Development of parallel chemical transport modeling using graphical processing units

This paper describes the need for and the proposed design of the high performance computation for chemical transport model, with particular emphasis on Aerosol Optical Depth (AOD) determination. Throughout this paper, the AOD measurements of the Ozone Monitoring Instrument (OMI) are assimilated with Polyphemus model. The performance of the model simulation is simulated on three NIVIDIA Graphics Processing Units (GPUs). Multiple issues related to model analysis are studied: time partitioning to reduce truncation slip, and dimension partitioning to enhance chemical transport models´ multi-processing. The steps, strategies, processes, and optimization approaches of the GPU computations for the chemical transport model implementation are outlined. Additionally, a technique to access matrix row and column randomly is discussed. The technique could be applied to a matrix that consists of different sizes to be suitable to different memory sizes of the GPUs accelerator unit. The performance and acceleration in terms of speedup of the proposed techniques are studied on selected graphics cards. Our proposed enhancements reach an average of 3.31 speedup.