Geomagnetically correlated autoregression model for short-term prediction of ionospheric parameters

Abstract

A new method for short-term prediction of ionospheric parameters is developed by incorporating the cross-correlation between the ionospheric characteristic of interest and the Ap index into the autocorrelation analysis. We consider the hourly time series of an ionospheric characteristic as composed of a periodic component and a random component. The periodic component containing the average diurnal variation is removed by using its relative deviations from the median values (Φ), which in the case of the critical frequency of the F2 layer, foF2, has the form: Φ = (foF2 - foF2med)/foF2med. The geomagnetically correlated autoregression model (GCAM) is an extrapolation model based on the weighted past data. The new term in the regression equation expresses linearly the dependence of Φ on magnetic activity by introducing a synthetic geomagnetic index G, which approximates the average dependence of Φ on hourly interpolated Kp. Using parametric expressions of the auto- and cross-correlation functions ensures the statistical sufficiency in GCAM; the parameters are then obtained by data fitting. Data from 2 years of high solar activity (1981-2) and 2 years of low solar activity (1985-6) were used to evaluate the prediction accuracy of GCAM. The mean square error in per cent of the 1-day prediction of foF2 relative to the median shows a large gain of accuracy of GCAM in the first 8-10 h of prediction relative to the median based prediction, a diurnal variation of errors and a steady offset of the GCAM prediction error from the median based prediction error. The GCAM error at the first hour is lowest, but gradually approaches the median error with a timescale of 8-10 h. A new error estimate, called `prediction efficiency' that is a good indicator of prediction performance during disturbed ionospheric conditions is defined.