A new approach involving the use of common empirical orthogonal functions (EOFs) in statistical downscaling of future global climate scenarios is proposed. The advantage of this method is that it minimizes the errors associated with the downscaling of future climate scenarios. The time series from the common EOF analysis are used both for the calibration of the statistical models and the prediction of future climate scenarios. This paper presents a systematic comparison between the common EOF approach and downscaling with a more conventional method based on the 'Perfect Prog' concept. Three different sets of experiments are carried out where the two downscaling methods are compared, and the robustness of the methods is explored taking the predictor field from different regions. The analysis indicates that climate scenarios derived using the common EOF approach are associated with smaller errors. Another important finding is that the choice of predictor domain may influence the downscaled results, and that the merit of downscaling may depend on this area as well as location and season. Copyright (C) 2001 Royal Meteorological Society.
