Kp forecast
The models
The Kp predictions are generated from four different models with different lead times. Each 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. The four different models are targeted at 0, 1, 2, and 3 hours lead time. Thus, the maximum prediction lead time is 3 hours 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 3-hour values (not just simple averages) to match the 3-hour Kp interval. The models are run once per minute thus providing 180 predictions for every 3-hour interval. In the comparison with the observed Kp, only the predicted Kp with timestamp closest to the standard 3-hour Kp interval is used. In the plot, all bars showing predicted Kp, except the last bar, therefore comes from one prediction for each 3-hour interval. As the prediction lead time is variable and does not match the standard 3-hour Kp interval, the last bar corresponds to the 3-hour interval containing the last predicted timestamp. The last bar will therefore change in height during the passage of time.
References
- The Kp-prediction models have been described in Wintoft et al. [2017] and further verification in Wintoft et al. [2018].
- The real-time solar wind data are obtained from SWPC.
- The real-time Kp index is obtained from GFZ.
- The main development of the current Kp forecast models was carried out within the EU project PROGRESS, No. 637302.
- The models have been implemented at the ESA 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.