Register now After registration you will be able to apply for this opportunity online.
Gaussian Processes for Ionospheric Modelling
This thesis explores ionospheric modeling using GNSS and potentially VLBI data, employing Gaussian Process regression to address the non-linear behaviors and noise inherent in such data. The study focuses on enhancing predictive accuracy and the quantification of uncertainties in ionospheric variations, which are essential for improving global navigation and communication systems.
This bachelor thesis delves into ionospheric modeling using Vertical Total Electron Content (VTEC) derived from GNSS and potentially VLBI data at Ionospheric Pierce Points (IPPs). The project aims to enhance models similar to the Global Ionospheric Maps provided by the International GNSS Service (IGS). We will use geographic, temporal, and potentially solar activity data, like solar flux, to map the ionosphere dynamically across space and time. We will consider various types of Gaussian Process (GP) regression to manage the varying noise levels in GNSS and VLBI data. We'll particularly explore Heteroscedastic Gaussian Processes and Multi-Output (Deep) Gaussian Processes to simultaneously handle complex correlations and biases in these data sources.
To ensure the robustness of our approach, we compare our GP regression with simple machine learning methods to validate the effectiveness of the advanced GP models. GPyTorch is utilized for its extensive support of the above GP configurations, which is crucial for comprehensive bias analysis and facilitates straightforward comparisons with the machine learning techniques implemented in PyTorch.
This research is motivated by the need to enhance global communication and navigation systems through more precise and reliable ionospheric predictions. It aims to establish a foundational study that integrates diverse geospatial data using sophisticated statistical techniques, leading the way for significant improvements in how we understand and predict ionospheric conditions.
This bachelor thesis delves into ionospheric modeling using Vertical Total Electron Content (VTEC) derived from GNSS and potentially VLBI data at Ionospheric Pierce Points (IPPs). The project aims to enhance models similar to the Global Ionospheric Maps provided by the International GNSS Service (IGS). We will use geographic, temporal, and potentially solar activity data, like solar flux, to map the ionosphere dynamically across space and time. We will consider various types of Gaussian Process (GP) regression to manage the varying noise levels in GNSS and VLBI data. We'll particularly explore Heteroscedastic Gaussian Processes and Multi-Output (Deep) Gaussian Processes to simultaneously handle complex correlations and biases in these data sources. To ensure the robustness of our approach, we compare our GP regression with simple machine learning methods to validate the effectiveness of the advanced GP models. GPyTorch is utilized for its extensive support of the above GP configurations, which is crucial for comprehensive bias analysis and facilitates straightforward comparisons with the machine learning techniques implemented in PyTorch. This research is motivated by the need to enhance global communication and navigation systems through more precise and reliable ionospheric predictions. It aims to establish a foundational study that integrates diverse geospatial data using sophisticated statistical techniques, leading the way for significant improvements in how we understand and predict ionospheric conditions.
The goal of this thesis is to improve the accuracy of ionospheric modeling by integrating GNSS and VLBI data through Gaussian Process regression. This integration will include various environmental and temporal parameters to develop a more precise and dynamic understanding of the ionosphere.
The goal of this thesis is to improve the accuracy of ionospheric modeling by integrating GNSS and VLBI data through Gaussian Process regression. This integration will include various environmental and temporal parameters to develop a more precise and dynamic understanding of the ionosphere.
Arno Rüegg (arrueegg@ethz.ch),
Marcel Iten (miten@ethz.ch),
Prof. Benedikt Soja (soja@ethz.ch)
Arno Rüegg (arrueegg@ethz.ch), Marcel Iten (miten@ethz.ch), Prof. Benedikt Soja (soja@ethz.ch)
