Evaluation of error characteristics of derived TEC with IRI-2016

Abstract

The total electron content (TEC) derived from the Global Navigation Satellite System (GNSS) has been widely used to study the ionosphere. The evaluation of the derivation method is usually based on the fitting residuals, which are the difference between the estimate and observation. However, there are few studies on the statistical characteristics of fitting residuals. The characteristics of fitting residuals and estimation errors are studied with international reference ionosphere (IRI-2016) and GNSS observations. First, the systematic estimation errors caused by mathematical modeling and related assumptions are discussed with IRI-2016. The errors are 1 TECU and 0.5 TECU for slant TEC (STEC) and vertical TEC (vTEC), respectively. The fitting residuals of STEC observations (ΔSTECr) are studied by different noise models, including independent and identically distributed (IID) Gaussian noise, independent and non identical distributed (INID) Gaussian noise and INID Laplace noise. The statistical analysis shows that the distribution of ΔSTECr for the whole observations of all receivers is Gaussian when the noise is IID Gaussian, while it is Laplace for INID Gaussian or Laplace noise. The ΔSTECr reflects the statistical characteristic of noise. The estimation error of STEC (ΔSTEC), which is the difference between the estimate (STECe) and the model STEC (STEC) is discussed with IRI-2016. The distribution of ΔSTEC is neither Gaussian nor Laplace, and STECe is a biased estimate of STEC. The distribution of estimation error of vTEC (ΔvTEC) is neither Gaussian nor Laplace. Finally, the ΔSTECr is studied with GNSS observations. The distribution of ΔSTECr is more consistent with Laplace than Gaussian, indicating that the noise of the GNSS observations should not be modeled as IID Gaussian distribution.