DST forecast
The models
The Dst prediction model consists of an ensemble of time-delay neural networks driven by measured solar wind at L1. The prediction lead time has two components: a variable propagation time from L1 to Earths bow shock, and the model lead time which is 1 hour. Thus, the maximum prediction lead time is 1 hour plus the propagation time.
The inputs to the model are solar wind magnetic field B and Bz (only Bz shown in plot), density n, speed V, and time of day and time of year. The 1-minute solar wind data are transformed into 1-hour value to match the 1-hour Dst interval. The models are run once per minute thus providing 60 predictions for every 1-hour interval. In the comparison with the observed Dst, only the predicted Dst with timestamp closest to the standard 1-hour Dst interval is used. In the plot, all bars showing predicted Dst, except the last bar, therefore comes from one prediction for each 1-hour interval. As the prediction lead time is variable and does not match the standard 1-hour Dst interval, the last bar corresponds to the 1-hour interval containing the last predicted timestamp. The last bar will therefore change in height during the passage of time.
References
- The Dst-prediction models have been described in Wintoft et al.[2018].
- The real-time solar wind data are obtained from SWPC.
- The real-time Dst index is obtained from World Data Center for Geomagnetism, Kyoto.
- The main development of the current Dst forecast model was carried out within the EU project PROGRESS, No. 637302.
- The models have been implemented at theESA SSA portal under the G-ESC, where additional verification is carried out in real-time.
- We are responsible for Regional Warning Center - Sweden, part of International Space Environment Service ISES.