The consistency between observations (TCCON, surface measurements and satellites) and CO₂ models in reproducing global CO₂ growth rate

Abstract. Atmospheric CO₂ growth is the primary driver of the global warming and the rate of this growth is a valuable indicator of the interannual changes in carbon cycle. Despite atmospheric CO₂ growth rate had been considered as the well-known quantity, the latest findings indicated that CO₂ models can considerably disagree in reproducing this rate. This study is aimed to advance our knowledge about temporal and spatial variations of annual CO₂ growth rate (AGR) by using CO₂ observations from the Total Column Observing Network (TCCON), CO₂ simulations from Carbon Tracker (CT) and Copernicus Atmospheric Monitoring System (CAMS) models being compared with the previously-reported global references of AGR from Global Carbon Budget (GCB) and satellite observations (SAT) for 2004–2019 years. TCCON and the CO₂ models revealed temporal AGR variations (AGRTCCON = 1.71–3.35 ppm, AGRCT = 1.64–3.15 ppm, AGRCAMS = 1.66–3.13 ppm) of very similar magnitude to the global CO₂ growth references (AGRGCB = 1.59–3.23 ppm, AGRSAT = 1.55–2.92 ppm). However, AGRTCCON estimates agree well with the references only during the 2010s (correlation coefficient, r = 0.68 vs GCB and r = 0.75 vs SAT) as the TCCON observational coverage has been substantially expanded since 2009. Moreover, AGRTCCON reasonably agrees (r = 0.67) with the strength of El-Nino Southern Oscillations (ENSO) in the 2010s. The highest atmospheric CO₂ growth (2015–2016) driven by the very strong El-Nino was accurately reproduced by TCCON which provided AGR of 2015–2016 years (3.29 ± 0.98 ppm) in very close agreement to the AGRSAT reference (3.23 ± 0.50 ppm). We further validated AGR simulations (CT and CAMS) versus the newly-acquired AGRTCCON (as point-location reference) for every TCCON site and found low agreement between the models and TCCON (r < 0.50) only at 3 out of 20 stations. This minor caveat has not affected the accuracy of global AGR simulations as they showed high agreement with SAT (r ≈ 0.76–0.78) and GCB (r ≈ 0.72–0.78) and reasonable agreement with TCCON (r = 0.65) global-scale references. The spatial correlation between CT and CAMS in simulating AGR (applied for every 3°×2° grid cell) is perfect (r = 0.99) for the modeling period (2004–2016). Similarly, land-wise intercomparison between CAMS and CT simulations of AGR yielded in perfect correlation for most MODIS land classes (median of land-dependent r > 0.98). From spatial perspective, the highest AGR estimates (> 20 % from the median) were observed in the regions of intense fossil fuel combustion (East Asia) or biomass burning (Amazon, Central Africa). Lack of ideal correlation and small disagreement between CT and CAMS (< 3.9 % difference between medians of global AGR estimates) are likely driven by the slight spatial disagreement between CT and CAMS in the aforementioned regions. To validate this statement, a sensitivity experiment is needed where in CO₂ inverse model, alongside with the current setup of a priori biomass burning fluxes, an alternative setup is assembled (multiple independent estimates of burned area and fire-dependent emission factors for various type of tropical fires can be used). In overall, our study showed that the current estimates of global atmospheric growth rate of CO₂ are consistent across a wide range of the different data sources and strengthening of carbon observational infrastructure (like covering more developing countries with ground-based CO₂ observations and providing more satellite CO₂ observations from cloudy and hazy regions) should improve the accuracy of CO₂ growth rate estimates on both local and global scales.