Evaluation of CMIP5 models using network analysis

Network analysis constitutes a powerful framework to validate climate models by quantifying their uncertainties while investigating how they reproduce climate modes of variability, teleconnections, and extremes. Here we propose a novel application of network analysis to evaluate current climate and future projections in 14 General Circulation Models (GCMs) participating in the CMIP5 project. The method proposed clusters gridded climate data into areas, regions homogeneous to the underlying climate variable. On a second step, links are inferred between areas and a climate network is constructed. The areas (nodes of the network) correspond to climate modes of variability while links between areas can be used as a proxy for climate teleconnections.

We perform an extensive model intercomparison in terms of the climate network that each model leads to, and we analyze a variety of fields, from SST to precipitation and geopotential height. We are able to identify common or specific errors both in terms of the spatial expression as well as the intensity of the climate modes and teleconnections that the models produce. Further we employ two distance metrics in order to quantify the differences between networks inferred from observational data and models, and from current and future modeled climates.