This award supports a three-year effort to use nonlinear techniques to improve understanding of Antarctic climate through studies of observational and forecast model data sets; improve and extend reconstructions of past Antarctic climate from ice-core data; and reconstruct data missing from the observational records, potentially into the pre-instrumental era. The intellectual merit of the proposed activity arises from the opportunity to improve understanding of the past, present and future climate of the Antarctic, a key component in the global climate system. Self-organizing maps (SOMs), an emerging, powerful nonlinear tool, will be used to classify free-atmosphere reanalysis data into archetypal patterns (SOM states). Feed-forward artificial neural networks (FF-ANNs) will then be trained to predict the preferred SOM states from ice-core data covering the instrumental era. The trained FF-ANNs will extend the reconstructions of SOM states to the full length of the ice core data, leading to long-term reconstruction of climate. Histories of surface conditions will be improved by filling data gaps in observational records using FF-ANNs and free-atmosphere reanalysis data. These records may also be extended into the pre-instrumental era using the above ice-core based reconstructions of the atmospheric circulation. The broader impacts of the project relate to activities with the Earth and Mineral Sciences Museum (co-located in the Geosciences building) which will bring project results/tools to a wider audience through development of interactive graphical visualizations/presentations for the Museum's fixed and traveling GeoWall displays. One or more undergraduates from the College will be involved in the project with an option to also present project results at a national meeting/workshop. The work will also contribute to the continuing development of an "early career" investigator, including the opportunity to continue building (and refining) relevant and useful skills in teaching, outreach, collaboration, etc.
