Data Science in Giant Planet Magnetospheres

每一颗巨大的行星都被行星磁层包围，其中包含一个复杂的相互作用网络，包括行星的中性和电离大气、光环系统和自然卫星、尘埃、中性气体、等离子体和辐射带，所有这些都嵌入在超音速太阳风中。揭开这些元素是如何相互作用的，以及在极大(百万公里)长的尺度上是什么物理过程在起作用，这是一个艰巨的挑战，过去40年来一直在研究，从先驱者10号飞越木星开始。理解这些系统的挑战包括：处理和关联数百GB的不同类型数据集；说明采样、分辨率和其他仪器偏差；以及推断航天器轨迹有限的大型系统中的进程状态。在快速发展的数据科学领域，这些都是每天都在解决的典型问题，这些问题可能会使我们研究行星磁层的方式发生一场革命。在这个项目中，学生将使用整个卡西尼号任务的数据来研究“土星的磁层”。特别是，将结合机器学习和互信息理论，利用现场测量(等离子体、高能粒子、磁场)和遥感测量(高能中性原子、极光)之间的关系来推断磁层的状态和工作中的物理过程。特别令人感兴趣的是，如何通过使用互信息理论将多个数据集与太阳风传播模型进行比较来确定太阳风活动的特征。具体地说，血浆数据中有哪些天然替代品可以告知系统的大小、规模和行为。项目目标1.使用机器学习从卡西尼号宇宙飞船的现场和等离子体测量中自动识别结构和事件。严格检查机器学习方法对自动磁层数据挖掘的有效性。应用互信息理论的方法来解释观测结果，并批判性地检验巨行星磁层动力学的物理驱动模型。

Each of the giant planets is surrounded by a planetary magnetosphere which contains a complex web of interacting elements, from a planet's neutral and ionised atmosphere, ring systems and natural satellites, populations of dust, neutral gas, plasma, and radiation belts, all embedded within the supersonic solar wind. Unravelling how these elements interact and what physical processes are at work over extremely large (million km) length scales is a formidable challenge and has been studied for the last four decades, starting with the Pioneer 10 flyby of Jupiter. The challenges of understanding these systems include processing and relating 100s of GB of heterogeneous datasets; accounting for sampling, resolution and other instrumental biases; and inferring the state of processes in large-scale systems with limited spacecraft trajectories. These are typical problems that are solved every day in the rapidly growing field of Data Science and which may enable a revolution in how we study planetary magnetospheres.In this project, the student will study 'Saturn's magnetosphere' using data from the entire Cassini mission. In particular, relationships between in situ measurements (plasma, energetic particles, magnetic fields) and remote-sensing measurements (energetic neutral atoms, aurorae) will be used to infer the state of the magnetosphere, and the physical processes at work, using a combination of machine learning and mutual information theory. Of particular interest is how the signatures of solar wind activity can be determined through comparing multiple data sets with solar wind propogation models using mutual information theory. In particular, what natural proxies are available in plasma data that can inform of of the systems size, scale, and behavior. Project goals1. Use machine learning to automatically recognise structures and events from in situ field and plasma measurements from the Cassini spacecraft.2. Critically examine the efficacy of machine learning methods for automated magnetospheric data mining.3. Apply methods from mutual information theory to interpret observations and critically examine physically-motivated models for dynamics in giant planet magnetospheres.