Accelerating the computation of air quality projections over India using novel computing

硝酸盐气溶胶很可能是未来空气污染事件的一个重要因素，随着人们对空气质量的担忧导致二氧化硫排放量下降，它将变得更加重要。然而，由于在气候模式中表示它的复杂性，它没有包括在当前一代的大多数地球系统模式中。对COVID-19封锁期间雾霾事件的案例研究也表明，二次化学反应对空气污染事件的重要性。对这些事件的全面评估需要复杂化学的表征，由于计算的限制，迄今为止还不可行。UKCA最近的工作已经实施了一个适合空气质量应用的方案，但这比目前在地球系统模型中使用的方案要大得多，也复杂得多。为了实现这样的模拟，需要新的计算技术，可以利用高度并行的计算设备，如gpu。具体来说，这个项目解决了两个研究问题。首先是什么先进的计算科学技术可以应用到UKCA科学代码中，使模型可移植到多个计算机体系结构中，并且，关键的是，在这些体系结构上表现良好？这就提供了解决第二个问题的方法：气溶胶减少对近期印度空气质量预测的影响是什么？与标准方案的预测相比，这些预测在硝酸盐气溶胶与这种改进的化学反应相结合的模拟中有何不同？UKCA盒模型的新开发功能将用于测试和实现gpu上计算昂贵的例程。这些是高度并行的计算设备，每瓦特的计算性能比传统的计算机处理器更高。它们需要新颖和并行的编程方法来开发。此外，处理器体系结构的激增意味着每种处理器都需要不同的方法和编码模式。通过遵循关注点分离的概念（英国气象局在开发新的大气模型Gung Ho/ LFRic时已经成功地开创了这一概念），代码既可以移植到不同的架构中，又可以在不同的架构上运行。计算能力的提高将使在多年代际模拟中使用更复杂的气溶胶和化学表征成为可能。由英国气象局开发的LFRic模型已经可以与UKCA一起运行，并计划在未来两年内进一步开发，因此该模型将足够成熟，可以部署新的计算能力，并实现以前不可行的计算。研究硝酸盐气溶胶对未来空气质量的影响的能力，以及捕捉二次化学反应所需的复杂化学反应的能力，是该领域的前沿。对印度空气质量案例研究的应用将是新颖的，而且，由于预计该区域的排放将发生巨大变化，将为有关排放及其对健康影响的重要政策问题提供信息。

Nitrate aerosol is likely to be an important contributor to future air pollution events and will become more important in time as emissions of sulphur dioxide decline due to air quality concerns. However, due to the complexity of representing it in climate models, it iis not included in the majority of the current generation of Earth System models. Case studies of haze events during the COVID-19 lockdowns also showed the importance of secondary chemical reactions to air pollution episodes. A comprehensive assessment of such events requires the representation of complex chemistry, and has not been feasible to date due to computational constraints. Recent work on UKCA has implemented a suitable scheme for air quality applications, but this is significantly larger and more complex than the current scheme used within the Earth System model. To enable such simulations, new computational techniques which can exploit highly parallel compute devices such as GPUs are required. Specifically, this project addresses two research questions. The first is what advanced computational science techniques can be applied to the UKCA science code to make the model portable to multiple computer architectures and, critically, performant on those architectures? This then provides the methodology to address the second question: what is the effect of aerosol reductions on projections of near-future Indian air quality, and how do these projections differ in simulations with nitrate aerosol coupled with this improved chemistry, compared to projections with the standard scheme? The newly developed functionality of the UKCA box model will be used to test and implement the computationally expensive routines on GPUs. These are highly parallel computational devices which have higher computational performance per watt than traditional computer processors. They require novel and parallel, programming methods to exploit. Moreover, the proliferation of processor architectures means different methods and coding patterns are required for each. By following the concept of the separation of concerns, which has been successfully pioneered by the Met Office in developing the new atmospheric model, Gung Ho/ LFRic, the code can be made both portable to different architectures and performant on them. The increase in computational power will enable the use of greater complexity in the representation of aerosols and chemistry in multi-decadal simulations. The LFRic model developed by the Met office can already run with UKCA and further development is planned in the next two years, thus the model will be sufficiently mature to deploy the new computational capability and enable a previously unfeasible calculation. The ability to examine the effect of nitrate aerosol, and the complex chemistry required to capture secondary chemical reactions, on future air quality is at the forefront of the field. The application to an Indian air quality case study will be novel, and, as large emission changes are anticipated in this region, will inform important policy questions regarding emissions and their health impacts.