Advanced Machine Learning Optimized by The Genetic Algorithm in Ionospheric Models Using Long-Term Multi-Instrument Observations

Abstract

The ionospheric delay is of paramount importance to radio communication, satellite navigation and positioning. It is necessary to predict high-accuracy ionospheric peak parameters for single frequency receivers. In this study, the state-of-the-art artificial neural network (ANN) technique optimized by the genetic algorithm is used to develop global ionospheric models for predicting foF2 and hmF2. The models are based on long-term multiple measurements including ionospheric peak frequency model (GIPFM) and global ionospheric peak height model (GIPHM). Predictions of the GIPFM and GIPHM are compared with the International Reference Ionosphere (IRI) model in 2009 and 2013 respectively. This comparison shows that the root-mean-square errors (RMSEs) of GIPFM are 0.82 MHz and 0.71 MHz in 2013 and 2009, respectively. This result is about 20%-35% lower than that of IRI. Additionally, the corresponding hmF2 median errors of GIPHM are 20% to 30% smaller than that of IRI. Furthermore, the ANN models present a good capability to capture the global or regional ionospheric spatial-temporal characteristics, e.g., the equatorial ionization anomaly and Weddell Sea anomaly. The study shows that the ANN-based model has a better agreement to reference value than the IRI model, not only along the Greenwich meridian, but also on a global scale. The approach proposed in this study has the potential to be a new three-dimensional electron density model combined with the inclusion of the upcoming Constellation Observing System for Meteorology, Ionosphere and Climate (COSMIC-2) data.Remote Sens. 2020, 12, 866 2 of 17 space-borne and ground-based instruments and data processed techniques, more attention has been paid to the development and improvement of these ionospheric models [3].In the previous studies, the artificial neural network (ANN), nonlinear least squares and AdaBoost techniques have been used to predict ionospheric variations with a satisfactory accuracy [4][5][6][7]. Among them, the ANN technique has proven to be a successful tool in the modeling of ionospheric variations as well as solving the forecast problems in many geophysical applications over a single station, regional area and global scale [3,[8][9][10]. The neural network was first applied to build the foF2 model over the Grahamstown station (33 • S, 26 • E) using the data range set of 1973-1983 [11]. This model can predict the daily noon foF2 value with a root-mean-square error (RMSE) of 0.95 MHz and the monthly averaged RMSE value was 0.48 MHz. Afterward, this approach has been widely applied in the Asia/Pacific sector [5], equatorial region [12], European region [13] and the auroral zone [14]. These studies showed an improvement in the accuracy of the ANN-based models over the International Reference Ionosphere (IRI) model.With the advancement of data acquisition, more ionospheric measurements have become available for the global modelling of foF2. For instance, Oyeyemi et al. [15] used the foF2 data from 59 globally distributed ionospheric stations to establish th...

A new method for improving the performance of an ionospheric model developed by multi-instrument measurements based on artificial neural network

“…In addition, these models successfully capture the prominent large-scale ionospheric phenomena, such as the diurnal and seasonal variations of equatorial ionization anomaly, ionospheric annual anomaly (Rishbeth et al, 2000), and the ionospheric trough. The powerful performance of ANN models has also been verified in Li et al's study (Li et al, 2020), the accuracies of foF2 and hmF2 models built by ANN technique were 20-30% higher than the International Reference Ionosphere (IRI) model, and the ANN models present a good capability to reproduce the global or regional ionospheric spatial-temporal characteristics. The above literatures demonstrate that the ANN technique could successfully extract the multivariable nonlinear relationship between ionospheric parameters and geographic location, solar-geomagnetic indices, thermospheric winds, etc.…”

“…For more details about the activation functions, the readers are referred to Specht's study (Specht, 1990). Hundreds of experiments indicated that the mean square error (MSE) of the neural network was minimized when the neurons in three hidden layers were 16, 14 and 12, respectively (Li et al, 2020). In this study, both COSMIC's profiles and ionosonde's observations were utilized to build global ionospheric models for predicting foF2 and hmF2 based on the neural network's architecture, respectively.…”

Application of a Multi‐Layer Artificial Neural Network in a 3‐D Global Electron Density Model Using the Long‐Term Observations of COSMIC, Fengyun‐3C, and Digisonde

“…This study is a further extension of the ANN algorithm proposed by Li, Zhao, He, et al., (2020). Based on the success of the ANN model in predicting ionospheric peak parameters foF2 and hmF2, this study uses the profiles obtained from multiple RO systems (COSMIC‐1, FY‐3C) and the observations of global Digisondes to develop a completely global three‐dimensional electron density model with an artificial neural network, namely ANN‐TDD.…”

“…This study is a further extension of the ANN algorithm proposed by [Li et al, 2020b]. Based on the success of the ANN model in predicting ionospheric peak…”

“…The RMSEs of the predicted hmF2 (F2‐layer peak height) residuals in 2009 and 2002 were 29 and 25 km, respectively. Compared to the Global Ionospheric Peak Height Model (GIPHM) developed using advanced machine learning techniques optimized by the genetic algorithm (Li, Zhao, He, et al., 2020), the predicted accuracy of the ANNIM was slightly lower than that of the following GIPHM model, which had predicted RMSEs of about 26 and 23 km in the solar moderate and minimum year, respectively. Then, the ANNIM model was extended to a complete global three‐dimensional ionospheric model (ANNIM‐3D; Gowtam et al., 2019).…”