A GPU-Implemented Lattice Boltzmann Model for Large Eddy Simulation of Turbulent Flows in and around Forest Shelterbelts

• Published in: Atmosphere Vol. 15; no. 6; p. 735
• Main Authors: Wang, Yansen, Zeng, Xiping, Decker, Jonathan, Dawson, Leelinda
• Format: Journal Article
• Language: English
• Published: Basel MDPI AG 01.06.2024
• Subjects: Agricultural land; Atmospheric boundary layer; atmospheric turbulence; Barriers; Central processing units; Computer applications; Computing time; CPUs; Drag; Flow velocity; Fluid dynamics; Forests; GPU computing; Graphics; Graphics processing units; Kinetic energy; Large eddy simulation; large eddy simulation of turbulent flow in and around forest shelterbelt; Large eddy simulations; lattice Boltzmann model; Mean winds; Microclimate; Mixed forests; Modelling; Observational studies; Parameter sensitivity; Porosity; Real time; Reynolds number; Sensitivity analysis; Shelterbelts; Simulation; Turbulent flow; Turbulent kinetic energy; Turbulent wind; Urban agriculture; Urban areas; Vortices; Wind barriers; Wind direction; Wind effects; Wind engineering; Wind speed; Winds
• ISSN: 2073-4433; 2073-4433
• DOI: 10.3390/atmos15060735

Using porous wind barriers for the microclimate modification of agricultural lands, urban areas, and surrounding roads is a ubiquitous practice. This study establishes a new method for numerically modeling the turbulent flow in and around forest shelterbelts using an advanced multiple-relaxation-time lattice Boltzmann model (MRTLBM). A detailed description is presented for a large eddy simulation (LES) of turbulent winds by implementing barrier element drag force in the MRTLBM framework. The model results for a forest shelterbelt are compared with a field observational dataset. The study indicated that our implementation of drag force in MRTLBM is an accurate method for modeling turbulent flows in and around forest patches. Sensitivity analyses of turbulent flow related to the shelterbelt structure parameters and wind directions are also carried out. The analysis indicated that the optimal wind shelter effect in reducing the mean wind speed and turbulent kinetic energy is maximized using a narrow, medium porosity shelterbelt, with the wind direction perpendicular to the shelterbelt. These conclusions are in agreement with other observational and modeling studies. Finally, the computational time of a central processing unit (CPU) and graphics processing unit (GPU) was compared for a large domain with 25 million grids to demonstrate the MRTLBM advantage of LES in regards to computational speed with a mixed forest and building environment. The GPU is approximately 300 times faster than a CPU, and real-time simulation for this large domain is achieved using the Nvidia V100 GPU.