A Near-Real-Time Forecasting Model of High-Frequency Radiowave Propagation Factor Fusion Based on the ICEEMDAN Decomposition and Bi-LSTM Methods

High-frequency (HF) communication achieves ultralong distances and even global propagation through the sky wave propagation mode via ionospheric refraction. M(3000)F2 is an essential parameter for HF sky wave propagation, defined as the ionospheric propagation factor at 3000 km in the F2 layer. However, the nonlinearity and nonstationarity of the M(3000)F2 make its near-real-time forecasting very challenging. This study presents a new hybrid method for improving the near-real-time forecasting accuracy of M(3000)F2 based on deep learning (DL) using improved complete ensemble empirical mode decomposition with adaptive noise (ICEEMDAN), correlation coefficients (CCs), and bi-directional long short-term memory (Bi-LSTM). We used M(3000)F2 data from five stations in China in 2014 as the data sample to discuss the model forecasting performance. The hybrid model initially employs ICEEMDAN to decompose the M(3000)F2 into subsequences and then uses CC to reconstruct the subsequences into approximate and detailed components. The reconstructed components are forecast with solar activity and geomagnetic activity parameters, and the results are combined and output, achieving a high-precision forecasting of M(3000)F2. For the validation and test sets, the average root mean square error (RMSE) is 0.10 and 0.12, and the relative RMSE (RRMSE) is 3.38% and 4.14%. Multiple comparative experiments are conducted with some mainstream models to validate the forecast accuracy and stability of the model, which outperforms comparative models, especially in low-latitude areas. The proposed method can support propagation prediction and frequency optimization of HF communication and has the potential for extension to provide high-precision forecasts of ionospheric parameters on a global scale.