Speaker
Description
Irregularities in the ionospheric layer can cause rapid fluctuations in the amplitude, phase, and direction of arrival of received GNSS signals, commonly referred to as ionospheric scintillation. These irregularities pose significant challenges to GNSS-based applications, especially in high-latitude regions where such disturbances are more pronounced. Recent advances in deep learning offer novel approaches to studying ionospheric scintillation and predicting related indices. This paper presents artificial intelligence (AI)-based models, leveraging solar wind and geomagnetic indices, to predict the Rate of TEC Index (ROTI), a key metric linked to high-latitude ionospheric disturbances [Fabbro et al., 2021].
A key objective of this initiative is to extend the prediction horizons of ionospheric indices, currently limited to little more than an hour [McGranaghan et al., 2018]. The long-term goal is to replace dependence on L1 data with methods based on solar images as proxies for solar activity. By leveraging advanced modeling techniques and index databases, this study contributes to enhancing the predictive capabilities and operational reliability of ionospheric forecasts.
The prediction of the ROTI index is particularly challenging due to the complex and nonlinear dynamics of the Earth’s magnetosphere and ionosphere. To address this issue, nonlinear models combining Long
Short-Term Memory (LSTM) networks and Convolutional LSTM (ConvLSTM) architectures have been developed. LSTM models have proven their efficiency in space weather domain to forecast geomagnetic indices specific to Earth magnetic current systems, like am index in [Gruet, 2018]. The choosen architectures incorporate two input branches: one for multivariate time series of solar wind and geomagnetic field measurements (1D input) and another for time series of 2D maps of the predicted index (2D input). The combined framework enables the prediction of 2D ROTI maps, capturing efficiently both temporal and spatial dynamics. The multivariate time series (1D input) describing the solar wind are intended to be supplemented and eventually replaced in the next phases of the project by outputs of neural networks capable of transforming solar images into proxies for solar activity.
The ROTI datasets used in this study were obtained from Norwegian Mapping Authority (NMA). These datasets provide extensive latitude coverage, enabling a comprehensive geographic analysis of ionospheric conditions in the Arctic region during the 2010–2020 period.Feature parameters for ROTI time series prediction were extracted from the five-minute resolution HRO dataset, available from NASA’s Space Physics Data.
The evaluation of the combined LSTM and ConvLSTM models demonstrated their potential to predict 2D ROTI maps with improved accuracy, offering promising insights for GNSS users operating in high latitude regions. The integration of solar wind and geomagnetic features into LSTM and ConvLSTMbased models provides a powerful tool for predicting ionospheric disturbances with practical implications for GNSS reliability.
These models seem to capture the spatio-temporal dynamics of ionospheric disturbances, making them well-suited for practical space weather applications. Future work will focus on extending the temporal and geographic coverage of the models to include mid- and low-latitude regions, as well as integrating geomagnetic proxies derived from neural network analysis of solar images. Efforts will be directed toward developing end-to-end operational systems
| Do you plan to attend in-person or online? | In-person |
|---|
