HR: 1340h
AN: H23A-1129    [Abstracts]
TI: Multiconstituent Reaction Identification in Groundwater Modeling
AU: * Thomas, B
EM: bthomas@ucla.edu
AF: UCLA, 5732 Boelter Hall Box 951593, Los Angeles, CA 90095 United States
AU: Yeh, W
EM: williamy@seas.ucla.edu
AF: UCLA, 5732 Boelter Hall Box 951593, Los Angeles, CA 90095 United States
AB: The selection of appropriate reaction terms for coupled advection-dispersion-reaction (ADR) groundwater systems is a difficult enterprise. Choosing a proper analytical form is crucial to the accurate prediction of system behavior. The decision can be daunting in the absence of accurate a priori biological, chemical, and geophysical properties of the site. Moreover, once the functional form is chosen, the analyst must then calibrate that reaction functionAŸA›A›ƒ_sAªA›ƒ_zA›s parameters. The calibration process itself can be challenging, especially in the case of complex empirical reaction models. Our research investigates the utility of avoiding the direct choice of a reaction function. Instead of picking a particular analytic reaction term, we construct a reaction function as best as possible given available data. The reaction term is developed by conjoining an optimal set of hyperplanes which approximate the functional geometry of the complex, multiconstituent reactions that the existing measurements suggest. The methodology employs a genetic algorithm (GA) to identify increasingly complex sets of junction nodes in the reaction space, sequential quadratic programming (SQP) to compute an ordinate at each node, and Delaunay triangulation to combine this information into a reaction surface of intersecting hyperplanes. The GA and the SQP ally with the ADR simulation itself to provide the most elementary functional approximation which satisfies the prediction and reliability requirements of the analyst. Results will be shown which demonstrate the utility of this multidimensional inverse modeling approach. Analyses will also be presented which compare an integral goodness of fit metric that indicates which member of a family of analytic functions the reaction surface approximation most closely resembles. In so doing, this methodology serves to provide both the most reliable reaction function which the data allow and to indicate to the researcher which analytic form is most likely to be present in the system studied.
DE: 1829 Groundwater hydrology
DE: 1832 Groundwater transport
SC: Hydrology [H]
MN: 2004 AGU Fall Meeting