Predicting ionospheric critical frequency of the F2 layer over Lycksele using the neural network improved by error compensation technology

Abstract

In this contribution, according to the critical frequency of the F2 layer (foF2) from the ionosonde station and the International Reference Ionosphere 2012 (IRI2012) model [the International Radio Consultative Committee (CCIR) coefficient], a single station model (called the Fusion model) for predicting the hourly value of foF2 over Lycksele (64.69°N, 18.80°E) is developed by the back-propagation neural network (BPNN) technology compensated for the model-deviation of the IRI2012 model. The input parameters of this BPNN-based foF2 prediction model include day number (day of the year), universal time, solar zenith angle, solar cycle information, geomagnetic activity (a 3-day running mean of the 3-h planetary magnetic index Ap and Kp) and the foF2 values from the IRI2012 model; the output parameter is the model-deviation. The data sets employed in this model are obtained from the National Geophysical Data Centre and National Aeronautics and Space Administration (NASA), about 215 614 sets, which are derived during the period from January 1958 to December 1984 (1960s data is missing). The data sets from 1971 to 1984 are selected for validation instead of for training use. Prediction values from the Fusion model, the General BPNN and IRI2012 models are each compared with the observed data. The results indicate that the proposed model is superior to the other models for hourly value of foF2 prediction over Lycksele. According to the statistical analysis of average root mean square error, the proposed model offers an improvement of 19.11% over the IRI2012 model.