Evaluating the effect of Tikhonov regularization schemes on predictions in a variable‐density groundwater model

By: , and 

Links

Abstract

Calibration of highly‐parameterized numerical models typically requires explicit Tikhonovtype regularization to stabilize the inversion process. This regularization can take the form of a preferred parameter values scheme or preferred relations between parameters, such as the preferred equality scheme. The resulting parameter distributions calibrate the model to a user‐defined acceptable level of model‐to‐measurement misfit, and also minimize regularization penalties on the total objective function. To evaluate the potential impact of these two regularization schemes on model predictive ability, a dataset generated from a synthetic model was used to calibrate a highly-parameterized variable‐density SEAWAT model. The key prediction is the length of time a synthetic pumping well will produce potable water. A bi‐objective Pareto analysis was used to explicitly characterize the relation between two competing objective function components: measurement error and regularization error. Results of the Pareto analysis indicate that both types of regularization schemes affect the predictive ability of the calibrated model.

Additional publication details

Publication type Conference Paper
Publication Subtype Conference Paper
Title Evaluating the effect of Tikhonov regularization schemes on predictions in a variable‐density groundwater model
Year Published 2010
Language English
Contributing office(s) Florida Water Science Center-Ft. Lauderdale
Description 5 p.
Larger Work Type Conference Paper
Larger Work Title SWIM21 – 21st Salt Water Intrusion Meeting Proceedings Book
First page 344
Last page 348
Conference Title 21st Salt Water Intrusion Meeting (SWIM21 – AZORES 2010)
Conference Location Azores, Portugal
Conference Date June 21-26, 2010
Online Only (Y/N) N
Additional Online Files (Y/N) N