Use of Artificial Intelligence to understand mountain weather and climate processes

利用人工智能了解山区天气和气候过程

• 批准号: 2442866
• 金额: --
• 依托单位: University of Leeds
• 依托单位国家: 英国
• 项目类别: Studentship
• 财政年份: 2020
• 资助国家: 英国
• 起止时间: 2020 至 无数据
• 项目状态: 已结题
• 来源: https://gtr.ukri.org/projects?ref=studentship-2442866
• 关键词: Use; Artificial; Intelligence; understand; mountain

This project will explore the potential of new data science approaches such as machine learning (ML) to develop understanding of complex flows over orography (hills and mountains) and improve their representation in models.The flow in mountainous regions is complex and poorly represented in global and weather and climate models. These flows have important effects on both the local weather and also by influencing the larger scale flow. For example, drainage currents affect the local conditions in mountain valleys, but also influence the turbulent exchange of heat, moisture and pollutants between the boundary layer and the free atmosphere across the larger mountain-range scales. The turbulence associated with breaking mountain waves is a dangerous hazard to aviation, but the process of wave breaking also imparts a drag force on the atmospheric flow, which affects the circulation on global scales.Current approaches (parametrizations) to representing these processes in models are based on simple theoretical concepts which are derived from highly idealised problems. Further advances are hampered by a lack of a theoretical framework to account for the complexity of the processes. Recent advances in Artificial Intelligence (AI) and Machine Learning (ML) provide an exciting opportunity to make further progress. This is an area which is now being embraced by the atmospheric science community. A number of recent papers have begun to investigate the potential of Machine Learning approaches to the problem of parametrising convection, but currently there has been little work on parametrising the effects of orographic processes. The student will use cutting edge high-resolution model simulations (e.g. of flows over UK, the Rockies and Himalayas) combined with observations obtained in field experiments and new machine learning techniques to develop new physical understanding of orographic flow phenomena and to derive novel methods to represent their effects in weather and climate models. A range of phenomena will be considered, starting with those whose effects are known to be important but the complexity is such that they are not currently represented in weather and climate models (e.g. mountain lee-waves). Other phenomena which could be considered include drainage flows, valley cold pools, fog, turbulent rotors and orographic enhancement of precipitation. The results of the project will then be used to assess the potential for new forecasting techniques and new parametrizations.Key questions might include:- How can AI techniques be used to synthesise and classify large atmospheric data sets (either model or observational)?- Which Machine Learning approaches are most suitable for representing orographic processes?- How well do ML techniques compare to traditional downscaling techniques or high resolution models for providing detailed forecasts in mountain areas (e.g. local temperature and fog forecasts or orographic rainfall)?- How well do ML techniques compare to physically-based parametrisations or high resolution models in predicting the bulk effects of orography on the large scale e.g. the orographic drag exerted on the atmosphere).- Which inputs (both type and quantity of data) provide the best and most efficient predictors in each case?- How well do networks trained in one area of the world predict in other geographic areas?The fusion of data science with simulation is a key theme of the 2020-2030 Met Office Science Strategy and the student will benefit from the growing use of machine learning techniques at the Met Office, plus their world-leading expertise in weather and climate simulation. The student will be based in the School of Earth and Environment at the University of Leeds, which is also growing activity in this field, with strong links to the School of Computing and the Leeds Institute for Data Analytics.

该项目将探索机器学习（ML）等新的数据科学方法的潜力，以发展对地形（丘陵和山脉）上复杂流动的理解，并改善其在模型中的表示。山区的流动很复杂，在全球和天气气候模型中表现不佳。这些气流对当地天气和大尺度气流都有重要影响。例如，排泄流影响山谷的当地条件，但也影响较大山脉范围内边界层与自由大气之间热量、水分和污染物的湍流交换。与破碎的山波相关联的湍流对航空是一种危险的危害，但是波破碎的过程也会对大气流动产生拖曳力，从而影响全球尺度上的环流。目前在模型中表示这些过程的方法（参数化）是基于来自高度理想化问题的简单理论概念。由于缺乏理论框架来解释这些过程的复杂性，进一步的进展受到阻碍。人工智能（AI）和机器学习（ML）的最新进展为取得进一步进展提供了令人兴奋的机会。这是一个现在被大气科学界所接受的领域。最近的一些论文已经开始研究机器学习方法在参数化对流问题上的潜力，但目前在参数化地形过程影响方面的工作很少。学生将使用尖端的高分辨率模型模拟（例如英国，落基山脉和喜马拉雅山脉的流量），结合实地实验和新的机器学习技术中获得的观察结果，以开发对地形流现象的新物理理解，并获得新的方法来表示其在天气和气候模型中的影响。将考虑一系列现象，首先是那些已知影响重要但复杂性如此之大，目前无法在天气和气候模型中表示的现象（例如山区背风波）。可以考虑的其他现象包括排水流、山谷冷池、雾、湍流转子和地形对降水的增强。该项目的结果将用于评估新的预测技术和新的参数化的潜力。关键问题可能包括：-人工智能技术如何用于合成和分类大型大气数据集（模型或观测）？-哪些机器学习方法最适合表示地形过程？与传统的降尺度技术或高分辨率模型相比，ML技术在山区提供详细预报（例如当地温度和雾预报或地形降雨）方面表现如何？ML技术与基于物理的参数化或高分辨率模型相比，在预测大规模地形的整体影响（例如地形对大气的阻力）方面有多好。在每种情况下，哪些输入（数据的类型和数量）提供了最好和最有效的预测因子？在世界某个地区训练的网络在其他地理区域的预测效果如何？数据科学与模拟的融合是2020-2030年气象局科学战略的一个关键主题，学生将受益于气象局越来越多地使用机器学习技术，以及他们在天气和气候模拟方面的世界领先的专业知识。该学生将在利兹大学的地球与环境学院工作，该学院在这一领域的活动也越来越多，与计算学院和利兹数据分析研究所有着密切的联系。