Evaluation of the Arctic surface radiation budget in CMIP5 models

The Arctic region is warming at a rate more than double the global average, a trend predicted to continue by all Coupled Model Intercomparison Project 5 (CMIP5) climate models. Despite this consistency, significant intermodel spread exists in the simulated Arctic climate related to differences in the Arctic surface radiation budget. Building upon previous work to characterize and understand surface radiation budget biases in climate models, the annual mean and seasonal cycle of the Arctic surface radiation budget in 17 CMIP5 models using the Historical‐forcing scenario is evaluated against state‐of‐the‐art Cloud and Earth's Radiant Energy System Surface Energy Balanced and Filled data. The CMIP5 multimodel ensemble is found to simulate longwave surface fluxes well during the sunlit months (~1 W m−2 differences in July) but exhibits significant wintertime biases (up to −19 W m−2). Shortwave fluxes show substantial across‐model spread during summer; the model standard deviation approaches 20 W m−2 in July. Applying a decomposition analysis to the cloud radiative effect (CRE) seasonal cycles, an unrealistic compensation is uncovered between the model‐simulated seasonal cycles of cloud fraction, all‐sky/clear‐sky flux differences, and surface albedo that enables models to simulate realistic CRE seasonal cycles with unrealistic individual contributions. This unrealistic behavior in models must be constrained to improve Arctic climate simulation; observational uncertainty is sufficient to do so. Lastly, biases in all and clear‐sky longwave downwelling fluxes positively correlate with model surface temperature in winter, while in summer surface temperature is most strongly related to clear‐sky upwelling radiation biases from surface albedo errors. Key Points Significant regional variations are found in Arctic surface radiation biases Unrealistic compensation contributes to realistic simulation of seasonal cycle Surface radiation biases influence regional surface temperature biases