Calibration of Swarm Plasma Densities Overestimation Using Neural Networks

Abstract Recent studies have shown that the measurements of Langmuir Probes (LPs) onboard ESA's Swarm mission overestimate ion densities on the nightside by up to 50%. The overestimation is due to the assumption of oxygen‐only plasma for ion density calculations, which is often violated at mid‐latitudes on the nightside. In this study, we present a calibration model that resolves the nighttime overestimation by Swarm LPs. Using observations by Swarm FacePlate (FP) as a reference, we develop a neural network (NN) model that adjusts LP data to the FP measurements. The model incorporates dependence on solar and geomagnetic conditions, parameterized by the P10.7 and Hp30 indices, location, day of the year and local time. Our model reveals a distinct double‐crest pattern in nighttime density overestimation by LPs, centered at ∼30° quasi‐dipole latitude in both hemispheres. This overestimation intensifies during low solar activity and shows strong seasonal dependence. During solstices, the crests are more pronounced in the local winter hemispheres, while during equinoxes the crests are weaker and exhibit hemispheric symmetry. This morphology aligns with the presence of light ions diffusing downward from the plasmasphere. Validating the LP data in conjunctions with Constellation Observing System for Meteorology, Ionosphere and Climate (COSMIC) observations showed a much stronger agreement after applying the developed correction: for Swarm B, nighttime correlation with COSMIC increased from 0.74 to 0.93. The NN‐calibrated LP data set has numerous applications in ionospheric research, and the developed model can provide useful insights into the ion composition in the topside ionosphere.