Modeling total solar irradiance from PMOD composite using feed-forward neural networks

The variability of the solar activity dominates the variability of the earth's atmosphere, which affects human life and technology on earth. To understand the effects of solar activity on earth's atmosphere different efforts are underway to model the variations of total solar irradiance (TSI) associated to the variations of photometric sunspot index (PSI) and core to wing ratio of Mg II index, for example, linear regression approach. In this study, feed-forward neural networks (NNs) algorithm, which takes the non-linear relationship between the dependent and independent variables, has been implemented to model daily TSI using PSI and Mg II index. First, data between 1978 and 2008 have been used to train and validate NNs, through which the parameters such as weights and biases are estimated. Therefore, NNs has been used to predict TSI between the years 2008 and 2013 from test data. The output of NNs have been compared with PMOD composite TSI and result has shown good agreement. Linear correlation between NNs predicted TSI and PMOD composite is found to be about 0.9307 for the years between 1978 and 2013. This means that NNs predicted TSI from solar proxies explains about 86.6% of the variance of TSI for solar cycles 21–24, and over 90% during solar cycle 23. Predicting TSI using NNs further strengthens the view that surface magnetism indeed plays a dominant role in modulating solar irradiance.