Collaborative Research: Understanding Climate Processes with Causal Discovery and Graphs of Information Flow in the Coupled Atmosphere-Land-Ocean System

该项目的目标是开发因果发现的图形模型，并将其应用于气候动力学领域的主题。在这些模型中，滞后相关性被用来识别耦合系统中组件之间的因果影响路径，并且通过将每个组件表示为有向图的一个节点来构建模型。节点之间的边是指定值，表示一个组件在给定另一个组件的先验知识的情况下是可预测的程度。因果发现中的一个关键问题是考虑间接效应，在间接效应中，A组分影响B组分，而B组分又影响C组分，而不是A组分直接影响C组分。用图形的术语来说，区别在于A和C之间是否存在边，或者A与C的连接是否仅通过B。这种评估是通过条件概率的统计计算完成的。PIS以前的工作应用因果发现来确定四种众所周知的遥相关模式之间的关系，这四种模式是西太平洋和东太平洋振荡(WPO和EPO)、北大西洋振荡(NAO)和太平洋/北美(PNA)模式。在这种情况下，该方法表明EPO可以在18天后影响NAO，而NAO可以在3到6天后影响PNA模式。虽然在大气科学中的应用是新的，但该方法已经成功地应用于其他学科，包括经济学、系统工程和生物信息学。该项目下的工作将进一步发展建模技术，目标是通过改进存储器访问的并行化和本地化，将计算速度提高100倍。还将开发可视化工具来显示结果。除了这些开发工作，PIS还将应用这项技术来研究气候动力学中的三个主题：亚洲夏季风的爆发动力学，北半球年模式(NAM)的亚季节变化，以及全球变暖模拟中信息流特征的变化。季风爆发方面的工作将集中在三个问题上：建立跨赤道气流的关键过程是什么？与孟加拉湾和南中国海的局地海气相互作用相比，与青藏高原有关的机械和热力强迫在开始时有多重要？这次爆发是大气不稳定的直接表现吗？对流层动力学、平流层-对流层耦合、对流层-地面耦合在决定NAM相变过程中的相对重要性。关于全球变暖的工作是基于对一个模拟的初步分析，在该模拟中，随着气候变暖，北部中纬度地区的信息流动变得更弱，这表明热带和中纬度地区的大气可预报性降低。通过将一个科学学科(经济学)发展的方法引入另一个科学学科(气候动力学)，这项工作产生了更广泛的影响。私营部门将通过一个门户网站向气候动力学研究人员社区提供他们的算法和软件，从而促进采用建模技术。他们还将在科学会议上提供指导，包括美国气象学会和美国地球物理联合会的年度会议。辅导材料还将在网上和发表的论文中提供。此外，该项目将支持和培训一名研究生和一名博士后研究员，从而为这一研究领域的未来劳动力提供支持。

The goal of this project is to develop graphical models of causal discovery and apply them to topics the field of climate dynamics. In these models lagged correlations are used to identify pathways of causal influence among components of a coupled system, and the model is constructed by representing each component as one node of a directed graph. Edges between node are assigned values representing the extent to which one component is predictable given prior knowledge of another. A key issue in causal discovery is accounting for indirect effects, in which component A influences component B which in turn influences component C, as opposed to the case in which component A directly influences component C. In graphical terms the distinction is whether or not an edge exists between A and C, or whether A's connection to C is only through B. This evaluation is done through a statistical calculation of conditional probabilities. Previous work by the PIs applied causal discovery to identify relationships among four well-known teleconnection patterns, the Western Pacific and Eastern Pacific Oscillations (WPO and EPO), the North Atlantic Oscillation (NAO), and the Pacific/North America (PNA) pattern. In this case the method shows that the EPO can affect the NAO 18 days later, while the NAO can influence the PNA pattern three to six days later. Although the application to atmospheric science is novel, the method has been used successfully in other disciplines including economics, systems engineering, and bioinformatics.Work under the project will further develop the modeling technique, with a goal of increasing computational speed by a factor of 100 through improvements in parallelization and localization of memory access. Visualization tools will also be developed to display results. In addition to these development efforts, the PIs will apply the technique to address three topics in climate dynamics: the onset dynamics of the Asian summer monsoon, the subseasonal variation of the Northern Hemisphere annual mode (NAM), and changes in the characteristics of information flow in global warming simulations. Work on monsoon onset will focus on three questions: What are the critical processes responsible for the establishment of the cross-equatorial flow? How important is the mechanical and thermal forcing associated with the Tibetan Plateau for the onset compared to local air-sea interaction in the Bay of Bengal and South China Sea? Is the onset a direct manifestation of atmospheric instability? For subseasonal NAM variability, the research goal is to investigate the relative importance of tropospheric dynamics, stratosphere-troposphere coupling, and troposphere-surface coupling in determining the phase transition of the NAM. The work on global warming is based on a preliminary analysis of one simulation in which information flows in the Northern midlatitudes become weaker as the climate warms, suggesting a reduction in atmospheric predictability in the tropics and mid-latitudes.The work has broader impacts through the introduction of a methodology developed in one scientific discipline (economics) into another (climate dynamics). The PIs will promote the adoption of the modeling technique by making their algorithms and software available to the community of climate dynamics researchers through a web portal. They will also offer tutorials at scientific conferences including the annual meetings of the American Meteorological Society and the American Geophysical Union. Tutorial materials will also be made available online and through published papers. In addition, the project will support and train a graduate students and a postdoctoral fellow, thereby providing for the future workforce in this research area.