HR: 0800h
AN: NG31B-0877 [Abstracts]
TI: Phase Changes and State Estimation for Non-linear Systems
AU: * Chin, T M
EM: mike.chin@jpl.nasa.gov
AF: Jet Propulsion Laboratory, M/S 238-332,
4800 Oak Grove Drive, Pasadena, CA 91109
United States
AU: Jewell, J B
EM: Jeffrey.Jewell@jpl.nasa.gov
AF: Jet Propulsion Laboratory, M/S 238-332,
4800 Oak Grove Drive, Pasadena, CA 91109
United States
AU: Turmon, M J
EM: turmon@aig.jpl.nasa.gov
AF: Jet Propulsion Laboratory, M/S 238-332,
4800 Oak Grove Drive, Pasadena, CA 91109
United States
AB:
Prediction of atmospheric and oceanic states is difficult
partly because of the lack of complete knowledge of the
present states. An important goal of the data assimilation
task is to provide such initial conditions for the atmospheric
and oceanic dynamic equations used for forecasting, by
updating the model results with some timely observational data.
Data assimilation practiced today is mostly based on realization
of linear estimation theory. In particular, a jointly-Gaussian
distribution of the estimation errors and quasi-linear treatment
of covariance dynamics have been assumed and practiced.
This does not take account of the weather and climatic "regimes"
(favored state values) that have been recognized both dynamically
(e.g. by Lorenz 1963) and empirically. A fundamental problem is
that a Gaussian distribution is uni-modal, while weather and climate
often display multi-modal characteristics known also as "dynamic regimes".
We have applied Monte Carlo numerical techniques to generic
data assimilation problems without the assumptions of Gaussian
distribution and quasi-linear dynamics. Specifically,
we have extended the well-known assimilation technique of
"ensemble Kalman filter" to a smoothing algorithm based
on some theoretical developments by Kitagawa (1996) and others.
The Monte-Carlo smoothing algorithm does yield more accurate
state estimate than the corresponding filtering algorithm.
A key component in both the filter and smoother is the generation
of the random perturbations that can sample the solution space
effectively and efficiently. When the size of ensemble is limited
as in the case with a typical data assimilation practice,
various empirically derived strategies to generate the stochastic
forcing have shown potential to retain some filter/smoother performance.
DE: 3220 Nonlinear dynamics
DE: 3260 Inverse theory
SC: Nonlinear Geophysics [NG]
MN: 2004 AGU Fall Meeting