This report presents the work conducted for the ARCHER2 eCSE05-3 project. The goal was to improve the scalability and object-resolving capability of the uDALES framework (Suter et al., 2022). There are three specific changes in- volved, which together constitute an upgrade of the codebase from version 1.0 to 2.0. Firstly, uDALES v1.0 has a one-dimensional domain decomposition, which means it is not able to effectively utilise the large number of processors available on ARCHER2. To address this, a two-dimensional domain decomposi- tion was implemented in uDALES v2.0 using the library 2DECOMP&FFT (Li and Laizet, 2010), and the model is shown to scale well for realistic use cases. Secondly, the pressure solver in uDALES v1.0 is not capable of using the Fast Fourier Transform (FFT) algorithm with inflow-outflow boundary conditions, instead relying on slower algorithms. This was made possible in uDALES v2.0, and the implementation has been validated using canonical urban cases. Finally, the immersed boundary method used in uDALES v1.0 only permits obstacles that align with the Cartesian computational grid, thus hindering its ability to model realistic urban geometries. uDALES v2.0 instead describes the geometry using an unstructured triangulated surface, which required a fundamental refor- mulation of the immersed boundary method and the pre-processing associated with the geometry. The implementation has been validated in both aligned and non-aligned cases.

This work was funded under the embedded CSE programme of the ARCHER2 UK National Supercomputing Service (http://www.archer2.ac.uk).