Enter Note Done
Go to previous page in this tab
Session
  • Presentation | S21C: From Task-Specific Machine Learning to Foundation Models in Seismology and Geodesy I Poster
  • Poster
  • Share Icon
    Bookmark Icon
  • S21C-0189: Probabilistic Neural Networks for Atmospheric Delay Modeling: Enhancing GNSS Positioning with Uncertainty-Aware Corrections
    • Schedule
    Notes
  • Board 0189‚ Hall EFG (Poster Hall)
    NOLA CC
    Set Timezone

Generic 'disconnected' Message
Author(s):
Benedikt Soja, Institute of Geodesy and Photogrammetry, ETH Zurich (First Author, Presenting Author)
Tomasz Hadaś, Institute of Geodesy and Geoinformatics, Wrocław University of Environmental and Life Sciences
Raul Orus Perez, Wave Interaction and Propagation Section, ESA-ESTEC
Muhammad Arqim Adil, Institute of Geodesy and Geoinformatics, Wrocław University of Environmental and Life Sciences
Matthias Aichinger-Rosenberger, Institute of Geodesy and Photogrammetry, ETH Zurich
Laura Crocetti, Institute of Geodesy and Photogrammetry, ETH Zurich
Junyang Gou, Institute of Geodesy and Photogrammetry, ETH Zurich
Kamil Kaźmierski, Institute of Geodesy and Geoinformatics, Wrocław University of Environmental and Life Sciences
Grzegorz Marut, Institute of Geodesy and Geoinformatics, Wrocław University of Environmental and Life Sciences
Shuyin Mao, Institute of Geodesy and Photogrammetry, ETH Zurich
Arno Rüegg, Institute of Geodesy and Photogrammetry, ETH Zurich
Matthias Schartner, Institute of Geodesy and Photogrammetry, ETH Zurich


Global Navigation Satellite Systems (GNSS), like GPS, are widely used for everything from smartphone navigation to high-precision scientific and engineering applications. However, signals from satellites are delayed as they pass through Earth’s atmosphere, which can reduce positioning accuracy. To fix this, we need models that estimate these delays precisely - and also tell us how reliable those estimates are.


This project, funded by the European Space Agency, uses machine learning - specifically a method called Probabilistic Neural Networks (PNNs) - to model delays caused by the atmosphere. Unlike traditional models, PNNs can not only predict these delays but also give a measure of uncertainty, which is important for providing trustworthy navigation solutions.


We developed and tested two versions of the model: one that uses only basic information like time and location, and another that includes extra data for more accurate results. We tested both in realistic navigation scenarios and found that they improved the accuracy and speed of positioning.




Scientific Discipline
Suggested Itineraries
Neighborhood
Type
Main Session
Discussion