European Space Weather Week 2025

Machine Learning-Based Prediction of GNSS Phase Scintillation in the Arctic Using Solar Wind and Geomagnetic Data

Not scheduled

20m

Idun

Poster CD1 - Combination of physics-based and data-driven methods for space weather forecasting CD1 - Combination of physics-based and data-driven methods for space weather forecasting

Speaker

Holger Lindholm (DTU Space, Technical University of Denmark)

Description

Ionospheric scintillation poses significant challenges to GNSS positioning accuracy and reliability, particularly in the Arctic Region, where geomagnetic activity is most pronounced. This study presents a machine learning framework for predicting phase scintillation events in GNSS receivers across the Arctic, with a primary focus on Greenland, using near real-time solar wind parameters and magnetic data.

Our approach leverages solar wind data from the L1 Lagrange point—including velocity, density, and B-field components—alongside ground-based geomagnetic measurements, to forecast scintillation events with a lead time corresponding to solar wind propagation from L1 to Earth. High-rate GNSS receivers (50–100 Hz) from the SWADO network in Greenland and the CHAIN network in Canada provide training data for σφ scintillation indices. In parallel, lower-rate receivers (1 Hz) from the GNET network are evaluated as potential sources for model validation and extended training.

A key component of this study is assessing the compatibility between high- and low-rate GNSS data for characterizing scintillation. Specifically, we investigate whether σφ values derived from 1 Hz data can reliably approximate those from high-rate data during periods of both low and high geomagnetic activity. Demonstrating sufficient agreement would justify incorporating 1 Hz data into the training process, thereby significantly increasing the spatial and temporal coverage of the dataset. This comparative analysis is essential to determining whether the model can scale across wider regions with fewer high-rate receivers.

By integrating these findings, the resulting predictive model not only forecasts scintillation events in near real-time but also demonstrates a path toward operational deployment in data-sparse regions. This work contributes both to model development and to a methodological evaluation of data sources for space weather applications, aiming to improve GNSS reliability in polar aviation, shipping, and scientific missions.