A neural network based error correction method for radio occultation electron density retrieval

Viet Cuong Pham, Jyh Ching Juang

Research output: Contribution to journalArticlepeer-review

1 Citation (Scopus)

Abstract

Abel inversion techniques have been widely employed to retrieve electron density profiles (EDPs) from radio occultation (RO) measurements, which are available by observing Global Navigation Satellite System (GNSS) satellites from low-earth-orbit (LEO) satellites. It is well known that the ordinary Abel inversion might introduce errors in the retrieval of EDPs when the spherical symmetry assumption is violated. The error, however, is case-dependent; therefore it is desirable to associate an error index or correction coefficient with respect to each retrieved EDP. Several error indices have been proposed but they only deal with electron density at the F2 peak and suffer from some drawbacks. In this paper we propose an artificial neural network (ANN) based error correction method for EDPs obtained by the ordinary Abel inversion. The ANN is first trained to learn the relationship between vertical total electron content (TEC) measurements and retrieval errors at the F2 peak, 220. km and 110. km altitudes; correction coefficients are then estimated to correct the retrieved EDPs at these three altitudes. Experiments using the NeQuick2 model and real FORMOSAT-3/COSMIC RO geometry show that the proposed method outperforms existing ones. Real incoherent scatter radar (ISR) measurements at the Jicamarca Radio Observatory and the global TEC map provided by the International GNSS Service (IGS) are also used to valid the proposed method.

Original languageEnglish
Pages (from-to)77-84
Number of pages8
JournalJournal of Atmospheric and Solar-Terrestrial Physics
Volume135
DOIs
Publication statusPublished - 2015 Dec 1

All Science Journal Classification (ASJC) codes

  • Geophysics
  • Atmospheric Science
  • Space and Planetary Science

Fingerprint Dive into the research topics of 'A neural network based error correction method for radio occultation electron density retrieval'. Together they form a unique fingerprint.

Cite this