Collaborative Research: Catch the waves - a machine learning approach to map brown dwarf and imaged exoplanet atmospheres in 3D

大气层及其云层对于系外行星和褐矮星如何随着时间冷却至关重要。大气和云是3D的。然而，用于模拟它们的典型技术需要大量的计算机时间。为了理解它们的性质，通常需要将它们简化为一维或二维结构。然而，了解云的三维结构对于理解它们如何影响大气是很重要的。由中佛罗里达大学和加州大学圣克鲁斯分校领导的一个团队将开发一种新方法，减少模拟云和大气所需的计算机时间。这项技术将允许对褐矮星和系外行星的三维大气进行表征。这项技术将有助于研究云如何随着重要的大气特性（年龄、重力、温度和金属丰度）而变化。它还将有助于测试预测这些大气天气的技术。因此，这项工作将增进对大气的了解。这项工作将支持研究生和本科生，并将形成他们的论文的基础。将创建动手活动和视频，以使K-12学生熟悉氛围和编程。这项提议的目标是实现棕矮星和系外行星大气的三维特征。为了做到这一点，将创建一个替代的辐射传输代码，该代码将利用神经网络的力量在当前代码所需的一小部分时间内模拟大气光谱。代理辐射传输代码将用于创建一个3D映射代码，该代码适合贝叶斯框架中的时间分辨观测。代码将在一个独立的、最先进的通用循环模型的输出上进行交叉验证。地面望远镜已经提供了很好的数据，可以绘制这些大气的三维地图，在未来的几十年里，适当数据的数量将大大增加。该代码将应用于现有的观测，以创建棕矮星大气的第一张3D地图。这项工作将为未来几十年的比较气候学创造独特的工具。代理辐射传输代码和3D映射代码将向社区开放源代码。该奖项反映了美国国家科学基金会的法定使命，并通过使用基金会的知识价值和更广泛的影响审查标准进行评估，被认为值得支持。

Atmospheres and their clouds are crucial for how exoplanet and brown dwarfs cool down over time. Atmospheres and clouds are 3D. The typical techniques used to model them though, require a lot of computer time. To understand their properties they generally need to be simplified to 1D or 2D structures. However, knowing the 3D structure of clouds is important to understand how they affect the atmosphere. A team led by the University of Central Florida and the University of California-Santa Cruz will develop a new method that reduces the computer time needed to model clouds and atmospheres. This technique will allow a characterization of brown dwarf and imaged exoplanet atmospheres in 3D. This technique will help the study of how clouds change with important atmospheric properties (age, gravity, temperature and metallicity). It will also help test techniques that predict the weather in these atmospheres. This work will thus improve understanding of atmospheres. This work will support a graduate and an undergraduate student, and will form the basis for their thesis. Hands-on activities and videos will be created to familiarize K-12 students with atmospheres and programming.The goal of this proposal is to enable the 3D characterization of brown dwarf and imaged exoplanet atmospheres. In order to do this a surrogate radiative transfer code will be created that will use the power of neural networks to model spectra of atmospheres in a fraction of the time current codes need. The surrogate radiative transfer code will be used to create a 3D mapping code that fits time resolved observations in a Bayesian framework. The code will be cross-validated on output from an independent, state-of-the-art General Circulation Model. Ground-based telescopes already give excellent data that enable the 3D mapping of these atmospheres and in the next decades the number of appropriate data will increase considerably. The code will be applied on existing observations to create the first 3D maps of brown dwarf atmospheres. The proposed work will create tools that will uniquely enable comparative climatology in the next decades. The surrogate radiative transfer code and the 3D mapping code will be provided open source to the community.This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.