| Summary: | Over recent years, global navigation satellite systems (GNSSs) have been increasingly used to study near-Earth space. The basis for such studies is the total electron content (TEC) data. Standard procedures for detecting TEC wave signatures include variation selection and detrending. Our experience showed that the inaccurate procedure causes artifacts in datasets which might affect data interpretation, particularly in automated processing. We analyzed the features of various detrending and variation selection methods. We split the problem of the GNSS data filtering into two subproblems: detrending and variation selection. We examined centered moving average, centered moving median, 6th-order polynomial, Hodrick–Prescott filter, L1 filter, cubic smoothing spline, double-applied moving average for the GNSS-TEC detrending problem, and centered moving average, centered moving median, Butterworth filter, type I Chebyshev filter for the GNSS-TEC variation selection problem in this paper. We carried out the analysis based on both model and experimental data. Modeling was based on simple analytical models as well as the International Reference Ionosphere. Analysis of TEC variations of 2–10 min, 10–20 min, and 20–60 min under insufficient detrending conditions showed that the higher errors appear for the longer periods (20–60 min). For the detrending problem, the smoothing cubic spline provided the best results among the algorithms discussed in this paper. The spline detrending featured the minimal value of the mean bias error (MBE) and the root-mean-square error (RMSE), as well as high correlation coefficient. The 6th-order polynomial also produced good results. Spline detrending does not introduce a RMSE more than 0.25 TECU and MBE > 0.35 TECU for IRI trends, while, for the 6th-order polynomial, these errors can exceed 1 TECU in some cases. However, in 95% of observations the RMSE and MBE do not exceed 0.05 TECU. For the variation selection, the centered moving average filter showed the best performance among the algorithms discussed in this paper.
|
|---|
