PNNL

Abstract

This study assesses the relative importance of time integration error in present-day climate simulations conducted with the atmosphere component of the Energy Exascale Earth System Model version 1 (EAMv1). We show that a factor-of-6 reduction of time step sizes in all major components leads to significant changes in the long-term mean climate. Examples of such changes include a low-latitude warming in the near-surface levels and cooling aloft, as well as drying throughout the troposphere accompanied by cloud fraction decreases. These changes correspond to a non-negligible degradation in the agreement between the simulated and observed climate statistics, suggesting that sizable time integration errors are compensated by other errors or by parameter tuning in the default model. Attribution of the time-step sensitivities is done by separately shortening time steps used in various components of EAM or by revising the numerical coupling between processes. The step sizes used by the dynamical core and the radiation parameterization are found to have relatively small impacts. The step size used by the collectively sub-cycled stratiform cloud macro- and microphysics parameterizations appears to be the primary factor affecting cloud fraction and the longwave cloud radiative effect in the upper troposphere. The coupling of these cloud parameterizations with the rest of EAM has a marked impact on the cloud fraction and cloud radiative effects of the subtropical marine stratocumulus. The step sizes used by other processes (including deep convection) and their coupling have moderate impacts on high clouds in the tropics and over the storm tracks as well as trade cumulus in the subtropics. Additional analyses also reveal that the changes in the subtropics are primarily thermodynamic responses of the clouds and their atmospheric environment; circulation changes are negligible. These results provide clues for follow-up investigations that aim at understanding the root causes of the time-step sensitivities and reducing time integration errors in the model.