Ensemble Data Assimilation for Reanalysis

Jeff Whitaker

Climate Diagnostics Center,Boulder, CO

Abstract:
Ensemble data assimilation techniques hold promise for significantly improving analysis accuracy by providing flow-dependant estimates for first-guess error. Flow-dependant first-gues errors are expected to have the largest impact on analysis accuracy when observations are sparse. This makes ensemble data assimilation ideal for reanalysis, especially in the pre-satellite (and even the pre-radiosonde) era. At CDC we have developed an ensemble data assimilation system based on a T62 version of the MRF that runs efficiently on distributed-memory clusters. Results will be presented to demonstrate the potential of the system for producing a reanalysis in the pre-radiosonde era using only surface observations. Analyses produced for Dec 2001 using only surface pressure observations at 1935 densities (about 600-700 in the Northern Hemisphere at 00 and 12 UTC) are reasonably close to those produced by CDAS (using several orders of magnitude more observations). RMS differences in 500 mb height are approximately 30 meters (roughly equivalent to a 2.5 day forecast error), while the CDAS itself produces analyses that are only half as accurate given the same observations. These results suggest that a 100+ year reanalysis is feasible.