Summary: | Most models in the fifth phase of the Coupled Model Intercomparison Project (CMIP5) underestimate the surface air temperature over China in both winter and summer. Understanding the weather regime in association with the simulated temperature variability is of high interest to get insight into those biases. Based on the weather regime method, we investigated the contributions of large-scale dynamics and non-dynamical processes to temperature biases and inter-model spread. The weather regimes associated with the observational temperature patterns were obtained through a k-means clustering algorithm applied to daily 500 hPa geopotential height anomalies. Here we identified the clustering number of weather regimes using the classifiability and reproducibility indices which can provide the optimal clustering number to obtain objective clustering. Both indices suggested the weather regimes in East Asia can be classified as four clusters in winter (December–January–February) and three in summer (June–July–August). The results indicated that the first and second weather regimes were related to the cold temperature anomalies in China during winter, and the three weather regimes could not effectively classify the temperature patterns during summer. The ensemble mean of 23 CMIP5 models overestimated the occurrence frequencies of the second weather regime, which corresponds to a weaker high latitude westerly jet over East Asia during winter. The 500 hPa geopotential height anomalies and the inter-model spread over the Tibetan Plateau may be associated with the limited ability of the CMIP5 models in simulating the thermal effects of plateau in summer. We also found that the non-dynamical processes had major contribution to the ensemble-mean biases, and the large-scale dynamics played a minor role. The non-dynamical processes substantially affected the inter-model spread, especially over the Tibetan Plateau and the Sichuan Basin, during both winter and summer. The results suggested that improving the simulation of regional processes may help to improve model performance. The use of multi-model mean is recommended since it performs better than most of individual models.
|