The determination of ionospheric key quantities such as the maximum electron density of the F2 layer NmF2, the corresponding F2 peak height hmF2 and the F2 scale height HF2 are of high relevance in 4-D ionosphere modeling to provide information on the vertical structure of the electron density (Ne). The Ne distribution with respect to height can, for instance, be modeled by the commonly accepted F2 Chapman layer. An adequate and observation driven description of the vertical Ne variation can be obtained from electron density profiles (EDPs) derived by ionospheric radio occultation measurements between GPS and low Earth orbiter (LEO) satellites. For these purposes, the six FORMOSAT-3/COSMIC (F3/C) satellites provide an excellent opportunity to collect EDPs that cover most of the ionospheric region, in particular the F2 layer. For the contents of this paper, F3/C EDPs have been exploited to determine NmF2, hmF2 and HF2 within a regional modeling approach. As mathematical base functions, endpoint-interpolating polynomial B-splines are considered to model the key parameters with respect to longitude, latitude and time. The description of deterministic processes and the verification of this modeling approach have been published previously in Limberger et al. (2013), whereas this paper should be considered as an extension dealing with related correlation studies, a topic to which less attention has been paid in the literature. Relations between the B-spline series coefficients regarding specific key parameters as well as dependencies between the three F2 Chapman key parameters are in the main focus. Dependencies are interpreted from the post-derived correlation matrices as a result of (1) a simulated scenario without data gaps by taking dense, homogenously distributed profiles into account and (2) two real data scenarios on 1 July 2008 and 1 July 2012 including sparsely, inhomogeneously distributed F3/C EDPs. Moderate correlations between hmF2 and HF2 as well as inverse correlations between NmF2 and HF2 are reflected from the simulation. By means of the real data studies, it becomes obvious that the sparse measurement distribution leads to an increased weighting of the prior information and suppresses the parameter correlations which play an important role regarding the parameter estimability. The currently implemented stochastic model is in need of improvement and does not consider stochastic correlations which consequently cannot occur.

Peer Reviewed