Improving VTEC-derived IRI empirical model using COSMIC2 Radio Occultation observations

Due to the inadequate spatial distribution of data in certain regions, empirical ionospheric models suffer from limited accuracy. To address this limitation, the present study developed a method to enhance the Total Electron Content (TEC) predictions of the International Reference Ionosphere 2020 (IRI-2020) model using ionospheric profile (ionPrf) products derived from the Constellation Observing System for Meteorology, Ionosphere, and Climate-2 (COSMIC2) measurements. COSMIC2 represents the next-generation follow-on to the original COSMIC mission. It consists of six satellites in orbits with a 24-degree inclination, primarily providing coverage between 30°N and 30°S latitudes. By contrast, the original COSMIC satellites operated at an 87-degree inclination, achieving near-global coverage. The proposed enhancement approach comprises two main components: a background model and correction terms. The IRI-2020 serves as the background, while the corrections are modeled via spherical harmonics expansion, yielding 256 unknown coefficients per time interval. The resulting improved TEC maps, guided by the spatial distribution of COSMIC2 observations, offer coverage from 30°S to 30°N across all longitudes. However, COSMIC2 data remain sparse near 45°N and 45°S latitudes. To assess the method, January 14, 2022, was selected as the test date. Geomagnetic conditions remained quiet (steady state) from 00:00 to 21:00 UT, shifting to disturbed (storm) conditions from 21:00 to 24:00 UT, while solar activity stayed stable throughout the day. The generated TEC maps feature a spatial resolution of 5° in longitude and 2.5° in latitude, with a temporal resolution of 2 hours. This interval was chosen because a 2-hour period typically contains more than 256 COSMIC2 observations, sufficient for reliable estimation of the spherical harmonic coefficients. Estimating these coefficients involves solving an ill-posed inverse problem using regularization techniques. Two approaches were compared in this study: the direct Tikhonov regularization and the recursive Least Squares with Quadratic Regularization (LSQR). For validation, modeled Vertical Total Electron Content (VTEC) values were compared against independent measurements from five International GNSS Service (IGS) GPS stations—two situated over the ocean and three on land. Performances were evaluated separately for quiet and storm geomagnetic conditions using Root Mean Square Error (RMSE) and Normalized Root Mean Square Error (NRMSE). Overall, the proposed method substantially reduced RMSE at both land-based and ocean-based stations, even during geomagnetic storms, although its effectiveness depended on the proximity and distribution of Radio Occultation (RO) observations within the region. Relative to the original IRI-2020 model, the recursive LSQR method reduced NRMSE by approximately 1.75% in quiet conditions and 3.86% during storm conditions, while the direct Tikhonov method achieved reductions of 1.34% and 3.68%, respectively. In most 2-hour intervals, LSQR outperformed Tikhonov at individual GPS stations, irrespective of land or sea location. Nevertheless, in one case during storm conditions, neither method could produce a reliable correction, owing to the unfavorable spatial distribution of available RO observations.