[1] We have compared six Dst forecast models using 63 intense geomagnetic storms (Dst ≤ À100 nT) that occurred from 1998 to 2006. For comparison, we estimated linear correlation coefficients and RMS errors between the observed Dst data and the predicted Dst during the geomagnetic storm period as well as the difference of the value of minimum Dst (DDst min ) and the difference in the absolute value of Dst minimum time (Dt Dst ) between the observed and the predicted. As a result, we found that the model by Li (2002, 2006) gives the best prediction for all parameters when all 63 events are considered. The model gives the average values: the linear correlation coefficient of 0.94, the RMS error of 14.8 nT, the DDst min of 7.7 nT, and the absolute value of Dt Dst of 1.5 hour. For further comparison, we classified the storm events into two groups according to the magnitude of Dst. We found that the model of Lee (2002, 2006) is better than the other models for the events having À100 ≤ Dst < À200 nT, and three recent models (the model of Wang et al. (2003), the model of Li (2002, 2006), and the model of Boynton et al. (2011b)) are better than the other three models for the events having Dst ≤ À200 nT.

[1] We have made a comparison of near-real time Kp forecast models based on neural network (NN) and support vector machine (SVM) algorithms. For this, we consider four models as follows: (1) a NN model using solar wind data, (2) a SVM model using solar wind data, (3) a NN model using solar wind data and preliminary Kp values from ground-based magnetometers, and (4) a SVM model using the same input data used in model 3. For the comparison, we estimate the correlation coefficients and the RMS errors, and the mean absolute errors between the observed Kp and the predicted one. As a result, we find that model 3 shows the best performance. The correlation coefficients, the RMS error, and the mean absolute error of model 3 are 0.93, 0.48, and 0.61, respectively. For the forecast evaluation of high magnetic activity occurrence for the four models, we present contingency tables and their statistical parameters such as probability of detection yes, false alarm ratio, bias, critical success index, and true skill statistic. From a comparison of these statistical parameters, we find that model 3 is superior to the other models for the forecasting of high magnetic activities (Kp 6).

The total electron content (TEC) is the total number of electrons along a path between a radio transmitter and a receiver. The unit of TEC (TECU) is defined as 10 16 electrons/m 2 , and 1 TECU corresponds to a 0.163 m range delay on L1 frequency. TEC is used as one of the major parameters in space weather and an indicator of ionospheric disturbance. The accurate measurement of TEC provides us with useful information on satellite communications, navigation, national defense, and aviation. The International Global Navigation Satellite Systems (GNSS) Service (IGS) TEC maps are calculated based on measured observational data at more than 380 IGS stations. They have been used as a reference data for comparison together with JASON satellite data (e.g.

In this study, we make a model, which is called DeepIRI, to generate improved International Reference Ionosphere (IRI) total electron content (TEC) maps by deep learning based on conditional Generative Adversarial Networks. For this we consider 48,901 pairs of IRI TEC maps and International Global Navigation Satellite Systems (GNSS) Service (IGS) TEC maps from 2001 to 2011 for training the model. We evaluate the model by comparing IGS TEC maps and DeepIRI TEC ones from 2013 to 2017. The DeepIRI TEC maps that our model generated are much more consistent with the corresponding IGS TEC maps than the IRI TEC ones. Especially, ionospheric peak structures are successfully generated in DeepIRI TEC maps while they are not in IRI-2016 ones. From the average differences between IRI and IGS TEC maps, our model greatly improved the IRI TEC at low-latitude region around the equatorial anomaly. These results show that our model can improve the global TEC prediction ability of the IRI-2016. Our study suggests a sufficient possibility to generate DeepIRI global TEC maps in near real time if IRI is generated in time. Our approach can be applied to make improved versions of empirical models if more realistic observations are available with a time delay. Plain Language SummaryThe International Reference Ionosphere (IRI) is one of the most widely used standard models of the ionosphere. The IRI total electron content (TEC) is very important to monitor ionospheric disturbances. This paper makes a model to improve IRI TEC maps by deep learning based on conditional Generative Adversarial Networks, which is a general purpose solution to resolve image-to-image translation problems. We consider many pairs of IRI TEC maps and IGS TEC maps for training the model. Our model greatly improves the global IRI TEC maps, which are very consistent with the observations.

No abstract