Assessing an earth system model using data assimilation/flux-inversion models

To improve future climate predictions, it is essential that models realistically capture the present-day observed climate and its variability. Often natural carbon cycle mechanisms, such as land and ocean exchanges to the atmosphere, are a major contributor to uncertainties in climate projections from earth system models (ESMs). Fortunately, the recent advancement of numerical modeling techniques provides a new detailed and comprehensive view of the natural carbon cycle patterns and behavior that influence the climate system across the globe. In this study, we consider a land carbon model, SiB4, and two versions of the observational data assimilation (DA) and flux-inversion model, CarbonTracker. Using these carbon models, we can evaluate how well an ESM (here ESM2G) can reproduce the present-day carbon cycle annual cycle, inter-annual variability, and trends. As SiB4's net land flux has not been tested against global observations and DA/flux-inversion models, our evaluation of ESM2G also will extend to SiB4. We show that ESM2G and SiB4 overestimate the flux of CO₂ into the atmosphere each year over a 15-year period due to terrestrial carbon cycle processes in specific regions. This study demonstrates that both models' weak land carbon sink is mainly located in regions such as the African tropics and the Eurasian boreal and temperate regions. We show ESM2G estimates an unrealistically large seasonality in the land carbon cycle found across global tropical regions and the Southern hemisphere. This paper demonstrates how well ESM2G simulates the terrestrial carbon cycle in the recent past as an indicator of how it might behave for future climate predictions. Additionally, this study suggests the comparisons of ESMs against these newly developed numerical carbon cycle models may be beneficial to future ESM inter-comparison projects.