Spectral analysis of the time series for average annual values of the globally averaged surface temperature anomaly shows the presence of harmonics of the lunar nodal cycle with a period of 18.6 years, which can be used to predict the values of the series. Three models of the series were considered: autoregression AR(p), combined model of autoregression – integrated moving average ARIMA(p,d,q) and artificial neural network. It is shown that the ARIMA(4,1,4) model gives the best results for predicting the global temperature anomaly for three years.
