Are global km-scale climate models becoming indistinguishable from observations?

Simulating global climate has been a challenge and aspiration ever since the advent of numerical modeling. Today, global climate models have become essential tools to understand the climate system, project future changes, and inform mitigation and adaptation decisions. In that, they build on a long history of development, from the first attempts to couple atmospheric and ocean models in the late 1960s, to the emergence of Earth system models in the 2000s, and the development of the first km-scale models today.

In this talk, we show that the latest models provide global climate information with previously unprecedented accuracy. The two next-generation km-scale models included in our analysis (ICON Sapphire and IFS) even simulate temperature fields indistinguishable from observation-based references for the first time. We place this step-change in model fidelity in the context of nine observation-based datasets (20CR, ERA40, ERA-Interim, ERA5, JRA55, MERRA, MERRA2, NCAR-NCEP) and over 150 global climate models developed over the past three decades (from CMIP2 to CMIP6) in an extensive model evaluation. Based on this comparison, we discuss emerging challenges for model evaluation as the choice of the reference dataset starts to dominate model error for the latest models.