ECE MS Thesis Defense: Kshitij Duraphe

ECE MS Thesis Defense: Kshitij Duraphe

Advisor: Professor Joshua Semeter

Committee: Professor Toshi Nishimura and Professor Michael Hirsch

Title: 'Data Driven Techniques to Advance our Understanding of the STEVE Phenomenon

Abstract: Strong Thermal Emission Velocity Enhancement (STEVE) is an optical phenomenon in Earth’s ionosphere that has recently captured the attention of both citizen and professional scientists. STEVE appears as a mauve-white elongated band stretching for a few thousand kilometers across the night sky during the recovery phase of an auroral substorm. It is sometimes accompanied by quasiperiodic collimated green rays known as the ‘picket fence’. The picket fence has been observed to be localized to a section of the STEVE.

STEVE has been observed to occur during the recovery phase of a substorm, which is an abrupt global-scale change in Earth’s magnetotail that releases energy stored in the nightside magnetosphere into two nightside polar ionospheres via field-aligned currents and particle precipitation.

This thesis employs data fusion and image processing methodologies from a remote sensing perspective to advance our understanding of the STEVE phenomenon. Additionally, it offers insights into the application of deep learning techniques in this field.

This thesis is divided into three parts. In the first part, we examine the morphology and velocity of features based on citizen science observations of STEVE. High-resolution imaging of the complete life cycle of a STEVE with a picket fence reveals fine-scale white pillar-like structures seemingly accelerating as they move westward, with an average velocity of ~13km/s. Individual pickets of the picket fence move much slower, at speeds of ~1km/s. This provides weak evidence for acceleration of features in STEVE, though we are limited by the geometry of the system.

The second part examines Total Electron Content (TEC) variations during a STEVE. TEC is a measurement of the ionospheric activity and has not been examined for a STEVE in the literature yet, largely due to a lack of Global Navigation Satellite System (GNSS) receivers in locations where a STEVE exists. We observe high-resolution (15-second) changes in TEC during a STEVE event on the 26 March 2008, which occurred over a region with a large amount of GNSS receivers. We observe a sudden jump in TEC as the STEVE propagates towards the equator and another jump as the auroral oval expands. For another substorm in the same region on 1 March 2011, we do not see this two-stage jump.

The third part examines the feasibility of detecting a STEVE as it occurs, based on data from ASIs. We adapt and enhance an algorithm originally designed for identifying faint sources in astronomical images to detect STEVE phenomena within faint and noisy All-Sky Imager (ASI) images. We then examine the performance of the ConvNeXt computer vision model in classifying ASI images with and without STEVE. We finally discuss the difficulty in applying computer vision models for the detection of this phenomenon, which is largely due to the low occurrence rate, and give some insight into strategies such as active learning and open set recognition which can potentially be used to identify STEVE in different ASI data.