A Deep (Learning) Dive into Solar Active Region Evolution and Flare Production

点击翻译按钮获取中文摘要

Solar flares, alongside coronal mass ejections (CMEs), are major contributors to space weather - changing conditions in the near-Earth space, magnetosphere and Earth's upper atmosphere. Flares mostly occur in active regions (ARs); volumes of the solar atmosphere defined by the magnetic field. Plasma flows move magnetic field around and, after enough energy accumulates and conditions are suitable, ARs release stored energy as flares/CMEs. However, the conditions required to initiate flares/CMEs are unclear, limiting our ability to forecast them.Recently, we built an infrastructure (FLARECAST) to explore many AR properties via machine-learning (ML) forecasting. However, we found that most information used by the ML methods is contained in a small number of AR properties. This is due to significant information redundancy since most AR properties were calculated from the same magnetic-field images. This is amplified by several AR properties aiming to quantify essentially the same (not directly observable) physical characteristics (e.g., proxies for free magnetic energy). Whole-image data has only recently begun being explored for forecasting via Deep Learning (DL), which should show improvement over previous ML methods that depend crucially on subjective choices of what AR properties to extract from magnetic-field images; DL explores all information in each magnetic-field image and its relation to supervised labels of flaring/non-flaring.In this project you will use vectormagnetic field observations understand the evolution of AR magnetic fields and their relation to flare occurrence for a large statistical sample, encompassing flare-quiet ARs to those with flaring activity of high frequency and magnitude. You will seek to understand the physics leading to flares with the ultimate aim of improving our capacity to forecast them. You will have access to state-of-the-art ML/DL methods and will apply these in large-scale processing of 10s-100s TB of data. This PhD concerns the magnetic conditions that power/initiate flares/CMEs, with the potential to impact on the quality/timeliness of forecasting adverse space-weather conditions and increasing the operational capacity of space-weather forecast centres (e.g., Met Office). As the project develops, you will have the opportunity to collaborate with national/international colleagues in the space-weather forecasting community.