Different methods of modeling the variability in the monthly mean concentrations of atmospheric CO2 at Mauna Loa

Abstract

A detailed analysis of the time series of the monthly average atmospheric CO2 concentrations from March 1958 to December 1985 has been carried out. Seasonal decomposition of the CO2 time series has been done in order to obtain estimations of the trend, seasonal, and irregular components. The seasonal variation includes a cycle of 6‐month period superimposed to the strong annual cycle. These two harmonic explain more than 95% of the seasonal variation of the CO2 concentration. Three different models for the atmospheric CO2 concentrations have been analyzed: univariate stochastic model, curve fitting and a model with exogenous variables. A quadratic function and a four‐harmonic Fourier series with the amplitude increasing in time have been fitted to the CO2 trend and seasonal component, respectively. Global monthly fossil fuel combustion data and the smoothed sea surface temperature off the Peruvian coast have been used as exogenous variables explaining the CO2 trend for the third model. Smoothed sea surface temperature cross‐correlated significantly with the anomalies of the time derivative of the CO2 trend at lags of 12 and 13 months. The results of the third model seem to confirm that changes in the upwelling and the CO2 outgassing from the eastern tropical Pacific Ocean are responsible of changes in the CO2 growth rate observed at Mauna Loa. It is estimated that the CO trend has risen about 0.5 ppm per El Niño event. The model with exogenous variables offered the closest fit to the measured data and seems to be the best model in forecasting the monthly average atmospheric CO2 concentration during the years 1986–1989.