Multi‐model assessment of global temperature variability on different time scales

This study explores the near-surface air temperature (TAS) prediction skills at different time scales via coupled climate models that were involved in the Coupled Model Intercomparison Project phase 5 (CMIP5). The simulation skills of the global mean TAS are first assessed between the observations and models; then, the temporal variability is separated into three parts (the linear trend, decadal variability and inter-annual variability), and each part is compared with the observations. It is found that the global mean TAS anomaly and the decadal variability are well captured by the model, while the inter-annual variability is poorly presented. In all the three parts of different time scales, there are larger differences among the 18 models, and the ensemble mean of CMIP5 is the closest to the observations. Besides, the TAS decadal variability is better presented for global ocean than for global land. In the assessment of climate oscillation, it is found that the models can well reproduce the TAS decadal variation patterns correlated with the Pacific Decadal Oscillation (PDO) and the inter-annual variation patterns correlated with El Nino-Southern Oscillation (ENSO) but poor for the decadal variation patterns related to the Atlantic Multidecadal Oscillation (AMO). This study provides a reference assessing the simulation skills of climate models and an indicator evaluating the advantage of CMIP6 in comparison with CMIP5.