Monthly Forecasting at ECMWF

Abstract

A monthly forecasting system based on 32-day coupled ocean?atmosphere integrations has been set up at ECMWF. This system has run routinely since March 2002 every 2 weeks, and 45 cases from March 2002 to December 2003 have been verified. Results of this validation suggest that the model displays some skill in predicting weekly averaged 2-m temperature, precipitation, and mean sea level pressure anomalies relative to the climate of the past 12 years. For days 12?18, probabilistic scores indicate that the monthly forecasting system performs generally better than both climatology and the persistence of the previous weekly probabilities, suggesting that forecasts at that time range could be useful. After about 20 days of forecast, the model displays some skill in predicting events with a large threshold. At that time range, the performance of the system depends strongly on the geographical location, with Europe being a particularly difficult region. However, the model displays some useful skill after 20 days over North America, Asia, and the southern extratropics. In order to calibrate the monthly forecasting system, a 5-member hindcast over the 12 years preceding the real-time forecast has been produced. Probabilistic scores computed with the hindcast confirm the main results obtained with the real-time forecast. The scores display strong seasonal variability, with the model being particularly skillful in winter.