HR: 1340h
AN: H33D-0500    [Abstracts]
TI: KNOWLEDGE DISCOVERY IN HYDROLOGIC DATA: A FRAMEWORK FOR SIMULATION AND PREDICTION
AU: * Khalil, A F
EM: akhalil@cc.usu.edu
AF: Abedalrazq F. Khalil, 1600 Canyon Road Utah Water Research Lab, Logan, Ut 84322 United States
AB: Uncertainty, non-stationarity, noise, and paucity of data all limit the prediction capabilities of hydrologic models. In this paper, we adopt a Bayesian predictive approach for forecasting that combines the features of excellent generalization properties and sparse representation. There are three novelties in the resulting framework: first, the uncertainty in model parameters is incorporated in the prediction; second, a multi-objective optimization algorithm is employed to account for the uncertainty in model structure (i.e., optimal model selection); and third, this framework allows AŸA›A›ƒ_sAªA<�"detection of shiftsAŸA›A›ƒ_sAªA›ƒ_zA› in the sense that, as we observe the behavior of a process through time, our framework detects drift (e.g., changes in the hydrologic processes caused by climatic changes) and thence initiates adaptations in model structure in response to a recognized shift in the underlying processes. Finally, given knowledge of some state and exogenous conditions, the framework is applied in an on-line fashion to provide probabilistic forecasts of future system states. For many hydrologic systems of practical interest, these forecasts can be accomplished in real-time and can provide valuable management information.
DE: 1899 General or miscellaneous
DE: 1894 Instruments and techniques
SC: Hydrology [H]
MN: 2004 AGU Fall Meeting