Improving Estimation and
Prediction of Extremes via Conditional Bias-Penalized Kalman Filter
D.J. Seo
University of Texas at Arlington
29 Nov, 9:30 am, in 2155
Abstract:
Kalman filter (KF) and its variants such as ensemble Kalman filter
(EnKF) are widely used for real-time state updating and prediction in
environmental science and engineering. In many applications, estimation
and prediction specifically of extremes are of particular importance.
Because KF is a least squares solution, it is subject to conditional
biases (CB) which arise from the error-in-variable, or attenuation,
effects when the model dynamics are highly uncertain, the observations
have large errors and/or the system being modeled is not very
predictable. In this presentation, we first describe conditional
bias-penalized Kalman filter (CBPKF) and comparatively evaluate CBPKF
with KF through synthetic experiments. We then present comparative
evaluation of conditional bias-penalized ensemble Kalman filter (CEnKF)
with EnKF for assimilation of streamflow observations into a lumped
hydrologic model of the NWS for streamflow prediction. The results show
that CBPKF reduces root mean square error over KF by 10 to 20% or more
over the tails of the distribution of the true state, and that CEnKF
significantly improves skill of ensemble streamflow prediction over
EnKF for lead times of up to the fast response time of the catchment.
Also shared in this presentation are the areas of future research,
including reduction in computational amount. With the ability to reduce
CB explicitly, CBPKF and CEnKF provide significant new additions to the
existing suite of data assimilation techniques for improved analysis
and prediction of extreme states of uncertain environmental systems.