Seasonal cycles in total column CO2: Where does the model-observation mismatch come from?

Atmospheric inversions using a network of surface flask measurements of carbon dioxide are routinely done to estimate the surface flux of the gas, for example [W. Peters et al, PNAS 2007]. The large-scale characteristics of the carbon dioxide mixing ratio in the surface layer are well-constrained by surface flask measurements, and therefore most transport models -- using optimized surface fluxes from atmospheric inversions -- faithfully reproduce spatiotemporal features of the surface layer carbon dioxide concentration, such as the phase and amplitude of the seasonal cycle. In contrast, the phase and amplitude of the total column carbon dioxide mixing ratio are not well-reproduced by most transport models -- the amplitude, in particular, is underestimated [Z. Yang et al, GRL 2007]. Consequently, atmospheric inversions using total column measurements result in erroneous estimates of surface fluxes. Given the recent initiatives to measure carbon dioxide using satellites and surface FTS stations -- both of which report total column mixing ratios -- this is a serious limitation of most transport models that needs to be overcome to effectively use satellite and surface FTS products for atmospheric inversions. In this work, we first demonstrate the general disagreement between models and measurements of total column carbon dioxide mixing ratio using six different transport models and three different surface FTS stations. Then we focus on one specific transport model -- namely, TM5 -- to analyze the reasons behind this disagreement. We find two problematic regions in the atmosphere. Firstly, by comparing TM5 concentration fields with aircraft data and modeling studies [Andrews et al, JGR 2001], we see that TM5 has unrealistically short tracer transport times across the tropopause and within the stratosphere. This causes the seasonal cycle of carbon dioxide in the entire stratosphere to be out of phase with the surface layer, reducing the seasonal amplitude in the total column mixing ratio. Secondly, by comparing TM5 concentration fields with aircraft samples (such as CONTRAIL), we see that TM5 underestimates the amplitude of the seasonal cycle of carbon dioxide in the free troposphere. Both of these play into the disagreement between modeled and observed total column mixing ratio. Having identified two important factors behind the disagreement between modeled and observed total column carbon dioxide, we present an attempt to close the error budget by quantifying the influence of these factors, and discuss options for reducing their impact on the results of atmospheric inversions.