Constantine Dovrolis, Ilias Foudalis - College of Computing

Annalisa Bracco - Earth and Atmospheric Sciences

Georgia Tech


Network analysis provides a powerful, but only marginally explored, framework to validate climate models, quantify uncertainties and investigate teleconnections, assessing their strength, range, and impacts on the climate system.

Our purpose is to develop a fast, scalable and cutting-edge computational toolbox that will help examine, quantify, understand and visualize climate sensitivity, while constraining Coupled Global Circulation Models (CGCMs) and Earth System Models (EaSMs) towards observations. The main objectives of this project are (i) understanding and explaining, at a fundamental level, the causes and manifestations of climate sensitivity in models; and (ii) validating GCMs and EaSMs with respect to their faithful representation of current climate, including the correct sensitivity, as well as to the robustness of future-climate projections using powerful numerical and visualization tools.

Key topics for investigation are:

  • Inferring static or dynamic spatial climate networks from climate variables
  • Comparing and contrasting the networks derived for observational and reanalysis data-sets with networks resulting from various CGCMs, focusing on the new CMIP5 outputs
  • Investigating how network feedbacks can trigger climate shifts, and examine the underlying network of each model to determine whether it has the necessary structure to reproduce such events
  • Understanding how network effects impact the uncertainty factor in climate models


We propose a new approach to apply network analysis to climate science. We first apply a novel network-based clustering method to group the initial set of grid cells in “areas”, i.e., in geographical regions that are highly homogeneous in terms of the underlying climate variable. These areas represent the nodes of the inferred network. Links between areas (i.e., the edges of the network) represent non-local dependencies between different regions over a certain time period. These inter-area links are weighted, and their magnitude depends on both the cumulative anomaly of each area and the cross-correlation between the two cumulative anomalies. The final network is represented as a complete weighted graph.



This research is supported by The Office of Biological and Environmental Research (BER) of the U.S. Department of Energy (DOE), an Earth System Modeling (ESM) project part of the SciDAC (Scientific Discovery through Advanced Computing) program.