GNGTS 2022 - Atti del 40° Convegno Nazionale

120 GNGTS 2022 Sessione 1.2 identity line, the estimated regression coefficients, slope and intercept, must equal to 1 and 0, respectively. For this purpose, the t-statistic is used to check the null hypothesis that slopes of the linear models are equal to 1 and intercept are equal to 0. The probability values of the test are reported in Tab. 1. Fig. 1 - a) Forecast at 1-, 5-, and 10-step plotted against observed measurements for 100 predictions at the 14 series. b) Forecast of the global CO2 signal (calculated after Liuzzo et al. , 2013) at 1-, 5-, and 10-step plotted against obser- vations for 100 predictions. c) Portion of the global signal and forecasts at 1-, 5-, and 10-step. Tab. 1 - Values of width of windows for the average component ( m ) and for the stochastic components ( n ) used for the estimates of the forecasts for the single time series and for the global signal. All the indicators described in the text are also reported. Time series Combined global signal m n RMSE R 2 pval slo pval int RMSE R 2 pval slo pval int 1-day 2 1.2 0.072 0.831 0.022 0.008 0.143 0.615 0.052 0.107 5-days 6 1.2 0.077 0.806 0.046 0.104 0.206 0.105 8.1e-16 3.4e-15 10-days 11 1.2 0.082 0.786 0.371 0.793 0.171 0.424 3.7e-11 1.5e-8