Skill and uncertainty in surface wind fields from general circulation models: Intercomparison of bias between AGCM, AOGCM and ESM global simulations

Abstract Understanding the reliability of global climate models (GCMs) to reproduce the historical surface wind fields is integral part of building robust projections of surface wind‐climate, and other wind‐dependent geophysical climatic variables. Understanding the skill of atmosphere‐only models (AGCM), coupled atmosphere–ocean models (AOGCM) and fully coupled earth system models (ESM) is likewise paramount to assess any systematic model improvements. In this paper, we systematically assess whether surface wind fields obtained from 28 CMIP5 GCMs can represent large‐scale spatial patterns and temporal variability of historical surface winds. We show that inter‐model uncertainty is typically 2–4 times larger than the uncertainty associated with GCM internal variability, although the latter can be significant within specific regions. We also find that CMIP5 models are typically capable of reliably reproducing large‐scale spatial patterns of historical near‐surface winds, but considerable uncertainty lies within the CMIP5 ensemble with strong latitudinal dependence. CMIP5 models show limitations in their ability to reliably represent inter‐annual and inter‐seasonal variability particularly within tropical‐cyclone‐affected regions. In further analysis, we quantify and intercompare historical wind bias from different types of models with different dynamical cores, based on multiple CMIP5 diagnostic experiments. We find that bias in surface wind fields are largely intrinsic to the atmospheric components of the models, and that the inclusion of carbon‐cycle dynamics has insignificant effect on simulated surface winds (at decadal time‐scales). Inconsistencies between AGCM and AOGCM simulations are largely driven by errors in sea surface temperatures (SST); though such differences are not statistically significant relative to the inter‐model uncertainty within the CMIP5 ensemble. These results show that the dominant source of bias in simulated wind fields lies in the underlying physics of the atmospheric component of the models.