Proceedings

Proceedings are still available from me (hlundstedt@solar.stanford.edu or henrik@lund.irf.se).

The following report was published in STEP International, Vol. 4, No. 4, April 1994.

The first international workshop on the use of artificial intelligence (AI) techniques in solar-terrestrial physics was held in Lund, Sweden, September 22-24, 1993. The workshop was sponsored by NOAA Space Environment Laboratory, the NASA/Goddard Space Flight Center, Laboratory for Extra- terrestrial Physics, the Swedish Science Research Council, and Lund University. The workshop general chairman was Henrik Lundstedt of Lund University.

The workshop's three-day schedule included sessions on theory, applications and data bases, demonstrations, and ample time for discussion. the sessions consisted of invited reviews, short contributions, computer presentations, and poster. The value of data access via Internet was shown when YOHKOH images, Stanford Solar Magnetograms, and NOAA's solar and geophysical images were displayed and discussed on the final day of the workshop.

The majority of presentations of the presentations at the workshop dealt with neural network applications of Multi-Layer-Error-Back-Propagation (MLBP) and Self-Organizing Map (SOM) neural nets, and with traditional expert systems and fuzzy expert systems. It was, however, concluded by John Freeman that it is the complete set of AI tools that can lead to successful missions. An AI system should be designed to make use of the optimum tools for the job. The system may even need to react in real-time by changing tools as the situation changes. Ludwik Liszka exemplified the use of hybrid neural networks, consisting of a SOM and a MLBP net. Neither were genetic algorithms in optimizing MLBP networks nor were the combination of fuzzy systems and neural networks such as simplified Fuzzy ART Maps exemplified. The solar- terrestrial community clearly enters a new and exciting area of research by using AI tools.

The following areas where AI tools could be used, and are used, were demonstrated:

Jo Ann Joselyn discussed how AI could be used by the NOAA Space Environment Operations Center.

Robert L. McPherron exemplified how expert systems and fuzzy expert systems could be used to assist space weather forecasters by monitoring the quality of the input data streams, activating alarms when certain values exceed a specific threshold.

Ata Etemedi showed how AI tools could be used for detecting geophysical events thereby reducing the enormous amount of data involved and, possibly, achieving video-rate performance at telemetering.

Several papers dealt with the classification of structures; for example, in the solar wind (Predeep, Gothakaar and Shyam Khoragade, and Ralph Snel and Henrik Lundstedt), in solar radio spectrograms (Andre Csillaghy and Hans Hinterberger), and in VHF electromagnetic pulses primarily due to lightning (Kurt Moore el al.).

A major subject in solar-terrestrial physics is forecasting or predicting space weather. The prediction of solar flares and solar activity were discussed by David Shaw, Takehiko Aso, Shinichi Watari, and Ludwik Liszka.

The majority of papers were devoted to the prediction of geomagnetic activity. Neural networks predictions were demonstrated by T. Detman et al., J. Freeman et al., A.W.P. Thompson and P. Wintoft, and H. Lundstedt, and R. Snel. Non-neural net predictions were presented by D. Vassiliadis, R.L. McPherron, and Todd Hoeksema. G.A. Stringer and R.L. McPherron presented neural network predictions of relativistic electrons at geosynchronous orbits - a subject of current interest in connection with the damage that recently occurred with the two Canadian Anik spacecraft.

Three papers by David Boteler, Ari Viljanen, and H. Lundstedt dealt with the prediction of geomagnetically-induced currents. It was interesting to note that potential users of space weather forecasts were represented including attendees from power companies in Sweden and Canada showing the important role solar-terrestrial physics plays in our society.

The workshop proceedings, edited by Jo Ann Joselyn, H. Lundstedt, and Joanna Trolinger, were published in December 1993. The papers demonstrate new directions for scientific studies. We anticipate that these techniques will be eagerly tested and developed in the wider community; the possibilities are limitless.