CAREER: End-to-End Active Region-based Heliospheric Forecasting System Using Multi-spacecraft Data and Machine Learning

太阳耀斑是最具影响力的太阳爆发活动之一。它们是由于磁重联，高度波动的磁场坍缩到较低的能量状态，向太空释放能量。这个项目的重点是通过在太阳观测中使用机器学习技术来预测太阳耀斑。更广泛的影响包括通过在犹他州公共广播电台组织每年一次的空间天气小组讨论来影响犹他州从中学到大学各个教育层次的学生。每年夏天，美国土著学生的研究经验将得到资助。此外，PI是一位早期职业女性科学家，她将为她的应用数据挖掘课程开发一个新的应用实验室组件。这项研究有可能通过减少与太空天气事件有关的国防和太空探索任务的潜在风险来增加国家安全和美国的竞争力。从太阳光球矢量磁图中提取的活动区磁场参数通常用于预测太阳耀斑的发生。尽管最近在太阳耀斑预测方面取得了进展，但要有效地结合不同的星载仪器的观测数据，跨越多个太阳周期，以训练稳健和无偏的太阳耀斑模型，仍然存在重大障碍。该项目是一个为期五年的研究项目，旨在利用最先进的机器学习模型来发现不同太阳周期中极端太阳耀斑的驱动因素，评估活跃区域对太阳瞬变事件的影响，并将所学知识转移到日球层的其他模型中。为了实现这一愿景，该项目将开发一个高空间分辨率的活动区域矢量磁图数据集，该数据集跨越两个太阳周期，基于NASA太阳动力学观测站、太阳和日光层观测站和日出（推力1）上的三个磁图。新的高质量和高分辨率磁场图将允许研究在太阳耀斑预测背景下从未检查过的小范围活动区域的物理特征。该项目将生成全面的磁场参数多变量时间序列（MVTS）数据集，这对数据科学和空间气象社区建模各种太阳现象都很有用（推力2）。最后，该项目将建立一个准确、稳健的太阳耀斑预测模型，并使用学习到的预测模式初始化其他太阳事件预测模型（推力3）。这项CAREER提案的最终目标是利用天体物理学领域的跨领域应用ML方法来提高我们对驱动不同类型太阳耀斑的活动区域物理属性的理解，并使科学家能够对太阳耀斑事件的预测进行比较，可重复和数据驱动的研究。这项研究的副产品之一将是一个前所未有的全面的太阳耀斑目录，补充了母活动区域的磁场参数多变量时间序列数据，这些数据将通过应用程序编程接口（API）免费提供，以实现其广泛的潜在用途（例如，进行统计研究，训练基于ml的模型和基于物理的模型）。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。

Solar flares are one of the most impactful solar eruptive activities that occur. They are due to magnetic reconnection, where highly fluctuating magnetic fields collapse to a lower energy state releasing energy into space. This project focuses on prediction of solar flares through using machine learning techniques on solar observations. The broader impacts include public outreach to impact students at all educational levels in Utah from middle-school to college level by organizing a yearly space-weather panel discussion on Utah Public Radio. Indigenous American student research experiences will be funded every summer. In addition, the PI, an early career woman scientist, will develop a new applied laboratory component to her applied data mining classes. The research has the potential to increase national security and US competitiveness by lessening the potential risk related to defense and space exploration missions from space weather events. Active region magnetic field parameters, extracted from solar photospheric vector magnetograms, have been routinely used to predict solar flare occurrences. Despite recent advancements in solar flare prediction, there are significant barriers to efficiently combine different space-borne instruments’ observations spanning multiple solar cycles, to train robust and unbiased solar flare models. This project is a five-year research program that aims at leveraging state-of-the-art machine learning models to discover the driving factors of extreme solar flares in different solar cycles, assess the impact of active regions’ properties on solar transient events, and transfer the learned knowledge to other models of the heliosphere. To achieve the vision, the project will develop a high-spatial resolution active regions vector magnetogram dataset, that spans two solar cycles, based on three magnetographs on-board NASA’s Solar Dynamics Observatory, Solar and Heliospheric Observatory and Hinode (Thrust 1). The new high- quality and high-resolution magnetic field maps will allow the study of small-scale active regions’ physical characteristics that were never examined in the context of solar flare prediction. The project will generate comprehensive magnetic field parameters multivariate time series (MVTS) dataset useful to both Data Science and Space Weather communities for modeling various solar phenomena (Thrust 2). Finally, the project will build an accurate and robust solar flare prediction model and use the learned predictive patterns to initialize other solar events predictive models (Thrust 3). The end goal of this CAREER proposal, is to leverage the cross-field of applied ML methods in the field of astrophysics to improve our understanding of the physical attributes of active regions that drive different types of solar flares, and enable scientists to perform comparative, reproducible, and data-driven studies on the prediction of solar flare events. One of the by-products of this research will be an unprecedented comprehensive solar flare catalog supplemented with parent active regions’ magnetic field parameters’ multivariate time series data that will be freely available through Application Programming Interface (API) for its wide potential usage (e.g., conduct statistical studies, train ML-based and physics-based models).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.