Scintillation Identification Based on Spectral Features

Performance of machine learning (ML) methods to identify scintillation events is analyzed for a variety of scenarios. It shows that spectrum embodies various features on scintillation variation. Different methods can be developed with ML to find out potential scintillation impacts, and satisfied results generally could be arrived with accuracy of 95%. It also point out that the descending trend existed over Fresnel frequency is essential to distinguish a potential scintillation. So selecting a proper frequency band to make spectral features more distinguishable plays an important role in ML realization. It further shows that precise GNSS routine observations with the sampling rate of 1 Hz can be served to recognize scintillation event if a sound spectrum range has been chosen. When a set of parameters on spectrum characteristics could be derived and used for ML training, better performance can even be expected.