HR: 0800h
AN: SF41A-0751 [Abstracts]
TI: Modular Inversion and Data Assimilation
AU: * Egbert, G D
EM: egbert@coas.oregonstate.edu
AF: Oregon State University, College of Oceanic and Atmospheric Sciences
, Corvallis, OR 97331-5503
United States
AB:
An important application of computational modeling in the
geosciences is to problems in parameter estimation, inversion, and data
assimilation. These applications can require complex software
systems that combine simulations
of Earth processes and systems with data processing and mathematical optimization
algorithms. A modular approach to software design for
these systems offers many obvious advantages:
for example, optimization and data handling components
can be developed once, and applied across a range of problems
or numerical models. Developing numerical modeling modules
that will be useful for inversion and assimilation applications
also presents some challenges. For example, for many optimization
strategies tangent linear and/or adjoint models are required.
These modules are comparable in complexity to the original code,
but tools and procedures have been developed (e.g., adjoint compilers)
to facilitate development of these modules given a working forward code.
Here we discuss recent experiences with two modular software systems being
developed for inversion and data assimilation. The first is
a small modular inversion system for electromagnetic induction
data that we are developing ourselves, primarily for prototyping of
efficient inversion algorithms. The second is a much larger project,
the Inverse Ocean Modeling (IOM)
system being developed in the physical oceanographic research community for
variational data assimilation in a range of ocean modeling applications.
One important, if unsurprising, lesson from these projects
is that even a bit of foresight in initial development
of modular forward modeling codes and interfaces can greatly simplify
incorporation of these modules into data assimilation and inversion schemes.
Since inversion, data assimilation, and parameter estimation are likely
to remain very important in solid Earth studies, this issue
deserves some consideration
as the geoscience community prepares to invest in new software
infrastructure for future research needs.
DE: 3230 Numerical solutions
DE: 3260 Inverse theory
DE: 1515 Geomagnetic induction
SC: Special Focus: Advances in Data Acquisition, Management, Analysis and Display [SF]
MN: 2004 AGU Fall Meeting