An exploratory analysis of long-term trends in atmospheric CO₂ concentrations

A new methodological approach for the analysis of monitoring data is discussed. The main ideas are illustrated for the example of the CO₂ problem. The analysis of CO₂ concentrations obtained from a global network of monitoring stations permitted us to construct a nonparametric evaluation of the spatial-temporal distribution of this field. We propose a parabolic parameterization of the long-term tendency of this field as a function of time (in one-year time steps). A function of the predictive ability of a model is defined on the basis of the technique of “supervised training.” This function is computed for a parabolic model and it is shown that this model constructed for the first 15 years of observations evaluates the tendency for the next 15 years quite well. The main problem that we solve in this paper is how to correlate the projections of different models for the carbon cycle and different scenarios of the annual release of carbon into the atmosphere with the projections that reflect parameterization of the trends of CO₂-monitoring data.