Seoul National University
July 1, 2009
Mesa Laboratory- Directors Conference Room (Brown Bag)
An Improvement of Climate Predictions by Regularized CCA
This paper proposes a statistical method to perform reliable predictions of future climate using a regularized canonical correlation analysis. Unlike the classical approach that depends solely on model output or observations, the proposed method combines the information from ensembles of atmosphere-ocean general circulation models (AOGCMs) and observed climate responses to improve the prediction power of future climate. We focus on predictions of future precipitation and temperature changes on both global and regional scales. The proposed statistical technique is based on a coupling of the classical canonical correlation analysis and the regularization principle designed to improve prediction ability and to handle high-dimensional data in which the number of variables is larger than the number of observations. Prediction results from numerical experiments and real observations of temperature and precipitation demonstrate the promising empirical properties of the proposed approaches.