Exploiting machine-learning to provide dynamical, microphysical, radiative and electrifying insight from observations of deep convective cloud

平衡气候敏感性（即二氧化碳增加一倍导致的变暖）是评估二氧化碳排放风险的基本指标。然而，40年来，气候敏感性的可信值一直不确定，其中云反馈是一个特别不确定的组成部分（Sherwood et al., 2020）。由深层对流产生的热带高云（如砧云）是一种重要的云类型，当涉及到反馈时。IPCC评估报告6最近评估了热带高云量的负反馈（Forster et al., 2021）。然而，由于缺乏对微物理和湍流对变暖的反应的理解，这一结论的可信度很低。云冰微物理学的参数化特别差。在2022年7月至8月，DCMEX活动成功地收集了大量关于新墨西哥州马格达莱纳山脉上空对流云发展的观测数据。FAAM的BAe-146飞机测量了云内的云微物理和动力学，同时多普勒雷达和自动摄像机从附近监测云的发展。气溶胶测量数据，包括冰核粒子（INP），是在飞机上和位于山脉顶峰的朗缪尔实验室收集的。现在可以结合卫星数据和最近开发的气象局统一模型CASIM微物理方案建模来分析这个广泛的数据集。总的来说，这些数据将通过改善全球气候模式中微物理过程的表征来支持减少气候敏感性的不确定性。已经获得了气溶胶的详细观测和云冰粒子的图像，其中包含了非常复杂的信息。气溶胶的特性、冰粒的大小和形状以及它们在云中的位置以不同的方式改变微物理过程和云辐射效应（Voigtländer et al., 2018; Gasparini et al., 2019; Diedenhoven et al., 2020）。气溶胶和冰过程的复杂性，以及云的响应，证明了使用新的分析技术（如机器学习）来获得洞察力是合理的。博士候选人将通过探索一系列分析技术来建立正在进行的DCMEX研究。最初的重点将放在从监督和无监督机器学习技术中可以学到什么，这些技术正迅速成为云和气候研究的关键工具（Gagne等人，2017；Beucler等人，2021；Kashinath等人，2021；Gettelman等人，2021）。最初，这项工作将侧重于了解冰粒子的形成过程，以支持UM-CASIM模型的发展。然后，研究将扩展到将动态微物理过程与辐射的影响以及风暴的电气化联系起来。通过将分析扩展到考虑电气化，我们建立了对全球测量变量闪电的理解，闪电是我们感兴趣的深层对流风暴的一个关键特征。考虑到在DCMEX研究区域运行的详细卫星（GOES GLM）和地面闪电探测网络，这是一个不容错过的机会。本文将解决以下研究问题：1)云内冰图像是否可以分类并与其形成动力学相关？inp在其中扮演什么角色？2)机器学习是否能够利用观测约束UM-CASIM模型参数和过程？3)云动力学和微物理如何影响深层对流砧处的辐射？4)闪电活动与云的微物理和动力学有什么关系？5)闪电可以作为云过程的一个指标，导致砧辐射特性的变化吗？

The equilibrium climate sensitivity (i.e. the warming from a doubling of CO2) is a fundamental metric for assessing the risks arising from CO2 emissions. Yet the plausible values of climate sensitivity have remained stubbornly uncertain for 40 years, with cloud feedbacks a particularly uncertain component (Sherwood et al., 2020). Tropical high cloud (e.g. anvils), produced by deep convection, is an important cloud type when it comes to feedbacks. The IPCC Assessment Report 6 recently assessed there to be a negative feedback from tropical high cloud amount (Forster et al., 2021). This, however, came with low confidence that arises, in part, from the lack of understanding of the response of microphysics and turbulence to warming. Cloud ice microphysics is particularly poorly parametrised. In July-August 2022, the DCMEX campaign successfully collected a vast set of observations of developing convective clouds over the Magdalena Mountains, New Mexico. The FAAM BAe-146 aircraft measured cloud microphysics and dynamics within the clouds whilst Doppler radars and automated cameras monitored the development of the clouds from nearby. Aerosol measurements, including of Ice Nucleating Particles (INP), were collected on the aircraft and at Langmuir Laboratory on the summit of the mountain range. This extensive dataset can now be analysed in combination with satellite data and modelling with the recently developed Met Office Unified Model CASIM microphysics scheme. Altogether, the data will support the reduction of climate sensitivity uncertainty by improving the representation of microphysical processes in global climate models.Detailed observations of aerosol and imagery of cloud ice particles have been obtained which contain a great complexity of information. The characteristics of aerosol, and the size and shape of ice particles, as well as their position within the cloud, modify the microphysical processes and cloud radiative effect in different ways (Voigtländer et al., 2018; Gasparini et al., 2019; Diedenhoven et al, 2020). The complexity of the aerosol and ice processes, and the cloud response, justifies using novel analysis techniques, such as machine learning, to gain insight. The PhD candidate will build upon ongoing DCMEX research by exploring a range of analysis techniques. An initial focus will be on what can be learned from both supervised and unsupervised machine learning techniques, which are fast becoming key tools in the study of clouds and climate (Gagne et al., 2017; Beucler et al., 2021; Kashinath et al, 2021;,Gettelman et al., 2021). Initially, the work will focus on understanding the ice particle formation processes to support the development of the UM-CASIM model. The research will then expand to relate the dynamical-microphysical processes to impacts on radiation, and also electrification of storms. By extending the analysis to consider electrification, we build understanding of a globally measured variable, lightning, which is a key feature of the deep convective storms of interest. This is an opportunity not to be missed given the detailed satellite (GOES GLM) and ground-based lightning detection networks in operation over the DCMEX study region. The following research questions will be addressed:1) Can in-cloud ice images be categorised and related to their formation dynamics? And what role do INPs play in this?2) Is machine-learning able to constrain UM-CASIM model parameters and processes using observations?3) How is the radiation at the deep convective anvil affected by cloud dynamics and microphysics?4) What is the relationship between lightning activity and cloud microphysics and dynamics?5) Can lightning be used as an indicator of cloud processes that result in variations of anvil radiative properties?