HR: 0800h
AN: H21A-1311    [Abstracts]
TI: Parameter estimation of hydrologic models using data assimilation
AU: * Kaheil, Y H
EM: yasir@cc.usu.edu
AF: Utah Water Reseach Lab Civil and Env. Eng Dept, 8200 Old Main Hill, Logan, UT 84322-8200 United States
AB: The uncertainties associated with the modeling of hydrologic systems sometimes demand that data should be incorporated in an on-line fashion in order to understand the behavior of the system. This paper represents a Bayesian strategy to estimate parameters for hydrologic models in an iterative mode. The paper presents a modified technique called localized Bayesian recursive estimation (LoBaRE) that efficiently identifies the optimum parameter region, avoiding convergence to a single best parameter set. The LoBaRE methodology is tested for parameter estimation for two different types of models: a support vector machine (SVM) model for predicting soil moisture, and the Sacramento Soil Moisture Accounting (SAC-SMA) model for estimating streamflow. The SAC-SMA model has 13 parameters that must be determined. The SVM model has three parameters. Bayesian inference is used to estimate the best parameter set in an iterative fashion. This is done by narrowing the sampling space by imposing uncertainty bounds on the posterior best parameter set and/or updating the "parent" bounds based on their fitness. The new approach results in fast convergence towards the optimal parameter set using minimum training/calibration data and evaluation of fewer parameter sets. The efficacy of the localized methodology is also compared with the previously used Bayesian recursive estimation (BaRE) algorithm.
DE: 0555 Neural networks, fuzzy logic, machine learning
DE: 1805 Computational hydrology
DE: 1816 Estimation and forecasting
DE: 1866 Soil moisture
DE: 1869 Stochastic hydrology
SC: Hydrology [H]
MN: Fall Meeting 2005