HR: 12:05h
AN: NG42A-08 [Abstracts]
TI: Variational Data Assimilation and the Indirect Representer Method: Application to crustal
geodynamics
AU: * Egbert, G D
EM: egbert@coas.oregonstate.edu
AF: Colleg of Oceanic and Atmospheric Sciences, 104 COAS Admin Bldg
Oregon State University, Corvallis, OR 97331-5503
United States
AB:
One standard approach to DA, now well developed in the atmospheric and oceanic
sciences, is to prescribe error covariances
which define a priori hypotheses on the magnitude and spatial structure
of model and data errors, and then optimize the tradeoff between
fitting the data and satisfying prior model assumptions by minimizing a quadratic
penalty functional. Minimization of such a penalty functional
using gradient based search methods is often referred to as variational
data assimilation, 4D-Var, or the generalized inverse method (GIM).
The GIM has proven to be particularly well suited to scientific
(as opposed to operational) studies in oceanography,
and this approach is likely also to prove very useful
for a wide range of scientific studies in geodynamics.
Over the past decade optimization methods that make GIM
more practical for large non-linear DA problems have been developed
and applied in oceanography. These so-called "indirect
representer" methods involve minimization of
the quadratic penalty functional through a series of conjugate
gradient solutions of the linearized Euler-Lagrange equations,
recast in the data space. A modular implementation of the indirect
representer approach, the Inverse Ocean Modeling (IOM) system,
is currently being developed through an NSF funded ITR project.
This modular system simplifies implementation of the GIM
for any client dynamical model, but requires coding of a
tangent linear (TL) to the non-linear dynamical
model along with the adjoint (ADJ) of this TL.
I will summarize the GIM approach to DA,
the indirect representer method, and the IOM.
As a specific application in geodynamics, I will discuss development of TL
and ADJ codes, and coupling with the IOM,
for the Geophysical Finite Element Simulation Tool (GEOFEST),
which has been extensively used for visco-elastic modeling of
Earth deformation associated with the earthquake cycle.
One important lesson from this experience is that with reasonably
modular modeling code, development of TL and ADJ codes need not
be so onerous as to preclude consideration of
gradient based search methods.
DE: 0594 Instruments and techniques
DE: 3260 Inverse theory
DE: 8194 Instruments and techniques
SC: Nonlinear Geophysics [NG]
MN: Fall Meeting 2005