Collaborative Research: Frameworks: Community-Based Weather and Climate Simulation With a Global Storm-Resolving Model

全球地球系统模型（ESM）使用数学方程来模拟天气和气候。ESM包括大气、海洋、陆地表面、冰和植被的动态。它们可以用来预测公众和政策制定者的使用情况。今天的ESM使用粗网格，网格宽约100公里。像雷暴这样的重要天气系统太小了，无法用这种网格来模拟。改进ESM的一种方法是使用可以直接模拟雷暴的更精细的网格，但这种模型只能在非常强大的计算机上运行。该项目名为EarthWorks，将利用高性能计算的最新发展，创建一个能够解决风暴的ESM。EarthWorks还将使用人工智能来改进和加速模型，并使用最先进的方法来限制模型运行时产生的数据量。EarthWorks ESM将通过剥离和修改广泛使用的社区地球系统模型的最新版本来构建。修改后的模型将在一个非常高分辨率的网格上代表大气、海洋和陆地表面，网格单元宽约4公里。它将提高预报技能，并对过去、现在和未来的气候进行更逼真的模拟。该项目将使该模型及其输出公开提供给所有科学家使用，开放源代码社区地球系统模型（CESM）是由一个庞大的研究社区开发和应用于科学问题。它是美国气候研究界的关键基础设施。在CESM的大气和海洋组成部分中，表示质量守恒、动量守恒和热力学能量守恒的偏微分方程的绝热项使用所谓的动力学核心进行数值求解。大气和海洋模式还包括参数化表示，称为参数化，旨在包括风暴和云过程的影响，这些影响发生在尺度太小而无法在模型的网格上表示。尽管许多科学家进行了数十年的工作，但今天的参数化仍然存在问题，并限制了ESM在许多社会相关应用中的效用。幸运的是，计算机能力的最新进展使得无参数化成为可能，因为在整个地球仪上使用了几公里数量级的网格间距。这些“全球风暴解析模型”（GSRM）只能在当今最快的计算机上运行。GSRM在世界各地的十几个建模中心正在积极开发中。然而，不幸的是，CESM目前的制定方式使其无法作为GSRM运行。这个项目称为EarthWorks，将通过剥离和密集修改CESM的副本来创建一个新的、公开可用的GSRM。为了实现这一目标，研究人员将使用最近开发的与大气和海洋密切相关的动力核心。模型的所有组件都将使用相同的高分辨率网格。这种高分辨率将使消除深积云对流的特别麻烦的参数化成为可能（即，雷暴），从而减少困扰当前ESM的系统性偏差。Earthworks将利用目前由高性能计算供应商推向市场的pre-exascale和exascale技术。新的exascale ESM将在强大的图形处理器单元（GPU）上运行计算密集型组件，并利用节点级任务并行性来异步执行模型的其余部分。组件模型代码已接近完成，目前正在GPU上进行测试。EarthWorks将使用简化的组件耦合方法，在可行的情况下结合机器学习，并利用有损压缩技术和并行I/O工具来处理模型运行时生成的大量数据。完成的模型将是简单的，强大的，并有良好的文档记录。该项目将把它应用于数值天气预报和气候模拟方面的紧迫科学问题。该模型及其输入数据集将通过GitHub向广大研究界公开提供。该奖项反映了NSF的法定使命，并且通过使用基金会的知识价值和更广泛的影响审查标准进行评估，被认为值得支持。

Global Earth System Models (ESMs) use mathematical equations to simulate both weather and climate. ESMs include the dynamics of the atmosphere, oceans, land surface, ice, and vegetation. They can be used to make predictions of use to the public and policymakers. Today’s ESMs use coarse grids with cells about 100 km wide. Important weather systems like thunderstorms are too small to be simulated with such grids. One way to improve ESMs is to use finer grids that can directly simulate thunderstorms, but such models can only be run on very powerful computers. This project, called EarthWorks, will create an ESM capable of resolving storms by taking advantage of recent developments in high performance computing. EarthWorks will also use artificial intelligence to improve and speed up the model, and state-of-the-art methods to limit the amount of data produced as the model runs. The EarthWorks ESM will be built by spinning off and modifying a copy of the most recent version of the widely used Community Earth System Model. The modified model will represent the atmosphere, the oceans, and the land surface on a single very high-resolution grid, with grid cells about 4 km wide. It will have improved forecast skill, and produce more realistic simulations of past, present, and future climates. The project will make the model and its output openly available for use by all scientists.The open-source Community Earth System Model (CESM) is both developed and applied to scientific problems by a large community of researchers. It is critical infrastructure for the U.S. climate research community. In the atmosphere and ocean components of the CESM, the adiabatic terms of the partial differential equations that express conservation of mass, momentum, and thermodynamic energy are solved numerically using what is called a dynamical core. Atmosphere and ocean models also include parametric representations, called parameterizations, that are designed to include the effects of storm and cloud processes that occur on scales too small to be represented on the model's grid. Despite decades of work by many scientists, today's parameterizations are still problematic and limit the utility of ESMs for many applications of societal relevance. Fortunately, recent advances in computer power have made it possible to parameterize less, by using grid spacings on the order of a few kilometers over the entire globe. These "global storm-resolving models" (GSRMs) can only be run on today's fastest computers. GSRMs are under very active development at a dozen or so modeling centers around the world. Unfortunately, however, the current formulation of the CESM prevents it from being run as a GSRM. This project, called EarthWorks, will create a new, openly available GSRM by spinning off and intensively modifying a copy of the CESM. To accomplish this goal, the researchers will use recently developed and closely related dynamical cores for the atmosphere and ocean. All components of the model will use the same very high-resolution grid. This high resolution will make it possible to eliminate the particularly troublesome parameterization of deep cumulus convection (i.e., thunderstorms), and thereby reduce systematic biases that plague current ESMs. Earthworks will exploit the pre-exascale and exascale technologies now being brought to market by high performance computing vendors. The new exascale ESM will run the most computationally intensive components on powerful graphics processor units (GPUs), and exploit node-level task parallelism to execute the rest of the model asynchronously. The component model codes are close to completion and are currently being tested on GPUs. EarthWorks will use a simplified component-coupling approach, incorporate machine learning where feasible, and leverage lossy compression techniques and parallel I/O tools to deal with the enormous data volumes that will be generated as the model runs. The completed model will be simple, powerful, and well documented. The project will apply it to pressing scientific problems in both numerical weather prediction and climate simulation. The model and its input datasets will be made openly available to the broad research community, via GitHub.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.