Planetary Kp index forecast using autoregressive models

Abstract

The geomagnetic Kp index is derived from the K index measurements obtained from thirteen stations located around the Earth geomagnetic latitudes between $48^\circ$ and $63^\circ$. This index is processed every three hours, is quasi-logarithmic and estimates the geomagnetic activity. The Kp values fall within a range of 0 to 9 and are organized as a set of 28 discrete values. The data set is important because it is used as one of the many input parameters of magnetospheric and ionospheric models. The objective of this work is to use historical data from the Kp index to develop a methodology to make a prediction in a time interval of at least three hours. Five different models to forecast geomagnetic indices Kp and ap are tested. Time series of values of Kp index from 1932 to 15/12/2012 at 21:00 UT are used as input to the models. The purpose of the model is to predict the three measured values after the last measured value of the Kp index (it means the next 9 hours values). The AR model provides the lowest computational cost with satisfactory results. The ARIMA model is efficient for predicting Kp index during geomagnetic disturbance conditions. This paper provides a quick and efficient way to make a prediction of Kp index, without using satellite data. Although it is reported that the forecast results are better when satellite data are used. In the literature we find that the linear correlation between predicted values and actual values is $77\%$, which is better than the $68.5\%$ obtained in this work. However, taking into account that our results are based only on Kp stochastic time series, the correlation value can be considered satisfactory.