基于数值模拟与深度学习的大气湍流诊断方法研究

Atmospheric turbulence is a kind of complicated and multi-scaled atmospheric motion which produces deep influence to mankind in living and work. Traditionally, atmospheric turbulence monitoring is mainly performed with numerical weather prediction model and diagnosis method due to the lack of effective real-time and large-scale observations, and it has some disadvantages of the uncertainties. In recent years, feature recognition for weather system through deep learning and satellite image has been applied, meanwhile the environment favorable to the generation of air turbulence is strongly associated with weather system. Thus, this project will combine the aircraft observation, numerical weather prediction and diagnosis method to derive the quantitative relationship between air turbulence and potential environment. Meanwhile, this project will combine the advantages of numerical weather prediction in atmospheric physics and deep learning technology in feature recognition to reveal the correlation between potential environment and weather system. This study is expected to provide a technology path for real-time monitoring with better understanding of formation mechanism for air turbulence.

大气湍流是一种复杂的尺度多变的大气运动形式，对人类工作生活有重要的影响。由于缺乏有效的实时大范围观测手段，现有的大气湍流监测方法主要基于数值天气预报模式与潜势诊断进行的，存在较大的不确定性。近年来，通过机器学习在卫星云图上对天气系统进行特征识别的方法逐渐被应用，而大气湍流发生的潜势环境与天气系统有密切的关系。因此，本申请项目将利用数值天气预报模式，结合飞机观测研究大气湍流发生与潜势环境之间的定量关系；同时融合模式在动热力场描述和深度学习在特征识别方面的技术优势，揭示大气湍流潜势环境与天气系统之间的联系。本申请项目的开展将加深对大气湍流发生机制的认知，为实现大气湍流的实时监测提供切实可行的技术途径。

大气湍流是一种复杂的尺度多变的大气运动形式，对空中航行安全有重要的影响。由于缺乏有效的实时大范围观测手段，现有的大气湍流监测方法主要是基于数值天气预报模式与诊断进行的，可靠性较低。本项目建立了2019-2022年的飞机原位湍流观测数据集，综合采用飞机与卫星遥感等观测、数值模拟和动力学诊断等方法，对大气湍流发生的潜势环境开展研究，揭示了中国区域大气湍流发生的主要规律。研究了基于深度学习的大气湍流诊断方法，建立了大气湍流诊断模型。项目所形成的成果有效丰富了中国区域大气湍流的观测数据，给出了中国区域大气湍流的发生规律及主要影响因素，为提升我国大气湍流的预报准确率提供了切实可行的技术途径。