Using atmospheric CO₂ to benchmark the performance of carbon cycle feedbacks in Earth System models can help identify new emergent constraints and reduce large uncertainties in climate projections. However, previous benchmarking efforts have often been limited to specific variables and/or to a subset of models. Recently, the development of an open source and modular community-based model evaluation system, the International Land Model Benchmarking (ILAMB) project, has provided a means for comprehensive land-model benchmarking. ILAMB includes observations across a range of spatial scales that constrain carbon, hydrology, and energy fluxes in model evaluations.

Here, we systematically evaluate terrestrial carbon fluxes from CMIP5 and nascent CMIP6 models using CO₂ benchmarks within ILAMB. For this purpose, we first developed several diagnostics to use atmospheric CO₂ mole fraction at surface sites in the NOAA ESRL network to evaluate terrestrial carbon fluxes at regional to global scale. Further, we used a pulse-response function that leverages pre-computed atmospheric transport to emulate seasonal and interannual CO₂ dynamics from any gridded land flux product. Using ILAMB, we then evaluated the patterns of atmospheric CO₂ variability arising from land model simulations in terms of amplitude, seasonal phasing, interannual variability, and its relationship to climate variability. Finally, individual model performance in CO₂ diagnostics was put into context by using other variables within ILAMB, including fine scale observations of Net Ecosystem Exchange and gridded data related to ecosystem productivity. The systematic benchmarking of land models will aid both model development and our understanding of the interactions between carbon cycle and climate change.