The Data Assimilation Research Testbed (DART) at NCAR is a modular
software environment which can be used to perform
ensemble filter data assimilation experiments. DART filters
combined with NCAR's Community Atmosphere Model (CAM) have been used to
assimilate reanalysis observations from January,
2003. The results are compared to the NCEP operational system
from that time. Future operational implementations of ensemble filters
are, for practical reasons, likely to be limited in the number of
ensemble members that can be run. As a consequence, sampling error is a
fundamental issue that must be dealt
with using three-dimensional covariance localization. A key advantage
of ensemble filter data assimilation systems is that they
can be readily applied to problems such as adaptive observations,
routine network design and re-analysis. For these applications,
four-dimensional as opposed to three-dimensional localization
is necessary. Using examples in a hierarchy of atmospheric-like
prediction models, the importance of four-dimensional
localization in ensemble filter applications to adaptive observations,
routine network design and re-analysis will be discussed.
Shree P. Khare and J.
L. Anderson