Forecasting Global Annual Average CO2 Concentrations

This study aims to enhance the accuracy of CO₂ level forecasts, compare the efficacy of different predictive models, and provide insights for policy development. Employing time series and regression analysis techniques, the study uses historical data from global monitoring stations (1979- 2022) to model the annual mean concentration of atmospheric CO2 The results reveal that the ARIMA (1,1,1) model outperforms the simple linear regression model in predictive accuracy. Nevertheless, the regression model came across a technical problem as residuals are significantly autocorrelated. The Augmented Dickey-Fuller test was applied to ensure stationarity of the first difference of the original series. The model was trained using data from 1979 to 2022 and validated for 2023. The errors of the ARIMA(1,1,1) was found to be white noise. The ARIMA model projected CO₂ concentrations of 419.5, 421.8 and 424.2 for the years 2023, 2024, and 2025 respectively, with a percentage error of just 0.048% for the 2023. In contrast, the corresponding percentage of error for the simple linear regression model was -1.236%. These findings underscore the ARIMA model’s superior performance in forecasting future CO₂ levels and its suitability for environmental monitoring and climate change mitigation strategies. This research provides a valuable methodological framework for future atmospheric science studies and informs policy decisions aimed at addressing rising CO₂ concentrations. Keywords: ARIMA; CO₂; Forecasting; Regression; Time series