Massively Parallel Solvers for Computational Fluid Dynamics on Multi-block Cartesian Grids

Computational fluid dynamics (CFD) plays a central role in numerical weather prediction (NWP) models for forecasting and understanding the Earth as a system. Often these models produce results with large uncertainty and can fail to predict the intensity of weather events with desired accuracy. This is especially true for forecasting microscale wind over arbitrarily complex terrain, which is one of the least understood subjects in the atmospheric sciences, but is fundamental to many crucial problems, such as predicting microscale contaminant dispersion. Growing evidence indicates that these shortcomings arise from a combination of inadequate modeling of microscale physical processes within the Earth's atmosphere and inherent numerical errors in the CFD methods. To address these shortcomings, the PIs are developing a massively parallel, open-source CFD software package GEM3D (https://github.com/GEM3D) to study microscale atmospheric flows over complex terrain with numerical resolutions that are much finer than any current software allows. Our package uses block structured Cartesian adaptive mesh refinement (AMR) methods to provide a practical, accurate method for geometrically modeling complex terrain and large-eddy simulation techniques with novel dynamic sub-grid scale models for physically modeling incompressible flows in complex terrain. To make near-real time forecasting possible, the software is being designed and developed to exploit the massively parallel, multi-GPU (graphics processing unit) computing paradigm efficiently.

Download : ai-ca-ha-pe-se-wr-2018.pdf