Characteristics of the temperature correlation network of climate models.

Temperature correlation network has been commonly used in climate monitoring, diagnosing and prediction using reanalysis data, however its application in the network analysis of climate dynamical models hasn't been deeply studied. We construct a temperature correlation network based on near-surface 2m air temperature of four climate models by comparing their capability of properly capturing the structural characteristics of the temperature networks, and further conduct a comparative analysis of the topological differences among different models. The features of temperature correlation networks varied significantly among the four models, in which the ECMWF-SYSTEM5 model network has the highest connectivity among the four models, while the NCEP_CFS2 model has the lowest one. It is also revealed that the model with higher connectivity normally has a stronger correlation between the nodes of the air temperature correlation network, which is likely attributed to the model's stronger teleconnection, regional consistency and smaller standard deviation between predicted temperature series of most two grid points. It is also implied that the model's prediction skills have a probable relationship with the network structure. For each model, the 1-month lead prediction has the highest prediction skill corresponding to the model having the connectivity close to the observation. With the increase of prediction lead times, connectivity bias has a quick rise, and the prediction skill has an obvious decrease. However, for different models at the same prediction lead time, it is not the case that the larger the connectivity deviation the lower the prediction skill, for example, the ECMWF_SYSTEM5 model has the highest prediction skill and the largest connectivity deviation, we find that the ECMWF_SYSTEM5 model network possesses significantly higher connectivity and more distinctive small-world characteristics, which implies a stable network structure helps to improve the prediction skills of the model. Therefore, this study can deepen our understanding of climate models and provide guidance for improving the ability of models to properly simulate climate features. [ABSTRACT FROM AUTHOR]