The degree of perturbation of the magnetosphere is usually evaluated with the help of special geomagnetic indices. Planetary perturbations during geomagnetic storms are measured by the Dst index, which is the deviation of variation of the magnetic field from the undisturbed level, averaged over the values measured at the control chain of magnetic stations located in the low latitudes. The global source of information about the measured values of the Dst index is the World Data Center) in Kyoto (Japan).

As the main cause of disturbances of the magnetosphere are the plasma fluxes originating from the solar corona (the solar wind, SW), the input data used for forecasting of the values of the Dst index are SW plasma parameters and the interplanetary magnetic field (IMF), measured by the ACE (Advanced Composition Explorer) spacecraft.

To predict the hourly values of the Dst index, artificial neural networks (ANN) of perceptron type are used. ANN belong to the family of adaptive algorithms. This means that in their work, ANN do not rely on any physical model, but only on the characteristics of the data. Due to exceptional complexity of the subject of the research, at present time there are no physical models, which could carry out forecasting of the values of the Dst index with reasonable accuracy. ANN learn the necessary information from historical data, in the process of learning by examples.

The engineering model of space environment presented here is intended for prediction of hourly average values of the Dst index. It uses the following data:
? data on the values of IMF ? its components (Bx, By, Bz) and |B| (IMF module);
? data on the parameters of SW plasma: SW speed (v), the density of protons (np), the translational temperature of protons (T);
? data on the previously measured values of the Dst index.

In the ?Analysis? section of this site there is a ready-made template ?Prediction of geomagnetic indices?, which will allow you to view all the data used to forecast the Dst index.

To ensure the possibility to take into account the history of changes of the parameters, the so-called delay embedding of time series is used. This means that in addition to the current values of the parameters, the neural network is fed with the hourly data about the values of the specified parameters during the latest 24 hours.

Loading of the data on the values of the Dst index and forecast update are performed twice per hour. At the 33rd minute of every hour, the preliminary value for the current hour is loaded, at the 3rd minute of the next hour it is replaced by the final one. The forecast is carried one hour ahead in respect to the last loaded value. Thus, the forecast horizon is from 0.5 to 1.5 hours in respect to the current moment of time. Historical data are shown on the graph in red, the preliminary value for the current hour in blue, the forecast of the model in green.

By default, the graph of changes of the Dst index is presented for the last 24 hours, together with the forecast of the model. To see the changes over a longer time period select the desired interval with the help of the corresponding radio buttons. The graph will be redrawn automatically.