Henrik Lundstedt

Prediction of geomagnetic storms from solar wind data using Elman Recurrent Neural Networks

In order to accurately predict geomagnetic storms, we exploit Elman recurrent neural networks to predict the Dst index one hour in advance only from solar wind data. The input parameters are the interplanetary magnetic field z-component Bz (GSM), the solar wind plasma number density n and the solar wind velocity V. The solar wind data and the geomagnetic index Dst are selected from observations during the period 1963 to 1987, covering 8620h and containing 97 storms and 10 quiet periods. These data are grouped into three data sets; a training set 4877h, a validation set 1978h and a test set 1765h. It is found that different strengths of the geomagnetic storms are accurately predicted, and so are all phases of the storms. As an average for the out-of-sample performance, the correlation coefficient between the predicted and the observed Dst is 0.91. The predicted average relative variance is 0.17, i.e. 83 percent of the observed Dst variance is predictable by the solar wind. The predicted root-mean-square error is 16 nT.

Geomagnetic storm predictions from solar wind data with the use of dynamic neural networks

Dynamic neural networks have been shown as an encouraging alternative to traditional approaches for nonlinear temporal predictions. We use partially recurrent neural networks to study solar wind-magnetosphere coupling by predicting geomagnetic storms. The solar wind and Dst data used in this study are selected from the period 1963 to 1992. Statistical cross-correlation analyses and neural networks are applied to finding the best coupling functions. It is found that the results from both studies are consistent and that the coupling functions P 1/3 VB S , P 1/2 V B s , V 2 B S , VB S , and VB z are well suited for predicting geomagnetic storms. Also, the relative importance of solar wind parameters is investigated in detail, aimed at finding the optimal combinations of solar wind parameters for predicting geomagnetic storms. The two basic combinations giving accurate prediction are B s , n, V and B z , n, V. Addition of B, F, or B y can further improve predictions. We find that the best combination outperforms the best coupling functions in terms of prediction accuracy. Geomagnetic storms are very accurately predicted 1-2 hours in advance from the solar wind alone. The accurate predictions 1-2 hours ahead might well imply that the internal dynamic magnetospheric processes in forming geomagnetic storms occur on a timescale of about 1 hour. The predictions 3-5 hours ahead are useful in practical operation according to their acceptable accuracy. The prediction goodness is decreased with increasing prediction time. We consider this as a result of the fact that the internal magnetospheric dynamics limits the short-term prediction accuracy. The short-term predictability of geomagnetic storms and the validity of the results from cross-correlation analyses are discussed.

Response of the auroral electrojets to the solar wind modeled with neural networks

The dissipative processes in the Earths magnetosphere, such as the ring current and the auroral electrojets, depend on both the external solar wind forcing and factors internal to the magnetosphere. Previous studies have shown that artificial neural networks are able to compute the ring current index Dst very accurately from only solar wind data. In this study, we use neural networks to model the response of the auroral electrojets to the solar wind conditions. The solar wind input to the networks consist of 5-min averaged data from the Earth-orbiting spacecraft IMP 8, while the output is the auroral electrojet index AE. The relationships between the solar wind and the AE index, as modeled by the neural networks, are investigated in a parameter study. The relative importance of individual solar wind variables is studied, as well as the abilities of various coupling functions. It is shown that the use of individual solar wind variables as input to a neural network is superior to the use of corresponding coupling functions. The nonlinear neural networks are related to earlier linear techniques, and the abilities of linear networks (linear filters) are compared to those of nonlinear networks. It is found that a nonlinear network with n, V, By, and B z as input during 100 min can account for 76% of the variance (r≃0.87) in the AE index. No influence of B x is found. With the coupling function p 1/2 V 2 B S as input to a nonlinear network, 71% of the AE index variance is predicted. These results are averaged over a large test set (∼ 330 hours) of data not used to train the networks. The test data are from 1973-1974 and include a diverse set of conditions, ranging from almost quiet to exceptionally disturbed.

Neural networks and predictions of solar−terrestrial effects

Abstract Neural networks for predictions of solar-terrestrial effects, such as geomagnetic induced currents (GICs), are presented. The following assumptions are made: the geomagnetic activity is mainly controlled by the southward B z -component of the solar wind. There are three major solar causes of the southward B z -components, namely the solar sector boundaries (SSB), the coronal mass ejections (CME) and the coronal holes (CH). The mean GIC size has an exponential relation to the geomagnetic activity index K p . The neural networks were trained with solar input data from various U.S. data bases: SSB-data from CSSA, Stanford, California, CME-data and solar wind-data from NOAA/SEL, Boulder, Colorado and finally CH-data from SacPeak/AFGL.

A THIRD CATALOGUE OF HIGH-SPEED PLASMA STREAMS IN THE SOLAR WIND - DATA FOR 1978-1982

Interplanetary magnetic field data obtained from near-Earth spacecraft are used to compile a catalogue of high-velocity solar wind streams for the period June 1978 through October 1982 (Bartels rotations 1980–2039). The compilation of high-velocity streams is a continuation of two previous lists which covered the periods 1964–1975 and 1975–1978, respectively. The total number of high-speed solar wind streams listed in the catalogues is 637 of which 520 are corotating streams and 117 are classified as flare-associated streams.

A catalogue of high-speed plasma streams in the solar wind 1975–78

Interplanetary magnetic field data compiled by King (1977, 1979) are used to obtain a list of high velocity solar wind streams for the period January 1975 through June 1978. The catalogue is a continuation of a previous list which covered 1964–75. The total number of high-speed solar wind streams listed in the catalogues is 455.

Influence of interplanetary interaction regions on geomagnetic disturbances and tropospheric circulation

Abstract Proposed solar wind-magnetosphere energy coupling functions are studied. An empirical formula proposed by Svalgaard (1977) is found to predict the geomagnetic activity quite well. The influence of solar wind interaction regions on the tropospheric circulation, through a suggested cosmic ray mechanism, was investigated. The cosmic ray intensity at Earth clearly showed a decrease at the time of passage of an interaction region. It is suggested that the well-known dip in the Vorticity Area Index may be caused by an interaction-modulated decrease in cosmic ray intensity.

Solar origin of geomagnetic storms and predictions

Abstract Changes of the large-scale solar magnetic fields are described and related to the occurrence of solar coronal phenomena which are associated with geomagnetic storms. Only for the very largest geomagnetic storms is there agreement on the coronal origin. However, when and where coronal mass ejections occur are still very difficult questions to answer. Artificial neural networks have been used to forecast geomagnetic storms either from daily solar input data or from hourly solar wind data. With solar data as input, predictions one-three days or even a month in advance are possible, while using solar wind data as input predictions about an hour in advance are possible. The latter predictions have been very successful. Finally, the effects of geomagnetic storms on power and satellite systems are reviewed.

Magnetograph observations with the Swedish solar telescope on La Palma

A high-resolution videomagnetograph that records the images of opposite circular polarization simultaneously has been constructed for the Swedish vacuum solar telescope at La Palma. Magnetograms are obtained by off-line integration of bursts consisting of typically 50 frames of 20 ms exposures, with bad frames rejected, and the frame-to-frame image motion of the remaining frames compensated for by cross-correlation techniques. The short exposures combined with frame selection and elimination of image motion optimizes the resolution and thereby also the S/N, allowing good magnetograms to be obtained with an effective exposure time of less than 1 s at an image scale of 0.1″ pixel−1. The advantages and limitations of the system are discussed and compared with other techniques of making filter magnetograms are discussed.

The coronal source of the slow solar wind

Periods of very low solar wind velocity at 1 AU, during the interval from 1977 to 1983, are identified and mapped back to the coronal source surface at 2.5 R⊙. In total 25 such low-velocity events were found. Inferred source locations were characterized with respect to their position relative to the coronal neutral line. The study showed that in 17 out of 25 cases the slow solar wind originated across a coronal neutral line. In the remaining cases the source was either along the neutral line or insides a warp. A prediction of the IMF polarity to be expected at Earth, from the computed coronal magnetic field, was also done. It failed clearly only in four cases out of 25 events. In three cases the prediction was uncertain because of missing data. Possible explanations of why the potential model sometimes predicts a wrong polarity are discussed.