Realism of simulated TOA radiation imbalance

A realistic representation of the top-of-atmosphere (TOA) radiation imbalance by fully coupled global climate models (GCMs) is key for the models climate sensitivity and projections of global and local warming. Recent research showed that atmospheric model simulations forced with observed sea surface temperature (SST) patterns realistically simulate the TOA radiation imbalance but it is unclear whether this holds for fully coupled GCMs with free air-sea interaction and persistent SST mean state biases. To justly evaluate the consistency of GCM simulations with the observed TOA imbalance, we use initial-condition large ensembles of seven GCMs simulating year 2000-2020 with 30-100 members each. This approach accounts for internal variability and model biases in TOA radiation and freely evolving SST. We find that in the global mean, the GCMs simulate the observed interannual variability in TOA net radiation but underestimate the observed trend. Regionally, the ensemble simulations largely cover the observed trends but have consistent biases in some regions such as the East Pacific and the Southern Ocean. For these regions, some GCMs have a few individual ensemble members that match the observed trends in SSTs and also simulate the observed trends in TOA net radiation correctly. This suggests that some GCMs are able to realistically simulate the TOA radiation imbalance and that inconsistencies in TOA net radiation are caused by biases in surface forcing. Our findings confirm the key role of simulating ocean-atmosphere interactions correctly and imply that GCMs underestimate the (local) sensitivity of the pattern effect.