Sunspot cycle prediction using multivariate regression and binary mixture of Laplace distribution model

Abstract

Prediction of sunspot cycle is a vital activity in space mission planning and various engineering decision making. In the present study, the sunspot cycle prediction has been carried out by a hybrid model which employs multivariate regression technique and the binary mixture of Laplace distribution (BMLD) function. The Expectation Maximization (EM) algorithm is being applied to the multivariate regression analysis to obtain a robust prediction of the sunspot cycle. Sunspot cycle 24 has been predicted using this technique. Multivariate regression model has been derived based on the available cycles 1 to 23. This model could predict cycle 24 as an average of previous cycles. Prediction from this model has been refined to capture the cycle characteristics such as bimodal peak at the high solar activity period by incorporating a predicted peak sunspot number from the BMLD model. This revised prediction has shown more accuracy in forecasting the major discrete features of sunspot cycle like maximum amplitude, the Gnevyshev gap, time duration from peak to peak amplitude, and the epoch of peak amplitude. This refined prediction shows that cycle 24 will be having a peak amplitude of 78 with an uncertainty of ±25. Moreover, the present forecast says that, cycle 24 will be having double peak with a strong second peak compared to the first peak. This hypothesis is found to be true with the realized data of cycle 24. Further, this techniques have been validated by predicting sunspot cycles 22 and 23. A preliminary level prediction of sunspot cycle 25 also been carried out using the technique presented here. Present study predicts that, cycle 25 also will be a modest cycle like the present cycle 24, and the peak amplitude may vary in a band of 75-95.