DRAFT – April 23 2003
WORKSHOP RECOMMENDATIONS ON
ENSEMBLE-BASED DATA ASSIMILATION
1. STATISTICAL CONSIDERATIONS
a)
Pursue
basic research on how well information is extracted from observations;
Develop
measures of related information content
b)
Explore
relationship and optimal resource allocation between model resolution,
ensemble
size, and extent of localization in background error covariance
estimation
c)
Study
hybrid methods that combine ensemble-based assimilation methods with
traditional approaches in defining background error covariance information
2. ISSUES RELATED TO THE USE OF IMPERFECT
MODELS
a)
Enhance
NWP models so they can represent model related uncertainty in
ensemble
forecasting applications
b)
Advance
knowledge related to, and develop new methods for
i.
Reducing
bias in background ensemble forecasts
ii.
Estimating
the covariance in model related forecast errors by, e. g.,
studying the
relative value of additive vs. multiplicative noise in variance
inflation
3. HIGH SPATIAL RESOLUTION APPLICATIONS
a)
Study
the relative merits of using limited area vs. variable spatial resolution
models in high resolution data assimilation applications
b)
Ensure
that in limited area model applications uncertainties present on the larger
scales are properly accounted for through the specification of the boundary
conditions
c)
Study
the effect of, and develop methods to cope with the highly non-linear and
spatially and temporally intermittent processes that are present on the small
scales
a)
Demonstrate that ensemble-based schemes, like 4D-VAR, can successfully
assimilate observations spread arbitrarily over, say, a 6-12 hour time period
b)
Carry
out high resolution experiments with realistically high data density for
testing prior to considering operational applications
c)
Provide the community with a common set of models and data sets with the level
of sophistication ranging from simple to the most complex operational setting so
that advances made by different groups can be easily and meaningfully compared