| Popis: |
In this thesis we explore methods for analysing flaring active regions, in particular studying X-class flares. We use global magnetic field models and machine learning techniques to carry out this analysis. Using both potential field source surface (PFSS) models and magnetohydrostatic (MHS) models, the global magnetic skeletons for dates where X-class flares occurred are created. This allows the investigation of topological features found around flaring active regions. The flares analysed all have observable signatures found in Atmospheric Imaging Assembly data (onboard the Solar Dynamics Observatory), in the form of solar flare ribbons which can be mapped by eye to the footpoints of the separatrix structures located in the active regions. Additionally, we consider techniques for identifying and locating the solar flare ribbons observed. The first technique utilises a convolutional neural network trained using images of M and C-class flares to allow the detection and classification of the types of flare ribbons observed. This includes two-ribbon, compact and limb flares, as well as quiet sun images. After training the network and identifying the flare ribbons in the data, we present an edge detection method which identifies the edges of the flare ribbons, making it easier to compare with the topological features previously found in the global field models. To find the best edge, two methods are presented which correct saturated pixels in the flare ribbon data. Afterwards the corrected images are passed into the edge detector which returns the ribbon edges, which are subsequently compared to the topological features previously found by calculating the Hausdorff and modified Hausdorff distances. Overall these methods could be put into an automated pipeline which would identify solar flare ribbons in the observations using a CNN, then subsequently creating 3D magnetic field models to investigate the topology around the flare. With the final step taking both the observational and modelled data to be processed by the edge detection method and subsequently outputting a metric which identifies whether they are related. Note however this pipeline was not created in this thesis. |
