Dacian Daescu (PI)
Portland State University
daescu@pdx.edu
Development of a New Methodology for Adaptive Observations in the Framework
of Four-Dimensional Variational Data Assimilation
The issue of adaptive observations is linked to the skillful prediction
of high-impact weather which is a scientific and societal challenge.
In this work we aim to develop a new methodology for adaptive (targeted)
observations strategies in the context of atmospheric four-dimensional
variational (4D-Var) data assimilation. This is a three-year collaborative
research proposal between Prof. Michael Navon and Prof. Gordon Erlebacher from
Florida State University and Prof. Dacian Daescu from Portland State
University to investigate optimal sampling strategies designed in the
time-space domain that fully account for the characteristics of the data assimilation
system. We will build on current adjoint based methods (e.g. singular
vectors, sensitivity to analysis and to observations) to develop new
targeting strategies that take into account the particular details of
the 4D-Var assimilation scheme and the interaction between time distributed
adaptive observations, the routine observational network, and the background
estimate of the initial conditions. These novel methods are expected
to outperform objective targeting strategies with a low additional computational
effort. The proposed research will advance the current status of
atmospheric targeted observations strategies in various aspects: for
the first time the temporal dimension of the 4D-Var procedure will be
incorporated in the optimal sampling design by simultaneously accounting
for multiple targeting instants in the assimilation window; shortcomings
of the conventional methods to adjust to the 4D targeting framework will
be highlighted; to identify optimal observational locations, novel
targeting methods will consider not only the spatial distribution of
the sensitivity fields, but also their dynamical interaction in the presence
of data. This very much-needed capability of a 4D adaptive sampling
methodology was not previously investigated. The methodology proposed
will identify multiple time varying areas for deploying in situ observational
resources following the flow regime. Adjoint modeling will be used to
obtain time-dependent sensitivities of the forecast to the analysis state
and data. The impact of the model error on the identification of the
targeting region will be further investigated. Adaptive methods
proven to be feasible for operational use will be validated using first
a 2-D global shallow-water model (Lin and Rood, QJRMS 1997) then, in
collaboration with NASA/GMAO, the 3-D finite volume General Circulation
Model within the GEOS-4 assimilation and forecast system. These techniques
will be first used in a posteriori analysis of the experimental results,
then applications to a priori experimental planning will be also investigated.
A successful completion of the methodology developed in this work has therefore
a significant relevance for future NASA field experiments (e.g., CAMEX-5). The
validation stage will include a comparative study with modern tools for
targeted observations such as the Ensemble Transform Kalman Filter approach.
Novel visualization techniques will be developed to allow faster evaluation and
synthesis of complex results. The proposed research will have a substantial
impact on the future development and use of adaptive observations in
meteorology and oceanography. The society as a whole will benefit from
this research through an improved prediction of high impact weather events.
+ Back to Participant Listing |
