A global 3-D electron density reconstruction model based on radio occultation data and neural networks

The accurate representation of the ionospheric electron density in 3-dimensions is a challenging problem because nature horizontal and vertical structures on both small large scales. This paper presents development global three-dimensional (3-D) reconstruction based radio occultation data during 2006–2019 neural networks. We demonstrate that developed model COSMIC dataset only capable reproducing different features when compared to independent datasets from ionosondes incoherent scatter radars (ISR) low, middle high latitude regions. Following some existing modelling efforts similar or related technique we divided into fine resolution grid cells 5 ∘ × 1 (geographic latitudes/longitudes) followed by network subroutine per cell later combining all 864 sub-models compile one model. approach has been demonstrated be appropriate enabling networks learn, reproduce generalise local behaviour density. Based ISR data, 3D improves maximum F2 layer ( NmF2 ) prediction 10%–20% IRI 2016 quiet conditions. For estimation ionosonde ordinary critical frequency foF2 2009 at 1200 UT (universal time), 3-D gives average root mean square error (RMSE) values 0.83 MHz, 1.06 MHz 1.16 0.92 1.09 1.01 over Africa–European, American Asian sectors respectively making their performances statistically comparable. Compared consistently shows better performance for hmF2 results almost investigated periods. • A artificial When benchmarked ISR, suggest new applicable future assimilation especially areas which are not sufficiently covered ground-based instrumentation such as ionosondes.