Sunspot Emergence Prediction: Machine Learning Based Space Weather Assessment

Singh, Vishal (2021) Sunspot Emergence Prediction: Machine Learning Based Space Weather Assessment. Masters thesis, Indian Institute of Science Education and Research Kolkata.

Text (MS Dissertation of Vishal Singh (16MS007)) 16MS007_Thesis_file.pdf - Submitted Version Restricted to Repository staff only Download (2MB)

Abstract

There is a growing recognition that the environmental conditions we refer to as "space weather" have an impact on the technical infrastructure that drives the world’s interconnected economies. As a result, better forecasts, environmental standards, and infrastructure design are required to better protect society from space weather. With research observatories on the ground and in space, significant progress has been done and continues to be made. However, the space weather domain is huge, spanning from deep within the Sun to well beyond planetary orbits. With the disruptive potential of solar eruptive phenomena like solar flares and coronal mass ejections, there is a strong need for improving our capabilities in being able to predict and prepare for such events. During these events, radiation and mass ejections can cause geomagnetic storms, which can disrupt our satellites and communication systems. Being aware of these episodes beforehand allows us to be ready to deal with their effects. In this context, Machine Learning is a newly emerging valuable tool to effectively and accurately make these predictions. While substantial work has been done in predicting solar flares using machine learning based data driven models, the field of predicting sunspots remains relatively underdeveloped. It is currently not feasible to anticipate their occurrence using a physical model, and one observational prescription has not proven to be reliable. In this study we have identified issues in previously done machine learning based studies and attempt to find possible solutions. We primarily use Dopplergams time series as the source of data for our models. Machine Learning provides with a large number of options to choose from. For this study, we work with Convolutional Neural Networks, a special kind of neural network designed for dealing with images. The dopplergram time series is used to extract relevant snapshots and used in the aforementioned models.

Actions (login required)

View Item