6.10
Improved skill of ENSO coupled model probability forecasts by Bayesian combination with empirical forecasts
Caio A. S. Coelho, University of Reading and ECMWF, Reading, Berks, United Kingdom; and S. Pezzulli, F. J. Doblas-Reyes, and D. B. Stephenson
After the particularly strong El Niņo 1982-83 event, several efforts have been made to forecast ENSO using both empirical (statistical) and physically derived numerical model approaches. However, the comparative skill of these two methods is a subject of substantial inquiry and some controversy. Although considerable efforts have been directed to the development and improvement of these two different methods, not many studies have attempted to combine these types of forecast. This study presents a new simple approach for combining empirical with dynamical ensemble forecasts in order to make better probability forecasts of the Niņo-3 index.
A simple Bayesian approach has been used to forecast December Niņo-3 index from forecasts started in August (5-month lead time). The empirical model was defined as a linear regression between December and July historical Niņo-3 index data (1950-2001). Dynamical ensemble forecasts for the period 1987-98 were provided by ECMWF, as part of the Development of a European Multi-model Ensemble system for seasonal to inTERannual prediction (DEMETER) project. The empirical and dynamical ensemble forecasts alone show similar skill of around 54% (mean absolute error skill score relative to climatology). However, the combined forecast has a much better skill of 68% than the separate forecasts. The combined forecast also provides a better and more realistic uncertainty estimate, which is underestimated by the ensemble system.
Session 6, Weather and Forecasting I
Tuesday, 25 March 2003, 1:30 PM-5:00 PM
Previous paper Next paper