Global Atmospheric CO2 Simulations with the IAP-AACM Model Using an Improved Vertical Diffusion Scheme and Evaluation with Multi-Source Data

Accurately simulating the spatiotemporal distribution of global atmospheric CO2 remains challenging yet essential for reducing uncertainties in carbon source-sink inversions, quantifying the climate effects of heterogeneous CO2 fields, and supporting the development of CO2 observation networks. In this study, we simulated global atmospheric CO2 concentrations (2010–2019) at a horizontal spatial resolution of 1° × 1° using the Aerosol and Atmospheric Chemistry Model of the Institute of Atmospheric Physics (IAP-AACM) without data assimilation, with initial fields and flux data from the CarbonTracker CT2022 (CT2022) reanalysis product. The simulations were comprehensively evaluated against CT2022 and observations from ground-based (NOAA GML), airborne (ObsPack), and satellite (OCO-2) platforms. The results indicate that across all evaluated surface stations, CT2022 exhibits poorer overall statistical performance (R = 0.69, RMSE = 5.37 ppm, MB = 2.05 ppm) primarily due to noticeable overestimations at unassimilated ground stations, while IAP-AACM maintains robust performance across the surface network (R = 0.84, RMSE = 2.62 ppm, MB = 0.28 ppm). Vertically, airborne observations across eight global campaigns confirm that IAP-AACM accurately reproduces the vertical distribution of CO2, maintaining strong correlations (R = 0.72–1.00) and performance comparable to the CT2022 reanalysis (R = 0.86–1.00). In terms of total column CO2 concentrations (XCO2), IAP-AACM exhibits strong agreement with satellite retrievals annually (R = 0.97, RMSE = 1.08 ppm, MB = 0.26 ppm), with seasonal metrics remaining consistently robust across all four seasons (R = 0.96–0.97, RMSE = 0.98–1.20 ppm, MB = 0.13–0.37 ppm), demonstrating large-scale transport fidelity on par with the CT2022 reanalysis. Finally, across representative ObsPack land sites, unassimilated IAP-AACM achieves a high median correlation (R = 0.97), low error (RMSE = 2.01 ppm), and low mean bias (MB = −0.45 ppm), closely approaching the assimilated CT2022 reanalysis product (R = 0.98, RMSE = 1.40 ppm, MB = −0.07 ppm). Further analysis indicates that the optimized IAP-AACM exhibits robust performance under stable boundary layer conditions, where the revised diffusion scheme produces higher vertical diffusion coefficients that help mitigate excessive near-surface CO2 accumulation during nighttime. Overall, the optimized IAP-AACM effectively simulates the spatiotemporal distribution of global atmospheric CO2, serving as a reliable tool to support advanced research.