Please use this identifier to cite or link to this item:
Scopus Web of Science® Altmetric
Type: Journal article
Title: Model smoothing strategies to remove microscale discontinuities and spurious secondary optima in objective functions in hydrological calibration
Author: Kavetski, D.
Kaszuba, J.
Citation: Water Resources Research, 2007; 43(3):1-9
Publisher: Amer Geophysical Union
Issue Date: 2007
ISSN: 0043-1397
Statement of
Dmitri Kavetski and George Kuczera
Abstract: <jats:p>Environmental processes frequently exhibit threshold‐type behavior, e.g., the initiation of fluxes such as snowmelt, recharge, and quick flow. Incorporating such thresholds into hydrological models introduces discontinuities into the objective functions used in model calibration, making parameter estimation unnecessarily more difficult. Moreover, this study shows that model thresholds can produce spurious multimodality in least squares objective functions even if the underlying model is near linear in its parameters. In contrast, smoothing the model with respect to its parameters and inputs yields differentiable objective functions and, in some cases, can also improve its macroscale characteristics by removing spurious secondary optima. This simplifies model calibration and sensitivity analysis by reducing the complexity of objective functions and permitting the use of powerful derivative‐based analysis methods such as Newton‐type optimization and Hessian‐based uncertainty assessment. This paper details smoothing strategies for several classes of thresholds and discontinuities commonly found in hydrological models, including step and angle discontinuities in the constitutive functions and flux constraints arising from conservation laws in the governing differential equations. The smoothing algorithms and their parameters are selected to ensure infinite differentiability of the model and its objective functions while preserving the macroscale behavior of the original governing equations. The improvements in the structure of the model and its objective functions are illustrated empirically for a degree‐day‐based snow model. The smoothing techniques are general and can be applied to other models with thresholds.</jats:p>
Description: Extent: 9p.
Rights: Copyright 2007 by the American Geophysical Union
DOI: 10.1029/2006WR005195
Published version:
Appears in Collections:Aurora harvest
Civil and Environmental Engineering publications
Environment Institute publications

Files in This Item:
There are no files associated with this item.

Items in DSpace are protected by copyright, with all rights reserved, unless otherwise indicated.