Forecasting the ionospheric foF2 parameter using a support vector machine technique

The ionosphere is highly variable due to the influence of the solar, geomagnetic and other sources. The F2 layer critical frequency, foF2, is one of the most important ionospheric pa-rameters. For many communication and guidance applications in practice, forecasting of foF2 1 hour in advance is very important. So we propose a SVM approach to ionospheric short-term forecasting. SVM is a combination of a kernel-based architecture and a structural risk minimization (SRM) principle. SRM minimizes an upper bound on the expected risk, which provides solid theoretical grounds for optimizing the generalization ability of SVM. SVM per-forms a mapping process from the input to a higher-dimensional feature space by using a kernel function, and finds out the relations between the input and the targets. The purpose of this paper is to introduce SVM, as a new approach, to forecast the one-hour-ahead ionospheric foF2. The hourly values of foF2 are used in this paper, depending on the availability, from the ionospheric stations Guangzhou, Haikou, Chongqing, Beijing, Lanzhou, Changchun, and Manzhouli located in China. To test our approach, we have used data from China. The re-sults have been compared to other prediction techniques. At this preliminary stage, this paper shows the potential application of this technique for forecasting foF2. The calculation results show that the forecasting errors at high solar activity are usually larger than that at low solar activity for the same station. As an example at Changchun station, foF2 is 0.46MHz and the RMS is 0.66 MHz in 1981 at high solar activity, while foF2 is 0.33 MHzand RMS is 0.46MHz in 1986. Compared with low solar activity, foF2 and RMS at high solar activity are larger by 0.13 MHz and 0.20MHz respectively. The forecasting errors at lower latitudes are usually larger. An important reason is that both the values of foF2 and their variations are usually larger at low latitudes, which makes foF2 and RMS larger, no doubt because of the increased ionospheric variability in the low latitude region Because foF2 variations during the different seasons and solar activity are different,foF2 and RMS for the different seasons in 1986 at these seven stations are debated. Equinoxes includes the months March, April, September and October, summer includes May, June, July, August and winter includes January, February, November and De-cember. In order to verify the predictive ability of the SVM model under both geomagnetically quiet and disturbed conditions, we consider three cases of disturbed ionospheric conditions that occurred in 4-6 March 1981, 1-3 March 1982 and 8-10 February 1986 respectively. These periods were considered for this test due to there being sufficient continuous data available. By introducing the SVM method for the ionospheric short-term forecasting, we propose a simple and practical forecast method for predictions of one hour ahead. the prediction quality has been quantitatively estimated at different stations and seasons. This indicates that the perfor-mance of the SVM model is superior to that of the autocorrelation and persistence models and comparable to that of the NN model.