HR: 08:30h
AN: H11B-03    [PDF]
TI: Combined Parameter and State Estimation of Hydrological Models Using Ensemble Kalman Filter
AU: * Moradkhani, H
EM: moradkha@uci.edu
AF: Civil & Environmental Eng. UC Irvine, Civil & Environmental Engineering E/4130 Engineering Gateway, Irvine, CA 92697 United States
AU: Gupta, H
EM: hoshin@hwr.arizona.edu
AF: Department of Hydrology, University of Arizona, Department of Hydrology, University of Arizona, Tucson, AZ 85721 United States
AU: Sorooshian, S V
EM: soroosh@uci.edu
AF: Civil & Environmental Eng. UC Irvine, Civil & Environmental Engineering E/4130 Engineering Gateway, Irvine, CA 92697 United States
AU: Houser, P R
EM: Paul.R.Houser@nasa.gov
AF: Hydrological Sciences Branch, NASA GSFC, Hydrological Sciences Branch, NASA-GSFC, Greenbelt, MD 20771 United States
AB: The response of hydrologic models is mainly defined by parameters, as the physical and generally time-invariant representation of watershed characteristics and states, as storages of water that are propagated by model physics. Some of the parameters may be measured directly but others are not easily estimated, therefore adjustment of initial values or calibration is needed. The model calibration process attempts to minimize the systematic bias arising due to inaccurate parameterization. Calibration is commonly accomplished in a batch-processing scheme where the available data is used at once to minimize the long-term bias in the simulation. On the other hand, to estimate the state variables of a system, data assimilation techniques, which are sequential data processing algorithms, are commonly used. Unlike the batch-processing scheme, data assimilation does not require long data record to be kept in storage. It has the capability to estimate the variables at each time step. Ensemble Kalman filter (EnKF) is an efficient data assimilation procedure for state estimation in hydrological modeling. The main goal of this study is to use the state augmentation technique in the context of EnKF to estimate both state variables and parameters simultaneously in a conceptual rainfall-runoff model. In this procedure one can detect the time variation of parameters while minimizing the short-term bias in the system. Different sources of uncertainties including forcing data (precipitation) error and output (streamflow) error are considered. The impact of the magnitude of perturbation against the number of ensemble members on the efficiency and accuracy of the estimation is investigated and a procedure for tuning these hyper-parameters is proposed.
DE: 1831 Groundwater quality
DE: 1832 Groundwater transport
DE: 1871 Surface water quality
DE: 1884 Water supply
DE: 1899 General or miscellaneous
SC: Hydrology [H]
MN: 2003 Fall Meeting