The Impact of Radiative Transfer at Reduced Spectral Resolution in Large‐Eddy Simulations of Convective Clouds

Many radiative transfer schemes approximate the spectral integration over ∼105 to ∼106 wavelengths with correlated k‐distributions methods that typically require only 101–102 spectral integration points (g‐points). The exact number of g‐points is then chosen as an optimal balance between computational costs and accuracy, normally assessed in terms of a number of radiative quantities. How this radiative accuracy propagates to simulation accuracy, however, is not straightforward. In this study, we therefore explore the sensitivity of cloud properties in large‐eddy simulations (LES) to the accuracy of radiative fluxes and heating rates. We first generate smaller sets of g‐points from existing k‐distributions by repeatedly combining adjacent g‐points while maintaining the highest possible accuracy on a chosen set of radiative metrics. Next, we perform three sets of LES with varying cloud—radiation coupling pathways, and therefore different requirements for the accuracy of the radiative transfer computations, to investigate how these smaller and thus less accurate k‐distributions affect simulation characteristics. The decrease in radiative accuracy with 3–4 times smaller k‐distributions results in biases in cloud properties that are relative small compared to their temporal fluctuations. These results show potential for speeding up radiative transfer computations in cloud‐resolving models by reducing the resolved spectral detail. However, more statistically converged simulations and a wider set of case studies is required to fully assess the robustness of our results. Plain Language Summary Radiation emitted by the sun and the earth drives our weather and climate. Atmospheric models therefore need to compute how solar and thermal radiation interacts with cloud droplets, aerosols, and gas molecules. These computations can be performed very accurately by doing independent calculations for many wavelengths, but that is very time‐consuming. Here, we study to what extent the accuracy of the radiation computations affects the accuracy of simulations of the atmosphere. We find that the reduction in radiative accuracy corresponding to 3–4 times fewer representative computations results in relative small mean errors in cloud properties compared to the temporal fluctuations. In further research, we should aim to better understand whether the mean errors are small enough to represent accurate simulations. Key Points We investigate the sensitivity of moist convection to th