Abstract
Global Navigation Satellite System (GNSS) Radio Occultation (RO) has provided high-quality atmospheric data assimilated in Numerical Weather Prediction (NWP) models and climatology studies for more than 20 years. In the satellite–satellite GNSS-RO geometry, the measurements are susceptible to ionospheric scintillation depending on the solar and geomagnetic activity, seasons, geographical location and local time. This study investigates the application of the Support Vector Machine (SVM) algorithm in developing an automatic detection model of F-layer scintillation in GNSS-RO measurements using power spectral density (PSD). The model is intended for future analyses on the influence of space weather and solar activity on RO data products over long time periods. A novel data set of occultations is used to train the SVM algorithm. The data set is composed of events at low latitudes on 15–20 March 2015 (St. Patrick’s Day geomagnetic storm, high solar flux) and 14–19 May 2018 (quiet period, low solar flux). A few conditional criteria were first applied to a total of 5340 occultations to define a set of 858 scintillation candidates. Models were trained with scintillation indices and PSDs as training features and were either linear or Gaussian kernel. The investigations also show that besides the intensity PSD, the (excess) phase PSD has a positive contribution in increasing the detection of true positives.
Highlights
The threats to Global Navigation Satellite System (GNSS) operation caused by the ionosphere are widely known
The amplitude scintillation is quantified by the S4 index, which corresponds to the fourth moment of the received signal amplitude [2,53], normally computed as the standard deviation of the normalised signal intensity, h( I − h I i)2 i in which I is the intensity of the Radio Occultation (RO) signal, h i denotes the expectation operator and I is the filtered version of the signal intensity used as a reference in the calculation
GNSS-RO measurements performed by Meteorological Operational (MetOp)-A/B at low latitudes during two intervals of the 24th solar cycle have been analysed
Summary
The threats to Global Navigation Satellite System (GNSS) operation caused by the ionosphere are widely known. A similar approach has been used in the detection of phase scintillation at high latitudes using σφ and phase PSD as training features [37] Both models were trained and validated with L-band measurements by ground-based receivers. An SVM-based model has been developed to predict ionospheric scintillation at high latitudes, using a large variety of training features besides scintillation indices and spectral information, namely, geographical location, local time, solar wind and particle precipitation data, and geomagnetic indices. Different training features are evaluated, such as amplitude and phase scintillation index and power spectral density, to define the classifier with the best overall performance in terms of accuracy, precision, recall and Receiver Operating Characteristic (ROC) curves.
Sign-in/Register to view highlights & summary Sign-in/Register to access full text options 
Free) 
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have
Schedule a call
