HR: 0830h
AN: H11F-0905    [PDF]
TI: Automated Parameterization of a Transpiration Model: A Comparative Study of Bayesian Analysis and a Procedure Based on Fuzzy Set Theory
AU: * Samanta, S
EM: ssamanta@wisc.edu
AF: Department of Forest Ecology and management, University of Wisconsin, 1630 Linden Dr., Madison, WI 53706 United States
AU: Mackay, D S
EM: dsmackay@buffalo.edu
AF: Department of Geography, 105 Wilkeson Quad., University at Buffalo, Buffalo, NY 14261 United States
AB: Bayesian inference using Markov chain simulation methods is a class of extremely useful techniques for drawing inference from uncertain data and parameterization of stochastic models. However, most simulation models in geosciences are deterministic and designed to capture the underlying mechanisms explicitly. Such models may not be easily formulated in stochastic terms. In addition, sometimes there is considerable uncertainty regarding the model structure and its components, and the errors may not be randomly distributed. Even when inputs are considered to have negligible uncertainty, due to the hierarchical nature and existence of feedbacks in many simulation models, it is difficult to set up an appropriate likelihood function for a Bayesian analysis of the model. Consequently, parameterization and uncertainty analysis of these models require considerable a priori knowledge about the joint distribution of parameters as well as the expected error distribution. These information may not always be available. In contrast, an alternative procedure developed using possibility theory in the context of fuzzy sets does not require the above a priori knowledge. The underlying assumption in this case is that the goodness of fit of the model output to a set of observed data can be interpreted as the membership grade function of a fuzzy set comprising of acceptable model-parameter combinations. In this study, the proposed method is compared extensively with Markov chain simulation procedure under various error models. The canopy conductance sub-model embedded within the Penman-Monteith equation for transpiration is used for this study. The analysis was conducted using synthetic data as well as measurements from Chequamegon Ecosystem Atmosphere Study (ChEAS) site in Wisconsin. Parameterization and uncertainty analysis using the fuzzy set based technique were found to be consistent with those obtained from the Bayesian analysis.
DE: 1694 Instruments and techniques
DE: 1818 Evapotranspiration
DE: 1894 Instruments and techniques
SC: Hydrology [H]
MN: 2003 Fall Meeting