Search

We present a new model for the probability that the disturbance storm time (Dst) index exceeds −100 nT, with a lead time between 1 and 3 days. Dst provides essential information about the strength of the ring current around the Earth caused by the protons and electrons from the solar wind, and it is routinely used as a proxy for geomagnetic storms. The model is developed using an ensemble of Convolutional Neural Networks that are trained using Solar and Heliospheric Observatory (SoHO) images (Michelson Doppler Imager, Extreme ultraviolet Imaging Telescope, and Large Angle and Spectrometric Coronagraph). The relationship between the SoHO images and the solar wind has been investigated by many researchers, but these studies have not explicitly considered using SoHO images to predict the Dst index. This work presents a novel methodology to train the individual models and to learn the optimal ensemble weights iteratively, by using a customized class-balanced mean square error (CB-MSE) loss function tied to a least-squares based ensemble. The proposed model can predict the probability that Dst < −100 nT 24 hr ahead with a True Skill Statistic (TSS) of 0.62 and Matthews Correlation Coefficient (MCC) of 0.37. The weighted TSS and MCC is 0.68 and 0.47, respectively. An additional validation during non-Earth-directed CME periods is also conducted which yields a good TSS and MCC score.

We present a new model for the probability that the disturbance storm time (Dst) index exceeds −100 nT, with a lead time between 1 and 3 days. Dst provides essential information about the strength of the ring current around the Earth caused by the protons and electrons from the solar wind, and it is routinely used as a proxy for geomagnetic storms. The model is developed using an ensemble of Convolutional Neural Networks that are trained using Solar and Heliospheric Observatory (SoHO) images (Michelson Doppler Imager, Extreme ultraviolet Imaging Telescope, and Large Angle and Spectrometric Coronagraph). The relationship between the SoHO images and the solar wind has been investigated by many researchers, but these studies have not explicitly considered using SoHO images to predict the Dst index. This work presents a novel methodology to train the individual models and to learn the optimal ensemble weights iteratively, by using a customized class-balanced mean square error (CB-MSE) loss function tied to a least-squares based ensemble. The proposed model can predict the probability that Dst < −100 nT 24 hr ahead with a True Skill Statistic (TSS) of 0.62 and Matthews Correlation Coefficient (MCC) of 0.37. The weighted TSS and MCC is 0.68 and 0.47, respectively. An additional validation during non-Earth-directed CME periods is also conducted which yields a good TSS and MCC score.

The ionosphere plays an important role in satellite navigation, radio communication, and space weather prediction. However, it is still a challenging mission to develop a model with high predictability that captures the horizontal-vertical features of ionospheric electrodynamics. In this study, multiple observations during 2005–2019 from space-borne global navigation satellite system (GNSS) radio occultation (RO) systems (COSMIC and FY-3C) and the Digisonde Global Ionosphere Radio Observatory are utilized to develop a completely global ionospheric three-dimensional electron density model based on an artificial neural network, namely ANN-TDD. The correlation coefficients of the predicted profiles all exceed 0.96 for the training, validation and test datasets, and the minimum root-mean-square error of the predicted residuals is 7.8 × 104 el/cm3. Under quiet space weather, the predicted accuracy of the ANN-TDD is 30%–60% higher than the IRI-2016 at the Millstone Hill and Jicamarca incoherent scatter radars. However, the ANN-TDD is less capable of predicting ionospheric dynamic evolution under severe geomagnetic storms compared to the IRI-2016 with the STORM option activated. Additionally, the ANN-TDD successfully reproduces the large-scale horizontal-vertical ionospheric electrodynamic features, including seasonal variation and hemispheric asymmetries. These features agree well with the structure revealed by the RO profiles derived from the FORMOSAT/COSMIC-2 mission. Furthermore, the ANN-TDD successfully captures the prominent regional ionospheric patterns, including the equatorial ionization anomaly, Weddell Sea anomaly and mid-latitude summer nighttime anomaly. The new model is expected to play an important role in the application of GNSS navigation and in the explanation of the physical mechanisms involved.

The ionosphere plays an important role in satellite navigation, radio communication, and space weather prediction. However, it is still a challenging mission to develop a model with high predictability that captures the horizontal-vertical features of ionospheric electrodynamics. In this study, multiple observations during 2005–2019 from space-borne global navigation satellite system (GNSS) radio occultation (RO) systems (COSMIC and FY-3C) and the Digisonde Global Ionosphere Radio Observatory are utilized to develop a completely global ionospheric three-dimensional electron density model based on an artificial neural network, namely ANN-TDD. The correlation coefficients of the predicted profiles all exceed 0.96 for the training, validation and test datasets, and the minimum root-mean-square error of the predicted residuals is 7.8 × 104 el/cm3. Under quiet space weather, the predicted accuracy of the ANN-TDD is 30%–60% higher than the IRI-2016 at the Millstone Hill and Jicamarca incoherent scatter radars. However, the ANN-TDD is less capable of predicting ionospheric dynamic evolution under severe geomagnetic storms compared to the IRI-2016 with the STORM option activated. Additionally, the ANN-TDD successfully reproduces the large-scale horizontal-vertical ionospheric electrodynamic features, including seasonal variation and hemispheric asymmetries. These features agree well with the structure revealed by the RO profiles derived from the FORMOSAT/COSMIC-2 mission. Furthermore, the ANN-TDD successfully captures the prominent regional ionospheric patterns, including the equatorial ionization anomaly, Weddell Sea anomaly and mid-latitude summer nighttime anomaly. The new model is expected to play an important role in the application of GNSS navigation and in the explanation of the physical mechanisms involved.

Magnetic reconnection is a fundamental process that quickly releases magnetic energy stored in a plasma. Identifying from simulation outputs where reconnection is taking place is nontrivial and, in general, has to be performed by human experts. Hence, it would be valuable if such an identification process could be automated. Here, we demonstrate that a machine-learning algorithm can help to identify reconnection in 2D simulations of collisionless plasma turbulence. Using a Hybrid Vlasov Maxwell model, a data set containing over 2000 potential reconnection events was generated and subsequently labeled by human experts. We test and compare two machine-learning approaches with different configurations on this data set. The best results are obtained with a convolutional neural network combined with an "image cropping"step that zooms in on potential reconnection sites. With this method, more than 70% of reconnection events can be identified correctly. The importance of different physical variables is evaluated by studying how they affect the accuracy of predictions. Finally, we also discuss various possible causes for wrong predictions from the proposed model.

Magnetic reconnection is a fundamental process that quickly releases magnetic energy stored in a plasma.Identifying, from simulation outputs, where reconnection is taking place is non-trivial and, in general, has to be performed by human experts. Hence, it would be valuable if such an identification process could be automated. Here, we demonstrate that a machine learning algorithm can help to identify reconnection in 2D simulations of collisionless plasma turbulence. Using a Hybrid Vlasov Maxwell (HVM) model, a data set containing over 2000 potential reconnection events was generated and subsequently labeled by human experts. We test and compare two machine learning approaches with different configurations on this data set. The best results are obtained with a convolutional neural network (CNN) combined with an 'image cropping' step that zooms in on potential reconnection sites. With this method, more than 70% of reconnection events can be identified correctly. The importance of different physical variables is evaluated by studying how they affect the accuracy of predictions. Finally, we also discuss various possible causes for wrong predictions from the proposed model.

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 andWeddell 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.

There are remarkable ionospheric discrepancies between space-borne (COSMIC) measurements and ground-based (ionosonde) observations, the discrepancies could decrease the accuracies of the ionospheric model developed by multi-source data seriously. To reduce the discrepancies between two observational systems, the peak frequency (foF2) and peak height (hmF2) derived from the COSMIC and ionosonde data are used to develop the ionospheric models by an artificial neural network (ANN) method, respectively. The averaged root-mean-square errors (RMSEs) of COSPF (COSMIC peak frequency model), COSPH (COSMIC peak height model), IONOPF (Ionosonde peak frequency model) and IONOPH (Ionosonde peak height model) are 0.58 MHz, 19.59 km, 0.92 MHz and 23.40 km, respectively. The results indicate that the discrepancies between these models are dependent on universal time, geographic latitude and seasons. The peak frequencies measured by COSMIC are generally larger than ionosonde's observations in the nighttime or middle-latitudes with the amplitude of lower than 25%, while the averaged peak height derived from COSMIC is smaller than ionosonde's data in the polar regions. The differences between ANN-based maps and references show that the di

Many thermospheric mass density (TMD) variations have been recognized in observations and physical simulations; however, their impact on the low-Earth-orbit satellites has not been fully evaluated. The present study investigates the quantitative impact of periodic spatiotemporal TMD variations modulated by the empirical DTM2013 model. Also considered are two small-scale variations, that is, the equatorial mass anomaly and the midnight density maximum, which are reproduced by the Thermosphere-Ionosphere-Electrodynamics General Circulation Model. This investigation is performed through a 1-day orbit prediction (OP) simulation for a 400-km circular orbit. The results show that the impact of TMD variations during solar maximum is 1 order of magnitude larger than that during solar minimum. The dominant impact has been found in the along-track direction. Semiannual and semidiurnal variations in TMD exert the most significant impact on OP among the intra-annual and intradiurnal variations, respectively. The zero mean periodic variations in TMD may not significantly affect the predicted orbit but increase the orbital uncertainty if their periods are shorter than the time span of OP. Additionally, the equatorial mass anomaly creates a mean orbit difference of 50 m (5 m) with a standard deviation of 30 m (3 m) in 1-day OP during high (low) solar activity. The midnight density maximum exhibits a stronger impact in the order of 150±30 and 15±6 m during solar maximum and solar minimum, respectively. This study makes clear that careful selection of TMD variations is of great importance to balance the trade-off between efficiency and accuracy in OP problems.

Numerical simulations have been performed on Marconi at CINECA (Italy) under the ISCRA initiative. The corresponding data can be found at: https://doi.org/10.5281/zenodo.3935887

The codes and their instructions referring to a journal article titled 'Identifying magnetic reconnection in 2D Hybrid Vlasov Maxwell simulations with Convolutional Neural Networks'.

Many thermospheric mass density (TMD) variations have been recognized in observations and physical simulations; however, their impact on the low-Earth-orbit satellites has not been fully evaluated. The present study investigates the quantitative impact of periodic spatiotemporal TMD variations modulated by the empirical DTM2013 model. Also considered are two small-scale variations, that is, the equatorial mass anomaly and the midnight density maximum, which are reproduced by the Thermosphere-Ionosphere-Electrodynamics General Circulation Model. This investigation is performed through a 1-day orbit prediction (OP) simulation for a 400-km circular orbit. The results show that the impact of TMD variations during solar maximum is 1 order of magnitude larger than that during solar minimum. The dominant impact has been found in the along-track direction. Semiannual and semidiurnal variations in TMD exert the most significant impact on OP among the intra-annual and intradiurnal variations, respectively. The zero mean periodic variations in TMD may not significantly affect the predicted orbit but increase the orbital uncertainty if their periods are shorter than the time span of OP. Additionally, the equatorial mass anomaly creates a mean orbit difference of 50 m (5 m) with a standard deviation of 30 m (3 m) in 1-day OP during high (low) solar activity. The midnight density maximum exhibits a stronger impact in the order of 150±30 and 15±6 m during solar maximum and solar minimum, respectively. This study makes clear that careful selection of TMD variations is of great importance to balance the trade-off between efficiency and accuracy in OP problems.

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 andWeddell 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.

There are remarkable ionospheric discrepancies between space-borne (COSMIC) measurements and ground-based (ionosonde) observations, the discrepancies could decrease the accuracies of the ionospheric model developed by multi-source data seriously. To reduce the discrepancies between two observational systems, the peak frequency (foF2) and peak height (hmF2) derived from the COSMIC and ionosonde data are used to develop the ionospheric models by an artificial neural network (ANN) method, respectively. The averaged root-mean-square errors (RMSEs) of COSPF (COSMIC peak frequency model), COSPH (COSMIC peak height model), IONOPF (Ionosonde peak frequency model) and IONOPH (Ionosonde peak height model) are 0.58 MHz, 19.59 km, 0.92 MHz and 23.40 km, respectively. The results indicate that the discrepancies between these models are dependent on universal time, geographic latitude and seasons. The peak frequencies measured by COSMIC are generally larger than ionosonde's observations in the nighttime or middle-latitudes with the amplitude of lower than 25%, while the averaged peak height derived from COSMIC is smaller than ionosonde's data in the polar regions. The differences between ANN-based maps and references show that the discrepancies between two ionospheric detecting techniques are proportional to the intensity of solar radiation. Besides, a new method based on the ANN technique is proposed to reduce the discrepancies for improving ionospheric models developed by multiple measurements, the results indicate that the RMSEs of ANN models optimized by the method are 14–25% lower than the models without the application of the method. Furthermore, the ionospheric model built by the multiple measurements with the application of the method is more powerful in capturing the ionospheric dynamic physics features, such as equatorial ionization, Weddell Sea, mid-latitude summer nighttime and winter anomalies. In conclusion, the new method is significant in improving the accuracy an

Magnetic reconnection is a fundamental process that quickly releases magnetic energy stored in a plasma.Identifying, from simulation outputs, where reconnection is taking place is non-trivial and, in general, has to be performed by human experts. Hence, it would be valuable if such an identification process could be automated. Here, we demonstrate that a machine learning algorithm can help to identify reconnection in 2D simulations of collisionless plasma turbulence. Using a Hybrid Vlasov Maxwell (HVM) model, a data set containing over 2000 potential reconnection events was generated and subsequently labeled by human experts. We test and compare two machine learning approaches with different configurations on this data set. The best results are obtained with a convolutional neural network (CNN) combined with an 'image cropping' step that zooms in on potential reconnection sites. With this method, more than 70% of reconnection events can be identified correctly. The importance of different physical variables is evaluated by studying how they affect the accuracy of predictions. Finally, we also discuss various possible causes for wrong predictions from the proposed model.

Magnetic reconnection is a fundamental process that quickly releases magnetic energy stored in a plasma. Identifying from simulation outputs where reconnection is taking place is nontrivial and, in general, has to be performed by human experts. Hence, it would be valuable if such an identification process could be automated. Here, we demonstrate that a machine-learning algorithm can help to identify reconnection in 2D simulations of collisionless plasma turbulence. Using a Hybrid Vlasov Maxwell model, a data set containing over 2000 potential reconnection events was generated and subsequently labeled by human experts. We test and compare two machine-learning approaches with different configurations on this data set. The best results are obtained with a convolutional neural network combined with an "image cropping"step that zooms in on potential reconnection sites. With this method, more than 70% of reconnection events can be identified correctly. The importance of different physical variables is evaluated by studying how they affect the accuracy of predictions. Finally, we also discuss various possible causes for wrong predictions from the proposed model.