19th Conference on Probability and Statistics

10.1

Improved statistical seasonal forecasts using extended training data

PAPER WITHDRAWN

Daniel S. Wilks, Cornell University, Ithaca, NY

Statistical seasonal forecasts for gridded North American temperatures are computed using Pacific sea-surface temperature predictors, comparing long (1880 through most recent year) and short (1950 through most recent year) training samples. Use of the longer training series substantially improves the forecasts in winter, and improves forecasts for the longer lead times in other seasons, even though the older reconstructed predictor fields are based on less complete and presumably less reliable information. Forecasts made using canonical correlation analysis and maximum covariance analysis (MCA) perform similarly overall, although the best forecasts in winter are achieved with MCA forecasts that also include a predictor representing the warming trend in recent years.

Session 10, Climate Forecasting
Thursday, 24 January 2008, 8:45 AM-9:45 AM, 219

Next paper

Browse or search entire meeting

AMS Home Page