Nonlinear prediction of the ionospheric parameter ƒ o F 2 on hourly, daily, and monthly timescales

Abstract

An application of nonlinear radial basis function (RBF) neural networks (NNs) to improve the accuracy of the prediction of ionospheric parameters is presented. Principal component analysis is also adopted for the purposes of noise and dimension reduction. Hourly, daily, and monthly predictive models have been created for the Slough, England, United Kingdom, f o F 2 time series. The quality of the model predictions is evaluated by comparison with corresponding predictions from reference persistence or recurrence models. Each RBF NN offers a significant improvement over the performance of the corresponding reference model. The noonday model gives a performance improvement of ∼60% over the baseline persistence model, for a 1 day ahead prediction. For a 1 hour ahead prediction the hourly model offers an improvement of ∼45% over the baseline 24 hour recurrence model. Finally, the monthly median model gives a performance improvement of ∼40% over the baseline persistence model, for a 1 month ahead prediction.