Secular behavior of solar magnetic activity: Nonstationary time-series analysis of the sunspot record

The sunspot record of solar magnetic activity is studied as a nonstationary time series by means of a previously developed algorithm for treating perturbed dynamical systems. This approach incorporates secular changes into the modeling process through an external driving parameter, whose temporal behavior is shown to correspond in this case to the long-term trend of the sunspot record. Our method is able to reduce by approximately 13% the prediction error of this series when compared to the standard stationary approach. Such a reduction is remarkable in view of the benchmark status of the sunspot record in the statistical literature and, moreover, the fact that this gain is obtained over the performance of an already very competitive modeling technique based on ensembles of artificial neural networks.