Ensemble Data Assimilation for Seamless Prediction (Invited Presentation)

The current state-of-the-art tools for weather/climate-system prediction vary depending on the spatial and temporal scales of the prediction. Different models and assimilation systems may be used for mesoscale predictions with lead times O(days), synoptic scale predictions with lead times O(weeks), and seasonal predictions with lead times O(months). Meanwhile, models that resolve increasingly smaller scales are being implemented as computer resources increase. Building a single prediction system that can generate predictions for all of these scales could allow more efficient use of both computational and development resources. This talk examines issues related to the use of an ensemble data assimilation system for such a “universal” prediction system. For instance, is it practical to do data assimilation in a high-resolution coupled model of the entire climate system using all available observations? Are there advantages to running multiple assimilations at different resolutions and for different components of the climate system? How should one proceed if short-term forecasts are only needed over limited geographical regions? A brief discussion of the question of “intrinsic” predictability in this context concludes the presentation.