Publicação
Forecasting the Impact of Space Weather in Satellite Orbits
| Resumo: | Satellites are a critical enabler of everyday technologies. From communications to GPS, from meteorology to the Internet, satellites allow modernity as we know it. Although they often go unnoticed by most people, satellites face constant challenges, such as Space Weather (SW). It comprises events that proceed mainly from the Sun and propagate through space. These events may disrupt satellite operations not only by altering their position but also by interfering with communication and onboard instruments. Such a disturbance may disrupt Earth systems as well. The most dangerous SW phenomenon for satellites is a Coronal Mass Ejection (CME). This event can alter the position of a satellite by hundreds of meters. If a large-scale CME hits Earth, it may cause planet-wide electricity and communications blackouts, which could be disastrous. Over the past several decades, researchers have attempted to forecast CMEs with some success, though many challenges remain, including limited accuracy and timeliness of existing predictions. Recent advances in Machine Learning (ML) in the field of space sciences provide promising avenues to improve CME forecasts. These methods can model the complex patterns of space data to provide better results than traditional methods. This dissertation aims to evaluate the viability of ML approaches for predicting CMEs by training models on rich solar wind data collected from the SOHO spacecraft, whose mission is to study the Sun. An extensive and carefully constructed dataset is used, featuring multiple solar wind parameters over several years. I thoroughly evaluate a range of ML algorithms along with preprocessing techniques and data splitting strategies suited for this problem. The results indicate that some models can achieve balanced accuracies above 70%, demonstrating meaningful predictive potential. This work contributes novel insights into the application of ML to Space Weather prediction and provides a robust dataset for the scientific community to use. |
|---|---|
| Autores principais: | Ramos, Inês Gonçalves |
| Assunto: | Machine Learning Space Weather Coronal Mass Ejection Satellite Orbits Aprendizagem Computacional Meteorologia Espacial Ejeção de Massa Coronal Órbitas de Satélites |
| Ano: | 2025 |
| País: | Portugal |
| Tipo de documento: | dissertação de mestrado |
| Tipo de acesso: | acesso aberto |
| Instituição associada: | Universidade de Coimbra |
| Idioma: | inglês |
| Origem: | Estudo Geral - Universidade de Coimbra |
| Resumo: | Satellites are a critical enabler of everyday technologies. From communications to GPS, from meteorology to the Internet, satellites allow modernity as we know it. Although they often go unnoticed by most people, satellites face constant challenges, such as Space Weather (SW). It comprises events that proceed mainly from the Sun and propagate through space. These events may disrupt satellite operations not only by altering their position but also by interfering with communication and onboard instruments. Such a disturbance may disrupt Earth systems as well. The most dangerous SW phenomenon for satellites is a Coronal Mass Ejection (CME). This event can alter the position of a satellite by hundreds of meters. If a large-scale CME hits Earth, it may cause planet-wide electricity and communications blackouts, which could be disastrous. Over the past several decades, researchers have attempted to forecast CMEs with some success, though many challenges remain, including limited accuracy and timeliness of existing predictions. Recent advances in Machine Learning (ML) in the field of space sciences provide promising avenues to improve CME forecasts. These methods can model the complex patterns of space data to provide better results than traditional methods. This dissertation aims to evaluate the viability of ML approaches for predicting CMEs by training models on rich solar wind data collected from the SOHO spacecraft, whose mission is to study the Sun. An extensive and carefully constructed dataset is used, featuring multiple solar wind parameters over several years. I thoroughly evaluate a range of ML algorithms along with preprocessing techniques and data splitting strategies suited for this problem. The results indicate that some models can achieve balanced accuracies above 70%, demonstrating meaningful predictive potential. This work contributes novel insights into the application of ML to Space Weather prediction and provides a robust dataset for the scientific community to use. |
|---|