HR: 1340h
AN: NG43B-0584    [Abstracts]
TI: Variational Data Assimilation for Short-Term Dynamical Processes in Earth
AU: * Tandon, K
EM: kush@coas.oregonstate.edu
AF: College of Oceanic & Atmospheric Sciences, 104 COAS Admin. Bldg. Oregon State University, Corvallis, OR 97331 United States
AU: Egbert, G
EM: egbert@coas.oregonstate.edu
AF: College of Oceanic & Atmospheric Sciences, 104 COAS Admin. Bldg. Oregon State University, Corvallis, OR 97331 United States
AU: Lyzenga, G
EM: Gregory.A.Lyzenga@jpl.nasa.gov
AF: Jet Propulsion Laboratory, M/S 126-347 4800 Oak Grove Drive, Pasadena, CA 90089 United States
AB: Numerical modeling of short-term geodynamic processes, such as inter-seismic strain buildup, presents opportunities for application of data assimilation (DA) methods to problems of broad societal impact. Together with high-resolution datasets obtained through initiatives like EARTHSCOPE, DA techniques will perhaps ultimately prove useful for improved predictive modeling of earthquake cycles, similar to what is commonly done now in forecasting of weather and seasonal climate variability. The NASA/JPL Geophysical Finite Element Simulation Tool (GeoFEST), a software package for visco-elastic modeling of dynamic stress and strain evolution in the crust, is one of the major simulation tools for QuakeSim, a NASA Earth Science Enterprise project, and been used extensively by other researchers for studying crustal deformation and the earthquake cycle. Here we describe initial efforts to develop and apply variational DA methods based on the GeoFEST package. A key feature of our development is to make use of a modular Inverse Ocean Modeling (IOM) system being developed for ocean data assimilation. In comparing a dynamical model to data three types of error must be considered: errors in the actual data (including spatial and temporal scales that the model is not intended to represent), errors in model physics or parameterizations, and errors in model inputs (initial conditions, boundary conditions, and forcing). Variational DA provides a framework for combining and inter-comparing data and dynamic models, explicitly allowing for all of the above mentioned error types. Similar to the generalized inverse methods familiar to geophysicists, minimization of the model and data error is accomplished in variational DA through a gradient-based search. Such schemes require coding of the tangent linear (TL) of the non-linear dynamical model, along with the adjoint (ADJ) of the TL. We will discuss development of TL and ADJ modules for GeoFEST. The relatively modular nature of GeoFEST allows significant code reuse, and only minor modifications to a subset of existing subroutines are required for the TL and ADJ codes. Interfacing of the new TL and ADJ codes with the IOM will also be discussed. As an initial illustration of the power of the methods we are developing, we will present results of synthetic DA experiments that test the power of different data types (e.g., GPS, InSAR, strainmeters) to correct different types of errors in model inputs and physics. These sorts of analysis can be very useful in rational design of observing systems, for example.
DE: 0500 COMPUTATIONAL GEOPHYSICS (3200, 3252, 7833)
DE: 1213 Earth's interior: dynamics (1507, 7207, 7208, 8115, 8120)
DE: 3200 MATHEMATICAL GEOPHYSICS (0500, 4400, 7833)
DE: 3260 Inverse theory
DE: 7290 Computational seismology
SC: Nonlinear Geophysics [NG]
MN: Fall Meeting 2005