Prediction of Ionospheric F2 Layer Height with Bi-parametric Deep Learning Network Based on HFSWR Data

Typhoons and other severe convective weather events annually lead to immeasurable financial losses in coastal areas and pose a threat to the safety of people. The height of the F2 layer of the ionosphere changes during the occurrence of severe sea states such as typhoons. The prediction of the ionospheric F2 layer height can be utilized to obtain the variations of the ionosphere over a period of time and then enable the analysis of typhoon wind speeds and other data, which is of great significance in protecting the lives of the coastal people and reducing the damage to property. The paper proposed a bi-parametric LSTM-CNN prediction model which fused the high frequency surface wave radar (HFSWR) ionospheric F2 layer height (hF2) data with the largest F2 layer height (hmF2) data acquired from the ionosonde as an input matrix. The fused data matrix was subsequently utilized as an input to the model for predicting the ionospheric F2 layer height data from HFSWR observations. The forecasting properties of the models were estimated through comparison with other prevalent ionospheric parameter prediction models such as BP (backward propagation), SARIMA (seasonal difference autoregressive moving average), and LSTM (long and short-term memory network). In this research, the forecasting performance of several models were evaluated with the root-mean-square error (RMSE), the median absolute deviation (MAD) and the mean absolute error (MAE). The MAE magnitudes of LSTM-CNN, BP, LSTM, and SARIMA models were 1.062, 3.467, 2.341, and 6.196, respectively. the two parametric LSTM-CNN model proposed in the paper performed the best in forecasting properties among the four predictive models, which was followed by the LSTM model, while the worst forecasting results came from the SARIMA model. Eventually, through the evaluation criteria and deviation results, which could be concluded that the proposed two-parameter LSTM-CNN model had a superior predictive performance.