A neural network-based ionospheric model for the auroral zone

Abstract A new empirical model for the lower ionosphere in the auroral zone, called IMAZ, has been developed, tested and refined for use in the International Reference Ionosphere (IRI) global model. Available ionospheric data have been used to train neural networks (NNs) to predict the high latitude electron density profile. Data from the European Incoherent Scatter Radar (EISCAT), based near Tromso, Norway (69.58°N, 19.23°E), combined with rocket-borne measurements (from 61° to 69° geomagnetic latitude) make up the database of reliable D- and E-region data. NNs were trained with different combinations of the following input parameters: day number, time of day, total absorption, local magnetic K index, planetary Ap index, 10.7 cm solar radio flux, solar zenith angle and pressure surface. The output that the NNs were trained to predict was the electron density for a given set of input parameters. The criteria for determining the optimum NN are (a) the root mean square (RMS) error between the measured and predicted output values, and (b) the ability to reproduce the absorption they are meant to represent. An optimum input space was determined and then adapted to suit the requirements of the IRI community. In addition, the true quiet electron densities were simulated and added to the database, thus allowing the final model to be valid for riometer absorptions down to 0 dB. This paper discusses the development of a NN-based model for the high-latitude, lower ionosphere, and presents results from the version developed specifically for the IRI user community.

[1]  M. Jespersen,et al.  Electron and positive ion density measurements during conditions of auroral absorption , 1968 .

[2]  W. Swider,et al.  Effective electron loss coefficient of the disturbed daytime D region , 1975 .

[3]  P. Stauning.  High-latitude D- and E-region investigations using imaging riometer observations , 1996 .

[4]  H. Voss,et al.  Electron density and energetic particle precipitation observed during the eclipse of 26 February 1979 , 1983 .

[5]  Lj.R. Cander,et al.  Short-Term Prediction of foF2 using Time-Delay Neural Network , 1999 .

[6]  S. Kirkwood,et al.  The high-latitude lower ionosphere observed by EISCAT , 1987 .

[7]  L. McKinnell,et al.  Towards neural network based models for the ionospheric D-region , 2003 .

[8]  Klaus Torkar,et al.  Typical behaviour of the high latitude lower ionosphere , 1995 .

[9]  Lee-Anne Williscroft,et al.  Neural networks, foF2, sunspot number and magnetic activity , 1996 .

[10]  Dieter Bilitza,et al.  International reference ionosphere—Status 1995/96 , 1997 .

[11]  S. Kirkwood,et al.  The disturbed auroral ionosphere based on EISCAT and rocket data , 2002 .

[12]  S. Hyakin,et al.  Neural Networks: A Comprehensive Foundation , 1994 .

[13]  K. Torkar,et al.  Quantitative measurements with wide-beam riometers , 2002 .

[14]  L. McKinnell,et al.  A new approach to modelling the daytime lower ionosphere at auroral latitudes , 2004 .

[15]  Ersin Tulunay,et al.  Forecasting of ionospheric critical frequency using neural networks , 1997 .

[16]  L. McKinnell,et al.  The development of a neural network based short term foF2 forecast program , 2000 .

[17]  F. Lübken,et al.  The quiet auroral ionosphere and its neutral background , 2004 .

[18]  Lee-Anne McKinnell,et al.  On the predictability of f0F2 using neural networks , 2000 .