Geoscientific Model Development (Apr 2019)

A statistical and process-oriented evaluation of cloud radiative effects in high-resolution global models

  • M. A. Thomas,
  • A. Devasthale,
  • T. Koenigk,
  • K. Wyser,
  • M. Roberts,
  • C. Roberts,
  • K. Lohmann

DOI
https://doi.org/10.5194/gmd-12-1679-2019
Journal volume & issue
Vol. 12
pp. 1679 – 1702

Abstract

Read online

This study evaluates the impact of atmospheric horizontal resolution on the representation of cloud radiative effects (CREs) in an ensemble of global climate model simulations following the protocols of the High Resolution Model Intercomparison Project (HighResMIP). We compare results from four European modelling centres, each of which provides data from “standard”- and “high”-resolution model configurations. Simulated radiative fluxes are compared with observation-based estimates derived from the Clouds and Earth's Radiant Energy System (CERES) dataset. Model CRE biases are evaluated using both conventional statistics (e.g. time and spatial averages) and after conditioning on the phase of two modes of internal climate variability, namely the El Niño–Southern Oscillation (ENSO) and the North Atlantic Oscillation (NAO). Simulated top-of-atmosphere (TOA) and surface CREs show large biases over the polar regions, particularly over regions where seasonal sea-ice variability is strongest. Increasing atmospheric resolution does not significantly improve these biases. The spatial structure of the cloud radiative response to ENSO and NAO variability is simulated reasonably well by all model configurations considered in this study. However, it is difficult to identify a systematic impact of atmospheric resolution on the associated CRE errors. Mean absolute CRE errors conditioned on the ENSO phase are relatively large (5–10 W m−2) and show differences between models. We suggest this is a consequence of differences in the parameterization of SW radiative transfer and the treatment of cloud optical properties rather than a result of differences in resolution. In contrast, mean absolute CRE errors conditioned on the NAO phase are generally smaller (0–2 W m−2) and more similar across models. Although the regional details of CRE biases show some sensitivity to atmospheric resolution within a particular model, it is difficult to identify patterns that hold across all models. This apparent insensitivity to increased atmospheric horizontal resolution indicates that physical parameterizations play a dominant role in determining the behaviour of cloud–radiation feedbacks. However, we note that these results are obtained from atmosphere-only simulations and the impact of changes in atmospheric resolution may be different in the presence of coupled climate feedbacks.