An Interactive Multi-Model Consensus on Climate Change


  1. To develop a scheme to identify and combine the best elements of different climate models, in an automated fashion, that is significantly superior to the weighted averaging of model outputs.
  2. To resolve some differences among a set of three models over the extent, regional characteristics, and near-term characteristics of climate change in the 21st century.
  3. To lay a theoretical foundation for the run-time fusion of a more extensive set of climate models that may be more intimately connected.
  4. To understand the effect of dynamical changes in the models, as greenhouse gas levels change, on the skill of the fused configuration.

Climate models used by the Intergovernmental Panel on Climate Change (IPCC) differ in projections of globally averaged temperature increase in the next century by as much as a factor of two, and differ completely in regard to projections for specific regions of the globe. The state of the art in combining the different projections is simply to form a weighted average of model outputs. This project instead combines the models themselves by allowing them to exchange information as they run, with weights attached to the different variables in the different models determined by training on 20th century data. This innovative approach to reducing uncertainty might be compared to a group of scientists resolving their differences through dialogue, rather than simply voting or averaging their opinions. Methods to be employed: The tracking of real weather by a numerical weather prediction system has been described as an instance of chaos synchronization: the chaotic model is made to synchronize with the chaotic natural system by passing noisy information about a few variables from the latter system to the former system. Here, the models are made to synchronize with each other as well as with reality. It has been shown that the synchronization phenomenon can commonly be extended so that model parameters are synchronized as well as states – an instance of machine learning. Here, we choose these adaptable parameters to be the strengths of the connections linking the different models - a more manageable set than the internal parameters of the models themselves. The connection strengths after training quantify the reliability of the different components of the different models. The desynchronization among the models quantifies uncertainty in the projections of the fused model. Potential impact: The project will potentially lead to more reliable climate projections, particularly for local regions and in the short term. Reduced uncertainty as to the impact of greenhouse gases will increase the societal impact of the projections and lead to better policy. Major participants: PI Kocarev is a recognized expert on chaos synchronization. PI'sDuane and Tribbia have pioneered in applications of synchronization to geophysical phenomena and to data assimilation. Co-PI Kirtman introduced the \interactive ensemble" of large models, here modifed to include adaptive connections. PI Tsonis has described climate regime shifts in terms of desynchronization of internal modes. Collaborator Selten is an expert and originator of the Ecbilt model, a model that will be used to bridge the gap between idealized \toy" model studies and the less tractable large climate models.

Project Term: 
2010 to 2013
Project Type: 
University Funded Research


None Available

Research Highlights:

None Available