured grid models. As a viable alternative to unstructured meshes, we propose block structured grids (BSGs) consisting of a topologically unstructured mesh of blocks, each of which is partitioned using a structured grid. Our methodology allows to automatically generate BSGs for realistic ocean domains with a prescribed number of blocks of given resolution already load-balanced for execution on a parallel computer of given configuration. To allow generating BSGs for ocean domain geometries and topographies of varying complexity our approach supports three different types of BSGs:
1. standard BSGs -- very computationally efficient but only practical for rather simple geometries
2. masked BSGs -- an extension of the standard BSGs that permits masking of elements to allow meshing of small geometric features using large blocks
3. hybrid BSGs -- an entirely new method combining structured with unstructured blocks to offer the optimal compromise between geometric accuracy and computational efficiency
We present grid generation techniques, validation of simulation results, and computational performance evaluations for the proposed methods.
How to cite: Schmalfuß, J., Faghih-Naini, S., Zint, D., and Aizinger, V.: Block-structured grids for finite element models of coastal ocean, EGU General Assembly 2024, Vienna, Austria, 14–19 Apr 2024, EGU24-16005, https://doi.org/10.5194/egusphere-egu24-16005, 2024.