CDS&E: Harnessing Graphical Processing Units (GPUs) to Accelerate the Computational Efficiency of Air Quality Modeling Systems for Four-Dimensional Air Pollution Predictions

空气污染是一个全国性和全球性的问题，对人类的健康和福祉具有重大的不利影响。空气质量模型模拟对于了解历史污染事件和预测未来空气质量趋势是必不可少的。然而，由于在大范围(例如，区域气盆)上以高空间分辨率模拟的案例研究的处理速度较慢，因此空气质量模型模拟的计算代价可能很高。这项研究的目标是探索使用图形处理单元(GPU)来加速社区多尺度空气质量(CMAQ)模型的计算密集型例程/模块，CMAQ模型是美国环保局和州政府机构在全国范围内采用的开源化学品运输模型，用于评估空气质量，以供监管决策。为了推进这一目标，该项目的首席调查人员建议执行一个综合的计算建模和模拟程序，该程序旨在模拟加州南海岸空气盆地(SCAB)的臭氧形成，并评估该地区臭氧形成的气象驱动因素，该地区的臭氧浓度在数十年的下降后最近已回升至1994年的水平。这一项目的成功完成将通过开发和部署更快、计算效率更高的模型/软件来支持监管空气质量监测，从而造福社会。学生教育和培训将进一步造福社会，包括指导两名博士生。监管空气质量建模和模拟需要控制大空间域上的偏微分方程(PDE)的模型的高分辨率数值解。由于图形处理单元(GPU)可以以相当的成本执行比中央处理单元(CPU)更快的浮点运算，因此它们可以为求解高维PDE系统提供显著的计算速度提升和节省。在这个项目中，PIS建议调查GPU的使用，以加速社区多尺度空气质量(CMAQ)模型的数值密集型例程/模块，EPA和州机构使用该模型来评估空气质量，以制定监管决策。CMAQ控制方程采用进程拆分方法求解，其中进程模块是串联执行的。这种方法通过将瓶颈模块迁移到GPU来提高它们的模拟时间。为了推进项目的总体目标，私人投资促进机构建议首先专注于CMAQ模型的GPU增强型气相化学解算器(GPC)的开发和硬件实现。这项工作的具体任务将包括1)通过控制方程的并行化和矢量化来加速CMAQ GPC，2)精度和灵敏度测试，3)评估GPU并行化对GPC反应速度的影响，以及4)使用基于现有空气质量数据集的案例研究来验证和应用新的CMAQ-GPU模型。这项拟议研究的成功完成可能会导致一个更快、计算效率更高的CMAQ模型/软件，以支持在未来气候情景和极端天气事件(例如热浪和野火)的气象条件下对空气质量(例如臭氧和颗粒物的形成)的预测模拟。该奖项由环境工程和计算和数据启用科学与工程(CDS&AMP；E)NSF/ENG/CBET分部的项目。该奖项反映了NSF的法定使命，并通过使用基金会的智力优势和更广泛的影响审查标准进行评估，被认为值得支持。

Air pollution is a national and global problem with significant adverse impacts on human health and wellbeing. Air quality model simulations are essential for understanding historical pollution episodes and predicting future air quality trends. However, air quality model simulations can be computationally expensive due to slow processing speeds for case studies simulated over a large area (e.g., a regional air basin) at high spatial resolutions. The goal of this research is to explore the use of graphical processing units (GPUs) to accelerate computationally intensive routines/modules of the Community Multiscale Air Quality (CMAQ) model, an open-source chemical transport model employed nationwide by EPA and state agencies to assess air quality for regulatory decision making. To advance this goal, the Principal Investigators (PIs) of this project propose to carry out an integrated computational modeling and simulation program structured to simulate ozone formation in the California South Coast Air Basin (SCAB) and evaluate the meteorological drivers of ozone formation in the SCAB where recent ozone concentrations have rebounded to 1994 levels after decades of decline. The successful completion of this project will benefit society through the development and deployment of faster and more computationally efficient models/software to support regulatory air quality monitoring. Further benefits to society will be achieved through student education and training including the mentoring of two doctoral students.Regulatory air quality modeling and simulations require high-resolution numerical solutions of the model governing partial differential equations (PDEs) over large spatial domains. Because graphical processing units (GPU) can carry out floating point operations at higher speeds than central processing units (CPUs) at comparable costs, they could provide significant computational speed enhancements and savings for solving systems of high-dimensional PDEs. In this project, the PIs propose to investigate the utilization of GPUs to accelerate numerically intensive routines/modules of the Community Multiscale Air Quality (CMAQ) model used by EPA and state agencies to assess air quality for regulatory decision making. CMAQ governing equations are solved using a process splitting approach where process modules are executed in series. This approach facilitates the improvement of simulation times for bottleneck modules by migrating them to GPUs. To advance the overarching goal of the project, the PIs propose to initially focus on the development and hardware implementation of the GPU enhanced gas phase chemical solver (GPCS) of the CMAQ model. Specific tasks for this effort will include 1) the acceleration of the CMAQ GPCS through parallelization and vectorization of the governing equations, 2) precision and sensitivity tests, 3) evaluations of the impact of GPU parallelization on GPCS reaction rates, and 4) validation and applications of the new CMAQ-GPU model using case studies based on available air quality datasets. The successful completion of the proposed research could lead to a faster and more computationally efficient CMAQ model/software to support predictive simulations of air quality (e.g., ozone and particulate formations) under future climate scenarios and meteorological conditions from extreme weather events (e.g., heat waves and wildfires) that are expected to exacerbate air pollution nationwide.This award is jointly funded by the Environmental Engineering and the Computational and Data-enabled Science and Engineering (CDS&E) programs of the NSF/ENG/CBET Division.This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.